Secretion of a VEGF-blocking scFv enhances CAR T-cell potency

Abstract

CAR T-cell therapy is an effective treatment strategy in B-cell malignancies, however, its efficacy in solid tumors remains limited. VEGF-targeted drugs are used as antitumor agents to target abnormal tumor vasculature, however, toxicities associated with systemic VEGF blockade limit their maximal therapeutic benefit. Increasing evidence suggests a role for VEGF in the immunosuppressive tumor microenvironment (TME), including through direct induction of T cell–effector dysfunction. Herein, we show that CAR T cells from patients treated with FDA-approved CAR T-cell products express members of the VEGF signaling pathway and this expression is correlated with patient non-response. To overcome putative VEGF-induced CAR T-cell dysfunction and deliver local VEGF blockade, we generated CAR T cells that secrete a VEGF-targeting scFv to block T-cell and tumor-derived VEGF within the TME. These CAR T cells potently inhibited VEGF signaling and angiogenesis in vitro, and exhibited enhanced activation, cytotoxicity, proliferation, and effector function across different antigen and solid tumor contexts. VEGF scFv–secreting CAR T cells showed improved tumor control in immunocompromised murine metastatic and orthotopic models of ovarian and lung cancer. These findings suggest that CAR T cell–secreted VEGF blockade augments CAR T-cell performance, inhibits VEGF without systemic toxicity, and warrants further development.

Introduction

Chimeric antigen receptor (CAR) T cells drive dramatic and durable remissions in patients with B-cell malignancies[1–8]. However, obtaining responses in solid tumors has been more challenging, and is constrained by T-cell exhaustion, limited suitable tumor-specific antigens, and suppression by the tumor microenvironment (TME)[9]. There is a dearth of highly effective “armoring” strategies to augment CAR T-cell function and modulate the TME.

Blockade of vascular endothelial growth factor (VEGF) signaling to limit tumor angiogenesis and tumor autocrine/paracrine signaling is a long-standing antitumor strategy, with FDA-approved drugs employing this approach used to treat a variety of solid tumor types[10]. The success of these agents is predicated on an imbalance of pro-angiogenic signals (like overproduction of VEGF) in the TME leading to the formation of abnormal tumor vasculature. This deranged milieu promotes tumor growth and induces immunosuppression[11]. Currently, there are more than 10 FDA-approved products targeting the VEGF signaling axis with many more in clinical trials[12]. Patients treated with VEGF blocking agents have seen benefits including increased quality of life, progression-free survival, and overall survival across a range of cancers including non-small cell lung cancer (NSCLC)[13,14], renal cell carcinoma (RCC)[15–17], ovarian cancer[18], and glioblastoma[10]. However, doses of systemically administered VEGF-targeted agents are limited by cardiovascular and renal toxicities[19], constraining their maximal theoretical antitumor benefit.

Beyond an angiogenic-specific effect, there is mounting evidence that VEGF signaling limits the antitumor potency of T cells[20–23]. VEGF has been shown to impair T-cell proliferation in vitro, which can be ameliorated with VEGF blockade[20,23]. Multiple groups have demonstrated clear dose-dependent increases in exhaustion-associated T-cell markers after co-culture with VEGF which coincide with loss of hallmarks of effector function[20,21]. Furthermore, systemic VEGF blockade results in increased infiltration of effector T cells (including CAR T cells) in syngeneic murine models of solid tumor malignancies[22,24]. Most convincingly, conditional knockout of the VEGF receptor (VEGFR) in T cells in a syngeneic model of colorectal cancer significantly decreased tumor volume and prolonged survival[20]. These data support the idea that VEGF blockade may not only be beneficial for traditional antitumor effects but also bolster T-cell proliferation and effector function.

CAR T cells represent a class of highly efficient living delivery vehicles capable of deploying biologics to the TME when they would otherwise be toxic with systemic administration[25–27]. In this study, we leveraged this concept by engineering CAR T cells to secrete a VEGF-blocking single-chain variable fragment (scFv) (CAR^αVEGF). To achieve this, we employed mesothelin (Meso)-targeting 4–1BB CAR and our previously published, optimized, ligand-based, CD70-targeting 41BB CAR with a modified hinge[28]. Using this approach, we have demonstrated activation-dependent VEGFR upregulation in CAR T cells that was recapitulated in the CAR T cells of patients treated with FDA-approved CAR T-cell therapies. In these patients, we saw an association between VEGFR upregulation and patient non-response. We further found that secreted VEGF-blockade resulted in abrogation of VEGF signaling across a range of physiologically relevant concentrations and interruption of in vitro angiogenesis. CAR^αVEGF displayed enhanced activation, cytotoxicity, proliferation, and production of effector cytokines across both targets and a range of histologies including both solid and liquid tumors. Finally, CAR^αVEGF drove enhanced antitumor responses against orthotopic metastatic models of ovarian and lung cancer.

Methods

Re-analysis of published data from patients treated with CD19 CAR T-cell therapy

An annotated scanpy object corresponding to 10x Chromium 5’ single-cell data published in [29] were downloaded from the publication’s github page (https://github.com/getzlab/Haradhvala_et_al_2022, file CART_fulldataset_clustered.h5ad). Analysis was carried out using python 3.7.3 and scanpy v1.6. Cells were grouped by timepoint and cell type (Monocyte, CAR-negative T-cell, CAR-positive T-cell) based on the previously published annotations. Pseudobulk expression measurements were computed for a given gene g and cell population C by summing transcript counts across cells and normalizing to transcripts per million:

where n_c^g and N_c are the gene-level and total UMI counts in a given cell in a particular cell set C, respectively. Only cell populations with at least 200 cells in a given patient were included, yielding n=133 measurements taken from N=32 patients, including N=20 baseline, N=31 IP, N=29 D7 CAR-negative, and N=22 D7 CAR-positive sorted PBMC samples.

