Research ArticleCancer

Chlorotoxin-directed CAR T cells for specific and effective targeting of glioblastoma

See allHide authors and affiliations

Science Translational Medicine  04 Mar 2020:
Vol. 12, Issue 533, eaaw2672
DOI: 10.1126/scitranslmed.aaw2672

Stinging glioblastoma

Chlorotoxin derived from scorpion venom has previously been shown to bind glioblastoma cells. Wang et al. designed a chimeric antigen receptor (CAR) based on chlorotoxin to surmount limitations of other glioblastoma-targeted CARs that have not been able to overcome tumor heterogeneity and antigen escape. They demonstrated that chlorotoxin binding captures a broader array of primary tumors than staining for previously identified antigenic targets. Chlorotoxin-directed CAR T cells were safe in mice and induced regression of orthotopic glioblastoma xenografts with no evidence of antigen escape. These toxin-based CAR T cells are distinct from conventional CAR design and could one day be used to deliver a poisonous blow to glioblastoma.


Although chimeric antigen receptor (CAR) T cells have demonstrated signs of antitumor activity against glioblastoma (GBM), tumor heterogeneity remains a critical challenge. To achieve broader and more effective GBM targeting, we developed a peptide-bearing CAR exploiting the GBM-binding potential of chlorotoxin (CLTX). We find that CLTX peptide binds a great proportion of tumors and constituent tumor cells. CAR T cells using CLTX as the targeting domain (CLTX-CAR T cells) mediate potent anti-GBM activity and efficiently target tumors lacking expression of other GBM-associated antigens. Treatment with CLTX-CAR T cells resulted in tumor regression in orthotopic xenograft GBM tumor models. CLTX-CAR T cells do not exhibit observable off-target effector activity against normal cells or after adoptive transfer into mice. Effective targeting by CLTX-CAR T cells requires cell surface expression of matrix metalloproteinase–2. Our results pioneer a peptide toxin in CAR design, expanding the repertoire of tumor-selective CAR T cells with the potential to reduce antigen escape.


Glioblastoma (GBM) is the most common type of primary brain tumor (PBT). Despite increasingly aggressive treatments incorporating surgery, chemotherapy, and radiotherapy, survival of patients with GBM has only modestly improved over the past several decades (1). Such poor prognosis has prompted the development of advanced therapies, among which is immunotherapy using T cells engineered to express chimeric antigen receptors (CARs) (2, 3). CAR T cell therapy redirects the cytotoxic activity of T lymphocytes independent of major histocompatibility complex restriction and without need for antigen priming. This cellular therapy, therefore, provides a strategy to generate de novo antitumor immunity, which may help overcome the challenges of highly heterogeneous expression of targetable tumor antigens, as well as the lack of intrinsic immunogenicity for tumors such as GBMs with low mutational burdens (4, 5). We and others have demonstrated that CAR T cell therapy can be successfully translated for the treatment of GBM (69), demonstrating safety; evidence for antitumor activity; and, in one case, the potential for mediating complete tumor remission (7).

Despite encouraging evidence of clinical safety and bioactivity for GBM-targeted CAR T cells, the overall response rates have been unsatisfyingly low, especially as compared to the remarkable clinical responses achieved against B cell malignancies (10, 11). One of the major obstacles limiting CAR T cell therapeutic efficacy has been tumor heterogeneity, which is particularly substantial in GBMs. The classification of GBM subtypes has illustrated the heterogeneity across patients, and more recent studies using single-cell sequencing also revealed considerable genetic variations among intratumoral subpopulations, as well as plasticity between different cellular states (12, 13). Efforts to develop CAR T cell immunotherapy must contend with this high diversity of potential target antigen expression. For example, CAR T cells targeting interleukin 13 (IL13) receptor α2 (IL13Rα2) are under active clinical development (7, 14) because we and others have reported that expression of IL13Rα2 is frequently found on GBM tumors and on a high proportion of cells within these tumors (15). However, after treating patients with IL13Rα2-targeted CAR T cells, instances of tumor recurrence with loss and/or reduced expression of IL13Rα2 have been observed (7, 14). Similar results have been reported after epidermal growth factor receptor (EGFR) variant III (EGFRvIII)–targeted immunotherapies, with lower EGFRvIII expressions in recurrent tumors after therapy (9, 16). In general, tumors are able to rapidly adapt to the selection pressures imposed by immunotherapies, resulting in relapsed tumors with distinct intratumoral cellular profiles (17), so-called antigen escape. The clinical performance of CAR T cell therapy against B cell malignancies is greatly aided by the homogeneous expression of CD19 as a target antigen on all B cell lineages and malignancies (18). Therapeutic outcomes for GBM-targeting CAR T cell designs would thus be expected to benefit from immunotherapies with broader tumor recognition, but progress has been limited by the scarcity of brain tumor antigen candidates that are both widely expressed and highly specific.

An opportunity to extend the repertoire of target antigens amenable to CAR T cell therapy is presented by the tumor-binding potential of some naturally derived molecules (19). One example is chlorotoxin (CLTX), a 36–amino acid peptide isolated from the venom of the death stalker scorpion Leiurus quinquestriatus (20). The GBM-binding potential of CLTX was first identified through conjugation with the radioisotope 131I (21). Subsequently, CLTX has been established to bind broadly and specifically to GBM and other neuroectodermal tumors while showing minimal cross-reactivity with nonmalignant cells in brain and elsewhere (22). Although the precise cell surface receptor for CLTX on GBM cells remains unclear, CLTX binding impairs GBM cell migration and invasiveness (23, 24). CLTX itself is noncytotoxic to tumor and normal tissues (25, 26). Consequently, efforts have focused on using CLTX for tumor-specific delivery of cytotoxic agents (27). Preclinically, CLTX has been used to coat a variety of vehicles for delivery of chemotherapeutics and small interfering RNAs (28, 29). Furthermore, an early clinical study in patients with GBM demonstrated that 131I-conjugated CLTX radiotherapy delivered into the postresection cavity containing residual tumor was well tolerated and without major neurotoxicity (25). More recently, the utility of fluorescently labeled CLTX to specifically mark tumor cells during surgery has been demonstrated in preclinical models and confirmed in a clinical study of glial tumors showing safety of intravenously delivered CLTX and tumor-specific uptake (26, 3032). These studies indicate that CLTX can be safely used for tumor-specific targeting.

Here, we sought to develop a CAR T cell using CLTX peptide as the tumor-targeting domain. We optimized the design of CLTX-CAR T cells and tested their antitumor activity against a panel of patient-derived GBM lines and in orthotopic mouse xenograft models. We also evaluated potential off-tumor targeting and toxicity against normal human cells and tissues and in CLTX cross-reactive mouse models. These studies aim to expand the repertoire of GBM-targeting CAR T cells and address the challenge of tumor heterogeneity by achieving broad tumor cell coverage.


