The most-studied tumor associated antigen in neuroblastoma is GD2. GD2 is a disialoganglioside that is highly and nearly universally expressed on neuroblastoma tissue (44) and likely plays a role in tumor immune evasion (45). It is a natural choice as a target for CAR T cell therapy in neuroblastoma based on the success of anti-GD2 monoclonal antibody therapy (6, 10–12).

One of the first CAR T cells products tested in children was a first generation anti-GD2 CAR (containing only the CD3ζ endodomain but no costimulatory domain). In preclinical models, Rossig et al. demonstrated that GD2 was a viable CAR T cell target for neuroblastoma (46). To translate the preclinical promise of anti-GD2 CAR T cells into patients, Pule et al. aimed to treat patients in a manner that could enhance CAR persistence. CAR T cells with first generation signaling domains (CD3ζ only) had previously demonstrated limited persistence in human trials for other indications, indicating that the CD3ζ only intracellular domain was not sufficient for optimal activity (47–49). Rather than endowing the CAR with embedded costimulation, Pule and colleagues generated a T cell product that could receive physiologic costimulation through engagement of a native T cell receptor (TCR).

These researchers drew on experience from clinical trials in which Epstein Barr Virus-specific cytotoxic T lymphocytes (EBV-CTLs) were adoptively transferred to patients with EBV-associated malignancies (50–53). In those trials, T cell persistence of at least 3 months was seen even with relatively low doses of EBV-CTLs. Adding tumor specificity with a CAR construct was a logical next step to take advantage of the longevity of EBV-CTLs. A Phase I trial (NCT00085930) tested this approach by infusing EBV-CTLs co-expressing a first generation anti-GD2 CAR into relapsed and refractory neuroblastoma patients who were seropositive for EBV viral capsid antigen (27).

In this trial, EBV-specific lymphocytes were extracted from eleven patients with refractory or recurrent neuroblastoma, transduced with retrovirus encoding a GD2 CAR molecule (containing the single chain variable fragment (scFv) derived from Dinutuximab, 14g2a), and stimulated ex vivo with autologous EBV-transformed lymphoblastoid cell lines (LCLs). This product was called GD2 CAR-CTL. Concurrently, bulk T cells were transduced with the same GD2 CAR but activated through the native TCR with anti-CD3 antibodies (GD2 CAR-ATC). Each patient received between 2 × 10⁷ and 1 × 10⁸ cells/m² of both GD2 CAR-CTL and GD2 CAR-ATC. A 12-base pair mutation between the receptor stop codon and the 3′ LTR allowed for comparison of in vivo durability of the two cell types by RT-PCR. There was little to no detection of GD2 CAR-ATCs after 2 weeks, but clear persistence of the EBV specific GD2 CAR-CTLs until on average 6 weeks, demonstrating that costimulation is vital for CAR T cell persistence. Four of the eight patients (50%) with evaluable tumors had a partial or complete response, though all later progressed. Responses included one patient with a complete response of an extradural parietal lesion as measured by MIBG, one patient with a complete response of extensive bone marrow disease, and two patients with significant tumor necrosis confirmed by imaging and biopsies. These data support the hypothesis that ongoing costimulation increases persistence in vivo and results in increased efficacy and durability of response. A subsequent study with longer follow up determined that even low levels of persistent cells correlated strongly with slower time to disease progression (28).

While using viral specific CTLs takes advantage of the native TCR machinery with physiologic stimulation, there is some evidence that co-engagement of a CAR and TCR can result in T cell exhaustion and decreased CAR persistence (54). Most CAR constructs now rely on embedded costimulation. The same group from Baylor produced a third generation CAR containing both the CD28 and OX40 costimulatory domains. Preclinical studies demonstrated that incorporation of tandem costimulation domains increased expansion of the engineered T cell product and augmented cytokine release (55, 56), which prompted testing this construct in clinical trials.

The third generation anti-GD2 CAR was administered to eleven patients with relapsed or refractory neuroblastoma. Patients were treated in one of three cohorts: GD2 CAR T cells alone, GD2 CAR T cells after lymphodepleting chemotherapy, or GD2 CAR T cells after lymphodepleting chemotherapy given with the PD-1 inhibitor pembrolizumab. Patients who received lymphodepletion with or without checkpoint blockade had increased expansion of their CAR T cells and longer CAR T cell persistence. Anti-PD-1 therapy did not appear to dramatically affect these parameters or efficacy. Unfortunately, even after patients received proper lymphodepletion, this CAR was found to have minimal activity with no measurable responses (43). One explanation for the lack of long-term persistence seen in this trial is tonic signaling of the CAR T cell caused by aggregation of the 14g2a anti-GD2 scFv, leading to T cell exhaustion and limited anti-tumor efficacy (57). T cell exhaustion, which will be further discussed below, has emerged as an important factor that can limit CAR efficacy and is highly dependent on costimulation molecules (57, 58).

