Antitumor Effects of CAR T Cells Redirected to the EDB Splice Variant of Fibronectin

Genetically modified T cells expressing chimeric antigen receptors (CAR T cells) have potent antitumor activity in patients with B-cell lineage malignancies targeting CD19, CD22, CD30, and/or BCMA (1–3). However, CAR T cells so far have had limited antitumor activity in early-phase clinical studies for patients with solid tumors and brain tumors (4, 5). This lack of efficacy is most likely multifactorial, including a limited array of targetable antigens, heterogeneous antigen expression, and the immunosuppressive tumor microenvironment (TME; refs. 4, 6).

The majority of tumor-associated antigens (TAA) that have been targeted with CAR T cells are differentially expressed cell surface molecules (6, 7). However, tumor cells secrete extracellular matrix (ECM) molecules that adhere to their cell surface and thus could serve as CAR targets. One prominent molecule of the ECM is fibronectin, a ubiquitously expressed protein. However, tumor cells secrete oncofetal splice variants, which are not expressed in normal adult tissues (8). One of these, extra domain B fibronectin (EDB or EIIIB), has been studied extensively (9). It is expressed in a broad range of solid tumors, including lung, breast, and prostate cancers, and high-grade glioma, making it a pan-cancer TAA (10, 11). Expression in endothelial cells of the tumor vasculature has been described (12). EDB belongs to fibronectins that are only present in the ECM and is not present in so-called plasma fibronectin. Like other ECM family fibronectins, it binds integrins on the cell surface through RGD motifs (12). Antibody–drug conjugates, based on the high-affinity EDB-specific mAb L19 (13), have shown potent antitumor activity in preclinical animal models (14, 15), and early-phase clinical testing is in progress. L19 mAb–based conjugates have been utilized successfully to image tumors in adult patients (16, 17).

To investigate the ability of EDB-CAR T cells to directly target cancer cells and endothelial cells of the tumor vasculature, we generated T cells expressing an EDB-CAR based on the L19 mAb. We demonstrated that EDB-CAR T cells recognized and killed EDB-positive tumor cells in an antigen-dependent manner in vitro and had potent antitumor activity in three xenograft mouse models without apparent “on-target/off-cancer” toxicity. EDB secretion by tumor cells enabled EDB-CAR T cells to target endothelial cells of the tumor vasculature. Thus, targeting proteins of the ECM, like EBD, have the potential to improve current CAR T-cell therapy for a broad range of solid tumors in which EDB is expressed.

The U87 (glioma), A549 (lung cancer), A673 (Ewing sarcoma), 293T, and HUVEC cell lines were purchased from the ATTC. The lung metastatic osteosarcoma cell line LM7 was kindly provided by Dr. Eugenie Kleinerman (MD Anderson Cancer Center, Houston, TX) in 2011. Primary fibroblast cell lines were previously established (18). The generation of the A549 cell line expressing an enhanced green fluorescence protein firefly luciferase fusion gene (GFP.ffluc) was previously described (4). All cell lines were grown in DMEM or RPMI (GE Healthcare Life Sciences HyClone Laboratories) supplemented with 10% FBS (GE Healthcare Life Sciences HyClone) and 2 mmol/L Glutamax (Invitrogen) for one to three passages after thaw. HUVEC cells were cultured two passages after thaw using vascular basal media (ATCC) supplemented with components of the endothelial cell growth kit (ATCC). The U87 fibronectin knockout cell line (U87^FN−/−) was generated by the Center for Advanced Genome Engineering (CAGE) at St. Jude Children's Research Hospital (St. Jude) using CRISPR/Cas9 gene-editing technology. Briefly, 400,000 U87 cells were transiently transfected with precomplexed ribonuclear proteins consisting of 150 pmol of chemically modified sgRNA (5′CCUAUAGAAUUGGAGACACC3′, Synthego) and 35 pmol of Cas9 protein (St. Jude Protein Production Core) via nucleofection (Lonza, 4D-Nucleofector X-unit). Two clones were identified via targeted deep-sequencing using gene-specific primers with partial Illumina adapter overhangs (hFN1.F—5′CTACACGACGCTCTTCCGATCTagtgtaataccttgcagcaccagagc3′, overhangs shown in upper case, hFN1.R—5′CAGACGTGTGCTCTTCCGATCTtcttgacctgcttccccatttcccg3′, overhangs shown in upper case). Next-generation sequencing analysis of clones was performed using CRIS.py (available on GitHub: https://github.com/patrickc01/CRIS.py; ref. 19). Two hFN1 knockout clones were identified. The genotype of both clones is indicated below and shown in Supplementary Fig. S1. We selected 2H3 for our experiments. Cell lines were authenticated using the ATCC's human short tandem repeat profiling cell authentication service and routinely checked for Mycoplasma by the MycoAlert Mycoplasma Detection Kit (Lonza).

The generation of the SFG retroviral vectors encoding the EphA2-CAR with a CD28 costimulatory domain, or GFP.ffluc, has been previously described (20). In-fusion cloning (Takara Bio) was used to generate the EDB-CAR with a CD28 costimulatory domain and a nonfunctional EDB-CAR with mutated immunoreceptor tyrosine–based activation motifs (ITAM; mutEDB-CAR) using our retroviral vector as a template, which encodes a EphA2-CAR.CD28ζ expression cassette, a 2A sequence, and truncated CD19 (20). The amino acid sequence of the EDB-CAR and mutEDB-CAR is shown in Supplementary Fig. S2. The EDB-specific single-chain variable fragment (scFv) was derived from the human mAb L19 (13) and synthesized by GeneArt (Thermo Fisher Scientific). RD114-pseudotyped retroviral particles were generated by transient transfection of 293T cells as previously described (20).