CLTX binds to a broad spectrum of GBM cells

Previous studies have documented the GBM-selective binding properties of CLTX (21, 22). However, CLTX binding, in relation to other tumor-associated antigens or to GBM subpopulations, has not been specifically examined. Therefore, we first evaluated CLTX binding to freshly dissociated tumor cells from surgical resection specimens. These PBT cells were examined by flow cytometry for binding of Cy5.5-conjugated CLTX peptide (CLTX-Cy5.5) and compared with expression of IL13Rα2, human epidermal growth factor receptor 2 (HER2), and EGFR, three receptors being clinically evaluated as CAR T cell immunotherapy targets for GBM (9, 14, 33). Strong CLTX-Cy5.5 binding was observed for almost all patient tumors, with greater than 80% of cells binding CLTX (Fig. 1, A and B, left). Across 23 tumor samples from 15 different patients, only samples from 2 patients with GBM (PBT114 and PBT131) showed CLTX-Cy5.5 binding in less than 40% of total cells. At the same time, expression of immunotherapy targets IL13Rα2, HER2, and EGFR varied widely between patient tumors. CLTX-Cy5.5 binding appeared to be independent of other antigens and was observed on tumors with both high and low expression of IL13Rα2, HER2, and EGFR (see representative flow cytometry histograms in Fig. 1A).

Fig. 1 CLTX binds broadly to freshly dispersed primary GBM cells and to cultured tumor lines.

(A) Freshly dispersed viable [4′,6-diamidino-2-phenylindole–negative (DAPI)] patient brain tumor (PBT) GBM cells (CD45CD31) were immunostained for expression of IL13Rα2, HER2, and EGFR or binding by CLTX-Cy5.5. Percentages of stained cells (blue) above isotype control (gray) are indicated in each histogram. (B) Summary of results [% positive as in (A)] for 23 freshly dispersed (primary) PBTs (left) and 22 cultured PBT–tumor spheres (TS) (right). nd, not done. (C) Phenotype of a GBM xenograft established by stereotactic injection and engraftment of 1 × 105 PBT106-TS cells in the right forebrain of an NSG mouse. Tumor-bearing mouse brain was harvested 97 days after cell injection, and paraffin sections were stained with antibodies against IL13Rα2 and EGFR and with biotin-conjugated CLTX. Staining was visualized by fluorochrome-conjugated secondary antibodies or streptavidin and DAPI to identify nuclei. Scale bars, 20 μm. Images are representative of four stained slides from the GBM xenograft. (D) Freshly dispersed GBM samples separated into CD44+ and CD44 (left; n = 11) or CD133+ and CD133 (right; n = 15) subpopulations and are examined for differences in CLTX-Cy5.5 staining [measured as mean fluorescence intensity (MFI)]. Only GBM samples with >20% CD44+ or CD133+ fractions were analyzed. (E) PBT-TS lines (n = 7), maintained in neural stem cell medium (TS) or differentiation medium (Dif) for 14 days, were evaluated for CLTX-Cy5.5 staining and CD133 expression (MFI). (D and E) P values were evaluated by paired two-way Student’s t test.

We also examined CLTX-Cy5.5 binding to low-passage patient-derived GBM tumor sphere (PBT-TS) lines, expanded under conditions favoring cancer stem cell–like phenotypes (34). Similar to dissociated patient tumor samples, 21 of 22 PBT-TS lines (95.5%) showed greater than 70% CLTX-Cy5.5 binding (Fig. 1B, right, and fig. S1A), displaying a wider coverage compared to IL13Rα2, HER2, and EGFR, as well as the relatively widely expressed GBM-associated antigen EphA2 (35, 36). To evaluate CLTX binding in engrafted tumors, GBM orthotopic xenografts were examined by fluorescence microscopy using biotin-conjugated CLTX peptide. Consistent with the analyses of freshly resected patient tumors, CLTX displayed consistent binding to all engrafted GBM xenograft tumors established using five different patient-derived PBT-TS lines and marked a greater proportion of tumor cells as compared to the expression of IL13Rα2 and EGFR (Fig. 1C and fig. S1B). Together, these studies confirmed the capacity of CLTX to bind to a high percentage of patient GBM tumors, as well as to most of GBM cells within each tumor.

GBM tumors are highly heterogeneous, composed of phenotypically distinct cellular subpopulations. Within GBM tumors, stem-like cells [glioma stem cells (GSCs)] display self-renewal and tumor-initiation capacity (37), identifying this population as a therapeutically important component of a GBM-targeted CAR T cell therapy (33, 38, 39). Hence, we examined CLTX-Cy5.5 binding with respect to this stem cell–like population. In one approach, using freshly dissociated primary PBTs, we distinguished two (often nonoverlapping) GSC populations by expression of surface markers CD133 and CD44 (Fig. 1B, left) (40, 41). We observed CLTX-Cy5.5 binding to both GSC populations and non-GSCs, with slightly higher fluorescence intensity in stem cell–like (CD44+ or CD133+) versus differentiated (CD44 or CD133) cells (Fig. 1D). In another approach, we used PBT-TS lines and varied culture conditions to favor growth of GSCs or, alternatively, to promote differentiation (38). Although in vitro differentiation led to reduced expression of the GSC marker CD133, it did not reduce CLTX-Cy5.5 binding (Fig. 1E). These studies demonstrated that CLTX binding, although showing some preference for CD133+ GSCs in freshly dispersed tumor samples, remains robust on both stem-like and more differentiated GBM cells. Together, these studies confirm the broad GBM-binding potential of CLTX, providing rationale for investigating its use for CAR T cell immunotherapy.

CLTX-CAR T cell design can be optimized through modifying nontargeting domains

We next sought to design an optimized CAR incorporating the CLTX peptide as the tumor-targeting domain. The initial CLTX-CAR was generated using the backbone of our CD19-targeted CAR that has shown safety and clinical activity against B cell malignancies (42). This CLTX-CAR construct is composed of the CLTX peptide, an IgG4Fc(EQ) spacer, and a CD28 costimulatory domain (Fig. 2A) and is referred to as CLTX-EQ-28ζ. Cytotoxicity of CLTX-EQ-28ζ CAR T cells was evaluated against a panel of GBM PBT-TS lines. CLTX-EQ-28ζ CAR T cells formed immunological synapse-like structures with tumor cells within 2 hours (Fig. 2B). The number of CLTX-CAR T cell–tumor cell conjugates was equivalent to IL13Rα2-targeted CAR T cells between 2.5 and 5 hours, although the formation rate appeared to be slower with CLTX-CAR T cells (0.5 to 1.5 hours) (fig. S2A). Coculture with GBM cells stimulated CLTX-EQ-28ζ CAR T cells to up-regulate T cell activation markers CD69 and 4-1BB (CD137) and to degranulate as measured by cell surface CD107a expression (Fig. 2C and fig. S2B). Furthermore, CLTX-EQ-28ζ CAR T cells efficiently killed GBM cells in 48-hour coculture assays at an effector-to-target (E:T) ratio of 1:4 (Fig. 2D). CLTX-EQ-28ζ CAR T cells were effective against PBT-TS lines from four different patients who displayed distinct IL13Rα2, HER2, and EGFR expression profiles. In particular, although IL13Rα2-targeted CAR T cells failed to respond to the PBT-TS lines with low-to-negative IL13Rα2 expression (PBT003-4-TS and PBT138-TS), CLTX-EQ-28ζ CAR T cells recognized and responded to all four TS lines (Fig. 2, B to D). Furthermore, using PBT003-4-TS cells, which show negligible expression of IL13Rα2, HER2, or EGFR, we demonstrated that coculture with CLTX-EQ-28ζ, but not the combination of CAR T cells targeting all three other antigens, mediated tumor cell elimination (movies S1 and S2). Lentiviral transduction of PBT003-4-TS to overexpress IL13Rα2, HER2, or EGFR did not reduce CLTX-EQ-28ζ cytotoxic activity (fig. S2, C and D). We thus conclude that CLTX-EQ-28ζ can recognize and target GBM cells and its cytotoxicity is consistent with the broad GBM-binding potential of the CLTX peptide while independent of other GBM-associated antigens.

