Glioblastoma (GBM) is a non-T-cell–inflamed cancer characterized by an immunosuppressive microenvironment that impedes dendritic cell maturation and T-cell cytotoxicity. Proangiogenic cytokines such as VEGF and angiopoietin-2 (Ang-2) have high expression in glioblastoma in a cell-specific manner and not only drive tumor angiogenesis and vascular permeability but also negatively regulate T-lymphocyte and innate immune cell responses. Consequently, the alleviation of immunosuppression might be a prerequisite for successful immune checkpoint therapy in GBM. We here combined antiangiogenic and immune checkpoint therapy and demonstrated improved therapeutic efficacy in syngeneic, orthotopic GBM models. We observed that blockade of VEGF, Ang-2, and programmed cell death protein-1 (PD-1) significantly extended survival compared with vascular targeting alone. In the GBM microenvironment, triple therapy increased the numbers of CTLs, which inversely correlated with myeloid-derived suppressor cells and regulatory T cells. Transcriptome analysis of GBM microvessels indicated a global vascular normalization that was highest after triple therapy. Our results propose a rationale to overcome tumor immunosuppression and the current limitations of VEGF monotherapy by integrating the synergistic effects of VEGF/Ang-2 and PD-1 blockade to reinforce antitumor immunity through a normalized vasculature.

The current standard therapy for GBM consisting of gross total surgery, followed by combined radiation and chemotherapy with temozolomide, results in overall survival of 15–25 months (1–3). The addition of antiangiogenic therapy with bevacizumab, a humanized monoclonal anti-VEGFA antibody, was not successful in first-line therapy despite advances in other cancer entities (4, 5). Hence, an urgent need exists for new treatment strategies. In this context, immune checkpoint therapy (ICT) offers new therapeutic avenues for the treatment of GBM (4, 6). Therapeutic targeting with immunomodulatory antibodies to enhance T-cell responses, such as antibodies to CTL-associated protein 4 (CTLA-4) and programmed cell death protein 1 (PD-1), have demonstrated a favorable outcome in various cancer entities, with the best efficacy in patients with advanced melanoma and lung cancer (4, 7). Whether ICT provides a treatment option for primary GBM is currently being investigated in clinical trials (NCT02617589, NCT02667587). In recurrent GBM, ICT has not met its primary endpoint compared to treatment with bevacizumab (NCT02017717; ref. 8). The unique central nervous system (CNS) microenvironment appears to be challenging for the efficacy of ICT in GBM (5, 9). GBM is a non-T-cell–inflamed cancer whereby immunosuppressive factors such as VEGF and TGFβ have been acknowledged to contribute to T-cell paucity (4, 6, 10). Studies have demonstrated that VEGF-targeting alleviates immunosuppression and is permissive for T-lymphocyte entry (4, 11, 12) and combined VEGF/angiopoietin-2 (Ang-2) therapy efficiently targets both the vasculature and immune cells in GBM models (13–15). VEGF and Ang-2 are known to act in concert to nourish tumor neovessel growth (16). Likewise, they operate as immunosuppressants and modulate immune cell recruitment with nonoverlapping downstream signaling pathways (9, 17). We hypothesized that alleviating this hostile microenvironment characterized by immunosuppression, abnormal vessel growth, hypoxia, and necrosis might improve immunotherapy. Motivated by the success of checkpoint inhibitors in melanoma, lung cancer (7), and rodent cancer models (18, 19), we analyzed tissue sections of patients with bevacizumab-treated GBM and identified PD-1/PD-L1 signaling as a promising target. We here identified antiangiogenic therapy as a treatment option to overcome resistance to ICT in GBM. ICT led to increased survival when combined with anti-VEGF/Ang-2 therapy by creating an immunostimulatory microenvironment. Bioinformatic analysis of glioma microvessels indicated that dual antiangiogenic therapy led to a global vascular normalization that was potentiated by anti–PD-1 therapy. We provide a rational combinatorial approach to improve the efficacy of ICT by integrating the synergistic effects of VEGF/Ang-2 and PD-1 blockade to reinforce antitumor immunity through a normalized vasculature.

Mice

Animal experiments were performed according to the principles of laboratory animal care and according to German national laws. The study was approved by the local ethics committee (Regierungspraesidium Darmstadt, FK/1031). In vivo experiments were performed with 8-week-old, female C57BL/6 or B6C3F1 mice that were obtained from Charles River.

Cell lines

GL261 (CVCL_Y003, gift from M. Machein, Department of Neurosurgery, Freiburg University Medical School, Freiburg, Germany), PD-L1–deficient (generated by CRISPR-Cas9, see below), and Tu-2449 glioma cells (CVCL_D318, generated by J. Weissenberger; ref. 20) were maintained in DMEM-GlutaMAX-I (Gibco) supplemented with 10% FBS (Sera Plus, Pan Biotech) and 1% penicillin and streptomycin (Sigma) at 5% CO2 at 37°C. Glioma cells were utilized one passage after thawing, authenticated microscopically and by their growth pattern. Cells were screened for Mycoplasma routinely (#A3744,0020 AppliChem).

Generation of PD-L1–deficient GL261 glioblastoma cells via CRISPR-Cas9

GL261 cells were cultivated in DMEM-GlutaMAX-I (Gibco) supplemented with 10% FBS (Sera Plus, Pan Biotech) and 1% penicillin and streptomycin (Sigma). Cells were passaged each 48 hours with trypsin (Sigma) and plated at 8 × 104 cells/cm2. Cells were cultivated at subconfluence and electroporated after cell dissociation at a cell concentration of 7 × 106 cells/mL using Bio-Rad Gene Pulser II (300 V/150 μF) with 10 μg of plasmid coding for wild-type SpCas9-guide RNA and mCherry. Cell sorting was conducted with BD FACSAria III sorter (BD Biosciences) 48 hours postelectroporation. mCherry-positive single cells were cultured to obtain pure clonal population. Clones #1 and #2 were amplified and isolated DNA was characterized by PCR and sequencing. PD-L1–deficient clones #1 and #2 were obtained following targeted hybridization of sgRNA#1: GACGTCAAGCTGCAGGACGC and sgRNA#2: GTATGGCAGCAACGTCACGA.

Immunocompetent intracranial GBM models and therapy

Intracerebral implantation of GL261, PD-L1–deficient GL261, and Tu-2449 glioblastoma cells was performed in 8-week-old female C57Bl/6 or B6C3F1 (Tu-2449) mice as described previously (13, 21). Mice were anesthetized by intraperitoneal injection with ketamine (100 mg/kg, Pfizer) and xylazine (10 mg/kg, Bayer) and fixed in a stereotactic device (Stoelting). A total of 1 × 105 GL261 glioma cells in 2 μL PBS were injected in the mouse striatum using a Hamilton syringe equipped with a 30G needle. Coordinates relative to the bregma were: 0.5 mm (anterior), 2 mm (mediolateral), and 3.5 mm (dorsoventral). The health status of tumor-bearing mice was checked on a daily routine. Mice were sacrificed when neurologic symptoms appeared (starting at 3 weeks after intracerebral implantation). Twenty percent loss of body weight or prolonged weight loss accompanied with reduced water/food uptake, or hunched back, rough coat, paresis, tremor, ataxia, and lethargia were considered as criteria for termination according to the Society of Laboratory Animals (GV-Solas; http://www.gv-solas.de). Five days postimplantation, antiangiogenic therapy with AMG386 (trebananib, Ang‐1/Ang‐2 peptibody, Amgen; ref. 22) and aflibercept (zaltrap, VEGF-‐trap, Sanofi; ref. 23) was initiated. AMG386 and aflibercept were administered subcutaneously (s.c.) 2×/week at doses of 5.6 mg/kg (22) and 25 mg/kg (23), respectively. Modes of action of AMG386 and aflibercept have been discussed in detail previously (13). Therapy was continued until mice became symptomatic (endpoint). For some experiments, animals were terminated 21 days post GL261 implantation. Monotherapy with anti–PD-1 (RPM1-14, BioXCell, 10 mg/kg) was started on day 10 by intraperitoneal injection (i.p.) 2×/week for a total of 8 injections. Untreated animals received an equal amount of rat IgG (clone 2A3, BioXCell, 10 mg/kg i.p.). T cells were depleted with anti-CD8 (clone 53-6.7 IgG2a; BioXCell, 10 mg/kg i.p.) on day 14 post intracranial implantation 2×/week as an intervention therapy. Tumor-bearing brains were removed and processed for further analyses as described below.

Lectin perfusion assay

For vascular perfusion studies, 100 μL DyLight 488-labeled Lycopersicon Escultentum lectin (1 mg/mL, Vector Laboratories) was intravenously injected 30 minutes prior to animal sacrifice. GBM-bearing mice were then anesthetized, perfused by intracardiac injection with 1% paraformaldehyde (PFA), and brains were subsequently harvested and processed for cryosectioning and immunofluorescence staining.

Immunofluorescence staining of mouse brain tumors

Whole brains were removed, embedded in Sakura Tissue-Tek O.C.T. (Thermo Fisher Scientific), and frozen on dry ice. Ten-micron–thick cyrosections were fixed in ice-cold 95% ethanol for 5 minutes and acetone at room temperature for 1 minute. Consecutive washing was carried out in PBSA solution (150 mmol/L NaCl, 10 mmol/L Na2HPO4, 10 mmol/L KH2PO4, 1% BSA, and 0.1% Triton X-100, pH 7.5). The following antibodies were applied for 1 hour at room temperature in antibody dilution buffer (0.5% BSA, 0.25% Triton X-100 in PBS, pH 7.2): rat anti-mouse CD31 (clone MEC 13.3, 1:300, BD Pharmingen), mouse monoclonal anti-desmin (clone D33, 1:300, DAKO), rat anti-mouse GLUT-1 (polyclonal, C-Terminus, 1:300, Millipore; Supplementary Table S1). After washing with PBSA, DyLight-labeled secondary antibodies (Thermo Fisher Scientific; Supplementary Table S2) were applied for 1 hour at room temperature. Slides were postfixed in 4% PFA, counterstained with 4,6-Diamidine-2´-phenylindole dihydrochloride (DAPI; Invitrogen), and embedded in Aqua PolyMount (Polysciences).

Immunofluorescence staining of murine vibratome sections

GL261-bearing mice were anesthetized and transcardially perfused with PBS (Gibco) for 4 minutes. Brains were dissected and postfixed overnight in 4% paraformaldehyde. Serial 50-μm coronal sections were cut on a Leica Vibratome (VT 1000S). The following primary antibodies were applied: mouse monoclonal anti-desmin (1:100, clone D33, DAKO), rabbit polyclonal anti-collagen IV (1:250, Biorbyt), rabbit polyclonal anti-CD8a (1:500, Synaptic Systems), and rat monoclonal anti-CD31 (1:40, clone SZ31, Dianova; Supplementary Table S1). Sections were incubated with primary antibodies at 4°C overnight followed by incubation with the following secondary antibodies at room temperature for 2 hours: goat anti-mouse IgG Alexa FluorR 488, goat anti-rabbit IgG Alexa FluorR 568, goat anti-rabbit Alexa FluorR 488, and goat anti-rat IgG Alexa FluorR 568 (all 1:400, Thermo Fisher Scientific; Supplementary Table S2). Antigen retrieval was performed prior to the staining by heating in 10 mmol/L citric acid buffer, pH 6.0. Images were taken on a Nikon C1 Spectral Imaging Confocal Laser Scanning Microscope System, using NIS Elements Microscope Imaging Software (Nikon Instruments). Video clips were prepared using Imaris 7.6.5, Bitplane Scientific Software.

Quantification of mouse immunofluorescence staining

Five to 12 images of each complete tumor (n = 4–5 per group) were taken using a Nikon C1 Spectral Imaging Confocal Laser Scanning Microscope System. HALO 2.0 Image Analysis Software (Indica Labs Informed Pathology) was used for the quantification. Pericyte coverage was defined as desmin+ area normalized to 5,000 μm2 of CD31+ area. Perivascular T cells were determined by spatial analysis of intratumoral CD8+ cells in relation to CD31+ endothelial cells over a distance of 100 μm. GLUT-1 expression was analyzed in vessels and tumor cells [mean fluorescence intensity (MFI) in 50 CD31+ cells and 1,000 tumor cells, respectively]. Lectin perfusion was quantified as the percentage of lectin+ area per total CD31+ area.

Hematoxylin and eosin staining of mouse tumor tissues

Twenty-one days post GL261 cell inoculation, mouse brains (n = 4 each treatment group) were dissected, embedded in Sakura Tissue-Tec O.C.T. Compound (Thermo Fisher Scientific), and frozen on dry ice. Ten-micron–thick cryosections were cut, dried for 10 minutes at 37°C, fixed in 4% PFA for 10 minutes at room temperature, and washed in deionized water prior to incubation in 20% Mayer hematoxylin solution for 4 minutes (Merck Chemicals). After washing in running tap water, slides were placed in a 0.25% alcoholic solution of Eosin Y for 30 seconds (Waldeck). The sections were washed in deionized water and dehydrated in increasing ethanol solutions (2 × 70%, 2 × 96%, 2 × 100%, for 10–15 seconds each). After clarification with xylene for 5 minutes (VWR), samples were mounted with Eukitt mounting medium (Th. Geyer). Brightfield images were obtained using a Nikon Eclipse 80i microscope and NIS Elements imaging software (Nikon Instruments). Necrotic areas (percentage of complete tumor) were assessed using HALO 2.0 Image Analysis Software (Indica Labs Informed Pathology; 5–12 sections/brain depending on the tumor size).