Cell lines

Cell lines were obtained from the American Type Culture Collection (ATCC) (786-O [obtained in 2022], JeKo-1 [2019], AsPC-1 [2019], BxPC-3 [2019], Capan-2 [2019], Panc-1 [2017], NCI-H226 [2019], SKOV3[2021], OVCAR3[2022], A549, HEK293T [2016], Jurkat, [2021], K562 [2016]). Nomo1 was a gift from Dr. Christopher Ott (Massachusetts General Hospital Center for Cancer Research, received in 2021). OVCAR4 was a gift from Dr. Oladapo Yeku (Massachusetts General Hospital received in 2021). All cells except A549 were cultured in R10 medium, which comprised RPMI1640 medium (Thermo Fischer Scientific, 61870036) supplemented with 10% FBS (Gibco A52567) and 1% Pen Strep (Thermo Fisher Scientific 15140163). A549 cell lines were cultured in EMEM with L-glutamine (ATCC, 30–2003) supplemented with 10% FBS and 1% Pen Strep. Cells were genetically engineered via lentiviral transduction to express click beetle green and green fluorescent protein (CBG-GFP) and sorted to purity. Cell lines were validated using short tandem repeat (STR) testing at least every 3 years and were kept to less than 30 passages prior to use in experiments. K562s were transduced with Meso and CD70 using lentiviral transduction. Mycoplasma testing was performed on all new cell lines and regularly thereafter using MycoAlert (Lonza) according to the manufacturer’s instructions.

CAR Construction

CAR transgenes (Meso^αVEGF, 70^αVEGF, Meso^αCD19, 70^αCD19; Fig. S1) were designed in Geneious (version 2023.2.1), then synthesized and cloned into third generation lentiviral vectors under a human EF-1a promoter (Genscript). A novel mesothelin-targeting monoclonal antibody, termed A2A11, was generated by LifeTein via immunization of mice with human mesothelin (Fig. S2). Sequencing of the antibody variable domain of the A2A11 hybridoma was performed by GenScript and the resulting sequence was incorporated into our second-generation CAR construct (see A2A11 scFv below). The generation of our CD70-targeting CAR T-cell construct is described in detail in[28]. The VEGF scFv has previously been described [48, 49], and we have previously described the use of a CD19-targeted scFv [25].

A2A11 scFv

METDTILLWVLLLWVPGSTGDIVLTQSPASLAVSLGQRATISCKASQSVDYDGDSYMNWYQQKPGQPPKLLIYAASNLESGIPARFSGSGSGTDFTLNIHPVEEEDAATYYCQQSNEDPSTFGGGTKLEVKGGGGSGGGGSGGGGSGGGGSMECNWILPFILSVTSGVYSEILLQQTGTVLARPGTSVKMSCKASGYTFTNYRMHWVKQRPGQGLEWIGGIYPGNSDTNYNQKFKDKAKLTAVTSTSTANMELSSLTNEDSAVYYCLRGIRGSYFDYWGQGTTLTVSS

Lentiviral Production

HEK293T cells were used to produce Lentivirus harboring CAR plasmids. HEK293T cells were plated at 3×10⁶ cells/plate in a 150mm cell culture plate. After 2 days, HEK293T cells at 80% confluence were transfected with CAR plasmid using Lipofectamine 3000 transfection reagent (Invitrogen, L3000001) and packaging vector (all from Genscript) that included 7μg pMGH\VSV-gL, 18 μg pMGH\RSV-Rev, and 18μg pMGH\GAG-POL per 150mm cell culture plate. Lentivirus was harvested at 24 and 48 hours and concentrated via 2 hour ultracentrifugation at 25,000rpm. Lentivirus was titered using Sup T1 cells prior to transduction.

CAR T-cell Production

Primary T cells were obtained from healthy donor leukapheresis products purchased from the Massachusetts General Hospital blood bank under an IRB-exempt protocol and isolated using (Stem cell tech). Isolated primary T cells were cultured in R10. T cells were activated using anti-CD3/28 beads (Life Technologies, 40203D) and 20 IUs/mL of human recombinant IL2 (PeproTech, 200–02) at a 1:3 ratio. Cells were transduced with lentivirus 24 hours after activation at a multiplicity of infection (MOI) of 5. Anti-CD3/28 beads were removed via magnetic de-beading on day 6. Media was doubled every 2–3 days until day 14 when transduction efficiency was measured and cells were frozen after normalization across donors. Supplementary Table S1 contains donor and transduction information for all experiments in the manuscript.

Stimulation Assays

For VEGFR1 expression assays, CAR T cells were stimulated with irradiated (100 grays) CD70- or Meso-expressing K562s at a 1:1 ratio for 96 hours or with PMA cell stimulation cocktail (eBioscience, 00–4970-03) for 4 hours. After incubation, cells were stained for CD8 and VEGFR1 via flow cytometry. For CD69 stimulation assays, Meso- and CD70-CAR T cells were stimulated with either SKOV3 tumor cells for 18 hours or recombinant human CD70 protein (Abcam, ab271444) for 24 hours, respectively. Tumor cells were plated a few hours prior to allow time to adhere. Plates were coated with recombinant CD70 protein the day before assay, incubated at 4C overnight, and washed before plating CAR T cells. CAR T cells were thawed and cultured in R10 with IL2 (20 IU/mL) overnight before being plated.

Reporter Assays

Reporter assays were performed using a VEGF BioAssay Kit (Promega, GA2001) according to manufacturer’s instructions. Jurkats transduced with 70^α19 and 70^αVEGF were seeded at 0.5×10⁶ cells/mL and expanded for 1 week. Supernatant was collected and filtered to obtain cell-free anti-CD19 or anti-VEGF-containing medium. VEGF from 786-O cell line was harvested from a confluent plate after 1 week of culture.

Angiogenesis Assay

Normal Human Dermal Fibroblasts (NHDFs) (Lonza, CC-2511) were seeded at 15,000 cells/well in a flat bottom 96 well plate in EGM-2 medium (Lonza CC-3162) for 3 days until confluent. After 3 days, Human Umbilical Vein Endothelial Cells (HUVECs) (Angio Proteomie, CAP0001GFP) were seeded onto the confluent layer of NHDFs at 5,000 cells/well and treatment conditions were added (recombinant VEGF at 4 ng/ml (R&D Systems 293-VE-010), suramin sodium salt at 0.25 mg/ml (Millipore Sigma, S2671–25MG), culture supernatant from confluent 70^αVEGF, 70^aCD19, or untransduced Jurkat T cells diluted 1:4). Media was changed every two days. The assay was imaged and analyzed using the Sartorius Incucyte Angiogenesis Software package every 6 hours for 14 days.

