Preclinical Models in Chimeric Antigen Receptor

Abstract

Cancer immunotherapy has enormous potential in inducing long-term remission in cancer patients, and chimeric antigen receptor (CAR)-engineered T cells have been largely successful in treating hematological malignancies in the clinic. CAR-T therapy has not been as effective in treating solid tumors, in part due to the immunosuppressive tumor microenvironment. Additionally, CAR-T therapy can cause dangerous side effects, including off-tumor toxicity, cytokine release syndrome, and neurotoxicity. Animal models of CAR-T therapy often fail to predict such adverse events and frequently overestimate the efficacy of the treatment. Nearly all preclinical CAR-T studies have been performed in mice, including syngeneic, xenograft, transgenic, and humanized mouse models. Recently, a few studies have used primate models to mimic clinical side effects better. To date, no single model perfectly recapitulates the human immune system and tumor microenvironment, and some models have revealed CAR-T limitations that were contradicted or missed entirely in other models. Careful model selection based on the primary goals of the study is a crucial step in evaluating CAR-T treatment. Advancements are being made in preclinical models, with the ultimate objective of providing safer, more effective CAR-T therapy to patients.

Introduction

Current cancer treatment options include surgery, radiation, chemotherapy, or some combination of these three therapies. More recently, cancer immunotherapy has gained attention as a possible “fourth pillar” of cancer treatment, as it has the potential to induce long-term remission by harnessing the power of the patient's own immune system against malignantly transformed cells. One area of cancer immunotherapy that has seen great clinical success is chimeric antigen receptor (CAR)-engineered T cells. CARs are synthetic receptors consisting of an extracellular antigen-binding domain, a flexible hinge region, a transmembrane domain, and intracellular signaling and co-stimulatory domains. The extracellular portion of the CAR is often an antibody-derived single-chain variable fragment (scFv) that recognizes and binds to a tumor-associated antigen (TAA) present on the cancer cell surface. Binding of the scFv to the cognate antigen triggers an intracellular signaling cascade to activate the T cell against the target cell, typically by activating the CD3ζ T-cell receptor complex, as well as one or more co-stimulatory domains in the cytoplasmic region of the CAR.¹

CAR-T therapies have yielded remarkable results against hematological cancers in the clinic,² but have been much less effective when treating patients with solid tumors.³ One major obstacle presented in solid tumors but not in blood cancers is the tumor microenvironment (TME), which is hostile to CAR-T cells due its acidic and hypoxic environment, as well as the presence of immunosuppressive cells and soluble factors.⁴ Additionally, CAR-T therapy often comes with serious side effects, some of which have resulted in patient death. The most common adverse side effect of CAR-T therapy is cytokine release syndrome (CRS), during which the over-activation of T cells can trigger the massive release of cytokines and subsequent systemic inflammatory response. Severe CRS can result in vascular leak and hypotension, leading to multi-organ failure and patient death.^5,6 Neurotoxic side effects such as aphasia, hallucinations, and seizures can result from CAR-T treatment and are linked to CRS, although the exact connections remain unclear.⁷ On-target off-tumor effects are another serious safety concern in CAR-T therapy. Most TAAs are expressed not only on tumor cells but also on healthy cells, albeit often at much lower levels. On-target off-tumor toxicity occurs when CAR-T cells attack healthy tissues that express the cognate antigen.⁸ Some of these safety concerns can be addressed effectively in the clinic, but others have been unanticipated with devastating consequences.

There are many strategies for making CARs safer or more effective, such as the inclusion of suicide switches,⁹ co-expression with better-matched chemokine receptors or inflammatory cytokines,¹⁰ dual-targeting or co-stimulation-only CARs,¹¹ and combination therapy with checkpoint inhibitors.¹² Fine-tuning the CARs themselves is important to advancing this therapy, but equally crucial are the preclinical models used in testing CAR safety and efficacy. Many animal models used in the studies of CAR-T therapies do not reflect the obstacles to achieving clinical efficacy, and nor do they predict potentially life-threatening safety concerns. This review highlights both the benefits and the shortcomings of various animal models used in CAR-T therapies, including syngeneic, human xenograft, immunocompetent transgenic, and humanized transgenic mice (Fig. 1), as well as primate studies. Careful selection of the preclinical model based on the aims of the study (Table 1) can lead to better prediction of both effectiveness and toxicities of CAR-T therapy in human patients.

Figure 1.

Schematic of the tumor, chimeric antigen receptor–engineered T-cell therapy (CAR-T), tumor-associated antigen, and immune components in four different mouse models: syngeneic, xenograft, transgenic, and humanized.^7,8,13,14 TAA, tumor-associated antigen.

Table 1.

Summary of five different preclinical models used in CAR-T studies, their advantages, disadvantages, and suggested uses^7,8,13,14

Model	Components	Advantages	Disadvantages	Suggested uses
Syngeneic mouse	Murine tumor, CAR-T, and immune system	Intact host immune system; allows the study of off-tumor effects	All cells must be of murine origin; mouse biology and immunity differ from human	Examine host immune effects; evaluate CAR safety when transgenic mouse is not available
Xenograft mouse	Human tumor and CAR-T	Allows the study of human cells; more straightforward validation of proof-of-concept studies	No host immune system; no study of off-tumor effects	Test complex CAR designs/proof-of-concept studies for efficacy; probe questions specific to human biology such as cytokine use
Transgenic mouse	Murine tumor, CAR-T, and immune system with human TAA	Allows study of human TAAs while preserving the host immune system; good for observing off-tumor effects	Not available for all TAAs; all cells must be of murine origin	Evaluate on-target off-tumor effects as well as effects of host immune system
Humanized mouse	Human tumor, CAR-T, and additional human immune system components	Allows the study of human cells but with additional immune cells to mimic TME or potential off-target effects	Humanized immune system still primitive; few existing studies	Observe effects of specific immune cells on CAR-T efficacy; test for HSPC ablation
Primate	Macaque CAR-T and immune system	Similar to human immune system and physiology; able to mimic clinical toxicities	Expensive; small study size; no implanted tumors to test for efficacy	Test for safety after extensive validation in mouse models; last step before clinical use

CAR-T, chimeric antigen receptor–engineered T-cell therapy; TAA, tumor-associated antigen; TME, tumor microenvironment; HSPC, hematopoietic stem/progenitor cell.