Human peripheral blood mononuclear cells (PBMC) were obtained from 11 healthy donors under an Institutional Review Board–approved protocol at St. Jude. To generate CAR T cells, we used our previously described protocol (4). Briefly, 1 × 10⁶ PBMCs were stimulated on non–tissue culture-treated 24-well plates (Corning), which were precoated with anti-CD3 (0.1 mg/mL) and anti-CD28 (0.1 mg/mL; BD). Recombinant human IL7 and IL15 (IL7: 10 ng/mL; IL15: 5 ng/mL; PeproTech) were added to cultures the next day. On day 2, T cells were transduced with 0.5 mL retroviral particle supernatant on RetroNectin (10 mg/well; Clontech)-coated plates. On day 5, transduced T cells were transferred into tissue culture 24-well plates and expanded with IL7 and IL15 (IL7: 10 ng/mL; IL15: 5 ng/mL) for another 2 to 7 days. Nontransduced (NT) T cells were prepared in the same way, except for no retrovirus was added. For the generation of GFP.ffluc-expressing EDB-CAR T cells, activated T cells were transduced with two retroviral vectors, one encoding the EDB-CAR and the other GFP.ffluc. All experiments were performed 7 to 14 days after transduction using unsorted “bulk” CAR T cells. Biological replicates were performed using PBMCs from different healthy donors.

A FACSCanto II (BD) instrument was used to acquire flow cytometry data, which were analyzed using FlowJo v10 (FlowJo). For surface staining of tumor cells, HUVECs, and newly generated CAR T cells, samples were washed with and stained in PBS (Lonza) with 1% FBS (HyClone). For all experiments, matched isotypes or known negatives (e.g., NT T cells) served as gating controls along with positive control (e.g., anti-CD4 in all colors). LIVE/DEAD Fixable Aqua Dead Cell Stain Kit (Invitrogen) was used as a viability dye. T cells were evaluated for CAR expression after transduction with an anti-human IgG, F(ab')2 fragment–specific AF647 (Jackson ImmunoResearch), or anti-CD19-PE (clone J3–119, Beckman Coulter). The following antibodies specific for human T-cell surface markers were used: anti–CD3-APC (clone UTCH1, Beckman Coulter), anti–CD4-Pacific Blue (clone SK3, BioLegend), anti–CD8-PerCPCy5.5 (clone SK1, BioLegend), anti–CD19-BV421 (clone HIB19, BD), anti–CCR7-AF488 (clone G043H7, BioLegend), anti–CD45RA-APC-H7 (clone HI100, BD), F(ab') Goat IgG (AF647: 005600003, The Jackson Laboratory). For detecting EDB expression in tumor cells and fibroblasts, we synthesized an L19 mAb with a hIgG1 heavy chain and human kappa light chain, which recognizes human and murine EDB based on the L19 sequence of our CAR and a previous report in the literature (Supplementary Fig. S3; Thermo Fisher Scientific; ref. 21). The antibody was conjugated using the Lightning-Link R-PE Antibody Labeling Kit (Novus Bio). For intracellular staining, cells were permeabilized using the Cytofix/Cytoperm Fixation/Permeabilization Solution Kit (BD). Anti-human EphA2-APC (clone 371805, R&D) was used to detect EphA2 expression in HUVEC cells.

mRNA extraction from 1 × 10⁶ to 1 × 10⁷ LM7, A549, and U87 tumor cells and fibroblasts was performed using the Maxwell RSC SimplyRNA Blood kit (Promega AS1380) on a Maxwell RSC instrument. RT-qPCR was performed according to the manufacturer's instructions using a one-step kit utilizing 800 ng of template and 3 biological replicates, normalized to GAPDH for delta Ct (Thermo Fisher Scientific). Previously published sequences were used for EDB PCR primers (11), and GAPDH PCR primers (GAPDH control reagents) were purchased from Life Technologies. Reactions were completed on the Applied Bioscience QuantStudio 6 Flex and analyzed using QuantStudio software (Thermo Fisher Scientific).

HUVEC capillary assays were performed using the In Vitro Angiogenesis Assay Kit (Millipore) per the manufacturer's instructions seeding 10,000 cells per well. Note that 30,000 CAR- or NT T cells were added to these capillary structures as judged by microscopic examination. Capillary structures were imaged on a Nikon Ti 12 hours after the CAR or NT T cells.

Note that 1 × 10⁶ NT or CAR T cells from healthy donors were cocultured alone or with 5 × 10⁵ LM7, U87, U87^FN−/−, or A549 cells without the provision of exogenous cytokine. For recombinant protein studies, recombinant human fibronectin (rhFN)-EDB (Abcam) was added at increasing concentrations (0.5, 1, and 10 ng/mL) for 3 hours at 37°C. Plates were washed, and 5 × 10⁵ T cells were plated. After 24 hours, supernatant was collected and frozen at −20°C for later analysis.

