Single cell RNA sequencing datasets of ECs from GBM and paired non-malignant control brain tissue were downloaded from the Gene Expression Omnibus (GEO) database (GSE162631). Information for cells and samples were obtained in the previous study (4). Only ECs from peripheral endothelial cell type I (Pe1), tumor core endothelial cell type I (Co1) and tumor core endothelial cell type II (Co2), which were represented in four patients, in the original study were selected for downstream analysis. 416 ECs from Pe1 cluster were considered as non-malignant brain endothelial cells. 634 ECs from Co1 and Co2 clusters were considered as tumor endothelial cells.

The expression of ETS1 in different human GBM anatomic regions, including leading edge region, infiltrating tumor region, cellular tumor core region, pseudopalisading necrosis region and microvascular proliferation region, was obtained from Ivy GAP database (http://glioblastoma.alleninstitute.org).

Ethical permission for using patient samples was granted by the Ethics Committee of Hunan University of Medicine (HUM-HE-2019-015). Glioblastoma samples were collected retrospectively at the First Affiliated Hospital of Hunan University of Medicine. For sample details, see Supplementary Table S1.

Murine brain endothelial cell (bEND.3) was purchased from Chinese Academy of Sciences Cell Bank. The CT-2A mouse glioma cell line was a gift from Dr. L. Zhang, Shaanxi Normal University. bEND.3 and CT-2A cells were cultured in Dulbecco’s modified Eagle’s medium (DMEM) (ThermoFisher, 10566016) supplemented with 10% fetal calf serum (FCS) (ThermoFisher, 10091155).

For Ets1 knockdown, bEND.3 cells were seeded and incubated with control siRNA or siRNA to Ets1 at concentration of 10 nmol/L using siRNA-mate (GenePharma) according to the manufacturer’s manual. The sequence of siRNA for Ets1 targeting was listed as following: control siRNA (siNT), sense: UUC​UCC​GAA​CGU​GUC​ACG​UTT, anti-sense: ACG​UGA​CAC​GUU​CGG​AGA​ATT; siEts1-331, sense: GGA​CAA​GCC​UGU​CAU​UCC​UTT, anti-sense: AGG​AAU​GAC​AGG​CUU​GUC​CTT; siEts1-946, sense: GGA​AUU​ACU​CAC​UGA​UAA​GTT, anti-sense: CUU​AUC​AGU​GAG​UAA​UUC​CTT. Experiments were performed on day 2–3 after siRNA transfection. Ets1/Ets1 knockdown efficiency was determined by qPCR and western blot 48 h after transfection.

Cell proliferations were determined using Cell Counting Kit-8 (CCK8, Beyotime Biotechnology). In brief, 200 μl of cell suspension (2.5 × 10⁴ cells/ml) was placed into each well of Primaria 96-well plates (BD Biosciences), performing triplicates for each time-point. Three wells with media alone were used for determination of background in each experiment. On day 1 and day 7, 20 μl CCK8 was added to each well, the plate was incubated at 37°C for 2 more hours. The number of living cells were indicated by the absorbance at 450 nm detected by INFINITE M NANO absorbance plate reader (TECAN). The cell density on day 7 was normalized to that on day 1 shown as the proliferation index. The experiment was repeated four times.

To determine the effect of VEGFA, TNFα or TGFβ2 on Ets1 expression in bEND.3 cells, the bEND.3 cells were seeded on the 12-well plates (8 × 10⁴ cells/well). At the second day, the cells were starved in DMEM with 1% FCS for overnight, followed by VEGFA (50 ng/ml; Peprotech), TNFα (2 ng/ml; Peprotech) or TGFβ2 (10 ng/ml, Peprotech) stimulations for 48 h. The experiment was repeated three times with 3 samples per group in each experiment. To determine whether Ets1 was up-regulated by tumors through TGFβ signaling in vitro, bEND.3 cell were seeded on the 12-well plates (8 × 10⁴ cells/well). After starvation in DMEM with 1% FCS for overnight, the bEND.3 cells were cultured with conditioned medium from CT-2A glioma cells for 48 h. The conditioned medium was collected from confluent CT-2A glioma cells culture, and then filtered through 0.2 μm Nalgene Syringe filter (ThermoFisher, 720-1320), followed by pre-treatment with 50 μg/ml anti-TGFβ neutralizing antibody (ThermoFisher, 16-9243-85) or control antibody (R&D system, MAB002) overnight (12 h), before incubation with bEND.3 cells. The experiment was repeated three times with 3 samples per group in each experiment.