Syngeneic (immunocompetent Allograft) Mouse Models

Syngeneic or immunocompetent allograft mouse models use CAR-T cells, tumors, and target antigens that are all murine derived. The main advantage to this model is that the mice are immunocompetent, allowing for the observation of CAR-T cells within the context of a functional immune system. Additionally, syngeneic models can reveal on-target off-tumor toxicities, as healthy murine tissue can express the target antigen at low levels, mirroring the expression patterns in human patients.¹³ However, the syngeneic model has drawbacks, as mouse biology does not always accurately recapitulate human biology. For example, murine immune systems differ from that in humans, and syngeneic models have been largely unable to mimic CRS. Murine CAR-T cells have much shorter persistence and are more susceptible to activation-induced cell death compared to human CAR-T cells, and syngeneic models do not provide much insight into the mechanisms of human CAR-T cells.¹⁴

Syngeneic models targeting hematologic and solid-tumor TAAs

CD19 is the paradigm antigen for CAR-T therapy, and many clinical trials have used CD19-targeting CAR-T regimens. An early study of anti-CD19 CARs used a syngeneic mouse model to demonstrate that a single infusion of CAR-T cells resulted in the long-term elimination of lymphoma and was more effective than using the anti-CD19 monoclonal antibody for antibody-derived cell cytotoxicity. Preconditioning via sub-lethal irradiation of the mice was critical to the efficacy of the treatment. They explored different CAR constructs and showed that a mutant variety of the CD3ζ domain resulted in better persistence than wild-type CD3ζ. This model was useful in detecting B-cell aplasia as a potential toxicity, which is also seen in the clinic—7 months after CAR-T injection, mice were found to be lymphoma-free, but healthy B cells were undetectable as well.¹⁵ Another study confirmed that anti-CD19 CAR-T therapy eradicated both leukemia and healthy B cells in a syngeneic mouse model, and that CAR-T persistence and efficacy were enhanced with cyclophosphamide preconditioning.¹⁶ A CD19 CAR-T study published a few years later was able to determine that toxicities were dependent both on the dosage and on the presence of co-stimulatory domains in the CAR construct. Building on a previous study by the same group,¹⁷ it was shown that first-generation CARs (without co-stimulatory domains) killed lymphoma cells but did not persist or cause adverse side effects. In contrast, second-generation CARs with CD28 co-stimulatory domains induced B-cell aplasia and chronic toxicity accompanied by an increase in CD11b⁺Gr-1⁺ myeloid-derived suppressor cells (MDSCs), which play a role in immunosuppression in the TME. Elevated serum interferon (IFN)-γ and tumor necrosis factor-α levels upon treatment with second-generation CAR-T cells indicated possible CRS. This particular model used BALB/c mice,¹⁸ but other syngeneic models used C3H or C7BL/6 mice and did not show these toxicities. Thus, while syngeneic models are important in evaluating CAR safety, side effects may vary between mouse strains.

CAR-T therapy has been successful in targeting CD19-expressing blood cancers, but more accurate preclinical models are needed to study the safety and efficacy of CARs targeting antigens expressed on solid tumors. Carcinoembryonic antigen (CEA) is one such TAA that serves as a potential target for CAR-T treatment. A syngeneic model was used to examine the immunosuppressive TME in a hepatic cancer treated with anti-CEA CARs. The authors reported that MDSCs from the liver dampened the effect of the therapy through programmed death (PD)-1/PD-L1 interactions. CAR-T function was rescued upon PD-L1 blockade, indicating that CAR-T treatment of solid tumors may be more successful if combined with checkpoint blockade regimens.¹⁹

A syngeneic model testing anti-NKG2D CAR-T cells also demonstrated potential toxicities and strain-specific effects. The CARs were better expressed in BALB/c splenocytes than in C57BL/6 splenocytes, but moribundity and morbidity were also more severe in BALB/c after adoptive transfer. Preconditioning with cyclophosphamide worsened the toxicity, which became apparent just hours after CAR-T injection in both tumor-bearing and tumor-free control mice, indicating on-target off-tumor effects. An immunocompromised mouse xenograft model was used to test one of the CAR constructs used in the syngeneic model and failed to highlight these obvious safety concerns.²⁰ Another study of anti-NKG2D CAR-T therapy emphasized the role of host immune cells and a CAR-T-mediated shift from an immunosuppressive to an immunostimulatory TME. The secretion of granulocyte-macrophage colony-stimulating factor and IFN-γ by CAR-T cells activated host macrophages, which in turn killed tumor cells and inhibited immunosuppressive responses such as interleukin (IL)-10 secretion. This study demonstrated the importance of endogenous immune effects after CAR-T injection, as host macrophage activity was necessary for complete tumor eradication in an advanced ovarian cancer model.²¹

Syngeneic models targeting TME markers

Further testing the impact of CAR-T cells on the TME was performed using CARs targeting VEGF-R, which is upregulated in the TME to promote angiogenesis. Anti-VEGF-R CAR-T cells were effective with no adverse effects in two mouse strains, enhancing the infiltration, retention, and proliferation of tumor-specific T-cell receptor–engineered T cells.²² A later paper by the same group tested anti-VEGF-R CARs in five cancer models in two mouse strains. While tumor size decreased after treatment, mice relapsed after 3 weeks, and lymphodepletion preconditioning had no effect on the treatment efficacy. In alignment with the anti-CD19 and anti-NKG2D CAR studies mentioned previously, no toxicity was observed in C57BL/6 mice, but there was severe toxicity in BALB/c mice if a high dose with CD4⁺ cells was used.²³ This study emphasizes the importance of carefully choosing mouse strains or testing in multiple strains to ensure toxicity is modeled accurately.