Another Phase I trial of anti-GD2 CARs is underway in the United Kingdom (NCT02761915) utilizing an scFv based on a previously described humanized murine antibody KM8138 (59) that is fused to a CD28 costimulatory domain and CD3ζ. Based on promising preclinical data (60), this trial is enrolling children with relapsed or refractory neuroblastoma and evaluable disease in a dose escalation model. Preliminary results presented in abstract form demonstrate minor clinical response by imaging criteria and cytokine release syndrome (CRS) in at least one patient at higher dose levels, but CAR T cell persistence also appears to be limited (30). A fourth generation GD2 CAR (including CD28, 4-1BB, and CD27 costimulatory domains in addition to CD3ζ) is also being tested in a multi-institutional Chinese Phase II trial for high-risk neuroblastoma patients. An abstract presented in 2017 reported 15% of 34 patients with a partial response and no dose limiting toxicities. Two patients had significant tumor regression, one with two bulky lesions that regressed by >90% each and one with a reduction in retroperitoneal tumor dimensions and standardized uptake value (SUV) by PET scan measured 2 months after CAR T cell therapy (31).

Despite mixed results in the early GD2 CAR clinical trials, this target remains an area of intense focus. There are currently many ongoing preclinical studies focused on targeting GD2 as well as five open clinical trials of CAR T cells directed against GD2 for neuroblastoma patients (NCT03373097, NCT02761915, NCT02765243, NCT03294954, NCT02919046). While the experience thus far with GD2 CARs in clinical trials has established safety and feasibility, limited T cell persistence has emerged as a major hurdle to success.

Another target of interest in neuroblastoma is L1-CAM, an adhesion molecule that is overexpressed on neuroblastoma. Monoclonal antibody CE7 preferentially binds to a tumor-specific epitope of L1-CAM (61). The mechanism of tumor specificity has not been elucidated, but appears to be glycosylation-dependent (62–64). A first generation CAR containing the CE7 scFv, a CD4 transmembrane domain, and the CD3ζ intracellular signaling domain (CE7R CAR) demonstrated preclinical activity in xenograft models of neuroblastoma (65). A clinical construct was designed to include a selection-suicide fusion protein composed of hygromycin phosphotransferase and thymidine kinase (HyTK), allowing for CAR ablation with ganciclovir in the case of unforeseen toxicity. In a Phase I clinical trial of escalating doses of CE7R HyTK CD8+ CAR T cells, the authors demonstrated safety and observed no off-tumor, on-target toxicity. However, only one of six patients had a significant clinical response. That patient had limited disease burden, whereas the patients with higher disease burden had progressive disease. All patients ultimately died of their disease (29). Similar to GD2, lack of persistence of CAR T cells was also a major limiting factor in this study, which may have been related to the lack of costimulation in the CAR or to immunogenicity of the suicide HyTK protein (66).

To enhance the activity and persistence of L1-CAM directed CARs, the researchers then generated a second generation CAR (2G CE7 CAR) containing a 4-1BB costimulation domain and a truncated extracellular epidermal growth factor receptor (EGFRt) domain in place of the HyTK suicide switch (allowing for an alternative ablation strategy with cetuximab) (67, 68). Reassuringly, there was no significant clinical toxicity in non-human primates treated with 2G CE7 CAR T cells at doses 10–100 times higher than the doses employed in the clinical trial, though these primates did not have antigen positive malignancies (69).

A Phase I trial with the 2G CE7 CAR in rotation with a similar third generation product that also includes a CD28 endodomain is currently underway at Seattle Children's Hospital for recurrent or refractory high risk neuroblastoma patients (NCT02311621). Patients receive anti-L1-CAM CAR T cells in a defined ratio of 1:1 CD4:CD8 T cells. This strategy is based on previous successes of this controlled strategy for CAR T cell treatment of B-ALL and non-Hodgkin lymphoma at the same institution (70–72). Further study is required to determine the utility of a defined CD4:CD8 T cell product as this has not been tested in a randomized clinical trial, and equally impressive response rates have been obtained using non-selected populations of T cells or PBMCs after transduction (16, 17, 19, 20).

