Therapeutically targeting glypican-2 via single-domain antibody-based chimeric antigen receptors and immunotoxins in neuroblastoma

Significance

Neuroblastoma is a childhood cancer that remains an important clinical challenge. It is fatal in almost half of the patients despite advances in multimodal treatments. In this report, we show that the cell surface glycoprotein glypican-2 (GPC2) is overexpressed in neuroblastomas when compared with normal tissues and that a high expression level is correlated with poor survival of neuroblastoma. We also describe that GPC2 has proliferative effects in neuroblastoma via activating Wnt signaling and its downstream target genes including N-Myc, a major driver for neuroblastoma tumorigenesis. We have produced the immunotoxins and chimeric antigen receptor T cells that target GPC2 and exhibit promising antitumor activities in cell and mouse models. This study suggests GPC2 as a promising target in neuroblastoma.

Abstract

Neuroblastoma is a childhood cancer that is fatal in almost half of patients despite intense multimodality treatment. This cancer is derived from neuroendocrine tissue located in the sympathetic nervous system. Glypican-2 (GPC2) is a cell surface heparan sulfate proteoglycan that is important for neuronal cell adhesion and neurite outgrowth. In this study, we find that GPC2 protein is highly expressed in about half of neuroblastoma cases and that high GPC2 expression correlates with poor overall survival compared with patients with low GPC2 expression. We demonstrate that silencing of GPC2 by CRISPR-Cas9 or siRNA results in the inhibition of neuroblastoma tumor cell growth. GPC2 silencing inactivates Wnt/β-catenin signaling and reduces the expression of the target gene N-Myc, an oncogenic driver of neuroblastoma tumorigenesis. We have isolated human single-domain antibodies specific for GPC2 by phage display technology and found that the single-domain antibodies can inhibit active β-catenin signaling by disrupting the interaction of GPC2 and Wnt3a. To explore GPC2 as a potential target in neuroblastoma, we have developed two forms of antibody therapeutics, immunotoxins and chimeric antigen receptor (CAR) T cells. Immunotoxin treatment was demonstrated to inhibit neuroblastoma growth in mice. CAR T cells targeting GPC2 eliminated tumors in a disseminated neuroblastoma mouse model where tumor metastasis had spread to multiple clinically relevant sites, including spine, skull, legs, and pelvis. This study suggests GPC2 as a promising therapeutic target in neuroblastoma.

Neuroblastoma is the most common extracranial solid tumor of children and the most frequently diagnosed neoplasm during infancy (1). It accounts for 8–10% of all childhood cancers and 15% of all pediatric cancer mortality (2). This cancer is derived from neuroendocrine tissue located in the sympathetic nervous system. Neuroblastoma is a complex and heterogeneous disease, with nearly 50% of patients having a high-risk phenotype. Despite the use of intensive multimodal treatments, these high-risk phenotype cancers often present with widespread dissemination and poor long-term survival (3). Approximately 45% of patients receiving standard therapy relapse and ultimately succumb from metastatic disease (4). As such, there is an urgent need for an effective neuroblastoma treatment.

One of the major challenges for the treatment of neuroblastoma and other pediatric cancers is the lack of effective therapeutic targets. Interestingly, glypican-2 (GPC2) was recently found as one of several mRNA transcripts that were highly expressed in multiple pediatric cancers, including neuroblastoma (5). Whereas these candidate markers have shown promise at the mRNA level, they have yet to be validated at the protein level. GPC2 belongs to a six member human glypican family of proteins (6). All of these proteins are attached to the cell surface by a GPI anchor (6). GPC2 is unique among the glypican family because it is expressed in the nervous system (7). It is known to participate in cellular adhesion and is thought to regulate the growth and guidance of axons (8). The role of GPC2 in the pathogenesis of neuroblastoma or any other cancers has not been reported.

The Wnt/β-catenin signaling pathway is highly conserved and plays an essential role in various processes of embryonic development (9). Additionally, it has been linked to pathogenesis of numerous adult and pediatric tumors (10). Wnt/β-catenin signaling has been shown to mediate neural crest cell fate and neural stem cell expansion. This signaling pathway may be of particular relevance to neuroblastoma cells because they arise from migratory neural crest-derived neuroblasts (11–13). Glypicans play an important role in developmental morphogenesis and are known to be modulators for the Wnt signaling pathway (14–17). It has been shown that GPC3, another member of the glypican family, may function as a coreceptor for Wnt and facilitate the Wnt/Frizzled binding in hepatocellular carcinoma cells (18, 19).

Antibody-based therapeutics are of growing significance in the field of cancer therapy. Antibodies can be used as vehicles to deliver a variety of effector molecules such as cytotoxic drugs or toxins to tumor cells. Recombinant immunotoxins are chimeric proteins composed of an antibody fragment fused to a toxin, such as the 38-kDa truncated fragment of Pseudomonas exotoxin (PE38). By combining the specificity of an antibody with the protein synthesis inhibitory domain from the exotoxin, it is possible to directly target cancer cells (20–24). Another antibody-based therapy that is emerging with clinical applications involves chimeric antigen receptor (CAR)-expressing T cells. CARs are composed of an antibody fragment fused to a transmembrane domain linked to a T-cell signaling moiety. T-cell expressing CARs (CAR T cells) have the ability to bind antigen directly, whereas normal T cells require antigen presented in MHC molecules. In recent years, CAR T-cell immunotherapy has emerged as one of the most promising approaches to treat leukemia (25–29). CAR T-cell immunotherapy has not been as successful in the treatment of solid tumors, in part due to the lack of tumor-specific targets. To improve engineered T-cell therapies in solid tumors, we will need to identify tumor antigens that can be safely and effectively targeted to discriminate cancers from normal tissues.

