The murine brain endothelioma cell line bEnd3 and the murine immortalized heart endothelial cell line H5V were cultured in Dulbecco’s Modified Eagle’s Medium (DMEM)-Glutamax comprising 4500 mg/L glucose and 110 mg/L sodium pyruvate, and supplemented with 10% heat-inactivated fetal bovine serum (FBS; Gibco, Thermo Fisher Scientific), 100 U/mL penicillin and 100 µg/mL streptomycin sulfate. The melanoma cell line B16-F10 was grown as a monolayer in DMEM-Glutamax supplemented with 10% FBS, 100 U/mL penicillin and 100 µg/mL streptomycin sulfate. Cells were passaged twice weekly to be maintained under exponential growth conditions and were routinely tested for mycoplasma contamination. The Phoenix Eco retroviral ecotropic packaging cell line, derived from immortalized normal human embryonic kidney cells, was maintained in RPMI 1640-Glutamax medium supplemented with 10% FBS, 100 U/mL penicillin and 100 µg/mL streptomycin sulfate. Primary murine T cells were cultured in RPMI 1640-Glutamax medium supplemented with 10% FBS, 100 U/mL penicillin, 100 µg/mL streptomycin sulfate, 1 mM sodium pyruvate, 50 µM β-mercaptoethanol and 10 mM non-essential amino acids (referred to as murine T-cell culture medium). T cells were activated 24 hours prior to transduction in murine T-cell culture medium supplemented with 50IU/mL of human IL-2, and expanded from day 2 after transduction onwards in murine T-cell culture medium supplemented with 10 ng/ml of both hIL-7 and hIL-15 as previously described.40

The retroviral murine stem cell virus (MSCV)-based splice-gag vector (pMSGV), comprising the MSCV long terminal repeat, was used as the backbone for the anti-VEGFR-2 CAR construct. The 2G anti-VEGFR-2 CAR (DC101-28z) construct comprising the anti-murine VEGFR-2 scFv DC101, the CD8α hinge (H) and transmembrane region, followed by the endodomains of CD28 and CD3ζ, was kindly provided by Dr Steven A. Rosenberg (National Cancer Institute). As a control, a retroviral vector pMSGV encoding the marker Thy1.1 was built.

Retrovirus was produced and primary murine T cells transduced and expanded as previously described.40 CAR-T cells were maintained at a cell density of 0.5–1×10⁶ cells/mL to ensure optimal expansion.

The three-dimensional (3D) structure of the VEGFR-2:DC101 complex was obtained by homology modeling using the experimental structure of ramucirumab in complex with domain 3 of VEGFR-2 (PDB ID 3S3643) at 3.2 Å resolution. The sequence alignment between the DC101 and ramucirumab sequences was performed using the program MUSCLE.44 Based on this sequence alignment, 1000 structural models were generated using the MODELLER program45 and ranked according to the DOPE energy score.46 The top-ranked model according to DOPE was retained as the final model. Structural superimposition as well as molecular visualization and analysis were performed using the UCSF Chimera software.47

VEGFR-2-28-z CAR expression was detected on T cells 7 days post-transduction by incubation with soluble recombinant murine (m)VEGFR-2-hIgG-Fc fusion protein (R&D Systems) followed by staining with PE-labeled goat anti-IgG Fc (eBioscience). VEGFR-2 expression by murine endothelial cell lines was assessed by cell-surface staining with rat anti-VEGFR-2 antibody (Clone Avas12, BioLegend), and adhesion molecule surface expression was assessed with anti-ICAM-1 (Clone YN1/1.7.4, BioLegend) and anti-VCAM-1 (Clone MVCAM.A, BioLegend) antibodies. To discriminate dead cells, staining with 7-aminoactinomycin D (7-AAD, BioLegend) was performed. Data were acquired on a BD flow cytometer and analyzed using FlowJo software (Tree Star). For the ex vivo phenotypic analysis of CAR-T cells (from the spleens and tumors of differently treated mice), the following antibodies were used: CD45 (30F/11), CD3ε (145-2C11), CD8α (53-6.7), CD45.1 (A20), PD-1 (29F.1A12), LAG-3 (C9B7W), CTLA-4 (UC10-4B9) and Ki67 (SolA15). Antibodies were purchased from eBioscience and BioLegend or produced in-house from hybridomas by the flow cytometry platform. To exclude dead cells, the Live/dead fixable Aqua Dead cell stain kit was used according to the manufacturer’s instructions (Molecular Probes, Life Technologies).

