Glioblastoma (GBM) is the deadliest form of brain cancer. It is a highly angiogenic and immunosuppressive malignancy. Although immune checkpoint blockade therapies have revolutionized treatment for many types of cancer, their therapeutic efficacy in GBM has been far less than expected or even ineffective. In this study, we found that the genomic signature of glioma-derived endothelial cells (GdEC) correlates with an immunosuppressive state and poor prognosis of patients with glioma. We established an in vitro model of GdEC differentiation for drug screening and used this to determine that cyclic adenosine monophosphate (cAMP) activators could effectively block GdEC formation by inducing oxidative stress. Furthermore, cAMP activators impaired GdEC differentiation in vivo, normalized the tumor vessels, and altered the tumor immune profile, especially increasing the influx and function of CD8+ effector T cells. Dual blockade of GdECs and PD-1 induced tumor regression and established antitumor immune memory. Thus, our study reveals that endothelial transdifferentiation of GBM shapes an endothelial immune cell barrier and supports the clinical development of combining GdEC blockade and immunotherapy for GBM.

See related Spotlight by Lee et al., p. 1300

Glioblastoma (GBM) is an aggressive brain tumor, and patients have a median survival of less than 21 months despite surgical resection, radiotherapy, and chemotherapy (1, 2). Although immune checkpoint blockade (ICB) therapy has revolutionized current cancer treatment strategies and has shown remarkable success in treating various tumors (3–5), it has shown little clinical benefit in GBM (6–9).

The response of patients with cancer to ICB therapy is dependent on the immune component of the tumor microenvironment (TME; ref. 10). The TME of GBM is highly immunosuppressive and differs from that of other malignancies as a result of the immune privilege status of the brain, low tumor mutational burden, and vascular dysplasia (7, 11, 12). Amid the bourgeoning successes of cancer immunotherapy, GBM has emerged as a model of resistance to immunotherapy (1, 6).

Mounting evidence shows that ongoing tumor angiogenesis contributes to immune evasion through the induction of a highly immunosuppressive TME (12). The key cytokines of angiogenesis affect immune responses. For example, VEGF has been found to inhibit T-cell development and function (13) and to promote T-cell exhaustion through the upregulation of immune checkpoint proteins (14). On the other hand, circulating immune cells rely on a functional vascular network to enter tumor tissues (12). The vasculature of tumors is morphologically abnormal and dysfunctional, which results in aberrant blood perfusion and oxygenation (15). This abnormal vasculature also directly inhibits leukocyte adhesion and extravasation (16, 17). Therefore, the angiogenic tumor vasculature establishes a physical barrier for circulating leukocytes.

GBM is a highly angiogenetic malignancy that has diverse vasculature cell populations (18). It has been reported that the abundant vasculature of GBM is partly derived from glioma stem cells (GSC), which have the capacity to differentiate into endothelial cells (EC) that are termed glioma-derived endothelial cells (GdEC; refs. 18–23) and pericytes (24–26), both of which contribute to GBM vascularization in vivo. However, the effect of GdECs on the GBM TME has not been elucidated; nor is it known whether targeting GdECs can promote the efficacy of immunotherapy.

In this study, we reveal that GdECs contribute to shaping an endothelial immune cell barrier and the immunosuppressive microenvironment in GBM. GdEC blockade led to remodeling of the tumor immune microenvironment and improved the therapeutic efficacy of ICB. Our findings identify a mechanism of tumor immune escape by which tumor cells directly transdifferentiate into endothelial-like cells to limit antitumor CD8+ T-cell immunity.

Cell culture and reagents

U87, GL261, CT2A, and B16F10 cells were cultured in DMEM (Gibco, 11965092) supplemented with 10% FBS (Gibco, 10099141) and penicillin/streptomycin (Gibco, 15070063) at 37°C under 5% CO2. Human umbilical vein endothelial cells (HUVEC) were cultured in endothelial cell medium (ECM; ScienCell, 1001) at 37°C under 5% CO2. GL261 and CT2A were provided by Prof. Guangmei Yan (Sun Yat-Sen University; Guangzhou, P.R. China) in 2018 and Prof. Jun Chen (Sun Yat-Sen University, Guangzhou, P.R. China) in 2021, respectively. U87 and HUVEC were purchased from the ATCC in 2018. B16F10 cells were purchased from ATCC in 2021.

Primary patient-derived GSCs (GSC1, GSC11, GSC21, GSC23, and GSC24) were provided by Prof. Guangmei Yan (Sun Yat-sen University, Guangzhou, P.R. China) in 2017. These primary patient-derived GSCs were cultured under clonal conditions in low-adherence plates (Corning, 3471) and maintained in DMEM/F12 medium (Gibco, 11320033) supplemented with 2% B27 (Gibco, 17504044), 20 ng/mL basic FGF (bFGF; Peprotech, 100-47), and 10 ng/mL EGF (Peprotech, 100-47) at 37°C under 5% CO2. Tumor spheres were reseeded every 4 days after dissociation with Accutase (Innovative Cell Technologies, A6964).

All cell lines were tested and found to be free of Mycoplasma, while no further cell authentication assays were carried out. Cells were expanded into low passage numbers and working stocks were frozen down. Cells were passaged two to three times post thaw before use for in vitro and in vivo analyses. The GFP and luciferase cell lines were generated by transducing the parental cells with lentivirus for 4 hours, and the medium was replaced with fresh medium. A total of 48 hours after transduction, the cells were selected for 7 days with the treatment of puromycin (Invivogen, ant-pr-1).

The following reagents were used in the study as indicated below: ibudilast (10 mmol/L, dissolved in DMSO, HY-B0763, MedChemExpress), N-acetylcysteine (10 mmol/L, dissolved in DMSO, S1623, Selleck Chemicals), MnTBAP (10 mmol/L, dissolved in DMSO, HY-126397, MedChemExpress), anti-mouse PD-1 (CD279; BE0146, RRID: AB_10949053, BioXCell), rat IgG2b (BE0090, RRID: AB_1107780, BioXCell), anti-mouse CD4 (BE0003-1, RRID: AB_1107636, BioXCell), and anti-mouse CD8 (BE0061, RRID: AB_1125541, BioXCell).

Compound library screened for ability to block GdEC formation

The 40 small molecules used were as follows: SB431542 (S1067), Galunisertib (S2230), RepSox (S7223), IWR-1-endo (S7086), XAV-939 (S1180), PRI-724 (S8968), Vismodegib (S1082), Sonidegib (S2151), PF-5274857 (S2777), FLI-06 (S7399), Crenigacestat (S7169), IMR-1 (S8280), RO4929097 (S1575), BIX01294 (S8006), Tubacin (S2239), IOX1 (S7234), Daminozide (S4800), JIB-04 (S7281), GSK-J4 (S7070), GSK-J1 (S7070), CPI-455 (S8287), ML324 (S7296), Fenretinide (S5233), Ivosidenib (S8206), Vorasidenib (S8611), BAY1436032 (S8530), AGI-5198 (S7185), Enasidenib (S8205), AGI-6780 (S7241), BPTES (S7753), CB-839 (S7655), UPGL00004 (S8778), Dynasore (S8047), Dyngo-4a (S7163). These drugs were dissolved in DMSO to prepare 10 mmol/L concentration solutions. Cyclopamine (10 mmol/L, dissolved in ethanol, S1146) and dbcAMP (100 mmol/L, dissolved in double-distilled water (ddH2O), S7858), Forskolin (20 mmol/L, dissolved in DMSO, S2449), Mdivi-1 (25 mmol/L, dissolved in DMSO, S7162), and IWP-2 (10 mmol/L, dissolved in dimethyl formamide, S7085). All the compounds were purchased from Selleck Chemicals.

Cell counting kit‐8 assay

To select a working concentration of the compounds in the library screened for ability to block GdEC formation, GSC11 cells were seeded in low-adherence 96‐well plates (Corning, 3474) at a density of 10,000 cells per well. The compounds were added immediately after the cells were seeded. A total of 48 hours after compound treatment, 10 μL of cell counting kit‐8 (CCK‐8) (Dojindo Laboratories, CK04) was added to each well and incubated for 3 hours. Absorbance was then read at 450 nm by a microplate reader (Winooski). The 10% inhibitory concentration (IC10) was measured and selected as the working concentration for the subsequent screening.

Matrigel-based tube formation assay for screening compounds for ability to block GdEC formation

The tubular formation assay was performed as described previously (20). A total of 200 μL of matrigel (Corning, 356234) was added into 24-well plates and allowed to polymerize for 20 minutes at 37°C. A total of 120,000 cells (GSC1, GSC11, GSC21, GSC24, HUVEC, or U87 cells) per well were then seeded on top of the matrigel. Then ECM (ScienCell, 1001) was added, and the cells were cultured for 8 to 12 hours. Tube formation images were acquired using a Nikon ECLIPSE Ti microscope (Nikon) and tube length measurements were performed with Image-Pro Plus software.

Cell metabolism measurement

Cellular oxidative phosphorylation and glycolysis were measured by real-time oxygen consumption rate (OCR) and extracellular acidification rate (ECAR) with the Seahorse Bioscience extracellular flux analyzer (XF24, Seahorse Bioscience) as described previously (27). GSC11 cells were treated with dbcAMP (1 mmol/L) for 24 or 48 hours, then 25,000 cells were seeded in specific 24-well plates in an appropriate growth medium and incubated overnight. Cells were washed and immersed with an unbuffered medium [XF base medium minimal DMEM (Agilent Technologies, 13417002) supplemented with 1% glutamine (Gibco, 25030081), 4.5 g/L glucose (Agilent Technologies, 103015)] and incubated in the absence of CO2 for 1 hour before measurements. For OCR measurement, we used seahorse XF Cell Mito Stress Test Kit (Agilent Technologies, 103015) and for ECAR measurement we used seahorse XF glycolytic rate assay kit (Agilent Technologies, 103344). The OCR and ECAR were measured in a typical 8-minute cycle of the mix (2–4 minutes), dwell (2 minutes), and measure (2–4 minutes), as recommended by Seahorse Bioscience.

Sphere formation assay

GSCs (1,000 cells/mL) were cultured in ultra-low attachment 24-well plates (Corning, 3473) in DMEM/F12 medium supplemented with 2% B27, 20 ng/mL bFGF, and 10 ng/mL EGF. dbcAMP was added immediately after the cells were seeded. 24 or 48 hours after dbcAMP treatment, the reagent was removed by centrifugation and replaced with fresh GSC culture medium for culturing for 10 days. Tumor spheres with a diameter >75 μm were counted. Five replicate wells were included in each analysis.

Limiting dilution assay

GSC1 and GSC11 cells were treated with ibudilast (10 μmol/L) for 48 hours, then the cells were plated at decreasing densities (400, 200, 100, 50, 25, and 1 cells/well) in ultra-low attachment 96-well plates (Corning) with fresh GSC culture medium. Each well was examined for the formation of tumor spheres after 14 days. Each density was seeded in replicates of 10. The self-renewal capacity of the cells was calculated using Extreme Limiting Dilution Analysis (ELDA) software, available at http://bioinf.wehi.edu.au/software/elda/.