Electron microscopy of mouse GBM samples

For ultrastructural analysis, GL261-bearing mice were anesthetized 21 days after GL261 cell inoculation and transcardially perfused with PBS for 1 minute and 4% PFA/0.1 mol/L cacodylate buffer (pH 7.4) for 4 minutes (n = 5 per treatment group). Brains were dissected and postfixed in 4% PFA/2.5% glutaraldehyde/0.1 mol/L cacodylate buffer (pH 7.4) overnight. Small pieces of tissue containing GL261 tumors were postfixed in 1% OsO4 (Sigma-Aldrich), dehydrated in graded ethanol solutions (30%, 50%, 70%, 80%, 96%, 2 × 100% for 45 minutes at room temperature) and propylene oxide (2 × 100% for 45 minutes at room temperature; Sigma-Aldrich) prior to embedding in Epon (Sigma Aldrich). The polymerization was performed for 24 hours at 60°C. Ultrathin sections were cut on a Leica Ultracut UCT (Leica Microsystems), contrast enhanced with 1.5% uranyl acetate dehydrate (Serva)/0.2 mol/L sodium acetate/0.2 mol acetic acid and 3% lead citrate (Leica Microsystems) using Leica EM Stain (Leica Microsystems). Samples were analyzed using FEI Tecnai Spirit BioTWIN electron microscope at 120 kV (FEI Europe B.V.). Images were taken with an Eagle 4K CCD bottom-mount camera (FEI Europe B.V.).

Flow cytometry of mouse GL261 tumors

After dissection, GL261 tumors were transferred into ice cold Hank balanced salt solution (HBSS), gently minced, and incubated with HBSS containing collagenase P (0.2 mg/mL, Roche), dispase II (0.8 mg/mL, Roche), DNase I (0.01 mg/mL, Sigma), and collagenase A (0.3 mg/mL, Roche) for 60 minutes at 37°C as described previously (13). The digestion was stopped by adding FBS (Sera Plus, Pan Biotech) on ice and samples were spun down at 250 × g for 10 minutes at 4°C. The pellet was resuspended in 25% BSA/PBS, centrifuged at 2,000 × g (30 minutes at 4°C), and the myelin-containing supernatant was discarded. The sample was resuspended in 1 mL HBSS, filtered through a 40-μm mesh, and washed in HBSS (centrifugation at 250 × g for 10 minutes at 4°C). Lysis of erythrocytes was performed for 10 minutes at room temperature prior to staining (Red blood cell lysis buffer, Roche). Subsequently, 2 mL staining buffer (5% FBS in PBS) was added and cells were centrifuged at 250 × g for 10 minutes at 4°C. Cells were preincubated with rat anti-mouse FcγIII/II receptor (CD16/CD32)-blocking antibodies (≤1 μg/million cells/100 μL, BD Pharmingen) for 5 minutes at 4°C and stained with the fluorochrome-conjugated surface antibodies (0.25–1 μg, Supplementary Table S3). DAPI (Invitrogen, myeloid cells) or Fixable Viability Dye eFluor 780 (1:1,000, eBioscience, lymphoid cells) were used for dead cell exclusion.

For intracellular staining, cells were incubated with live/dead Fixable Viability Dye eFluor 780 (1:1,000, eBioscience). Cells were fixed with 1 mL of Fix/Perm (BD Biosciences) for 30 minutes, washed twice with Perm Wash (BD Biosciences), and stained with fluorochrome-conjugated antibodies (Supplementary Table S4) for 30 minutes on ice. Samples were acquired using a FACSCanto II flow cytometer (BD Biosciences) and further analyzed using FlowJo analytic software (v. 10.0.8, FLOWJO, LLC). Background fluorescence was determined by fluorescence minus one controls.

Hypoxia of GL261 cells

Subconfluent GL261 cells were cultured in DMEM-GlutaMAX-I (Gibco) supplemented with 10% FBS (Sera Plus, Pan Biotech) and 1% penicillin and streptomycin (Sigma) at 5% CO2 and 37°C. After 24 hours, the medium was removed and the cells were incubated in serum-free DMEM (Gibco) adjusted to 2 mmol/L glucose (Gibco). For hypoxic exposure, cells were placed in sealed Gas Pak EZ pouches for anaerobic culture with indicator (Becton-Dickinson) for 48 hours. For flow cytometry analysis, cells were detached with Accutase (Sigma), washed with flow cytometry buffer (PBS/5% FBS), and stained with rat anti-mouse PD-L1 (clone 10F.9G2, BioLegend). Samples were acquired using a FACSCanto II flow cytometer (BD Biosciences). Analysis was performed using FlowJo analytic software (v. 10.0.8, FLOWJO, LLC).

Isolation of mouse brain tumor microvessels

Brain tumors were dissected and rolled on a Whatman filter membrane (Schleicher & Schuell) to remove meninges as described previously (24, 25). In three independent experiments, GL261 tumor tissue and contralateral brain hemispheres from 3 animals for each treatment group were dissected, pooled, and homogenized in microvessel buffer (MVB; 15 mmol/L HEPES, 147 mmol/L NaCl, 4 mmol/L KCl, 3 mmol/L CaCl2, 1.2 mmol/L MgCl2, 5 mmol/L glucose, and 0.5% BSA, pH 7.4) using a dounce homogenizer (Wheaton, 0.025 mm clearance) and centrifuged at 400 × g for 10 minutes at 4°C. The pellet was resuspended in 25% BSA and centrifuged at 2,000 × g for 30 minutes to remove myelin. The microvessel pellet was resuspended in MVB and filtered through a 40 μm nylon mesh (BD Biosciences) to wash off nuclei, erythrocytes, and dead cells. Microvessels trapped on top of the mesh were lysed in RLTplus buffer (Qiagen) and stored at −80°C until use. The purity of the MBMV isolation was previously confirmed and showed a significant enrichment of endothelial cell markers (Cldn5, Cdh5) and negligible presence of other neurovascular unit (NVU) cell types (Aqp4, Ng2, Dcx; refs. 24, 25).

RNA sequencing of GL261 tumor microvessels

RNA was isolated from mouse brain tumor microvessels using the mRNAeasy Micro Kit (Qiagen; GEO repository number: GSE130324). To avoid contamination by genomic DNA samples were treated by on-column DNase digestion (DNase-Free DNase Set, Qiagen). Total RNA and library integrity were verified with LabChip Gx Touch 24 (PerkinElmer). One-hundred nanograms of total RNA was used as input for SMARTer Stranded Total RNA Sample Prep Kit - HI Mammalian (Clontech). Sequencing was performed on the NextSeq500 instrument (Illumina) using v2 chemistry, resulting in average of 25M (million) reads per library with 1 × 75 bp single-end setup. The resulting raw reads were assessed for quality, adapter content, and duplication rates with FastQC (http://www.bioinformatics.babraham.ac.uk/projects/fastqc). Reaper version 13–100 was employed to trim reads after a quality drop below a mean of Q20 (1 error in 100 base pairs) in a window of 10 nucleotides (26). Only reads between 30 and 150 nucleotides were cleared for further analyses. Trimmed and filtered reads were aligned versus the Ensembl mouse genome version mm10 (GRCm38) using STAR 2.4.0a with the parameter “–outFilterMismatchNoverLmax 0.1″ to increase the maximum ratio of mismatches to mapped length to 10% (27). The number of reads aligning to genes was counted with feature Counts 1.4.5-p1 tool from the R Subread package (Bioconductor; ref. 28). Only reads mapping at least partially inside exons were admitted and aggregated per gene. Reads overlapping multiple genes or aligning to multiple regions were excluded. Differentially expressed genes were identified using DESeq2 version 1.62 (Bioconductor; ref. 29). Only genes with a minimum fold change of ±1.5 (log2 ±0.59), a maximum Benjamini–Hochberg corrected P value of 0.05, and a minimum combined mean of 5 reads were deemed to be significantly differentially expressed. The Ensemble annotation was enriched with UniProt data (release 06.06.2014) based on Ensembl gene identifiers (Activities at the Universal Protein Resource (UniProt; www.uniprot.org). Dimension reduction analyses (principal component analysis; PCA) were performed on DESeq2 normalized and regularized log transformed counts using the R packages FactoMineR and factoextra. Differentially expressed genes (DEGs) were submitted to gene set enrichment analyses with KOBAS (http://kobas.cbi.pku.edu.cn; ref. 30). The resulting bubble plot shows pathways with Benjamini–Hochberg corrected P value < 0.05 (represented by dashed line). The larger gray circles are scaled to the number of genes comprising the respective pathway, while the smaller colored circles represent the subset found to be DEGs.

IHC and analysis of human GBM samples

Formalin-fixed paraffin-embedded (FFPE) brain tumor tissue specimen derived from 30 patients with GBM who received bevacizumab during their clinical course (see Table 1, obtained from 2009 and 2015) were collected and classified according to WHO criteria by board-certified neuropathologists (P.N. Harter and K.H. Plate; ref. 2). Patients with initial brain tumor resection/biopsy and reresection/rebiopsy or autopsy after bevacizumab treatment were included. Patients who did not receive bevacizumab were excluded from the study. Samples were stored at room temperature. Written consent was obtained from all patients. The study was conducted in accordance with ethical standards such as the Declaration of Helsinki. The study protocol was endorsed by the local ethical committee (GS 04/09 SNO-09-2018). Treatment-naïve and bevacizumab-treated matched-pair samples of individual patients were compared in this study. Biopsies were stained and analyzed independently in two laboratories (Edinger Institute, Frankfurt and Institute of Pathology, University of Cologne, Cologne, Germany).

Table 1.

Clinical data of patients with GBM treated with bevacizumab

Patient IDSexAge at tumor resection (years)MGMT statusDuration of BEV treatment (days)Termination of BEV treatment prior to recurrent tumor resection (days)Age at resection of recurrent tumor (years)PD-L1 expression pre/post-BEV
n.a. 50 n.a. n.a. n.a. n.a. ++/+++++ 
46 170 96 46 −/+ 
70 258 34 71 −/++++ 
43 140 31 43 −/+ 
62 210 169 63 −/− 
51 91 179 51 −/− 
30 407 204 31 −/− 
52 84 35 52 −/+ 
49 160 n.a. 49. −/− 
10 41 n.a. n.a. n.a. −/− 
11 56 n.a. 288 45 57 +/+ 
12 38 n.a. n.a. n.a. −/− 
13 55 n.a. 150 n.a. −/+ 
14 72 26 403 73 +/+++++ 
15 63 280 33 64 −/+ 
16 47 214 133 48 −/− 
17 45 n.a. n.a. n.a. n.a. −/− 
18 49 n.a. 84 32 50 −/− 
19 41 217 23 42 −/− 
20 47 n.a. n.a. n.a. n.a. −/+ 
21 46 n.a. n.a. n.a. n.a. −/− 
22 n.a. 65 n.a. n.a. n.a. n.a. −/++ 
23 41 n.a. n.a. n.a. n.a. −/− 
24 n.a. n.a. n.a. n.a. n.a. −/− 
25 45 281 34 46 −/− 
26 65 n.a. 62 36 65 −/− 
27 66 293 67 −/− 
28 53 n.a. 114 27 53 −/− 
29 62 n.a. n.a. n.a. n.a. −/− 
30 51 90 30 51 −/− 
Patient IDSexAge at tumor resection (years)MGMT statusDuration of BEV treatment (days)Termination of BEV treatment prior to recurrent tumor resection (days)Age at resection of recurrent tumor (years)PD-L1 expression pre/post-BEV
n.a. 50 n.a. n.a. n.a. n.a. ++/+++++ 
46 170 96 46 −/+ 
70 258 34 71 −/++++ 
43 140 31 43 −/+ 
62 210 169 63 −/− 
51 91 179 51 −/− 
30 407 204 31 −/− 
52 84 35 52 −/+ 
49 160 n.a. 49. −/− 
10 41 n.a. n.a. n.a. −/− 
11 56 n.a. 288 45 57 +/+ 
12 38 n.a. n.a. n.a. −/− 
13 55 n.a. 150 n.a. −/+ 
14 72 26 403 73 +/+++++ 
15 63 280 33 64 −/+ 
16 47 214 133 48 −/− 
17 45 n.a. n.a. n.a. n.a. −/− 
18 49 n.a. 84 32 50 −/− 
19 41 217 23 42 −/− 
20 47 n.a. n.a. n.a. n.a. −/+ 
21 46 n.a. n.a. n.a. n.a. −/− 
22 n.a. 65 n.a. n.a. n.a. n.a. −/++ 
23 41 n.a. n.a. n.a. n.a. −/− 
24 n.a. n.a. n.a. n.a. n.a. −/− 
25 45 281 34 46 −/− 
26 65 n.a. 62 36 65 −/− 
27 66 293 67 −/− 
28 53 n.a. 114 27 53 −/− 
29 62 n.a. n.a. n.a. n.a. −/− 
30 51 90 30 51 −/− 

NOTE: Scoring, < 6,5 MFI, no PD-L1 expression (−); 6,5–24,99 MFI, PD-L1 expression (+ to ++++); ≥ 25 MFI, high PD-L1 expression (+++++); MFI, mean fluorescence intensity/100 cells.