Long Term Proliferation Assays

CAR T cells were stimulated with irradiated Meso-expressing K562s at a 1:1 ratio. Every 3 days, CAR T cells were quantified and restimulated at 1:1 ratio and phenotyped for expression of PD1, TIM3, and LAG3 surface markers via flow cytometry.

Cytotoxicity Assays

All cytotoxicity assays were performed using the Incucyte Live Cell Analysis System (Incucyte SX5). CAR T cells contain an mCherry reporter, detectable in the red incucyte channel. Tumor cells contain a GFP reporter detectable in the green incucyte channel. Tumor cells were plated and given 4–6 hours to adhere. For suspension tumor cell lines, plates were pre-coated with CD71 unconjugated antigen (BioLegend, 334102) to ensure suspension tumor adhesion. CAR T cells were thawed and cultured in IL2 overnight before being plated onto adhered tumor cells. Plates were imaged every 1–2 hours using the green and red fluorescence channels. Tumor cell and CAR T-cell quantification represent the total area of green or red fluorescence detected in each image, which can be analyzed using the incucyte software. Green and red fluorescence area values were normalized to the values at time 0 where specified.

Cytokine Analysis

Cytokines were measured using Ella Automated Immunoassay System with Multianalyte assay chips (Bio-Techne, SPCKE-CS- 010597). Supernatant was collected from cytotoxicity assays at 72 hours, or as specified, and frozen until day of quantification. Assay was performed following manufacturer’s protocol.

ELISA

VEGF ELISA (Invitrogen, BMS277–2) was performed following manufacturer’s instructions. Supernatant was collected from CAR T cells on day 13 of production. ELISA’s for Granzyme B (R&D Systems DY2906–05), TNF-α (R&D Systems DY210–05), IFN-γ (R&D Systems DY285B), and IL2 (R&D Systems, DY202) were performed according to the manufacturer’s recommendations.

Nanostring

96-well round bottom plates were coated with 50 μL/well recombinant human CD70 protein (Abcam, ab271444) diluted in PBS to a final concentration of 10 μg/mL. Plates were incubated at 37°C for two hours before dumping and adding 1×10⁵ cells per well at a concentration of 1×10⁶ cells/mL. Samples were incubated for approximately 48 hours before being washed in PBS and resuspended at a concentration of 1×10⁴ cells/μL. Cells were lysed using iScript RT-qPCR Sample Preparation Reagent (Bio-rad, 1708898) and stored at −80°C until use. 5 μL of each sample were processed according to manufacturer protocol (Nanostring, “Lyse and Go protocol” MAN-10051–06, CAR-T characterization panel codeset 115000343, 10,00 cells/μL per sample direct lysate) and incubated overnight at 65°C in the thermocycler and allowed to hybridize with the panel probes. After overnight incubation in the thermocycler, cell lysates were loaded into the nCounter cartridge (NAA-AKIT), which analyzed gene expression using the CAR-T characterization panel. The resulting data was analyzed using nSolver Advanced Analysis software with samples being stratified by CAR construct and donors.

Flow Cytometry

The following antibodies were used: anti-mouse IgG (Cell Signaling Technology 4410S) VEGFR1 (R&D systems, FAB321V-100ug), Mouse IgG1k Isotype (R&D systems-IC002V), CD69 (Biolegend, 310918), CD8 (BD-560179, BD-647458), CD4 (BD-651850), CD3 (641406), CD71 (Biolegend-334104), PD1(BD Horizon-563789), TIM3 (BD Horizon-565566), LAG3 (BD Pharmingen-565716), CCR7 (BD Pharmingen-561271), CD45RA (Biolegend-304126), CD45 (Biolegend-368562), TOX (Miltenyi Biotec, 130-118-335), Human IgG1 isotype (Miltenyi Biotec, 130-120-709) . Cells were stained for 10 minutes at room temperature in the dark followed by 2x washes with FACS buffer (2% FBS in PBS). Cells were resuspended in DAPI (Thermo Fisher Scientific D1306)-containing FACS buffer to determine live/dead cell separation. Cells were acquired on a BD Fortessa X20 and analysis was performed using FlowJo software (version 10.10.0).

Intracellular Flow Cytometry

Cells were washed 2x in PBS and then stained with Fixable Viability Stain 575V (BD, 565694) at room temperature for 10 minutes. Cells were then fixed with the BD Cytofix/Cytoperm Fixation/Permeabilization Kit (554714) according to manufacturer protocol and stained at 4°C for 30 minutes. Fixed cells were washed and then flowed in PBS on a BD Fortessa X20 and analysis was performed using FlowJo software (version 10.10.0).

In vivo models

Nod-scid gamma (NSG) mice were obtained from Jackson Laboratories and bred under pathogen-free conditions at the MGH Center for Comparative Medicine. Mice were maintained in 12:12 h light:dark cycles at 30–70% humidity and a room temperature of 21.1–24.5 °C. Experiments were performed in 6–10 week old male and female mice under MGH Institutional Animal Care and Use Committee-approved protocols. All tumors contained the CBG bioluminescent reporter. For the A549 model, 1×10⁶ CBG-GFP A549 tumor cells were injected via tail vein on day −7. For the SKOV3 model, 1×10⁶ CBG-GFP SKOV3 tumor cells were injected intraperitoneally (IP) on day −7. On day −1, animals were randomized and injected IV with the indicated quantity of CAR T cells. Tumor bioluminescence was imaged once per week using a Spectral Ami HT imaging system after injecting mice IP with D-luciferin (Fisher Scientific PI88294. All mouse procedures were performed by a technician who was blinded to experimental groups. For systemic VEGF-specific antibody administration clone 2G11–2A05 from BioXCell was utilized IV at a dose of 5mg/kg 2x weekly. Rat IgG2a isotype (BE0089-R025mg) was given as a control. For the orthotopic 786-O model, mouse flanks were shaved, sterilized, and then injected transcutaneously after palpating the kidney with a 50:50 mixture of matrigel/PBS with 1.5×10⁶ 786-O tumor cells and allowed to engraft for 14 days prior to T-cell injection. Engraftment was monitored via a CBG bioluminescent reporter.