Cytokines were measured using human IFNγ and IL2 ELISA kits (R&D Systems). Supernatant was diluted 1:10 when required (if absorbance was above the detection limit, the ELISA was rerun with dilution). Note that 100 μL of supernatant in technical triplicates was analyzed for each biological replicate on a Tecan Infinite M Plex plate reader utilizing Magellan software (Tecan). Concentration was calculated using a standard curve included in the ELISA kits as per the manufacturer's directions.

Note that 1 × 10⁶ cells/mL of U87, U87FN^−/−, A549, A673, and LM7 cells and fibroblasts were harvested after 72 hours of culture. Cells were lysed with cell lysis buffer (Sigma), and collected media were snap frozen. Human fibronectin was detected by ELISA (R&D, DFBN10); 50 μL of undiluted media or cell lysate was analyzed for each biological replicate on Tecan Infinite M Plex plate reader utilizing Magellan software (Tecan). Concentration was calculated using a standard curve included in the ELISA kits as per the manufacturer's directions.

Animal experiments followed a protocol approved by St. Jude Institutional Animal Care and Use Committee. All experiments utilized 6- to 8-week-old NSG mice obtained from St. Jude NSG colony.

Mice were injected s.c. with 2 × 10⁶ tumor cells (U87, U87FN^−/−, A549, or A673) in Matrigel (Corning; 1:1 diluted in PBS). For the in vivo study in which mice received an admixture of U87 and U87^FN−/− cells, cells were mixed prior to subcutaneous injection to achieve the following percentages of U87 cells: 100%, 90%, 70%, 10%, and 0%. On day 7, mice received a single i.v. dose of 1 × 10⁶ fresh NT or CAR T cells from healthy donors via tail vein injection. Tumor growth was assessed by serial, weekly, caliper measurements. Mice were euthanized when (i) they met physical euthanasia criteria (significant weight loss, signs of distress); (ii) the tumor burden was approximately 3,000 mm³; or (iii) recommended by St. Jude veterinary staff. For the rechallenge experiments, mice received a single i.v. dose of 1 × 10⁶ fresh tumor cells from healthy donors via tail vein injection. For studies to determine the presence of CD31-positive endothelial cells, tumors were harvested after euthanasia and placed in formalin for paraffin embedding. For studies to determine the presence of FN or EBD, tumors were harvested after euthanasia, and half of the tumor was collected for fresh-frozen processing and the other half placed in formalin for paraffin embedding.

Mice were injected i.v. with 2 × 10⁶ tumor cells (A549) via tail vein injection, and on day 7, received a single i.v. dose of 1 × 10⁶ fresh NT or CAR T cells. Mice were euthanized when they reached (i) the bioluminescence flux endpoint of 2 × 10¹⁰ on two consecutive measurements (as described below) and/or (ii) the above-mentioned general euthanasia criteria.

Non–tumor-bearing mice received a single i.v. dose of 1 × 10⁷ NT or 1 × 10⁶ or 1 × 10⁷ CAR T cells expressing GFP.ffLuc. Infused T cells were tracked by bioluminescence imaging (as described below), and on day 14 after T-cell injection, kidneys, liver, lung, and spleen were harvested after euthanasia for analysis.

Mice were injected s.c. with 2 × 10⁶ U87 or A673 tumor cells in Matrigel (Corning; 1:1 diluted in PBS). At day 14, single-cell suspensions were prepared from harvested xenograft tumors, minced (∼0.5 mm pieces), and digested in collagenase II (Worthington, 50 μL enzyme per 0.1 g tissue) for 30 minutes at 37°C, filtered through a 70 μm cell strainer, and enriched for CD31-positive cells using MicroBeads (Miltenyi). MACS enrichment was performed using a Miltenyi autoMACS ProSeparator [Possel DS program with following reagents: anti-CD31 REAfinity PE (Miltenyi), anti-PE MicroBeads (Miltenyi)]. CD31-enriched cells were stained for DAPI (Invitrogen) and gated as single, viable cells, and the top 25% CD31-PE⁺ were sorted using the BDFACS Aria III. After sorting, 15,000 cells were plated into 0.1 mg/mL anti-murine CD31 (BD)-coated wells, and after 24 hours, 30,000 CAR- or NT T cells were added per well. After 24 hours, T-cell activation was assessed by determining the concentration of IFNy in culture media using an ELISA (Human IFNy, see Materials and Methods' section “Fibronectin ELISA” for detail); in addition, the cytolytic activity was determined using a standard MTS assay (see Materials and Methods' section “MTS Assay” for detail).

A CellTiter96 AQueous One Solution Cell Proliferation Assay (Promega) was utilized to assess CAR T-cell cytotoxicity. In a 96-well plate, 12,500 U87, A549, and U87^FN−/−, 10,000 HUVECs or CD31-positive sorted cells, or 15,000 LM7 cells or primary fibroblasts were cocultured with serial dilutions of CAR T cells and NT T cells at the indicated effector-to-target (E:T) ratios in the figure panels. Media and tumor cells alone served as controls. Each condition was plated in triplicate. After 5 days for tumor cells, 3 days for HUVECs, and 24 hours for CD31⁺ sorted cells, the media and T cells were removed by gently pipetting up and down to avoid disrupting adherent tumor cells, and live cells were determined according to the manufacturer's instructions utilizing a Tecan Infinite M Plex plate reader utilizing Magellan software (Tecan). Percent live tumor cells were determined by the following formula: (sample-media alone)/(tumor alone-media alone) × 100.