Abbreviations: BEV, bevacizumab; F, female; M, male; M, methylated; MGMT, O6-Methylguanine-DNA-Methyltransferase; n.a., not available; U, unmethylated.

Paraffin-embedded samples were cut into 3-μm sections on a Leica microtome (SM2000R, Leica Microsystems). Sections were deparaffinized in xylene (2 × 10 minutes, VWR) and rehydrated in decreasing concentrations of ethanol (100%, 96%, 70% 2 times 5 minutes each), followed by incubation in a tap water bath for 5 minutes. Immunofluorescence staining with mouse anti-human PD-L1 (clone 5H1, 1:1,000), mouse anti-human CD68 (clone PG-M1, 1:5,000), mouse anti-human CD8 (C8/144B, 1:5,000), mouse anti-human Ki67 (clone MIB-1, 1:1,000; Supplementary Table S5) was performed using the multiplex thyramide signal amplification (TSA) system according to the manufacturer's protocol (Opal 4-color IHC, NEL810001KT, PerkinElmer). The staining was performed sequentially with repeating runs of antigen retrieval by heating in 0.1 mol/L citric acid buffer pH 6.0, (10xAR6 buffer, AR6001KT, PerkinElmer) quenching of endogenous peroxidase activity with 3% hydrogen peroxide for 30 minutes at room temperature, incubation with blocking reagent (0.5%, FP1020, PerkinElmer) for 30 minutes at room temperature, and primary antibody incubation (for concentrations, see Supplementary Table S5) overnight at 4°C. Species-specific horseradish peroxidase (HRP)–conjugated secondary antibody (goat anti-mouse IgG-HRP-labeled, 10 μg/mL, NEF822001EA, PerkinElmer) was applied for 30 minutes at room temperature, followed by incubation with fluorescently labeled tyramide (1:50 in amplification diluent, NEL810001KT, PerkinElmer) for 10 minutes at room temperature.

For quantitative analysis of PD-L1 expression pre- and post-bevacizumab therapy (N = 30), images were obtained with a Nikon Eclipse 80i microscope (Nikon Instruments, Inc.) using the same parameter and camera settings for each image taken (10 images/sample). The mean intensity of PD-L1 expression in tumor cells was determined using NIS Elements Microscope Imaging Software (Nikon Instruments, Inc.) and recorded as three-step MFI score within 100 cells using following thresholds: < 6,5 MFI, no PD-L1 expression (−); 6,5–24,99 MFI, PD-L1 expression (+ to ++++); ≥ 25 MFI, high PD-L1 expression (+++++). Colabeling analysis of PD-L1, CD8, and CD68 expression was performed on multispectral images taken at the Vectra 3.0 Imaging System for Quantitative Pathology and using InForm V2.0.2 Image Analysis Software (Akoya Biosciences Inc.). For quantitative analysis of proliferating CD8+ cells (n = 10 biopsy samples), multispectral images were taken and unmixed using the Vectra 3.0 Imaging System for Quantitative Pathology (Akoya Biosciences Inc.), and analyzed by HALO 2.0 image analysis software (Indica Labs Informed Pathology).

In a second approach, GBM tissue specimens of the same patient cohort were processed for IHC with mouse anti-human PD-L1 clone 22C3 (Supplementary Table S5), an antibody that was established for PD-L1 expression scoring in non–small cell lung cancer (NSCLC; ref. 31). The PD-L1 IHC 22C3 pharmDx assay (code SK006, DAKO Agilent Technologies) was employed for IHC detection of PD-L1 by using the Envision FLEX visualization system on a DAKO Autostainer Link 48 IHC staining platform (DAKO Agilent Technologies) according to the manufacturer's instruction (Dako Agilent, Code SK006, see detailed staining procedure protocol herein). Tumor cells were considered as PD-L1 IHC-positive if it showed linear membranous staining of any intensity that could be complete/circular or partial. Glioblastoma specimens pre/post-bevacizumab tested for PD-L1 expression were scored and divided into three levels based on a tumor proportion score (TPS): TPS < 1%: no PD-L1 expression; TPS 1–49%: PD-L1 expression, TPS ≥ 50%: high PD-L1 expression (31).

Statistical analyses

Statistical analyses were performed using GraphPad Prism software (GraphPad Inc.) and JMP 11.0 software (SAS). Statistical tests applied are indicated in the figure legends. For all nonsurvival statistical analyses of two experimental groups, an unpaired, two-tailed Student t test was performed. For multiple comparison analyses of more than two groups, one-way ANOVA followed by Tukey posttest, Wilcoxon signed-rank test, or Kruskall-Wallis test followed by Dunn posttest were used. Kaplan–Meier survival curves were analyzed using the log-rank test. P < 0.05 were considered statistically significant and indicated with asterisks (*, P < 0.05; **, P < 0.01; ***, P < 0.001; ****, P < 0.0001). Data are represented as mean ± SEM if not otherwise indicated.

PD-L1 is a potential target for ICT in GBM

Although PD-L1 maintains immune homeostasis under normal conditions, tumors often exploit the PD-L1 pathway to inhibit antitumor immune responses (7). To test whether PD-1/PD-L1 signaling is a potential therapeutic target in GBM, we determined PD-L1 expression in biopsies before and after bevacizumab therapy using two different antibody approaches (Fig. 1; Table 1; Supplementary Fig. S1; Supplementary Table S6; refs. 31, 32). In a blinded IHC study, we investigated a cohort of 30 GBM matched-pair biopsies that were stained and analyzed independently in two laboratories. Images of representative GBM biopsies stained with anti–PD-L1 (clone 5H1; ref. 32) are displayed in Fig. 1A. Approximately 10% of resected, patients with treatment-naïve GBM exhibited PD-L1 expression (Fig. 1B; Table 1). Treatment with bevacizumab resulted in significantly increased PD-L1 expression in 36% of the patients (P < 0.001; Fig. 1B, left/bottom). Results obtained with clone 22C3, an antibody that was established for PD-L1 expression scoring in NSCLC (31), also showed increased PD-L1 expression in post-bevacizumab specimens, although on a lower overall level compared with clone 5H1 (Supplementary Fig. S1; Supplementary Table S6). Overall, anti-VEGF therapy led to a significant upregulation of PD-L1 on glioblastoma cells.

Figure 1.

Anti-VEGF (bevacizumab)–induced PD-L1 expression is a therapeutic target in human GBM. A, PD-L1, CD68, and CD8 expression was determined by immunofluorescence using the TSA amplification technique (PerkinElmer). Images of representative GBM biopsies pre- (top, scale bar, 20 μm) and post-bevacizumab (middle and bottom, scale bars, 50 μm) are displayed. PD-L1 (red), CD68 (white), and CD8 (green). B, Quantification of PD-L1, CD68, and CD8 expression (Wilcoxon signed-rank test). Matched-pair analyses of treatment-naïve and bevacizumab-treated human GBM specimens (left) and the statistical impact displayed as the difference in PD-L1 expression within this cohort (below, n = 30). Kruskal–Wallis test followed by Dunn posttest (*, P < 0.05; **, P < 0.01; ***, P < 0.001; ns, not significant). Bev, bevacizumab.

Figure 1.

Anti-VEGF (bevacizumab)–induced PD-L1 expression is a therapeutic target in human GBM. A, PD-L1, CD68, and CD8 expression was determined by immunofluorescence using the TSA amplification technique (PerkinElmer). Images of representative GBM biopsies pre- (top, scale bar, 20 μm) and post-bevacizumab (middle and bottom, scale bars, 50 μm) are displayed. PD-L1 (red), CD68 (white), and CD8 (green). B, Quantification of PD-L1, CD68, and CD8 expression (Wilcoxon signed-rank test). Matched-pair analyses of treatment-naïve and bevacizumab-treated human GBM specimens (left) and the statistical impact displayed as the difference in PD-L1 expression within this cohort (below, n = 30). Kruskal–Wallis test followed by Dunn posttest (*, P < 0.05; **, P < 0.01; ***, P < 0.001; ns, not significant). Bev, bevacizumab.

Close modal

In addition to its expression on glioblastoma cells, PD-L1 was detected on human tumor-infiltrating macrophages (mean value = 4.33%, n = 3) and CD8+ cytotoxic T cells (mean value = 2.66%, n = 3), whereby overall CD68+ macrophages numbers were significantly reduced and CD8+ T cells remained almost unchanged after bevacizumab therapy (medians CD68 pre/post-bevacizumab: 21.99/15.68, P < 0.0121; medians CD8 pre/post-bevacizumab: 0.24/0.26, P < 0.9003; Fig. 1B). However, CD8+ T-cell numbers pre/post-bevacizumab were heterogeneous among individual patients, being significantly increased in 11 of 30 (P = 0.0067), decreased in 10 of 30 (P = 0.0059), and remaining unchanged in 9 of 30 (P = 0.4436) patients (Fig. 1B). In future studies, it will be of interest to identify biomarkers predicting CD8+ T-cell responses in bevacizumab-treated patients.

These observations were supported by data in a syngeneic GBM mouse model where PD-L1 was upregulated by hypoxia and upon antiangiogenic therapy (Supplementary Fig. S2A and S2B). The significant upregulation of PD-L1 post antiangiogenic therapy may identify a barrier/immune escape mechanism in the GBM microenvironment, arguing for PD-L1 targeted therapies in GBM. We further identified a significant contribution of tumoral PD-L1 on increasing survival by employing a CRISPR/Cas9-mediated genome editing approach to delete PD-L1 in GL261 cells (Supplementary Fig. S2C). We, therefore, aimed to explore the PD-1/PD-L1 signaling pathway in experimental GBM using a therapeutic approach.

PD-1/PD-L1 pathway targeting in a mouse model of glioblastoma

Although T cells are not abundant in GBM (6), we have previously shown that targeting VEGF leads to an improved infiltration of T lymphocytes, which, however, is not sufficient to elicit durable antitumor immune responses (13). Having established that the PD-1/PD-L1 pathway is activated upon VEGF targeting (Fig. 1), we aimed to test a combination of anti–PD-1 and dual antiangiogenic therapy in the orthotopic, syngeneic GL261 model. GL261 cells were stereotactically implanted in the striatum of C57BL/6 mice. Once tumors were established, antiangiogenic therapy using aflibercept (VEGF-trap; ref. 23) and trebananib (AMG386, an Ang-1/Ang-2 bispecific peptibody that neutralizes both Tie2 ligands with a predominance for Ang-2 due to a 30-fold higher binding capacity; ref. 22) was initiated 5 days post intracerebral implantation (see ref. 13 for more details on aflibercept and AMG386). The kinetics and treatment schedules are displayed in Fig. 2A. Intracranial tumor experiments (Fig. 2A) were additionally performed in the syngeneic Tu-2449 glioma model (20, 21) with similar survival results (Supplementary Fig. S2D).

Figure 2.

The combination of anti-VEGF, anti–Ang-2, and anti–PD-1 therapy increases survival and leads to tumor vessel normalization in GL261 glioma. A, Diagram depicting the timeline and treatment schedule in the syngeneic, orthotopic GL261 glioma model. B, Kaplan–Meier survival curves and median survival of GL261-bearing mice after monotherapy and dual therapy with aflibercept/AMG386 and anti–PD-1 [n = 15 (three independent experiments) for IgG, anti–PD-1, aflibercept/AMG386, aflibercept/AMG386/anti–PD-1; n = 10 (one independent experiment) for aflibercept, AMG386, aflibercept/anti–PD-1, AMG386/anti–PD-1]. Right, flow cytometry analysis of the cytotoxic effects of different therapies (IgG control, n = 16; anti–PD-1, n = 9; aflibercept/AMG386, n = 21; aflibercept/AMG386/anti–PD-1, n = 9; two–four independent experiments). C, Representative H&E images of GL261-bearing mice 21 days postimplantation (n = 4/group) and graph displaying percent necrosis in the different treatment and control groups. D, Images and morphologic analysis of pericyte coverage (top; desmin, red; CD31, green; DAPI nuclear staining, blue), hypoxia (middle, evidenced by GLUT-1 expression; GLUT-1, green; CD31, red; DAPI nuclear staining, blue) and lectin perfusion (bottom; DyLight-488 L. esculentum lectin, green; CD31, red; DAPI nuclear staining, blue) in GL261 tumors after antiangiogenic and immunotherapy (IgG control, anti–PD-1, aflibercept/AMG386, aflibercept/AMG386/anti–PD-1) 21 days postimplantation. Single colors are displayed in gray scale. Scale bars, 25 μm. E–H, Analyses of pericyte coverage (desmin+), lectin perfusion, and hypoxia (GLUT-1+) in the different treatment groups 21 days postimplantation (E: control, n = 5; anti–PD-1, n = 4; aflibercept/AMG386, n = 5; aflibercept/AMG386/anti–PD-1, n = 4; F: all treatments, n = 4; and G and H: control, n = 4; anti–PD-1, n = 4; aflibercept/AMG386, n = 5; aflibercept/AMG386/anti–PD-1, n = 4). Values are mean + SEM (B) and mean + SD (E–H). Statistical analyses were performed using log-rank test (B) and one-way ANOVA multiple comparison with Tukey posttest (B and E–H; *, P < 0.05; **, P < 0.01; ***, P < 0.001; ****, P < 0.0001).