Tumor Digestion

In the A549 lung cancer model, mice were euthanized on day 14, and one main bronchus was insufflated, ligated, and fixed for IHC. The other was digested using the Miltenyi Biotec Tumor Dissociation Kit, Human (130–095-929) on the gentleMACS Octo Dissociator with Heaters (130–096-427) using the gentleMACS program 37C_h_TDK_2. Manufacturer’s instructions were followed for the lung tissue digestion. The samples were stained using the indicated antibodies and evaluated by flow cytometry.

Immunohistochemistry

Mouse lungs from the A549 lung cancer model and kidneys from the orthotopic 786-O renal cell carcinoma model were washed in PBS and then incubated overnight in 4% paraformaldehyde (PF Thermo-Fisher Scientific AAJ19943K2) and then stored in 70% ethanol until staining. The MGH histopathology core performed tissue staining. Tissues were stained with CD3 Ventana 790–4341, CD31 abcam ab28364, invitrogen MA5–32038. Slides were scanned on an Axio Scan.Z1 system using ZEN 2 (blue edition) software. Images were processed using Qupath (v 0.4.4) and underwent white balance correction as indicated in the individual figure legends.

Mouse Blood Quantification

Blood was collected from mice via facial bleeds for the A549 model. 50μL of blood was incubated with mouse dump antibodies, Ly6G/Ly6C-APC, NK1.1-APC, CD11b-APC, TER-119/Erythroid-APC (BioLegend 108412, 108710, 101212, 116212) in TruCount tubes (BD, 340334). Red blood cells were lysed using RBC lysis buffer (Thermo Scientific, 89901). Blood samples were run on a BD Fortessa X20 and CAR T cells were identified via mCherry reporters. Data was analyzed using FlowJo software (version 10.10.0).

Statistical Methods

All analysis was performed using GraphPad prism (version 10.2.3). Data indicates mean values and error represents standard error of mean. Number of repeated technical, biological, and experimental replicates are described within the relevant figure legends. P values are reported on figures and significance established at a p value less than 0.05. All statistical tests are two-tailed.

Data Availability Statement

Single-cell RNA sequencing data from figure one is available: Gene expression matrices from the scRNA data have been previously deposited with the Gene Expression Omnibus (GEO accession GSE197268). Raw sequencing data is available on the database of Genotypes and Phenotypes (phs002922.v1.p1). Other data generated in this study are available upon request from the corresponding author.

Results

To understand VEGF pathway member expression in CAR T-cell patients we conducted a reanalysis of a single-cell RNA sequencing dataset we previously generated from patients treated with two FDA-approved CAR T-cell products: axicabtagene ciloleucel (axi-cel), which has a CD19 CAR with a CD28z costimulatory domain, and tisagenlecleucel (tisa-cel), which has a CD19 CAR with a 41BB costimulatory domain [29]. In that study, samples were collected at baseline prior to CAR T-cell production, from the infusion product itself, and at day 7 post-infusion from the peripheral blood. In the reanalysis, we found that the CAR T cells in the infusion products expressed VEGF signaling pathway members, namely Flt1 (VEGFR1), VEGFA, and NRP2 (Fig. 1A, n=32 patients). This expression was almost exclusively restricted to tisa-cel (Fig. 1B). Furthermore, infusion product CAR T cells from non-responding tisa-cel patients had significantly higher expression of VEGFR1 than responding patients (Fig. 1C).

Figure 1: CAR T cells express members of the VEGF signaling family.

A, Single cell RNA sequencing data derived from[29] showing expression of KDR (VEGFR2), FLT1 (VEGFR1), VEGFA , NRP1, and NRP2 in CAR T cells from patients that received Tisa-cel and Axi-cel at baseline (prior to CAR T cell production), pre-infusion (post CAR T-cell production), and day 7 after CAR-T cell infusion into the patient (separated into CAR⁺ and CAR⁻ populations). Only cell populations with at least 200 cells in a given patient were included, yielding n=133 measurements taken from N=32 patients, including N=20 baseline, N=31 infusion product, N=29 Day +7 CAR-negative, and N=22 Day +7 CAR-positive sorted PBMC samples. B, Comparison of VEGF signaling family members among Tisa-cel (4–1BB) and Axi-cel (CD28z) patient pre-infusion products. C, Expression of VEGFR1, VEGFR2, NRP1, and NRP2 at baseline or in the infusion product (IP) or monocytes from responding and non-responding patients that received Tisa-cel or Axi-cel. P-values represent a wilcoxon ranksum test, the q-values are FDR-corrected by Benjamini-Hochberg. D, Mean fluorescence intensity (MFI) of VEGFR1 expression in meso-targeting CD8⁺CAR⁺ T cells (or untransduced T cells, UTD) with (+) and without (−) stimulation with MESO⁺K562s for 96 hours and representative histograms. Symbols represent technical triplicates from 2 normal donors (NDs), bars represent mean±SEM, p values by unpaired t-test). E, MFI of VEGFR1 expression in CD70 targeting CD8⁺ CAR⁺ T cells after activation with CD70⁺K562s for 96 hours and representative histograms. Symbols represent technical triplicates from 3 NDs, bars represent mean±SEM, p-value by unpaired t-test) F, VEGF concentration in culture supernatant following mesothelin CAR T-cell production measured by ELISA. (Symbols represent technical triplicates from 2 NDs, bars represent mean±SEM, p-value by one-way ANOVA). IP-infusion product

To corroborate these findings and determine if they could be extended to other antigen contexts, we performed an in vitro experiment using antigen-expressing K562s to stimulate CAR T cells. Upon stimulation through the CAR with CD70- or Meso-expressing K562s, Meso- (Fig. 1D, Supplementary Fig. S3) and CD70- (Fig. 1E, Supplementary Fig. S4) targeting 41BB CAR T cells increased expression of VEGFR1. However, VEGFR1 expression in activated Meso-targeting CD28z CAR T cells remained unchanged on activation, paralleling our finding in the tisa-cel vs. axi-cel patient CAR T-cell scRNAseq data. While CD70-targeting CD28z CAR T cells did experience a slight upregulation of VEGFR1 expression upon activation (Supplementary Fig. S4A), upregulation of VEGFR1 expression was still greater in CD70-targeting 41BB CAR T cells (Fig. 1E). Using CAR-independent activation with PMA, we observed a slight increase in the expression of VEGFR1 in Meso- (Supplementary Fig. S4B) but not CD70-targeting CAR T cells (Supplementary Fig. S4C). In addition to the expression of the receptor, we also found that activated CAR T cells secreted VEGF at levels comparable to non-transduced T cells, further supporting the idea that the VEGF signaling pathway may be relevant to CAR T-cell behavior (Fig. 1F).