Mice were injected i.p. with 150 mg/kg of D-luciferin (Thermo Fisher Scientific) 5 to 10 minutes before imaging, anesthetized with 1.5% isoflurane purchased through St. Jude's Center for In Vivo Imaging and Therapeutics, and imaged with a Xenogen IVIS-200 imaging system. Emitted photons were quantified using Living Image software (Caliper Life Sciences). Mice were imaged once per week to track tumor burden, or 1 to 5 times per week to track T cells.

Mice where anesthetized with 1.5% isoflurane and received a single i.v. dose of Angiosense 750 (2 nmol/100 μL; PerkinElmer; NEV10011EX) as recommended by the manufacturer. Administered Angiosense 750 was allowed to equilibrate in mice for 24 hours before imaging using a Xenogen IVIS-200 imaging system. Fluorescence signals were quantified using Living Image software (Caliper Life). Relative fluorescent units were calculated by measuring tumor fluorescence subtracted from the background of mouse autofluorescence divided by tumor volume.

For CD31 IHC, tumor samples were fixed and embedded in paraffin, and 8 μm sections were cut and mounted on slides. The sections were then processed and stained with anti-CD31 (Dianova, clone SZ31, cat. DIA 310). CD31 was only scored on tumors ± 700 mm³ from the average tumor size in order to best control for the impact tumor size variation has on vascularization. CD31 expression was calculated by independent blind-scoring, and images were acquired using an Axio Scan Z.1. The tissue (kidneys, liver, lung, or spleen) of the safety study (see “Xenograft Mouse Models” section) was fixed in 10% neutral-buffered formalin, embedded in paraffin, sectioned at 4 μm, and stained with hematoxylin and eosin (H&E). Stained H&E sections were reviewed by light microscopy and interpreted by a board-certified veterinary pathologist (CAR T cells were considered safe if no tissue damage was observed).

To detect fibronectin expression, a Leica BOND-MAX–automated stainer (Leica Biosystems) was used with the manufacturer-supplied protocol IHC F, and the anti-fibronectin clone EP5 (Santa Cruz Biotechnology, SC-8422). All reagents were provided by Leica Biosystems, including the Leica Bond Dewax solution (AR9222), Leica Bond wash solution (AR9590), 3% to 4% (v/v) hydrogen peroxide for a 5-minute peroxide block (included in Bond Polymer Refine Detection kit DS9800), and HIER solution using the Bond Epitope Retrieval Solution 2 (ER2, AR9640) which was performed for 30 minutes. The primary antibody was incubated for 30 minutes. The Bond Polymer Refine Detection Kit reagents were used for visualization (DS9800).

To detect EBD expression, tumors were fixed in ice-cold acetone for 20 minutes and immunostained using the L19 mAb on a Ventana DISCOVERY ULTRA–automated stainer (Ventana Medical Systems, Inc.). HIER was carried out for 32 minutes at 37°C using cell conditioning media 2 (Roche, CC2, cat no. 950–223), followed by visualization with a rabbit anti-human IgG Fc Secondary antibody horseradish peroxidase (HRP; Thermo Fischer Scientific, #31423), OmniMap anti-rabbit HRP (760–4311), and the DISCOVERY ChromoMap DAB kit (760–159). Unaffected, morphologically normal murine epithelium served as an internal-negative tissue control to compare with positive immunoreactivity that was visualized in human tumor cells, the collagenous interstitium, and murine folliculosebaceous units.

Both the fibronectin and EDB IHCs were evaluated using visual estimations of immunoreactivity to determine the presence or absence of the protein marker as well as comparison of immunoreactivity in experimental samples with what was observed in the positive tissue controls.

Exon quantification in reads-per-kilobase-million (RPKM) was manually downloaded from the legacy GDC data portal (https://portal.gdc.cancer.gov/legacy-archive/search/f). The exons in these mRNAseq samples were quantified by The Cancer Genome Atlas (TCGA) RNAseqV2 pipeline (https://webshare.bioinf.unc.edu/public/mRNAseq_TCGA/UNC_mRNAseq_summary.pdf). Representative TCGA samples of solid and brain tumors were selected for analysis, including adrenocortical carcinoma (n = 79), bladder urothelial carcinoma (n = 166), breast-invasive carcinoma (n = 118), cervical squamous cell carcinoma and endocervical adenocarcinoma (n = 200), cholangiocarcinoma (n = 57), esophageal carcinoma, glioblastoma multiforme (n = 174), head and neck squamous cell carcinoma (n = 135), kidney renal clear cell carcinoma (n = 103), kidney renal papillary cell carcinoma (n = 200), lower-grade glioma (n = 200), liver hepatocellular carcinoma (n = 188), lung adenocarcinoma (n = 153), lung squamous cell carcinoma (n = 134), mesothelioma (n = 80), ovarian serous cystadenocarcinoma (n = 82), pancreatic adenocarcinoma (n = 183), pheochromocytoma and paraganglioma (n = 187), prostate adenocarcinoma (n = 200), rectum adenocarcinoma (n = 177), sarcoma (n = 265), skin cutaneous melanoma (n = 200), stomach adenocarcinoma (n = 154), testicular germ cell tumors (n = 156), thyroid carcinoma (n = 200), thymoma (n = 244), uterine corpus endometrial carcinoma (n = 200), uterine carcinosarcoma (n = 57), and uveal melanoma (n = 80). Two hematologic malignancies were also incorporated, including lymphoid neoplasm diffuse large B-cell lymphoma (DLBC; n = 48) and acute myeloid leukemia (AML; n = 95). A java program was used to extract the EDB exon expression from the RPKM files. The EDB exon expression boxplot was generated using an R package ggplot2 under R version 3.5.1. Code from our analysis is available in the GitHub repository (https://github.com/gatechatl/EDB_Exon_Expression).