Figure 2.

The combination of anti-VEGF, anti–Ang-2, and anti–PD-1 therapy increases survival and leads to tumor vessel normalization in GL261 glioma. A, Diagram depicting the timeline and treatment schedule in the syngeneic, orthotopic GL261 glioma model. B, Kaplan–Meier survival curves and median survival of GL261-bearing mice after monotherapy and dual therapy with aflibercept/AMG386 and anti–PD-1 [n = 15 (three independent experiments) for IgG, anti–PD-1, aflibercept/AMG386, aflibercept/AMG386/anti–PD-1; n = 10 (one independent experiment) for aflibercept, AMG386, aflibercept/anti–PD-1, AMG386/anti–PD-1]. Right, flow cytometry analysis of the cytotoxic effects of different therapies (IgG control, n = 16; anti–PD-1, n = 9; aflibercept/AMG386, n = 21; aflibercept/AMG386/anti–PD-1, n = 9; two–four independent experiments). C, Representative H&E images of GL261-bearing mice 21 days postimplantation (n = 4/group) and graph displaying percent necrosis in the different treatment and control groups. D, Images and morphologic analysis of pericyte coverage (top; desmin, red; CD31, green; DAPI nuclear staining, blue), hypoxia (middle, evidenced by GLUT-1 expression; GLUT-1, green; CD31, red; DAPI nuclear staining, blue) and lectin perfusion (bottom; DyLight-488 L. esculentum lectin, green; CD31, red; DAPI nuclear staining, blue) in GL261 tumors after antiangiogenic and immunotherapy (IgG control, anti–PD-1, aflibercept/AMG386, aflibercept/AMG386/anti–PD-1) 21 days postimplantation. Single colors are displayed in gray scale. Scale bars, 25 μm. E–H, Analyses of pericyte coverage (desmin+), lectin perfusion, and hypoxia (GLUT-1+) in the different treatment groups 21 days postimplantation (E: control, n = 5; anti–PD-1, n = 4; aflibercept/AMG386, n = 5; aflibercept/AMG386/anti–PD-1, n = 4; F: all treatments, n = 4; and G and H: control, n = 4; anti–PD-1, n = 4; aflibercept/AMG386, n = 5; aflibercept/AMG386/anti–PD-1, n = 4). Values are mean + SEM (B) and mean + SD (E–H). Statistical analyses were performed using log-rank test (B) and one-way ANOVA multiple comparison with Tukey posttest (B and E–H; *, P < 0.05; **, P < 0.01; ***, P < 0.001; ****, P < 0.0001).

Close modal

GL261-bearing mice became symptomatic 21 days post tumor cell inoculation if left untreated, and dual anti-VEGF/Ang-2 treatment significantly improved survival compared with AMG386 or aflibercept monotherapies (Fig. 2B). Anti–PD-1 was administered 10 days after surgical implantation. Anti–PD-1 monotherapy led to improved survival (median 27 days) compared with controls (Fig. 2B). Combinations of either AMG386 or aflibercept with anti–PD-1 significantly improved survival, but the therapies showed less or similar efficacy than AMG386/aflibercept combination therapy (median survival AMG386/anti–PD-1: 33.5 days; aflibercept/anti–PD-1: 46.5 days; AMG386/aflibercept: 45 days). If checkpoint blockade was combined with aflibercept/AMG386, survival was improved to a greater extent than with individual or dual therapies (median 54 days; Fig. 2B). Flow cytometry analysis of enzymatically dissected tumors revealed decreased numbers of live cells in the TME, which were further reflective of effective tumor targeting upon double/triple therapies (Fig. 2B). Our findings suggest that ICT in combination with dual antiangiogenic therapy offers potential therapeutic avenues for GBM therapy.

Three weeks post tumor cell inoculation, tumor morphology differed in the treatment groups, evidenced by hematoxylin and eosin (H&E) staining (Fig. 2C). Signs of necrosis were evident and significantly increased upon antiangiogenic treatment but were reduced upon the addition of anti–PD-1 (Fig. 2C). Compared with IgG control and anti–PD-1 treatment, dual antiangiogenic and triple therapy led to vessel normalization with improved pericyte coverage (Fig. 2D and E; Supplementary Fig. S3; Supplementary Movies S1–S4), which in turn may facilitate immune cell recruitment and drug delivery. Tumor vessels that were normalized by aflibercept/AMG386 displayed improved perfusion, evidenced by intravenous injection of lectin, which was further significantly increased by the addition of anti–PD-1 (triple therapy; Fig. 2D and F). Perfusion of brain vessels from contralateral hemispheres was not affected by the different therapies (Supplementary Fig. S3B). Nonetheless, double and triple therapy increased hypoxia in both endothelial cells (Fig. 2D and G) and tumors cells (triple therapy; Fig. 2D and H). Although the combination of angiogenesis and ICT led to a survival benefit, hypoxia may have been caused by persistent tumor-promoting innate immune cells (such as M2 macrophages; ref. 33). At the same time, vessel normalization allows the recruitment of adaptive immune cells that may help to eradicate tumor cells (11). We, thus, commenced detailed analyses of cells within the tumor microenvironment (TME) by flow cytometry.

Assessing the GBM microenvironment after anti-VEGF/Ang-2 and anti–PD-1 therapy

To understand the underlying mechanisms that led to improved survival after anti-VEGF/Ang-2 and anti–PD-1 therapy, we assessed the immune cell composition in the GBM microenvironment. At 21 days (d) and upon the development of neurologic symptoms (endpoint), gliomas were dissected and processed for flow cytometry analyses (to discriminate between adaptive and innate immune cells (Fig. 3). The gating strategy for glioma-infiltrating lymphocytes is displayed in Supplementary Fig. S4. The number of CD45high cells significantly declined at the endpoint upon aflibercept/AMG386 and aflibercept/AMG386/anti–PD-1 treatment but not 21 days postsurgery. CD3+ T lymphocytes were increased in all therapy groups compared with controls (Fig. 3A; endpoint). We identified an increase in CD8+ cytotoxic T cells (endpoint vs. 21d) in the anti–PD-1 and aflibercept/AMG386/anti–PD-1 therapy regimen, whereas CD4+ Th cells declined in those groups (Fig. 3A; endpoint vs. 21d). Immunosuppressive FoxP3+ cells constituted a major fraction of CD4+ T cells (Fig. 3A). Overall, we observed less frequent CD4/FoxP3+ regulatory T cells (Tregs) upon triple therapy (Fig. 3A; endpoint, aflibercept/AMG386/anti–PD-1 vs. IgG). Although Tregs increased significantly in the control and aflibercept/AMG386 groups, this increase was diminished by the addition of ICT, which was also reflected by the ratio of CD8 and FoxP3+ cells (Fig. 3A). Our flow cytometry analyses demonstrated that the addition of anti–PD-1, single and in combination with aflibercept/AMG386, led to the infiltration of cytotoxic T cells, which may have glioma-eradicating capacities. This supports the survival data, where PD-1 targeting alone led to increased survival, which was further significantly extended when combined with anti-VEGF/Ang-2 (Fig. 2B). Our findings demonstrated that the combination of ICT and dual antiangiogenic therapy created a favorable immune environment with increased cytotoxic T cells that was accompanied by a Treg decrease. Compared with controls at endpoint analyses, we also observed a significant increase of B lymphocytes in the aflibercept/AMG386 and aflibercept/AMG386/anti–PD-1–treated group reflective for therapy resistance (Fig. 3A). In the context of the specialized GBM microenvironment, B lymphocytes are known to support tumor angiogenesis and regulate macrophage phenotypes to interfere with progression and antitumor immune responses (33).

Figure 3.

Dual antiangiogenic and ICT alter the immune cell composition in the GBM microenvironment. A, Flow cytometry analysis showing the individual immune cell subpopulations in GL261 tumors after employing the different therapy regimens (IgG control, anti–PD-1, aflibercept/AMG386, aflibercept/AMG386/anti–PD-1). The analysis was performed 21 days (21d) postimplantation and at the endpoint upon the development of neurologic symptoms. B, Glioma-infiltrating myeloid cells are displayed. Gating strategy was employed as described previously (13). Statistical analysis was performed using one-way ANOVA multiple comparison with Tukey posttest. Values are mean + SEM. CD45high 21d (n = 5/group), CD45high endpoint (control, n = 12; anti–PD-1, n = 7; aflibercept/AMG386, n = 10; aflibercept/AMG386/anti–PD-1, n = 4). CD3 21d (n = 5/group), CD3 endpoint (control, n = 14; anti–PD-1, n = 9; aflibercept/AMG386, n = 21; aflibercept/AMG386/anti–PD-1, n = 10). CD4 21d (n = 5/group), CD4 endpoint (control, n = 20; anti–PD-1, n = 9; aflibercept/AMG386, n = 21; aflibercept/AMG386/anti–PD-1 n = 10). CD8 21d (n = 5/group), CD8 endpoint (control, n = 16; anti–PD-1, n = 9; aflibercept/AMG386, n = 21; aflibercept/AMG386/anti–PD-1, n = 10). CD19 21d (n = 5/group), CD19 endpoint (control, n = 16; anti–PD-1, n = 9; aflibercept/AMG386, n = 21; aflibercept/AMG386/anti–PD-1, n = 9). MDSCs 21d (n = 5/group), MDSC endpoint (control, n = 16; anti–PD-1, n = 8; aflibercept/AMG386, n = 20; aflibercept/AMG386/anti–PD-1, n = 9). CD206 21d (n = 5/group), CD206 endpoint (control, n = 18; anti–PD-1, n = 8; aflibercept/AMG386, n = 16; aflibercept/AMG386/anti–PD-1, n = 11). MHCIIhigh 21d (n = 5/group), MHCIIhigh endpoint (control, n = 12; anti–PD-1, n = 7; aflibercept/AMG386, n = 15; aflibercept/AMG386/anti–PD-1, n = 9). MHCIIlow 21d (n = 5/group), MHCIIlow endpoint (control, n = 13; anti–PD-1, n = 6; aflibercept/AMG386, n = 15; aflibercept/AMG386/anti–PD-1, n = 9). CD45low 21 (n = 5/group), CD45low endpoint (control, n = 8; anti–PD-1, n = 4; aflibercept/AMG386, n = 16; aflibercept/AMG386/anti–PD-1, n = 9; *, P < 0.05; **, P < 0.01; ***, P < 0.001; ****, P < 0.0001; one–four independent experiments).

Figure 3.

Dual antiangiogenic and ICT alter the immune cell composition in the GBM microenvironment. A, Flow cytometry analysis showing the individual immune cell subpopulations in GL261 tumors after employing the different therapy regimens (IgG control, anti–PD-1, aflibercept/AMG386, aflibercept/AMG386/anti–PD-1). The analysis was performed 21 days (21d) postimplantation and at the endpoint upon the development of neurologic symptoms. B, Glioma-infiltrating myeloid cells are displayed. Gating strategy was employed as described previously (13). Statistical analysis was performed using one-way ANOVA multiple comparison with Tukey posttest. Values are mean + SEM. CD45high 21d (n = 5/group), CD45high endpoint (control, n = 12; anti–PD-1, n = 7; aflibercept/AMG386, n = 10; aflibercept/AMG386/anti–PD-1, n = 4). CD3 21d (n = 5/group), CD3 endpoint (control, n = 14; anti–PD-1, n = 9; aflibercept/AMG386, n = 21; aflibercept/AMG386/anti–PD-1, n = 10). CD4 21d (n = 5/group), CD4 endpoint (control, n = 20; anti–PD-1, n = 9; aflibercept/AMG386, n = 21; aflibercept/AMG386/anti–PD-1 n = 10). CD8 21d (n = 5/group), CD8 endpoint (control, n = 16; anti–PD-1, n = 9; aflibercept/AMG386, n = 21; aflibercept/AMG386/anti–PD-1, n = 10). CD19 21d (n = 5/group), CD19 endpoint (control, n = 16; anti–PD-1, n = 9; aflibercept/AMG386, n = 21; aflibercept/AMG386/anti–PD-1, n = 9). MDSCs 21d (n = 5/group), MDSC endpoint (control, n = 16; anti–PD-1, n = 8; aflibercept/AMG386, n = 20; aflibercept/AMG386/anti–PD-1, n = 9). CD206 21d (n = 5/group), CD206 endpoint (control, n = 18; anti–PD-1, n = 8; aflibercept/AMG386, n = 16; aflibercept/AMG386/anti–PD-1, n = 11). MHCIIhigh 21d (n = 5/group), MHCIIhigh endpoint (control, n = 12; anti–PD-1, n = 7; aflibercept/AMG386, n = 15; aflibercept/AMG386/anti–PD-1, n = 9). MHCIIlow 21d (n = 5/group), MHCIIlow endpoint (control, n = 13; anti–PD-1, n = 6; aflibercept/AMG386, n = 15; aflibercept/AMG386/anti–PD-1, n = 9). CD45low 21 (n = 5/group), CD45low endpoint (control, n = 8; anti–PD-1, n = 4; aflibercept/AMG386, n = 16; aflibercept/AMG386/anti–PD-1, n = 9; *, P < 0.05; **, P < 0.01; ***, P < 0.001; ****, P < 0.0001; one–four independent experiments).