Annexin V/7-AAD, 7-Aminoactinomycin D assay

Cell death was quantified with an APC Annexin V Apoptosis Detection Kit with 7-AAD, 7-Aminoactinomycin D (7-AAD; BioLegend, 640930) according to the manufacturer's instructions. GSC1 and GSC11 cells were treated with ibudilast (10 μmol/L) for 48 hours, then the cells were digested with Accutase (Innovative Cell Technologies, A6964) and stained at room temperature for 15 minutes with Annexin V-APC and 7-AAD staining solution. All samples were measured by a flow cytometer (CytoFLEX, Beckman). Data were analyzed with FlowJo software (v.10; Tree Star). Technicians acquiring and gating the data were blinded to the treatments.

cAMP assay

GSC1 and GSC11 cells were seeded in a low-adherence T75 culture flask and cultured for 48 hours. After starvation in serum-free medium for 12 hours, cells were then treated with ibudilast (10 μmol/L) for 2 hours. Then, the cells were washed with precooled PBS three times, and the cell lysates were collected to detect intracellular cyclic adenosine monophosphate (cAMP) levels using the cAMP Parameter Assay Kit (KGE002B, R&D Systems) according to the manufacturer's protocol. Absorbance was then read at 450 nm by a microplate reader (Winooski).

Flow cytometry analysis

For mitochondria reactive oxygen species (ROS) measurements, ROS production was measured with Cell ROX Deep Red reagent (Thermo Fisher Scientific, C10422). Cells were treated with dbcAMP (1 mmol/L) or MnTBAP (80 μmol/L) for the indicated time and then stained with 5 μmol/L of CellROX Deep Red Reagent by adding the probe to the complete medium and incubating the cells at 37°C for 30 minutes before analysis.

For cell proliferation assessment, EdU (5-ethynyl-2´-deoxyuridine) activity was measured using the Click-iT EdU Alexa Fluor 647 Flow Cytometry Assay Kit (Thermo Fisher Scientific, C10424). Cells were incubated with 10 μmol/L EdU for 2 hours at 37°C. Then, harvested cells were added 100 μL of Click-iT fixative for 15 minutes at room temperature, protected from light. The supernatant was removed by centrifugation, and cells were resuspended in 100 μL of 1X Click-iT saponin-based permeabilization for 15 minutes. 0.5 mL of Click-iT reaction cocktail was added for 30 minutes.

For CD31, CD34, CD105, and CD133 expression measurements, cells were harvested and incubated with FcR blocking reagent (Miltenyi Biotec, 130-059-901) and then stained using fluorescently conjugated anti-human CD31 (Invitrogen, 63-0319-42); anti-human CD34 (BioLegend, 343612); anti-human CD105 (BioLegend, 323218), and anti-human CD133 (Miltenyi Biotec, 130-113-106), and incubated for 20 minutes in the dark in the refrigerator (4°C).

For in vivo lineage tracing of GdEC, GL261 xenograft was separated, harvested, and passed through a 70 mm strainer, then resuspended in FACS buffer (2% inactivated FBS in PBS). Dead cells were stained with a Zombie Red Fixable Viability Kit (BioLegend, 423109). Cells were stained with fluorochrome-conjugated anti-human CD31, anti-human CD34, and anti-human CD105. We performed cell surface staining according to BioLegend's Cell Surface Immunofluorescence Staining Protocol (https://www.biolegend.com/protocols/cell-surface-flow-cytometry-staining-protocol/4283/).

For analysis of tumor-infiltrating immune cells, GL261 and CT2A tumors were harvested, minced, and dissociated to single-cell suspension using Tumor Dissociation Kit (Miltenyi Biotec, 130-095-929), then passed through a 70 μm screen (Falcon, 352350), and resuspended in cell staining buffer (BioLegend, 420201). Dead cells were stained with Zombie Red Fixable Viability Kit. Then cells were stained with fluorochrome-conjugated anti-mouse antibodies from BioLegend [anti-mouse CD45, 103138; anti-mouse CD3ε, 100330; anti-mouse CD4, 100528; anti-mouse CD8a, 100730; anti-mouse/human CD11b, 101228; anti-mouse Ly-6G/Ly-6C (Gr-1), 108407]. We performed dead cell and cell surface staining as described in BioLegend's Cell Surface Immunofluorescence Staining Protocol. Intracellular staining (anti-mouse/rat/human FOXP3, BioLegend, 320014; anti-human/mouse Granzyme B, BioLegend, 515408) was also performed following the BioLegend's Intracellular Flow Cytometry Staining Protocol (https://www.biolegend.com/protocols/intracellular-flow-cytometry-staining-protocol/4260/).

All samples were run on a CytoFLEX flow cytometer (Beckman) and were analyzed with FlowJo software (v.10; Tree Star). Technicians acquiring the data were blinded to the treatments.

Subcutaneous and intracranial xenograft GBM models

Six- to 8-week-old female mice (C57BL/6 and nude mice) were obtained from GemPharmatech and housed in a pathogen-free animal facility. All animal experiments were approved by the Institutional Ethics Committee of the School of Medicine, Sun Yat-sen University (Guangzhou, P.R. China).

For subcutaneous xenograft GBM models, the hind flanks of nude mice were subcutaneously injected with 2 × 105 GSC1 cells expressing luciferase to generate a subcutaneous xenograft GBM model. After 5 days, the mice were randomly divided into two groups (n = 11), and were intraperitoneally injected with vehicle or 10 mg/kg bevacizumab (two times a week), respectively, for 2 weeks.

For intracranial GBM models, dissociated glioma cells were injected 2 mm lateral and 0.5 mm anterior to the bregma and 2.5 mm below the skull of mice to generate intracranial xenograft GBM models as described previously (27). Nude mice were orthotopically transplanted with 3 × 105 human GSC1 cells expressing luciferase. C57BL/6 mice were orthotopically transplanted with 3 × 105 mouse GL261 cells or CT2A cells. The mice were intragastrically administrated with vehicle or 20 mg/kg ibudilast (five times a week) for 2 weeks.

For the combination experiment, C57BL/6 mice were orthotopically transplanted with 3 × 105 GL261 cells or CT2A cells. Mice were randomly divided into four groups: a control group, an ibudilast group, an anti–PD-1 group, and a combination group. The mice were intragastrically administrated with vehicle or 20 mg/kg ibudilast (five times a week) for 2 weeks. Anti–PD-1 (10 mg/kg) or isotype control antibodies rat IgG2b (10 mg/kg) were intraperitoneally administered three times (three times a week).

For the rechallenge experiment, long-term surviving mice were rechallenged with an increased dose of cells (6 × 105 GL261 or B16F10) in the contralateral hemisphere. Age-matched naïve mice were implanted with the same number of cells as the control group.

For the immune cell depletion study, C57BL/6 mice were orthotopically transplanted with 3 × 105 GL261 cells. Mice were randomly divided into three groups: an isotype group, an anti-CD4 group, and an anti-CD8 group. Anti-mouse CD4, anti-mouse CD8, or IgG2b isotype antibody (10 mg/kg) were intraperitoneally injected on days 3, 5, 7, 9, and 11. Mice were treated with anti–PD-1 (10 mg/kg) on days 7, 9, 11, or 20 mg/kg ibudilast (five times a week) respectively for 2 weeks.

The mice were monitored daily, and a Kaplan–Meier survival curve was generated using GraphPad Prism 6.0 software package (GraphPad). The points on the curves represent glioma-related deaths (P was determined by log-rank analysis). Mouse brains were dissected and fixed in formalin for hematoxylin and eosin (H&E) and IHC staining.

Bioluminescent imaging and analysis

Mice were intraperitoneally injected with 1.5 mg of d-luciferin (Promega, P1042; 15 mg/mL in PBS). Imaging was completed within 10 minutes of d-luciferin injection using an IVIS Lumina S5 Imaging System and the data were analyzed using Aura Imaging Software. For the bioluminescence image plots, photon flux was evaluated for each mouse using an oval region of interest encompassing the mouse head.

Tissue histologic staining and image quantification

For H&E staining, tumors and indicated organs were fixed overnight in 4% paraformaldehyde (PFA). The samples were dehydrated at a gradient concentration, embedded in paraffin, and cut into 5 μm sections followed by H&E standard staining. Briefly, tissue sections were processed at room temperature by dipping the slides in a series of baths as follow: xylene (10 minutes; twice), gradual rehydration using 100%, 90%, and 70% ethanol treatment (5 minutes each), distilled water (30 seconds), hematoxylin (5 minutes; H&E Staining Kit, Beyotime Biotechnology, C0105S); distilled water (1 minute); 0.5% eosin Y (2 minutes; H&E Staining Kit, Beyotime Biotechnology, C0105S); 95% Ethanol (30 seconds; twice); 100% ethanol (30 seconds; twice); and xylene (30 seconds; twice). Neutral balsam mounting medium (Solarbio, G8590) and glass cover slips (CITOTEST, 80330-0130) were used to mount the slides. The H&E staining images were acquired using a Nikon ECLIPSE Ti microscope (Nikon).

For IHC staining, tissue sections were processed for antigen retrieval with Tris-EDTA buffer (Servicebio, G1203) followed by incubation with primary antibodies (4°C, overnight). Then sections were incubated with a secondary antibody (Dako REAL EnVision, K5007), and nuclei were counterstained with hematoxylin. Primary antibodies used were anti-Ki67 (1:500, Cell Signaling Technology, 9449S), anti-Malondialdehyde (MDA; 1:200, Abcam ab27642), anti-human CD31 (1:100, Abcam ab187377), and anti-mouse CD31 (1:200, Servicebio, GB12063). The IHC staining images were acquired using a Nikon ECLIPSE Ti microscope (Nikon).

For immunofluorescence staining, tissues, and indicated organs were fixed overnight in 1% PFA. The samples were dehydrated in 20% sucrose solution overnight, embedded in optimal cutting temperature (SAKURA, 4583), and cut into 10 μm sections. Frozen sections were processed for antigen retrieval with Tris-EDTA buffer (Servicebio, G1203) followed by blocking with 5% donkey serum (Abcam, ab7475) in Phosphate Buffer Solution (PBST; 0.2% Triton X-100 in PBS), and then incubated overnight at 4°C with the following primary antibodies: anti-GFP (1:500, Servicebio, GB11062), anti-human CD31 (1:100, Abcam ab187377), anti-mouse CD31 (1:200, Servicebio, GB12063), and anti-α-Smooth Muscle (1:100, Sigma, A5228). The samples were then incubated for 1 hour at room temperature with the following secondary antibodies: 488 donkey anti-mouse IgG (Invitrogen, A21202) or 555 donkey anti-rabbit IgG (Invitrogen, A31570).

For hypoxic staining, Hypoxyprobe-1 (60 mg/kg, Hypoxyprobe, HP1-100) was intraperitoneally injected 60 minutes before perfusion fixation. The tumors were then harvested, sectioned, and stained with FITC-conjugated anti-Hypoxyprobe (Hypoxyprobe, HP1-100).

For the in vivo vascular leakage assay, mice were injected in the tail vein with 100 μL of rhodamine-conjugated dextran (25 mg/mL, 70 kDa, Sigma-Aldrich, 46945) 30 minutes before euthanasia. Mice were perfused by intracardiac injection of 1% PFA to remove circulating dextran. The tumors were then harvested and sectioned.

Nuclei were stained with Hoechst (Sigma, 14533). Then the sections were mounted with a fluorescent mounting medium (DAKO, S3023) and immunofluorescent images were acquired using a Nikon A1 spectral confocal microscope (Nikon).

Single-cell RNA sequencing analysis

Mice bearing GL261 tumors were intragastrically administrated with vehicle or 20 mg/kg ibudilast (five times a week) for 2 weeks. Tumor-infiltrating lymphocytes were isolated using the Brain Tumor Dissociation Kit (#130-095-942, Miltenyi Biotec) and sorted for CD45+ lymphocytes by EasySep Mouse TIL (CD45) Positive Selection Kit (#100-0350, STEMCELL Technologies).