In a recently presented abstract, the researchers reported that L1-CAM CAR T cells infiltrate sites of disease in patients but appear to be causing off-tumor toxicity with transient skin rash (where the CAR T cells may colocalize with L1-CAM expressing normal cells) and poorly understood hyponatremia in some patients. Although these toxicities have all been transient and the trial is ongoing (42), the early finding of possible off-tumor, on-target toxicity is a reminder of the difficulty of identifying appropriate CAR-T cell targets (discussed further below).

CAR T cell persistence is essential for durable clinical responses (16, 47, 73–75). Long term follow-up of Baylor's first generation anti-GD2 CAR T cell trial demonstrated that time to disease progression was significantly delayed in patients whose T cells were detectable for longer (27, 28). In the trial of a first generation L1-CAM CAR, the only patient of six with a clinical response had detectable CAR T cells in the blood 56 days after treatment, while patients without objective response had shorter persistence (29).

CAR T cell persistence may be diminished due to T cell exhaustion. T cell exhaustion has primarily been studied in the setting of chronic antigen exposure including for viral infections (76, 77) and cancer (78–81). Exhausted T cells upregulate inhibitory receptors after excessive and continuous stimulation over a matter of days to weeks and exhibit diminished effector functions. T cell exhaustion appears to be partially reversible. This is fundamentally different from T cell senescence, which typically occurs over months to years, is associated with telomere shortening, and represents a terminally differentiated state without potential for reversibility or proliferation (82).

An exhausted CAR T cell phenotype has recently been described in GD2 CAR T cells, driven by antigen-independent tonic signaling (57). Long et al. explored why GD2 CAR T cells containing the 14g2a scFv appeared to be less functional than CD19 CAR T cells. The authors found that unlike the CD19 CAR, the GD2 CAR aggregated on the surface of T cells and subsequently triggered low level tonic signaling in the absence of antigen, which ultimately resulted in T cell exhaustion. Additionally, they demonstrated that integration of the CD28 costimulatory domain into tonically signaling CAR T cells amplified this phenotype, while inclusion of a 4-1BB costimulatory domain protected against T cell exhaustion (57). This finding is in line with clinical studies of CD19 CAR T cells, as those with 4-1BB costimulatory domains demonstrate long term persistence while those with CD28 costimulatory domains do not (16, 18, 19). Our group plans to open a clinical trial of GD2 CAR T cells with a 4-1BB costimulatory domain in early 2019, which will be the first such trial in North America.

Persistence can be affected by factors extrinsic to the CAR molecule. Early CAR T cell trials did not incorporate lymphodepletion prior to CAR T cell infusion, which may have compromised expansion of the engineered T cells (18, 43, 71). Lymphodepleting chemotherapy improves engraftment and efficacy and has become a standard part of CAR T cell regimens (83, 84). The mechanism of increased activity after lymphodepletion is thought to be depletion of regulatory immune cells and/or a reflexive increase in homeostatic cytokines IL-7 and IL-15 that drive CAR T cell proliferation (84–86). Given that endogenous cytokines may increase CAR efficacy, some groups have focused on increasing CAR potency by programming CAR T cells to secrete immunostimulatory cytokines locally (87, 88). Systemic infusion of cytokines is often associated with unacceptable toxicity (89–91), and overexpression of the cytokine receptor does not overcome a dearth of cytokines in the tumor microenvironment (92). Therefore, providing local and inducible cytokine release by the CAR T cells themselves is an attractive strategy. Initial reports have demonstrated improved potency of CD19 CAR when co-expressed with IL-7 (93), IL-12 (94), IL-15 (95), membrane bound chimeric IL-15 (88), and IL-21 (93). Further studies will be required to translate these results clinically and to see if this can be generalized to solid tumors and to neuroblastoma specifically.

Anti-carcinoembryonic antigen (CEA) CAR T cells were engineered to produce IL-12 only after engagement with target antigen by placing IL-12 under the control of a nuclear factor of activated T cells (NFAT) promoter. In a colon cancer model, CEA CAR T cells that expressed inducible IL-12 mediated greater tumor regression and abrogation of antigen negative tumor outgrowth. This effect was likely enhanced by activated macrophages that infiltrated the tumors in response to the locally secreted IL-12. (96). An alternative system combines oncolytic viruses that secrete cytokines IL-15 and CCL5 with anti-GD2 CAR T cell therapy in xenograft models of neuroblastoma in order to increase T cell infiltration and persistence (97).

