Selective Targeting of Glioblastoma with EGFRvIII/EGFR Bitargeted Chimeric Antigen Receptor T Cell

Glioblastoma is the most common and aggressive brain malignancy without effective treatment options (1, 2). In the United States, approximately 10,000 new cases are diagnosed annually (3). Standard-of-care procedures include surgical resection followed by radiotherapy and/or chemotherapy with the alkylating agent temozolomide (4). Despite therapeutic advances, the prognosis of glioblastoma remains very poor, with a median survival of less than 2 years (5). The resistance of glioblastoma to standard therapies results in a high rate of tumor recurrence, and the lack of tumor specificity of the treatment may result in a significant damage to healthy brain tissues (6). There is consequently a high unmet medical need for a safe and efficacious tumor-selective therapy against glioblastoma.

Chimeric antigen receptor (CAR)–engineered T (CAR T) cells are a promising strategy for cancer treatment (7). A classic CAR T structure includes an extracellular tumor-targeting single-chain variable fragment from an antibody (scFv domain), a short transmembrane domain, and a tandemly assembled intracellular costimulatory domain with intracellular T-cell signaling moieties. Preclinical and clinical studies have shown CAR T cells redirected to IL13Rα2 and HER2 could induce tumor regression in glioblastoma (8–11), suggesting that CAR T cells could be developed as next-generation therapeutics for glioblastoma treatment.

Epidermal growth factor receptor (EGFR) is amplified in approximately 40% to 60% of glioblastomas (12, 13). EGFRvIII, a cancer-specific variant of EGFR, is found in approximately 31% of glioblastoma patients (12). Antibodies to EGFR and small-molecule inhibitors targeting EGFR have been used to treat patients with glioblastoma, but were not clinically efficacious (14, 15). CAR T cells targeting EGFRvIII can selectively eliminate EGFRvIII-expressing glioblastoma cells (14, 16, 17). Although CAR T cells efficiently infiltrated and eliminated EGFRvIII tumor cells in some patients, neither partial nor complete responses were achieved in the clinical trials (18).

An increasing body of data suggests that intratumoral heterogeneity is one of the principal causes of tumor relapse (19, 20). The poor clinical outcome of glioblastoma patients treated with EGFRvIII-targeted CAR T cells might be attributed, in part, to the fact that EGFRvIII was only expressed in a subpopulation of tumor cells (21, 22). We hypothesized that bitargeted CAR T cells that recognize both wild-type EGFR and EGFRvIII overexpressed by glioma cells would show better efficacy than CAR T cells targeting EGFRvIII alone. Monoclonal antibody 806 (mAb 806) targets a conformationally exposed epitope, EGFR^287-302, that is found in both wild-type EGFR that is expressed on tumor cells and the mutated oncogenic form of EGFR, EGFRvIII (23, 24). ABT-806, a humanized variant of mAb 806, shows minimal skin toxicity in treated patients (25). ABT-414, an antibody–drug conjugate composed of ABT-806 and a potent antimicrotubule agent, monomethyl auristatin F (MMAF) had no skin toxicity in human studies with an objective response rate of approximately 6.8% (26). This evidence suggests that the immune-oncology therapeutics targeting EGFR^287–302 have potential for improved safety and efficacy. Therefore, here we generated CAR T cells recognizing EGFR^287–302 and characterized their safety and efficacy for the treatment of glioblastoma.

The human glioblastoma cell lines U87MG, U251, human epidermoid carcinoma cell line A431, human oral adenosquamous carcinoma cell line CAL27, and the human embryonic kidney cell line 293T were obtained from ATCC. Human hepatocyte L02 cell and human prostate epithelial RWPE-1 cell were purchased from the Cell Bank of the Chinese Academy of Sciences (Shanghai, China). U87MG-EGFR and U251-EGFR cells stably expressing human EGFR protein were generated in our laboratory by transducing the U87MG and U251 cells with a VSV-G pseudotyped lentiviral vector encoding wild-type EGFR. U87MG-EGFRvIII and U251-EGFRvIII cells stably expressing EGFRvIII protein were generated in our laboratory by transducing the U87MG and U251 cells with a VSV-G pseudotyped lentiviral vector encoding exon 2–7-deleted EGFR. U251-luci-EGFR and U251-luci-EGFRvIII cells expressing firefly luciferase were also established by lentiviral transduction. All cells were cultured in Dulbecco's Modified Eagle Medium with 10% FCS (Life Technologies) with 100 μg/mL penicillin and 100 U/mL streptomycin (Life Technologies). Human primary keratinocytes were isolated from healthy human skin and cultured in EpiLife Medium with 60 μmol/L calcium (Life Technologies) with Human Keratinocyte Growth Supplement (HKGS; Life Technologies). The cell lines were authenticated by using short tandem repeat analysis. Mycoplasma contamination testing was routinely performed by using PCR every 3 months in our laboratory. These cell lines had been cultured for 3 to 4 years in our lab and were cryopreserved to create a working cell bank in which each vial of cells was subject to subculture for up to 4 weeks after recovery. All cells were maintained at 37°C in humidified air with 5% CO₂.