Single-cell RNA sequencing (scRNA-seq) libraries were generated using the 10X Genomics Chromium Controller Instrument and Chromium Single Cell 3′ V3 Reagent Kits (10X Genomics). CD45+ cell solutions with 1,000 cells/μL were loaded into each channel to generate single-cell Gel Bead-In-Emulsions (GEM). After the reverse transcription step, GEMs were broken and barcoded-cDNA was purified and amplified. The amplified barcoded cDNA was fragmented, A-tailed, ligated with adaptors and index PCR amplified. The final libraries were quantified using the Qubit High Sensitivity DNA assay (Thermo Fisher Scientific, Q32852) and the size distribution of the libraries were determined using a High Sensitivity DNA chip on a Bioanalyzer 2200 (Agilent). All libraries were sequenced by NovaSeq (Illumina) on a 150 bp paired-end run.

Statistical analysis of scRNA-seq data

scRNA-seq data analysis was performed by NovelBio Co. Ltd. with NovelBrain Cloud Analysis Platform (www.novelbrain.com). We applied fastp with a default parameter filtering for the adaptor sequence and removed the low-quality reads to achieve the clean data. The feature-barcode matrices were obtained by aligning reads to the mouse genome (mm10, ensembl 100) using CellRanger v6.1.1. We applied down sample analysis among samples sequenced according to the mapped barcoded reads per cell of each sample and finally achieved the aggregated matrix. Cells containing over 800 expressed genes and a mitochondria unique molecular identifier (UMI) rate below 10% passed the cell quality filtering and mitochondria genes were removed in the expression table.

Seurat package (version: 4.0.3, https://satijalab.org/seurat/) was used for cell normalization. Regression was based on the expression table according to the UMI counts of each sample and percent of mitochondria rate to obtain the scaled data. Principal component analysis (PCA) was constructed on the basis of the scaled data with top 2,000 high variable genes and top 10 principals used for t-distributed stochastic neighbor embedding (t-SNE) construction and uniform manifold approximation and projection construction. Utilizing a graph-based cluster method, we acquired the unsupervised cell cluster result based on the PCA top 10 principal and we calculated the marker genes by FindAllMarkers function and the Wilcoxon rank-sum test algorithm using following criteria:1. log2FC > 0.25; 2. Pvalue < 0.05; 3. min.pct > 0.1. To identify the cell type detailed, clusters of the same cell type were selected for re-t-SNE analysis. They were graphed on the basis of clustering and marker analysis. The scRNA-seq data have been deposited in the Gene Expression Omnibus (GEO) under accession no. GSE205198.

Morphometric analyses

The density measurements for blood vessels, hypoxic area, pericytes, leakage area, and perfusion area were performed with ImageJ software. To quantify the fraction of Ki67+ and MDA+ cells, a semiautomated IHC Profiler software (ImageJ) was utilized to automatically segment cytoplasm and nuclei as described previously (28). To quantify human blood vessel density, the human CD31+ area was analyzed automatically using CellProfiler on random 0.42 mm2 areas in the intratumoral regions of brain sections. Coverage of αSMA+ pericytes was also calculated as corresponding fluorescent positive length along the CD31+ blood vessels in random 0.42 mm2 intratumoral regions. Area of hypoxia was quantified as a ratio of Hypoxyprobe-1+ to unstained region in random 0.42 mm2 areas. The extent of vascular leakage was quantified as a percentage of dextran+ in random 0.42 mm2 areas. Quantification of positive staining was done on five fields of view at 200X magnification from at least 3 mice.

Circulating antibody detection

The serum of tumor-bearing mice was isolated for detection of antibodies against the GL261 cell lysates (29). Cell lysates of GL261 cells were separated by electrophoresis and transferred to nitrocellulose membranes. The membrane was blocked in milk for 1 hour at room temperature and then probed with mouse serum at a 1:200 dilution in 5% BSA, overnight at 4°C. The membrane was washed and probed with horseradish peroxidase (HRP) conjugated to mouse IgG (Sigma, AP160P). The immunoconjugates were detected with an immobilon Western HRP substrate (Millipore, WBKLS0100) using a chemiluminescence imaging system (Bio-Rad).

Transcriptome data processing and functional enrichment analysis

Total RNA was extracted from GSC11 cells using TRIzol reagent (Life Technologies, 15596026CN). Samples were sent to the Beijing Genomics Institution for RNA sequencing (RNA-seq) analysis. Samples were run on Illumina PE150. The RNA-seq reads were mapped and quantified using the Bowtie2 (v2.2.5), and RSEM software package (v1.2.8). The “DEseq2” method was used for differential gene analysis. The RNA-seq data were deposited in the GEO database (GSE197990). Gene set enrichment analysis (GSEA; GSEA software, v4.1.0) and Kyoto Encyclopedia of Genes and Genomes (KEGG) was used to perform the biological functional enrichment analysis. Raw P values from the GSEA were corrected for multiple testing by FDR. Pathways with corrected P values of less than 0.05 were considered significant.

Bioinformatics analysis

The Cancer Genome Atlas (TCGA) glioblastoma and lower grade glioma datasets including gene expression and patient survival information were downloaded from cBioPortal publicly available platforms (https://www.cbioportal.org/). The four glioma patient datasets, including Brain Lower Grade Glioma (TCGA, PanCancer Atlas; n = 514), Glioblastoma (TCGA; n = 206; ref. 30), Glioblastoma Multiforme (TCGA, Firehose Legacy; n = 401; Source data from https://gdac.broadinstitute.org/runs/stddata__2016_01_28/data/GBM/20160128/), and Glioblastoma Multiforme (TCGA, PanCancer Atlas; n = 160). TCGA GBM& low-grade glioma (LGG) merged datasets included Brain Lower Grade Glioma (TCGA, PanCancer Atlas) and Glioblastoma Multiforme (TCGA, PanCancer Atlas) datasets.

The GdEC signature genes were derived from a previous report (ref. 23; Supplementary Table S1). The GdEC score was calculated by single-sample GSEA (ssGSEA) in the The Cancer Genome Atlas (TCGA). For TIMER or CIBERSORT analysis, TCGA-GBM patients (RNA-seq, n = 151) were divided by quartile (n = 37 in Fig. 1A) or trisection (n = 50 in Supplementary Fig. S1A). For overall survival (OS) analysis, TCGA-GBM or LGG patients were divided by quartile.

Figure 1.

A GdEC-related signature correlates with the immunosuppressive microenvironment and poor prognosis in GBM. A and B, Bioinformatics analyses of the TCGA GBM dataset. The GdEC scores were calculated by ssGSEA. The infiltration of different immune cells was evaluated by the TIMER or CIBERSORT algorithm. A, TCGA GBM samples (RNA-seq, n = 151) were divided by quartile: GdEC-high (red column; n = 37) and GdEC-low (blue column; n = 37). B, Correlation analyses of the GdEC scores and the infiltration of immune cells in TCGA GBM merged datasets. The goodness of fit was quantified and shown as r (n = 74). C, The OS of patients with glioma was compared between the GdEC-high group and GdEC-low group from TCGA-GBM & LGG dataset. The patients with glioma were divided by quartile. Data represent the mean ± SD. *, P < 0.05; **, P < 0.01; ***, P < 0.001. P values are determined by two-tailed unpaired Student t test (A); Spearman correlation coefficient test (B); log-rank test (C).

Figure 1.

A GdEC-related signature correlates with the immunosuppressive microenvironment and poor prognosis in GBM. A and B, Bioinformatics analyses of the TCGA GBM dataset. The GdEC scores were calculated by ssGSEA. The infiltration of different immune cells was evaluated by the TIMER or CIBERSORT algorithm. A, TCGA GBM samples (RNA-seq, n = 151) were divided by quartile: GdEC-high (red column; n = 37) and GdEC-low (blue column; n = 37). B, Correlation analyses of the GdEC scores and the infiltration of immune cells in TCGA GBM merged datasets. The goodness of fit was quantified and shown as r (n = 74). C, The OS of patients with glioma was compared between the GdEC-high group and GdEC-low group from TCGA-GBM & LGG dataset. The patients with glioma were divided by quartile. Data represent the mean ± SD. *, P < 0.05; **, P < 0.01; ***, P < 0.001. P values are determined by two-tailed unpaired Student t test (A); Spearman correlation coefficient test (B); log-rank test (C).

Close modal

The infiltrating score of each immune cell was acquired using the Tumor Immune Estimation Resource and CIBERSORT algorithm (TIMER, https://cistrome.shinyapps.io/timer/; CIBERSORT, https://cibersortx.stanford.edu/). Spearman correlation coefficient (Bioinformatics online platforms; http://www.bioinformatics.com.cn/) was used to evaluate the correlation between two variables.

Statistical analysis

Comparisons of tubular formation, seahorse assay, flow cytometric data, and bioluminescent imaging assay were performed using the unpaired Student t test. The quantification of Ki67+ and MDA+ cells, blood vessels, hypoxic area, pericytes, leakage area, and perfusion area was done using Welch t test. A one-way ANOVA was used for single comparisons with more than two groups. Survival data were analyzed using the log-rank test. Correlation data were analyzed using the Spearman correlation coefficient test. Error bars represent the SD of the mean values from either independent experiments or independent samples. Statistical comparisons were performed in GraphPad Prism 8.

Data availability

The data generated in this study are available within the article and its Supplementary Data or from the corresponding author upon reasonable request. Expression profile data generated in this study have been deposited in GEO at GSE205198 and GSE197990.

GdEC-related signature correlates with immunosuppressive state and poor prognosis of patients with glioma

To investigate a correlation between GdECs and the immune state in glioma, we analyzed two public human glioma datasets from TCGA (TCGA-GBM and TCGA-LGG). We performed ssGSEA using the previously reported GdEC gene set (Supplementary Table S1; ref. 23) and analyzed the infiltration of immune cells by either TIMER or CIBERSORT algorithm.

In the cohort with the top GdEC enrichment score (GdEC-high), we observed less infiltration of CD8+ T cells and CD4+ Th cells (Fig. 1A; Supplementary Fig. S1A), compared with the cohort with the bottom GdEC enrichment score (GdEC-low). Conversely, the GdEC-high cohort had higher infiltration of myeloid cells, macrophages, neutrophils, and B cells (Fig. 1A; Supplementary Fig. S1A). Furthermore, the GdEC scores were negatively correlated with infiltration of CD8+ T cells and CD4+ Th cells, and positively correlated with infiltration of myeloid cells, macrophages, neutrophils, and B cells (Fig. 1B; Supplementary Fig. S1B).

To evaluate the clinical significance of GdECs, we correlated the GdEC signatures with OS in TCGA databases. We determined that the GdEC signatures negatively correlated with OS in both TCGA-GBM and TCGA-LGG (Fig. 1C). This suggests that GdECs hamper immune effector cell infiltration and contribute to the poor prognosis of patients with glioma.

Drug screening identifies cAMP activators as the optimal blocker of GdEC differentiation

The above bioinformatics analysis prompted us to explore the immunoregulatory and therapeutic effects of GdEC blockade. We first established an in vitro model of GdEC differentiation using human GSCs as reported previously (21, 23). We found that human GSCs could form tubular structures on Matrigel (Supplementary Fig. S2A) and the expression of EC markers (CD31, CD105, CD34) showed a time-dependent increase upon culture in ECM (Supplementary Fig. S2B–S2D), which validated the capacity of GSCs to differentiate into GdECs in vitro. However, neither anti-VEGF treatment nor VEGF cytokine supplement affected GdEC formation (Supplementary Fig. S3A). In subcutaneous GSC1 xenografts, immunofluorescence analysis with antibodies specific for either human or mouse CD31 revealed that anti-VEGF treatment had no inhibitory effect on ECs expressing human-CD31 (hCD31, an indicator of GSC-derived vessels), but significantly reduced the number of mouse-CD31+ (mCD31, an indicator of host-derived vessels) ECs (Supplementary Fig. S3B–S3D). These data indicated that the differentiation of GSCs to GdECs could not be inhibited by anti-VEGF treatment.