Given the previously established effector dysfunction induced by VEGF on T cells [20,21,23] and our finding that VEGFR1 was upregulated in infusion product CAR T cells from non-responding tisa-cel patients, we generated Meso- and CD70-targeted 41BB-costimulated CAR T cells that secrete either a VEGF-blocking scFv, Meso^αVEGFand 70^αVEGF, or a CD19-specific scFv control, Meso^αCD19 and 70^αCD19 (Supplementary Fig. S1). We hypothesized that CAR T-cell secretion of a VEGF blocking scFv could block both tumor-derived and CAR T cell–derived VEGF, thereby augmenting CAR T-cell antitumor potency. To determine if our construct secreted a biologically active inhibitor of VEGF signaling, we took supernatant from 70^αVEGF and utilized it in a HEK293T VEGFR reporter assay, demonstrating blockade of VEGF signaling across several log-fold of physiologically relevant concentrations (Fig. 2A). This supernatant also completely blocked signaling from VEGF containing supernatant taken from the VHL deficient 786-O RCC cell line, which has unrestrained VEGFA production (Fig. 2B)

Figure 2: CAR^αVEGF blocks VEGF signaling and angiogenesis in vitro.

A and B, Bioluminescence of HEK-293T VEGFR reporter cells treated with supernatant containing either anti-VEGF or anti-CD19 and recombinant VEGF (rVEGF) (Data from technical triplicates, points represent the mean, error bars represent SEM, p-value by two-way ANOVA) (A) or supernatant from confluent VEGF-producing 786O tumor cells (data from technical triplicates, lines represent the median, p-value by one-way ANOVA) (B). C, Schematic of in vitro angiogenesis assay to visualize blood vessel formation upon blocking VEGF. GFP⁺ human umbilical vein endothelial cells (HUVECs) were cocultured with normal human dermal fibroblasts (NHDFs), suramin, rVEGF, and culture supernatant from CAR^αCD19 or CAR^αVEGF constructs and primitive blood vessel formation was imaged over time on the incucyte. D, Representative images of GFP⁺ HUVECs with vessels labeled by the automated incucyte angiogenesis software package. E, HUVEC blood vessel network length (length of blood vessel network (mm) per mm² in image). (data representative of 2 technical replicates, mean±SEM, p-value by two-way ANOVA).

Given the known role of VEGF in promoting angiogenesis [11,30], we next sought to determine if our secreted anti-VEGF scFv was sufficient to inhibit angiogenesis in vitro. To model in vitro angiogenesis we used HUVECs on a NHDF matrix, as has been previously established [31–34] (Fig. 2C). HUVECs and NHDFs were cultured in the presence or absence of recombinant VEGF and the VEGF signaling inhibitor suramin [35]. Anti-VEGF–containing supernatant significantly reduced (p=0.009) blood vessel network length (primitive vessels per square millimeter (mm/mm²)) relative to the anti-CD19–containing supernatant control during 70 hours of co-culture (Fig. 2D,E).

After demonstrating that secreted anti-VEGF effectively abrogated VEGF signaling in reporter and angiogenesis assays at physiologically relevant concentrations, we endeavored to characterize how secreted VEGF blockade affected CAR T-cell activation and proliferation. In a short-term co-culture of Meso CAR T cells and SKOV3 targets, we observed higher levels of the activation marker CD69 in Meso^αVEGF than Meso^αCD19 (Fig. 3A, Supplementary Fig. S5). CD69 expression after CD70 stimulation of CD70 CAR T cells with plate-bound protein was more donor dependent, with one of two showing higher levels of CD69 in the 70^αVEGF relative to the 70^αCD19 CAR T cells. (Fig. 3B, Supplementary Fig. S5). To better understand the differences between CAR^αVEGF and controls in the absence of other potentially confounding cell types we stimulated these CAR T cells via well-plates coated with recombinant antigen and performed Nanostring targeted RNA expression analysis looking specifically at genes involved in CAR T-cell biology (Fig. 3C, Supplementary Fig. S6). Across n=4 donors we found divergent expression programs in the 70^αVEGF relative to the 70^αCD19 CAR T cells with the former exhibiting higher pathway flux in a number of areas related to enhanced CAR T-cell performance, including Type II interferon signaling, which was the most differentially expressed pathway (q=0.000328 Fig. 3D). Longer term co-culture of Meso CAR T cells with recursive stimulation using irradiated Meso-expressing K562 cells revealed significantly higher proliferation of Meso^αVEGF than Meso^αCD19 (p=0.0017) (Fig. 3E). Given the prior demonstration of a TOX-dependent induction of non–CAR T-cell dysfunction [20], we performed intracellular TOX staining after 96-hours of stimulation with irradiated K562s expressing the cognate CAR antigen for both Meso and CD70-targeted CARs, finding no significant differences between the constructs or their effects on bystander non–CAR T cells (Supplementary Fig. S7).

Figure 3: CAR^αVEGF enhances proliferation and effector function across target antigens and tumor types in vitro.