CAR-T cells were incubated with 10 µg/mL recombinant mVEGFR-2-hIgG-Fc fusion protein (R&D Systems) in the presence of control protein or soluble murine VEGF-A (VEGF-164, BioLegend), VEGF-C (BioLegend), or VEGF-D (R&D Systems), at a VEGF/VEGFR-2 molar ratio of 3:1 in a total volume of 10 μL. Alternatively, VEGF-A was incubated with anti-VEGF-A antibody (Clone B20‐4.1.1; a kind gift from Genentech, San Francisco, California, USA) at a VEGF-A:anti-VEGF-A antibody molar ratio of 1:3 for 30 min at 37°C prior its addition to CAR-T cells. The cells were then washed and stained to evaluate mVEGFR-2-hIgG-Fc binding levels on the CAR-T cells as described above.

Adhesion assays were performed using the Vybrant Cell Adhesion Assay Kit (Thermo Fisher Scientific) according to the manufacturer’s instructions. Briefly, CAR-T cells were washed twice with phosphate-buffered saline (PBS) and then resuspended in RPMI at 5×10⁶ cells/mL. Calcein AM stock solution was added at a final concentration of 5 µM. Samples were mixed well and incubated at 37°C for 30 min. After labeling with calcein AM, CAR-T cells were washed twice and resuspended in plain RPMI at 1×10⁶ cells/mL. Subsequently, 100 µL of the calcein-labeled CAR-T cell suspension (1×10⁵ cells unless otherwise indicated) was added to the prepared microplate wells containing confluent endothelial cell monolayers and incubated at 37°C for 90 min. The plates were washed four times to remove non-adherent calcein-labeled cells and then 200 µL of PBS was added to each well and fluorescence was measured. Prior to T-cell addition, target cells (where indicated) were exposed to bovine serum albumin (BSA) or VEGF-A or VEGF-A/anti-VEGF-A antibody complexes.

Cytokine release was measured following: (1) exposure of T cells to 96-well plates coated with 1 µg/mL VEGFR-2 or control protein (BSA), (2) co-culture of 0.5×10⁵ T cells with 0.5×10⁵ target cells in 96-well flat bottom plates, and (3) stimulation with anti-(α)CD3/CD28 dynabeads at a bead:cell ratio of 2:1. For the assays with immobilized antigen, the 96-well plates were treated with BSA, VEGF-A, VEGF-C, VEGF-D or VEGF-A/anti-VEGF-A antibody complexes at the indicated molar ratios for 1 hour at 37°C prior to T-cell addition. For the co-culture assays, the target cells were treated with 0.17 µg/mL VEGF-A or VEGF-A/VEGF-A antibody complexes at molar ratio 1:3 or BSA for 1 hour at 37°C prior to T-cell addition. For the assays of T cell stimulation with αCD3/CD28 dynabeads, VEGF-A, VEGF-C, or VEGF-D were added at a concentration of 0.17 or 0.5 µg/mL. At 20–24 hours post stimulation with immobilized antigen or target cells or αCD3/CD28 dynabeads, the supernatants were assayed for the presence of interferon-γ (IFN-γ) by commercial ELISA Kit (BioLegend). The levels of IL-2, tumor necrosis factor-α (TNF-α) and macrophage inflammatory protein-1 alpha (MIP-1α) were quantified using the Cytokine Bead Array (CBA) as described by the manufacturer (BD Biosciences).