To identify potential strategies for blocking GdEC differentiation, we performed a dual-step drug screening using 40 anti-GSC agents that target proteins involved in self-renewal, cell metabolism, or epigenetic regulation (Supplementary Table S2). First, we used a CCK-8 assay to measure the dose–response curves of drugs and calculate the IC10 in GSC11 cells (Supplementary Table S2; Fig. 2A). Second, we performed a tube formation assay to test the endothelial lineage differentiation capacity of GSCs in the presence of the different compounds. Through this drug screening, agents were ranked according to the inhibitory effect on tube length (Fig. 2B and C). Two cAMP activators, dbcAMP and forskolin, showed optimal inhibitory effects in our screening (Fig. 2B and C). We found that dbcAMP and forskolin also significantly decreased the tube length in other GSC cultures, including GSC1, GSC21, and GSC24 (Fig. 2D). However, cAMP activators had no inhibitory effect on tube formation by HUVECs (Fig. 2D). These data suggested that VEGF-independent GdEC differentiation could be blocked by cAMP activators in vitro.

Figure 2.

Drug screening identifies cAMP activators as the optimal GdEC-differentiation blockers. A, The outline of the drug-screening protocol. In the first screen, GSC11 cells were seeded on 24-well plates and treated with drugs for 48 hours, then IC10 (10% inhibiting concentration) was determined via CCK-8 assay. The IC10 was selected as the working concentration for the second screening. In the second screen, a tube formation assay was performed to detect the GdECs differentiation inhibition rate. B, The representative images of the tube formation assay in GSC11 cells treated with different 40 drugs for 48 hours. C, The quantitation of tube length in B (n = 3). D, The tube formation assays in different GSCs (GSC1, GSC21, and GSC24) and HUVECs were treated with 1 mmol/L dbcAMP, or 50 μmol/L forskolin for 48 hours. The quantitation of tube length was shown in the right (n = 3). Data represent the mean ± SD. N.S., not significant; *, P < 0.05; **, P < 0.01; ***, P < 0.001; different from the control group. P values are determined by one-way ANOVA in C and D. Scale bar, 100 μm in B and D.

Figure 2.

Drug screening identifies cAMP activators as the optimal GdEC-differentiation blockers. A, The outline of the drug-screening protocol. In the first screen, GSC11 cells were seeded on 24-well plates and treated with drugs for 48 hours, then IC10 (10% inhibiting concentration) was determined via CCK-8 assay. The IC10 was selected as the working concentration for the second screening. In the second screen, a tube formation assay was performed to detect the GdECs differentiation inhibition rate. B, The representative images of the tube formation assay in GSC11 cells treated with different 40 drugs for 48 hours. C, The quantitation of tube length in B (n = 3). D, The tube formation assays in different GSCs (GSC1, GSC21, and GSC24) and HUVECs were treated with 1 mmol/L dbcAMP, or 50 μmol/L forskolin for 48 hours. The quantitation of tube length was shown in the right (n = 3). Data represent the mean ± SD. N.S., not significant; *, P < 0.05; **, P < 0.01; ***, P < 0.001; different from the control group. P values are determined by one-way ANOVA in C and D. Scale bar, 100 μm in B and D.

Close modal

cAMP activators block GdEC differentiation via inducing oxidative stress

We investigated the effect of dbcAMP treatment on global gene expression patterns of GSC11 using RNA-seq. Gene Ontology (GO) analysis showed that oxidative phosphorylation was strikingly upregulated in dbcAMP-treated GSC11 (Fig. 3A). GSEA also revealed that dbcAMP treatment enhanced oxidative phosphorylation (Fig. 3B). We evaluated cell metabolism by measuring the OCR and ECAR, finding that dbcAMP treatment markedly elevated the basal OCR and maximal OCR of GSC11 (Fig. 3C), but it did not influence the ECAR (Fig. 3D). In addition, dbcAMP treatment also significantly enhanced the proton leak level of OCR (Fig. 3C), indicating enhancement of mitochondria-derived ROS (mtROS). The MitoSOX staining assay confirmed that dbcAMP treatment could increase mtROS in a time-dependent manner (Fig. 3E).

Figure 3.

cAMP activators inhibit GdEC differentiation by inducing excessive mtROS and oxidative stress. A, GO analysis of transcriptome data was performed by comparing the dbcAMP treatment group with the control group in GSC11 cells. The top 10 enriched biological process pathways were shown. B, GSEA of the gene set of KEGG oxidative phosphorylation. The plot showed enrichment of gene signatures associated with oxidative phosphorylation in the dbcAMP treatment group. FDR, false discovery rate; NES, normalized enrichment score. KEGG, Kyoto Encylopedia of Genes and Genomes. C and D, The measurements of OCR and ECAR in GSC11 cells. Cells were treated with dbcAMP for the indicated time, and then the OCR (C) and the ECAR (D) were monitored by using the seahorse bioscience extracellular flux analyzer in real time. The dotted lines indicate the incubation of the indicated compounds. The basal, maximal, and proton leak of OCR and the maximal ECAR were calculated (n = 4). E and F, The level of mitochondrial ROS was detected by mitoSOX Red staining. E, GSC11 cells were treated with 1 mmol/L dbcAMP for the indicated time (n = 3). F, GSC11 cells were treated with 1 mmol/L dbcAMP with or without 1 mmol/L NAC or 80 μmol/L MnTBAP for 48 hours (n = 3). Then the cells were stained with mitoSOX Red for flow cytometry analysis. The representative plot was shown in the left; the quantitation of the relative mean fluorescence intensity (MFI) of mitoSOX Red was shown in the right. G, Tube formation of GSC11 cells treated with 1 mmol/L dbcAMP with or without 1 mmol/L NAC or 80 μmol/L MnTBAP. The quantitation of tube length was shown in the right (n = 3). H, Cell proliferation was measured by the EdU-labeling assay. GSC11 cells were treated with 1 mmol/L dbcAMP for the indicated time. The percentage of proliferating EdU-positive cells was determined by flow cytometry. The quantitation of the EdU-positive cell was shown in the right (n = 3). I, Quantification of tumorspheres in GSC11 treated with 1 mmol/L dbcAMP for the indicated time (n = 5). J, The expression of CD133 was detected by flow cytometry. GSC11 cells were treated with 1 mmol/L dbcAMP for the indicated time (n = 3). Then the cells were stained with fluorochrome-conjugated anti-CD133 antibody for flow cytometry analysis. The representative plot was shown in the left; the quantitation of the relative MFI of CD133 was shown in the right. Data represent the mean ± SD. N.S., not significant; *, P < 0.05; **, P < 0.01; ***, P < 0.001. P values are determined by one-way ANOVA (C–J). Scale bar, 100 μm in G.

Figure 3.

cAMP activators inhibit GdEC differentiation by inducing excessive mtROS and oxidative stress. A, GO analysis of transcriptome data was performed by comparing the dbcAMP treatment group with the control group in GSC11 cells. The top 10 enriched biological process pathways were shown. B, GSEA of the gene set of KEGG oxidative phosphorylation. The plot showed enrichment of gene signatures associated with oxidative phosphorylation in the dbcAMP treatment group. FDR, false discovery rate; NES, normalized enrichment score. KEGG, Kyoto Encylopedia of Genes and Genomes. C and D, The measurements of OCR and ECAR in GSC11 cells. Cells were treated with dbcAMP for the indicated time, and then the OCR (C) and the ECAR (D) were monitored by using the seahorse bioscience extracellular flux analyzer in real time. The dotted lines indicate the incubation of the indicated compounds. The basal, maximal, and proton leak of OCR and the maximal ECAR were calculated (n = 4). E and F, The level of mitochondrial ROS was detected by mitoSOX Red staining. E, GSC11 cells were treated with 1 mmol/L dbcAMP for the indicated time (n = 3). F, GSC11 cells were treated with 1 mmol/L dbcAMP with or without 1 mmol/L NAC or 80 μmol/L MnTBAP for 48 hours (n = 3). Then the cells were stained with mitoSOX Red for flow cytometry analysis. The representative plot was shown in the left; the quantitation of the relative mean fluorescence intensity (MFI) of mitoSOX Red was shown in the right. G, Tube formation of GSC11 cells treated with 1 mmol/L dbcAMP with or without 1 mmol/L NAC or 80 μmol/L MnTBAP. The quantitation of tube length was shown in the right (n = 3). H, Cell proliferation was measured by the EdU-labeling assay. GSC11 cells were treated with 1 mmol/L dbcAMP for the indicated time. The percentage of proliferating EdU-positive cells was determined by flow cytometry. The quantitation of the EdU-positive cell was shown in the right (n = 3). I, Quantification of tumorspheres in GSC11 treated with 1 mmol/L dbcAMP for the indicated time (n = 5). J, The expression of CD133 was detected by flow cytometry. GSC11 cells were treated with 1 mmol/L dbcAMP for the indicated time (n = 3). Then the cells were stained with fluorochrome-conjugated anti-CD133 antibody for flow cytometry analysis. The representative plot was shown in the left; the quantitation of the relative MFI of CD133 was shown in the right. Data represent the mean ± SD. N.S., not significant; *, P < 0.05; **, P < 0.01; ***, P < 0.001. P values are determined by one-way ANOVA (C–J). Scale bar, 100 μm in G.

Close modal

We then tested whether scavenging mtROS could rescue the inhibitory effect of dbcAMP treatment on GdEC formation from GSC11. We found that antioxidant N-acetyl cysteine and MnTBAP (a superoxide scavenger) treatment significantly reduced mtROS accumulation (Fig. 3F) and rescued the tube formation upon dbcAMP treatment (Fig. 3G). Notably, the “cell-cycle pathway” and “p53 pathway” were also enriched in the treated group compared with the control group by GO analysis (Fig. 3A). Flow cytometry analysis confirmed that EdU+ proliferating cells decreased in number under the dbcAMP treatment (Fig. 3H). We observed that dbcAMP also inhibited the sphere-forming ability and CD133 expression level of GSC11 (Fig. 3I and J), indicating that dbcAMP treatment also impaired the stemness of GSCs. Collectively, these data demonstrated that the cAMP activator blocked GdEC differentiation via induction of mtROS.

A cAMP activator blocks GdEC differentiation and normalizes the tumor vessels in a mouse GBM model

Although dbcAMP and forskolin are widely used in research to increase intracellular cAMP levels, they are not clinically approved drugs. To determine the therapeutic value of GdEC-differentiation blockade, we utilized a clinically approved cAMP activator, ibudilast, as the therapeutic agent in vivo (31). Ibudilast is a phosphodiesterase inhibitor and can cross the blood–brain barrier (31–33). We confirmed that ibudilast treatment could significantly increase intracellular cAMP levels in GSCs in vitro (Fig. 4A) and inhibited tube formation by GSCs, without obvious cytotoxicity (Fig. 4B; Supplementary Fig. S4A). In vitro limiting dilution assays revealed that ibudilast treatment also decreased sphere formation by GSCs (Supplementary Fig. S4B).

Figure 4.