Another TAA used to target the TME instead of the tumor cells directly is fibroblast activation protein (FAP), which is expressed by immunosuppressive tumor stromal fibroblasts. Anti-FAP CAR-T cells that recognized both murine and human antigen were tested in a variety of subcutaneous tumor models in two mouse strains. These CARs had limited antitumor effects but serious bone-related toxicities. Later in the study, it was discovered that FAP is strongly expressed on multipotent bone-marrow stromal cells in both mice and humans, and the authors concluded against using FAP as a CAR-T target.²⁴ Conversely, another group showed that CARs targeting murine FAP inhibited multiple subcutaneous tumors with minimal off-tumor toxicity by augmenting the endogenous CD8⁺ T-cell response and increasing numbers of tumor-infiltrating lymphocytes. While the mouse strains used were the same, the disparities between these two studies may be attributed to a number of differences: the study with no toxicities used 4-1BB instead of CD28 as a co-stimulatory domain, had no preconditioning prior to CAR-T injection, injected a lower dose, did not administer exogenous cytokine during treatment, and the CARs recognized a different epitope, which was sensitive only to cells that expressed high amounts of FAP.²⁵

Syngeneic models are important in elucidating cytokine-induced changes in the TME and involvement of host immunity after CAR-T administration, and they can be used to focus on the effects of CAR-T cells on the TME instead of just on the direct lysis of tumor cells. They can also be used to study the persistence and safety of different CAR constructs in a host with an intact immune system. Widely varying results in studies using the same TAA but different mouse strains, as seen in anti-CD19, anti-NKG2D, and anti-VEGF studies, caution against using a single mouse strain to determine safety before moving to clinical trials. For example, BALB/c mice have higher expression of pan-NKG2D ligands, which may explain the increased toxicity of the treatment—possible CRS and severe lung pathology were observed in this strain when treated with anti-NKG2D CARs.²⁰ In the case of the anti-VEGF CAR study, the distinct microbiomes in each mouse strain may also play a role in the varying responses to CAR-T injection. Additionally, the BALB/c CAR-T cells had a higher percentage of CD4⁺ T cells, possibly contributing to adverse side effects, and toxicity was limited to tumor-bearing mice.²³ The anti-FAP CAR-T studies highlight that not only mouse strain, but also CAR construct, dosage, and treatment regimen can lead to vastly different responses in immune competent mouse models.

Human Xenograft Mouse Models

Human xenograft mouse models, referred to simply as “xenograft models” in this review, utilize an immunocompromised mouse injected with human tumor and CAR-T cells. The process of breeding mice that did not reject human cells made its first major breakthrough with the development of T-cell deficient athymic nude mice in 1966.²⁶ Thirty years later, human cell engraftment was greatly improved by crossing non-obese diabetic (NOD) mice, which had impaired innate immunity, with severe combined immunodeficiency (SCID) mice, which lacked an adaptive immune system. In the early 2000s, genetic engineering further diminished immunity by introducing a mutation in the IL-2 receptor γ chain gene. The loss of signal transduction from multiple cytokines impaired lymph node, T-cell, B-cell, and natural-killer (NK) cell development.²⁷ Currently, most xenograft mouse models of CAR-T therapy use the NOD-SCID-IL2rγ^null (NSG) mouse strain.²⁸ Xenograft models have the advantage of permitting the study of human CAR-T cells against human cancer cells, although this is essentially performed in a vacuum, as interactions with other immune cells or healthy human tissues are nonexistent. Nonetheless, xenograft mice serve as a valuable model for testing CAR-T efficacy and for validating proof-of-concept studies.

Xenograft models with dual murine and human TAA recognition

A few groups have published studies in which the CAR construct used in xenograft models reacts with both human and murine antigen. One group combined anti-FAP CARs that targeted human and murine antigen with CARs that recognized the human TAA EphA2 in SCID-beige mice.²⁹ The anti-FAP CAR recognized mouse stroma that surrounded the human tumor xenografts and was instrumental in reducing tumor growth. Treatment with both CAR types enhanced antitumor efficacy compared to treatment with either CAR as a single agent. This model allowed observation of CAR-T therapy against the tumor cells themselves and the TME simultaneously, although further analysis is needed in an immunocompetent model to determine if the anti-FAP CARs help reverse immunosuppression. Another study used promiscuous CARs that targeted several ErbB dimers (murine and human) in SCID-beige mice to evaluate safety. In addition to proving the efficacy of the CAR construct, this study highlighted the importance of the route of administration: intravenous or intratumoral administration caused tumor regression without toxicity, but intraperitoneal injections caused tumor regression with dose-dependent side effects.³⁰ Safety data can be gleaned even in an immunocompromised model, although follow-up studies in immune-competent mice may be needed to explore the mechanisms behind the observed toxicity.