All experiments were performed at least in triplicates. For comparison between two groups, a two-tailed t test was used. For comparisons of three or more groups, values were log-transformed as needed and analyzed by ANOVA with Tukey posttest. Survival was analyzed by the Kaplan–Meier method and the log-rank test. Bioluminescence imaging data were analyzed using either ANOVA, t test, or AUC. Statistical and AUC analyses were conducted with Prism software (Version 9.0.0, GraphPad Software).

We generated a retroviral construct encoding an EDB-specific CAR consisting of an EDB-specific scFv derived from the L19 mAb (13), a short hinge/CD28 transmembrane domain, and a CD28ζ signaling domain. The vector also encoded a 2A sequence and truncated CD19 (tCD19; Fig. 1A). EDB-CAR T cells were generated by standard retroviral transduction, and 7 to 10 days after transduction, CAR expression was determined by flow cytometric analysis using anti-F(ab')₂. A mean of 35.6% of T cells were CAR-positive (Fig. 1B and C). Retroviral T-cell transduction was confirmed using anti-CD19 (Supplementary Fig. S4A and S4B). EDB-CAR T cells contained a mixture of CD4- and CD8-positive T cells, and further T-cell subset analysis revealed the presence of naïve, central memory, effector memory, and terminally differentiated memory T cells (Supplementary Fig. S4C and S4D).

To initially demonstrate that EDB-CAR T cells recognized EDB, NT or EDB-CAR T cells were cultured on plates coated with increasing amounts of recombinant human FN-EDB (rhEDB) protein. After 24 hours of culture, only EDB-CAR T cells produced significant amounts of IFNγ (Fig. 1D), demonstrating activation of EDB-CAR by its target antigen. We confirmed EDB expression in a broad range of solid tumors and brain tumors from TCGA (Supplementary Fig. S5) and focused our functional studies on sarcoma (osteosarcoma, LM7; Ewing's sarcoma, A673), lung adenocarcinoma (A549), and high-grade glioma (U87). All four cell lines expressed EDB as judged by RT-qPCR using primary human fibroblasts from two healthy donors as negative controls (Fig. 1E). To confirm EDB expression, we performed cell surface and intracellular staining for EDB using the L19 mAb. Note that 85.6% to 99.5% of U87 and A549 cells were EBD-positive by cell surface or intracellular staining; 73.5% of A673 cells were EBD-positive by intracellular staining, with <10% cell surface staining; and approximately 14% of LM7 cells were EBD-positive by cell surface and intracellular staining (Supplementary Fig. S6A). Normal fibroblast and fibronectin-knockout U87 cells (U87^FN-/) served as negative controls. ELISAs for fibronectin demonstrated a significant concentration of fibronectin in LM7 cells and normal fibroblasts compared with U87^FN−/− controls in media and cell lysates of U87, A549, and A673 cells (Supplementary Fig. S6B and S6C). EDB-CAR T cells produced significant amounts of IFNγ and IL2 in 48-hour coculture assays with all four cell lines compared with NT T cells (Fig. 1F). EDB-CAR T cells also had significant cytolytic activity, in contrast to NT T cells, in a standard MTS-based cytotoxicity assay (Fig. 1G). In contrast, EDB-CAR T cells had no cytolytic activity against primary human fibroblasts even at high E:T ratios of 16:1 (Fig. 1H). EDB-CAR T cells did not recognize primary human fibroblasts, determined by IFNγ production in coculture assays (Supplementary Fig. S7).

To provide further evidence that target cell recognition depended on the expression of a functional EDB-CAR in T cells and EDB in target cells, we conducted experiments with the U87^FN−/− cells as targets and generated a nonfunctional EDB-CAR by mutating the three ITAMs in the zeta signaling domain of the CAR (mutEDB-CAR). mutEDB-CAR T cells did not recognize or kill U87 cells, in contrast to EDB-CAR in T cells (Fig. 1I), and U87^FN−/− cells were not recognized or killed by EDB-CAR T cells (Fig. 1J). Thus, EDB-CAR T cells recognized and killed EDB-positive tumor cells in an EDB-specific manner.

Because EDB is secreted, we wanted to address if there was bystander killing of targets that do not express EDB (Supplementary Fig. S8A). First, we demonstrated that conditioned media of U87 cells contained EDB, which could activate EDB-CAR T cells, as judged by IFNγ production, in contrast to conditioned media from U87^FN−/− cells (Supplementary Fig. S8B). Next, we cocultured EDB-positive (U87) or EDB-negative (U87^FN−/−, fibroblasts) target cells with NT or EDB-CAR T cells in the presence of conditioned media from EDB-positive or EBD-negative cells and performed an MTS assay. EDB-positive U87 cells were killed by EDB-CAR T cells regardless of the added conditioned media. In contrast, significant killing of EBD-negative target cells was only observed in the presence of conditioned media from EDB-positive cells (Supplementary Fig. S8C).