The cAMP activator ibudilast inhibits GdEC formation and induces normalization of tumor vessels in vivo. A, GSC1 and GSC11 cells were treated with ibudilast (10 μmol/L) for 2 hours. Intracellular cAMP levels were then detected using ELISA Kit (n = 3). B, The tube formation assays in different GSCs (GSC1, GSC11, GSC21, and GSC24) and HUVECs were treated with 10 μmol/L ibudilast for 48 hours. The quantitation of tube length was shown in the right (n = 3). Scale bar, 100 μm. C–F, The therapeutic effect of cAMP activator ibudilast in nude mice bearing GSC1 intracranial xenograft. C, Diagram depicting the drug treatment schedule. D, Tumor growth was monitored via bioluminescence images of luciferase activity on day 7 (before the drug treatment) and day 16 (after the drug treatment). The quantitative analysis of bioluminescence images on day 16 was shown in the right (n = 7). E, H&E staining of mouse brain sections. Scale bar, 1 cm. F, Kaplan–Meier survival curve of mice. G–I, Tumor tissues from mice bearing GSC-1 tumor in C were evaluated through IHC analysis for human-specific CD31 (hCD31, an indicator of GdEC; G), ki67 (a marker of proliferation; H) and MDA (a marker of oxidative stress; I). The representative images were shown in the left; the quantitation of IHC staining was shown in the right (n = 3 tumors, five fields per tumor). Scale bar, 100 μm. J–L, The vascular pathology of GSC1 tumors from C was evaluated through immunofluorescence analysis for hypoxia (J), αSMA+ pericyte coverage (K), and dextran leakage (L). The representative images were shown in the left; the quantitation of immunofluorescence staining was shown in the right (n = 3 tumors, five fields per tumor). Scale bar, 100 μm. Data represent the mean ± SD. N.S., not significant; *, P < 0.05; **, P < 0.01; ***, P < 0.001. P values are determined by two-tailed unpaired Student t test (AC, G–L); log-rank test (F). DAPI, 4′,6-diamidino-2-phenylindole.

Figure 4.

The cAMP activator ibudilast inhibits GdEC formation and induces normalization of tumor vessels in vivo. A, GSC1 and GSC11 cells were treated with ibudilast (10 μmol/L) for 2 hours. Intracellular cAMP levels were then detected using ELISA Kit (n = 3). B, The tube formation assays in different GSCs (GSC1, GSC11, GSC21, and GSC24) and HUVECs were treated with 10 μmol/L ibudilast for 48 hours. The quantitation of tube length was shown in the right (n = 3). Scale bar, 100 μm. C–F, The therapeutic effect of cAMP activator ibudilast in nude mice bearing GSC1 intracranial xenograft. C, Diagram depicting the drug treatment schedule. D, Tumor growth was monitored via bioluminescence images of luciferase activity on day 7 (before the drug treatment) and day 16 (after the drug treatment). The quantitative analysis of bioluminescence images on day 16 was shown in the right (n = 7). E, H&E staining of mouse brain sections. Scale bar, 1 cm. F, Kaplan–Meier survival curve of mice. G–I, Tumor tissues from mice bearing GSC-1 tumor in C were evaluated through IHC analysis for human-specific CD31 (hCD31, an indicator of GdEC; G), ki67 (a marker of proliferation; H) and MDA (a marker of oxidative stress; I). The representative images were shown in the left; the quantitation of IHC staining was shown in the right (n = 3 tumors, five fields per tumor). Scale bar, 100 μm. J–L, The vascular pathology of GSC1 tumors from C was evaluated through immunofluorescence analysis for hypoxia (J), αSMA+ pericyte coverage (K), and dextran leakage (L). The representative images were shown in the left; the quantitation of immunofluorescence staining was shown in the right (n = 3 tumors, five fields per tumor). Scale bar, 100 μm. Data represent the mean ± SD. N.S., not significant; *, P < 0.05; **, P < 0.01; ***, P < 0.001. P values are determined by two-tailed unpaired Student t test (AC, G–L); log-rank test (F). DAPI, 4′,6-diamidino-2-phenylindole.

Close modal

We next developed an orthotopic GBM model by transplanting GSC1 cells in nude mice (Fig. 4C). Bioluminescent imaging of orthotopic xenografts demonstrated that ibudilast treatment reduced tumor growth (Fig. 4D and E). In addition, Kaplan–Meier analysis showed that mice treated with ibudilast had a significantly longer survival time than controls (Fig. 4F). At the end of the experiment, IHC staining revealed that ibudilast treatment significantly decreased hCD31+ vessel density (Fig. 4G; Supplementary Fig. S5A) but did not influence mCD31+ vessel density (Supplementary Fig. S4C), and it reduced the percentage of proliferative Ki67+ cells in the brain tumor tissue (Fig. 4H). The measurement of peroxidized lipids by MDA antibody revealed that ibudilast treatment also induced oxidative stress in vivo (Fig. 4I).

We further investigated the effect of GdEC-differentiation blockade on tumor vessel pathology. Ibudilast treatment led to a marked reduction in intratumoral hypoxia (Fig. 4J), a significant increase in α-SMA+ pericytes coverage on tumor vessels (Fig. 4K), and a reduction in vascular leakage of FITC-dextran (Fig. 4L). These data suggested that treatment with the cAMP activator ibudilast decreased GdEC differentiation and normalized the tumor vessels in the mouse GBM model.

GdEC-differentiation blockade remodels the tumor immune profile and enhances effector T-cell infiltration

To explore the immunoregulatory effect of GdEC-differentiation blockade, we designed an in vivo cell lineage tracing experiment to monitor GdEC formation by introducing GFP into the GL261 mouse GBM cell line (GL261-GFP) and developed orthotopic tumors in immune-competent C57BL/6 mice (Fig. 5A). Flow cytometry analysis revealed that a small fraction of GFP+ cells expressed the EC markers CD31, CD34, and CD105 (Fig. 5B; Supplementary Fig. S6A), and immunofluorescent staining revealed that partial vessels in the tumor area were GFP+CD31+ double-positive vessels (Supplementary Fig. S7A), indicating that GdEC differentiation occurred in this mouse GBM model. Ibudilast treatment significantly reduced the ratio of GFP+/CD31+, GFP+/CD34+, and GFP+/CD105+ cells among GFP+ cells (Fig. 5B), suggesting a decrease in GdEC formation.

Figure 5.

cAMP activator ibudilast blocks GdEC formation and reshapes the immune profile in the mouse GBM model. A and B,In vivo cell lineage tracing assay was performed to monitor the effect of ibudilast treatment on the GdECs formation in C57BL/6 mice bearing GL261-GFP tumors. A, Diagram depicting the drug treatment schedule. GL261-GFP cells were orthotopically injected to develop the GBM model. B, Flow cytometry analysis of the expression of endothelial markers in GFP-labeled cells. After the ibudilast treatment, tumors were harvested and dissociated. Then cells were stained with fluorochrome-conjugated anti-CD31, anti-CD34, and anti-CD105 antibodies. The ratio of GFP+/CD31+, GFP+/CD34+ and GFP+/CD105+ were analyzed (n = 4). C–F, The tumor-infiltrating immune cells were isolated from C57BL/6 mice bearing GL261 tumor and obtained using CD45 magnetic cell separation, and were subjected to scRNA-seq. C, t-SNE view of 10,972 CD45+ cells (including 6,205 cells from the control group and 4,767 cells from the ibudilast-treated group). The cell clusters were colored (left) and annotated by cell type (right). D, Marker gene expression for each cell type, where dot size and color represented the percentage of marker gene expression and the averaged expression value, respectively. E, The bar chart showed the sample origin of each cluster. F, The pie chart showed the cell type composition of each sample. Data represent the mean ± SD. *, P < 0.05; **, P < 0.01. P values are determined by two-tailed unpaired Student t test (B). DC, dendritic cell; NK, natural killer.

Figure 5.

cAMP activator ibudilast blocks GdEC formation and reshapes the immune profile in the mouse GBM model. A and B,In vivo cell lineage tracing assay was performed to monitor the effect of ibudilast treatment on the GdECs formation in C57BL/6 mice bearing GL261-GFP tumors. A, Diagram depicting the drug treatment schedule. GL261-GFP cells were orthotopically injected to develop the GBM model. B, Flow cytometry analysis of the expression of endothelial markers in GFP-labeled cells. After the ibudilast treatment, tumors were harvested and dissociated. Then cells were stained with fluorochrome-conjugated anti-CD31, anti-CD34, and anti-CD105 antibodies. The ratio of GFP+/CD31+, GFP+/CD34+ and GFP+/CD105+ were analyzed (n = 4). C–F, The tumor-infiltrating immune cells were isolated from C57BL/6 mice bearing GL261 tumor and obtained using CD45 magnetic cell separation, and were subjected to scRNA-seq. C, t-SNE view of 10,972 CD45+ cells (including 6,205 cells from the control group and 4,767 cells from the ibudilast-treated group). The cell clusters were colored (left) and annotated by cell type (right). D, Marker gene expression for each cell type, where dot size and color represented the percentage of marker gene expression and the averaged expression value, respectively. E, The bar chart showed the sample origin of each cluster. F, The pie chart showed the cell type composition of each sample. Data represent the mean ± SD. *, P < 0.05; **, P < 0.01. P values are determined by two-tailed unpaired Student t test (B). DC, dendritic cell; NK, natural killer.

Close modal

Next, we evaluated the immune profile of intracranial GL261 xenografts via scRNA-seq of CD45+ cells (Supplementary Fig. S8A). After standard data processing and quality control procedures, we obtained transcriptomic profiles for 10,972 cells: 6,205 cells from control tumors and 4,767 cells from ibudilast-treated tumors (Supplementary Fig. S8A).

We identified 17 clusters from the scRNA-seq profiles using t-SNE visualization and classified them into T cells, B cells, natural killer cells, monocytes, mast cells, and dendritic cells based on the canonical markers for each cell population (Fig. 5C and D; Supplementary Fig. S8B). The sample origin of each cluster substantially differed (Fig. 5E). The monocyte clusters (6, 1, 4, 11, 10, 3, and 0 subclusters) mainly originated from the control group; however, the T-cell clusters (12, 7, and 13 subclusters) largely originated from the ibudilast-treated group (Fig. 5E), indicating that GdEC-differentiation blockade could reshape the TME of GBM. By analyzing the cell composition of the samples, we found that monocytes were the predominant populations in both groups (Fig. 5F). An elevated influx of T cells was observed in the ibudilast-treated group, compared with the control group (Fig. 5F).

T-cell and monocyte populations were reanalyzed using t-SNE, yielding eight subclusters in T cells and 12 subclusters in the monocytes (Fig. 6A). Within the T-cell population, subclusters 4 and 6 expressed genes encoding markers of actively dividing cells (ki67 and top2a). Subclusters 1 and 0 contained markers of effector T cells (ifng, gzmb, and prf1). Subclusters 5 and 7 expressed genes s associated with naïve T cells (tcf7 and lef1). Several transcripts upregulated in subcluster 3 (foxp3 and il2ra) were associated with regulatory T cells (Treg) and markers in subcluster 2 (bhlhe40 and icos) were associated with Th1 cells (Supplementary Fig. S9A). Ibudilast treatment increased the percentage of CD8+ effector T cells, from 40.0% to 50.2% (Fig. 6B). Furthermore, we identified two cell types within the monocytic clusters: microglia (tmem119, cx3cr1, and p2ry12) and monocyte-derived macrophages (ccr2, s100a6, itga4, and ly6c2), as reported previously (refs. 34, 35; Fig. 6A; Supplementary Fig. S9B). We found that ibudilast treatment decreased the frequency of microglia, but enhanced the presence of monocyte-derived macrophages (Fig. 6C).

Figure 6.

GdEC-differentiation blockade by a cAMP activator enhances the infiltration of CD8+ effector T cells. A, t-SNE view of the T-cell population (left) and monocyte population (right), color-coded by re-evaluated clusters. Recluster analysis of the T-cell population yielded cycling T cells, CD8+ effector T cells, CD4+ Th cells (Th1 cells), CD4+ Tregs, and CD8+ naïve subclusters. Recluster analysis of the monocyte population yielded microglia and monocyte-derived macrophage subclusters. B, The pie chart showed the T-cell composition of each sample. C, The bar chart showed the monocytes composition of each sample.