Xenograft models for the validation of proof-of-concept studies

Many proof-of-concept studies testing the selectivity of more complex or novel CAR constructs utilize xenograft models. One area of interest is in creating CARs that are dependent on multiple TAAs for complete activation to minimize off-tumor effects or antigen escape. One such study used NSG mice to test combinatorial antigen targeting to reduce on-target off-tumor effects. One CAR targeted prostate-specific membrane antigen but had suboptimal activation, while a second CAR targeted prostate stem-cell antigen and had a co-stimulatory receptor; the T cells were fully activated only if both antigens were present on the target cell. Tumors expressing only one of these antigens continued to grow, while tumors with both antigens elicited a strong CAR-T response.¹¹ Another study used bispecific CARs targeting both CD19 and CD20 in order to minimize antigen escape in CD19⁻ leukemia. In contrast to CAR-T cells that only targeted CD19, bispecific CAR-T cells were able to eradicate mixed populations of leukemic cells in NSG mice.³¹ Another study used masked CARs with the goal of improving tumor-specific CAR-T activation. The scFvs of anti-EGFR CARs were masked by a peptide with a cleavable linker. Proteases expressed by tumor cells, but not healthy tissue, cleaved this linker, exposing and subsequently activating the CAR in NSG mice.³² Synthetic Notch (synNotch) receptors control transcriptional activation in the presence of specified TAAs, and by replacing the traditional CAR intracellular domains with a synNotch domain, one group was able to control CAR-T cytokine profiles in a xenograft model.³³ Another study tested a switchable CAR that had the flexibility to target multiple antigens: the CAR targeted a specific peptide neo-epitope introduced on a TAA-binding antibody so the CAR-T cells were activated by the antibody instead of directly by the tumor cells. These switchable CARs had comparable efficacy to traditional CARs in a B-cell leukemia model in NSG mice. The activation of the CAR-T cells was controlled by escalating doses of antibodies, which has the potential to reduce the risk of CRS,³⁴ although this cannot be validated in a xenograft model. Finally, a few proof-of-concept studies have used NSG mice to validate CARs that recognize TAAs with antibody mimetic proteins instead of traditional scFvs. One study demonstrated that anti-Her2 CARs with designed ankyrin repeat proteins controlled tumor growth as well as scFV-based CARs,³⁵ while another found that adnectin-based anti-EGFR CARs had comparable antitumor effects to scFv-based CARs.³⁶

In certain instances, xenograft models are necessary to test aspects of CAR-T therapy specific to human components, as mice do not faithfully capture all features of human biology. For example, one group tested the efficacy of various CAR constructs that constitutively expressed different human cytokines, including IL-2, IL-7, IL-15, and IL-21. Each cytokine improved the antitumor effects in NSG mice but with varying degrees of efficacy and through different mechanisms. This severely immunocompromised model was prudently selected to demonstrate the effects of cytokines on adoptively transferred human T cells only, without confounding effects of host immune cells (e.g., IL-15-mediated augmentation of host NK cell activity).¹⁰

Xenograft models have been instrumental in the development of second- and-third generation CARs. Second-generation CARs containing CD28 or 4-1BB co-stimulatory domains were described in the early 2000s,^37,38 but xenograft models have elucidated some key differences between the two domains. NSG mice were used to test exhaustion in two second-generation human CAR-T cells. Anti-GD2 CAR-T cells did not persist or inhibit tumor growth due to T-cell exhaustion caused by tonic signaling. Exhaustion was prevented with 4-1BB co-stimulation and exacerbated by CD28 co-stimulation.³⁹ Another study used second- and third-generation anti-mesothelin CARs with either or both CD28 and 4-1BB co-stimulation. 4-1BB was found to promote multifunctional CAR-T cells that secreted multiple cytokines, while CD28 promoted faster tumor decline.⁴⁰ These facets of CAR-T therapy would not have been recapitulated in a model that required the use of murine T cells, thus necessitating a xenograft model.

Xenograft models with limited human immune cell interactions

Additional human immune cells can be injected into xenograft mice to determine their interactions with CAR-T cells. One study observed the recruitment of human NK cells to the tumor site after CAR-T injection. NSG mice with renal cell cancer were treated with CAIX-targeting CAR-T cells that secreted PD-L1 antibodies. The secreted PD-L1 antibodies were able to recruit human NK cells to the tumor site via binding to the Fcγ receptor on the NK cell surface.⁴¹ However, this model was limited by the short survival period of human NK cells in NSG mice, and safety limitations of this TAA were not explored. CAIX has low expression in human bile ducts and has resulted in CAR-T clinical toxicity. Therefore, follow-up studies in transgenic mice are advised. Some xenograft mouse models have endeavored to examine the effects of immunosuppressive cells on CAR-T efficacy by injecting human regulatory T cells (Tregs) as well as CAR-T cells. A study in SCID-beige mice found that the antitumor effects of anti-CD19 CAR-T with CD28 co-stimulation were nullified by Tregs. The xenograft model allowed for the titration of Treg to CAR-T ratios, and ratios as low as 1:8 resulted in full suppression of antitumor activity.⁴² Another study also demonstrated that CAR-T cells with CD28 co-stimulation were ineffective against solid CEA⁺ tumors due to high Treg infiltration in nude mice. The researchers first used an immune competent model but moved to a xenograft model for greater control over immune cell interactions and to define further the role of Tregs in CAR-T efficacy. Because IL-2 promotes Treg proliferation, mutating the CD28 domain to suppress IL-2 secretion resulted in decreased Treg accumulation and improved antitumor efficacy.⁴³ Another group also found that CAR-T cells with IL-2 administration were rendered ineffective by Tregs in a GD2⁺ tumor model in NSG mice. When transduced with IL-7Rα and expanded with IL-7 instead of IL-2, these CAR-T cells controlled tumor growth regardless of the presence of Tregs.⁴⁴

Xenograft models for the study of patient-derived tumors

Finally, a few groups have established patient-derived xenograft (PDX) models by implanting a primary tumor biopsy instead of injecting tumor cell lines to capture natural tumor heterogeneity better. Additionally, unlike tumor cell lines, PDXs are not exposed to artificial environments involving high serum content and frequent passages. The first report in 2010 matched patient tumor and T cells in NSG mice to show that a cancer patient's autologous T cells can be used in CAR-T therapy with antitumor effects. The study also found that multiple doses of CAR-T cells transfected with nonintegrating vector systems was effective.⁴⁵ Two additional PDX CAR-T reports were recently published in 2017. One tested hepatocarcinoma xenografts established from three patients' primary tumors. All PDXs preserved original tumor characteristics such as morphology, immunological markers, and gene expression, and CAR-T therapy suppressed tumor growth in one patient line and eradicated tumors entirely in the other two lines. The PDX that was more resistant to CAR-T therapy was more aggressive and had higher PD-L1 expression.⁴⁶ This model may more accurately predict patient response and be useful in determining the course of treatment, such as a combination of checkpoint blockade and CAR-T therapies. Another group tested various depletion strategies to remove anti-CD123 CAR-T cells from NSG mice after successfully treating acute myeloid leukemia (AML) PDXs to minimize unwanted chronic hematological toxicities. Robust response and long-term remission required CAR-T persistence for at least 4 weeks before CAR-T ablation. This model efficiently evaluated CAR-T ablation strategies in vivo, but it cannot predict on-target off-tumor effects or prove that hematopoietic stem cells would not be impacted by the CAR-T treatment even after ablation. The authors concluded that the answer to how human CAR-T cells orchestrate off-target toxicities may be found only in early clinical trials.⁴⁷