A, CD69 surface expression of meso-targeting CAR T-cell constructs cultured with SKOV3 for 18 hours. B, CD69 surface expression of CD70-targeting CAR T-cell constructs cultured with plate bound CD70 antigen for 24 hours. C, 70^αVEGF and 70^αCD19 CAR T cells (n=4 donor T cells each) were incubated for 48 hours on plates coated with recombinant CD70 protein, then lysed and evaluated via targeted RNA expression. Heatmap results highlight individual pathway scores for each donor and construct. See supplemental figure 5 for additional details. D, Pathway score for the most differentially scored pathway from 3C. Symbols represent individual T-cell donors. Dashed line represents the median and the dotted line represents the top and bottom quartiles. All pathways were subjected to paired t tests and corrected for multiple comparisons with the two-stage step-up method of Bejanmini, Kreiger, and Yekutieli with a desired FDR (q) of 1%. E, Fold expansion of Meso^αCD19 and meso^αVEGF CAR-T cells from restimulation assays with meso⁺K562 ( 2 NDs and 2 technical replicates, mean±SEM, p-values by 2 way ANOVA). F and G, Cytotoxicity and proliferation of meso-targeting CAR T cells cultured with SKOV3 at 3:1 E:T ratio (2 NDs with 2+ technical replicates) (F) and OVCAR3 at 3:1 E:T (2 NDs) (G). H and I, Concentration of VEGF, IFNγ, TNFα, and Granzyme B in supernatant taken from endpoint of associated cytotoxicity assay, SKOV3 (H) and OVCAR3 (I) measured using Ella automated immunoassay system ( 2 ND with 2+ technical replicates). J and K, Cytotoxicity and proliferation of meso-targeting CAR T cells cultured with Nomo-1 at 3:1 E:T (1 ND, representative of 2 NDs) (J) and A549 at 3:1 E:T (2 NDs) (K), images depict GFP-expressing tumor and mcherry-expressing CAR T cells from the same respective wells at time 0 and after 72 hours of co-culture. L, M, and N, Cytotoxicity assays of mcherry+ CD70^αCD19 and CD70^αVEGF CAR T cells cultured with GFP⁺ 786O at 2:1 E:T (3 NDs) (L), SKOV3 at 3:1 E:T ( 2 NDs) (M), and Nomo-1 at 3:1 E:T (1 ND) (N). O, Concentration of VEGF from supernatant of cytotoxicity assay in N (2 NDs with 2+ technical replicates). Cytotoxicity assays were performed using incucyte imaging system and wells were imaged every 1–2 hours. Data represents values normalized to the starting condition of red or green fluorescence. All cytotoxicity assays show mean±SEM with p-values by two-way ANOVA. All cytokine assays show mean±SEM with p-values by one-way ANOVA. ND-normal donor.

To further understand the cytolytic and proliferative capacity of our anti-VEGF–secreting CAR T cells, we performed co-cultures with our constructs and a number of solid and liquid tumor cell lines. Upon co-culture with the natively Meso-expressing SKOV3 and OVCAR3 ovarian cancer cell lines, Meso^αVEGF CAR T cells demonstrated enhanced antitumor activity compared to controls, and this was coupled with a significant increase in CAR T-cell expansion (p<0.0001 and p=0.0003, respectively; Fig. 3F,G). Interrogation of the supernatant following SKOV3 and OVCAR3 co-culture revealed complete blockade of detectable VEGF (p<0.0001) by the anti-VEGF CAR T cells and an increase in effector cytokines such as IFNγ, TNFα, Granzyme B, and IL2 (Fig. 3H,I). Similar enhanced antitumor activity and Meso^αVEGF CAR T-cell proliferation was seen with another mesothelin-expressing cell line, Nomo-1 (AML) (Fig. 3J). Upon co-culture with the Meso-expressing A549 lung cancer cell line Meso^αVEGF CAR T-cell proliferation was significantly augmented, however, there was no additional cytotoxicity benefit (Fig. 3K). Culture of tumor cell lines with anti-VEGF–containing supernatant alone had no effect on tumor growth of multiple cell lines, suggesting that the antitumor effect is CAR T-cell dependent (Supplementary Fig. S8).

To determine any putative antitumor and proliferative benefits of CAR T cell–secreted VEGF blockade in the context of another antigen target, 70^αVEGF CAR T cells were cultured with natively CD70-expressing tumor cells. Culture of 70^αVEGF CAR T cells with 786-O RCC (Fig. 3L), SKOV3 ovarian cancer (Fig. 3M), and Nomo-1 AML (Fig. 3N) cell lines, which resulted in enhanced proliferation and substantial decrease in detectable VEGF (Fig. 3O), demonstrated that VEGF blockade induced proliferative and cytotoxic benefits are applicable in multiple CAR-antigen contexts. While the proliferative benefit of CAR^αVEGF was generally consistent across tumor-antigen contexts, it did not always lead to enhanced anti-cytolytic activity at the effector:target ratios and time points evaluated. Analysis of T-cell subsets and exhaustion-associated markers at the end of production (prior to tumor co-culture) revealed no obvious differences between anti-CD19– and anti-VEGF–secreting CAR T cells in either Meso- or CD70-targeting contexts (Supplementary Fig. S9).