Fig. 2 Effector activity of CLTX-CAR T cells.

(A) Diagram of a CAR incorporating a CLTX tumor-targeting domain, an IgG4-Fc spacer domain with EQ mutations, a CD28 transmembrane domain, and intracellular costimulatory and signaling domains (CD28 and CD3ζ). (B) Immunological synapse formation at 2 hours after adding CLTX-CAR T cells to dissociated PBT003-4-TS or PBT106-TS GBM cells (E:T ratio, 1:1). (Top) A representative image of an immunological synapse indicated by colocalization of phosphorylated CD3ζ (p-CD3ζ) and polarized F-actin accumulation at the interface between the CLTX-EQ-28ζ CAR T cell (arrowhead, CAR T cell) and PBT003-4 tumor cell. Scale bars, 5 μm. (Bottom) Number of immunological synapses per 5000 tumor nuclei—for mock-transduced (mock), CLTX-CAR, or IL13Rα2-CAR T cells—cocultured with dissociated PBT003-4-TS (IL13Rα2) or PBT106-TS (IL13Rα2+) GBM cells. Five fields were counted for each group. Shown is mean ± SEM; ***P < 0.001. (C) Degranulation of mock, CLTX-CAR, or IL13Rα2-CAR T cells (% CD107a) after 5-hour coculture with either IL13Rα2 (PBT003-4-TS and PBT138-TS) or IL13Rα2+ (PBT030-2-TS and PBT106-TS) GBM cells (E:T ratio, 1:1). CAR T cells (gated as CD3+ CD19t+) were analyzed by flow cytometry for surface CD107a expression as a marker of degranulation. Shown are means ± SEM of % CD107a+ cells in duplicate wells. (D) Percent tumor cell killing of different PBT-TS cells cocultured with mock, CLTX-CAR, or IL13Rα2-CAR T cells (E:T ratio, 1:4; 48 hours), calculated against the numbers of viable tumor cells when cultured in the absence of CLTX-CAR T cells. Shown are means ± SEM of % cell killing in duplicate wells. (C and D) ns, not significant (P > 0.05); *P < 0.05; **P < 0.01; and ***P < 0.001 comparing CAR versus mock, by one-way ANOVA with Bonferroni’s multiple comparison correction.

The cytolytic activity of CAR T cells is greatly influenced by regions outside of the antigen-targeting domain, including the spacer (43) and the costimulatory domains (44). Using the CLTX-EQ-28ζ CAR as a reference, we first addressed the impact of spacer length by generating CAR constructs to compare IgG4Fc(EQ) (239 amino acids) with three shorter spacers: immunoglobulin G4 (IgG4)–Fc with the CH2 domain deleted (∆CH2) (129 amino acids), CD8 hinge (CD8h) (44 amino acids), and a short synthetic linker (L) (10 amino acids) (Fig. 3A and data file S2). We observed that CAR T cell–mediated tumor killing was greatly reduced with ∆CH2 or L spacers, whereas the CD8h spacer retained CAR function similar to that of IgG4Fc(EQ) (Fig. 3B). We next evaluated contributions of costimulatory signals by generating CARs bearing CD28 or 4-1BB costimulatory domains in the context of the more efficacious spacers IgG4Fc(EQ) and CD8h (Fig. 3A). CAR T cells incorporating a CD28 costimulation domain consistently showed higher killing potency as compared with CARs incorporating a 4-1BB costimulation domain (Fig. 3C).

Fig. 3 Optimization of CLTX-CAR design for effector potency.

(A) Schemas of CLTX-CARs incorporating different spacers [IgG4Fc(EQ), IgG4(ΔCH2), L, and CD8h] and costimulatory domains (CD28 and 4-1BB). (B) Evaluation of alternative spacers for CLTX-CAR T cells with shared CD28 costimulatory domain, tested against PBT003-4-TS or U251T GBM cells. Indices of efficacy were degranulation (% CD107a+; top) and cytotoxicity (% target cell killing; bottom). (C) Evaluation of alternative costimulatory domains for CLTX-CAR T cells with IgG4Fc(EQ) or CD8h spacers, tested against PBT003-4-TS or U251T GBM cells. Measured were degranulation (top) and cytotoxicity (bottom). (B and C) Shown are means ± SEM of duplicate wells; *P < 0.05, **P < 0.01, and ***P < 0.001 by one-way ANOVA with Bonferroni’s multiple comparison correction. (D) Evaluation in a repetitive rechallenge assay of CLTX-CAR T cells with alternative spacers [IgG4Fc(EQ) or CD8h] and shared CD28 costimulatory domain. CAR T cells were cocultured with PBT003-4-TS cells (4000 CAR T cells and 16,000 tumor cells), repetitively challenged with 32,000 tumor cells every 48 hours [arrows at D2 (day 2), D4, and D6], and numbers of remaining viable tumor cells were quantified at the indicated time points (D1, D3, D5, and D7). *P = 0.017, **P = 0.002, and ***P < 0.001. (E) Evaluation of alternative spacer domains by comparison of coexpression of inhibitory receptors PD-1, LAG-3, and TIM-3 by CLTX-EQ-28ζ and CLTX-CD8h-28ζ CAR T cells at day 4 of rechallenge with PBT003-4-TS tumor cells. (D and E) Shown are means ± SEM of triplicate wells; **P < 0.01 by unpaired Student’s t test.

We then considered possible mechanisms underlying functional differences across CLTX-CAR constructs. For the six CLTX-CAR constructs evaluated, measures of tumor-dependent activation for each construct were consistent across multiple assays: killing potency (% killing), extent of degranulation (% CD107a+), and secretion of cytokine interferon γ (IFNγ) (Fig. 3, B and C, and fig. S3A). Moreover, consistent with the stronger cytotoxic effector functions of CLTX-EQ-28ζ and CLTX-CD8h-28ζ CARs, both also showed more frequent expression of T cell activation markers 4-1BB and CD69 (fig. S3B), and inhibitory molecule programmed cell death–1 (PD-1) (which is linked to T cell activation) (fig. S3C) after tumor cell exposure. These results therefore suggest that differences in initial activation were the major contributor to variations in effector function between various CLTX-CAR designs.

We next returned to the comparison of effector potency between CLTX-EQ-28ζ and CLTX-CD8h-28ζ CAR T cells. First, screening degranulation and cytokine production against a panel of 10 PBT-TS lines, we observed that CLTX-EQ-28ζ CAR T cells displayed higher degranulation but lower IFNγ production as compared to CLTX-CD8h-28ζ CAR T cells (fig. S3, D and E). Next, the capacities of CLTX-EQ-28ζ and CLTX-CD8h-28ζ CAR T cells to maintain long-term antitumor activity were evaluated in a repetitive tumor challenge assay in vitro (45, 46). We observed that CLTX-EQ-28ζ, but not CLTX-CD8h-28ζ, CAR T cells retained activity through multiple rounds of tumor challenge (Fig. 3D and fig. S3F). Because both CLTX-EQ-28ζ and CLTX-CD8h-28ζ CAR T cells up-regulated PD-1 expression upon tumor stimulation, we then assessed coexpression of three T cell inhibitory receptors (PD-1, LAG-3, and TIM-3), which together mark exhausted T cells (45, 47), and observed that the absence of durable effector function in CLTX-CD8h-28ζ CAR T cells was associated with an exhausted phenotype featuring coexpression of all three inhibitory receptors (P = 0.009) (Fig. 3E). Considering together these observations about effector potency, we adopted the CLTX-EQ-28ζ CAR as the optimal design for subsequent evaluation.