A scratch wound was applied on confluent cell monolayers using a 200 ul tip. Pictures were taken at 0 (T0) and 24 h (T24) post-scratching using a Primovert iLED microscope (Zeiss, Germany). Migration was measured with the fiji/ImageJ software and is expressed as % wound closure (gap area at T0 minus gap area at T24 in % of gap area at T0). The experiment was repeated three times with 4 samples per group in each experiment.

Transwell inserts with 8 µm pore size (CLS3422-48 EA, Corning) for 24-well plates were used and 5 × 10⁴ cell were seeded in the upper chamber in medium without fetal bovine serum (Biological Industries). Regular culture medium containing fetal bovine serum was added to the lower chamber. After 24 h, cells were fixed with 4% formaldehyde, and permeabilized with methanol. Non-migrated cells were removed from the upper surface of the membrane and the membrane was cut off and stained with Hoechst 33258 (Sigma-Aldrich). To quantify the cells that had migrated through the membrane, pictures were taken at four different fields using an Axio Imager upright microscope (Zeiss, Germany). The experiment was repeated three times with 8 samples per group in each experiment.

Total RNA was extracted with RNeasy Mini Kit (Qiagen, 74104). cDNA from total RNA was synthesized using random hexamer primers and SuperScript III reverse transcriptase (ThermoFisher, 18080093) according to the manufacturer’s instructions. qPCR was performed on Thermal Cycler iQ5 multicolor Real-Time PCR detection system (Bio-Rad) using TB Green Premix Ex TaqTMII (Takara, RR820A) with 0.25 μM reverse and forward primer per well. Gene expression was normalized to the house-keeping gene hypoxanthine guanine phosphoribosyl transferase (Hprt) according to the following formula: relative expression of gene X = 2^{-(CT Hprt− CT gene X)}. The sequences of primers for qPCR were listed as following: Angpt2, forward: CCT​CGA​CTA​CGA​CGA​CTC​AGT, reverse: TCT​GCA​CCA​CAT​TCT​GTT​GGA; Mcam, forward: CCC​AAA​CTG​GTG​TGC​GTC​TT, reverse: GGA​AAA​TCA​GTA​TCT​GCC​TCT​CC; Sox4: forward: CGG​CTG​CAT​CGT​TCT​CTC​C, reverse: CGC​TTC​ACT​TTC​TTG​TCG​GC; Vegfa, forward: CTG​CCG​TCC​GAT​TGA​GAC​C, reverse: CCC​CTC​CTT​GTA​CCA​CTG​TC; Kdr, forward: TTT​GGC​AAA​TAC​AAC​CCT​TCA​GA, reverse: GCA​GAA​GAT​ACT​GTC​ACC​ACC; Itga1, forward: CCT​TCC​CTC​GGA​TGT​GAG​TCA, reverse: AAG​TTC​TCC​CCG​TAT​GGT​AAG​A; Notch4, forward: CTC​TTG​CCA​CTC​AAT​TTC​CCT, reverse: TTG​CAG​AGT​TGG​GTA​TCC​CTG.

Cells were lysed in Pierce LDS sample buffer (ThermoFisher, 84788) and protein concentration was determined using BCA protein assay kit (BCA, Beyotime Biotechnology). 10 μg of protein was loaded on the gel. Samples were separated on NuPAGE 4–12% Bis-Tris gels (ThermoFisher, NP0335BOX) using MOPS SDS running buffer (ThermoFisher, NP0050), and then transferred to a Hybond-C Extra filter (GE Healthcare). Membranes were blocked with 5% milk in tris-buffered saline plus 0.01% Tween, and incubated with primary antibodies (anti-Ets1 antibody: abcam, ab220361; anti-β-actin antibody: abcam, ab8227) diluted in blocking solution overnight at 4°C. Then, membranes were incubated with horseradish peroxidase (HRP)-conjugated secondary antibodies (abcam, ab205718), and detected using Pierce ECL plus substrate (ThermoFisher, 32134).