Figure 6.

GdEC-differentiation blockade by a cAMP activator enhances the infiltration of CD8+ effector T cells. A, t-SNE view of the T-cell population (left) and monocyte population (right), color-coded by re-evaluated clusters. Recluster analysis of the T-cell population yielded cycling T cells, CD8+ effector T cells, CD4+ Th cells (Th1 cells), CD4+ Tregs, and CD8+ naïve subclusters. Recluster analysis of the monocyte population yielded microglia and monocyte-derived macrophage subclusters. B, The pie chart showed the T-cell composition of each sample. C, The bar chart showed the monocytes composition of each sample.

Close modal

Our data supported that GdEC-differentiation blockade remodeled the tumor immune microenvironment in GBM by reshaping the composition of tumor-infiltrating T cells and enhancing the influx of CD8+ effector T cells.

Dual blockade of GdEC differentiation and PD-1 induces tumor regression and specific antitumor immunity

The above results revealed that GdEC-differentiation blockade enhanced the infiltration of effector T cells. Therefore, we reasoned that dual blockade of GdEC differentiation and PD-1 might augment the overall antitumor immune response and sought to evaluate the combinational effects. In the orthotopic GBM model with GL261 cells, Kaplan–Meier analysis showed that combination treatment of ibudilast and anti–PD-1 significantly extended the survival time of mice, compared with single treatment groups or control group (Fig. 7A and B). The combination treatment also induced tumor regression in 25% (5/20) of mice and resulted in several long-term survivors (Fig. 7B and C). In addition, there was no observable change in body weight (Supplementary Fig. S10A). No signs of vital tissue toxicity were observed in the combination treatment group (Supplementary Fig. S10B).

Figure 7.

Dual blockade of GdEC differentiation and PD-1 induces tumor regression, forms tumor-specific antitumor immunity, and reshapes the tumor immune microenvironment. A, Diagram depicting ibudilast and anti–PD-1 treatment schedule in the mice bearing GL261 intracranial tumors. B, Kaplan–Meier survival curve of mice bearing GL261 intracranial tumors (n =  7 in control, ibudilast, and anti–PD-1 group; n = 20 in combination group). Animal survival was monitored until day 80. C, The H&E staining of the mice brain section bearing GL261 tumors. Scale bar, 1 cm. D, Surviving GL261 mice in the combination group were rechallenged on day 80 with a 2-fold increase in the number of homologous GL261 cells (6 × 105) or heterologous B16F10 cells (6 × 105) in the hemisphere contralateral to the primary injected tumor and survival time was monitored (n = 2 or 3). Naïve mice of similar age (4 months) were implanted with GL261 or B16F10 cells as controls (n = 3). E, Detection of in vivo circulating tumor antibodies against GL261 cells detected by probing GL261 lysates with serum from the mice described in A. Each lane represents serum from a single mouse. F, Diagram depicting ibudilast and anti–PD-1 treatment schedule in the mice bearing CT2A intracranial tumors. G, Kaplan–Meier survival curve of mice bearing CT2A intracranial tumors (n =  8 in each group). Animal survival was monitored until day 60. H, The H&E staining of the mice brain section bearing CT2A tumors. Scale bar, 1 cm. I and J, Flow cytometry analysis of immune cell infiltration and function in GL261 intracranial xenograft. I, Percentages of CD4+ T cells, CD8+ T cells among CD3+ cells, and Tregs (Foxp3+) among CD4+ T cells were analyzed. J, The function of CD8+ T cells was evaluated by measuring GranzB expression. K–M, Immune cell depletion experiments. C57BL/6 mice were injected with either anti-mouse CD8, anti-mouse CD4, or isotype antibody for five doses (10 mg/kg). K–L, Validation of immune cell depletion. M, Kaplan–Meier survival curve of the mice bearing GL261 tumors in immune cells depletion experiments. The mice were treated with a combination treatment of ibudilast and anti–PD-1 (n =  5). N and O, Flow cytometry analysis of immune cell infiltration and function in CT2A intracranial xenografts. N, Percentages of CD4+ T cells, CD8+ T cells among CD3+ cells, and Tregs (Foxp3+) among CD4+ T cells were analyzed. O, The function of CD8+ T cells was evaluated by measuring granzyme B expression. Data represent the mean ± SD. N.S., not significant; *, P < 0.05; **, P < 0.01; ***, P < 0.001. P values are determined by log-rank analysis (B, D, G, and M); one-way ANOVA (I–J and N-O); by two-tailed unpaired Student t test (L).

Figure 7.

Dual blockade of GdEC differentiation and PD-1 induces tumor regression, forms tumor-specific antitumor immunity, and reshapes the tumor immune microenvironment. A, Diagram depicting ibudilast and anti–PD-1 treatment schedule in the mice bearing GL261 intracranial tumors. B, Kaplan–Meier survival curve of mice bearing GL261 intracranial tumors (n =  7 in control, ibudilast, and anti–PD-1 group; n = 20 in combination group). Animal survival was monitored until day 80. C, The H&E staining of the mice brain section bearing GL261 tumors. Scale bar, 1 cm. D, Surviving GL261 mice in the combination group were rechallenged on day 80 with a 2-fold increase in the number of homologous GL261 cells (6 × 105) or heterologous B16F10 cells (6 × 105) in the hemisphere contralateral to the primary injected tumor and survival time was monitored (n = 2 or 3). Naïve mice of similar age (4 months) were implanted with GL261 or B16F10 cells as controls (n = 3). E, Detection of in vivo circulating tumor antibodies against GL261 cells detected by probing GL261 lysates with serum from the mice described in A. Each lane represents serum from a single mouse. F, Diagram depicting ibudilast and anti–PD-1 treatment schedule in the mice bearing CT2A intracranial tumors. G, Kaplan–Meier survival curve of mice bearing CT2A intracranial tumors (n =  8 in each group). Animal survival was monitored until day 60. H, The H&E staining of the mice brain section bearing CT2A tumors. Scale bar, 1 cm. I and J, Flow cytometry analysis of immune cell infiltration and function in GL261 intracranial xenograft. I, Percentages of CD4+ T cells, CD8+ T cells among CD3+ cells, and Tregs (Foxp3+) among CD4+ T cells were analyzed. J, The function of CD8+ T cells was evaluated by measuring GranzB expression. K–M, Immune cell depletion experiments. C57BL/6 mice were injected with either anti-mouse CD8, anti-mouse CD4, or isotype antibody for five doses (10 mg/kg). K–L, Validation of immune cell depletion. M, Kaplan–Meier survival curve of the mice bearing GL261 tumors in immune cells depletion experiments. The mice were treated with a combination treatment of ibudilast and anti–PD-1 (n =  5). N and O, Flow cytometry analysis of immune cell infiltration and function in CT2A intracranial xenografts. N, Percentages of CD4+ T cells, CD8+ T cells among CD3+ cells, and Tregs (Foxp3+) among CD4+ T cells were analyzed. O, The function of CD8+ T cells was evaluated by measuring granzyme B expression. Data represent the mean ± SD. N.S., not significant; *, P < 0.05; **, P < 0.01; ***, P < 0.001. P values are determined by log-rank analysis (B, D, G, and M); one-way ANOVA (I–J and N-O); by two-tailed unpaired Student t test (L).

Close modal

To determine the durability of the observed therapeutic effect, the long-term survivors were rechallenged on day 80 with 2-fold autologous GL261 cells or with 2-fold heterologous B16F10 melanoma cells contralaterally to the primary tumor injection site (Fig. 7D). Mice rechallenged with GL261 cells did not succumb to the new tumor, while all the mice rechallenged with B16F10 cells and age-matched naïve mice succumbed to tumors by day 40 (Fig. 7D). Circulating antibodies against GL261 cells were detected at higher levels in mice that received the combination therapy, compared with those who received monotherapy or controls (Fig. 7E).

The CT2A orthotopic GBM model is a highly clinically relevant model (36, 37). We found that combination treatment with ibudilast and anti–PD-1 significantly lengthened the survival time of mice (Fig. 7F and G) and significantly inhibited tumor growth (Fig. 7H) in the CT2A model. These data demonstrated that the combination treatment of a cAMP activator and PD-1 blockade potentially eradicated orthotopic GBM in immunocompetent mice and established durable tumor-specific immunity.

Dual blockade of GdEC differentiation and PD-1 reshapes the TME and this is dependent on CD8+ T cells

To characterize TME reshaping following combination treatment with ibudilast and anti–PD-1, we analyzed tumor-infiltrating immune cells via flow cytometry (Supplementary Fig. S6B). In the GL261 murine model, combination treatment increased the proportion of tumor-infiltrating CD45+ cells, compared with the monotherapy groups (Supplementary Fig. S10C). The combination treatment also increased the number of CD8+ T cells (Teff) and lowered the quantity of tumor-infiltrating Tregs, which led to a marked increase in the ratio of Teffs to Tregs (Fig. 7I; Supplementary Fig. S11A). Combination treatment also significantly increased granzyme-B (GranzB) production in CD8+ T cells, indicating enhanced activity of CD8+ T cells (Fig. 7J; Supplementary Fig. S11B). Overall, these data demonstrate that combination therapy with GdEC-differentiation blockade and anti–PD-1 augments the number of functionally active CD8+ T cells within tumors.

We further investigated which immune cells mediated the antitumor activity through in vivo depletion experiments using antibodies against CD4 and CD8 (Supplementary Fig. S10D). The depletion antibodies effectively eliminated CD4+ and CD8+ T cells (Fig. 7K and L). Depletion studies demonstrated that tumor regression was mainly dependent on CD8+ T cells, and not on CD4+ T cells (Fig. 7M; Supplementary Fig. S10E).

In the CT2A murine model, we found that the combination treatment also elevated the number of Teffs and decreased the amount of Tregs, resulting in a rise in the ratio of Teffs to Tregs (Fig. 7N; Supplementary Fig. S11C). The GranzB production of CD8+ T cells was also significantly increased in the combination group (Fig. 7O; Supplementary Fig. S11D). These data verified that the combination of a cAMP activator and PD-1 blockade induced systematic changes in the immune microenvironment of GBM and the therapeutic effect mainly depended on CD8+ cells.

Ibudilast has been reported to block the MIF–CD74 interaction and reduce the function of myeloid-derived suppressor cells (MDSC) to enhance CD8+ T-cell activity (38). However, in our model, ibudilast (20 mg/kg) treatment did not affect the proportion of MDSCs (CD45+GR1+CD11b+; Supplementary Fig. S6C and S11E), suggesting ibudilast treatment increased CD8+ T cells by blocking GdEC differentiation, and not by influencing the MDSCs.

Although quite a few studies have demonstrated the existence of GdECs in gliomas (18–21, 23, 39), the interaction between GdECs and the immune microenvironment has not been studied yet. Here, we reveal that the endothelial transdifferentiation of GBM can shape an endothelial immune cell barrier—what we believe to be a novel mechanism of tumor immune escape. Through drug screening, we identified cAMP signaling activators as the optimal blockers of GdEC differentiation. Given that ibudilast is a clinically approved drug for treating asthma and stroke (40), our combination strategy of using ibudilast and PD-1 blockade has a clear route toward clinical translation.