CLTX-CAR T cells mediate antitumor activity against established GBM xenografts

To test the in vivo antitumor activity of CLTX-EQ-28ζ CAR T cells, xenograft tumors were established using two patient-derived PBT-TS lines with distinct antigen expression patterns: PBT003-4-TS (<15% expression of IL13Rα2, HER2, or EGFR) and PBT106-TS (>65% expression of IL13Rα2, HER2, and EGFR). Against subcutaneously engrafted GBMs, intratumoral administration of CLTX-EQ-28ζ CAR T cells resulted in tumor regression, whereas tumors injected with mock-transduced T cells displayed growth kinetics similar to tumor-only controls (fig. S4A). Next, the antitumor function of CLTX-EQ-28ζ CAR T cells was evaluated in orthotopic GBM models (48). Consistent with the subcutaneous tumor model, intracranial tumor (ICT) administered CLTX-EQ-28ζ CAR T cells, but not mock T cells; controlled tumor growth; and prolonged the survival of mice bearing PBT003-4-TS or PBT106-TS tumors (Fig. 4, A to D). We then examined the contribution of delivery routes to CLTX-CAR T cell therapeutic efficacy (fig. S4B) and observed that the intracerebroventricular (ICV) route also mediated potent antitumor activity but required slightly higher CAR T cell doses to obtain similar efficacy compared with ICT delivery (fig. S4, B to D). By contrast, intravenous delivery yielded minor therapeutic benefit (fig. S4, B to D), consistent with our previous results observed with other CAR T cells (49, 50). The in vivo antitumor potency of CLTX-CAR T cells was similar compared with IL13Rα2-CAR T cells (fig. S4E).

Fig. 4 In vivo antitumor activity of CLTX-CAR T cells.

ffLuc+ PBT106-TS (A and B) or ffLuc+ PBT003-4-TS (C and D) GBM cells were stereotactically implanted into the right forebrain of NSG mice (1 × 105 cells per mouse). On day 8 (PBT106-TS) or 12 (PBT003-4-TS) after tumor implantation, mice received either no treatment (Tumor only; n = 5 to 6), intracranial tumor (ICT) treatment with 1 × 106 mock-transduced T cells (Mock; n = 6), or CLTX-EQ-28ζ CAR T cells (CLTX-CAR; n = 6 to 8). (A and C) Kaplan-Meier survival analysis with log-rank (Mantel-Cox) test comparing CLTX-EQ-28ζ CAR T cell– and mock T cell–treated groups. (B and D) Tumor volumes over time monitored using bioluminescent imaging. (E) Tumors harvested from NSG mice bearing PBT003-4-TS GBM xenografts that were either untreated (n = 5) or relapsed from those treated with CLTX-CAR T cells (n = 3) were dissociated into single-cell suspensions and stained with CLTX-Cy5.5. Lines indicate mean MFI ± SEM. (F) Immunochemical staining for CD3 and granzyme B on mouse brain sections at day 14 after T cell injection and 7 days after tumor clearance (top) or on the relapsed tumor (bottom). (G) PD-L1 staining on untreated (top) or relapsed (bottom) PBT003-4-TS tumors. (H) Expression of IFNγ receptor A (IFNγRA) on GBM cells dissociated from PBT003-4-TS and PBT106-TS. Percentages of immunoreactive cells (blue) above that of isotype control staining (gray) are indicated in each histogram. (I) PBT003-4-TS and PBT106-TS GBM cells were cocultured with mock or CLTX-CAR T cells for 24 hours, and condition media (CM) were transferred to fresh GBM cells and cultured for 24 hours. Numbers indicate PD-L1 MFI above isotype control. Shown are means ± SEM of triplicate wells; **P < 0.01 by unpaired Student’s t test.

Although CLTX-CAR T cells had potently targeted both PBT003-4-TS and PBT106-TS in vitro, the responses of mice with engrafted tumors were not equivalent. After ICT CLTX-CAR T cell administration, all mice previously bearing PBT106-TS tumors remained tumor free for over 170 days, whereas only a subset of PBT003-4-TS tumor–bearing mice achieved similar long-term tumor eradication (Fig. 4, A to D, and fig. S4F). To investigate mechanisms underlying this variation in antitumor response, we examined recurrent PBT003-4-TS tumors after CLTX-EQ-28ζ CAR T cell therapy. We first observed that CLTX-Cy5.5 binding to recurrent tumors was similar to untreated tumors (Fig. 4E), suggesting that antigen escape did not account for tumor relapse. Examining the CAR T cells within these orthotopic tumors, we found that during the primary antitumor response, granzyme B–expressing CAR T cells were detected 14 days after adoptive transfer (Fig. 4F, top row). By comparison, the persisting CLTX-EQ-28ζ CAR T cells in the recurrent tumors only weakly stained for granzyme B (Fig. 4F, bottom row). At the same time, these relapsed PBT003-4-TS tumors displayed increased expression of programmed cell death–ligand 1 (PD-L1) as compared to untreated tumors (Fig. 4G). It is well established that PD-L1 expression can be induced by IFNγ signaling leading to tumor adaptive resistance to immunotherapy (51). Consistent with this, PD-L1 induction on PBT003-4-TS tumors after CAR treatment was associated with higher IFNγ receptor A (IFNγRA) expression before implantation (Fig. 4H). Furthermore, we observed greater induction of PD-L1 on PBT003-4-TS cells, as compared to PBT106-TS, when incubated with conditioned media from CLTX-CAR T cells cocultured with each tumor line (Fig. 4I). This difference was also evident in response to recombinant IFNγ treatment (fig. S4G). Although other immunosuppressive mechanisms may be involved, these observations suggest that immunostimulatory cytokines produced by activated CAR T cells can induce GBM adaptive resistance pathways, such as PD-L1 induction, and hinder tumor eradication by CAR T cells.

CLTX-CAR T cells exhibit minimal off-target effects

Prior clinical studies using CLTX to deliver radiation therapy and imaging reagents to tumor sites for patients with GBM have not reported notable adverse events (25, 32). Consistent with these reports, we observed limited to undetectable CLTX-Cy5.5 binding to a panel of human nontumor cells, including peripheral blood mononuclear cells (PBMCs); human embryonic kidney 293T cells; and induced pluripotent stem cell (iPSC)–differentiated astrocytes (iPSC-Ast), neural progenitor cells (iPSC-NPCs), and immortalized fetal brain–derived neural stem cell (FB-NSC) line LM-NSC008 (Fig. 5A). Furthermore, these normal cells were insufficient to trigger activation of CLTX-EQ-28ζ CAR T cells during coculture. As indicated by degranulation and in vitro cytotoxicity assays (Fig. 5, B and C), CLTX-EQ-28ζ CAR T cells did not target PBMC or 293T cells, or, in particular, neural lineage cells iPSC-Ast, iPSC-NPC and FB-NSC, exhibiting in all cases effector activity comparable to the mock T cell–negative controls.

