Chimeric Antigen Receptor– and TCR-Modified T Cells Enter Main Street and Wall Street

Presently there are three types of adoptive cell transfer (ACT) using effector T cells that are advancing on a path toward regulatory approval (Fig. 1). Tumor-infiltrating lymphocytes (TILs) have been developed with slow but continuing progress during several decades. Recently, an international phase III randomized trial has begun for patients with metastatic melanoma. Lion Biotechnologies has been formed to commercialize TIL therapies for melanoma and other tumors that have suitable T cell infiltration.

In contrast to TILs, gene transfer–based strategies have been developed to overcome the consequences of immune tolerance on the tumor-specific T cell repertoire. These approaches provide the potential to efficiently redirect T cells to tissues by transferring chimeric Ag receptors (CARs) composed of Ab-binding domains fused to T cell signaling domains, or by transferring cells expressing TCR α/β heterodimers. The infusion of gene-modified T cells directed to specific targets offers the possibility to endow the immune system with reactivities that are not naturally present. This approach has the additional benefit of rapid tumor eradication that is usually seen with cytotoxic chemotherapy or with targeted therapies, and it contrasts to the delayed effects that are usually observed with vaccines and T cell checkpoint therapies.

Cell therapies are ultimately personalized in that, with rare exceptions, they are comprised of autologous, patient-derived T cells. For this reason, ACT is primarily being developed based on an unprecedented reliance on academic and pharmaceutical industry partnerships. In this model, academia and industry are coexisting, with the former developing and testing new ideas regarding cellular engineering and the latter scaling to achieve global impact on health care. Such academic and industrial partnerships have recently emerged at numerous institutions worldwide, including the University of Pennsylvania with Novartis, Baylor College of Medicine with Bluebird Bio and Celgene, the Memorial Sloan Kettering Cancer Center and the Fred Hutchinson Cancer Research Center with Juno Therapeutics, the National Cancer Institute with Kite Pharma, and the Cellular Biomedicine Group with the Chinese PLA General Hospital. Overall, there are now dozens of companies in the cell therapy field representing billions of dollars in investments (1). The influence of these partnerships remains uncertain, as the merger of academic intellectual freedom with the big business focus on value will surely create conflict. Pursuit of extramural grant funding and the rights to intellectual property will be intense topics of conversation between academic investigators, who created this field, and the pharmaceutical companies that seek to license the science.

It is interesting to note from a historical perspective that some of the first forms of ACT involving gene-modified T cells were conducted almost two decades ago in patients with advanced HIV-1/AIDS (2), and that many of the results from trials conducted in HIV-1–infected patients have informed current concepts in the field of cancer, as exemplified by the demonstration that CAR T cells could survive for more than a decade in HIV-1/AIDS patients (3). These initial trials were done to control drug-resistant forms of HIV-1 infection. However, the current challenge in the field is to develop cellular therapies with the potential to eliminate the reservoir of HIV-1 that is resistant to current antiviral therapies (4). The field has been energized by an extraordinary experiment conducted by Gero Hütter et al. (5) in Berlin in a patient who has apparently been cured of HIV-1 infection following an allogeneic hematopoietic stem cell transplant and ACT from an HIV-1⁻ homozygous CCR5 Delta32 donor. There are a number of approaches to induce a cell-intrinsic resistance to HIV-1 infection and to target the reservoir of HIV-1 by gene-modified ACT (6). Recent studies suggest that ACT with appropriately targeted CD8⁺ CTLs can clear HIV-1 latent reservoirs in humanized mice (7), providing additional rationale for the application of ACT using genetically modified T cells for the therapy of HIV-1 infection and other chronic infections that often fail to be controlled by the endogenous immune system. It is likely that the incremental improvements in the construction of mice with humanized immune systems will greatly accelerate the development of successful ACT for HIV-1 infection.