In the present study, we found that GPC2 protein was specifically overexpressed in neuroblastoma compared with its expression in normal peripheral nerve tissues and its high expression correlated with a poor prognosis for patients with neuroblastoma. We also found that down-modulation of GPC2 via siRNA or CRISPR-Cas9 suppressed neuroblastoma cell growth by attenuating Wnt/β-catenin signaling and reduced the expression of N-myc, a Wnt target gene and an oncogenic driver for neuroblastoma pathogenesis. Furthermore, we identified a group of human single-domain antibodies (LH1–LH7) to GPC2 by phage display technology. To evaluate their potential use for the treatment of neuroblastoma, we constructed immunotoxins using these single-domain antibodies and demonstrated that the LH7–PE38 immunotoxin inhibited growth of neuroblastoma xenograft tumors in mice. In addition, we produced CARs targeting GPC2 and expressed them in T cells isolated from multiple healthy donors. Here we found that CAR T cells could potently kill neuroblastoma cells. Notably, anti-GPC2 CAR T cells were effective in suppressing the dissemination of neuroblastomas in our mouse xenograft model.

Results

Generation of Anti-GPC2 Human Single Domain Antibodies.

To identify the antibodies specific for GPC2, we decided to screen a phage-display engineered human V_H single-domain antibody library. Enrichment was determined by counting the number of phages recaptured after each round of panning. Four rounds of panning resulted in an ≈1,000-fold enrichment of eluted phage (Fig. 1A). Phage pools after two rounds of panning exhibited enhanced binding to GPC2, whereas no binding to BSA was found with pooled phage from any of the four rounds of panning (Fig. 1B). At the end of the fourth round of panning, 27 individual clones were confirmed to be GPC2 binders by monoclonal phage ELISA. Subsequent sequencing analysis revealed seven unique binders (LH1–LH7). The GPC2–hFc OD₄₅₀ values of all seven clones were at least fivefold higher than that of BSA (Fig. 1C), further indicating the specificity of the phages to GPC2. Notably, LH1 and LH7 were the two most abundant binders and combined for 14 of the 27 identified (Fig. 1D). The LH5 clone was excluded from further study due to its low affinity for GPC2. To determine whether the identified clones would bind to other members of the human glypican family, we performed monoclonal phage ELISA using recombinant human GPC1 and GPC3. We also included mouse GPC2 to determine whether any of the binders had cross-species capabilities. As shown in Fig. 1E, the five clones LH1, LH2, LH4, LH6, and LH7 specifically bound to human GPC2, but not human GPC1 or GPC3. Interestingly, LH4 and LH6 are cross-reactive against both human and mouse GPC2 proteins. LH3 showed the highest binding signal to human GPC2 among all binders, but it was slightly cross-reactive with GPC1 and GPC3. To determine binding kinetics, we produced a LH7–Fc fusion protein and incubated it with GPC2 protein in solution on the Octet platform. The K_d value of the LH7–Fc fusion for GPC2 was 9.8 nM (Fig. 1F). Taken together, we have successfully identified a group of high-affinity anti-GPC2 human single-domain antibodies by phage display.

Fig. 1.

Isolation of GPC2-specific human single-domain antibodies by phage display. (A) Phage-displayed single-domain antibody fragments were selected against recombinant GPC2–hFc after four rounds of panning. A gradual increase in phage titers was observed during each round of panning. (B) Polyclonal phage ELISA from the output phage of each round of panning. BSA was used as an irrelevant antigen. (C) Monoclonal phage ELISA of the seven GPC2 binders. (D) Distribution of unique sequences of GPC2 binders in 27 selected phage clones. (E) Monoclonal phage ELISA analysis of cross-reactivity of GPC2 binders to human GPC1 and GPC3 and mouse GPC2. (F) Octet association and dissociation kinetic analysis for the interaction between various concentrations of the LH7 antibody and human GPC2. All data are represented as mean ± SEM of three independent experiments.

GPC2 Is Highly Expressed in Neuroblastoma but Not in Normal Human Tissues.

A microarray study showed that GPC2 mRNA was overexpressed in a panel of pediatric cancers, including neuroblastoma (5). To examine the expression of GPC2 protein in neuroblastoma, we used anti-GPC2 antibodies as research tools to examine established human cell lines and clinical tissue samples. Our Western blotting data demonstrated that GPC2 was highly expressed in five neuroblastoma cell lines, including LAN1, IMR5, LAN5, IMR32, and NBEB (Fig. 2A). GPC2 was only weakly expressed in the SKNSH, the sixth neuroblastoma cell line that we evaluated. To assess the clinical relevance of our observation, GPC2 protein levels were measured in human specimens from patients with neuroblastoma or nonmalignant disease by immunohistochemistry using the LH7 antibody. As shown in Fig. 2B, GPC2 labeling was readily apparent in specimens derived from patients with neuroblastoma (i–iv), but essentially undetectable in normal peripheral nerves from patients with nonmalignant disease (v and vi). Neuroblastoma tumor tissues showed strong GPC2 staining in 13 of the 25 cases (52%) (SI Appendix, Fig. S1). To further analyze GPC2 expression in normal human tissues, we used a US Food and Drug Administration (FDA)-recommended human normal tissue array and probed it with the LH7 antibody. As shown in Fig. 2C, no significant GPC2 staining was observed in the normal tissues, including essential organs such as the brain, heart, lung, and kidney. These results would suggest a tumor-specific expression of GPC2. The complete panel of all 32 types of normal tissues stained for GPC2 expression is shown in SI Appendix, Fig. S2. In addition, we also measured GPC2 mRNA levels in a human normal tissue array by quantitative real-time PCR. GPC2 mRNA expression was not found in any normal tissues except for a moderate mRNA expression in thymus and testis (SI Appendix, Fig. S3). However, our immunohistochemistry analysis showed no specific binding of the LH7 antibody for either testis (C1) or thymus (D2) (SI Appendix, Fig. S2). These data strongly support tumor-specific expression of GPC2 and suggest it as a promising neuroblastoma biomarker.

Fig. 2.

