The Future Is Now: Chimeric Antigen Receptors as New Targeted Therapies for Childhood Cancer

Cell therapy for cancer is crossing the threshold of clinical activity, as trials in both adult and pediatric oncology have now shown unquestionable clinical responses (1–7). Historically, cell-based therapies have focused on cytolytic T cells targeting MHC-restricted antigens. Although this approach remains promising, its efficacy and applicability are limited by the need to enhance the affinity of naturally occurring T-cell receptors (TCR) that target most tumor antigens, and the tendency for tumors to downregulate MHC molecules. Furthermore, targeting MHC-restricted antigens is particularly challenging for pediatric cancer and other rare tumors because immunodominant epitopes have not been defined for most MHC alleles, and the difficulty of targeting a rare tumor in a limited population that expresses a particular MHC allele severely affects feasibility and applicability. For these reasons, chimeric antigen receptors (CAR; Fig. 1), which harness the potent cytolytic capacity of expanded effector populations with MHC-unrestricted targeting, are receiving increasing attention and represent a promising new therapeutic modality for childhood cancer. Below, we discuss the progress that has been made in this young field, as well as the challenges that remain to be overcome.

CARs (historically referred to as T-bodies) were first generated in 1989 by Eshhar and colleagues as fusions of antibody and TCR subunits that, when expressed on T cells, mediated MHC-independent T-cell activation (8–10). Although early results provided proof-of-principle, T-cell activation was not robust, and therefore CAR technologies were gradually optimized via modification and addition of subunits to increase signaling potency. The structure has evolved over the last 2 decades to most commonly incorporate a single-chain variable fragment (scFv) derived from a monoclonal antibody (mAb) plus the signaling motif from the TCR ζ chain [referred to as a first-generation CAR (Fig. 1; ref. 11)]. A major advance was made when 1 (second-generation CAR) or 2 (third-generation CAR) costimulatory activating motifs from CD28, 4-1BB (CD137), and/or CD134 (OX-40) were incorporated into CARs, leading to enhanced proliferation, cytotoxicity, and persistence in vivo (Fig. 1; refs. 3, 12–15). Beyond the general principle that CARs that incorporate costimulatory signals are more potent, it remains unclear whether any particular costimulatory molecule is superior to another, and to what extent the beneficial effects of any or all costimulators can be imparted by signaling during the expansion process without incorporating the molecule into the CAR itself (15–18). Of note, incorporation of the 4-1BB signaling motif into a CAR was recently reported to result in more robust proliferation in vitro compared with CD28 (15). These 2 costimulatory domains have not been compared in controlled clinical trials as yet; however, 2 of 3 patients with refractory chronic lymphocytic leukemia (CLL) treated with an anti-CD19 CAR that used 4-1BB as its costimulatory signal achieved durable (1 year) complete remissions, and the third achieved a partial remission (3, 19).

Multiple methods can be used to introduce CARs into effector cells. The most common approach uses γ retroviruses, which efficiently and stably integrate the receptor sequence into the target cell genome (6, 20, 21). Thus far, this approach has proved to be safe for transduction of mature T cells, although concerns remain that insertional mutagenesis or immune responses against retroviral antigens could occur (22, 23). Similarly, lentiviral transduction of the CAR sequence into expanding T cells has also been shown to be a safe and successful approach (3, 15). Nonviral methods, such as the use of transposon-based systems and direct RNA transfection (24–27), are also being explored. These methods are likely to be less costly to develop (25), and RNA transfection has the potential for self-limited expression in cases where permanent expression may be undesirable. In cases where permanent expression systems are used, some investigators have incorporated suicide genes, such as the herpes simplex thymidine kinase (TK) gene, or an inducible caspase 9 protein that can be activated by specific drugs and eradicate the genetically engineered cells if adverse effects occur following adoptive transfer (28–30).