Mouse xenografts are useful in screening for basic CAR-T efficacy and for answering questions specific to human biology. Several proof-of-concept studies with more complicated CAR designs have been validated in xenograft models, although additional studies in immune competent hosts are required to evaluate CAR safety. Xenograft models are appropriate when probing aspects of human biology in CAR-T therapy, such as the effects of different cytokines or co-stimulatory domains on treatment efficacy, as well as interactions between CAR-T and other human immune cells. The hostile TME includes Tregs and MDSCs, yet it is largely ignored or unaddressed in preclinical immunocompromised models. Tregs are elevated in the peripheral blood and TME of cancer patients and correlate with a worse prognosis.⁴² Tregs may partly explain the disappointing CAR-T clinical trial results in solid tumors, but their inclusion in xenograft models may yield more accurate results. Lastly, xenograft models are needed to study PDXs with limited host immune rejection. PDX models are increasingly used in tumor immunology studies, as they may be more clinically relevant than cancer cell lines.

Immunocompetent Transgenic Mouse Models

Less commonly, immunocompetent transgenic mice, herein referred to as simply “transgenic mice,” have been employed in CAR-T studies to determine treatment safety better. These mice express a human TAA transgene to highlight the potential for on-target off-tumor effects in specific TAA-expressing healthy tissues. As most TAAs are not exclusively expressed on tumors but also at lower levels on healthy tissues, transgenic mice serve as an important model for the observation of unwanted side effects of TAA-specific immunotherapies such as CAR-T treatment. Transgenic mice typically have a murine TAA knockout and human TAA knockin, and can be bred such that TAA expression patterns and levels are similar to those found in humans.⁴⁸ Transgenic models utilize murine T cells and have an intact immune system, like syngeneic models, but allow for the study of human TAA-specific CAR constructs, like xenograft models.

Transgenic models for anti-CD19 CARs

CD19-targeting CARs are well-studied, but most of these are performed in syngeneic or xenograft models. However, one group performed a study analyzing anti-CD19 CAR-T cells in a transgenic mouse to observe the effects of preconditioning in an immune-competent host. These C57BL/6 mice had knockout murine CD19 and knockin human CD19 with expression limited to B cells, mimicking CD19 expression patterns observed in human patients. Preconditioned mice displayed enhanced survival, decreased numbers of regulatory T cells (Tregs), and increased IL-12 secretion by CAR-T cells targeting human CD19. These effects were not seen in mice receiving treatment without preconditioning. The researchers then modified the CAR to secrete murine IL-12 constitutively. This CAR eradicated the tumor, resulted in long-term survival, and induced B-cell aplasia in a preconditioning-independent manner in transgenic mice. No toxicities were observed in the mice after treatment other than B-cell aplasia.⁴⁹

Transgenic models for anti-CEA CARs

Several studies have been performed in transgenic mice to evaluate anti-CEA CARs. CEA is a desirable CAR target, as it is overexpressed on many cancers of the gastrointestinal and pulmonary tracts, but it is not tumor selective and is also expressed in healthy intestine and lung tissue. Transgenic C57BL/6 mice expressing human CEA in the intestinal and pulmonary tracts underwent anti-CEA CAR-T therapy. The CAR-T cells produced long-term tumor eradication in an orthotopic pancreatic adenocarcinoma model, but also heavily infiltrated the intestines and lung in transgenic but not in wild-type mice. Despite the dense infiltrates, mice did not develop signs of an autoimmune inflammatory response, and no pathology was found in the healthy tissue. CEA was also secreted in the serum in this model, as is seen in human patients, but soluble CEA did not hinder CAR-T function.⁵⁰ A later study using the same model indicated that long-term survival of CAR-T cells, but not of host T cells, are necessary for the complete rejection of secondary tumor challenges. This study again confirmed the apparent safety of the CAR, as no healthy tissue damage was observed.⁵¹ Another study transduced Tregs with anti-CEA CAR to prevent autoimmunity: CAR-Tregs were redirected to the gastrointestinal tract and dampened the effects of colitis. Transgenic mice developed colitis and subsequent colorectal cancer when injected with non-Treg CAR-T cells—this is contrary to the previously mentioned studies that did not find CAR-T-mediated autoimmunity. However, colitis severity and tumor burden were reduced when mice were subsequently injected with CAR-Tregs. Thus, CAR-Tregs were able to suppress autoimmune disease and, though counterintuitive, had an anticancer effect by downregulating inflammatory mediators.⁵²