M27-scFv, an anti–EGFR-specific single-chain variable fragment (scFv), was derived from the humanized EGFR antibody M27 that recognizes the residues 287–302 of EGFR. The sequence encoding the M27-scFv antibody in the VL-VH orientation was cloned by PCR amplification from a plasmid encoding scFv-M27-Fc. Another EGFR-specific scFv, 806-scFv, was derived from the chimeric antibody ch806 (24) that binds amino acids 287 to 302 of EGFR as well. The sequence encoding the 806 scFv antibody in the VL-VH orientation was obtained by PCR amplification from a plasmid encoding scFv-806-Fc. M27-28BBZ and 806-28BBZ CARs contained extracellular domains of the human CD8α signal peptide (nucleotides 1–63, GenBank NM 001768.6), M27-scFv or 806-scFv, CD8α hinge (nucleotides 412–546, GenBank NM 001768.6), and CD28 transmembrane domain (nucleotides 457–537, GenBank NM 006139.3). The intracellular domains consisted of CD28 (“28”; nucleotides 538–660, GenBank NM 006139.3), CD137 (4-1BB, “BB”; nucleotides 640–765, GenBank NM 001561.5) costimulatory domain and CD3ζ (“Z”; nucleotides 154–492, GenBank NM 198253.2) costimulatory polypeptide. M27-28Z CAR had an extracellular domain that consisted of human CD8α signal peptide (nucleotides 1–63, GenBank NM 001768.6), M27-scFv, and a CD8α hinge (nucleotides 412–546, GenBank NM 001768.6). The intracellular signaling domain of M27-28Z CAR consisted of CD28 (nucleotides 538–660, GenBank NM 006139.3) and CD3ζ (nucleotides 154–492, GenBank NM 198253.2) molecules. The 28BBZ and 28Z fragments were produced by PCR amplification with a plasmid template encoding the corresponding fragments (27). The DNA sequences of scFv-28BBZ or scFv-28Z were further cloned by PCR amplification. Both of these fragments were designed to have a MluI site at the 5′ end and a SalI site at the 3′ end. The synthesized fragments were digested with MluI and SalI restriction enzymes (New England Biolabs) and then inserted into a similarly digested pRRLSIN.cPPT-GFP.WPRE vector plasmid. The sequence integrity of all the vectors described in this paper was confirmed by DNA sequencing. The mock construct was transduced by using the pRRLSIN.cPPT-GFP.WPRE lentiviral vector.

Human embryonic kidney 293T cells were seeded at 1.5 × 10⁷ per 15-cm dish for 24 hours prior to transfection. Expression vector pRRLSIN.cPPT-GFP.WPRE (mock), or pRRLSIN-M27-28Z, or pRRLSIN-M27-28BBZ, or pRRLSIN-806-28BBZ was mixed with two lentiviral packaging plasmids pMDLg/pRRE and pRSV-Rev plus an envelope expressing plasmid pCMV-VSV-G (from Addgene) to reconstitute a transfection DNA mixture in the polyethylenimine-based DNA transfection reagent. 293T cells were transfected with the reconstituted DNA mixture as mentioned above. The viral supernatants were harvested and filtered (0.45-μm filter) at 72 hours after transfection. The lentiviral particles were subsequently concentrated 30-fold by ultracentrifugation (Beckman Optima XL-100 K) for 2 hours at 28,000 rpm.

Peripheral blood mononuclear cells (PBMC) were cultured in AIM-V medium (Life Technologies) in the presence of 2% human AB serum (Huayueyang Biotechnology) and recombinant human IL2 (Huaxin High Biotechnology). For the transduction of primary T cells, PBMCs were activated for 48 hours by Dynabeads Human T-Activator CD3/CD28 (Life Technologies) at a bead:cell ratio of 2:1 before infection. The activated T cells were transduced with lentiviral vectors at a multiplicity of infection of 15 in a 24-well plate coated with RetroNectin (Takara). The transduced T cells were cultured at a density of 5 × 10⁵ cells/mL in the presence of rhIL2 (500 IU/mL).