There is increasing interest in harnessing regulatory T cells (Tregs) to downregulate unwanted immune responses. The efficacy of this approach has been shown in preclinical models of autoimmune disease and allograft rejection, as the adoptive transfer of Tregs can prevent lethal graft-versus-host disease and autoimmune diabetes in mice (8, 9). The adoptive transfer of expanded cord blood Tregs has been shown to be safe and feasible in patients undergoing allogeneic hematopoietic stem cell transplantation (10). Distinct cell culture methodologies have been developed to optimize the expansion and function of natural and induced Tregs (11, 12). The need for differential cell culture approaches is likely due to differences in signal transduction between Tregs and effector T cells (13). The principles of synthetic engineering have been applied to Tregs; for example, CAR Tregs show promise in preclinical models of autoimmunity (14, 15). The concept of using a CAR in Tregs is centered around directing Tregs to a vulnerable tissue via the CAR, where their engagement would result in suppression of autoimmunity. Specifically, modifying mouse Tregs with a CAR targeting myelin basic protein–specific T cells protected against autoimmune encephalitis, and the concept is being pursued for colitis- and diabetes-targeting Ags in the colon and pancreatic islet, respectively (16–18).

Animal models have played a key role in developing cellular therapies. There are two major variations used by investigators: the syngeneic mouse model using mouse T cells and mouse Ags, and the xenograft model using immunodeficient mice and human T cells and tumor cells (19, 20). During the decades of development of cellular therapies, these models have produced often contradictory and conflicting conclusions (20, 21). The considerable number of variables involved in producing a cellular therapy is one of the factors behind such results. The structure of the CAR, the cellular composition of the T cells (CD4/CD8 or both), the choice of costimulatory domain, the method of gene transfer, the method and time of ex vivo expansion, the use or not of lymphodepleted hosts, and the amount and type of tumor burden must all be controlled to draw accurate conclusions. As one example, the use of the CD28 costimulatory domain in the CAR structure was found to be beneficial in a xenograft model that did not use CD28 stimulation in the ex vivo expansion (22).

The syngeneic mouse model has the inherent benefit of modeling an intact immune system, whereas the xenograft system has the advantage of allowing the study of human cells. Xenograft systems have accurately predicted efficacy in several reports studying chimeric Ag receptors (23–25). These systems failed to predict any of the toxicity, specifically the cytokine release syndrome (CRS) described in pediatric acute lymphoblastic leukemia (ALL) treated with anti-CD19 CAR T cells (see below) (26, 27). Syngeneic mouse models have not yet mimicked the CRS, although they have predicted some efficacy (28–30). In part, this is due to subtle yet important differences in the ability to ex vivo stimulate and expand T cells of human or murine origin. As clinical trials are obviously focused on human T cells, the understanding of influences of human T cell costimulation are of paramount importance and support the utility of the xenograft model. More sophisticated models using humanized mice such as the MISTRG strain (immunodeficient mice with transgenic expression of human cytokines such as GM-CSF, M-CSF, IL-3, and thrombopoietin) provide species-specific cytokine support for human cells, allow for improved engraftment of some such cells, and may allow for modeling both efficacy and toxicity with human cells (31). Additional hybrid models such as the SCID/beige mouse with human T cells interacting with mouse monocyte lineages may provide additional insights (32). A truly efficacious syngeneic mouse model, however, is likely to provide less confounding results than artificially humanized mice but still requires more development to improve on the actionable data generated by present xenograft models. Thus, syngeneic and humanized xenograft mouse models provide complementary insights, with syngeneic systems being most suited to probe murine cell biology with an intact immune cell milieu, whereas humanized systems allow for human cell engraftment and in vivo human cell–cell interaction.