Shum et al. recently described a system in which a constitutively active IL-7 receptor was co-expressed with a second generation GD2 CAR. This resulted in improved efficacy of GD2 CAR T cells in vitro and in a murine xenograft model of neuroblastoma (98). This modification did not lead to malignant transformation in short term assays, an important safety consideration as the IL-7 receptor was derived from a patient with T cell acute lymphoblastic leukemia (99). However, implementation of such a strategy into clinical trials will require caution due to the potential for delayed malignant transformation. These approaches to increase potency and persistence of CAR T cells are beginning to undergo testing in early clinical trials (NCT03635632), and may help to improve efficacy, durability, and ultimately clinical outcomes.

Choosing an optimal CAR T cell target in neuroblastoma and more generally in solid tumors is a daunting task. Much of the success of CD19 and CD22 CAR T cells hinges on the restriction of these targets to lymphoblasts and normal B cells, which are in large part dispensable with appropriate supportive measures (16–20). An ideal CAR target antigen is highly and homogeneously expressed on tumor cells with minimal expression on vital tissues. Fulfillment of these criteria is difficult for solid tumor antigens, as many antigens are expressed in cells of related origin.

Many antigens overexpressed on neuroblastoma are often present at lower levels in peripheral nerves and/or on other neural tissue (100–102), so an important consideration in the development of anti-GD2 CAR T cells is the potential for off-tumor, on-target toxicity. Anti-GD2 monoclonal antibodies cause pain requiring continuous infusion of narcotics for analgesia (103–106) due to their interaction with peripheral nerves and possibly engagement of the complement system (107). However, clinical trials of CAR T cells targeting GD2 have not resulted in toxicity despite clear signs of on-tumor efficacy (27, 28, 30, 43). Still, due to toxicity concerns, most anti-GD2 CAR T cells for clinical trials have been designed to include a “suicide switch” to allow for rapid ablation.

Oncologists have approached adoptive T cell therapy for solid tumors cautiously due to the overlap of antigen expression with normal tissues. There have been several incidents of off-tumor toxicity in human trials using engineered high affinity TCRs against MAGE-A3 (108) and MAGE-A12 (109) that cross reacted with normal tissue. Additionally, one patient with metastatic colon cancer died after treatment with a HER2-targeted CAR (110). The initial case report of this incident noted that there was pulmonary infiltration by CAR T cells that could be due off-tumor, on-target toxicity. However, that patient was administered a dose of CAR T cells that was found to be 100 times the maximum tolerated dose of CD19 CAR T cells as well as exogenous IL-2. She was found to have very high levels of circulating cytokines and our recent understanding of the toxicities associated with CAR T cells indicates that this was more likely to be caused by CRS than off-tumor, on-target toxicity (111). This is further supported by recent efforts at Baylor College of Medicine to target HER2 on pediatric sarcomas using CAR T cells. In carefully designed dose escalation trials conducted without and with lymphodepletion, anti-HER2 CAR T cells elicited no off-tumor, on-target toxicity but resulted in clinically significant responses including a complete response in a patient with metastatic rhabdomyosarcoma (112, 113).

One possible explanation for the lack of toxicity for both GD2 and HER2 CARs is the differential in antigen density between tumor cells and normal tissue. Antigen density is emerging as an important consideration for CAR efficacy. When our group engineered a CAR against Anaplastic Lymphoma Kinase (ALK) on neuroblastoma, there was a clear correlation between the number of surface molecules of target antigen and ALK CAR T cell efficacy. A threshold number of target molecules was required to elicit effector functions (114). In vivo, ALK CAR T cell efficacy was only seen when ALK expression was high on tumor cells. Similarly, in a Phase I trial of CD22 CAR T cells of children with ALL, after initially achieving a complete response, most patients relapsed with leukemia expressing lower levels of CD22 than their pre-treatment samples, apparently below the threshold for CAR efficacy (20). Others have found a similar relationship of CAR efficacy and antigen density in preclinical studies of CARs for targets including CD123, CD20, HER2, EGFR, and CD30 (115–121). This represents a paradigm shift in the field as it opens up potential therapeutic windows for targets expressed at low levels on normal tissue as long as expression on tumor is high (111).