Fig. 5 Off-target evaluation of CLTX-CAR T cells.

(A) CLTX-Cy5.5 staining on viable peripheral blood mononuclear cells (PBMCs), human embryonic kidney 293T cells, iPSC-derived astrocytes (iPSC-Ast), and neural progenitor cells (iPSC-NPC) or human fetal brain–neural stem cells (FB-NSC line LM-NSC008) evaluated by flow cytometry. Percentages of stained cells (blue) above control (gray) are indicated in each histogram. (B and C) Degranulation (B) and cytotoxicity (C) of mock-transduced (Mock) or CLTX-EQ-28ζ CAR+ (CLTX-CAR) T cells against nonmalignant cell lines, with PBT003-4-TS GBM cells as a positive control. Data are means ± SEM of duplicate wells; **P < 0.01 when compared with mock-transduced T cells using unpaired Student’s t test. (D) CLTX-Cy5.5 staining on mouse GBM cell lines. Percentages of positively stained cells (blue) above that of isotype control (gray) are indicated in each histogram. (E) Immunofluorescence phenotype of a GBM xenograft established by stereotactic injection of 1 × 105 PBT106-TS cells into the right forebrain of an NSG mouse. The tumor-bearing mouse brain was harvested 93 days after cell injection. Paraffin sections were stained with DAPI to identify nuclei and CLTX-Cy5.5 to depict the border between xenograft and normal mouse brain. Images are representative of four stained slides from the GBM xenograft. Scale bars, 20 μm. (F) PBT106-TS GBM cells were stereotactically implanted into the right forebrain of NSG mice (2 × 105 cells per mouse), and after 8 days, they received ICT administration of 1 × 106 CLTX-CAR T cells, intracerebroventricular (ICV) administration of 2 × 106 CLTX-CAR T cells, or intravenous (IV) administration of 1 × 107 CLTX-CAR T cells (n = 4, 7, 7, and 4, respectively). Body weights were measured before tumor injection, before CAR T cell injection, and 7 and 28 days after CAR T cell injection and compared with untreated NSG mice. Lines indicate means ± SEM.

The potential for toxicity was further assessed in mouse models, an approach justified by the observation that CLTX-Cy5.5 also binds to mouse GBM cells, as shown for the mouse KR158 and GL261 GBM lines (Fig. 5D). First, in mouse brains bearing GBM xenograft tumors, CLTX-Cy5.5 bound only to tumor cells but not surrounding normal brain tissue, thereby delineating the xenograft-tumor border (Fig. 5E). In addition, ICT and ICV regional administration of CLTX-EQ-28ζ CAR T cells in GBM-bearing mice in the experiments described above showed no evidence of adverse reaction, such as neurological symptoms and/or loss of body weight (Fig. 5F). After ICT administration of CLTX-EQ-28ζ CAR T cells to orthotopic GBM-bearing mice, no morphological alterations were evident in close examinations of multiple organs including brain, kidney, liver, spleen, intestine, colon, lung, and bladder (fig. S5A). The potential for systemic toxicity was first assessed by intravenous injection of 50 × 106 CLTX-EQ-28ζ CAR T cells into unmanipulated NSG (NOD/SCID/IL2R−/−) mice. Even at this high dose, CLTX-CAR T cells were well tolerated; all animals remained alert and active and did not show changes in body weight (fig. S5B). Furthermore, although both mock and CLTX-CAR T cells were detected in lung 7 days after systemic intravenous injection, neither expressed granzyme B (fig. S5, C and D). In addition, no T cells were detected at day 14 in the lung or at days 7 and 14 in intestine or bladder (fig. S5, C and D). Together, these results indicate the absence of CAR T cell activation or persistence in these organs. These observations are consistent with other preclinical studies of CLTX-redirected agents (26, 31) and strongly support our hypothesis that CLTX-EQ-28ζ CAR T cells specifically target GBM cells without causing observable toxicities.

Matrix metalloproteinase 2 expression is required for CLTX-CAR targeting

CLTX binding has been reported to associate with multiple membrane proteins, including membrane-associated matrix metalloproteinase 2 (MMP2), chloride channel CLCN3, and phospholipid protein annexin A2 (ANXA2) (23, 24, 52). To better understand target recognition, we used 15 different PBT-TS lines and iPSC-NPCs (negative control) as CLTX-EQ-28ζ CAR T cell targets and assessed the correlations of MMP2, CLCN3, and ANXA2 expression on target cells with CLTX-CAR T cell activation (CD107a degranulation). We found that the strength of activation on CLTX-EQ-28ζ CAR T cells varied between these different GBM target cells (Fig. 6A) and observed a strong correlation between CLTX-CAR T cell degranulation and target cell MMP2 expression (Fig. 6B) but no correlation with expression of CLCN3 or ANXA2 (fig. S6A). Consistent with this pattern, CLTX-Cy5.5 binding correlated with MMP2 expression on target cells (Fig. 6C). Differences in CLTX-CAR cytotoxic potential against different GBM cells were most evident at a low E:T ratio of 1:8 (fig. S6B) and correlated well with degranulation (fig. S6C). The extent of degranulation did not correlate with IFNγRA (CD119) expression on PBT-TS lines (fig. S6D), which was previously related to low in vivo CLTX-CAR T cell function through induction of PD-L1. These results suggest that for GBM cells, MMP2 is the primary mediator of CLTX-Cy5.5 binding and the activation of CLTX-CAR T cells.

Fig. 6 CLTX-CAR T cell effector activity requires MMP2 expression on target cells.

(A) Degranulation of CLTX-EQ-28ζ CAR T cells against iPSC-NPCs and GBM cells from indicated TS lines. Shown are means ± SEM of duplicate wells. ns denotes not significant between PBT003-4 and PBT106 using an unpaired Student’s t test. (B and C) MMP2 mRNA expression (2−∆Ct compared with actin) in the PBT-TS lines were correlated to their corresponding stimulation of CLTX-CAR T cell degranulation (B) and CLTX-Cy5.5 staining (C), with goodness of fit for linear regression (r2) indicated in each graph (n = 16). (D) Degranulation of CLTX-CAR T cells tested against PBT003-4-TS or PBT106-TS lines that had been transduced with scrambled shRNA or shRNA targeting MMP2 (shMMP2). (E) Cytolytic activity of CLTX-CAR T cells against scrambled shRNA- or shMMP2-transduced GBM cells over 72 hours of coculture (E:T ratio, 1:4). (D and E) Shown are means ± SEM of duplicate wells; *P < 0.05 compared with scrambled shRNA-transduced targets using an unpaired Student’s t test. (F) Scrambled shRNA- or shMMP2-transduced GBM cells (5 × 106) were injected subcutaneously into the right flank of NSG mice. Mock-transduced or CLTX-EQ-28ζ CAR T cells (3 × 106) were then injected into the tumors at day 14 (tumor diameter, ~5 mm; n = 4 in each group), and tumor size was monitored over time with caliper measurements. *P < 0.05 and **P < 0.01 using one-way ANOVA with Bonferroni’s multiple comparison tests. (G) CLTX-CAR T cells were cultured with the indicated concentrations of soluble MMP2 in the absence (top row) or presence (bottom row) of PBT003-4-TS cells (E:T ratio, 1:4) for 24 hours. CAR T cell activation was determined by flow cytometric analysis of 4-1BB and CD69 surface expression. Quadrants were drawn on the basis of control staining, and percentages of double-stained cells are indicated in each histogram. (H) Elimination (% Killing) of PBT003-4-TS cells by CLTX-CAR T cells after 48 hours of coculture (E:T ratio, 1:4) in the presence of different concentrations of soluble MMP2; means ± SEM of duplicate wells are shown. (I) Immunofluorescence staining of MMP2, CLCN3, and actin at 2 hours after initiating coculture of CLTX-CAR T cells and PBT003-4-TS cells (E:T ratio, 1:1); arrowhead, T cell.