Mouse studies were approved by the Animal Experiment Ethical Committee of Hunan University of Medicine (HUM-AE-2018-113). Six-week-old C57BL/6 mice were purchased from Vitalriver. During injection, mice were anesthetized with 2.5% isoflurane. CT-2A cells (5 × 10⁴) in 2 μl Dulbecco’s phosphate-buffered saline (DPBS) were injected into subventrical zone (coordinates: 1 mm anterior to bregma, 1.5 mm from the mid-line, and 2.7 mm below the cranial surface) using the Hamilton microtiter syringe. Pre-warmed pads were employed for mice until fully recovered. Tumor-bearing mice were administered daily with galunisertib (HY-13226, MedChemExpress) by oral gavage (150 mg/kg of galunisertib in 0.5% methylcellulose/Tween 80) starting from 7 days after tumor inoculation. 10 days after first treatment, tumor-bearing mice were sacrificed, and brains were collected for further analysis. The experiment was repeated two times with at least 10 mice per group in each experiment.

Immunohistochemistry (IHC) staining of ETS1, CD31 was performed on 6 μm paraffin sections of mouse brain or human tumor. Sections were deparaffinized and dehydrated prior to antigen retrieval followed by blocking with 3% bovine serum albumin (BSA) (Sigma-Aldrich, A7906) in phosphate-buffered saline (PBS). Then the sections were incubated with primary antibodies against Ets1 (Abcam, ab220361), CD31 (Dianova, DIA-310) followed by incubation with biotinylated secondary antibody (Vector Laboratories, BA-1000, anti-rabbit IgG; BA-9400, anti-rat IgG) and streptavidin conjugated to peroxidase (Vector Laboratories, SA-5014). The staining was detected with DAB substrate (Vector Laboratories, SK-4100) according to manufacturer’s instruction. The hematoxylin counterstaining was used to visualize nuclei. The images were acquired using NIS software (Nikon). The Ets1 staining was semi-quantified according to the fraction of positively stained vessels on a scale from 0 to 2 (no vessel stained, minority vessel stained, majority vessel stained). The vascular areas were quantified according to the area stained positive for CD31 using Image J software. The data is presented as CD31 positive area in a given area (μm² mm⁻²).

Tumor vascular space, indicated by the mean diameters of the vessels, were analyzed on CD31 immunohistochemical stained paraffin sections using eyepiece grid as previously described (36,37). In brief, the eyepiece grid with 10 × 10 squares (0.25 mm × 0.25 mm) was placed at tumor area. The number of vessels (Q_ves) and the number of test points hitting vessels (P_ves) were counted in the counting frame. 10–25 frames were quantified from each tumor depending on the tumor size. The mean vascular diameters were calculated based on the following formulation (37): d ( mean   s e c t i o n   d i a m e t e r   o f   v e s s e l s ; m m ) = 2 × ∑ P _ ves ∑ Q _ ves × A ( frame ) 2 π × P ( pcg ) A (frame) (area of one counting frame) = 0.0625 mm²; P (pcg) (number of test points in one point-counting grid) = 121.10 individual tumors were analyzed per group.

Statistical analysis was performed using GraphPad Prism software and R software. The Mann-Whitney test or t test was performed to determine statistically significant differences in the experiments with two groups. One-way ANOVA with Tukey’s multiple comparisons test was performed to determine statistically significant differences in the experiments with more than two groups. All statistical tests were two-sided.

To investigate the expression of ETS1 in GBM ECs, we reanalyzed a recently published dataset of single cell RNA-seq (scRNA-seq) of ECs from tumors and paired non-malignant brain tissue in 4 GBM patients (4). ETS1 was upregulated in ECs from tumor compared to non-malignant brain ECs (Student’s t test; p < 0.0001) (Figure 1A). The expression of ETS1 in distinct anatomical locations of 34 GBM samples was analyzed by using Ivy GAP database (http://glioblastoma.alleninstitute.org/) (38), which documented transcriptome from microdissected human GBM anatomic regions, including leading edge, infiltrating tumor region, cellular tumor core region, microvascular proliferation region and pseudopalisading necrosis region. In accordance with upregulation of ETS1 in tumor ECs, ETS1 expression was higher in microvascular proliferation region where tumor ECs were enriched due to active angiogenesis (Figure 1B). Upregulation of ETS1 in tumor vasculature was further confirmed in protein level by immunohistochemical staining of our in-house samples including 18 GBM tumors and paired control brain tissue (Figure 1C). Taken together, these results indicate that ETS1 is upregulated in GBM ECs.