Transgenic models for anti-Her2 CARs

Her2, like CEA, is commonly overexpressed in a variety of cancers but is also expressed at lower levels in healthy tissues. The first study in Her2 transgenic mice built on previous research from the group, which found that anti-Her2 CAR-T treatment eradicated lung metastases in a xenograft mouse model. They demonstrated that CAR-T treatment is also effective at tumor eradication in an immune-competent transgenic model, although survival was improved with prior lymphodepletion and IL-2 administration during treatment. No off-target effects were detected in the mammary tissue or brain, which expressed human Her2 at lower levels.⁵³ Two more studies used transgenic Her2 models to test CAR-T therapy combined with immune checkpoint blockade. CAR-T therapy was enhanced with PD-1 blockade but did not exacerbate targeting of Her2-expressing healthy tissues.^54,55 In addition to testing PD-1 blockade, one of these studies found that CAR-T cells generated from transgenic mice with an adenosine receptor 2A (A2aR) knockout were resistant to immunosuppression when transplanted to transgenic Her2 mice.⁵⁵ A2aR is implicated in the immunosuppression of T cells in the TME. This study demonstrates that transgenic models are not limited to modeling TAA expression but also can be used to study receptors with immunoregulatory functions such as A2aR. These transgenic models did not recapitulate the pulmonary toxicity as seen in one anti-Her2 CAR clinical trial, although another study did mirror these results. Using transgenic Her2 mice with spontaneously developing tumors, they found that a single CAR-T injection did not quell tumor growth and recurrence, but multiple doses controlled tumor relapse without damaging Her2⁺ healthy tissues. Multiple low doses did not prompt adverse effects, but a single high dose resulted in death within a few days due to cytokine storm, determined by elevated IFN-γ serum levels.⁵⁶

CARs targeting the TAAs expressed in each of these transgenic models—CD19, CEA, and Her2—have been tested in the clinic. In an anti-CD19 CAR-T clinical trial, patients who did not receive preconditioning had no objective response and progressive disease, even though those CARs demonstrated a strong antitumor response in preclinical xenograft models. Conversely, patients preconditioned with cyclophosphamide had much better outcomes.⁵⁷ This caveat was not detected in xenograft models, but it was confirmed in the transgenic model. A clinical trial of anti-CEA CAR-T induced severe autoimmunity (colitis and pneumonia) in all three patients, which was transient but required therapy withdrawal.⁵⁸ Two of the three CEA transgenic mouse studies did not predict this autoimmunity, possibly because, although the mice expressed high levels of human CEA, pancreatic cancer patients have a loss of colon mucosal folds, making these cells more accessible to CAR-T, and bacterial colonies in the gut may enhance immune cell function. These conditions cannot be replicated in transgenic mice and may contribute to disparate safety results. Similarly, the death of a patient after a high dose of anti-Her2 CAR-T cells was observed in a clinical trial.⁶ Yet, three of the four Her2 transgenic models did not predict this lethality. Improved preclinical models clearly are needed for CARs targeting these TAAs. The model that correctly predicted anti-Her2 CAR toxicity utilized spontaneously developing tumors. Grafted tumors do not recapitulate many of the properties of naturally occurring tumors. A transgenic model that spontaneously develops tumors may better mimic the clinical progression of human cancers and predict off-tumor toxicity.

Humanized Transgenic Mouse Models

Humanized transgenic mice, often called “humanized mice,” are immunocompromised mice implanted with human immune cells in addition to human tumor and CAR-T cells. They may bridge the gap between syngeneic and xenograft models, as they are tolerant to human cells yet have aspects of a human immune system. Several humanized mouse models have been used in the field of tumor immunology, but only a few have been utilized thus far in CAR-T studies. Humanized models can be relatively simple, such as NSG mice transplanted with human CD34⁺ hematopoietic stem/progenitor cells (HSPCs) prior to CAR-T treatment. HSPCs can regenerate myeloid and lymphoid compartments, although T-cell development is not optimal due to the lack of a thymus. In the humanized immune system (HIS) model, sub-lethally irradiated newborn BALB/c Rag2⁻/⁻γ_c ⁻/⁻ mice are injected with CD34⁺ cells for improved T-cell development. Newborn immunodeficient mice, including BALB/c Rag2⁻/⁻γ_c ⁻/⁻ mice, as well as NSG and NOG strains, may support better T-cell development, as they do not display the thymic involution and phagocytic activity against human immune cell engraftment seen in adult mice. Finally, the more complicated BLT model utilizes fetal bone marrow, liver, and thymus tissue to generate a wide variety of human immune cells and have a more complete reconstitution of T cells in vivo. ¹⁴ Humanized models can provide valuable information pertinent to CAR-T therapy, such as the ablation of HPSCs as an unwanted side effect.

Humanized models with transplanted CD34⁺ cells

Several groups have used NSG mice transplanted with human CD34⁺ cells to look for CAR-T toxicity against HSPCs. One group transplanted NSG mice with CD34⁺ cells and treated the mice with anti-CD19 CAR-T cells transduced with a suicide gene. Upon administration of the drug that activated the suicide gene, they observed that CD19⁺ cells derived from the HSPCs continued to proliferate, while CAR-T cells were eliminated. They used allogenic cells to exacerbate potential CRS further to show that the suicide gene was able to rescue mice from this dangerous condition and prevent unwanted B-cell aplasia.⁵⁹ Another study used CAR-T cells targeting adhesion receptor CD44v6, which prevented the engraftment of AML and multiple myeloma in immunocompromised mice. The group then demonstrated that these CAR-T cells spared normal hematopoietic stem cells in cytokine-engineered NSG mice that were infused with cord blood–derived HSPCs 4 weeks prior to treatment. The only toxicity observed was monocytopenia, which rebounded when CAR-T cells contracted; this reversibility indicated that the HSPC pool was spared and able to regenerate CD14⁺ cells.⁶⁰ Two more studies both used anti-CD123 CAR-T cells in CD34⁺ cell-transplanted NSG mice but had conflicting results. One found that anti-CD123 CAR-T cells eliminated primary AML PDXs but spared both cord blood–derived and adult bone marrow–derived CD34⁺ cells. This was not the case in CD33-targeting CAR-T cells, which significantly reduced CD34⁺ cells, indicating that CD123 may be a more desirable target in treating AML.⁶¹ However, the second study showed that anti-CD123 CAR-T therapy eradicated both primary AML PDXs in NSG mice and normal myelopoiesis in fetal liver-derived CD34⁺ cells in HIS mice.⁶² Several differences in study design may have contributed to these opposite conclusions, including the choice of co-stimulatory domains in the CAR constructs, different HSPC sources, the use of adult versus newborn mice, and CAR-T ex vivo expansion methods. Follow-up studies are needed to elucidate the key differences between these two findings. Another group used a HIS model to test anti-CD30 CAR-T cells, which eradicated lymphoma without lasting B-cell aplasia. CD30 is also expressed by HSPCs during activation but in much lower levels than in lymphoma cells. This difference in expression level, as well as the expression of granzyme B inhibitor by HSPCs, may have contributed to the protection of the engrafted HSPCs from CAR-T-mediated toxicity.⁶³