To further verify the dependence of CLTX-CAR activation on MMP2, we knocked down MMP2 expression in GBM cells using short hairpin RNA (shRNA) (shMMP2). The knockdown efficiently decreased MMP2 mRNA and soluble MMP2 secretion whereas resulting in only modest decreases in CLCN3 expression and minimal changes in ANXA2 expression (fig. S6E). We found that MMP2 knockdown in GBM cells substantially reduced CLTX-EQ-28ζ CAR T cell activation and cytotoxicity (Fig. 6, D and E, and fig. S6F). Furthermore, shMMP2 significantly reduced in vivo antitumor activity of CLTX-EQ-28ζ CAR T cells against established PBT003-4-TS and PBT106-TS tumors (P = 0.01 and P = 0.003, respectively) (Fig. 6F). Together, these results demonstrate that MMP2 is required for CLTX-CAR recognition and activation against tumor targets. MMP2 shows low to negligible expression in various human normal tissues in comparison to GBM (fig. S5E), which is consistent with our findings and other reports showing low CLTX-Cy5.5 binding against normal human cells and tissues (22).

MMP2 is a secreted MMP that can either associate with the membrane through interaction with αvβ3 integrin and MMP14 (also known as membrane type 1 MMP) or exist as a soluble enzyme (23, 53). Although our data and others indicate that CLTX binds to the membrane-bound form of MMP2 (23), we further investigated whether CLTX-CAR T cells may also respond to soluble MMP2. In the presence of soluble MMP2, we observed no activation of CLTX-CAR T cells (Fig. 6G) or reduction in CLTX-CAR T cell cytotoxicity against GBM cells (Fig. 6H); nor did adding soluble MMP2 restore CLTX-CAR activity after MMP2 knockdown in GBM target cells (fig. S6G). Consistent with CLTX-CAR recognition of MMP2, we observed that MMP2 was recruited to the CLTX-CAR T cell–GBM cell immunological synapse, and this interaction also colocalized with CLCN3 in PBT003-4 cells (Fig. 6I). The recruitment of MMP2 and CLCN3 to the immunological synapse did not occur when coculturing IL13Rα2-CAR T cells and GBM cells (fig. S6H). Moreover, we also observed colocalization of MMP2 and CLCN3 when CLTX peptide was applied to GBM cells (fig. S6I). These results suggest that interaction between CLTX-CAR T cells and GBM cells requires surface expression of MMP2 and involves CLCN3. Recruitment of CLCN3 after MMP2 interaction with CLTX peptide or CLTX-CAR T cells was consistent with another study using CLTX-coated liposomes (54). Together, our results suggest that membrane-associated MMP2 is necessary for CLTX-CAR T cell activation and that soluble MMP2 neither activates nor disrupts CLTX-CAR function.


This study demonstrates that CLTX, a peptide component of scorpion venom, can be successfully incorporated into a CAR construct to redirect cytotoxic T cells to target GBMs. The antitumor effects of CLTX-CAR T cells were robust and specific. Moreover, CLTX binding extended across a wider range of freshly dissociated patient tumor cells and patient-derived GBM cell lines than expression of IL13Rα2, HER2, EphA2, and EGFR antigens currently under consideration as CAR immunotherapy targets. Our study thus establishes that the intrinsic binding properties of a natural toxin peptide can be exploited in generating targeted CAR T cells.

The criteria for selection of GBM-associated immunotherapy targets are particularly stringent because of the sensitivity of the brain to off-target activity or reactions to the therapy (2). For the target antigens currently under clinical investigation, one selection criterion is their specific expression on GBM tumors. IL13Rα2 has negligible expression in normal brain and elsewhere (except as a cancer-testis antigen) (55, 56). EGFRvIII, the most common mutated form of EGFR, is restricted to multiple cancers including GBM (57). Other antigen targets such as EGFR are overexpressed by tumor GBM cells as compared to normal cells, and HER2, although moderately expressed in some normal tissues, shows only minimal expression on postnatal neurons or glial cells (58). Development of CLTX-CAR T cells was justified by previous reports of the safety of CLTX itself, as demonstrated in preclinical and clinical studies (25, 26, 31, 32). Specifically, no toxicity was observed in a previous clinical study using CLTX to deliver 131I to tumor sites (25) or in on-going clinical studies evaluating intravenous delivery of fluorophore-conjugated CLTX as a real-time imaging agent during surgery (32).

For the results presented here, we used mouse models to investigate potential toxicities of CLTX-CAR T cells. Safety evaluation in mouse models was supported by our observation that CLTX efficiently binds to mouse glioma cells and the fact that the putative target MMP2 is conserved between human and mouse. Therefore, although these murine models may only represent a surrogate for human safety, they provide relevant information about off-tumor targeting. Using these preclinical models, we determined that systemic (intravenous) and regional (ICT and ICV) delivery of CLTX-CAR T cells into both healthy and tumor-bearing mice did not show systemic toxicity. Furthermore, the organ distribution and persistence of systemically administered CAR T cells in tumor-bearing mice were similar to those of mock T cells. Overall, these results from preclinical models provide strong evidence for efficacy together with negligible off-tumor toxicity and support future clinical evaluation in patients with GBM. However, despite the clinical utilization of CLTX peptide, the potential immunogenicity of CLTX-CAR T cells may represent another safety consideration, which was not specifically addressed in these preclinical models.

Tumor recurrence remains a barrier to successful immunotherapy for GBM. Recurrence is commonly associated with the presence of a GSC subpopulation(s) resistant to radiotherapy and chemotherapy and characterized by high tumor-initiating potential (37). The virulence of GSCs has made them a critical consideration of GBM-targeted immunotherapy (59), and previous studies have demonstrated that GSCs can be as responsive to CAR T cell–mediated cytotoxicity as more differentiated GBM cells (33, 38, 39) despite their intrinsic immunosuppressive properties (60). Functional evaluations of CLTX-CAR T cells throughout this study have used patient-derived TS lines, which maintain stem cell–like properties under appropriate culture conditions (34). Furthermore, because the identification of GSCs using single surface markers may confer bias toward certain GBM molecular subtypes (61), we used two different markers, CD133 and CD44, to identify GSC-like subsets in primary tumor samples and showed that CLTX-Cy5.5 binding was similar both in GSC subsets and in non-GSC populations. Our results indicate that CLTX-CAR T cells inherit the binding properties of CLTX peptide and thus will be active against both GSC and non-GSC populations in patient tumors, suggesting the potential to target the “seeds” of GBM recurrence.