GPC2 expression in human neuroblastoma tumors and normal human tissues. (A) GPC2 protein levels in human neuroblastoma cell lines, including SKNSH, LAN1, IMR5, LAN5, IMR32, and NBEB as determined by Western blotting. (B) Expression of GPC2 in neuroblastoma tumors (i–iv) and normal nerve (v and vi) tissues as determined by immunohistochemistry. (C) Expression of GPC2 in human normal tissues, including nerve, brain, heart, lung, liver, stomach, small intestine, colon, pancreas, spleen, kidney, and thyroid as determined by immunohistochemistry. The tissues were labeled with 1 μg/mL LH7–mFc antibody. The final magnification of all images was 100×. (D) Kaplan–Meier analysis of overall survival in patients with neuroblastoma with high GPC2 mRNA expression (n = 18) and low GPC2 mRNA expression (n = 458) from the Kocak dataset in the R2 Genomics Analysis and Visualization Platform. (E) Kaplan–Meier analysis of event-free survival in patients with neuroblastoma with high GPC2 mRNA expression (n = 20) and low GPC2 mRNA expression (n = 456) from the Kocak dataset.

There has been evidence that GPC3 expression or other glypicans (e.g., GPC1) has been correlated with poor prognosis in hepatocellular carcinoma or other types of cancer (30–33). To analyze a possible correlation between GPC2 mRNA levels and survival of patients with neuroblastoma, we used the R2 Genomics Analysis and Visualization Platform. Patients with high GPC2 expression exhibited poorer overall survival and event-free survival compared with patients with low GPC2 expression (Fig. 2 D and E).

We examined the ability of the LH7 antibody to bind GPC2 on neuroblastoma cells by flow cytometry. LH7 showed specific binding to IMR5, LAN1, IMR32, and LAN5 neuroblastoma cells (SI Appendix, Fig. S4A). In addition, LH7 exhibited no binding to SKNSH cells, which is consistent with the low expression of GPC2 in this neuroblastoma cell line (Fig. 2A). Furthermore, we quantified the number of cell surface GPC2 sites per cell using flow cytometry. LAN5 and IMR32 cells expressing native GPC2 contain between 10⁴ and 10⁵ sites per cell, whereas LAN1 and IMR5 cells contain between 10³ and 10⁴ GPC2 sites per cell (SI Appendix, Fig. S4B). SKNSH cells showed an extremely low number of cell surface GPC2 sites, with only 433 sites per cell. Taken together, we have demonstrated that GPC2 is a tumor-specific cell surface antigen in neuroblastoma.

Silencing of GPC2 Inhibits Neuroblastoma Cell Growth via Suppression of Wnt/β-Catenin Signaling.

To analyze the role of GPC2 in neuroblastoma cell growth, we used siRNA and CRISPR-Cas9 techniques to silence GPC2 in two neuroblastoma cell models (IMR5 and LAN1). Three different GPC2 siRNAs were used to avoid potential off-target effects of siRNA. GPC2 knockdown efficiency was confirmed by Western blotting, which showed substantial reductions of GPC2 levels in both cell lines (Fig. 3A). As shown in Fig. 3B, GPC2 siRNAs suppressed the growth of neuroblastoma cells within 3 d of transfection by ≈40–50% compared with cells transfected with scrambled siRNA. To validate the oncogenic effect of GPC2 in neuroblastoma, we generated GPC2 KO neuroblastoma cells by using CRISPR-Cas9. Three single-guide RNAs (sgRNAs) targeting different GPC2 exons (exons 1–3) were transfected into IMR5 cells. Expression of GPC2 protein was almost completely abolished in the sgRNA-transfected cells (Fig. 3C). As shown in SI Appendix, Fig. S5A, a 25–50% reduction of growth was observed in GPC2 KO cells compared with vector control cells after 3 d of growth. In addition, KO of GPC2 induced apoptosis in neuroblastoma cells as measured by elevated expression of cleaved-Poly (ADP ribose) polymerase (PARP) (SI Appendix, Fig. S5B) and increased activity of caspase-3 and -7 (Fig. 3D).

Fig. 3.

Genetic silencing of GPC2 inhibits neuroblastoma tumor cell growth and induces apoptosis by suppressing Wnt/β-catenin signaling. (A) GPC2 protein expression in LAN1 and IMR5 neuroblastoma cells after siRNA-mediated knockdown of GPC2. (B) Inhibition of tumor cell growth by GPC2 siRNAs in both LAN1 and IMR5 cell lines. (C) GPC2 expression in IMR5 neuroblastoma cells after GPC2 KO using the CRISPR-Cas9 technique. GPC2 KO decreased active β-catenin protein levels at 72 h posttransfection. (D) Caspase 3/7 activity in IMR5 cells after treatment with GPC2 targeted sgRNA. (E) Protein expression of Wnt3a and Wnt11 in neuroblastoma cell lines. (F) Interaction between GPC2 and Wnt3a as determined by immunoprecipitation. (G) Reduction of active β-catenin levels by LH7 treatment after 6 h in HEK-293 SuperTopFlash cells that were stimulated with Wnt3a CM. (H) LH7 suppressed the expression of β-catenin in HEK-293 SuperTopFlash cells that were stimulated with LiCl and/or Wnt3a CM. Whole cell lysates were collected after 6 h of treatment. (I) The anti-GPC2 antibodies decreased TopFlash activity in Wnt3a-activated HEK-293 SuperTopFlash cells after 6 h of treatment. (J) N-Myc protein level in neuroblastoma cell lines as determined by Western blotting. (K) Inhibition of N-Myc expression by silencing GPC2 in neuroblastoma cells. (L) The proposed mechanism mediated by anti-GPC2 antibodies to inhibit neuroblastoma cell growth. Blockade of GPC2 suppresses the expression of β-catenin and its targeted genes, including N-Myc. All data are represented as mean ± SEM of three independent experiments. *P < 0.05, **P < 0.01.