We next tested the antitumor activity of our Meso^αVEGF CAR T cells in vivo in immunocompromised murine models of ovarian and lung cancer. As an orthotopic model of metastatic ovarian cancer, NSG mice were injected intraperitoneally with the SKOV3 ovarian cancer cell line followed by tail vein injection of CAR T cells seven days later (Fig. 4A). Mice treated with Meso^αVEGF CAR T cells demonstrated enhanced tumor control compared to Meso^αCD19 CAR T-cell controls (Fig. 4B). In a model of NSCLC, mice were injected IV with the A549 cell line, which engrafts in the lungs (Fig. 4C). In this model, mice treated with Meso^αVEGF CAR T cells demonstrated eradication of tumor while Meso^αCD19 CAR T-cell controls had unrestrained tumor growth (Fig. 4D). Number of CAR T cells in the peripheral blood (cells/μL) were similar between the groups, although Meso^αVEGF CAR T cells did increase relative to Meso^αCD19 CAR T cells at later time points (Supplementary Fig. S10A). In a repeat experiment with reduced number of CAR T cells injected, Meso^αVEGF CAR T cells also trended toward higher peripheral blood expansion, concordantly with improved tumor killing, athough this was not statistically significant (Supplementary Fig. S10B,C). We then repeated this experiment under stress conditions with a lower intravenous CAR T-cell dose, sacrificing the mice at day +14 and harvesting the lungs to evaluate for T-cell phenotype, exhaustion-associated markers, and immunohistochemistry (Fig. 4E,F; Supplementary Fig. S11). Meso^αVEGF CAR T cells adopted a more effector memory–like phenotype in the tissue at the expense of central memory cells and had lower levels of Tim3 (p=.056) and numerically lower Lag3 (p=0.11) (Fig. 4E). Immunohistochemistry revealed T-cell tumor homing and infiltration only in the Meso^αVEGF CAR T-cell treated mice (Fig. 4F, Supplementary Fig. S12). To assess reduction in tumor angiogenesis resulting from Meso^αVEGF-mediated blockade of VEGF signaling to endothelial cells, we stained for the endothelial marker murine CD31 and saw apparent decreases in CD31 staining adjacent to aggregates of Meso^αVEGF T cells (Supplementary Fig. S12). To further investigate the ability of CAR^αVEGF to block VEGF locally in vivo, we utilized an orthotopic model of RCC where 786-O tumor was injected transcutaneously into the kidneys of mice. Following engraftment, the mice were sacrificed 14 days after CAR T-cell injection and the tissue was analyzed by immunohistochemistry. We observed a paucity of VEGF staining in regions overlying T cells of 70^αVEGF-treated mice, but not nearby regions without T cells or in 70^αCD19- and UTD-treated mice, suggesting localized blockade (Supplementary Fig. S13).

Figure 4: VEGF scFv-secreting CAR T cells enhance anti-tumor activity against metastatic orthotopic models of lung and ovarian cancer.

A, schematic of SKOV3 orthotopic model of ovarian cancer. B, Bioluminescent imaging of SKOV3 (ovarian) tumor-bearing mice treated with mesothelin-targeting CAR T cells. (Data represents 3 pooled experiments with n= 20 mice, 2 NDs, mean±SEM, p-value by mixed effects model). C, schematic of A549 metastatic orthotopic model of lung cancer. D, Bioluminescent imaging of A549 (lung) tumor-bearing mice treated with mesothelin-targeting CAR T cells. (data displays 1ND, n=5, mean±SEM, p-value by 2-way ANOVA; representative results from one of two experiments). E, In a repeat experiment, NSG mice (n=5 mice per group) were engrafted intravenously with 1×10⁶ A549 human non-small cell lung cancer tumor cells on day −7 followed by intravenous treatment with 1×10⁶ Meso^αVEGF, Meso^αCD19, or untransduced T cells (UTD), on day 0 and sacrifice at day +14. Lungs were harvested, digested, and evaluated for the indicated flow cytometry markers. T-cell phenotype and exhaustion markers were gated on contemporaneously stained healthy unstimulated donor T cells (Naive, CCR7⁺CD45RA⁺; Central Memory, CCR7⁺CD45RA⁻; Effector Memory, CCR7⁻ CD45RA⁻; Terminally Differentiated Effector, CCR7⁻CD45RA⁺. Lines represent the median. P-values represent individual Mann-Whitney tests. F, Whole slide scans of IHC staining for human CD3 (T cells) and human CK7 (tumor as well as alveolus) from representative mice from each of the experimental groups described in E. Images have been manually white-balance corrected for clarity. See supplemental figure 11 for additional highlighted regions and replicates. Representative of n=5 mice per group. IV-intravenous; IP-intraperitoneal

Finally, given the existing use of systemic anti-VEGF agents for the treatment of multiple malignancies, we wanted to compare the potency of the CAR^αVEGF platform to systemic administration of a species cross-reactive antibody. Returning to our A549 lung model in NSG mice, we found that Meso^αVEGF CAR T-cell treatment outperformed twice weekly high dose treatment (5mg/kg) with systemic anti-VEGF in addition to anti-mesothelin CAR (Supplementary Fig. S14).

Discussion

In this study we characterized the expression of VEGF pathway members in the context of CAR T cells. We found elevation of VEGFR1 in non-responding patients after CD19-targeted CAR T-cell therapy and, along with substantial existing data suggesting that VEGF induces effector dysfunction[20,21,23], we pursued a strategy of CAR T-cell secreted VEGF blockade. We show that our CAR^αVEGF secrete a highly potent, biologically active inhibitor of VEGF signaling and angiogenesis in vitro and display superior cytotoxicity, proliferation, and effector cytokine production across antigen and tumor contexts. Finally, we demonstrate that targeted delivery of VEGF blockade improves antitumor activity in vivo against metastatic and orthotopic models of ovarian and lung cancer in immunocompromised mice.

A key finding of this work is that VEGF has a direct impact on CAR T cells. VEGF has now been repeatedly and convincingly demonstrated to be sufficient to disrupt T-cell effector function[20,21,23], however, our study presents what we believe to be the first characterization of VEGF effects specifically on CAR T cells. We find a higher level of tumor infiltration and a shift toward an intratumoral effector memory–like phenotype with localized anti-VEGF secretion. This increase in effector memory cells came at the expense of a reduction in cells with a central memory–like phenotype, the latter of which is typically associated with a long-lived response in the blood of patients [36]. However, in the tissues, it may be more beneficial to shift toward enhancing effector functions for practical cytotoxicity, something that VEGF typically inhibits [20]. Intriguingly, while VEGFR2 has been identified as the dominant receptor responsible for VEGF-mediated signaling [37], we found only elevated VEGFR1 expression in CAR T-cell products from patients. Furthermore, elevation of VEGFR1 was exclusively seen in 41BB costimulated (but not CD28 costimulated) CAR T cells, and it was higher among non-responding than responding patients suggesting the VEGF signaling axis may be limiting CAR T cell–treated patients receiving 41BB CAR T-cell treatment. We recapitulated the finding that VEGFR1 is elevated on activated 41BB costimulated CAR T cells in vitro. CD28 and 41BB costimulation in the context of CAR T cells are known to activate differential downstream transcriptional programs, with CD28 driving AP-1, NFkB, and NFAT, while 41BB primarily stimulates NFkB [38]. NFkB signaling (and specifically non-canonical NFkB) signaling is stronger in 41BB than CD28z CARs and may have a more substantive role in VEGF-pathway activation. This could explain why we see higher levels of VEGFR1 upon activation of 41BB-containing CARs than those containing CD28z across 3 different constructs including clinical grade CARs, however, additional work will be necessary to confirm these associations [39,40]. VEGFR1 is typically thought to act as a decoy receptor and modulator of VEGF-signaling flux, having higher binding affinity than VEGFR2, but weaker receptor tyrosine kinase (RTK) signaling activity [41]. However, recent efforts have expanded the role of VEGFR1, identifying VEGF-mediated signaling through VEGFR1 as driving downstream activation of PLCγ and PI3K pathways in macrophages[42]. Integration of these data led us to hypothesize that blocking VEGF signaling among 41BB CAR T cells might endow them with more CD28-like properties such as enhanced activation and proliferation[43]. Targeted RNA expression profiling revealed divergent expression programs between anti-VEGF–secreting CAR T cells including higher enhanced Type II interferon and other hallmarks of T-cell activation in the 70^αVEGF CAR T cells relative to the controls in the absence of confounding tumor cells, lending support to this supposition.