The choice of costimulatory signal has proven to be a critical element of CAR design. CAR T cells with CD28 (62) or 4-1BB (7, 63) costimulatory signals have been introduced to patients with hematological and solid tumors. In general, compared with CD28, 4-1BB costimulation has resulted in slower but more long-lasting T cell activation, as indicated by differential metabolic profiling (64), and consistent with CAR T cell expansion dynamics when coinfused into patients with B cell acute lymphoblastic leukemia (65). Therefore, it was unexpected that in this study, we found that CLTX-CAR T cells displayed degranulation and killing antitumor responses only when incorporating CD28 costimulation, whereas 4-1BB costimulation reduced both short-term and long-term effector activities. One possible explanation may be the nature of CLTX’s interaction with its receptors. Although the binding affinity between CLTX and membrane-bound MMP2 is not known, an early study identified a low-affinity CLTX receptor with high abundance on GBM cell membranes (21). Because CD28 has been reported to lower the affinity threshold for T cell receptor activation (66), it is possible that interaction between the CLTX tumor recognition domain and low-affinity GBM receptors requires CD28 costimulation to reach a threshold for initiating CLTX-CAR T cell activation. We noted that along with higher effector function, CLTX-EQ-28ζ and CLTX-CD8h-28ζ CAR T cells also more frequently expressed the exhaustion marker PD-1 compared to the other CLTX-CARs evaluated. This observation would be consistent with incomplete activation by other non–CD28-bearing CLTX-CAR T cells upon tumor binding because T cell exhaustion markers are also induced upon antigen engagement and serve as indicators of T cell activation (67). Further optimization of CLTX-CAR design may be achieved by protein design strategies to modulate its binding affinity through single amino acid mutations or replacements, which may be facilitated by the small size and helix-dominated structure of CLTX.

Our studies also provide important insights into the clarification of the interaction between CLTX and its receptor, which previously has not been well defined (27). The well-characterized inhibition of GBM cell migration and invasion by CLTX (23, 68) has been attributed to either the chloride channel CLCN3 through inhibiting the Cl flux required for cell shape changes during invasion (24) or to MMP2, thereby decreasing extracellular matrix cleavage (23). Interpretation of immunoprecipitation-based assays showing interaction between CLTX and its receptor(s) has been complicated by difficulties distinguishing between direct versus indirect association. Using genetic manipulations to knockdown MMP2, we determined that MMP2 expression is required for CLTX-CAR T cell activity. Together with previous discoveries that overexpressing MMP2 facilitates the binding of CLTX (30), our data suggest that membrane-associated MMP2 serves as the CLTX receptor or a critical component within a receptor complex. CLTX-CAR T cells did not respond to secreted MMP2, thus preserving their activation potential even in a soluble MMP2-rich microenvironment such as tumors (53). The development of CLTX-CAR T cells shows that CAR T cells are able to selectively target membrane-bound forms of secreted enzymes, in addition to the previously reported CAR designs against soluble proteins (69, 70). Because CAR T cell activation requires direct antigen engagement, our results suggest MMP2 as a direct target for CLTX, and this interaction may also recruit other associated proteins such as CLCN3. Similar patterns of CLTX’s association with tumor cells have been inferred by previous studies using CLTX-coated nanoparticles and fusion proteins (54, 71). Although CLCN3 expression appeared unrelated to CLTX-CAR activity, it might still participate in CLTX’s interaction with target cells. We also observed correlations between MMP2 expression, degranulation, in vivo CLTX-CAR efficacy, and CLTX-Cy5.5 staining intensity, suggesting that CLTX-Cy5.5 staining serves as an efficient strategy for identifying the potential responsiveness of GBM and other tumors to CLTX-CAR T cell therapy.

A generalizable finding arising from our in vivo CLTX-CAR studies is the observation that antitumor function of GBM-targeted CARs may be inhibited by adaptive resistance mechanisms of GBM tumors, an observation also reported in a clinical study showing up-regulated immunosuppressive pathways in patients with GBMs after treating with EGFRvIII-targeted CAR T cells (9). Although PD-L1 is expressed in a subset of GBMs (72), we found that PD-L1 could be strongly induced in GBM xenografts and TS lines after CAR T cell therapy. This induction was associated with IFNγ receptor expression, consistent with the mechanism of adaptive resistance characterized in metastatic melanomas (51). The inhibitory effect of PD-L1 on CLTX-CAR T cells is mostly on long-term in vivo function because initial CLTX-CAR activation (degranulation and in vitro killing) was not well correlated with the expression of tumor IFNγRA (CD119), which mediates PD-L1 induction. CLTX-CAR T cells displayed similar initial activation despite the differential IFNγRA expression of the two PBT-TS lines that we used for in vivo studies. Our results lead to the potential of correlating tumor IFNγ receptor expression with CLTX-CAR T cell therapeutic effect, as well as the possibility of combining CLTX-CAR T cells with checkpoint inhibitors for more effective GBM clearances.

Overall, we were able to observe broad GBM-targeting capability of CLTX-CAR T cells, consistent with the wide expression of its receptor, membrane-bound MMP2, in GBM samples (73). Of particular importance, CLTX-CAR T cells elicited potent cytotoxic responses against tumor cells with no or little expression of other targetable antigens (IL13Rα2, HER2, and EGFR). Furthermore, CLTX-CAR T cells efficiently eradicated GBM tumors in vivo with no observed toxicity, with the limitation that the GBM xenograft models may not fully recapitulate the invasive nature of GBM tumors in patients. We believe that CLTX-CAR T cells address two major hurdles to effective immunotherapy for GBM: reduction of antigen escape while maintaining tumor cell restriction. We suggest that CLTX-CAR T cells present a strategic combination of selective yet ubiquitous tumor targeting and are a candidate for clinical development as anti-GBM immunotherapy, mitigating antigen escape either as a single agent or in combination with other CAR T cells or immunotherapy strategies.


Study design

In this study, we evaluated the antitumor potency of CLTX-CAR T cells against GBM. First, CLTX binding was verified in various GBM samples. PBT cells were obtained from GBM resections at City of Hope under protocols approved by the City of Hope Internal Review Board. For experiments on functional evaluation, CAR T cells were generated from three different healthy donors and tested in vivo and in vitro against at least two independent GBM models. To ensure statistical power, for in vitro assays, two to five replicates within each condition were used to sufficiently represent intragroup variations and allow for defining statistical significance. In all in vivo experiments, 6- to 8-week-old NSG mice were used, and four to eight mice were included within each group to ensure statistical power, which enabled us to statistically distinguish tumor sizes and survival rates across groups. Before CAR T cell treatment, mice were randomized on the basis of bioluminescent imaging to ensure similar average tumor sizes across groups. The health condition of mice was monitored on a daily basis by the Department of Comparative Medicine at City of Hope, with euthanasia applied according to the American Veterinary Medical Association Guidelines. Investigators were not blinded when monitoring mice survival. For every mouse euthanized, the brain was collected to confirm the presence of GBM tumors. If a mouse died during imaging processes with no prior sign of tumor progression, then it was considered an anesthesia-related death and the mouse was excluded from survival analysis. The pathological conditions of mouse organs were determined by a mouse pathologist from the Veterinary Pathology Program at City of Hope. All primary data are reported in data file S1.