We hypothesized that GPC2 could be an extracellular modulator of Wnt signaling in neuroblastoma cells. GPC3 has been shown to interact with Wnt and suppress hepatocellular carcinoma cell proliferation (19). To determine whether GPC2 could affect Wnt signaling in neuroblastoma cells, active β-catenin levels were measured. As shown in Fig. 3C, the expression of active β-catenin was lower in GPC2 KO IMR5 cells than in vector control cells (Fig. 3C). Interestingly, we found the expression of Wnt3a and Wnt11 in all of the neuroblastoma cell lines expressing high levels of GPC2 (LAN1, IMR5, LAN5, IMR32, and NBEB). However, Wnt3a was undetectable and Wnt11 expression was extremely low in the SKNSH cell line that has low GPC2 expression (Fig. 3E). To determine the interaction of GPC2 and Wnt, we conducted a coimmunoprecipitation assay using Wnt3a-conditioned media (CM) and demonstrated that GPC2 could interact with Wnt3a (Fig. 3F). Furthermore, we used the luciferase-expressing HEK-293 SuperTopFlash cell model to analyze the function of GPC2. As shown in SI Appendix, Fig. S6, GPC2 was expressed in HEK-293 cells. Treating cells with the LH7 single-domain antibody decreased the Wnt3a-induced active β-catenin levels in a dose-dependent manner (Fig. 3G). Lithium chloride (LiCl) is a GSK3β inhibitor and an intracellular β-catenin signaling inducer (19). As shown in Fig. 3H, a combination of Wnt3a and LiCl showed synergistic elevation of β-catenin expression. Interestingly, the elevated β-catenin expression can be reduced by LH7 treatment, supporting the idea that the Wnt/β-catenin pathway can be directly modulated by the addition of the LH7 antibody. In addition to LH7, two other anti-GPC2 single-domain antibodies showed dose-dependent reduction of Wnt signaling, but to a lesser degree (Fig. 3I). Notably, LH7 at 100 μg/mL resulted in a 90% reduction of β-catenin signaling compared with control human IgG. The data indicate that the LH7 single-domain antibody has the most inhibitory effect on Wnt signaling.

MYCN amplification occurs in ≈25–33% of neuroblastoma cases and results in N-Myc protein overexpression (1). Patients with MYCN-amplified tumors usually have a very poor prognosis. In addition, studies have shown that Wnt/β-catenin signaling acts upstream of N-Myc to regulate lung and limb development (34, 35). As shown in Fig. 3J, N-Myc protein was found in GPC2 high-expressing neuroblastoma cells (LAN1, IMR5, LAN5, and IMR32) but not in GPC2 low-expressing SKNSH cells. Furthermore, we found that silencing of GPC2 suppressed N-Myc expression (Fig. 3K). Taken together, our data show that GPC2 is involved in Wnt signaling, and that targeting GPC2 by single-domain antibodies such as LH7 can suppress neuroblastoma cell growth by inhibiting Wnt signaling and down-regulating Wnt target genes including N-Myc, an oncogenic driver of neuroblastoma pathogenesis. Fig. 3L shows a working model based on our observations.

GPC2-Specific Immunotoxins Inhibit Neuroblastoma Growth.

To determine whether GPC2 could be used as a target of immunotoxins for the treatment of neuroblastoma, we constructed three immunotoxins using the LH1, LH4, and LH7 binding domains. All immunotoxins were expressed in Escherichia coli, refolded in vitro, and isolated with over 90% purity (Fig. 4A). The binding affinities of all three immunotoxins on purified GPC2 protein were measured by ELISA. As shown in SI Appendix, Fig. S7, the calculated EC₅₀ values for the three immunotoxins were in the range of 4.6 nM to 43.9 nM. We found that the EC₅₀ value (18 nM) for the LH7–PE38 monomeric immunotoxin in ELISA was similar to the K_d value (9.8 nM) of the LH7–Fc fusion protein (Fig. 1F), indicating the immunotoxin retained binding properties of the original single-domain antibody. To determine the cytotoxicity of all immunotoxins in vitro, we examined the inhibition of cell proliferation on a panel of cell lines using the WST cell proliferation assay. All three immunotoxins potently and selectively inhibited the growth of GPC2⁺ cell lines LAN1 and IMR5 with similar IC₅₀ values of 0.5–1.2 nM. LAN1 and IMR5 cells were poorly sensitive to an irrelevant immunotoxin targeting mesothelin (IC₅₀ for LAN1, 8 nM; IC₅₀ for IMR5, 34 nM). None of the immunotoxins affected the growth of low GPC2-expressing SKNSH cells (Fig. 4 B–D).

Fig. 4.

Recombinant immunotoxins against GPC2 inhibit neuroblastoma tumor growth in vitro and in vivo. (A) Purity of LH1–PE38 (molecular weight of 53 kDa), LH4–PE38 (molecular weight of 52 kDa), and LH7–PE38 (molecular weight of 52 kDa) as determined by SDS/PAGE. (B–D) Effectiveness of anti-GPC2 immunotoxins on the growth of IMR5 (B), LAN1 (C), and SKNSH (D) cell lines, as measured by the WST-8 assay. An antimesothelin immunotoxin was used as an irrelevant control immunotoxin. (E) Toxicity detection of LH7–PE38 in vivo. Athymic nu/nu nude mice were treated with indicated doses of immunotoxin i.v. every other day for a total of 10 injections. Each arrow indicates an individual injection (n = 5 per group). (F) Antitumor activity of LH7–PE38. Athymic nu/nu nude mice were s.c. inoculated with 1 × 10⁷ LAN1 cells mixed with Matrigel. When tumors reached an average volume of 150 mm³, mice were treated with a 0.4-mg/kg dose of LH7–PE38 i.v. every other day for 10 injections. Each arrow indicates an individual injection; n = 5 per group. *P < 0.05. (G) Body weight of the mice treated in F. (H) Representative pictures of tumors from control and treated mice at the end of study. Values represent mean ± SEM. IT, immunotoxin.