Having established that EDB-CAR T cells had significant EDB-specific antitumor activity in vitro, we next set out to evaluate their antitumor activity in vivo. We utilized three subcutaneous (U87, A673, and A549) and one intravenous lung (A549.GFP.ffluc) NSG tumor models and confirmed fibronectin (Supplementary Fig. S9) and EDB (Supplementary Fig. S10) expression in the subcutaneous tumors by IHC. On day 7 or 10 after tumor cell injection, mice received a single i.v. dose of 1 × 10⁶ NT or EDB-CAR T cells. In all models, EDB-CAR T cells had antitumor activity in comparison with mice injected with NT T cells, resulting in a significant survival advantage (Fig. 2A–F; Supplementary Fig. S11). To confirm that the observed antitumor activity depended on EDB expression by tumor cells, we used U87^FN−/− cells in the subcutaneous model. Although U87^FN−/− tumors readily grew in NSG mice, EDB-CAR T cells had no antitumor activity (Fig. 2G and H). To further confirm specificity and to get insight into the in vivo expansion of EDB-CAR T cells, we performed an in vivo experiment with mutEDB- and EDB-CAR T cells, which were also genetically modified to express GFP.ffluc, in the subcutaneous A549 model (Supplementary Fig. S12A). Only EDB-CAR T cells had significant antitumor activity (Supplementary Fig. S12B and S12C) and expanded and/or persisted longer than mutEDB-CAR T cells, determined by AUC analysis within the first 10 days after CAR T-cell infusion (Supplementary Fig. S12D). To assess long-term, functional persistence of EDB-CAR T cells, 5 mice that achieved a complete response (2 mice: subcutaneous A549 model; 3 mice: subcutaneous U87 model) were challenged with a second dose of their respective tumor cells on day 100. Five naïve mice served as controls. Tumors grew in 5 of 5 control mice, in contrast to only 1 of 5 mice that had previously been treated with EDB-CAR T cells (Fig. 2I).

Because we demonstrated “bystander killing” in our in vitro studies, we next assessed this in vivo by injecting admixtures of U87 and U87^FN−/− cells injected subcutaneously into mice (100%, 90%, 70%, 10%, and 0% U87 cells). On day 7, mice received a single dose of EDB-CAR T cells. Although the antitumor activity decreased with decreasing percentage of U87 cells, EDB-CAR T cells still had significant antitumor activity against 10% U87 tumors (Supplementary Fig. S13A–S13C).

There was a significant higher incidence of macroscopic necrosis of subcutaneous EDB-positive tumors <2,000 mm³, determined by visual inspection in mice that received EDB-CAR T cells (Fig. 2J). We therefore wanted to evaluate if EDB-CAR T cells targeted the tumor vasculature using two independent approaches. First, we injected intravenous NT or EDB-CAR T cells into subcutaneous A549 or A673 tumor–bearing mice. On day 10 or 14 after T-cell injection, tumors were processed for CD31 via IHC to enumerate endothelial cells. As an additional control, EphA2-CAR T cells were included in the subcutaneous A549 model. There was a significant reduction of intratumoral CD31-positive endothelial cells with EDB-CAR T-cell treatment in comparison with NT and EphA2-CAR T-cell treatment groups (Fig. 3A–D; Supplementary Fig. S14). The second approach relied on direct imaging of the tumor vasculature using AngioSense on day 14 after NT or EDB-CAR T-cell injection in the subcutaneous A673 model. Our results demonstrated that A673 tumors treated with EDB-CAR T cells contained a lower number of blood vessels (Fig. 3E–G).

To confirm that EDB-CAR T cells targeted vascular endothelial cells, we performed experiments with HUVEC cells and freshly isolated CD31-positive endothelial cells from U87 and A673 tumors. HUVEC cells expressed EDB by flow cytometric analysis (Supplementary Fig. S15A) and were recognized and killed by EDB-CAR T cells (Supplementary Fig. S15B and S15C). NT T cells served as negative, and EphA2-CAR T cells as positive controls. EDB-CAR T cells also killed capillaries formed by HUVEC cells in contrast to NT T cells (Supplementary Fig. S15D). EphA2-CAR T cells did not kill preformed capillaries because the CAR recognizes an epitope that is not accessible in cell structures that form tight junctions (22). EDB-CAR T cells also recognized and killed CD31-positive endothelial cells (Supplementary Fig. S16A–S16C).

To establish if endothelial cells produced sufficient EDB to become targets or if their killing relied on binding of tumor cell–secreted EDB on their cell surface, we took advantage of our U87 and U87^FN−/− models. U87 and U87^FN−/− subcutaneous tumor-bearing mice received on day 7 one intravenous dose of NT or EDB-CAR T cells. On day 10 after T-cell injection, mice were euthanized and tumors processed for CD31 IHC. In U87 tumors, EDB-CAR T cells induced a significant reduction of intratumoral CD31-positive endothelial cells in comparison with NT T cells (Fig. 3H, I, and K; Supplementary Fig. S17A). However, we observed no significant difference between the NT and EDB-CAR T-cell treatment groups in U87^FN−/− xenografts (Fig. 3J and K; Supplementary Fig. S17B), indicating that EDB secretion by neighboring tumor cells was essential for the observed antivascular activity of EDB-CAR T cells.