CD45.1⁺ congenic C57BL/6 mice were bred in the Epalinges UNIL animal facility. Female CD45.2⁺ C57BL/6 mice aged 6–8 weeks were purchased from Harlan (Harlan, Netherlands). All in vivo experiments were conducted in accordance with approval from the Service of Consumer and Veterinary Affairs (SCAV) of the Canton of Vaud.

B16 tumor cells were harvested with 0.05% trypsin, washed, and resuspended in PBS for injection. 1×10⁵ tumor cells were injected subcutaneously in the flank of C57BL/6 mice, aged 8–12 weeks. Nine days later (average tumor volume 20–40 mm³), the mice received 5 Gy of sublethal total body irradiation and grouped (n≥5 mice/group) for comparative average tumor volumes. On days 10 and 13, the mice were treated with intravenous injection of 8×10⁶ CD45.1⁺ CAR-T cells or control Thy1.1-transduced T cells. In some experiments, anti-VEGF-A antibody (40 µg) was coadministered on the days of ACT and then every 2–4 days. Mice were carefully monitored and tumor length (L; greatest longitudinal measurement) and width (W; greatest transverse measurement) were measured by caliper every 2–3 days. Tumor volumes (V) were calculated using the formula: V=(L×W²)/2. The average tumor volumes/group are plotted ±SEM. Mice were sacrificed once tumors reached 1000 mm³, or if they became distressed or moribund, according to regulations.

Tumors were embedded and frozen in Tissue-Tek OCT (Sakura Finetek). Cryostat sections (5–8 µm) were collected on Superfrost/Plus glass slides (Thermo Fisher Scientific) and then air-dried overnight before being stored at −20°C until further analysis. The day of the staining slides were warmed, rehydrated with PBS for 5 min at room temperature (RT), fixed in 10% neutral buffered formalin for 10 min at 4°C and washed three times in PBS. Sections were permeabilized with 0.5% Triton X-100 in PBS for 10 min at RT, washed (4×5 min at RT) and blocked using PBS with 1% BSA and 5% FBS for 2 hours at RT. Immunofluorescence primary stainings were performed with anti-CD31 (Clone MEC 13.3, BD Biosciences) and anti-VEGFR-2 (R&D Systems, Catalog # AF644) antibodies diluted in blocking solution and incubated on sections overnight at 4°C. Secondary staining was then performed for 1 hour at RT using donkey anti-rat IgG-Alexa488 and donkey anti-goat-Alexa 568 (Invitrogen). All slides were then stained with DAPI (Sigma-Aldrich, St Louis, Missouri, USA) and mounted using DABCO solution (Sigma-Aldrich). Images were acquired on a Zeiss AxioVision microscope with an AxioCam and processed using AxioVision software and Image J. To analyze the vessel presence in the tumor area as well as the CD31 MFI, the staining was evaluated on three to six different areas of each tumor. A similar procedure was performed for the VEGFR-2 staining and analysis.

RNA and cDNA preparation from dissociated tumors and relative quantification of the DC101 scFv using fast SYBR Green-based reagents (Applied Biosystems) was performed as previously described.40 The following primers were used for DC101 scFv: DC101-Forward: 5′-GCAACCCAAACTCCTCATCT-3′; DC101-Reverse: 5′-TATCATCAGCCTCCACAGGA-3′. The primers for GAPDH applied in the SYBR Green-based qPCR and used for normalization of the RNA levels were the following: GAPDH-Forward: 5′-AGGTCGGTGTGAACGGATTTG-3′ and GAPDH-Reverse: 5′-TGTAGA CCATGTAGTTGAGGTCA-3′. Evaluation of the relative mRNA expression for VEGFR-2, VEGF-A, CD8α and IFN-γ (normalized to GAPDH) was performed using TaqMan-based reagents and primers (Applied Biosystems). Each sample was run in triplicate and data were acquired using the 7500 Fast Real-Time PCR System (Applied Biosystems).