To evaluate the antitumor activity of LH7–PE38 in vivo, we s.c. inoculated nude mice with LAN1 cells. When tumor reached an average volume of 150 mm³, mice were treated with LH7–PE38 every other day for a total of 10 injections. The different dose concentrations were used to determine the relative toxicity of the LH7–PE38 immunotoxin. As shown in Fig. 4E, the mice tolerated 0.4 mg/kg LH7–PE38 well. However, mice treated with 0.8 mg/kg died after 5 injections. Only two mice survived after 10 injections at the 0.6 mg/kg dose. Notably, 0.4 mg/kg of LH7–PE38 inhibited tumor growth during treatment without affecting body weight (Fig. 4 F and G). At the end of treatment, tumor volumes in the LH7–PE38-treated group were significantly smaller than those in the control group (Fig. 4H). In addition to measuring body weight, we also performed toxicology studies to further evaluate any side effects of LH7–PE38 at 0.4 mg/kg. The LH7–PE38-treated mice had an increase in white blood cells, indicating that the immunotoxin could cause inflammatory effects in vivo (SI Appendix, Table S1). In addition, the LH7–PE38-treated group showed an increase in alanine aminotransferase; however, we did not find any gross evidence of liver damage following mouse necropsy. All organ weights of the treated mice were statistically similar to those of the control group, except for the spleen. No significant differences were detected in any other parameters measured. In conclusion, we found that immunotoxins based on our anti-GPC2 human single domains inhibited neuroblastoma cell proliferation both in vitro and in vivo.

GPC2 CAR T Cells Kill Neuroblastoma Cells.

To explore other therapeutic approaches, we constructed CARs containing anti-GPC2 antibody single domains linked to a CD8α hinge and transmembrane region, followed by the 4-1BB costimulatory signaling moiety and the cytoplasmic component of CD3ζ signaling molecule. The upstream GFP reporter was coexpressed with CAR using the “self-cleaving” T2A peptide (Fig. 5A). The genetically modified T cells began to expand after activation (Fig. 5B). On day 9, the expression of GPC2 CARs in the transduced T cells was demonstrated through GFP expression. The transduction efficiencies of CARs were between 21% and 47% (Fig. 5C). To determine whether T cells targeting GPC2 could specifically recognize and kill GPC2⁺ neuroblastoma cells, a luminescent-based cytolytic assay was established using the neuroblastoma cells engineered to express luciferase. As shown in Fig. 5D, IMR5 cells, which express high levels of GPC2, are resistant to mock-transduced T-cell–mediated killing. This resistance was true even at effector:target ratios as high as 8:1. Conversely, IMR5 cells were efficiently lysed by the GPC2 CAR T cells in a dose-dependent manner. In addition, anti-GPC2 CAR T cells demonstrated equivalent lytic capacity against LAN1 neuroblastoma cells (SI Appendix, Fig. S8). As expected, the mock-transduced T cells and GPC2 CAR T cells showed similarly low cytolytic activity against the low GPC2-expressing SKNSH cells (Fig. 5E). A cytokine production assay revealed that GPC2 CAR T cells produced significantly more IFN-γ and TNF-α after exposure to IMR5 cells than the mock T cells (Fig. 5 F and G). However, little to no induction of IFN-γ and TNF-α secretions were observed when mock or GPC2 CAR T cells were cocultured with SKNSH cells.

Fig. 5.

The CAR T cells targeting GPC2 kill neuroblastoma cells. (A) Schematic diagram of bicistronic lentiviral constructs expressing CARs targeting GPC2 along with GFP using the T2A ribosomal skipping sequence. (B) Timeline of CAR T-cell production. (C) GPC2-specific CAR expression on human T cells transduced with lentiviral particles was analyzed using flow cytometry by detection of GFP fluorescence. (D–E) Cytolytic activities of GPC2 targeting CAR T cells in cell assays. The luciferase-expressing IMR5 (D) and SKNSH (E) neuroblastoma cells were cocultured with mock or GPC2 CAR-transduced T cells at the indicated effector (E):target (T) ratios for 20 h, and specific lysis was measured using a luminescent-based cytolytic assay. (F–G) The above culture supernatants at an E:T ratio of eight were harvested to measure IFN-γ (F) and TNF-α (G) secretions via ELISA. All data are represented as mean ± SEM of three independent experiments. *P < 0.05, **P < 0.01.

We tested the killing ability of CAR T cells generated from eight individual human donors. At an effector:target ratio of 8:1, GPC2-specific CAR T cells lytic activity against IMR5 neuroblastoma cells ranged from 44% to 71%, with an average of 56% (Fig. 6A). Minimal cell lysis was observed in IMR5 cells treated with mock T cells (Fig. 6B). Next, we assessed the antitumor activity of GPC2-targeting CAR T cells in nude mice i.v. engrafted with luciferase-expressing IMR5 cells. Although LH3 CAR T cells were the most potent in the cell killing assay (Fig. 5D), the LH3 phage binder was also cross-reactive with other glypican members (e.g., GPC3) (Fig. 1E). Therefore, we chose LH7 for preclinical testing in neuroblastoma models. Bioluminescence imaging using IVIS showed that IMR5-bearing nude mice developed disseminated tumor lesions surrounding spine and bones (Fig. 6C). Notably, LH7 CAR T cells effectively suppressed metastatic tumors after 14 d of T-cell infusion, whereas mock T cells failed to reduce tumor burden (Fig. 6 C and D). Four of eight (50%) of the mice treated with CAR T cells targeting GPC2 were tumor-free at the end of this study. Neither mock T cell nor LH7 CAR T cell treatment affected mice body weight (SI Appendix, Fig. S9). The efficacy of GPC2-targeting CAR T cells was also evaluated in a LAN1 xenograft mouse model. LH7 CAR T cells initially led to a reduction in tumor size and significantly suppressed tumor growth compared with the control group at the end of the study (SI Appendix, Fig. S10). Taken together, we have demonstrated that our single-domain antibodies can be used to construct CAR T cells that are able to kill GPC2-expressing neuroblastoma cells in cell and mouse models.

Fig. 6.

GPC2-specific CAR T cells demonstrate potent activity in mice bearing human neuroblastomas. (A–B) Cytotoxic activity of LH7 CAR T cells derived from multiple donors. PMBCs were isolated from eight healthy donors. The luciferase-expressing IMR5 cells were cocultured with LH7 CAR-transduced T cells (A) or mock T cells (B) at the indicated effector (E):target (T) ratios for 20 h, and specific lysis was measured using a luminescent-based cytolytic assay. (C) LH7 CART demonstrated potent antitumor activity by suppressing metastatic IMR5 neuroblastoma cells as measured by bioluminescent imaging. Animals (n = 8 per group) were i.v. injected with a single infusion of 30 × 10⁶ mock T cells or LH7 CAR T cells. (D) Quantitation of bioluminescence in mice treated in C. Values represent mean ± SEM.