To determine the binding of scFv to target cells, 1 × 10⁶ cells were incubated with 5 μg/mL scFv-M27-Fc, scFv-806-Fc, or scFv-C225-Fc antibody for 45 minutes at 4°C. PBS was used as a negative control. After washing with FACS buffer (cold phosphate-buffered saline containing 1% newborn calf serum), the cells were incubated with FITC-conjugated goat antihuman IgG (H + L; Catalog Number: SA00003-12; Proteintech) for 45 minutes at 4°C. To quantify CAR expression, CAR T cells were incubated with 20 μg/mL biotinylated antihuman-EGFR-F(ab′)2 fragments at 4°C for 45 minutes. PBS was used as a negative control. After FACS buffer wash, the cells were incubated with PE-conjugated streptavidin (eBioscience) for 45 minutes at 4°C. Fluorophore-labeled CAR T cells were determined by using a BD FACSCelesta flow cytometer. Data were analyzed using FlowJo 7.6 software. Data are representative of three independent experiments.

To examine the persistence of CAR T cells, whole mouse blood collected by retro-orbital bleeding was analyzed as following: 50 μL of blood was incubated with the antibody against CD3-PerCP/CD4-FITC/CD8-PE in the dark for 15 minutes at room temperature. After lysis of red blood cells, cells were fixed with 0.45 mL 1 × BD FACS Lysing Solution (BD Biosciences) for 15 minutes at room temperature. The fixed cells were then subjected to flow-cytometric analysis by using FACSCelesta flow cytometer (BD Biosciences). Data from flow cytometry were further analyzed by using FlowJo 7.6 software. Absolute cell numbers per microliter of blood were determined by using TruCount tubes (BD Biosciences) as described in the manufacturer's instruction.

Each line of glioma cells was cocultured with target-selective CAR T cells or mock T cells at effector:target (E:T) ratios of 3:1, 1:1, and 1:3. After 18-hour culture, supernatant lactate dehydrogenase (LDH) activity was determined by using the CytoTox 96 Nonradioactive Cytotoxicity Kit (Promega) as previously described (27).

CAR T cells or mock T cells were cocultured with glioma cells at a 3:1 ratio in a 96-well culture plate. After 24-hour coculture, the release of IFNγ, TNFα, and IL2 cytokines from activated CAR T cells or from mock T cells was determined by using an ELISA kit (MultiSciences Biotech) as described in the manufacturer's instructions.

To generate subcutaneous xenograft models, 6- to 8-week-old female NOD/SCID mice were subcutaneously inoculated with 3 × 10⁶ U251-EGFR or 2 × 10⁶ U251-EGFRvIII cells at the right flank. After the tumor volume increased to 75 to 120 mm³, the mice were randomly divided into two groups (n = 6–10 per group) for receiving treatment regimen. The control group of mice received intravenous (i.v.) injection of mock T cells (activated T cells transduced with eGFP-CAR). The treatment group received i.v. injection of M27-28BBZ CAR T cells (activated T cells transduced with EGFR CAR). Twenty-four hours prior to CAR T-cell infusion, the mice were intraperitoneally (i.p.) injected with 100 mg/kg cyclophosphamide to deplete host lymphocytes and to enhance the tumor treatment efficacy of the administered T cells (28). Two doses of 1 × 10⁷ M27-28BBZ CAR T cells were i.v. injected via the tail vein in 200 μL of PBS on days 14 and 17. Tumors were measured by using calipers, and tumor volumes were calculated by using the formula V = (length × width²)/2. Animal body weights and tumor volume measurements were carried out twice weekly. The mice were euthanized when their body weight loss was greater than 20% of the initial weight, the mean tumor volume exceeded 2,000 mm³, or the tumors became ulcerated in the mock control groups.

To generate the U251-luci-EGFR orthotopic model, 1 × 10⁶ U251-luci-EGFR cells were intracranially implanted into 6- to 8-week-old female NOD/SCID mice (n = 8 mice per group). To generate the U251-luci-EGFR/U251-luci-EGFRvIII orthotopic model, the mixture of 7.5 × 10⁵ U251-luci-EGFR and 2.5 × 10⁵ U251-luci-EGFRvIII cells were intracranially implanted into 6- to 8-week-old female NOD/SCID mice (n = 7 mice per group). To generate the U251-luci-EGFRvIII orthotopic model, 5 × 10⁵ U251-luci-EGFRvIII cells were intracranially implanted into 6- to 8-week-old female NOD/SCID mice (n = 6 mice per group). Implantation was performed by using a stereotactic surgical device with injection of tumor cells at 1 mm right and at 1 mm anterior to the bregma and at 3 mm into the brain on day 0. Seven days after surgery, the mice were randomly divided into two groups. On day 8, the mice were intravenously injected with a single dose of 1 × 10⁷ CAR T cells in 200 μL of PBS via the tail vein. Bioluminescent measurements were used as surrogates for tumor volume. The transduction efficiency of CAR T cells used in the assays was ∼50%. All animal experiments were performed by following the protocol approved by the Shanghai Cancer Institute Experimental Animal Care Commission.