Isolation of PBT cells, establishment of neurospheres, and other cell lines

Resected brain tumor specimens were digested using a human tumor dissociation kit (Miltenyi Biotech Inc.) to generate PBT cells. TS lines were subsequently established from PBTs and maintained as described previously (38, 74). To generate cells for in vivo biophotonic imaging, these cells were engineered to express the firefly luciferase (ffLuc) reporter gene as previously described (38). Differentiation of TS lines was performed by withdrawal of EGF and fibroblast growth factor in the TS culture media and supplemented with 10% fetal calf serum (FCS), as described previously (38). FB-NSCs were established and characterized as reported previously (75). Astrocytes and NPCs were differentiated from healthy donor–derived iPSCs on the basis of established protocols (76).

DNA constructs

All CLTX-CAR constructs contain a CLTX peptide and the cytoplasmic domain of human CD3ζ, with different spacers including IgG4EQ [IgG4 with two point mutations (L235E and N297Q) (6)]; ∆CH2, IgG4-Fc with the CH2-domain deleted; CD8h; and L, a synthetic 10–amino acid short linker (data file S2). CAR constructs also contain CD4 or CD28 transmembrane domains and CD28 or 4-1BB costimulatory domains. All domains have been previously described (42, 44, 49). A truncated CD19 was also introduced in the construct to allow for potential enrichment and tracking of transduced cells. The ffLuc–green fluorescent protein construct for tumor biophotonic imaging was generated as described previously (38).

CAR T cell production

Blood products were obtained from healthy donors under protocols approved by the City of Hope Internal Review Board, and naïve/memory T cell (Tn/mem) isolation followed the procedures described in previous studies (49). Briefly, PBMCs were isolated by density gradient centrifugation over Ficoll-Paque (GE Healthcare) and then underwent sequential rounds of CliniMACS/AutoMACS depletion to remove CD14- and CD25-expressing cells, followed by a CD62L positive selection for Tn/mem cells. To generate CAR T cell products, T cells were stimulated with Dynabeads Human T expander CD3/CD28 (Invitrogen) at a 1:3 ratio (T cell:bead) and transduced with lentivirus to express CAR (multiplicity of infection, 2) in X-VIVO 15 (Lonza) containing 10% FCS with protamine sulfate (5 μg/ml) (APP Pharmaceuticals), recombinant human IL2 (rhIL2) (50 U/ml), and rhIL15 (0.5 ng/ml). Cultures were then maintained at 37°C and 5% CO2 under the same condition of media and cytokines (cytokines were replenished every other day). On day 7 after transduction, the CD3/CD28 Dynabeads were removed from cultures using the DynaMag-50 magnet (Invitrogen). CAR-transduced T cells were enriched by positive selection using anti-CD19 magnetic beads (STEMCELL Technologies). Cultures were propagated for 14 to 16 days before applying to assays or cryo-preservation. Mock-transduced T cells were generated by stimulating and culturing Tn/mem cells from the same donors as described above, without lentivirus addition.

GBM xenograft studies

All mouse experiments were approved by the City of Hope Institutional Animal Care and Use Committee (protocol no. 18059). Orthotopic GBM models were generated using NSG mice as previously described (48). Briefly, on day 0, ffLuc+ GBM cells (1 × 105 or 2 × 105) were stereotactically implanted into the right forebrain. After 8 (PBT106-TS) or 12 (PBT003-4-TS) days, mice were then treated by ICT application of 0.2 × 106, 0 5 × 106, or 1 × 106 CAR T cells, ICV route with 1 × 106 to 2 × 106 CAR T cells, or intravenous route with 1 × 106 to 10 × 106 CAR T cells (indicated in figure legends). Tumor volumes were determined by in vivo noninvasive optical biophotonic imaging using a Xenogen IVIS 100 as previously described (38). For subcutaneous tumor xenografts, GBM cells (5 × 106) were mixed with Matrigel (Corning) and injected into the right flank of NSG mice, and tumors were allowed to grow for 16 to 21 days until tumor sizes reached 5 mm by 5 mm. CAR T cells (2 × 106) were injected intratumorally, and tumors sizes were monitored until animal euthanasia when tumor sizes reached 15 mm by 15 mm. To acquire single cells for flow cytometric analysis, xenograft tumors were cut into pieces, physically dissociated, and filtered. The pathological conditions of mouse organs were determined by the Veterinary Pathology Program at City of Hope.

Statistical analysis

Data analysis was performed using Prism version 6.0 (GraphPad Software) and presented as stated in individual figure legends. Comparisons were determined using Student’s t test (two groups) or one-way analysis of variance (ANOVA) (three or more groups). For comparisons between three or more groups, Bonferroni’s multiple comparison tests were used to compare all or selected pairs of data (95% confidence intervals). Comparison of Kaplan-Meier survival data was performed using the log-rank (Mantel-Cox) test. Detailed comparisons in each experiment are described in the figure legends.


Materials and Methods

Fig. S1. Antigen expression and CLTX-Cy5.5 binding on GBMs.

Fig. S2. Activation of CLTX-CAR T cells after GBM stimulation.

Fig. S3. CLTX-EQ-28ζ and CLTX-CD8h-28ζ CARs initiate stronger T cell activity than other CLTX-CAR constructs.

Fig. S4. CLTX-CAR T cell targeting of GBM xenografts.

Fig. S5. CLTX-CAR T cells do not elicit off-tumor targeting in mouse models.

Fig. S6. MMP2 is necessary for CLTX-CAR T cell activation.

Movie S1. Killing of PBT003-4-TS–derived GBMs during a 72-hour coculture with the mixture of IL13Rα2-, HER2-, and EGFRvIII-targeted CAR T cells.

Movie S2. Killing of PBT003-4-TS–derived GBMs during a 72-hour coculture with CLTX-EQ-28ζ CAR T cells.

Data file S1. Primary data.

Data file S2. Amino acid sequence of spacers.


Acknowledgments: We thank the Department of Comparative Medicine and the cores of Synthetic and Biopolymer Chemistry, Small Animal Imaging, Light Microscopy, Mouse Pathology, and Solid Tumor Pathology, as well as B. Chang, J. Ruiz-Delgado, and L. Weng for technical assistance. We thank J. M. Olson for interactive discussions and intellectual feedback on this work. Funding: This work was supported by grants from the Ben and Catherine Ivy Foundation and NIH grant P30CA33572 (cores). D.W. is supported by NCI fellowship 5F99CA234923-02. Author contributions: Designing research studies: D.W., S.J.F., M.E.B., and C.E.B.; conducting experiments: D.W., R.S., W.-C.C., B.A., S.L.W., X.Y., A.B., and A.S.; data acquisition: D.W., R.S., and B.A.; analysis and interpretation: D.W., D.A., M.E.B., and C.E.B.; resources: M.G., K.A., L.L., Y.S., B.B., C.E.B., and S.J.F.; writing manuscript: D.W., D.A., J.R.O., M.E.B., and C.E.B.; supervision: S.J.F., M.E.B., and C.E.B. Competing interests: S.J.F. and C.E.B. receive royalty payments from Mustang Bio; all other authors declare that they have no competing interests. A patent associated with this study covering the CLTX-CAR has been held and submitted by City of Hope (WO2017066481A8) with M.E.B., C.E.B., S.J.F., and D.W. as inventors. Data and materials availability: All data associated with this study are present in the paper or the Supplementary Materials. PBT-TS lines are available from C.E.B. under a material transfer agreement with City of Hope.

Stay Connected to Science Translational Medicine

Navigate This Article