CAR specificity is most commonly endowed by an scFv derived from phage display or from mAbs raised against cell-surface antigens, although in some cases receptor ligands have been used (31, 32). It is generally assumed that a high-affinity antigen-binding domain that targets a plentiful tumor cell-surface antigen with limited expression on normal tissue is desirable; however, the precise and relative importance of affinity and target expression levels have not been defined, and very few antigens show exclusive expression on tumors. The risk–benefit ratio of any particular CAR-expressing T lymphocyte is driven largely by properties of the target, because targets with vital organ expression will induce off-tumor, on-target effects. Depending on the tissues in which the target is expressed, this could result in minimal or unacceptable toxicity. B-cell malignancies provide several attractive CAR targets because mAbs are available to target several B-cell surface antigens, and the mature B-cell compartment is considered to be relatively expendable, at least temporarily, in patients with hematologic malignancies. Thus, CARs targeting CD20 (33, 34), CD19 (20), and CD22 (35) have been developed, and many trials for pediatric B-cell precursor acute lymphoblastic leukemia (pre-B ALL) are under way or anticipated (Table 1). CARs that target CD30 (18, 36) and ROR1 for lymphomas (37) are under study as well and may also be applicable to pediatric hematologic malignancies.

Choosing CAR targets for solid tumors is potentially more challenging because solid-tumor–associated antigens often coexist on nonexpendable tissues. GD2 is a validated tumor antigen for which mAb-based targeting has proved to be safe (38). Efforts to target GD2 with mAbs are reviewed by Matthay and colleagues in this CCR Focus section (39). Not surprisingly, therefore, one of the first trials of CAR therapy used a GD2-CAR administered to children with advanced neuroblastoma. In that study, 3 of 11 patients with active disease experienced complete responses, no substantial toxicity was observed, and sustained clinical benefit was reported for several patients with long-term follow-up (1, 2). Of note, the CARs used in this trial were of the first-generation variety and incorporated only CD3 ζ chain signaling motifs. Current approaches that incorporate costimulatory signaling motifs may be even more active. In contrast to the favorable safety profile seen when targeting GD2, CAR therapy used to target Her2/Neu resulted in a patient death that was attributed to Her2/Neu expression on normal lung and/or cardiac tissue (40). It is important to note, however, that substantially higher numbers of CAR-transduced cells were administered in this case than were administered in many other trials, raising the prospect that lower doses of CARs targeting this antigen might also be safe. Indeed, a cautious dose-escalation study is under way in pediatric patients with osteosarcoma to determine whether Her2/Neu-CAR T cells have acceptable risk–benefit ratios (http://clinicaltrials.gov/ct2/show/NCT00902044). Nonetheless, this clinical experience illustrates that antigens that are safely targeted by antibody therapy cannot necessarily be assumed to be safe when targeted with CARs.

A full list of potential targets for CAR-based therapy of pediatric solid tumors is beyond the scope of this report, but this is an area of very active investigation (for review, see ref. 41). Targets that are currently under study include epidermal growth factor receptor (EGFR) vIII (42) and interleukin (IL)-13 receptor α2 (32, 43), both of which are expressed on a substantial number of pediatric gliomas. B7H3 is expressed by many sarcomas and pediatric brain tumors (44), and fibroblast growth receptor 2 (FGFR2) is highly expressed by most rhabdomyosarcomas (45). Substantially more work is needed to prioritize targets for testing in early clinical trials of CAR therapy in pediatric oncology, and this will be based on preclinical studies of killing activity and the relative risk of off-target effects.

Several groups have developed clinical-grade cell-manufacturing systems that are capable of producing large numbers of CAR-modified T cells. There is wide variation in the specific processes used, and to date there is no clear evidence that one is superior to another, although this remains an area of active investigation. One of the most commonly used methods employs anti-CD3 plus autologous or allogeneic feeder cells and high-dose IL-2 (Fig. 2A, I; refs. 46–49). Alternatively, artificial antigen-presenting cells (aAPC) that provide anti-CD3 plus costimulatory signals have been developed in an effort to replicate the physiologic process of T-cell activation by dendritic cells. Cell-based aAPCs, most often K562 cells modified to express costimulatory ligands (4-1BBL, CD80) and/or cytokines (IL-15 and IL-21), induce potent expansion of both natural killer cells and/or T cells (Fig. 2A, II; refs. 50–53). Bead-based aAPCs (Fig. 2A, III) employing anti-CD3 and anti-CD28 antibodies coupled to paramagnetic beads induce robust T-cell expansion without the need for exogenous cytokines (54, 55). All of these methods are sufficient to generate a several-hundred-fold expansion of cells over a 10- to 14-day period (Fig. 2B). In addition, investigators have developed approaches to expand donor or patient Epstein–Barr virus (EBV)- and cytomegalovirus (CMV)-specific T cells using viral antigens that engage endogenous TCRs and costimulatory pathways before and after CAR transduction (1, 18, 56). Such virus-specific T cells could prove useful for administering CAR T cells following allogeneic hematopoietic stem cell transplant because they are not predicted to induce graft-versus-host disease. In addition, the investigators hypothesize that exposure to intermittent viral reactivation in vivo will contribute to persistence following transfer. However, the production of CAR-modified, virus-specific cells generally requires a longer culture period compared with the polyclonal techniques described above (18).