Bioluminescence imaging was performed by using the IVIS system (IVIS, Xenogen). Briefly, tumor-bearing mice were intraperitoneally injected with d-luciferin (150 mg/kg) and imaged under isoflurane anesthesia after 10 minutes. The images were quantified by using Living Image software (Caliper Life Sciences).

To assess the infiltration of human T cells into tumors, formalin-fixed and paraffin-embedded tumor tissues were immunostained by using anti-CD3 (Thermo Fisher Scientific). A normal rabbit IgG was used as an isotype-matched control. The procedures were performed as previously described (27). Briefly, after deparaffinization and rehydration, the sections were exposed to 3% H₂O₂ in methanol to eliminate endogenous peroxidase activity and then blocked with bovine serum albumin (1%) for 30 minutes at room temperature (RT). After blocking, the sections were incubated with primary rabbit antihuman CD3 mAb overnight at 4°C. After PBS wash, the sections were incubated with peroxidase-conjugated secondary antibodies (ChemMate DAKO EnVision Detection Kit, Peroxidase/DAB, Rabbit/Mouse) for 45 minutes at RT. The sections were visualized by using a diaminobenzidine staining kit (Dako Corporation) and then counterstained with hematoxylin, dehydrated, cleared, mounted, and photographed. The DAB-immunostained sections were analyzed by bright-field microscopy using an Olympus microscope (OLYMPUS IX71) equipped with an image analysis software. CD3⁺ cells were quantified by measuring the number of stained T cells in each section by using the Image-Pro Plus (Media Cybernetics) software. Sections from 3 mice in each group were subjected to determination of T-cell infiltration for statistical analysis. The mean count of the three areas was obtained and expressed as the absolute number of CD3⁺ cells per 0.95 mm² (200× magnification).

All data were presented as the mean ± SE. Statistical significance was determined by using a two-way or one-way ANOVA with Bonferroni posttest for multisample comparisons or the unpaired two-tailed Student t test for comparisons between two samples. The overall survival statistics were calculated by using the log-rank test. GraphPad Prism 5.0 was used for statistical calculations. P < 0.05 was considered statistically significant.

To avoid anaphylaxis caused by mouse scFv-derived CAR T cells (29), we generated a panel of humanized antibody fragments derived from a mouse EGFR monoclonal antibody 7B3 that specifically bound EGFR^287-302 (Supplementary Table S1; Supplementary Fig. S1). EGFR- and EGFRvIII-transfected U87MG and U251, two cancer cell lines with endogenous EGFR overexpression (A431 and CAL27), normal cells L-02 (Human hepatocyte cell) and RWPE-1 (human prostate epithelial cell) as well as primary keratinocytes were used in the following biological assays. The expression of EGFR was confirmed by Western blot (Supplementary Fig. S2). After two rounds of affinity maturation and binding specificity enrichment, M27 scFv clone was selected for further characterization of its binding specificity. The scFv derived from anti-EGFR antibody C225 (cetuximab) and mAb 806 were used as controls (Supplementary Fig. S3). Flow-cytometric analysis showed that C225 scFv bound to most of the cells except for U87MG cells (Fig. 1A). The binding curves of C225 scFv to U251 cells and to the three healthy keratinocytes were similar. 806 scFv showed observable binding on EGFR- and EGFRvIII-overexpressing U87MG and U251 cells and three primary keratinocytes. The mean fluorescence intensity (MFI) of 806 scFv on U87MG-EGFRvIII, U251-EGFRvIII, U87MG-EGFR, and U251-EGFR cells was 22.6 ± 0.42, 58.1 ± 0.85, 1.57 ± 0.05, and 9.81 ± 0.08, respectively. The MFI of 806 scFv on three primary keratinocytes was 0.71 ± 0.04, 0.57 ± 0.02, and 0.60 ± 0.03. Humanized M27 scFv showed distinguishable binding on EGFRvIII-overexpressing cells and EGFR-overexpressing U87MG and U251 cells. The MFI of M27 scFv on U87MG-EGFRvIII, U251-EGFRvIII, U87MG-EGFR, and U251-EGFR cells was 12.2 ± 0.31, 19.4 ± 0.50, 0.39 ± 0.01, and 0.78 ± 0.02, respectively. The EC₅₀ of M27 and 806 scFv on U87 MG-EGFR cell was 3.9 and 2.4 μmol/L respectively, whereas the EC₅₀ of M27 and 806 scFv on U87 MG-EGFRvIII cell was 99.5 nmol/L and 66.4 nmol/L, respectively (Supplementary Fig. S4). These results indicated that M27 scFv had a lower affinity to EGFR- or EGFRvIII-overexpressing cells than 806 scFv. However, M27 scFv did not bind to any of the three primary keratinocytes tested. In addition, all anti-EGFR scFv demonstrated binding to endogenous EGFR-overexpressing cell lines A431 and CAL27. M27 scFv, however, showed relatively less affinity than other two scFv in binding to A431 and CAL27 (Fig. 1A). The MFI of different scFv proteins on the cells tested is shown (Fig. 1B). Additionally, we also evaluated the binding ability of M27 scFv to normal cells L-02 (human hepatocyte cell) and RWPE-1 (human prostate epithelial cell), which endogenously express EGFR. The results showed that M27 scFv also could not bind to these two cells (Supplementary Fig. S5). These data indicate that the M27 scFv specifically bound EGFR and EGFRvIII overexpressed on tumor cells but not EGFR expressed on normal cells.