CAR design has progressed during the past two decades (Fig. 2). Three laboratories reported the first generation CAR design in 1991 (33–35). Kuwana et al. (36) first created a chimeric receptor that resulted in MHC class I–independent T cell recognition. Roberts (37) and Finney and Lawson (38) designed the first second generation CARs encoding CD28 or 4-1BB signaling domains. In preclinical models, others found that 4-1BB but not CD28 provided advantages to CAR efficacy in a xenograft model, but used a CD28-based ex vivo expansion system (25, 39). Both CAR models showed preclinical efficacy, and both have demonstrated clinical efficacy (26, 27, 40). Recent studies by Long et al. (41) indicate that CD28-based CARs augment and accelerate T cell exhaustion, whereas 4-1BB–based CARs reduce T cell exhaustion. Our own studies agree with these findings and indicate that CD28-based endodomains can mediate constitutive signaling leading to terminal differentiation of effector T cells (42). The careful conclusion from these studies is not that one human endodomain is necessarily better than the other, but rather that the CD28 signal is key in producing effective clinical T cell products, and that 4-1BB promotes persistence of CAR T cells.

For more than a decade, many clinical trials using ACT had shown a variety of engineered cell products to be quite safe, but relatively nonefficacious. In the past 4 y, ACT has undergone a revolution in efficacy, which reflects that T cells can now be manufactured in such a way as to allow extensive in vivo proliferation (40, 43, 44) and even, in some cases, long-term persistence (3, 27). This has radically changed the efficacy of ACT, especially in the area of CAR T cells, but as is so often the case in cancer therapy, with efficacy comes toxicity.

The most prominent toxicity of CAR T cells for bone marrow–derived tumors is CRS. Highly proliferative T cells can produce CRS, which may range from high fever and myalgias to unstable hypotension and respiratory failure. This was an unexpected observation because it was not observed in preclinical animal models. A key insight into CRS came with the observation that, in addition to the expected effector cytokines such as IFN-γ, IL-6 can be quite elevated during the exponential proliferative phase of CAR T cell therapy (27). CRS is directly and possibly causally related to a complementary toxicity, which is macrophage activation syndrome (26). Fortunately, these insights have also resulted in a therapeutic option for severe CRS, which is IL-6 blockade using the IL-6 receptor antagonist tocilizumab. Several of the groups treating ALL with highly proliferative CAR T cells have confirmed both the IL-6 observation and the efficacy of tocilizumab for severe CRS (40, 45, 46), and we have observed a similar mechanism underlying severe CRS driven by the bispecific T cell–engaging Ab blinatumomab (47). Another important observation is that severe CRS is observed almost exclusively in patients with high disease burden (27). This means that as CAR T cells are deployed in patients earlier in the course of their disease, before the disease becomes refractory to cytoreduction prior to CAR infusion, the risk of severe CRS will be far lower.

CRS is a direct result of T cell proliferation. It is not dependent on the antigenic specificity of the engineered T cell, although there may be differences in CRS risk and symptoms depending on the disease being treated, even where the CAR is the same. For example, CRS is generally milder in CLL and diffuse large B cell lymphoma and more prominent in ALL. In addition to CRS, there are toxicities that are described as “on target” because they directly result from the antigenic specificity of the engineered T cell. Examples of this include tumor lysis syndrome, which is a direct result of tumor cell destruction (43, 48). B cell aplasia is an on-target but off-tissue toxicity, which is seen with CARs directed to B cell–expressed targets such as CD19. Because B cell aplasia is present as long as CD19 CAR T cells are present, absence of B cells serves as a useful pharmacodynamic marker of CAR T cell functional persistence (27). B cell aplasia produces a more profound hypogammaglobulinemia compared with patients infused with anti-CD20–specific mAbs, which must be treated with i.v. Ig replacement. Unlike the relative hypogammaglobulinemia seen with the anti-CD20 Ab rituximab, i.v. Ig replacement is absolutely required during prolonged CAR-mediated B cell aplasia. Two recent case reports of lethal toxicities related to engineered T cell infusion described a patient who received a Her-2 CAR (49) and two patients infused with an engineered TCR targeting MAGE-A3 (50, 51). In both cases, off-cancer expression of the target in normal tissues resulted in rapid and overwhelming cardiopulmonary toxicity. All of these on-target toxicities result from the inability of engineered T cells to distinguish between normal cells and cancer cells that express the targeted Ag. The toxicity from high-affinity TCR-engineered T cells may be a result of the affinity maturation process, as the “on target off tumor” aspect may be restricted to the high-affinity TCR that would not have happened physiologically. Careful screening of high-affinity TCRs will be needed to address the possibility of recognizing epitopes outside of the target protein as a result of high-affinity design process. Neurologic toxicity has also been reported after infusion of T cells engineered with an HLA-A2–restricted MAGE-A3–specific TCR (52).