Cells that have been expanded in each of the above systems are amenable to transduction, secrete cytokines, and lyse antigen-expressing cells (57, 58); however, emerging studies suggest that the nature of the starting cell population that is genetically modified and/or the method of expansion may affect the likelihood that CAR T cells will persist in vivo and effectively mediate antitumor effects. In models that emphasized generation of larger doses of CAR T cells at the expense of driving them toward short-lived T-effector cells (Fig. 2C), recent studies of adoptive immunotherapy correlated clinical benefit with in vivo cell expansion and persistence (2, 59–61). In a recent clinical study, administration of a very low number of CAR-specific T cells, calculated to yield an effector:target ratio of 1 CAR-transduced T cell to 93,000 CLL cells, yielded a dramatic clinical benefit (1, 3). This finding and the observation that administration of small numbers of T cells can mediate dramatic effects against large tumor burdens (62) emphasize the notion that cells that are capable of robust proliferation can overcome daunting tumor burdens.

Data gleaned from a variety of murine and human studies of adoptive T-cell immunotherapy show that cells bearing a terminally differentiated T-effector memory (T_EM) phenotype, which tend to dominate prolonged ex vivo cultures and systems using anti-CD3 plus IL-2, have a limited capacity to expand and persist in vivo following an encounter with the tumor (63). In contrast, T-central memory (T_CM) cells are more potent in several models (64), and it is predicted that this subset may be more effective in the setting of CAR-transduced T cells as well. Some studies showed enhanced efficacy when T_CM cells were used as a starting population for transduction (65), whereas in other studies investigators focused on expanding T_CM cells using APC-based systems (66). Surprisingly, in some cases, T_CM cells emerged in vivo following administration of a mixed population of predominantly T_EM cells (64). Virus-specific T cells showed enhanced persistence compared with nonviral CAR T cells, perhaps due to an increased frequency of T_CM in virus-specific populations (1). Recently, a new subset termed T stem cell memory (T_SCM) performed better than T_CM and T_EM in a preclinical model of CARs directed against the tumor antigen mesothelin (67). Studies are under way to determine optimal approaches for generating or maintaining T_SCM cells during expansion, and several agents, including rapamycin, IL-21 (26, 68), and Wnt signaling modulators (69), appear to be capable of limiting terminal differentiation and/or enhancing T_SCM cells. Finally, recent data suggest that the efficacy of CAR T-cell products correlates with the number of CD4⁺ T cells infused (2), and historical data provide evidence that CD4⁺ T cells themselves support the persistence of CD8⁺ T cells (70). Of interest, bead-based aAPCs preferentially expand CD4⁺ T cells compared with most other approaches (26, 66). Although CD4⁺ regulatory T cells (Treg) theoretically could be transduced with a CAR or develop from a non-Treg that was CAR transduced (Fig. 2A), culture conditions do not seem to select for this subset, and there is no evidence thus far that CAR-transduced Tregs mediate clinically significant immunosuppression.

Given the multitude of cell subsets that can be targeted with CAR therapy, and the vast array of approaches that can be used to transduce and expand CAR T cells, it would be highly desirable to obtain accurate preclinical models to predict clinical efficacy. Unfortunately, for most CARs, such studies are limited to xenograft models that incorporate adoptive transfer of human tumors and human T cells into immunodeficient mice. Although such models provide some insights into clinical activity, they may not provide an accurate assessment of persistence in humans, and provide little insight into toxicity, including CAR T-cell–induced autoimmunity. Fully murine models may be preferable but are not available for most CARs currently under study in humans. Thus, optimizing preclinical models to prioritize targets and cell-manufacturing methods represents an important area for future efforts.