We generated a series of recombinant lentiviral vectors encoding various CAR cassettes that had an EGFR^287–302 scFv antibody, an intracellular human T-cell costimulatory domain derived from human CD28, CD137, and CD3ζ chains, and a linker domain of human CD8 hinge and CD28 transmembrane regions (M27-28BBZ and 806-28BBZ). We also constructed a CAR cassette lacking the intracellular CD137 signaling domain (M27-28Z). The scheme of recombinant lentiviral constructs is shown in Fig. 2A.

To determine the expression of the EGFR CAR on the genetically modified T-cell surface, different CAR-transduced T cells were determined by flow cytometry using biotinylated human-EGFR-F(ab′)2 fragment antibody and PE-conjugated streptavidin. The transduction efficiency ranged from 51.9% to 74.8%. As a control, mock T cells were transduced with the lentiviral vector encoding GFP gene. The transduction efficiency of mock T cells was determined by eGFP expression (Fig. 2B).

To determine whether CAR T cells targeting EGFR/EGFRvIII could selectively recognize and eliminate EGFR/EGFRvIII-positive human glioblastoma cells, cytotoxicity assays were performed by incubating the genetically modified CAR T cells with either control cells or EGFR- or/and EGFRvIII-overexpressing glioblastoma cells. The results showed that both M27-28Z and M27-28BBZ CAR T cells efficiently lysed the EGFR/EGFRvIII-overexpressing glioblastoma cells but not untransfected U87MG and U251 cells (Fig. 2C and D). Mock T cells showed very weak cytotoxicity compared with M27-derived CAR T cells (Fig. 2C). The M27-28BBZ CAR T cells showed higher cytotoxic activities than M27-28Z CAR T cells in lysis of both EGFR- and EGFRvIII-overexpressing U87MG or U251 cells at 1:1 and 3:1 E:T ratio (P < 0.05; Fig. 2C and D). These data suggested that M27-28BBZ CAR T cells would be a better candidate for further evaluation and development.