Regulation of ETS1 expression in brain ECs is still unknown. To uncover the signal pathway mediating ETS1 upregulation, we analyzed the expression of Ets1 in bEND.3 cells upon stimulation of VEGFA, TNFα and TGFβ2, which can increase ETS1/Ets1 expression in human umbilical vein endothelial cells (HUVECs) or renal cells (39–41). Neither TNFα nor VEGFA could increase ETS1 expression in bEND.3 cells (Figure 2A). Notably, TGFβ treatment could upregulate ETS1 expression (Figure 2A), indicating a key role of TGFβ signaling on mediating ETS1 induction in brain ECs.

To determine whether ETS1 is upregulated in GBM ECs via TGFβ-dependent manner, bEND.3 cells were stimulated with CT-2A glioma cells conditioned medium with neutralizing antibody against TGFβ or control antibody, after which the expression of Ets1 was analyzed by qPCR. We found that conditioned medium from GBM tumor cells was sufficient to upregulate Ets1 expression in bEND.3 cells (Figure 2B), suggesting a direct effect of tumor cell on Ets1 expression in ECs. Notably, neutralizing TGFβ antibody treatment attenuated tumor conditioned medium induced upregulation of ETS1 expression in bEND.3 (Figure 2B). To evaluate the role of TGFβ signaling on Ets1 expression in tumor ECs in vivo, we employed CT-2A syngeneic orthotopic glioblastoma model. CT-2A tumor-bearing mice were treated with galunisertib, a potent TGFβ receptor kinase inhibitor (42,43). In accordance with in vitro findings, galunisertib treatment decreased Ets1 level in tumor vessels in vivo (Figure 2C). These results indicate that Ets1 upregulation in GBM ECs is dependent on TGFβ signaling.

Active angiogenesis is a key feature of GBM vessels (28). To investigate the role of Ets1 on angiogenesis, we determined whether Ets1 knockdown affects the migratory capacity and proliferation of bEND.3 cells. We used RNA interference to knockdown the expression of Ets1 in bEND.3 in vitro. Transfection of bEND.3 cells with siRNA to Ets1 resulted in an efficient downregulation of Ets1 in both RNA and protein levels (Figures 3A,B). Ets1 knockdown led to a significant reduction of bEND.3 cell migration compared to control cells (Figures 3C,D). These results were further supported by transwell migration assay showing that the migration of bEND.3 cells through membrane was inhibited by Ets1 knockdown (Figures 3E,F). In addition, knockdown of Ets1 inhibited proliferation of bEND.3 cells (Figure 3G). Taken together, these results suggest that Ets1 may regulate brain EC migration and proliferation.

To investigate the role of ETS1 on expression of genes associated with vascular abnormality in brain ECs. qPCR analysis revealed that Ets1 knockdown in brain ECs suppressed the expression of 6 out of 7 selected vascular abnormality associated genes (Figures 3H–N), including Mcam, Angpt2, Kdr, Vegfa, Sox4 and Itga1, suggesting a potential role of Ets1 on vascular abnormality.

We next set out whether TGFβ signaling, which is the upstream of ETS1 signaling (Figure 2C), affects the tumor angiogenesis and vascular abnormality. Quantification of vessel density, based on CD31 positive staining area, revealed decreased vascular area upon TGFβ inhibition (4A-4B). In addition, the space occupied by the blood vessel in CT-2A tumors, indicated by mean vascular diameter, was reduced in the tumors upon galunisertib treatment according to stereological quantification (36), (Figures 4A,C), indicative of an effect of TGFβ inhibition on angiogenesis and vascular normalization. We additionally analyzed the expression of a panel genes (Angpt2, Sox4, Vegfa, Kdr, Itga1, Mcam, Notch4 and Ets1) associated with vascular abnormality in ECs using RNA extracted from total glioma tumor tissues from control and treated mice. We found that 7 out of 8 selected genes including Angpt2, Sox4, Vegfa, Kdr, Mcam, Notch4 and Ets1 were downregulated in tumors upon galunisertib treatment (Figure 4D). Taken together, these results indicate that targeting TGFβ signaling, the upstream of ETS1, could suppress tumor angiogenesis and downregulate vascular abnormality associated genes.