Because murine and human EDB are 100% identical (9), we also used our NSG tumor model to assess safety. Injection of 1 × 10⁶ EDB-CAR T cells did not result in weight loss, and long-term follow-up showed continued weight gain of treated mice (Supplementary Fig. S18). We performed additional experiments with GFP.ffLuc-expressing EDB-CAR T cells to assess their expansion in non–tumor-bearing mice. Non–tumor-bearing mice received one i.v. dose of 1 × 10⁷ NT or 1 × 10⁶ or 1 × 10⁷ EDB-CAR T cells. Only after the injection of 1 × 10⁷ EDB-CAR T cells did we observed significant, transient expansion (Fig. 4A–D) in comparison with NT T cells. Mice that received 1 × 10⁷ NT or EDB-CAR T cells were euthanized on day 14, and major organs (kidney, liver, lung, and spleen) were subjected to pathologic analysis. No histologic differences were observed between examined tissues of both groups of mice (Fig. 4E).

Collectively, our studies demonstrated that T cells, expressing an EDB-CAR with an antigen recognition domain derived from the L19 mAb, recognized and killed EDB-secreting tumor cells. EDB-CAR T cells had potent antitumor activity in three xenograft models representing lung cancer, sarcoma, and high-grade glioma. EDB-secreting tumor cells also redirected EDB-CAR T cells to CD31-positive endothelial cells of the tumor vasculature.

Fibronectin is encoded by 47 exons and has 27 splice variants according to GTEx. EDB is encoded by exon 25, and 4 of 27 splice variants contain this exon. EDB is expressed in a broad range of malignancies, which we confirmed using 4,684 TCGA samples. Among 31 types of cancers, expression was high in common cancers, such as breast and lung cancers, and in cancers with poor prognosis, including pancreatic adenocarcinoma and high-grade glioma. Although cancer-specific secretion of EDB has been exploited by numerous investigators for imaging purposes, targeted radiotherapy, or targeted delivery of anticancer drugs including cytokines (16, 23), very few studies have been conducted that induce EDB-specific immune responses. One study finds that a prophylactic EDB vaccine reduces tumor growth and increases tumor necrosis without impairing wound healing in a preclinical fibrosarcoma model (24). Another study demonstrates that EDB-CAR T cells target tumor cells and the tumor vasculature in immune-competent animal models. However, no antitumor activity was observed in immunodeficient mice (25).

Here, we demonstrated in three tumor models using immunodeficient mice that tumor cells can be directly targeted with EDB-CAR T cells. For our studies, we used tumor cells for which we confirmed EDB and fibronectin expression by qPCR and flow cytometry (EDB) or ELISA (fibronectin) using U87^FN−/− and fibroblasts as controls. Three of four cell lines secreted significant amounts of fibronectin in comparison with U87^FN−/−. Fibroblast secreted only low amounts of fibronectin, which could be explained by either our culture conditions (26) or the presence of other ECM molecules (like collagen) secreted by fibroblasts that bind fibronectin and interfere with the performed ELISAs. Although by qPCR all four cell lines expressed EDB, flow cytometric analysis revealed that only 2 of 4 cell lines expressed cell surface EDB and 3 of 4 cell lines had intracellular expression We believe that these discrepancies are due to technical issues because all four cell lines were recognized and killed by EDB-CAR T cells, and we confirmed the specificity of our CAR by using EDB-negative fibroblasts and U87^FN−/− cells.

Being able to demonstrate antitumor activity by EDB-CAR T cells in vivo in xenograft models presents a significant advance because it highlights the feasibility of directly targeting cancer cells with CAR T cells that recognize secretory proteins that adhere to the cell surface. Our results also suggested bystander killing of EDB-negative tumor cells, presenting a distinct advantage in comparison with targeting membrane proteins. Other investigators have not observed antitumor activity of EDB-CAR T cells in xenograft models (7). These discrepant results are most likely explained by differences in CAR design: whereas the EDB-binding domain of the published CAR consists of a heavy-chain–only antibody (VHHs or nanobody), we used a standard scFv encoding the EDB-specific L19 mAb. The EDB-specific nanobody and L19 scFv most likely bind to distinct epitopes within the EDB domain.

EDB-CAR T cells had significant antivascular activity, which we confirmed by two independent methods. In vivo studies with U87^FN−/− cells revealed that EDB secretion by tumor cells was critical for the observed antivasculature activity. This is in contrast to several studies reporting that EDB is secreted by endothelial cells within the tumor vasculature (12). Because human and murine EDB are 100% identical, the murine origin of endothelial cells within xenograft tumors does not explain our finding (11). Possibly, lack of fibronectin secretion by the tumor cells could influence blood vessel formation. Indeed, the number of CD31-positive endothelial cells was lower in U87^FN−/− than U87 tumors of mice that received NT T cells; however, this difference did not reach statistical significance. Clearly, additional studies in other models are needed to further investigate our findings.