To determine the selective cytotoxicity of M27-28BBZ CAR T cells in lysis of tumor cells, CAR T cells were incubated with primary keratinocytes and EGFR/EGFRvIII-overexpressing tumor cells, respectively. 806 scFv-derived 806-28BBZ CAR T cells were also included as a control. The data showed that both 806- and M27-derived CAR T cells could efficiently lyse EGFR- and EGFRvIII-overexpressing U87MG and U251 cells. The cytotoxicities of both CAR T cells were similar in lysis of U251-EGFRvIII cells. 806-28BBZ CAR T cells showed slightly higher cytotoxic effect on U87MG-EGFRvIII cells than on M27-28BBZ CAR T cells (806-28BBZ vs. M27-28BBZ at 1:3 E:T: 24.3% ± 2.0% vs. 16.2% ± 0.9%; 806-28BBZ vs. M27-28BBZ at 1:1 E:T: 50.2% ± 3.7% vs. 43.0% ± 2.5%; 806-28BBZ vs. M27-28BBZ at 3:1 E:T: 68.9% ± 4.4% vs. 57.8% ± 4.7%; Fig. 3A and B). Unlike M27-28BBZ, 806-28BBZ CAR T cells demonstrated a cytotoxic effect on untransfected U251 cells at a 3:1 E:T ratio. At a 3:1 E:T ratio, both M27-28BBZ and 806-28BBZ CAR T cells could lyse almost all the U251-EGFR cells. At an E:T ratio of 1:1, M27 CAR T cells had a lower lysis capacity than the 806 CAR T cells in EGFR-transfected U87MG and U251 cells (Fig. 3A and B). In addition, both CAR T cells could effectively induce lysis of endogenous EGFR-overexpressing cancer cell lines A431 and CAL27. At E:T ratio of 3:1, no significant difference on the lysis capacity was observed between these two CAR T cells (Fig. 3C). The M27-28BBZ CAR T cells had no measurable cytotoxic effect on all three primary keratinocytes; however, the 806-28BBZ CAR T cells showed significantly higher cytotoxic effect on all three keratinocyte lines at 3:1 E:T ratio (806-28BBZ vs. M27-28BBZ on keratinocyte-1: 31.9% ± 1.6% vs. 16.9% ± 0.8%, P < 0.001; 806-28BBZ vs. M27-28BBZ on keratinocyte-2: 38.8% ± 2.0% vs. 11.7% ± 1.3%, P < 0.001; 806-28BBZ vs. M27-28BBZ on keratinocyte-3: 54.1% ± 3.3% vs. 20.6% ± 1.1%, P < 0.001; Fig. 3D). Additionally, M27-28BBZ CAR T cells had no cytotoxic effect on other normal cells L-02 and RWPE-1 (Supplementary Fig. S6). This indicated that M27-28BBZ CAR T cells could selectively lyse EGFR- and/or EGFRvIII-overexpressing tumor cells but not EGFR-expressing normal cells.

Cytokine secretion by CAR T cells in response to a target antigen indicates activation and maintenance of an antigen-specific immune response. The secretion of TNFα, IL2, and IFNγ from CAR T cells was determined to evaluate the activation of CAR T cells by antigen-expressing tumor cells. Activated M27-28BBZ CAR T cells secreted significantly more TNFα, IL2, and IFNγ than did mock T cells after incubation with EGFR/EGFRvIII-overexpressing tumor cells (P < 0.05, Fig. 3E). In addition, M27-28BBZ CAR T cells produced greater concentrations of cytokines in the presence of EGFRvIII-overexpressing tumor cells than in the presence of EGFR-overexpressing tumor cells. Secretion of these cytokines from CAR T cells was negligible after incubation with untransfected U87MG and U251 cells (Fig. 3E). Cytokine secretion was not induced in M27-28BBZ CAR T cells cocultured with three primary keratinocytes, further supporting that M27-28BBZ CAR T cells would not cross react with keratinocytes.

To determine the therapeutic efficacy of genetically modified T cells, M27-28BBZ CAR T cells and mock T cells were used to treat mice bearing U251-EGFR or U251-EGFRvIII subcutaneous tumors. The results demonstrated that M27-28BBZ CAR T cells could significantly inhibit tumor growth in both xenografts. On day 42 following tumor cell inoculation, significant reduction of tumor volume (P < 0.001, Fig. 4A) and tumor weight (P < 0.0001, Fig. 4B) of the U251-EGFRvIII tumors was observed in the group treated with M27-28BBZ CAR T cells when compared with the group of mock T-cell treatment. On day 49 following tumor cell inoculation, the tumor volume and tumor weight of U251-EGFR xenografts in the M27-28BBZ CAR T-cell group was also significantly lower than those in the mock T-cell group (P < 0.05; Fig. 4C and D). These results indicated that M27-28BBZ CAR T cells could efficiently suppress the growth of both U251- EGFR and U251-EGFRvIII cells in vivo.