Compelling evidence indicates that antiangiogenic therapy has the capacity to ameliorate antitumor immunity by inhibiting various immunosuppressive features of angiogenesis (41–43). Combinations of antiangiogenic agents and immunotherapy are currently being tested in over 90 clinical trials and five such combinations have been approved by the FDA in the past few years (12). The VEGF family represents one of the most well-validated signaling pathways in tumor angiogenesis. Several preclinical studies have shown the benefit of combining anti-VEGF agents with various immunotherapies, including ICB therapy and adoptively transferred engineered chimeric antigen receptor T cells (16, 17, 41, 44). Three recent phase III trials have shown the benefit of combining PD-1 or PD-L1 blockade with anti-VEGF or anti-VEGFR agents, leading to FDA approval for combination treatments in lung and kidney cancers (NCT02684006, NCT02853331, and NCT02366143; ref. 45).

Because VEGF signaling is also essential for normal tissue vascular angiogenesis, VEGF inhibitors in cancer therapies often result in hypertension and nephrotoxicity (46, 47). Thus, it would be beneficial to identify distinct pathways underlying tumor vascular abnormalities that are not essential for normal tissue vessel homeostasis. Mounting evidence shows that GSCs have the plasticity of transdifferentiating into different vascular components, including ECs (18–23) and pericytes (24–26). The role of GdECs was not involved and considered in the above preclinical studies and clinical trials because previous reports (21) along with our findings (Supplementary Fig. S3) have demonstrated that GdEC differentiation in tumor angiogenesis is VEGF independent. In this study, we presented data indicating that VEGF-independent endothelial plasticity of GBM is a therapeutic target for enhancing immunotherapy.

Elevated cAMP inhibits proliferation and sustained inhibition of cAMP production potentiates glioma growth (48–50). The clear correlation of low cAMP levels with enhanced brain tumorigenesis, brain tumor grade, and brain tumor growth has prompted efforts to develop cAMP-elevating approaches for brain tumor treatment. Accumulating evidence suggests that the reactivation of cAMP signaling or exposure of glioma cells to cAMP analogs can decrease the proliferation of glioma and inhibit the growth of xenografted brain tumors (27). In our study, we chose ibudilast as the cAMP activator for animal experiments because it is a small molecule for the treatment of asthma and stroke (40). Ibudilast inhibits several cyclic nucleotide phosphodiesterases, Toll-like receptors, and macrophage migration inhibitory factors, and can cross the blood–brain barrier, with possible salutary effects on the central nervous system (51). Ibudilast is also a MIF–CD74 interaction inhibitor, and 50 mg/kg ibudilast treatment reduces tumor infiltration by MDSCs (38). However, ibudilast treatment (20 mg/kg), in our model, did not significantly influence the infiltration of MDSCs. In this study, we revealed that ibudilast could inhibit GdEC differentiation and normalize the vascular microenvironment, leading to a decrease in intratumoral hypoxia and vessel leakage. These reports, along with our findings, support the idea of targeting cAMP signaling in treating GBM.

The contribution of ROS to cancer development has been somewhat controversial and is highly complex (52–55). Numerous lines of evidence support the concept that ROS contributes to cancer initiation and metastatic capacity (56, 57). Other studies also support the idea that tumor cells inherently carry a high burden of oxidative stress, suggesting that additional ROS burden or interference with these antioxidant pathways may selectively kill cancer cells (58–60). Here, our findings indicate that GSCs are a group of GBM cells that are extremely sensitive to oxidative stress. The ROS limitation pathway might be essential for maintaining the self-renewal capacity of GSCs.

Our study does have several limitations. We conducted our experiments with implanted GBM mouse models only. However, a spontaneous GBM mouse model would better translate to humans, and information regarding the TME of such a model would be valuable. In addition, our analyses of tumor vasculature were histologically at limited timepoints, which provides an incomplete depiction of tumor pathogenesis. This knowledge gap would benefit from live imaging to record tumor vascular dynamics in response to GdEC-differentiation blockade. To precisely analyze GdECs, scRNA-seq analyses of sorted CD31+ cells are required to further redefine GdEC scores. Finally, we propose that GdECs shape an endothelial–immune cell barrier, but the molecular mechanism needs clarity.

In summary, our study provides a better understanding of the biological and immunoregulatory functions of GdECs and supports the theory that GdEC-differentiation blockade by cAMP activators is a safe and effective approach to improve vascular functions and immunotherapy in GBM (Supplementary Fig. S12).

No disclosures were reported.

Z. Qin: Data curation, software, validation, investigation, visualization, methodology, writing–original draft, writing–review and editing. Y. Huang: Software, formal analysis, validation, investigation, visualization, writing–review and editing. Z. Li: Software, formal analysis, validation, investigation, visualization, writing–review and editing. G. Pan: Software, validation, visualization. L. Zheng: Software, validation, visualization. X. Xiao: Software, funding acquisition, validation. F. Wang: Software, validation, visualization. J. Chen: Software, supervision, validation. X. Chen: Validation, project administration. X. Lin: Resources, data curation, formal analysis, supervision, funding acquisition, project administration, writing–review and editing. K. Li: Project administration. G. Yan: Resources, data curation, formal analysis, supervision. H. Zhang: Resources, data curation, software, formal analysis, funding acquisition, validation, project administration, writing–review and editing. F. Xing: Conceptualization, resources, data curation, formal analysis, supervision, funding acquisition, validation, methodology, writing–original draft, project administration, writing–review and editing.

The work is supported by the National Natural Science Foundation of China (NSFC grant nos. 82173829 and 81803568, to F. Xing; grant no. 81972605, to H. Zhang), Guangdong Natural Science Funds of Distinguished Young Scholar (2021B1515020067, to H. Zhang), the Excellent Young Talent Program of Guangdong Provincial People's Hospital to F. Xing and Medical Scientific Research Foundation of Guangdong Province of China (A2022446, to X. Xiao.).

We thank Prof. Guangmei Yan (Sun Yat-sen University) for providing GSCs. The work is supported by the National Natural Science Foundation of China (NSFC grant nos. 82173829 and 81803568, to F. Xing; grant no. 81972605, to H. Zhang), Guangdong Natural Science Funds of Distinguished Young Scholar (2021B1515020067, to H. Zhang), the Excellent Young Talent Program of Guangdong Provincial People's Hospital to F. Xing and Medical Scientific Research Foundation of Guangdong Province of China (A2022446, to X. Xiao).

The publication costs of this article were defrayed in part by the payment of publication fees. Therefore, and solely to indicate this fact, this article is hereby marked “advertisement” in accordance with 18 USC section 1734.

Note: Supplementary data for this article are available at Cancer Immunology Research Online (http://cancerimmunolres.aacrjournals.org/).