Beyond the impact on T cells, substantial evidence has demonstrated the tumor angiogenesis–specific and immunosuppressive effects of VEGF in the TME and on other immune cells like regulatory T cells, macrophages, and dendritic cells [11,44]. It is estimated that VEGF concentration in the TME ranges from around 585–1125 pg/mL [45]. We found that CAR^αVEGF is able to block VEGF concentrations up to 0.1ug/mL (100,000pg/mL). While modeling in vitro angiogenesis disruption we used 4ng/mL (4,000pg/mL) of VEGF to model vascular growth disruption and see an inhibitory effect on vasculature growth at this concentration. While these in vitro assays do not precisely recapitulate the TME, we believe they demonstrate that our CAR T cells will produce sufficient VEGF-blockade to favorably augment the antitumor response. One limitation of our study is the lack of immunocompetent mouse models and a robust TME to recapitulate the effects of VEGF blockade on other immune cells. However, systemic VEGF blockade co-administered with CAR T-cell therapy in a syngeneic immunocompetent mouse model of glioblastoma recently demonstrated that, in addition to improved survival and cytotoxicity, combination therapy also acted synergistically to improve infiltration of CAR T cells into the TME[24]. Incorporating these results with our findings, we postulate that secreted VEGF blockade by CAR T cells may act synergistically against the tumor by attenuating VEGF-induced CAR T-cell effector dysfunction, normalizing tumor vasculature, and limiting the immunosuppressive effects of VEGF in the TME while avoiding the toxic effects engendered by systemic VEGF blockade[19]. We have previously leveraged this approach, demonstrating the potency, safety, and localization of secreted therapeutics by CAR T cells within the TME[25,26], including in patients[46]. With localized delivery only at the site of the tumor, our CAR^αVEGF technology represents a safer strategy, avoiding systemic toxicities, but imparts the same benefits of combined CAR T cell and systemic VEGF blockade.

Another advantage of our approach is the use of a VEGF-blocking scFv; a class of therapeutics that has a rapid systemic clearance compared to antibodies[47]. Typically, poor persistence is a drug-developmental liability, however, in our approach with targeted and constitutive delivery into the TME, rapid clearance allows a significant concentration gradient to develop within the TME relative to systemic circulation, thereby reducing the potential for off-target toxicity. Furthermore, scFvs are ideal secreted inhibitors since they have substantially greater tissue penetration[47,48], endowing our secreted therapeutic with a greater chance of infiltrating the dense TME. Compared to standard VEGF-blocking antibodies used clinically, like bevacizumab, our chosen scFv is much smaller (at 26 KDa compared to bevacizumab at 149 KDa) and has a stronger binding affinity to VEGF[47–49]. Finally, the smaller payload allows it to be more easily integrated into other size-restricted CAR T-cell technologies.

Our findings suggest that targeted delivery of VEGF blockade via CAR T cells has the potential to augment antitumor responses through perturbation of T-cell effector function. Before any clinical translation, it will be important to ensure that either regional or total-body anti-VEGF activity does not exceed levels that are toxic in humans. This may necessitate the use of inducible or druggable regulatory elements to constrain activity within a finite range. It is also possible that some benefit may be achieved through co-administration of systemic VEGF-blockade with CAR T cells rather than (or in addition to) CAR T-cell secretion of anti-VEGF molecules, as has been previously demonstrated [24]. Given that VEGF is a negative regulator of T-cell function it is conceivable that blockade of VEGF could lead to excessive T-cell activation, however, we do not see this in vivo, and in contrast, we see lower levels of Tim3 and PD-1 in CAR T cells isolated from the T cells in the lungs of mice.

Our work demonstrates the benefits of secreted VEGF blockade in two unrelated tumor antigen targets, suggesting that it may enhance the function of CAR T cells targeting other antigens as well. Additional work will be required to fully characterize the exact mechanism by which VEGF blockade enhances CAR T-cell function. In T cells, effector dysfunction has been shown to progress through a mechanism ultimately involving TOX signaling[20]. One potential hypothesis is that in addition to inhibiting VEGF signaling from the outside of the cell (autocrine and paracrine signaling), secreted VEGF blockade also blocks VEGF signaling that arises intracellularly. This ‘intracrine’ signaling has been demonstrated along the VEGF axis in endothelial cells[50]. Therefore, it is also possible that intracellularly produced anti-VEGF may bind and block intracrine as well as autocrine and paracrine VEGF.

In summary, CAR T cell–secreted VEGF blockade is a promising and clinically relevant strategy to improve CAR T-cell efficacy. Our data point to its potential to limit disruptive tumor angiogenesis, block harmful effects to the TME by overproduction of VEGF, and enhance CAR T-cell antitumor function. Further study of CAR^αVEGF technology is warranted to better understand its benefits and elucidate its mechanisms.

Synopsis:

Systemic toxicity limits the therapeutic benefit of VEGF-targeted agents. The authors show VEGF scFV–secreting CAR T cells mediate local VEGF blockade and augment CAR T-cell performance, suggesting they could enhance antitumor responses across targets and tumor types.