Humanized model using fetal bone marrow, liver, and thymus

Finally, one group used a BLT model to test CAR-T efficacy against human immunodeficiency virus (HIV). BLT mice have been used in HIV models previously, allowing HSPCs to differentiate in vivo. CARs targeting HIV glycoprotein gp120 were transduced into HSPCs, differentiated into both CAR-T and CAR-NK cells, and suppressed HIV replication in vivo. Responses were variable, and mice that responded well had a greater expansion of CAR-T cells and higher percentages of CD14⁺ antigen presenting cells (APCs). Low APC presence greatly impacted disease outcome and may be due to suboptimal development of the myeloid compartment in the BLT model.⁶⁴

These reports provide insight into the value of humanized mouse models in preclinical studies of CAR-T therapies. Humanized mice are well-suited for use in models of hematological cancers, as on-target off-tumor effects may result in the ablation of HSPCs. However, in these models, CAR-T cells were injected weeks after HSPC engraftment, which does not mimic the conditions of patients who are lymphodepleted just prior to CAR-T treatment. Furthermore, these studies performed separate tumor and HSPC experiments; a study design in which mice are engrafted with HSPCs, inoculated with tumor, and then treated with CAR-T may provide a more clinically relevant milieu. BLT models are said to recapitulate most closely the full human immune system, although defects in the myeloid compartment leave room for improvement. As humanized mice become more sophisticated, they may provide more clinically accurate models for CAR-T therapy.

Primate Models

Until a few years ago, all CAR-T preclinical studies had been performed in mice. However, significant differences exist between mouse and human physiology, TAA sequences, and immune systems. Primates have immune systems much more similar to humans and can serve as a large-animal model for CAR-T treatment. Murine models largely have not been able to replicate clinical toxicities such as CRS or neurotoxicity, but primate models may be more suitable for these purposes, given the physiological similarities to humans.

The first primate CAR-T study used ROR1-targeting CAR-T cells that had been previously validated in xenograft mouse models. ROR1 is highly conserved between humans and macaques and has similar tissue expression patterns, and a primate model can provide valuable information regarding the safety of ROR1-targeting CARs. Autologous macaque T cells expressing human CAR constructs were transferred to two macaques. These CAR-T cells trafficked to areas with high amounts of ROR1⁺ B cells, such as lymph nodes and bone marrow, and depleted ROR1⁺ B-cell precursors but not mature B cells. No toxicity was observed, even when high doses were administered; elevated IFN-γ and IL-6 were detected in the serum 1 day after administration but returned to normal levels by 3 days post transfer.⁹ Another primate study used CARs targeting L1 cell adhesion molecule; another used TAA, which is highly conserved between humans and macaques and has similar expression profiles. Doses that were 100 times higher than those used in clinical trials were found to be well tolerated in two macaques, with only a transient increase of IFN-γ and IL-6 levels in the serum and no observed toxicities. Again, this CAR construct was first screened for efficacy in a xenograft mouse model before testing for safety in a primate model, and a Phase I clinical trial was initiated following these studies.⁶⁵ A third primate study stood in sharp contrast to the other two reports in that profound toxicities, including B-cell aplasia, neurotoxicity, and CRS symptoms, were observed after administration of anti-CD20 CAR-T cells to three macaques. Elevated IL-6 and IL-8 serum levels mirrored CRS data gathered from clinical trials. Neurotoxicity symptoms included extremity tremors and behavioral abnormalities within 1 week after CAR-T injection but were controlled upon the administration of an anti-epileptic drug. Unlike the previous studies, which showed poor CAR-T persistence, anti-CD20 CAR-T cells underwent significant in vivo expansion and persisted past 6 weeks. Even though the dose used in this study was 10 times lower than the doses used in the other two studies, this was the only primate model that displayed adverse effects.⁶⁶ The exact reason behind these differences—whether it was due to cyclophosphamide preconditioning or due to the nature of the target antigen or CAR construct—remains to be seen.

One significant limitation of these primate models is a lack of tumor cells in the animals. The first study injected ROR1⁺ T cells to serve as target cells, but this likely did not achieve the antigen levels of a typical tumor burden. Tumors could chronically activate the CAR-T cells and potentially enhance their off-tumor response. In two of three studies, CAR-T cells did not persist past 6 weeks due to immunogenicity of the CAR construct, eliminating the observation of long-term effects on the macaques. Finally, large-animal models often necessitate using smaller groups; each of these primate studies tested only two or three macaques. Even so, the macaque immune system provides good insight into how CAR-T cells would interact with the human immune system. CRS and neurotoxicity stemming from CAR-T treatment are poorly understood, and no mouse models truly exhibit these phenomena, while primate models displayed side effects much like those observed in the clinic. Other TAAs that are highly conserved between macaques and humans include EGFRv06, folate receptor, and FAP.⁹ Primate studies of CARs targeting these antigens may provide useful safety information after validation in murine models.