Close modal

We next investigated whether ICT also influenced the intratumoral myeloid cell composition. We had previously shown that dual antiangiogenic therapy led to tumor-associated macrophage (TAM) depletion, whereas M2-polarized macrophages remained high in mouse GBM (13). TAM numbers declined after dual antiangiogenic therapy and upon the addition of anti–PD-1 (21d and endpoint, compared with control; Fig. 3B). When analyzing protumorigenic M2-polarized macrophages (CD45highCD11b+Gr-1F4/80+CD206+) 21 days post tumor cell inoculation, CD206+ infiltrates were not significantly different among the treatment groups (controls vs. anti–PD1, aflibercept/AMG386, aflibercept/AMG386/anti–PD-1; Fig. 3B). However, they significantly increased at the endpoint in all treatment groups, although numbers were not significantly influenced by the anti–PD-1 therapy itself (Fig. 3B). We hypothesized that CD206+ macrophages were potentially recruited from the circulation and represented reeducated TAMs as the tumor progresses. Their presence/predominance may, thus, contribute to tumor refractoriness to therapy by the secretion of growth and immunosuppressive factors. M1-polarized (CD45highCD11b+Gr-1F4/80+MHCIIhigh) macrophages were present in tumors and did not further increase during progression (Fig. 3B). Myeloid-derived suppressor cell (MDSC; CD45highCD11b+Gr-1+) numbers declined upon anti-VEGF/Ang-2/anti–PD-1 therapy (Fig. 3B, endpoint). Microglia significantly increased upon dual and triple therapy (Fig. 3B), indicative for tumor-eradicating properties. Overall, our findings demonstrated that combined vascular and immune checkpoint targeting led to reduced tumor-promoting MDSCs, while it did not interfere with M2-polarized macrophage numbers.

Depletion of CD8+ T cells reverses the survival benefit after triple therapy

Having established that CD8+ cytotoxic lymphocytes were increased after aflibercept/AMG386/anti–PD-1 therapy, we aimed to assess whether depletion of CD8 would attenuate the survival benefit. GL261-bearing mice receiving anti–PD-1 and aflibercept/AMG386 treatment were additionally given anti-CD8 twice per week starting from day 14 post tumor cell implantation (Fig. 4A). Using this strategy, we demonstrated that survival was alleviated when CD8+ cells were depleted in the aflibercept/AMG386/anti–PD-1 treatment group (Fig. 4B). Median survival was reduced to 46 days and resembled values observed after double therapy (Fig. 4B). This finding supports the hypothesis that tumor-infiltrating CD8+ T cells contributed to a favorable outcome in mouse GBM and provide a rationale for targeting adaptive immune cells in the specialized glioma TME. To reveal the mediators regulating CD8+ T-cell activity, we performed flow cytometry of tumor-infiltrating lymphocytes (TIL) and determined intracellular cytokines (IFNγ, TNFα) in addition to PD-1 expression, reflective for CD8+ T-cell exhaustion (Fig. 4C). Although CD8+ T-cell numbers were increased upon anti–PD-1 and aflibercept/AMG386/anti–PD-1 therapies (Fig. 3B), elevated IFNγ and TNFα were solely detected in double and triple therapies but not PD-1 monotherapy, indicative of a proinflammatory Th1 signature (Fig. 4C). Our results are in line with previous work (18, 19) that reported increased IFNγ upon anti–Ang-2 and/or anti-VEGF therapy in breast, pancreatic, and melanoma cancer models.

Figure 4.

CD8 blockade reverses the survival benefit achieved by the combination of anti-VEGF/Ang-2 and anti–PD-1 immunotherapy. A, Diagram depicting the timeline and the addition of anti-CD8 treatment to the dual antiangiogenic and ICT schedule in the GL261 glioma model. B, Kaplan–Meier survival analysis of GL261 tumor–bearing mice after different therapy regimens and median survival rates are displayed (IgG control, n = 15; anti–PD-1, n = 15; anti–PD-1/anti-CD8, n = 10; aflibercept/AMG386/anti–PD-1, n = 15; aflibercept/AMG386/anti–PD-1/anti-CD8, n = 10; two–three independent experiments). C, Flow cytometry plots showing IFNγ+, TNFα+, PD-1+, and Ki67+ expression in CD8+ glioma-infiltrating T lymphocytes after therapy (IgG control, anti–PD-1, aflibercept/AMG386, aflibercept/AMG386/anti–PD-1). The analyses were performed upon the development of neurologic symptoms (endpoint). IFNγ (control, n = 7; anti–PD-1, n = 6; aflibercept/AMG386, n = 6; aflibercept/AMG386/anti–PD-1, n = 4); TNFα (control, n = 7; anti–PD-1, n = 6; aflibercept/AMG386, n = 6; aflibercept/AMG386/anti–PD-1; n = 4); PD-1 (control, n = 7; anti–PD-1, n = 6; aflibercept/AMG386, n = 6; aflibercept/AMG386/anti–PD-1, n = 4); Ki67 (control, n = 7; anti–PD-1, n = 6; aflibercept/AMG386, n = 6; aflibercept/AMG386/anti–PD-1, n = 4; one independent experiment). D, Morphologic and quantitative analysis of CD8+ T-lymphocyte infiltration in GL261 glioma. Representative immunofluorescence images of mouse brain tumor sections 21 days postimplantation in the different treatment groups (CD31, green; CD8, red; DAPI nuclear staining, blue). The number and distance of CD8+ T cells to CD31+ glioma vessels in the different therapy regimen are displayed (n = 5/group). Scale bar, 20 μm. E, Representative immunofluorescence images of GBM biopsies [treatment-naïve, post-bevacizumab (Bev)] stained with anti-CD8 (green) and anti-Ki67 (red) and the quantification are displayed, n = 10. Scale bar, 10 μm. Values are mean + SEM (C) and mean + SD (D–E). Statistical analyses were performed using log-rank test (B), one-way ANOVA multiple comparison with Tukey posttest (C and D), and Student t test (E; *, P < 0.05; **, P < 0.01; ***, P < 0.001; ****, P < 0.0001).

Figure 4.

CD8 blockade reverses the survival benefit achieved by the combination of anti-VEGF/Ang-2 and anti–PD-1 immunotherapy. A, Diagram depicting the timeline and the addition of anti-CD8 treatment to the dual antiangiogenic and ICT schedule in the GL261 glioma model. B, Kaplan–Meier survival analysis of GL261 tumor–bearing mice after different therapy regimens and median survival rates are displayed (IgG control, n = 15; anti–PD-1, n = 15; anti–PD-1/anti-CD8, n = 10; aflibercept/AMG386/anti–PD-1, n = 15; aflibercept/AMG386/anti–PD-1/anti-CD8, n = 10; two–three independent experiments). C, Flow cytometry plots showing IFNγ+, TNFα+, PD-1+, and Ki67+ expression in CD8+ glioma-infiltrating T lymphocytes after therapy (IgG control, anti–PD-1, aflibercept/AMG386, aflibercept/AMG386/anti–PD-1). The analyses were performed upon the development of neurologic symptoms (endpoint). IFNγ (control, n = 7; anti–PD-1, n = 6; aflibercept/AMG386, n = 6; aflibercept/AMG386/anti–PD-1, n = 4); TNFα (control, n = 7; anti–PD-1, n = 6; aflibercept/AMG386, n = 6; aflibercept/AMG386/anti–PD-1; n = 4); PD-1 (control, n = 7; anti–PD-1, n = 6; aflibercept/AMG386, n = 6; aflibercept/AMG386/anti–PD-1, n = 4); Ki67 (control, n = 7; anti–PD-1, n = 6; aflibercept/AMG386, n = 6; aflibercept/AMG386/anti–PD-1, n = 4; one independent experiment). D, Morphologic and quantitative analysis of CD8+ T-lymphocyte infiltration in GL261 glioma. Representative immunofluorescence images of mouse brain tumor sections 21 days postimplantation in the different treatment groups (CD31, green; CD8, red; DAPI nuclear staining, blue). The number and distance of CD8+ T cells to CD31+ glioma vessels in the different therapy regimen are displayed (n = 5/group). Scale bar, 20 μm. E, Representative immunofluorescence images of GBM biopsies [treatment-naïve, post-bevacizumab (Bev)] stained with anti-CD8 (green) and anti-Ki67 (red) and the quantification are displayed, n = 10. Scale bar, 10 μm. Values are mean + SEM (C) and mean + SD (D–E). Statistical analyses were performed using log-rank test (B), one-way ANOVA multiple comparison with Tukey posttest (C and D), and Student t test (E; *, P < 0.05; **, P < 0.01; ***, P < 0.001; ****, P < 0.0001).

Close modal

Tumor-derived VEGF has previously been shown to enhance the expression of PD-1 and other inhibitory checkpoints (34). In line with this observation, upon anti–PD-1 treatment (i.e., when VEGF is present in the TME), but not after aflibercept/AMG386/anti–PD-1 therapy, CD8+ TILs upregulated PD-1, indicative for exhaustion (Fig. 4C). Proliferation of cytotoxic T cells was increased in the aflibercept/AMG38/anti–PD-1 group (Fig. 4C), and CD8+ T cells were preferentially associated with GBM vessels upon triple therapy (Fig. 4D). Concurrent with findings in the mouse model, we observed significantly increased numbers of proliferating CD8+ T cells in bevacizumab-treated versus treatment-naïve GBM biopsies (Fig. 4E). In conclusion, our data established that increased numbers of glioma-infiltrating CD8+ T cells are important contributors to antitumor responses in GBM by providing a proinflammatory cytokine signature particularly when combined with dual antiangiogenic therapy.

Triple therapy leads to global vascular normalization of GBM microvessels

We next analyzed the transcriptional profiles of GL261 brain tumors by RNA sequencing (RNA-seq). To understand how the synergistic effects of the combination therapies effectuated gene regulation in the vascular compartment, we analyzed the vessels of glial tumors 21 days postimplantation (Fig. 5). The tumor microvessel isolation procedure is displayed in Fig. 5A. Purity of the microvessel isolation and endothelial enrichment has been demonstrated in previous reports (24, 25). Volcano plots and heatmaps of the top 50 significant regulated genes among the treatment groups indicated several genes that were dysregulated in mouse GBM compared with sham-operated animals, which were also differentially regulated upon the addition of antiangiogenic therapy or further addition of anti–PD-1 (Fig. 5B and C). PCA plot demonstrated that the transcriptome of control tumor vessels shifted toward the profile of sham-operated animals upon double therapy, an effect that was more pronounced with triple therapy, suggesting a global normalization of tumor vasculature gene transcription with triple therapy (Fig. 5D; triple therapy shifts toward sham animals, circled). Venn diagrams illustrated that in the vasculature of untreated GL261 tumors, 7,851 genes were differentially regulated compared with sham controls (i.e., representative of normal brain vessels; Fig. 6A). Of these, 4,871 genes returned to baseline after anti-VEGF/Ang2 therapy and 5,331 genes after the combined treatment with anti-VEGF/Ang2 and anti–PD-1 (Fig. 6A). This indicated that anti–PD-1 therapy targeted an additional set of genes (981) compared with dual antiangiogenic therapy (4,350 that were common to double and triple therapy), and highlights the therapy benefit upon anti–PD-1 addition to vascular targeting (Fig. 6A). Molecules associated with a normalized vasculature or blood–brain barrier (BBB) function returned to baseline levels (Fig. 6B). For example, PDGFRB, ANGPT1, and TEK (TIE2) were downregulated in GL261 control tumors and upregulated following double and triple therapy (Fig. 6B). A normalized vasculature with an improved pericyte coverage and perfusion was also seen via histology (Fig. 2D; Supplementary Fig. S3A; Supplementary Movies S1–S4) and ultrastructural analysis (Supplementary Fig. S3C). At the same time, Angpt2 (also Vegfa, and Tgfb1 to some extent) was downregulated after double/triple therapy, respectively (Fig. 6B). Tight junction molecules, such as Occludin (Ocln), and other BBB-associated molecules, such as β-catenin (Ctnnb1) and GLUT-1 (Slc2a1), were upregulated upon triple therapy (Fig. 6B). Pathway enrichment analysis using KEGG pathway database in untreated tumors (7,851 genes from group 1, Fig. 6A) indicated dysregulated pathways in cancer, angiogenesis, cell adhesion molecules, and immune cell response (Fig. 6C and D).

Figure 5.