Discussion

In the present study, we reported that GPC2 protein expression levels were elevated in human neuroblastoma tumors compared with normal human tissues. Additionally, silencing of GPC2 decreased neuroblastoma cell viability and induced apoptosis. We also found that GPC2 modulated Wnt/β-catenin signaling and the expression of the key oncogenic driver gene N-Myc in neuroblastoma. By using phage display, we succeeded in identifying a group of human single-domain antibodies specific for GPC2. The immunotoxins and CAR T cells based on these antibody binding domains significantly inhibited neuroblastoma tumor cell growth. Our observations demonstrate an important role for GPC2 as a target candidate of antibody-based therapies for the treatment of neuroblastoma.

One current approach to treating high-risk patients with neuroblastoma is immunotherapy targeting a tumor-associated antigen, such as, the disialoganglioside GD2. Anti-GD2 antibodies have been tested in clinical trials for neuroblastoma, with proven safety and efficacy (3, 36). The FDA approved Unituxin, in combination with granulocyte-macrophage colony-stimulating factor (GM-CSF), interleukin-2 (IL-2), and 13-cis-retinoic acid (RA), for the treatment of patients with high-risk neuroblastoma in 2015. However, in patients with advanced disease, anti-GD2 antibodies show only limited activity. GD2 therapy is also associated with severe pain toxicity (37, 38). These challenges emphasize the urgent need to discover therapeutic targets for neuroblastoma therapy.

We found significant elevation of GPC2 protein levels in about 50% of neuroblastoma tumors obtained from clinical specimens (13 of 25). Immunohistochemistry confirmed that GPC2 protein expression was not detected in normal peripheral nerve tissues. Our studies also demonstrated that GPC2 protein was either present at low levels or not expressed in most normal tissues in our comprehensive array. This finding was especially true of the essential organ tissues included in our array. In addition, using the data from a study cohort in the R2 Genomics Analysis and Visualization Platform, we found that both overall survival and event-free survival of the patients with neuroblastoma with high GPC2 expression were significantly shorter than those with low GPC2 expression. Overall, we show that GPC2 protein is uniquely overexpressed in neuroblastoma and that its high expression may be correlated with poor survival. We also found an inhibitory effect of GPC2 in neuroblastoma cell growth using siRNA and CRISPR-Cas9 approaches. Our results suggest the possibility that the suppression of GPC2 expression in neuroblastoma cells may induce apoptosis. Taken together, our data show that GPC2 is a promising target candidate for neuroblastoma therapy.

Given the importance of Wnt/β-catenin signaling in neuroblastoma (10–12), we decided to determine whether GPC2 inhibition could suppress this signaling pathway in neuroblastoma cells. GPC2 inhibition by our anti-GPC2 antibodies or KO by sgRNA was found to reduce the expression of active β-catenin and suppress target genes that may regulate neuroblastoma cell proliferation and survival. We notice the discrepancy between the IC₅₀ (0.3–1.3 μM) of Wnt signaling inhibition and EC₅₀ of binding activity (5–50 nM) of single domains (SI Appendix, Fig. S7). For example, the IC₅₀ for the best single domain (LH7) is 0.3 μM (25 µg/mL), 17-fold more than its EC₅₀ (18 nM), likely because in the luciferase reporter assay, a saturation level of the antibody is needed to functionally block the interaction of GPC2 and Wnt and/or other associated proteins (e.g., Frizzled). We previously found that GPC3 could interact with Wnt3a to suppress hepatocellular carcinoma cell proliferation (19). Wnt11 is secreted in regions adjacent to the neural crest and could induce neural crest migration (39). It has been shown that Wnt11 mRNA was highly expressed in neuroblastoma clinical samples (40). Thus, we particularly determined the expression of Wnt3a and Wnt11 in neuroblastoma cell lines. Interestingly, Wnt3a and Wnt11 were expressed in GPC2 high-expressing neuroblastoma cells (e.g., LAN1, IMR5, LAN5, and IMR32). By contrast, Wnt3a and Wnt11 proteins were either not detected or poorly detected in GPC2 low-expressing SKNSH cells. The differences in Wnt protein expression are in agreement with the sensitivity of GPC2-targeted immunotoxins and CAR T cells in LAN1/IMR5 and SKNSH cell lines. Furthermore, we demonstrated that GPC2 could coimmunoprecipiate with Wnt3a. These observations are consistent with previous reports showing that activation of Wnt/β-catenin signaling contributes to the aggressiveness of neuroblastoma (41, 42) and indicate the role of GPC2 in modulation of Wnt signaling in neuroblastoma cells. It is possible that GPC2 can interact with other Wnt molecules beyond Wnt3a. Future studies are necessary to understand the interactions of various Wnt molecules and GPC2 in the tumor microenvironment to better understand the role of GPC2 in modulating Wnt signaling in cancer cells. N-Myc is a key driver for neuroblastoma tumorigenesis (43–45). Here we demonstrated that N-Myc was expressed in MYCN-amplified neuroblastoma cells, including LAN1, IMR5, LAN5, and IMR32 (Fig. 3J). However, N-Myc protein was not found in MYCN-nonamplified SKNSH neuroblastoma cells. Interestingly, in the present study we found genetic silencing of GPC2 significantly inhibited the expression of N-Myc in neuroblastoma cells. It has been shown that Wnt signaling can regulate N-Myc expression level and β-catenin may activate the promoter of N-Myc during development (34, 35). Our result indicates that GPC2 can down-regulate N-Myc expression by inhibiting Wnt/β-catenin signaling. Protein surfaces contain clefts that are relatively inaccessible to conventional antibodies as a result of steric hindrance. Interestingly, single-domain antibodies have the ability to bind in protein clefts or hidden substrate pockets not accessible to conventional antibodies (46, 47). We previously identified a human single-domain antibody that recognizes a cryptic functional site on GPC3 and inhibits Wnt signaling in liver cancer (48). In the present study, we isolated a group of seven representative binders specific for GPC2, and all of these single-domain antibodies significantly inhibited Wnt/β-catenin signaling in neuroblastoma cells. Together, these studies indicate that single-domain antibodies are an emerging class of promising therapeutic candidates that can inhibit the signaling related to the growth of cancer cells by blocking receptor–ligand interactions.