Given that the subcutaneous microenvironment might disturb the activity of CAR T cells, we evaluated the antitumor activities of M27-28BBZ CAR T cells in an intracranial glioblastoma xenograft that contained luciferase-transfected U251-EGFR and U251-EGFRvIII tumor cells. Previous studies have reported that the expression of EGFRvIII in some cells is frequently associated with amplified expression of EGFR and that the coexpression of both receptors within the tumor mass confers a worse prognosis in patients with glioblastoma (12, 30). To model heterogeneous expression of EGFR and EGFRvIII in glioblastoma, U251-luci-EGFR cells and U251-luci-EGFRvIII cells were mixed together at 3:1 ratio (EGFR-expressing cells grew much more slowly than EGFRvIII-expressing cells). The mixed cells were intracranially implanted into mice (day 0). After 6 days since glioblastoma xenografts were established, the engraftment of U251-EGFR and U251-EGFRvIII cells was confirmed by bioluminescence imaging. On day 7 following initial tumor engraftment, the mice were intravenously injected with one dose of 1 × 10⁷ M27-28BBZ CAR T cells or mock T cells. At day 28 after tumor inoculation (21 days after T-cell injection), the tumors in the U251-EGFR xenograft model showed a significant reduction in volume after treatment of M27-28BBZ CAR T cells compared with mock T cells (Fig. 5A and B). The survival of mice in M27-28BBZ CAR T cell–treatment group was significantly prolonged compared with that in the mock T cell–treatment group (P < 0.05, Fig. 5C). In the U251-EGFRvIII and U251-EGFR/EGFRvIII xenograft models, tumor growth in the M27-28BBZ treatment group was suppressed compared with that in the mock group at day 21 after tumor inoculation (14 days after T-cell injection; Fig. 5A and B). In both models, some mice treated with M27-28BBZ CAR T cells survived for greater than 80 days (U251-EGFR/EGFRvIII model: 2/6 mice, U251-EGFRvIII model: 1/6 mice). In M27-28BBZ CAR T cell–treatment groups, median survival time was increased from 26 to 60 days in the mice with U251-EGFRvIII xenografts and changed from 24 to 56 days in the mice with U251-EGFR/EGFRvIII xenografts compared with mice receiving mock T cells (P < 0.05; Fig. 5C).

To evaluate CAR T-cell persistence in vivo, tumor-bearing mice were euthanized at day 15 after intracranial tumor implantation (8 days after injection of CAR T cells). The persistence of CAR T cells in the murine peripheral blood and the infiltration of CAR T cells into brain tissues were determined in the orthotopic glioblastoma xenograft models. Flow-cytometric analysis showed a significant increase (P < 0.05) of both CD4⁺ and CD8⁺ T cells, with a predominance of CD8⁺ T cells, in the M27-28BBZ group compared with the mock group in all three models (Fig. 6A). More T cells in the EGFR/EGFRvIII heterogeneity model persisted than in EGFRvIII and EGFR groups. The infiltration of human T cells was determined in the tumor-bearing mouse brain tissues by immunostaining with antibody to human CD3 (Fig. 6B–D). Human CD3⁺ cells robustly infiltrated residual tumors after M27-28BBZ CAR T-cell infusion. In contrast, very few T cells could be detected in tumor lesions after mock T-cell treatment (Fig. 6B–D). More infiltrating T cells were found in EGFRvIII-overexpressing tumor lesions than in EGFR-overexpressing tumors (U251-EGFRvIII vs. U251-EGFR: 101.8 ± 19.9 vs. 62.1 ± 20.7 CD3⁺ cells/mm², P < 0.01; Fig. 6E). These data suggested that the injected M27-28BBZ CAR T cells infiltrated antigen-expressing brain tumors in vivo and were amplified and maintained.

Mounting evidence indicates that intratumoral heterogeneity may be a cause of drug resistance and tumor recurrence in cancer treatment (31). Glioblastoma is a highly heterogeneous tumor (32, 33) and one of the first cancers profiled in The Cancer Genome Atlas project (34). Genetic alterations including mutations, rearrangements, alternative splicing, and focal amplifications of EGFR have been found in 57% of glioblastomas (35). In addition, EGFR amplification and overexpression has been reported in approximately 50% to 60% of glioblastomas (36, 37). EGFR variants including EGFRvIII and de4 EGFR (38, 39) could increase proliferation and invasiveness of glioblastoma cells. One study demonstrated that EGFRvIII is invariably accompanied by EGFR amplification, although EGFR overexpression does not necessarily give rise to the EGFRvIII variant (23, 24). This partially explains why EGFRvIII-specific therapy was unable to completely eliminate glioblastoma. Another study showed that EGFRvIII expression was lost in 80% of relapsed tumors following vaccination with PEPvIII-KLH in Freund's complete adjuvant, suggesting that the EGFRvIII-negative cell population predominated at the time of tumor recurrence (40). Additionally, neither partial nor complete responses were observed in a total of 10 glioblastoma patients treated with EGFRvIII-specific CAR T cells, although EGFRvIII antigen completely disappeared in 2 patients after CAR T-cell infusion (18). This paradoxical clinical readout could be ascribed to heterogeneity of EGFRvIII expression. To overcome the heterogeneity of EGFRvIII, CAR-NK cells that could recognize both EGFR and EGFRvIII were developed (41, 42). However, the antibodies used in the CARs, such as C225, could bind to normal EGFR, which is expressed in normal brain tissues (39, 43). Thus, these CAR-NK cells may induce on-target, off-tumor toxicities in patients with glioblastoma even treated by locally administration.