Early studies of adoptive immunotherapy focused on generating large quantities of antigen-specific effectors to overcome the perceived limitations posed by overwhelming tumor cell numbers compared with available effectors. In contrast, current efforts are more focused on administering cells with robust proliferative capacity, and modulating the host immune milieu to support in vivo expansion and persistence. Lymphopenia induces profound changes in T-cell physiology, due primarily to accumulation of IL-7 and IL-15, homeostatic cytokines that support T-cell expansion and survival (71–73). As a result, most adoptive cell therapy protocols incorporate lymphotoxic therapies prior to cell transfer to increase the availability of such homeostatic cytokines. The results of nonrandomized trials support this approach (7, 74), but no definitive data are available to confirm that it is necessary.

In pediatric oncology, most candidates for adoptive immunotherapy are already lymphopenic due to previous exposure to cytotoxic regimens. GD2-CAR therapy administered without a lymphodepleting regimen showed antitumor activity in patients with neuroblastoma (1, 2), although it remains unclear to what extent these patients were lymphopenic at the time cell therapy was initiated. Lymphodepleting regimens may also enhance the efficacy of adoptive cell therapies by transiently reducing Treg numbers (75), and by inducing mucosal damage that results in systemic exposure to lipopolysaccharide and other bacterial byproducts, which activate the innate immune system (Fig. 3; refs. 76, 77). Preclinical models show that administration of homeostatic cytokines combined with targeted therapies to induce Treg depletion is more effective than nonspecific cytotoxic regimens in supporting adoptive immunotherapy (78). Therefore, efforts are under way to induce or administer the essential host factors that will support adoptively transferred cells while limiting the toxicity of the regimen. This is particularly relevant for the field of pediatric oncology, where the desire to limit late effects is a major factor driving the development of targeted therapies such as CAR-based regimens.

At the present time, because of the complexity and cost of generating CAR-based therapies, only a few specialized centers are capable of delivering such products. Further, the approach is currently supported entirely by academia, government, and private sources; to date, no biotechnology or pharmaceutical company has sponsored a CAR trial. Despite these limitations, however, several clinical trials have already been completed and many others are currently under way or planned, including some in pediatrics. Several key elements of successful CAR therapy have been elucidated in these trials. First, a proof-of-principle that CAR therapy can mediate a clinical benefit against childhood cancers was provided in the first trial undertaken in pediatrics. Most trials that used CARs lacking costimulatory motifs were unsuccessful due to poor persistence and/or expansion of the transduced T cells; however, a clinical trial of anti-GD2 CAR T cells in 19 patients with neuroblastoma who received a mixture of autologous, CAR-modified EBV-cytotoxic T lymphocytes (CTL) and anti-CD3–activated polyclonal CAR-modified T cells proved to be an exception (1, 2). These patients did not receive a preparative regimen, although they may have already been lymphopenic due to extensive prior therapy. Three of 11 evaluable patients had a complete response, and improved clinical outcomes were observed in patients in whom CAR T cells could be detected beyond 6 weeks. In this trial, improved clinical outcomes also correlated with higher levels of CD4⁺ cells and higher levels of central memory cells in the infused adoptive cell product (2). Of interest, although pain associated with anti-GD2 mAb therapy is commonly seen, no significant GD2-CAR–associated pain was observed in this trial, raising the prospect that in some cases, CAR-based therapy could be less toxic than mAb therapy.

Clinical trials in adults have provided evidence that CD19 is a safe and effective target for CAR therapy of B-cell malignancies (3, 4, 19). In these studies, patients who experienced clinically meaningful responses with lysis of large tumor burdens and maintenance of complete responses showed massive expansion of the infused cells, and persistence of cells for at least 6 to 12 months. Thus, clinical activity has been observed in several CAR clinical trials, including those targeting CD19 and GD2, both of which are antigens expressed on pediatric tumors. Several groups have shown that incorporation of a costimulatory domain(s) into a CAR leads to increased persistence (17, 79, 80), and results from nonrandomized studies suggest that a lymphodepleting regimen improves outcomes (60, 77, 80). Therefore, most studies that are currently planned or under way in pediatrics will incorporate costimulatory elements in the CAR design and use a lymphodepleting regimen prior to adoptive transfer.