For murine VEGF-A protein quantification in the serum, blood was drawn on the day of ACT and placed in standard 1.5 mL Eppendorf tubes followed by incubation in a standing position at RT for 30 min. The tubes were then centrifuged at 2000×g for 10 min at 4°C to separate the clot from the supernatant (serum) and mVEGF-A was measured in the serum by ELISA (Mouse VEGF DuoSet ELISA, RD Systems) according to the manufacturer’s instructions. For mVEGF-A quantification in the tumors, extracted tumors were smashed using glass potter tissue grinders (Thermo Fisher Scientific) in PBS buffer containing EDTA-free Protease Inhibitors (Thermo Fisher Scientific, Catalog # A32965) and phenylmethylsulfonyl fluoride (PMSF; Roth, Catalog # A32965) according to the manufacturer’s instructions. The smashed tumors were then transferred to 1.5 mL Eppendorf tubes and centrifuged at 12,000×g for 10 min at 4°C. The resulting supernatants were then subjected to another centrifugation before quantifying the levels of mVEGF-A by ELISA.

Spleens and tumors were harvested from the differently treated mice. Splenocytes were gently crushed through a 40 µm cell strainer followed by RBC lysis. Tumor fragments were cut into pieces with scissors and then digested in RPMI supplemented with 200 µg/mL Liberase TL (Roche) and 5 units/mL DNase I (Sigma-Aldrich) for 1 hour at 37°C on a rocker, followed by passage through a 40 µm cell strainer. The ex vivo CAR-T cell phenotype was assessed by flow cytometry using antibodies mentioned above.

All statistical analyses were performed using GraphPad Prism V.8 software. Analysis of differences between two groups was performed using an unpaired two-tailed Student’s t-test. Statistical analyses of three or more groups were performed using a one-way analysis of variance (ANOVA) followed by Tukey’s post hoc correction test. Statistical analysis of tumor growth curves was performed using a two-way repeated measures ANOVA followed by Newman-Keuls post hoc correction test. Significance levels are indicated with stars in the figures and are the following: *p<0.05, **p<0.01, ***p<0.001 and ****p<0.0001.

In our study, we used a 2G CAR comprising the anti-VEGFR-2 scFv DC10129 fused to the hinge and transmembrane domain of CD8α, followed by the endodomains of CD28 and CD3ζ (figure 1A). Primary murine T cells were efficiently engineered (>80% transduction, as described in a previous work40) with retrovirus encoding the CAR (or Thy1.1 for control T cells) as evaluated by soluble recombinant mVEGFR-2-hIgG-Fc fusion protein staining and flow cytometric analysis on day 7 post-transduction (figure 1B). The adoptive transfer of 8×10⁶ CD45.1⁺ anti-VEGFR-2-CAR-T cells in lymphodepleted B16 melanoma tumor-bearing mice on days 10 and 13 post-tumor cell engraftment (in vivo treatment schematic shown in figure 1C) failed to control tumor growth (figure 1D). This is in line with previous studies showing poor responses with this 2G CAR as a monotherapy in the context of several tumor models.27–32

We first questioned whether TA loss may account for poor tumor control by the CAR-T cells. Immunofluorescent analysis of B16 tumors following adoptive cell transfer (ACT), however, showed that this was not the case as the distribution and density of CD31 and VEGFR-2 was similar in both CAR-T cell and control Thy1.1-T cell treatment groups (figure 1E–G). We further evaluated VEGF-A gene expression levels and observed a significant increase in the tumors of CAR-T cell versus control-T cell-treated mice, unlike VEGFR-2 gene expression which did not differ (figure 1H). Intrigued by this observation, coupled with the previous demonstration that VEGF-A can cause clustering defects of adhesion molecules,42 we next sought to evaluate if soluble VEGF-A could impair the adhesion and function of VEGFR-2 directed CAR-T cells against target cells.