Severe toxicities including death have been reported in CAR T-cell immunotherapy due to on-target, off-tumor toxicities (44, 45). Thus, using cancer-selective antibodies as the targeting moiety of the CARs may enhance the safety of this therapy. MAb 806, which binds to the EGFR^287–302 epitope, can bind to EGFR and to EGFRvIII overexpressed in cancer cells, but not to EGFR in normal cells. Unexpectedly, our study revealed that 806 scFv could bind to primary keratinocytes, although with a much lower binding capacity than cetuximab. At doses up to 24 mg/kg, ABT-806 (25) only induced mild dermatitis acneiform in 12% (6/49) of the patients and one grade 3 morbilliform rash. Even more promising is that ABT-414, at a dose of 1.25 mg/kg, showed no skin toxicity and induced an objective response rate of 6.8% in the 66 patients treated (26). In combination with radiation and temozolomide, ABT-414 was well tolerated and showed promising clinical benefit in patients with newly diagnosed glioblastoma (46). These studies support using antibodies targeting the EGFR^287-302 epitope as an immunotherapeutic strategy for patients with glioblastoma. However, the study of ABT-414 had ocular grade 3/4 adverse events in 33% of patients overall, which may be ascribed to the accumulation of the toxic MMAF. Another consideration with antibody–drug conjugates is that they need internalization through receptor-mediated endocytosis. Its antitumor activities depend on the binding avidity of the antibody and the endocytosis capacity of the receptor. Unlike antibody–drug conjugates, CAR T cells do not possess the same compound-associated toxicities and are capable of killing cancer cells independent of receptor-mediated endocytosis. A single CAR T-cell may express thousands of CAR copies, which may increase its binding avidity on target cells and maybe the reason why CAR T cells comprising 806 scFv, a low-affinity EGFR binder, are capable of lysing normal keratinocytes in vitro. In this study, we identified a humanized scFv M27, which did not bind to normal cells, but could bind to both EGFR and EGFRvIII in tumor cells at a relatively lower affinity than 806. Like previous reports that EGFR antibodies could be tuned to low affinity to reduce CAR T cell–associated toxicity against normal cells (47, 48), M27 CAR T cells showed no toxicity against normal cells while retaining their tumor lysis capabilities. Additionally, M27-28BBZ CAR T cells have relatively higher cytotoxicity against cancer cells with EGFR- or EGFRvIII-overexpression, compared with that of M27-28Z CAR T cells. These data indicated that M27-28BBZ CAR T cells demonstrated effective antitumor activity with minimal clinically relevant on-target, off-tumor toxicity.

In intracranial tumor xenograft models, M27-28BBZ CAR T cells efficiently suppressed the growth of EGFR- and EGFRvIII-overexpressing tumors and increased survival of the mice. Its antitumor activity against EGFRvIII tumors appeared more potent than those against EGFR tumors. It might be explained by higher binding affinity of M27 scFv to EGFRvIII than to EGFR. The xenografts in the EGFR/EGFRvIII mixture group were eradicated by M27-28BBZ CAR T cells, and this may be ascribed to the enhanced amplification or persistence of M27-28BBZ CAR T cells in the presence of EGFRvIII-positive tumor cells. T-cell numbers in both blood and tumor tissues in the EGFR/EGFRvIII group were higher than those in the EGFR group, and increased T-cell infiltration and persistence might facilitate the elimination of tumor cells with EGFR overexpression.

Given their distinct tumor specificity and antitumor activities, M27-28BBZ CAR T cells represent a new bitargeted antitumor agent for the treatment of patients with glioblastoma. Additionally, as many other tumor types, including non–small cell lung cancer, breast cancer, and head and neck squamous carcinoma, are characterized by EGFR amplification and/or EGFRvIII mutations (23, 49, 50), M27-28BBZ CAR T cells also have the potential for the treatment of these cancer types.

No potential conflicts of interest were disclosed.

Conception and design: Z. Li

Development of methodology: H. Jiang, H. Gao, J. Kong, B. Song

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): P. Wang, B. Shi, H. Wang

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): H. Jiang, H. Gao

Writing, review, and/or revision of the manuscript: H. Jiang, Z. Li

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): H. Wang

Study supervision: Z. Li

This work was supported by funding from the Supporting Programs of Shanghai Science and Technology Innovation Action Plan (No. 18431902900), Shanghai Municipal Commission of Health and Family Planning (No. 201740124), the National Natural Science Foundation of China (No. 81672724 and No. 81472573), a grant from the State Key Laboratory of Oncogenes and Related Genes (No. 91-17-17), and the seed fund of Renji Hospital (No. RJZZ17-021).

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.