The immunotoxins based on our anti-GPC2 antibodies demonstrated highly specific and potent killing of neuroblastoma in both in vitro and in vivo mouse models. In the present study, the irrelevant antimesothelin immunotoxin shows modest background cytotoxicity at high concentrations. It has been shown that the immunotoxins (e.g., mPE24) containing only the enzymatic domain (domain III) has significantly less nonspecific cytotoxicity than PE38, suggesting that domain II may potentially contribute to nonspecific cytotoxicity (49). To improve therapeutic index and reduce potential side effects, future reengineering of the anti-GPC2 immunotoxin by removing domain II may be useful for clinical development. In mouse testing, the optimal dose appears to be 0.4 mg/kg, which is similar to the dose of other immunotoxins that are currently being evaluated in preclinical and clinical stages (including phase III) (50). Despite not observing any obvious side effects in mice treated with anti-GPC2 immunotoxins, we noted an increase in alanine aminotransferase. This increase could be indicative of liver damage, but we did not note any gross evidence of liver damage upon mouse necropsy. Further comprehensive preclinical studies are needed to validate the effect of anti-GPC2 immunotoxins systematically, including pharmacodynamics, pharmacokinetics, and influence on liver function, before identifying the most promising candidate for use in humans. Because PE38 is a bacterial protein, it is highly immunogenic in patients with solid tumors that have normal immune systems (51). Future efforts using immunosuppressive drugs or less immunogenic toxins should be pursued for clinical development of anti-GPC2 immunotoxins.

CAR T cells have been shown to be a promising T-cell–based immunotherapy in leukemia (25, 26, 29, 52, 53). CAR T cells targeting CD19 have resulted in sustained complete responses and shown complete response rates of ≈90% in patients with relapsed or refractory acute lymphoblastic leukemia (54). However, CAR T-cell therapies have not been successful in treating solid tumors. In the present study, we sought to evaluate the use of CAR T cells in treating neuroblastoma. We established an in vivo bioluminescent model of disseminated neuroblastoma in mice. Most neuroblastomas begin in the abdomen in the adrenal gland or next to the spinal cord, or in the chest. Neuroblastomas can spread to the bones, such as in the face, skull, pelvis, and legs. They can also spread to bone marrow, liver, lymph nodes, skin, and orbits. In our study, we frequently found disseminated tumors near the spine and in the bones of face, skull, legs, and pelvis, indicating the clinical relevance of our animal model. We then demonstrated that a single infusion of LH7 CAR T cells significantly suppressed the growth of metastatic neuroblastoma cells in mice. Whereas the immunotoxins can effectively inhibit neuroblastoma growth, CAR T cells targeting GPC2 can cause complete remission in 50% of treated mice. The persistence of CAR T cells is a major challenge in solid tumors. Preclinical and clinical studies investigating combinations of CAR T cells with immune checkpoint blockade therapies would be pursued and examined (54). Future comprehensive preclinical studies will be needed to validate GPC2-targeting CAR T-cell–based immunotherapy for neuroblastoma.

In conclusion, we found that the GPC2 protein is highly expressed in neuroblastoma and produced single-domain antibodies against GPC2. These antibodies can potentially be used either as immunotoxins or CARs in neuroblastoma treatment. This report establishes GPC2 as a potential therapeutic target in cancer. The single-domain antibodies described here are promising candidates that should be further evaluated as cancer therapeutics for the treatment of neuroblastoma and other GPC2⁺ cancers.

Materials and Methods

Cell Culture.

Human neuroblastoma cell lines, including SKNSH, LAN1, IMR5, IMR32, NBEB, and LAN5, were obtained from collaborators at the National Cancer Institute (NCI). IMR5, LAN1, and SKNSH cell lines were also transduced with lentiviruses expressing firefly luciferase, which were obtained from NCI-Frederick (55). Peripheral blood mononuclear cells (PBMCs) were isolated from the blood of eight healthy donors using Ficoll (GE Healthcare) according to the manufacturer’s instructions. The aforementioned cell lines were cultured in RPMI-1640 medium supplemented with 10% FBS, 1% l-glutamine, and 1% penicillin–streptomycin at 37 °C in a humidified atmosphere with 5% CO₂. The HEK-293T cell line was obtained from the American Type Culture Collection. The HEK-293 SuperTopFlash stable cell line was a kind gift from Jeremy Nathans, Johns Hopkins University, Baltimore. These two cell lines were grown in DMEM supplemented with 10% FBS, 1% l-glutamine, and 1% penicillin–streptomycin at 37 °C in a humidified atmosphere with 5% CO₂. All cell lines were authenticated by morphology and growth rate and were mycoplasma-free.

Phage Display and Biopanning.

A combinational engineered human V_H single-domain phage display library, with an estimated size of 2.5 × 10¹⁰, was used for the screening and has been previously described (56). The phage library was subjected to four rounds of panning on Nunc 96-well Maxisorp plate (Thermo Scientific) as described previously (57–59). Details are provided in the SI Appendix, SI Materials and Methods.

Antibody Binding Assay.

The binding kinetics of the LH7 antibody to GPC2 was determined using the Octet RED96 system (FortéBio) as described previously (54). Briefly, all experiments were performed at 30 °C and reagents were prepared in 0.1% BSA, 0.1% Tween20 PBS, pH 7.4 buffer. Biotinylated GPC2–hFc protein was immobilized onto streptavidin biosensors, which were subsequently used in association and dissociation measurements for a time window of 600 s and 1,800 s, respectively. Data analysis was performed using the FortéBio analysis software provided with the instrument.

Immunohistochemistry.