To evaluate changes to CAR-T cell adhesion in the presence of VEGF-A, we began by setting up an in vitro assay (as illustrated in figure 2A) in which fluorescently labeled T cells were deposited onto confluent monolayers of bEnd3 (VEGFR-2⁺) or H5V (VEGFR-2^-) endothelial cells (figure 2B). Low levels of adhesion were observed for Thy1.1-T cells to both bEnd3 and H5V cells (figure 2C). In contrast, anti-VEGFR-2 CAR-T cells strongly bound to bEnd3 endothelial cells but poorly to H5V cells (figure 2C,D). A repetition of the assay for 1×10⁵ CAR-T cell and Thy1.1-T cells revealed a significant drop in CAR-T cell adhesion to bEnd3 cells in the presence of soluble VEGF-A (figure 2E). However, we observed that bEnd3 exposure to VEGF-A did not change the expression levels of intracellular adhesion molecule-1 (ICAM-1), vascular cell adhesion molecule-1 (VCAM-1) and VEGFR-2 (figure 2F). While these observations did not preclude the possibility of clustering defects of adhesion molecules, we next sought to explore the possibility that soluble VEGF-A, a natural angiogenic ligand of VEGFR-2, interferes with target receptor binding by the DC101 scFv-based CAR.

We interrogated structural databases and identified the human anti-VEGFR-2 antibody ramucirumab, which has been demonstrated to sterically hinder VEGF-A engagement with VEGFR-2,43 as having sequence similarity to DC101; ~65% homology for the heavy chain and ~58% for the light chain (figure 2G, online supplemental figure S1). By homology modeling using ramucirumab in complex with domain (D)3 of VEGFR-2 (PDB ID 3S3643), we predicted the VEGFR-2:DC101 structure (figure 2H) which we then superimposed on the structure of the human VEGFR-2:VEGF-A heterotetramer (PDB ID 3V2A48). Superimposing the predicted and experimental structures revealed that in this model DC101 binds to D2 and D3 of VEGFR-2 and into the groove between them, and that residues 51-IEDKSNNYFIS-61 of DC101 occupy the same space as VEGF-A (figure 2I,J). Hence, the computational modeling indicated that bound VEGF-A may sterically hinder DC101 scFv engagement with VEGFR-2.

We next sought to experimentally test if soluble VEGF-A interferes with the DC101 scFv-based CAR binding to mVEGFR-2. We thus mixed anti-VEGFR-2 CAR-T cells with recombinant mVEGFR-2-hIgG-Fc that had been preincubated with VEGF-A or preassembled VEGF-A/anti-VEGF-A antibody complexes (figure 3A). As expected, the control Thy1.1-T cells did not bind to mVEGFR-2-hIgG-Fc under any conditions tested. Anti-VEGFR-2 CAR-T cells strongly bound mVEGFR-2-hIgG-Fc, but not if mVEGFR-2-hIgG-Fc had been preincubated with VEGF-A (figure 3B). In addition, anti-VEGFR-2 CAR-T cells mixed with VEGF-A/anti-VEGF-A antibody complexes strongly bound to soluble mVEGFR-2-hIgG-Fc (figure 3B).

Next, as depicted in figure 3C, we evaluated anti-VEGFR-2 CAR-T cell effector function against plate-captured target antigen (i.e., VEGFR-2) under the same conditions. Anti-VEGFR-2 CAR-T cells secreted high levels of IFN-γ and IL-2 in the presence of target antigen (figure 3D and F), while cytokine production was attenuated in the presence of soluble VEGF-A but not in the presence of VEGF-A/anti-VEGF-A antibody complexes (figure 3E and G). In contrast, we found that the other VEGFR-2 ligands, namely VEGF-C and VEGF-D, did not impair CAR-T cell binding with soluble mVEGFR-2-hIgG-Fc (online supplemental figure S2A,B), nor did they attenuate CAR-T cell effector function (assessed by IFN-γ production) as was observed for VEGF-A (online supplemental figure S2C). Finally, to address a direct impact of VEGF-A, VEGF-C or VEGF-D on effector function, CAR-T cells were activated with αCD3/CD28 beads in the presence of supraphysiologic concentrations of the different molecules but there were no changes in IFN-γ, MIP-1α, TNF-α and IL-2 production (online supplemental figure S3A–D). Taken together, the computational and experimental data indicate that soluble VEGF-A sterically hinders DC101 scFv-based CAR-T cell engagement with VEGFR-2, thereby attenuating effector function.