1.
Shergalis
A
,
Bankhead
A
3rd
,
Luesakul
U
,
Muangsin
N
,
Neamati
N
.
Current challenges and opportunities in treating glioblastoma
.
Pharmacol Rev
2018
;
70
:
412
45
.
2.
Cloughesy
TF
,
Cavenee
WK
,
Mischel
PS
.
Glioblastoma: from molecular pathology to targeted treatment
.
Annu Rev Pathol
2014
;
9
:
1
25
.
3.
Wolchok
JD
,
Kluger
H
,
Callahan
MK
,
Postow
MA
,
Rizvi
NA
,
Lesokhin
AM
, et al
.
Nivolumab plus ipilimumab in advanced melanoma
.
N Engl J Med
2013
;
369
:
122
33
.
4.
Garon
EB
,
Rizvi
NA
,
Hui
R
,
Leighl
N
,
Balmanoukian
AS
,
Eder
JP
, et al
.
Pembrolizumab for the treatment of non-small-cell lung cancer
.
N Engl J Med
2015
;
372
:
2018
28
.
5.
Ansell
SM
,
Lesokhin
AM
,
Borrello
I
,
Halwani
A
,
Scott
EC
,
Gutierrez
M
, et al
.
PD-1 blockade with nivolumab in relapsed or refractory Hodgkin's lymphoma
.
N Engl J Med
2015
;
372
:
311
9
.
6.
Lim
M
,
Xia
Y
,
Bettegowda
C
,
Weller
M
.
Current state of immunotherapy for glioblastoma
.
Nat Rev Clin Oncol
2018
;
15
:
422
42
.
7.
Jackson
CM
,
Choi
J
,
Lim
M
.
Mechanisms of immunotherapy resistance: lessons from glioblastoma
.
Nat Immunol
2019
;
20
:
1100
9
.
8.
Daubon
T
,
Hemadou
A
,
Romero Garmendia
I
,
Saleh
M
.
Glioblastoma immune landscape and the potential of new immunotherapies
.
Front Immunol
2020
;
11
:
585616
.
9.
Wang
H
,
Xu
T
,
Huang
Q
,
Jin
W
,
Chen
J
.
Immunotherapy for malignant glioma: current status and future directions
.
Trends Pharmacol Sci
2020
;
41
:
123
38
.
10.
Datta
M
,
Coussens
LM
,
Nishikawa
H
,
Hodi
FS
,
J
RK
.
Reprogramming the tumor microenvironment to improve immunotherapy: emerging strategies and combination therapies
.
Am Soc Clin Oncol Educ Book
2019
;
39
:
165
74
.
11.
Andersen
BM
,
Faust Akl
C
,
Wheeler
MA
,
Chiocca
EA
,
Reardon
DA
,
Q
FJ
.
Glial and myeloid heterogeneity in the brain tumour microenvironment
.
Nat Rev Cancer
2021
;
21
:
786
802
.
12.
Huinen
ZR
,
Huijbers
EJM
,
van Beijnum
JR
,
Nowak-Sliwinska
P
,
Griffioen
AW
.
Anti-angiogenic agents - overcoming tumour endothelial cell anergy and improving immunotherapy outcomes
.
Nat Rev Clin Oncol
2021
;
18
:
527
40
.
13.
Ohm
JE
,
Gabrilovich
DI
,
Sempowski
GD
,
Kisseleva
E
,
Parman
KS
,
Nadaf
S
, et al
.
VEGF inhibits T-cell development and may contribute to tumor-induced immune suppression
.
Blood
2003
;
101
:
4878
86
.
14.
Voron
T
,
Colussi
O
,
Marcheteau
E
,
Pernot
S
,
Nizard
M
,
Pointet
AL
, et al
.
VEGF-A modulates expression of inhibitory checkpoints on CD8+ T cells in tumors
.
J Exp Med
2015
;
212
:
139
48
.
15.
Jain
RK
,
di Tomaso
E
,
Duda
DG
,
Loeffler
JS
,
Sorensen
AG
,
Batchelor
TT
.
Angiogenesis in brain tumours
.
Nat Rev Neurosci
2007
;
8
:
610
22
.
16.
Fukumura
D
,
Kloepper
J
,
Amoozgar
Z
,
Duda
DG
,
J
RK
.
Enhancing cancer immunotherapy using antiangiogenics: opportunities and challenges
.
Nat Rev Clin Oncol
2018
;
15
:
325
40
.
17.
Huang
Y
,
Kim
BYS
,
Chan
CK
,
Hahn
SM
,
Weissman
IL
,
Jiang
W
.
Improving immune-vascular crosstalk for cancer immunotherapy
.
Nat Rev Immunol
2018
;
18
:
195
203
.
18.
Carlson
JC
,
Cantu Gutierrez
M
,
Lozzi
B
,
Huang-Hobbs
E
,
Turner
WD
,
Tepe
B
, et al
.
Identification of diverse tumor endothelial cell populations in malignant glioma
.
Neuro Oncol
2021
;
23
:
932
44
.
19.
Ricci-Vitiani
L
,
Pallini
R
,
Biffoni
M
,
Todaro
M
,
Invernici
G
,
Cenci
T
, et al
.
Tumour vascularization via endothelial differentiation of glioblastoma stem-like cells
.
Nature
2010
;
468
:
824
8
.
20.
Wang
R
,
Chadalavada
K
,
Wilshire
J
,
Kowalik
U
,
Hovinga
KE
,
Geber
A
, et al
.
Glioblastoma stem-like cells give rise to tumour endothelium
.
Nature
2010
;
468
:
829
33
.
21.
Soda
Y
,
Marumoto
T
,
Friedmann-Morvinski
D
,
Soda
M
,
Liu
F
,
Michiue
H
, et al
.
Transdifferentiation of glioblastoma cells into vascular endothelial cells
.
Proc Natl Acad Sci U S A
2011
;
108
:
4274
80
.
22.
Mei
X
,
Chen
YS
,
Chen
FR
,
Xi
SY
,
C
ZP
.
Glioblastoma stem cell differentiation into endothelial cells evidenced through live-cell imaging
.
Neuro Oncol
2017
;
19
:
1109
18
.
23.
Hu
B
,
Wang
Q
,
Wang
YA
,
Hua
S
,
Sauvé
CG
,
Ong
D
, et al
.
Epigenetic activation of WNT5A drives glioblastoma stem cell differentiation and invasive growth
.
Cell
2016
;
167
:
1281
95
.
24.
Cheng
L
,
Huang
Z
,
Zhou
W
,
Wu
Q
,
Donnola
S
,
Liu
JK
, et al
.
Glioblastoma stem cells generate vascular pericytes to support vessel function and tumor growth
.
Cell
2013
;
153
:
139
52
.
25.
Zhou
W
,
Chen
C
,
Shi
Y
,
Wu
Q
,
Gimple
RC
,
Fang
X
, et al
.
Targeting glioma stem cell-derived pericytes disrupts the blood-tumor barrier and improves chemotherapeutic efficacy
.
Cell Stem Cell
2017
;
21
:
591
603
.
26.
Segura-Collar
B
,
Garranzo-Asensio
M
,
Herranz
B
,
Hernández-SanMiguel
E
,
Cejalvo
T
,
Casas
BS
, et al
.
Tumor-derived pericytes driven by EGFR mutations govern the vascular and immune microenvironment of gliomas
.
Cancer Res
2021
;
81
:
2142
56
.
27.
Xing
F
,
Luan
Y
,
Cai
J
,
Wu
S
,
Mai
J
,
Gu
J
, et al
.
The anti-warburg effect elicited by the cAMP-PGC1alpha pathway drives differentiation of glioblastoma cells into astrocytes
.
Cell Rep
2017
;
18
:
468
81
.
28.
Jensen
EC
.
Quantitative analysis of histological staining and fluorescence using imageJ
.
Anat Rec
2013
;
296
:
378
81
.
29.
McGrail
DJ
,
Garnett
J
,
Yin
J
,
Dai
H
,
Shih
DJH
,
Lam
TNA
, et al
.
Proteome instability is a therapeutic vulnerability in mismatch repair-deficient cancer
.
Cancer Cell
2020
;
37
:
371
86
.
30.
Cancer Genome Atlas Research Network
.
Comprehensive genomic characterization defines human glioblastoma genes and core pathways
.
Nature
2008
;
455
:
1061
8
.
31.
Schwenkgrub
J
,
Zaremba
M
,
Mirowska-Guzel
D
,
Kurkowska-Jastrzębska
I
.
Ibudilast: a non‑selective phosphodiesterase inhibitor in brain disorders
.
Postepy Hig Med Dosw
2017
;
71
:
137
48
.
32.
Wang
J
,
He
XL
,
Meng
HY
,
Li
Y
,
Dmitriev
P
,
Tian
F
, et al
.
Robust myelination of regenerated axons induced by combined manipulations of GPR17 and microglia
.
Neuron
2020
;
108
:
876
86
.
33.
Wakita
H
,
Tomimoto
H
,
Akiguchi
I
,
Lin
JX
,
Ihara
M
,
Ohtani
R
, et al
.
Ibudilast, a phosphodiesterase inhibitor, protects against white matter damage under chronic cerebral hypoperfusion in the rat
.
Brain Res
2003
;
992
:
53
9
.
34.
Friebel
E
,
Kapolou
K
,
Unger
S
,
Núñez
NG
,
Utz
S
,
Rushing
EJ
, et al
.
Single-cell mapping of human brain cancer reveals tumor-specific instruction of tissue-invading leukocytes
.
Cell
2020
;
181
:
1626
42
.
35.
Klemm
F
,
Maas
RR
,
Bowman
RL
,
Kornete
M
,
Soukup
K
,
Nassiri
S
, et al
.
Interrogation of the microenvironmental landscape in brain tumors reveals disease-specific alterations of immune cells
.
Cell
2020
;
181
:
1643
60
.
36.
Liu
CJ
,
Schaettler
M
,
Blaha
DT
,
Bowman-Kirigin
JA
,
Kobayashi
DK
,
Livingstone
AJ
, et al
.
Treatment of an aggressive orthotopic murine glioblastoma model with combination checkpoint blockade and a multivalent neoantigen vaccine
.
Neuro Oncol
2020
;
22
:
1276
88
.
37.
Tritz
ZP
,
Ayasoufi
K
,
J
AJ
.
Anti-PD-1 checkpoint blockade monotherapy in the orthotopic GL261 glioma model: the devil is in the detail
.
Neurooncol Adv
2021
;
3
:
vdab066
.
38.
Alban
TJ
,
Bayik
D
,
Otvos
B
,
Rabljenovic
A
,
Leng
L
,
Jia-Shiun
L
, et al
.
Glioblastoma myeloid-derived suppressor cell subsets express differential macrophage migration inhibitory factor receptor profiles that can be targeted to reduce immune suppression
.
Front Immunol
2020
;
11
:
1191
.
39.
Chen
T
,
Zhang
F
,
Liu
J
,
Huang
Z
,
Zheng
Y
,
Deng
S
, et al
.
Dual role of WNT5A in promoting endothelial differentiation of glioma stem cells and angiogenesis of glioma derived endothelial cells
.
Oncogene
2021
;
40
:
5081
94
.
40.
Oliveros
G
,
Wallace
CH
,
Chaudry
O
,
Liu
Q
,
Qiu
Y
,
Xie
L
, et al
.
Repurposing ibudilast to mitigate Alzheimer's disease by targeting inflammation
.
Brain
2022
;
146
:
898
911
.
41.
Khan
KA
,
Kerbel
RS
.
Improving immunotherapy outcomes with anti-angiogenic treatments and vice versa
.
Nat Rev Clin Oncol
2018
;
15
:
310
24
.
42.
Tian
L
,
Goldstein
A
,
Wang
H
,
Ching Lo
H
,
Sun Kim
I
,
Welte
T
, et al
.
Mutual regulation of tumour vessel normalization and immunostimulatory reprogramming
.
Nature
2017
;
544
:
250
4
.
43.
Huijbers
EJM
,
Khan
KA
,
Kerbel
RS
,
Griffioen
AW
.
Tumors resurrect an embryonic vascular program to escape immunity
.
Sci Immunol
2022
;
7
:
eabm6388
.
44.
Ma
W
,
Wang
Y
,
Zhang
R
,
Yang
F
,
Zhang
D
,
Huang
M
, et al
.
Targeting PAK4 to reprogram the vascular microenvironment and improve CAR-T immunotherapy for glioblastoma
.
Nat Cancer
2021
;
2
:
83
97
.
45.
Munn
LL
,
Jain
RK
.
Vascular regulation of antitumor immunity
.
Science
2019
;
365
:
544
5
.
46.
Ferrara
N
.
The role of VEGF in the regulation of physiological and pathological angiogenesis
.
EXS
2005
:
209
31
.
47.
Hurwitz
H
,
Fehrenbacher
L
,
Novotny
W
,
Cartwright
T
,
Hainsworth
J
,
Heim
W
, et al
.
Bevacizumab plus irinotecan, fluorouracil, and leucovorin for metastatic colorectal cancer
.
N Engl J Med
2004
;
350
:
2335
42
.
48.
Yang
L
,
Jackson
E
,
Woerner
BM
,
Perry
A
,
Piwnica-Worms
D
,
Rubin
JB
.
Blocking CXCR4-mediated cyclic AMP suppression inhibits brain tumor growth in vivo
.
Cancer Res
2007
;
67
:
651
8
.
49.
Goldhoff
P
,
Warrington
NM
,
Limbrick
DD
Jr
,
Hope
A
,
Woerner
BM
,
Jackson
E
, et al
.
Targeted inhibition of cyclic AMP phosphodiesterase-4 promotes brain tumor regression
.
Clin Cancer Res
2008
;
14
:
7717
25
.
50.
Warrington
NM
,
Gianino
SM
,
Jackson
E
,
Goldhoff
P
,
Garbow
JR
,
Piwnica-Worms
D
, et al
.
Cyclic AMP suppression is sufficient to induce gliomagenesis in a mouse model of neurofibromatosis-1
.
Cancer Res
2010
;
70
:
5717
27
.
51.
Fox
RJ
,
Coffey
CS
,
Conwit
R
,
Cudkowicz
ME
,
Gleason
T
,
Goodman
A
, et al
.
Phase 2 trial of ibudilast in progressive multiple sclerosis
.
N Engl J Med
2018
;
379
:
846
55
.
52.
Paul
MK
,
Bisht
B
,
Darmawan
DO
,
Chiou
R
,
Ha
VL
,
Wallace
WD
, et al
.
Dynamic changes in intracellular ROS levels regulate airway basal stem cell homeostasis through Nrf2-dependent Notch signaling
.
Cell Stem Cell
2014
;
15
:
199
214
.
53.
Assi
M
.
The differential role of reactive oxygen species in early and late stages of cancer
.
Am J Physiol Regul Integr Comp Physiol
2017
;
313
:
R646
53
.
54.
Cheung
EC
,
DeNicola
GM
,
Nixon
C
,
Blyth
K
,
Labuschagne
CF
,
Tuveson
DA
, et al
.
Dynamic ROS control by TIGAR regulates the initiation and progression of pancreatic cancer
.
Cancer Cell
2020
;
37
:
168
82
.
55.
Tafani
M
,
Sansone
L
,
Limana
F
,
Arcangeli
T
,
De Santis
E
,
Polese
M
, et al
.
The interplay of reactive oxygen species, hypoxia, inflammation, and sirtuins in cancer initiation and progression
.
Oxid Med Cell Longev
2016
;
2016
:
3907147
.
56.
Arnandis
T
,
Monteiro
P
,
Adams
SD
,
Bridgeman
VL
,
Rajeeve
V
,
Gadaleta
E
, et al
.
Oxidative stress in cells with extra centrosomes drives non-cell-autonomous invasion
.
Dev Cell
2018
;
47
:
409
24
.
57.
Mantovani
A
,
Cassatella
MA
,
Costantini
C
,
Jaillon
S
.
Neutrophils in the activation and regulation of innate and adaptive immunity
.
Nat Rev Immunol
2011
;
11
:
519
31
.
58.
Gorrini
C
,
Harris
IS
,
Mak
TW
.
Modulation of oxidative stress as an anticancer strategy
.
Nat Rev Drug Discov
2013
;
12
:
931
47
.
59.
Le Gal
K
,
Ibrahim
MX
,
Wiel
C
,
Sayin
VI
,
Akula
MK
,
Karlsson
C
, et al
.
Antioxidants can increase melanoma metastasis in mice
.
Sci Transl Med
2015
;
7
:
308re8
.
60.
Piskounova
E
,
Agathocleous
M
,
Murphy
MM
,
Hu
Z
,
Huddlestun
SE
,
Zhao
Z
, et al
.
Oxidative stress inhibits distant metastasis by human melanoma cells
.
Nature
2015
;
527
:
186
91
.