Anti–PD-1 blockade combined with antiangiogenic therapy reshapes the transcriptomes in microvessels isolated from GL261 tumors. A, Diagram of the workflow for the isolation of brain tumor microvessels. The brain tissue was mechanically and enzymatically dissociated before removal of myelin and filtered through cell strainers. Isolated microvessels were microscopically inspected for purity. See bottom photograph. Scale bar, 20 μm. B, Volcano plots highlighting significant (P < 0.05) genes for comparison of treatment groups within the dataset. The indicated conditions are (clockwise): untreated (IgG) versus sham-operated, aflibercept/AMG386-treated versus IgG, aflibercept/AMG386/anti–PD-1-treated versus IgG, and aflibercept/AMG386-treated versus aflibercept/AMG386/anti–PD-1-treated. The log2 counts were plotted versus the log2 fold change (FC). Significantly upregulated genes (log2 fold change of more than ±1 and FDR < 1%) are displayed in green or red according to the groups. C, Heatmap of the top 50 most regulated genes in brain microvessels from IgG versus sham, aflibercept/AMG386, aflibercept/AMG386/anti–PD-1–treated mice, and between the dual- and triple-therapy groups. The color is based on raw Z score. D, PCA of all samples within the dataset showing differences in dimension 1 (29.8%) and 2 (10.3%) for all detected genes. Each dot corresponds to the pool of three brain tumors from each therapy group (of n = 3 independent experiments). The triple therapy is circled.

Figure 5.

Anti–PD-1 blockade combined with antiangiogenic therapy reshapes the transcriptomes in microvessels isolated from GL261 tumors. A, Diagram of the workflow for the isolation of brain tumor microvessels. The brain tissue was mechanically and enzymatically dissociated before removal of myelin and filtered through cell strainers. Isolated microvessels were microscopically inspected for purity. See bottom photograph. Scale bar, 20 μm. B, Volcano plots highlighting significant (P < 0.05) genes for comparison of treatment groups within the dataset. The indicated conditions are (clockwise): untreated (IgG) versus sham-operated, aflibercept/AMG386-treated versus IgG, aflibercept/AMG386/anti–PD-1-treated versus IgG, and aflibercept/AMG386-treated versus aflibercept/AMG386/anti–PD-1-treated. The log2 counts were plotted versus the log2 fold change (FC). Significantly upregulated genes (log2 fold change of more than ±1 and FDR < 1%) are displayed in green or red according to the groups. C, Heatmap of the top 50 most regulated genes in brain microvessels from IgG versus sham, aflibercept/AMG386, aflibercept/AMG386/anti–PD-1–treated mice, and between the dual- and triple-therapy groups. The color is based on raw Z score. D, PCA of all samples within the dataset showing differences in dimension 1 (29.8%) and 2 (10.3%) for all detected genes. Each dot corresponds to the pool of three brain tumors from each therapy group (of n = 3 independent experiments). The triple therapy is circled.

Close modal
Figure 6.

Bioinformatic analysis evidenced a normalization of cancer- and barrier-related pathways in glioma vessels after anti-VEGF/Ang-2 and anti–PD-1 immunotherapy. A, Venn diagrams showing the number of regulated genes between sham versus IgG (I: 7,851), sham versus double therapy (II: 3,335), and sham versus triple therapy (III: 2,915). Red, genes dysregulated in mouse GBM vessels; green, genes rescued by double therapy; purple, genes rescued by triple therapy; blue, genes rescued by both therapies (double and triple); and orange, genes rescued by triple therapy exclusively. B, Genes relating to BBB function, such as Pdgfrb, Angpt1, Tie2, Ocln, Ctnnb1, Glut-1, and Slc2a1, that were dysregulated in tumor vessels were brought back to sham (no tumor) levels. Tumor-promoting genes (Angpt2, Vegfa, Fasl, and Tgfb1) were downregulated after double/triple therapy. Statistical analysis was performed using DESeq2 algorithm as part of the RNA-seq analysis (P values: **, P < 0.01; ***, P < 0.001). C and D, Bubble plot of the top candidates according to KEGG pathway (KOBAS enrichment) analysis using genes significantly regulated in the different treatment groups [P < 0.05, absolute (log2FC) > 0.585]. Pathways significantly dysregulated in mouse GBM vessels (red bubbles) were unified with analyses from the genes normalized in the two therapy groups, including the common and exclusive triple-therapy group. The resulting pathways were sorted by amount of correlation between all contrasts (i.e., groups color coded in A). y-axis represents value of significance for each contrast, and size of bubble reflects the number of significant hits. Smallest size corresponds to nonsignificant test in the contrast. Red bubbles, sham versus IgG; blue bubbles, both therapies (double and triple); green bubbles, double therapy; purple bubbles, triple therapy; and orange bubbles, triple exclusive (see also A). Specific significant pathways have been highlighted with the corresponding color of the therapy group (colored boxes). n.s., not significant.

Figure 6.

Bioinformatic analysis evidenced a normalization of cancer- and barrier-related pathways in glioma vessels after anti-VEGF/Ang-2 and anti–PD-1 immunotherapy. A, Venn diagrams showing the number of regulated genes between sham versus IgG (I: 7,851), sham versus double therapy (II: 3,335), and sham versus triple therapy (III: 2,915). Red, genes dysregulated in mouse GBM vessels; green, genes rescued by double therapy; purple, genes rescued by triple therapy; blue, genes rescued by both therapies (double and triple); and orange, genes rescued by triple therapy exclusively. B, Genes relating to BBB function, such as Pdgfrb, Angpt1, Tie2, Ocln, Ctnnb1, Glut-1, and Slc2a1, that were dysregulated in tumor vessels were brought back to sham (no tumor) levels. Tumor-promoting genes (Angpt2, Vegfa, Fasl, and Tgfb1) were downregulated after double/triple therapy. Statistical analysis was performed using DESeq2 algorithm as part of the RNA-seq analysis (P values: **, P < 0.01; ***, P < 0.001). C and D, Bubble plot of the top candidates according to KEGG pathway (KOBAS enrichment) analysis using genes significantly regulated in the different treatment groups [P < 0.05, absolute (log2FC) > 0.585]. Pathways significantly dysregulated in mouse GBM vessels (red bubbles) were unified with analyses from the genes normalized in the two therapy groups, including the common and exclusive triple-therapy group. The resulting pathways were sorted by amount of correlation between all contrasts (i.e., groups color coded in A). y-axis represents value of significance for each contrast, and size of bubble reflects the number of significant hits. Smallest size corresponds to nonsignificant test in the contrast. Red bubbles, sham versus IgG; blue bubbles, both therapies (double and triple); green bubbles, double therapy; purple bubbles, triple therapy; and orange bubbles, triple exclusive (see also A). Specific significant pathways have been highlighted with the corresponding color of the therapy group (colored boxes). n.s., not significant.

Close modal

Corresponding analysis of genes that returned to baseline in double and triple therapy [groups 2 (4,871) and 3 (5,331); Fig. 6A] indicated that the pathways were normalized by the respective therapy (Fig. 6C and D). To analyze the significance of the genes normalized by the two therapies, common to both or exclusive to triple therapy, we performed bioinformatic pathway analyses of these subsets comparing them to genes dysregulated in untreated tumor samples when compared with normal sham microvessels. We observed that cancer-related pathways were dysregulated in untreated tumors (group 1) because they are above the significance threshold (Fig. 6C and D, red bubbles above the dotted line). Several of these pathways were normalized by the triple therapy, some also by double therapy, and a few exclusively by triple therapy, indicated by the corresponding bubbles closer or below the significance cutoff (dotted line). Specifically, PKG, cAMP, foxO, and insulin signaling pathways were normalized by triple therapy and to a lesser extent by double therapy (purple box, Fig. 6C), whereas p53, TGFβ, and Hippo signaling pathways were normalized by the dual antiangiogenic therapy (green boxes, Fig. 6C). This was also the case for adherens and tight junction pathways, indicating the beneficial effect of dual antiangiogenic therapy. Extended KEGG pathway enrichment analysis indicated that oxytocin, MAPK, and calcium signaling pathways were also normalized in the triple therapy group but mediated exclusively by the addition of anti–PD-1 (Fig. 6D, orange boxes). There were, however, several dysregulated pathways in mouse GBM that were not normalized by either therapy (Fig. 6D, red boxes). This group could provide candidate pathways for complete targeting in future therapies.

Immune checkpoint blockade is currently pursued in clinical trials in GBM, although it is a non-T-cell–inflamed cancer (10, 35). Anti–PD-1 monotherapy shows little therapeutic benefit in GBM (8). We here provided a rational combinatorial approach to improve the efficacy of immune therapy by integrating the synergistic effects of VEGF/Ang-2 and PD-1 blockade.

Considering the failure of antiangiogenic therapy with bevacizumab for newly diagnosed GBM and an unfavorable survival with standard therapy (1, 3), an urgent need exists for new treatment options to target GBM. The therapeutic efficacy of immune checkpoint inhibition in other cancer modalities is appealing (7, 10) and could be an option for GBM as well. However, immune checkpoint monotherapy using PD-1 antibodies is not successful in recurrent GBM (36), whereas clinical studies in newly diagnosed glioblastoma are ongoing (Clinical Trials.gov identifiers NCT02617589 and NCT02667587).

The aim of this study was to test in syngeneic, preclinical GBM models whether ICT is beneficial when combined with dual antiangiogenic therapy. We demonstrated that ICT led to improved survival following anti-VEGF/Ang-2 therapy by creating an immunostimulatory microenvironment nourished with CD8+ CTLs and reduced immunosuppressive MDSCs and FoxP3+ Tregs. We provided evidence at the transcriptional level that the combination of PD-1 and VEGF/Ang-2-targeting therapies normalized the gene expression in endothelial cells back to almost physiologic conditions.

Numerous immunosuppressive factors such as VEGF, IL10, and TGBβ promote the downregulation of the host immune system and lead to the exclusion of tumor-fighting immune cells (33). We and others have shown that antiangiogenic therapy targeting VEGF and Ang-2/Tie2 signaling alleviate immune suppression and improve survival in GBM models (13–15). However, efficient targeting of tumor cells was not achieved, as the adaptive host immune system was not fully activated to promote glioma cell eradication. We here aimed to test the hypothesis that anti-VEGF/Ang-2 therapy was beneficial when combined with anti–PD-1 in glioma models. Basis for our hypothesis was the finding that PD-L1 was upregulated in patients with GBM after bevacizumab treatment. Although differential outcomes on patient survival with regard to PD-L1 expression have been reported (37–39), the presence of PD-L1 in GBM serves as a checkpoint and marker for therapy responses (7). PD-L1 promotes immunosuppression by inhibiting T-cell function (4) and macrophage phagocytosis (40). We hypothesized that PD-L1/PD-1 targeting could be an alternative approach to alleviate immune suppression in GBM to turn a T-cell–poor (“cold”) into T-cell–rich (“hot”) GBM microenvironment.

In our study, the combination of dual antiangiogenic and PD-1 checkpoint therapy led to a significant increased overall survival. Compared with antiangiogenic therapy alone, the survival of glioma-bearing mice was significantly extended upon the addition of anti–PD-1. Antiangiogenic therapy with aflibercept, AMG386, and anti–PD-1 rendered a normalized vasculature with improved pericyte coverage at the histologic and the gene expression level. Disruption of vessel normalization has been reported to reduce T-lymphocyte infiltration in transgenic mice and xenograft models (41). A functional vasculature is a prerequisite for leukocyte migration and, thus, provides a barrier within the TME (9, 33, 42). Proper adhesion molecules need to be present for effective T-cell recruitment to eradicate glioma cells (11). VEGF and Ang-2 are among the factors that create an immunosuppressive environment, which also affects the expression of ICAM-1 and VCAM-1 (11).

In conjunction with a normalized vasculature, the increased expression of cell adhesion molecules provides a basis for effector lymphocyte recruitment. Our RNA-seq analysis showed decreased FasL expression with increasing therapies. FasL has previously been shown to act as checkpoint at the vascular level that prevents lymphocyte entry into the TME, whereby tumor-derived VEGF and IL10 cooperatively induced FasL expression in endothelial cells to promote T-effector cell death (12). The presence of FasL might, in part, explain that CD8+ effector cells are not frequent during early tumor progression. Indeed, CD8+ T cells increased in the anti–PD-1 and aflibercept/AMG386/anti–PD-1–treated groups later during tumor progression, which demonstrated that checkpoint therapy was able to efficiently reinvigorate effector cell infiltration in formerly T-cell–poor brain tumors. The survival benefit was superior in the triple combination group. When CD8+ T cells were depleted, the survival benefit was diminished more effectively in triple therapy than anti–PD-1 monotherapy, and CD8+ effector T cells were identified as crucial contributors to the extended survival upon aflibercept/AMG386/anti–PD-1 therapy, in line with findings in breast cancer models (41). CD8+ T lymphocytes in the anti–PD-1 single treatment group, however, showed different characteristics: they produced less IFNγ, proliferated less, and had higher PD-1 expression, indicative of exhaustion, compared with triple therapy. Indeed, VEGF has been demonstrated to contribute to T-cell exhaustion by the upregulation of immune checkpoints, including PD-1 (34).