Although EDB-CAR T cells had antitumor activity, only in the U87 model, 40% of animals were cured with a follow-up of 3 months. Limited antitumor activity might be explained by several factors. First, we used low numbers of CAR T cells (1 × 10⁶ per mice) in all of our therapeutic models. We also observed limited expansion of EDB-CAR T cells after infusion. In the future, we will therefore focus on optimizing CAR T-cell dosing and CAR design, and explore additional genetic modification of EDB-CAR T cells to enhance their effector function as reported in other model systems (27, 28). These include expression of cytokines and/or chimeric cytokine receptors, or deleting negative regulators (29–32). Combining EDB-CAR T cells with CAR T cells that target cell surface TAAs is another possibility. Last, studies in immune-competent tumor models are warranted, which have a TME that is closer to human tumors than the TME in NSG mice.

Several other CARs have been developed that target the tumor vasculature, including CARs targeting VEGFR1, VEGFR2, PSMA, and TEM8 (33–35). Although VEGFR2-CAR T cells had promising antitumor activity in preclinical models, clinical activity (clinicaltrials.gov website, NCT01218867) is limited, and systemic delivery of TEM8-CAR T cells in one preclinical model is toxic (35). Because human and murine EDB are 100% homologous, our xenograft studies allowed us to evaluate the safety of EDB-CAR T cells. First, we did not observe any clinical signs of toxicity, including weight loss, in mice treated with one intravenous dose of EDB-CAR T cells in any of our efficacy studies. Second, we performed studies in non–tumor-bearing mice, in which mice received a single 10-fold higher dose of GFP.ffLuc-expressing EDB-CAR T cells. Although there was transient, significant expansion at the 1 × 10⁷ cell dose, we did not observe any overt toxicity, and histologic examination of major organs 14 days after injection revealed no pathologic findings, indicating our EDB-CAR T cells are well-tolerated with no overt on-target/off-tumor toxicity. At present, it is unclear if the observed transient expansion at the 1 × 10⁷ CAR T-cell dose is due to the presence of low EDB in selected normal tissues or alloreactivity, and additional studies are needed and should also include careful examination of femur and hips because EDB-CAR T-cell expansion was observed at these sites by bioluminescence imaging.

Finally, our study highlights that a cancer-specific splice variant can be safely targeted with CAR T cells in vivo. Aberrant splicing is a “hallmark of cancer” (8), and numerous splice variants have been described (36). Thus, we believe that our study provides the impetus to carefully mine existing gene expression databases for splice variants that could serve as CAR targets. In conclusion, EDB-CAR T cells had potent antitumor and antivascular activity in preclinical solid tumor models without overt toxicity. Thus, targeting EDB with CAR T cells has the potential to improve current CAR T-cell therapy approaches for solid tumors.

J. Wagner reports grants from St. Jude Children's Research Hospital during the conduct of the study, as well as a patent for CARs for direct and indirect targeting of fibronectin + tumors pending. T.I. Shaw reports a patent for EBD CAR - SJ-19-0029 pending. J. Zhang reports a patent for EBD-CAR T cells pending. S. Gottschalk reports grants from NIH (P01CA096832 and R50CA211481), NCI (P30CA021765), Alex's Lemonade Stand Foundation, Alliance for Cancer Gene Therapy, and American Lebanese Syrian Associated Charities during the conduct of the study; personal fees from Immatics and grants from TESSA Therapeutics outside the submitted work; and a patent for CARs for direct and indirect targeting of fibronectin + tumors pending. No disclosures were reported by the other authors.

The content is solely the responsibility of the authors and does not necessarily represent the official views of the NIH.

J. Wagner: Conceptualization, data curation, formal analysis, funding acquisition, investigation, methodology, writing–original draft, project administration, writing–review and editing. E. Wickman: Data curation, formal analysis, investigation, writing–review and editing. T.I. Shaw: Data curation, software, investigation, methodology, writing–review and editing. A. Allo Anido: Investigation, methodology, writing–review and editing. D. Langfitt: Investigation, writing–review and editing. J. Zhang: Supervision, funding acquisition, investigation, methodology, writing–review and editing. S.N. Porter: Investigation, writing–review and editing. S.M. Pruett-Miller: Funding acquisition, investigation, methodology, writing–review and editing. H. Tillman: Investigation, methodology, writing–review and editing. G. Krenciute: Formal analysis, supervision, funding acquisition, writing–review and editing. S. Gottschalk: Conceptualization, resources, data curation, supervision, writing–original draft, project administration, writing–review and editing.

The authors thank the research staff of St. Jude Centers and Cores for their assistance with the conducted experiments.

The work was supported by the Alex Lemonade Stand Foundation and Cure4Cam Foundation (ALSF; Young Investigator Grant; J. Wagner), the Alliance for Cancer Gene Therapy (ACGT; S. Gottschalk), and the American Lebanese Syrian Associated Charities (G. Krenciute and S. Gottschalk). Animal imaging was performed by the Center for In Vivo Imaging and Therapeutics, which is supported in part by NIH grants P01CA096832 and R50CA211481. Cellular images were acquired at St. Jude Children's Research Hospital Cell & Tissue Imaging Center, which is supported by St. Jude Children's Research Hospital and NCI P30 CA021765. Gene editing of cell lines was performed by the CAGE, which is supported in part by NCI P30 CA021765.

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.