Before exploring VEGF-A blockade in vivo, we sought to set up competition assays in the context of target endothelial cells. As depicted in figure 4A, target bEnd3 monolayers were exposed to soluble VEGF-A, VEGF-A/anti-VEGF-A antibody complexes, or control protein, prior to the addition of anti-VEGFR-2 CAR T cells (or control Thy1.1-T cells). As was previously observed in the context of plate-captured target antigen, both adhesion (figure 4B) and cytokine/chemokine secretion (IFN-γ, IL-2, TNF-α and MIP-1α; figure 4C) were attenuated upon co-culture of anti-VEGFR-2 CAR-T cells and target cells in the presence of VEGF-A but not VEGF-A/anti-VEGFR-2 antibody complexes or control protein.

Subsequently, we investigated whether tumor control by VEGFR-2 CAR T cells is impaired by VEGF-A in vivo. We coadministered anti-VEGF-A antibody with anti-VEGFR-2 CAR-T cells in lymphodepleted mice engrafted with B16 melanoma (illustrated in figure 4D). In this tumor model, soluble VEGF-A is highly abundant in the tumors and also present, although at much lower levels, in the serum (figure 4E). As we previously observed (figure 1D), anti-VEGFR-2 CAR-T cells alone and Thy1.1-T cells failed to control tumor growth. Consistent with prior work,49 coadministration of anti-VEGF-A antibody with Thy1.1-T cells slowed tumor growth (not significant) as compared with Thy1.1-T cells alone (figure 4F), presumably by tumor vasculature normalization, and/or TME remodeling favoring endogenous immunity. However, the combination of anti-VEGF-A antibody and anti-VEGFR-2 CAR-T cells significantly controlled tumor growth in comparison to both CAR-T cells alone and Thy1.1-T cells combined with anti-VEGF-A antibody (figure 4F). Notably, mice that received the combinatorial therapy did not show signs of toxicity, undergoing similar weight changes as the other treatment groups (figure 4G). Furthermore, we observed that coadministration of anti-VEGF-A antibody and anti-VEGFR-2 CAR-T cells was associated with a higher abundance of CD8α, IFN-γ and CAR-T cells in the tumors in comparison to all other treatment groups as measured by RT-qPCR (figure 4H–J).

Finally, we sought to evaluate the impact of anti-VEGF-A antibody coadministration on CAR-T cell fitness in the tumor and spleen following ACT. By flow cytometric analyses, we evaluated expression of the inhibitory markers programmed cell death protein-1 (PD-1), lymphocyte activation gene 3 (LAG-3) and cytotoxic T lymphocyte associated protein-4 (CTLA-4), along with Ki67 as a marker of proliferation. We observed a lower proportion of PD-1⁺ Ki67^- CAR-T cells as well as LAG-3⁺ Ki67^- CAR-T cells in the tumors of mice treated with CAR-T cells and anti-VEGF-A antibody. In contrast, there were no significant differences in the phenotype for CAR-T cells engrafted in the spleen (figure 5A–C), nor in the proportion of tumor-infiltrating CTLA-4⁺ Ki67^- CAR-T cells (figure 5C) between the two treatment groups. Notably, we found that the combination of CAR-T cells with the anti-VEGF-A antibody did not augment VEGF-A mRNA levels in tumors as compared with CAR-T cell treatment alone (figure 5D). In summary, along with enhancing tumor control, anti-VEGF-A antibody coadministration supports CAR-T cell fitness in the TME and may alleviate exhaustion. Based on our in vitro data and abundance of VEGF-A in the TME, presumably anti-VEGF-A antibody treatment also limits competition for VEGFR-2 binding by the CAR-T cells, thereby allowing enhanced reactivity against the tumor vasculature.