The human neuroblastoma tissue array and normal tissue array were purchased from US Biomax. The tissue arrays were sent to Histoserv Inc. for staining. Details are provided in the SI Appendix, SI Materials and Methods.

Patient Samples Analysis.

The R2: Genomics Analysis and Visualization Platform (r2.amc.nl) was used to analyze the correlation between GPC2 mRNA expression and survival of patients with neuroblastoma. We chose the largest tumor neuroblastoma Kocak dataset (n = 649) to perform analysis of overall survival and event-free survival by the Kaplan–Meier method.

Immunoblotting and Immunoprecipitation.

Cells were harvested, vortexed in ice-cold lysis buffer (Cell Signaling Technology), and clarified by centrifugation at 10,000 × g for 10 min at 4 °C. Protein concentration was measured using a Coomassie blue assay (Pierce) following the manufacturer’s specifications. A total of 20 μg of cell lysates was loaded onto a 4–20% SDS/PAGE gel for electrophoresis. The anti-GPC2 antibody for Western blotting was purchased from Santa Cruz Biotechnology. The anti-active β-catenin and N-Myc antibodies were obtained from Millipore. The anti-Wnt3a and Wnt11 antibodies were purchased from Abcam. All other antibodies, including cleaved PARP, total β-catenin, β-actin, and GAPDH were obtained from Cell Signaling Technology.

Fifty micrograms of GPC2–hFc protein was incubated with Wnt3a CM for 4 h on ice. Protein A-Agarose beads (Roche) were added to the immune complex and incubate overnight at 4 °C. Immune complexes were washed three times with lysis buffer and subjected to immunoblotting with anti-Wnt3a antibody.

Production of Recombinant Immunotoxin.

Anti-GPC2 single-domain antibodies were cloned into pRB98 expression plasmid containing PE38. The expression and purification of recombinant immunotoxins were performed following our protocol as described previously (60).

Generation of GPC2-Specific CAR.

The anti-GPC2 CAR included the isolated anti-GPC2 single-domain antibody fragment, linked in-frame to the CD8α hinge and transmembrane regions, a 4-1BB costimulatory domain, and intracellular CD3ζ. The construct was engineered to express an upstream GFP reporter separated from the CAR by a T2A sequence. The sequence encoding the whole CAR construct was subcloned into the CMV promoter-containing lentiviral vector pLenti6.3/v5 (Invitrogen).

Lentivirus Production, T-Cell Transduction, and Expansion.

To produce viral supernatant, HEK-293T cells were cotransfected with GPC2–CAR lentiviral vectors and Mission viral packaging plasmids (Sigma-Aldrich) using Lipofectamine 2000 (Invitrogen) per the manufacturer’s protocol. The supernatant was collected at 72 h posttransfection and then concentrated using Lenti-X concentrator (Clontech) according to the manufacturer’s instructions.

PBMCs were purchased from Oklahoma Blood Institute and stimulated for 24 h with anti-CD3/anti-CD28 antibody-coated beads (Invitrogen) at a 2:1 bead-to-T cell ratio in growth medium supplemented with IL-2. Activated T cells were then transduced with the lentivirus expressing GPC2-specific CARs at a multiplicity of infection of five. Cells were counted every other day and fed with fresh growth medium every 2–3 d. Once T cells appeared to become quiescent, as determined by both decreased growth kinetics and cell size, they were used either for functional assays or cryopreserved.

T-Cell Effector Assays.

Effector T cells were cocultured with luciferase expressing neuroblastoma cells at different ratios for 20 h. At the end of the coculture incubation period, supernatant was saved for IFN-γ and TNF-α levels by ELISA (R&D Systems). The remaining tumor cells were lysed for 5 min. The luciferase activity in the lysates was measured using the Steady Glo luciferase assay system on Victor (PerkinElmer). Results were analyzed as percent killing based on luciferase activity in wells with tumor cells alone: % killing = 100 − [relative light units (RLU) from wells with effector and target cells]/(RLU from wells with target cells) × 100.

Animal Studies.

All mice were housed and treated under the protocol approved by the Institutional Animal Care and Use Committee at the NIH. For xenograft tumor studies, 5-wk-old female athymic nu/nu nude mice (NCI-Frederick) were given s.c. injections of 10 × 10⁶ LAN1 cells suspended in Matrigel (Corning). Tumor dimensions were determined using calipers, and tumor volume (in cubic millimeters) was calculated by the formula V = 1/2 ab², where a and b represent tumor length and width, respectively. For LH7–PE38 treatment, when average tumor size reached around 150 mm³, mice were i.v. injected with indicated doses every other day for 10 injections. For T-cell treatment, when tumor burden was ≈120 mm³, mice were injected i.p. with 200 mg/kg cyclophosphamide to deplete host lymphocyte compartments. After 24 h, 10 × 10⁶ of either mock T cells or LH7 CAR T cells were i.v. injected into mice on days 13, 20, and 27. Mice were given i.p. injection of 2,000 units of IL-2 twice a week following T-cell infusion. Mice were killed when the tumor size reached 1,500 mm³.

For the disseminated tumor study, 5-wk-old female athymic nu/nu nude mice were i.v. injected with 7 × 10⁶ luciferase-expressing IMR5 cells. Cyclophosphamide was injected i.p. at 200 mg/kg 24 h before any cell administration. Then animals were given single infusion of 30 × 10⁶ mock T cells or LH7 CAR T cells by tail vein injection. Disease was detected using the Xenogen IVIS Lumina (PerkinElmer). Nude mice were injected i.p. with 3 mg d-luciferin (PerkinElmer) and imaged 10 min later. Living Image software was used to analyze the bioluminescence signal flux for each mouse as photons per second per square centimeter per steradian (photons/s/cm²/sr). Mice were killed when they showed any sign of sickness or when bioluminescence signal reached 1 × 10⁹.

Statistics.

All experiments were repeated a minimum of three times to determine the reproducibility of the results. All error bars represent SEM. Statistical analysis of differences between samples was performed using the Student’s t test. A P value of <0.05 was considered statistically significant.