A further unexpected and as yet unexplained phenomenon seen with CAR T cell therapy against leukemia is the emergence of neurologic symptoms. Reported by several groups, these symptoms are varied but self-limiting, including delirium, dysphasia, akinetic mutism, and seizures (26, 27, 40, 45, 53). Although somewhat temporally related to the systemic CRS and certainly correlated with the presence of CAR cells in the cerebrospinal fluid, these symptoms do not appear to be modified by tocilizumab. The mechanism and target tissues of these symptoms remain to be determined.

Infusion of activated T cells carries the risk of autoimmunity. Vitiligo is seen with melanoma-directed ACT, and toxicities such as rash, colitis, and hypophysitis have been described rarely with activated T cell infusions (54, 55). This has not been a significant issue with CAR-modified T cells. When allogeneic T cells are used for the infusion, there is a potential risk of graft-versus-host disease. This might be a concern in patients who have previously undergone allogeneic stem cell transplant. Fortunately, the approach of collecting tolerized donor-origin T cells from the recipient (rather than going back to the donor) appears to have a very low risk of graft-versus-host disease (27). Certainly, one approach to limiting toxicity would be the use of suicide systems such as the elegant inducible caspase-9 system or the inclusion of defined surface targets such as CD20 (56). In the case of inducible caspase-9, a soluble activating agent causes the induction of the caspase system, resulting in apoptosis of the T cells. By including CD20 in the gene modification of the CAR T cells, a mAb such as rituximab could be used to clear the offending T cells.

At present there are a number of scientific and engineering challenges that are being addressed in the laboratory. Below we address some of the current challenges.

Initial cell therapies were primarily administered as highly differentiated CD8⁺ T cells. These CTLs cells had optimal cytotoxicity but did not have sufficient replicative capacity after infusion, and, with rare exceptions, the infused T cell products had poor persistence in the patients. It is now widely accepted that mixtures of CD4⁺ and CD8⁺ T cells are often preferred, likely because the CD4⁺ T cells provide growth factors and other signals to maintain function and survival of the infused CTLs. Additionally, studies in mice sometimes mislead approaches with human T cells. For instance, human T cells have limitations on replicative capacity by virtue of telomere degradation, a feature not encountered in mouse models, and human CD4 cells can exhibit more cytotoxicity than do mouse CD4 cells (57, 58). At issue is whether to separate subpopulations of cells from patients by flow cytometry or other methods and culture them independently. The optimal cell culture conditions for CD4 cells and CD8 cells are distinct, in part because the signal transduction pathways differ in CD4⁺ and CD8⁺ T cells (59). Additionally, cell separation approaches enable the removal of Tregs that are potentially harmful for therapies using effector T cells. Furthermore, tumor cells may need to be removed from the input T cells, an issue that is particularly challenging in patients with leukemia. However, the cost of cell manufacturing weighs heavily on the technologies used in good manufacturing practices, as cell sorting by flow cytometry or bead-based approaches can easily add $10,000 or more to the cost of goods for the infused product.