These early successes with CAR therapy, however, were not achieved without substantial toxicities. Flu-like symptoms such as fever, malaise, and myalgias are common after CAR T-cell infusion and likely relate to cytokine release by the infused cells, because they coincide with increases in IFN-γ (19). As discussed earlier, one adult patient treated with 10¹⁰ highly activated Her2/Neu-CAR T cells developed fatal immune activation syndrome resulting from low-level target expression in the lungs shortly after CAR T-cell infusion (40, 81). A second patient who received cyclophosphamide conditioning followed by 3 × 10⁷/kg anti-CD19 CAR T cells developed exacerbation of preexisting immune activation, presumably from an underlying infection, 6 hours after cell infusion and ultimately died (81, 82). As a result of these toxicities, recent trials have decreased the number of infused CAR cells in the initial cohorts (generally starting at 10⁶/kg), and incorporated close monitoring of clinical symptoms and serum cytokine levels. Other observed toxicities include autoimmunity mediated by on-target, off-tissue effects (e.g., hepatic toxicity) resulting from therapy with carbonic anhydrase IX-CAR, which has since been shown to be expressed on cells in the biliary tract (23). Furthermore, anti-CAR immune responses (23) may adversely affect the efficacy of the therapy itself.

Dramatic progress in our understanding of genomics (83, 84), cancer biology (85), and immunology (39) is fueling exciting new progress in the search for more effective and less toxic targeted therapies for childhood cancer (86). Among the most novel and promising of these approaches are CAR-based cell therapies that combine advances in genetic engineering and adoptive immunotherapy. Current research focused on optimizing this approach emphasizes (i) effective tumor targeting with limited off-tumor toxicity, (ii) optimized cell manufacturing to improve efficacy and render the therapies more exportable, and (iii) modulation of the host and/or cell product to increase in vivo expansion and/or persistence. In pediatric cancers, several candidate antigens have already been identified that provide acceptable tumor targeting, but much more work is needed in this arena, especially for pediatric solid tumors. Cell-surface antigens with robust tumor expression and limited normal tissue expression that could be effectively and safely targeted with this therapy must be systematically identified and prioritized. With regard to optimizing cell manufacturing, it is clear that many different approaches can reliably generate large numbers of CAR-expressing T cells. However, better preclinical models and surrogates for bioactivity are needed to compare manufacturing approaches, and carefully controlled clinical trials are needed to directly test the various CAR constructs and expansion techniques that are currently available. Finally, optimal modulation of host factors is also likely a key determinant of in vivo expansion and persistence, and recent insights suggest that lymphopenic hosts that receive cells bearing T_CM or T_SCM phenotypes experience optimal antitumor effects. Given the limitation in patient numbers that challenges early-phase pediatric trials, we anticipate that seminal insights in each of these arenas will be gleaned from clinical trials that are under way in adult oncology, or in trials that enroll both adult and pediatric patients. Ultimately, however, careful clinical trials in children will be needed to assess efficacy and oxicity. If favorable antitumor effects are observed, we anticipate that progress in manufacturing techniques will lead to reductions in the cost and complexity of generating these therapies, allowing them to be exported and/or made available to institutions where such approaches are currently not available.

We envision that CAR-based therapies will ultimately comprise multimodal regimens, likely incorporating host preparative regimens, novel elements in the cell expansion cocktail, and potentially cytokines to maximize in vivo expansion. Although most immunotherapies seem to be most active in the setting of minimal residual disease, impressive antitumor effects observed thus far in a small number of patients with bulky disease raise the prospect that CAR-based therapies may have activity in the setting of large tumor burdens. For studies undertaken in the setting of minimal residual disease, novel clinical trial endpoints are needed to determine whether results from early-phase trials warrant larger randomized trials with survival endpoints. Careful clinical trials will be needed to assess the optimal timing for incorporating CAR-based adoptive immunotherapies, and to compare CAR-based targeting of cell-surface antigens with mAb-based immunotherapies. In summary, CARs represent a novel and promising approach for targeted therapy of childhood cancer. Progress in this arena will be largely driven by academia and will require support for expensive early-phase clinical trials that promise to pave the way for a new form of targeted, exportable immunotherapy for children with cancer.

No potential conflicts of interest were disclosed.

Conception and design: D.W. Lee, D.M. Barrett, C.L. Mackall, R.J. Orentas, S.A. Grupp

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): D.W. Lee, R.J. Orentas, S.A. Grupp

Writing, review, and/or revision of the manuscript: D.W. Lee, D.M. Barrett, C.L. Mackall, R.J. Orentas, S.A. Grupp,

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): D.W. Lee, R.J. Orentas