Our data demonstrated a favorable outcome in the GBM models upon triple therapy, which is in line with literature that suggests an improved efficacy of antiangiogenic therapy (targeting Ang-2 and/or VEGF signaling) when combined with checkpoint therapy (targeting PD-1 or PD-L1) in preclinical animal models (18, 19). However, in a transgenic GBM model, the benefit was not as efficient as in non-brain tumor models, which may relate to tumor-intrinsic factors of the GBM model applied and/or the level of PD-L1 expression (19). In the GL261 model, which is associated with high expression of PD-L1, the addition of anti–PD-1 led to improved survival compared with controls in GL261 and even more efficiently in Tu-2449 gliomas, indicative of differences in the glioma microenvironment that possibly also influence the kinetics and frequency of CD8+ TIL influx. Reports also indicate that PD-1 clones applied in GBM models may lead to different outcomes with regard to CD8+ T-cell reinvigoration, therapy, efficacy, and long-term survival (43, 44). The timing of therapy may be crucial as suggested by a study demonstrating that neoadjuvant administration of anti–PD-1 enhances effector cell–mediated antitumor immune responses in recurrent GBM (45).

PD-L1 is expressed on both tumor and immune cells. When PD-L1 was eliminated by a genetic knockdown approach and PD-L1–deficient GL261 cells were implanted in the striatum of C57BL/6 mice, we observed significantly extended survival indicating an essential contribution of glioma cell–derived PD-L1, in line with previous observations (46). PD-L1 was upregulated upon aflibercept/AMG386 treatment, which leads to increased hypoxia, a known driver of PD-L1 expression (47). PD-L1 can further be induced via IFNγ secretion from CD8+ T cells (18, 19). In line with these observations, in the matched-pair biopsy cohort, compared with treatment-naïve GBM, PD-L1 was significantly upregulated after bevacizumab therapy. In recurrent GBM, PD-L1 may provide a therapeutic target, especially in conjunction with vascular targeting to enhance antitumor immunity.

Because brain tumors were not eradicated completely upon triple therapy, additional components of TME are potentially involved that prevent overall tumor clearance. Indeed, although cells of the innate immune system were diminished with antiangiogenic treatment, they remained high upon anti–PD-1 monotherapy. The addition of anti–PD-1 had no effect on CD206+ macrophages residing in the GBM microenvironment as immune-suppressors and effectuated fewer MDSCs and FoxP3+ Tregs, as reported previously, using the same GBM models (43, 44). In patients with GBM, MDSCs play a major role in glioma-induced T-cell suppression via the PD-1/PD-L1 axis (48). Elevated hypoxia observed upon vascular targeting may be a driving force for remaining M2-polarized macrophages as reported previously (33). Thus, it might be appealing to create a less hypoxic GBM milieu to improve therapy efficacy. Simultaneous targeting of Tie2 and Ang-2 leads to tumor vessel normalization, decreased hypoxia, improved drug delivery, and favorably alters the TME (49). Similarly, an improvement of BBB function in GBM may be achieved with inhibitors that activate Tie2 (25, 50–52). In particular, improving endothelial cell-barrier functions may be useful for edema management in patients with GBM, which is a major clinical need (53).

In an attempt to further understand the mechanisms underlying the survival benefit achieved by the combined targeting of the vasculature and the PD-1/PD-L1 pathway, RNA-seq analysis was performed on isolated tumor vessels. DEseq2-based analysis within the different treatment groups revealed a number of differentially regulated genes, uncovering treatment-specific transcriptomic changes. Although barrier-associated genes were downregulated in control tumors, genes that were associated with vascular stabilization [e.g., Pdgfrb, Angpt1, Tek (Tie2)] and BBB function were upregulated upon dual antiangiogenic or triple therapy. Some unfavorable cytokines (such as TGFβ) that were dysregulated in glioblastoma were restored close to levels obtained in the sham control group, indicative of the efficacy of checkpoint therapy in the glioma model. Venn diagram analysis showed a unique set of genes that were restored to basal levels (“normalized”) in the triple therapy, and were common to both double and triple therapy. Bioinformatic analysis of these two subgroups provided further insights into the molecular basis for the increased survival in the triple therapy, and vascular normalization and survival benefit in the dual therapy. Several pathways such as PKG, cAMP, foxO, and insulin signaling were normalized in the triple therapy. However, oxytocin, MAPK, and calcium signaling pathways were normalized exclusively in the triple therapy, indicating that the effect is specific to addition of anti–PD-1. At the same time, pathways relating to barrier function such as tight- and adherens junction pathways were normalized in both therapies, including pathways relating to dysregulated tumor cell growth, such Hippo signaling (54). However, several pathways such as Ras, JAK/STAT, chemokine, and T-cell receptor signaling pathways were not entirely normalized by either therapy.

Our analysis provides a resource for discovering therapeutic targets for potential complete eradication of tumor cells and increased survival. Our findings support the hypothesis that checkpoint therapy in combination with antiangiogenic therapy may have an impact on the efficacy in patients with GBM by the alleviation of immunosuppression and improving cancer immunity through a normalized vasculature that is permissive for T-cell reinvigoration (55). Our study provides evidence that ICT efficiently improved survival in preclinical GBM mouse models when combined with vascular-targeted therapy. This is in line with findings in other cancer models which suggest that antiangiogenic therapy targeting the angiopoietin and/or VEGF pathway elicits antitumor immunity that is enhanced by anti–PD-1/PD-L1 therapy (18, 19). Although the GBM environment facilitates immune evasion of tumor cells, for example, via TGFβ, IL10, IDO1, Tregs, and MDSCs (56), the combination of immune checkpoint and vascular-targeting therapies alleviated immunosuppression in GBM. Multimodal therapies, including ICT, thus, may be the future for GBM disease management to invoke tumor-targeting immune cells. Study findings imply that ICT is more efficacious in neoadjuvant compared with adjuvant therapy in recurrent GBM, but no data for first-line therapy are currently available (45). Clinical studies have been initiated with the aim to improve the immune–vascular crosstalk for cancer immunotherapy and to overcome resistance that is driven by the immunosuppressive microenvironment (35). Stratification of patients that benefit from combination treatment could further improve therapy outcomes. Studies associated PD-L1 expression with the mesenchymal GBM subtype and NF1 deficiency (39). Patients with NF1 loss-of-function may thus benefit from ICT. Hypermutation at diagnosis or at recurrence was associated with CD8+ T-cell enrichment, indicating that those patients could be implemented for PD-1/PD-L1 immunotherapy (57). Nonetheless, because PD-L1 expression in GBM is dynamic and detection is challenging, serum Ang-2 might also be of value to predict response to immune checkpoint inhibition (58). Integrating the synergy of anti-VEGF/Ang-2 and anti–PD-1 therapy may precipitate immediate clinical impact. The RNA-seq datasets of isolated vessels from the current work provide molecular insight into the therapeutic benefit of triple therapy and serve as a resource for translational therapies when combined with human datasets.

J.P. Steinbach reports receiving a commercial research grant from Merck, Germany, and speakers bureau honoraria from AbbVie, Boehringer, Roche, Bristol-Myers Squibb, Medac, and UCB. M. Glas is an advisory board member for AbbVie and Novocure and reports receiving speakers bureau honoraria from Novocure, Novartis, Bayer, Medac, Merck, and Kyowa Kirin. U. Herrlinger reports receiving a commercial research grant from Roche and speakers bureau honoraria from Medac, Bayer, Noxxon, Novartis, Daiichi Sankyo, Bayer, Janssen, and Bristol-Myers Squibb. R. Büttner reports receiving speakers bureau honoraria from MSD, Bristol-Myers Squibb, and AstraZeneca, has ownership interest (including patents) in Targos Molecular Pathology, and is a consultant/advisory board member for MSD, Bristol-Myers Squibb, and AstraZeneca. O.M. Grauer is a consultant/advisory board member for Bristol-Myers-Squibb. G. Tabatabai reports receiving commercial research grants from Roche Diagnostics, Medac, and Novocure, and speakers bureau honoraria from Medac, Novocure, AbbVie, and Bayer. No potential conflicts of interest were disclosed by the other authors.

Conception and design: K.H. Plate, Y. Reiss

Development of methodology: M. Di Tacchio, J. Macas, J. Weissenberger, K. Devraj, K.H. Plate, Y. Reiss

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): M. Di Tacchio, J. Macas, J. Weissenberger, K. Sommer, O. Bähr, C. Senft, U. Herrlinger, D. Krex, A. Weyerbrock, M. Timmer, R. Goldbrunner, A.H. Scheel, R. Büttner, O.M. Grauer, G. Tabatabai, P.N. Harter, S. Günther, K. Devraj, Y. Reiss

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): M. Di Tacchio, J. Macas, J. Weissenberger, O. Bähr, V. Seifert, A.H. Scheel, R. Büttner, P.N. Harter, S. Günther, K. Devraj, K.H. Plate, Y. Reiss

Writing, review, and/or revision of the manuscript: M. Di Tacchio, J. Macas, J. Weissenberger, K. Sommer, O. Bähr, J.P. Steinbach, C. Senft, M. Glas, U. Herrlinger, D. Krex, A. Weyerbrock, M. Timmer, R. Goldbrunner, O.M. Grauer, P.N. Harter, K. Devraj, K.H. Plate, Y. Reiss

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): M. Di Tacchio, K. Sommer, O. Bähr, J.P. Steinbach, V. Seifert, M. Glas, M. Meinhardt, A. Weyerbrock, M. Deckert, J. Schittenhelm

Study supervision: K.H. Plate, Y. Reiss

K.H. Plate and Y. Reiss thank Dr. Angela Coxon (Amgen) for providing AMG386 (Master Agreement No. 2010537481). This work was supported by the Clinical Translation Program “Glioma” from the Frankfurt Cancer Institute (FCI), the Collaborative Research Center “Vascular differentiation and remodeling” (CRC/Transregio23, Project C1), the Cluster of Excellence 147 “Cardiopulmonary system” (ECCPS) from the German Research Council (DFG; to K.H. Plate and Y. Reiss), and grants from the German Cancer Consortium (DKTK, Partnersite Frankfurt/Mainz; to K.H. Plate and Y. Reiss). M. Di Tacchio was supported by the GO-IN Goethe International Post-Doc Programme of the Goethe University Frankfurt (PCOFUND-GA-2011-291776). The authors are grateful for experimental support from Sonja Thom and Maryam I. Khel.

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.