As CARs are engineered to become more potent, they could also become more toxic due to recognition of lower levels of target. While several clinical trials of GD2 CAR T cells containing the 14g2a binder have been carried out without any reports of central or peripheral neurotoxicity (27, 28, 30, 43), one preclinical study of a high affinity GD2 CAR reported neurotoxicity and T cell infiltration in the brains of mice (122). However, studies of a CAR with the same high affinity binder in our laboratory do not cause neurotoxicity (123), calling into question whether the findings were truly due to off-tumor, on-target toxicity. The point remains, however, that as CAR T cells are better engineered to target low target antigen density tumor cells, there will be potential for increased toxicity and clinical trials must be conducted carefully.

In addition to GD2 and L1-CAM, researchers are investigating several novel target antigens for CAR T cell therapy in neuroblastoma, and preclinical data are summarized below. Figure 1 depicts the targets currently under investigation for CAR T cell therapy for neuroblastoma.

NK cells have long been recognized as important in neuroblastoma and killer cell immunoglobulin-like receptors (KIR) haplotypes are strongly correlated to survival (152–154). NK cells lack the specificity of T cells, but they have the capacity to kill infected and malignant cells without the prerequisite priming and sensitization to peptide-MHC complexes on the target cell surface. Instead, NK activity is regulated by a balance of activating and inhibitory receptors (155). Several trials are underway in which neuroblastoma patients receive adoptively transferred ex vivo expanded but unmanipulated NK cells (NCT02573896, NCT01857934, NCT02650648, NCT03209869).

Given their importance in control of neuroblastoma, researchers have attempted to augment the anti-tumor effects of NK cells in by imparting them with tumor antigen specific CARs. One group generated patient-derived NK cells expressing a second generation GD2-specific CAR, and demonstrated significant improvement in cytotoxicity against primary patient neuroblastoma cells compared to NK cells without a CAR (156). Similarly, expressing the GD2-CAR in an NK-92 cell line promoted in vitro cytotoxicity against neuroblastoma cell lines that were resistant to killing by the parental NK-92 cell line (157). NK cells do not have the same proliferative capacity as T cells, and clinical trials of adoptively transferred NK cells are often marked by short persistence and disappointing anti-tumor effect (158). The persistence of NK cells and invariant NK T cells can be increased by constitutive secretion of IL-15, an approach being studied in clinical trials for children with neuroblastoma at Baylor College of Medicine (NCT03294954) (159, 160).

In contrast to standard chemotherapy or “off the shelf” immunotherapies such as monoclonal antibodies, an important consideration for CAR T cell therapy is the ability to manufacture adequate quantities of a viable, maximally efficacious T cell product. Some patients have poor expansion and inadequate production of CAR T cells. One group hypothesized that myeloid derived suppressor cells (MDSC) in the apheresis product may interfere with T cell expansion, and found higher proportions of monocytes in PBMC concentrates to inversely correlate with fold expansion of CD19 and GD2 CAR T cells (161). CAR T cell quality is of particular concern for patients who have undergone chemotherapy, radiation, and/or stem cell transplant, all important elements of upfront neuroblastoma therapy. Data presented in abstract form describe T cell fitness in PBMC samples collected at diagnosis and after each cycle of chemotherapy from children with a wide variety of cancers including neuroblastoma. These data suggest that after chemotherapy, patients develop poor CAR T cell potential, defined by a low proportion of naïve T cells, mitochondrial dysfunction, and poor spare respiratory capacity (162). Further study is warranted to understand this phenomenon and whether it ultimately impacts CAR T cell efficacy in patients, as highly active CD19 CAR T cells have been successfully generated from most patients with heavily pretreated ALL (18, 19).

The immunosuppressive tumor microenvironment (TME) presents a significant barrier to successful CAR T cell therapy for neuroblastoma. Neuroblastoma tumors are intermixed with a suppressive cell population that includes tumor associated macrophages (TAMs) and regulatory T cells (Tregs). Presence of these cells predicts poor outcomes (34, 163). Tumors also express inhibitory ligands such as PD-L1 that dampen T cell responses (164–167). Furthermore, the TME contains an array of soluble factors such as TGF-β and IL-10 that act to directly inhibit T cells (34, 168–171). Finally, physical barriers such as stroma, extracellular matrix (ECM) and tumor associated vasculature prevent tumor infiltrating T cells from easily accessing their target (172–175).

RM has a pending patent application for the use of GD2 CAR T cells in H3K27M mutant gliomas. The remaining authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.