Related to the above issue, what is the optimal state of differentiation of the infused T cells? Studies in mice and humans now indicate that naive or central memory cells are preferable (60, 61). A challenge is how to enable this approach in elderly adults who may have very few naive T cells, a condition often aggravated by chemotherapy or other disease processes. Our data in leukemia patients treated with CD19-specific CARs indicate that replicative capacity is the most important predictive biomarker of success (48, 62). One approach is to isolate central memory or naive T cells from input lymphocytes obtained from whole blood. In contrast, a more robust and simple approach has been to use bulk T cells and then to culture the T cells under conditions that promote the maintenance of a less differentiated population of naive and central memory cells. This later approach is based on principles that specific costimulatory signals can promote selected fate and differentiation. For instance, CD28 stimulation can program CD4 cells to maintain central memory states (63, 64), and 4-1BB promotes the growth of CD8 central memory cells (65). In contrast, ICOS cosignaling can promote the outgrowth and stability of Th17 cells (66). The use of T cells with stem cell–like qualities is promising (67, 68), and enforced Wnt signaling promotes the propagation of memory stem cells (69).

It is unknown as to whether the striking and unexpected success of CD19 CAR T cells for the treatment of B cell malignancies can be recapitulated in patients with solid tumors. We find that CAR T cells can kill primary adenocarcinoma cells in vitro with an efficiency that is similar to leukemia targets. However, it remains to be determined whether the cell composition or the cell culture conditions will require modification to optimize trafficking and persistence of engineered T cells for patients with solid tumors. It is likely that the cell engineering to optimize trafficking of T cells to tumors that are compartmented, such as glioblastoma and pancreatic cancer, rather than disseminated as in hematologic malignancies, may differ. One approach that has been proposed is the use of targeting strategies such as genetically engineering various chemokines or chemokine receptors to promote homing of the infused T cells to tumor deposits (70, 71). Other approaches include preconditioning the tumor or the host with radiation (72–75), the injection of oncolytic vectors (76), and the direct intratumoral injection of the T cells (77).

Previous studies have shown that the optimal culture conditions for various T cell subsets are distinct (65, 78). This raises the related question as to whether various lymphocyte subsets of engineered T cells should be equipped with distinct signaling domains? To begin to address this issue we have assessed the use of endodomains comprised of CD28, 4-1BB, and ICOS in human CD4⁺ and CD8⁺ T cells. In humanized mice bearing adenocarcinoma xenografts, we find that CD4⁺ CAR T cells equipped with an ICOS signaling domain are superior, whereas 4-1BB domains are generally preferred in CD8⁺ T cells (79). It remains to be determined whether the benefits derived from the increased complexity of cell manufacturing will be justified in clinical scenarios. Furthermore, it is likely that the optimal composition of the T cell subsets and engineered signaling domains will differ for various tumor microenvironments. For instance, engineered Th17 cells are preferable to Th1 cells in some preclinical tumor models (80).

Perhaps the largest uncertainty with the use of engineered T cells is whether biosynthetic engineering to induce novel specificities and enhanced effector functions will result in unexpected off-target toxicities. This is a significant concern because on-target but off-tumor toxicity has occurred with CAR T cells (49, 81), and off-target toxicity has occurred with TCR-engineered T cells (50, 51). There are various approaches to mitigate these events. We have found that transfecting T cells with mRNA encoding the CAR to provide self-limited expression of the CAR in transferred T cells is useful to screen for immediate toxicities, and that with emergence of toxicity, it will rapidly abate when the infusions are terminated (82, 83). Various approaches to induce apoptosis of genetically modified T cells have been proposed (84, 85), and it is likely that these approaches will be incorporated so that enhanced effector functions can be matched with conditional ablation to achieve stringent safety requirements.

Based on ongoing trials, synthetic T cells expressing engineered CARs and TCRs are poised to gain widespread commercial approval. The resources of Wall Street are now fueling advances that were previously limited due to insufficient investment in this field. The ability to introduce or delete genes in infused T cells (86, 87) has the additional potential to provide novel cell products to overcome the immunosuppression in the tumor microenvironments and may ultimately eliminate the need for checkpoint therapy using systemic Ab blockade (88). It is likely that the advent of advanced genetic engineering technologies for ACT will enable significant progress in applying the principles of synthetic biology to cancer therapy, chronic infections, and autoimmunity.

S.A.G. and C.H.J. have intellectual property with engineered T cells that is owned by the University of Pennsylvania and has been licensed to Novartis. D.M.B. has no financial conflicts of interest.