1.
Stupp
R
,
Mason
WP
,
van den Bent
MJ
,
Weller
M
,
Fisher
B
,
Taphoorn
MJB
, et al
Radiotherapy plus concomitant and adjuvant temozolomide for glioblastoma
.
N Engl J Med
2005
;
352
:
987
96
.
2.
Louis
DN
,
Ohgaki
H
,
Wiestler
OD
,
Cavenee
WK
.
WHO classification of tumours of the central nervous system
.
4th ed. Vol. 1. Lyon (France):
International Agency for Research on Cancer
; 
2016
.
3.
Stupp
R
,
Taillibert
S
,
Kanner
A
,
Read
W
,
Steinberg
D
,
Lhermitte
B
, et al
Effect of tumor-treating fields plus maintenance temozolomide vs maintenance temozolomide alone on survival in patients with glioblastoma: a randomized clinical trial
.
JAMA
2017
;
318
:
2306
16
.
4.
Preusser
M
,
Lim
M
,
Hafler
DA
,
Reardon
DA
,
Sampson
JH
. 
Prospects of immune checkpoint modulators in the treatment of glioblastoma
.
Nat Rev Neurol
2015
;
11
:
504
14
.
5.
Quail
DF
,
Joyce
JA
. 
The microenvironmental landscape of brain tumors
.
Cancer Cell
2017
;
31
:
326
41
.
6.
Reardon
DA
,
Wen
PY
,
Wucherpfennig
KW
,
Sampson
JH
. 
Immunomodulation for glioblastoma
.
Curr Opin Neurol
2017
;
30
:
361
9
.
7.
Wei
SC
,
Duffy
CR
,
Allison
JP
. 
Fundamental mechanisms of immune checkpoint blockade therapy
.
Cancer Discov
2018
;
8
:
1069
86
.
8.
Omuro
A
,
Vlahovic
G
,
Lim
M
,
Sahebjam
S
,
Baehring
J
,
Cloughesy
T
, et al
Nivolumab with or without ipilimumab in patients with recurrent glioblastoma: results from exploratory phase I cohorts of CheckMate 143
.
Neuro Oncol
2018
;
20
:
674
86
.
9.
Liebner
S
,
Dijkhuizen
RM
,
Reiss
Y
,
Plate
KH
,
Agalliu
D
,
Constantin
G
. 
Functional morphology of the blood-brain barrier in health and disease
.
Acta Neuropathol
2018
;
135
:
311
36
.
10.
Jackson
CM
,
Choi
J
,
Lim
M
. 
Mechanisms of immunotherapy resistance: lessons from glioblastoma
.
Nat Immunol
2019
;
20
:
1100
9
.
11.
Motz
GT
,
Coukos
G
. 
Deciphering and reversing tumor immune suppression
.
Immunity
2013
;
39
:
61
73
.
12.
Motz
GT
,
Santoro
SP
,
Wang
L-P
,
Garrabrant
T
,
Lastra
RR
,
Hagemann
IS
, et al
Tumor endothelium FasL establishes a selective immune barrier promoting tolerance in tumors
.
Nat Med
2014
;
20
:
607
15
.
13.
Scholz
A
,
Harter
PN
,
Cremer
S
,
Yalcin
BH
,
Gurnik
S
,
Yamaji
M
, et al
Endothelial cell-derived angiopoietin-2 is a therapeutic target in treatment-naive and bevacizumab-resistant glioblastoma
.
EMBO Mol Med
2016
;
8
:
39
57
.
14.
Kloepper
J
,
Riedemann
L
,
Amoozgar
Z
,
Seano
G
,
Susek
K
,
Yu
V
, et al
Ang-2/VEGF bispecific antibody reprograms macrophages and resident microglia to anti-tumor phenotype and prolongs glioblastoma survival
.
Proc Natl Acad Sci U S A
2016
;
113
:
4476
81
.
15.
Peterson
TE
,
Kirkpatrick
ND
,
Huang
Y
,
Farrar
CT
,
Marijt
KA
,
Kloepper
J
, et al
Dual inhibition of Ang-2 and VEGF receptors normalizes tumor vasculature and prolongs survival in glioblastoma by altering macrophages
.
Proc Natl Acad Sci U S A
2016
;
113
:
4470
5
.
16.
Saharinen
P
,
Eklund
L
,
Alitalo
K
. 
Therapeutic targeting of the angiopoietin-TIE pathway
.
Nat Rev Drug Discov
2017
;
16
:
635
61
.
17.
Scholz
A
,
Plate
KH
,
Reiss
Y
. 
Angiopoietin-2: a multifaceted cytokine that functions in both angiogenesis and inflammation
.
Ann N Y Acad Sci
2015
;
1347
:
45
51
.
18.
Schmittnaegel
M
,
Rigamonti
N
,
Kadioglu
E
,
Cassará
A
,
Wyser Rmili
C
,
Kiialainen
A
, et al
Dual angiopoietin-2 and VEGFA inhibition elicits antitumor immunity that is enhanced by PD-1 checkpoint blockade
.
Sci Transl Med
2017
;
9
. pii: eaak9670.
19.
Allen
E
,
Jabouille
A
,
Rivera
LB
,
Lodewijckx
I
,
Missiaen
R
,
Steri
V
, et al
Combined antiangiogenic and anti-PD-L1 therapy stimulates tumor immunity through HEV formation
.
Sci Transl Med
2017
;
9
. pii: eaak9679.
20.
Weissenberger
J
,
Steinbach
JP
,
Malin
G
,
Spada
S
,
Rülicke
T
,
Aguzzi
A
. 
Development and malignant progression of astrocytomas in GFAP-v-src transgenic mice
.
Oncogene
1997
;
14
:
2005
13
.
21.
Smilowitz
HM
,
Weissenberger
J
,
Weis
J
,
Brown
JD
,
O'Neill
RJ
,
Laissue
JA
. 
Orthotopic transplantation of v-src-expressing glioma cell lines into immunocompetent mice: establishment of a new transplantable in vivo model for malignant glioma
.
J Neurosurg
2007
;
106
:
652
9
.
22.
Coxon
A
,
Bready
J
,
Min
H
,
Kaufman
S
,
Leal
J
,
Yu
D
, et al
Context-dependent role of angiopoietin-1 inhibition in the suppression of angiogenesis and tumor growth: implications for AMG 386, an angiopoietin-1/2-neutralizing peptibody
.
Mol Cancer Ther
2010
;
9
:
2641
51
.
23.
Holash
J
,
Davis
S
,
Papadopoulos
N
,
Croll
SD
,
Ho
L
,
Russell
M
, et al
VEGF-Trap: a VEGF blocker with potent antitumor effects
.
Proc Natl Acad Sci U S A
2002
;
99
:
11393
8
.
24.
Devraj
K
,
Poznanovic
S
,
Spahn
C
,
Schwall
G
,
Harter
PN
,
Mittelbronn
M
, et al
BACE-1 is expressed in the blood-brain barrier endothelium and is upregulated in a murine model of Alzheimer's disease
.
J Cereb Blood Flow Metab
2016
;
36
:
1281
94
.
25.
Gurnik
S
,
Devraj
K
,
Macas
J
,
Yamaji
M
,
Starke
J
,
Scholz
A
, et al
Angiopoietin-2-induced blood-brain barrier compromise and increased stroke size are rescued by VE-PTP-dependent restoration of Tie2 signaling
.
Acta Neuropathol
2016
;
131
:
753
73
.
26.
Davis
MPA
,
van Dongen
S
,
Abreu-Goodger
C
,
Bartonicek
N
,
Enright
AJ
. 
Kraken: a set of tools for quality control and analysis of high-throughput sequence data
.
Methods
2013
;
63
:
41
9
.
27.
Dobin
A
,
Davis
CA
,
Schlesinger
F
,
Drenkow
J
,
Zaleski
C
,
Jha
S
, et al
STAR: ultrafast universal RNA-seq aligner
.
Bioinformatics
2013
;
29
:
15
21
.
28.
Liao
Y
,
Smyth
GK
,
Shi
W
. 
featureCounts: an efficient general purpose program for assigning sequence reads to genomic features
.
Bioinformatics
2014
;
30
:
923
30
.
29.
Love
MI
,
Huber
W
,
Anders
S
. 
Moderated estimation of fold change and dispersion for RNA-seq data with DESeq2
.
Genome Biol
2014
;
15
:
550
.
30.
Xie
C
,
Mao
X
,
Huang
J
,
Ding
Y
,
Wu
J
,
Dong
S
, et al
KOBAS 2.0: a web server for annotation and identification of enriched pathways and diseases
.
Nucleic Acids Res
2011
;
39
:
W316
22
.
31.
Scheel
AH
,
Dietel
M
,
Heukamp
LC
,
Jöhrens
K
,
Kirchner
T
,
Reu
S
, et al
[Predictive PD-L1 immunohistochemistry for non-small cell lung cancer: current state of the art and experiences of the first German harmonization study]
.
Pathologe
2016
;
37
:
557
67
.
32.
Taube
JM
,
Anders
RA
,
Young
GD
,
Xu
H
,
Sharma
R
,
McMiller
TL
, et al
Colocalization of inflammatory response with B7-h1 expression in human melanocytic lesions supports an adaptive resistance mechanism of immune escape
.
Sci Transl Med
2012
;
4
:
127ra37
.
33.
Missiaen
R
,
Mazzone
M
,
Bergers
G
. 
The reciprocal function and regulation of tumor vessels and immune cells offers new therapeutic opportunities in cancer
.
Semin Cancer Biol
2018
;
52
:
107
16
.
34.
Voron
T
,
Colussi
O
,
Marcheteau
E
,
Pernot
S
,
Nizard
M
,
Pointet
A-L
, et al
VEGF-A modulates expression of inhibitory checkpoints on CD8+ T cells in tumors
.
J Exp Med
2015
;
212
:
139
48
.
35.
Rahma
OE
,
Hodi
FS
. 
The intersection between tumor angiogenesis and immune suppression
.
Clin Cancer Res
2019
;
25
:
5449
57
.
36.
Filley
AC
,
Henriquez
M
,
Dey
M
. 
Recurrent glioma clinical trial, CheckMate-143: the game is not over yet
.
Oncotarget
2017
;
8
:
91779
94
.
37.
Berghoff
AS
,
Kiesel
B
,
Widhalm
G
,
Rajky
O
,
Ricken
G
,
Wöhrer
A
, et al
Programmed death ligand 1 expression and tumor-infiltrating lymphocytes in glioblastoma
.
Neuro Oncol
2015
;
17
:
1064
75
.
38.
Nduom
EK
,
Wei
J
,
Yaghi
NK
,
Huang
N
,
Kong
L-Y
,
Gabrusiewicz
K
, et al
PD-L1 expression and prognostic impact in glioblastoma
.
Neuro Oncol
2016
;
18
:
195
205
.
39.
Heiland
DH
,
Haaker
G
,
Delev
D
,
Mercas
B
,
Masalha
W
,
Heynckes
S
, et al
Comprehensive analysis of PD-L1 expression in glioblastoma multiforme
.
Oncotarget
2017
;
8
:
42214
25
.
40.
Gordon
SR
,
Maute
RL
,
Dulken
BW
,
Hutter
G
,
George
BM
,
McCracken
MN
, et al
PD-1 expression by tumour-associated macrophages inhibits phagocytosis and tumour immunity
.
Nature
2017
;
545
:
495
9
.
41.
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
.
42.
Jain
RK
. 
Antiangiogenesis strategies revisited: from starving tumors to alleviating hypoxia
.
Cancer Cell
2014
;
26
:
605
22
.
43.
Reardon
DA
,
Gokhale
PC
,
Klein
SR
,
Ligon
KL
,
Rodig
SJ
,
Ramkissoon
SH
, et al
Glioblastoma eradication following immune checkpoint blockade in an orthotopic, immunocompetent model
.
Cancer Immunol Res
2016
;
4
:
124
35
.
44.
Antonios
JP
,
Soto
H
,
Everson
RG
,
Orpilla
J
,
Moughon
D
,
Shin
N
, et al
PD-1 blockade enhances the vaccination-induced immune response in glioma
.
JCI Insight
2016
;
1
. pii:e87059.
45.
Cloughesy
TF
,
Mochizuki
AY
,
Orpilla
JR
,
Hugo
W
,
Lee
AH
,
Davidson
TB
, et al
Neoadjuvant anti-PD-1 immunotherapy promotes a survival benefit with intratumoral and systemic immune responses in recurrent glioblastoma
.
Nat Med
2019
;
25
:
477
86
.
46.
Juneja
VR
,
McGuire
KA
,
Manguso
RT
,
LaFleur
MW
,
Collins
N
,
Haining
WN
, et al
PD-L1 on tumor cells is sufficient for immune evasion in immunogenic tumors and inhibits CD8 T cell cytotoxicity
.
J Exp Med
2017
;
214
:
895
904
.
47.
Noman
MZ
,
Desantis
G
,
Janji
B
,
Hasmim
M
,
Karray
S
,
Dessen
P
, et al
PD-L1 is a novel direct target of HIF-1α, and its blockade under hypoxia enhanced MDSC-mediated T cell activation
.
J Exp Med
2014
;
211
:
781
90
.
48.
Dubinski
D
,
Wölfer
J
,
Hasselblatt
M
,
Schneider-Hohendorf
T
,
Bogdahn
U
,
Stummer
W
, et al
CD4+ T effector memory cell dysfunction is associated with the accumulation of granulocytic myeloid-derived suppressor cells in glioblastoma patients
.
Neuro Oncol
2016
;
18
:
807
18
.
49.
Park
J-S
,
Kim
I-K
,
Han
S
,
Park
I
,
Kim
C
,
Bae
J
, et al
Normalization of tumor vessels by Tie2 activation and Ang2 inhibition enhances drug delivery and produces a favorable tumor microenvironment
.
Cancer Cell
2016
;
30
:
953
67
.
50.
Frye
M
,
Dierkes
M
,
Küppers
V
,
Vockel
M
,
Tomm
J
,
Zeuschner
D
, et al
Interfering with VE-PTP stabilizes endothelial junctions in vivo via Tie-2 in the absence of VE-cadherin
.
J Exp Med
2015
;
212
:
2267
87
.
51.
Shen
J
,
Frye
M
,
Lee
BL
,
Reinardy
JL
,
McClung
JM
,
Ding
K
, et al
Targeting VE-PTP activates TIE2 and stabilizes the ocular vasculature
.
J Clin Invest
2014
;
124
:
4564
76
.
52.
Goel
S
,
Gupta
N
,
Walcott
BP
,
Snuderl
M
,
Kesler
CT
,
Kirkpatrick
ND
, et al
Effects of vascular-endothelial protein tyrosine phosphatase inhibition on breast cancer vasculature and metastatic progression
.
J Natl Cancer Inst
2013
;
105
:
1188
201
.
53.
Dubinski
D
,
Hattingen
E
,
Senft
C
,
Seifert
V
,
Peters
KG
,
Reiss
Y
, et al
Controversial roles for dexamethasone in glioblastoma - opportunities for novel vascular targeting therapies
.
J Cereb Blood Flow Metab
2019
;
39
:
1460
8
.
54.
Kim
J
,
Kim
YH
,
Kim
J
,
Park
DY
,
Bae
H
,
Lee
D-H
, et al
YAP/TAZ regulates sprouting angiogenesis and vascular barrier maturation
.
J Clin Invest
2017
;
127
:
3441
61
.
55.
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
.
56.
Reardon
DA
,
Wucherpfennig
K
,
Chiocca
EA
. 
Immunotherapy for glioblastoma: on the sidelines or in the game?
Discov Med
2017
;
24
:
201
8
.
57.
Wang
Q
,
Hu
B
,
Hu
X
,
Kim
H
,
Squatrito
M
,
Scarpace
L
, et al
Tumor evolution of glioma-intrinsic gene expression subtypes associates with immunological changes in the microenvironment
.
Cancer Cell
2017
;
32
:
42
6
.
58.
Wu
X
,
Giobbie-Hurder
A
,
Liao
X
,
Connelly
C
,
Connolly
EM
,
Li
J
, et al
Angiopoietin-2 as a biomarker and target for immune checkpoint therapy
.
Cancer Immunol Res
2017
;
5
:
17
28
.