Conclusions and Future Directions

For decades, mouse models have been the link between in vitro experiments and clinical trials. Mice are small, easy to handle, relatively inexpensive, have short life-spans, and reproduce quickly, making them ideal for scientific research in many ways. However, mice are far from the optimal preclinical model for cancer immunotherapy, although breakthroughs are being made in the humanization of murine immune systems. Many CAR-T studies are done in human xenograft models, where it is hard to delineate between xenogeneic rejection, allogeneic response of human CAR-T cells to the tumor, and the actual CAR-T therapeutic efficacy in causing tumor regression. Furthermore, the lack of host immune system does not allow testing of the TME, the tumor's metastatic potential, or the host response to CAR-T cells. Only a few studies have used xenograft mice to study the effects of Tregs on CAR-T efficacy, but studies including other immunosuppressive cells are lacking; for example, transferring human MDSCs to mice prior to CAR-T injection will provide more information on the effects of the immunosuppressive TME. Syngeneic models address some of these concerns but do not allow the evaluation of optimal human CAR-T ex vivo preparation. Each murine model has advantages and shortcomings, and multiple model types may provide a better understanding of a particular CAR-T regimen than a single model. Syngeneic models, which have an intact host immune system but require murine cells, and xenograft models, which are tolerant to human cells but are immunodeficient, can be complimentary and paired to test for CAR-T efficacy as well as safety.

Immunocompetent transgenic mice have been utilized for only three CAR-T TAAs to date, but they have the potential to provide information that neither syngeneic nor xenograft models can give, as transgenic models can reveal off-tumor toxicity specific to human TAAs. Creating transgenic mouse strains for additional human TAAs is a laborious task, but one that may generate many more valuable CAR-T studies. However, most transgenic mouse models mentioned in this review failed to recapitulate the toxicities observed in the clinic. One transgenic model that did accurately model lethal clinical toxicities expressed the human Her2 gene under a mouse mammary tumor virus promoter, resulting in spontaneously developing tumors. Testing CAR-T therapy in mice with such spontaneous cancers has been lacking but may be more clinically accurate, as spontaneous tumors have a similar progression as seen in cancer patients.

Within the last two decades, great strides have been made in using humanized transgenic mice to reflect the human immune system better in an animal model. CAR-T studies in humanized mice have provided safety data regarding HSPC ablation. However, these models are still in the early stages of development, and there is great opportunity for improvement. Multiple groups tested the effects of CAR-T treatment by using mice engrafted with CD34⁺ HSPCs, although studies with mice that have both CD34⁺ cells and cancer cells concurrently are needed. CD34⁺ cells can be isolated from a variety of sources, including adult bone marrow, peripheral blood, cord blood, and fetal liver tissue. Cord blood is thought to be the best source to date, as cells are easily obtainable and result in higher engraftment compared to other sources.⁶⁷ The disparate results seen in the CAR-T studies previously mentioned may be in part due to the different sources of CD34⁺ cells. More research is needed into which source most reliably reflects clinical outcomes. Furthermore, the proportions of lymphoid and myeloid cells that differentiate from CD34⁺ cells in mice are much different from those found in humans, and the established immune cells tend to be naïve.⁶⁸ To address these concerns, humanized mouse models are being modified with human cytokine knockins to promote better human immune cell development.¹⁴

Finally, primate models have been used very recently to monitor unwanted CAR-T side effects, although these studies include a small number of animals and do not test CAR-T antitumor efficacy. The macaque immune system is like the human immune system, and this is the only model in which CAR-T-mediated neurotoxicity has been observed. The primate model can be useful in studying CARs against TAAs that are highly conserved between humans and macaques. However, due to ethical and monetary reasons, macaque studies should be performed after extensive validation in mouse models, as a last step before initiating clinical trials. Ultimately, there is no single perfect preclinical model for CAR-T therapy, but advancements in breeding transgenic mouse strains, improvements in the humanization of murine immune systems, and the combination of multiple animal models will provide more information of different aspects of CAR-T treatment. The evolution and refinement of preclinical models will lead to improved prediction of CAR-T safety and efficacy in the clinic.

Footnotes

Acknowledgments

We thank the constant support of nanomedicine research from the Ming Hsieh Institute for Research on Engineering-Medicine for Cancer.

Author Disclosure

The authors declare no competing financial interests.

References

Sadelain

, Brentjens

, Rivière

. The basic principles of chimeric antigen receptor design. Cancer Discov, 2013; 3:388–398.

Maude

, Frey

, Shaw

, et al. Chimeric antigen receptor T cells for sustained remissions in leukemia. New Engl J Med, 2014; 371:1507–1517.

Newick

, Moon

, Albelda

. Chimeric antigen receptor T-cell therapy for solid tumors. Oncolytics, 2016; 3:16006.

Albini

, Sporn

. The tumor microenvironment as a target for chemoprevention. Nature Rev, 2007; 7:139–147.

Brentjens

, Yeh

, Bernal

, et al. Treatment of chronic lymphocytic leukemia with genetically targeted autologous T cells: case report of an unforeseenadverse event in a Phase I clinical trial. Mol Ther, 2010; 18:666–668.

Morgan

, Yang

, Kitano

, et al. Case report of a serious adverse event following the administration of T cells transduced with a chimeric antigen receptor recognizing ERBB2. Mol Ther, 2010; 18:843–851.

Wegner

. Chimeric antigen receptor T cells for the treatment of cancer and the future of preclinical models for predicting their toxicities. Immunother, 2017; 9:669–680.

Barrett

, Grupp

, June

. Chimeric antigen receptor (CAR) and T cell receptor (TCR) modified T cells enter Main Street and Wall Street. J Immunol, 2015; 195:755–761.

Berger

, Sommermeyer

, Hudecek

, et al. Safety of targeting ROR1 in primates with chimeric antigen receptor–modified T cells. Cancer Immunol Res, 2015; 3:206–216.

10.

Markley

, Sadelain

. IL-7 and IL-21 are superior to IL-2 and IL-15 in promoting human T cell–mediated rejection of systemic lymphoma in immunodeficient mice. Blood, 2010; 115:3508–3519.

11.

Kloss

, Condomines

, Cartellieri

, et al. Combinatorial antigen recognition with balanced signaling promotes selective tumor eradication by engineered T cells. Nat Biotechnol, 2013; 31:71–75.

12.

Gargett

, Yu

, Dotti

, et al. GD2-specific CAR T cells undergo potent activation and deletion following antigen encounter but can be protected from activation-induced cell death by PD-1 blockade. Mol Ther, 2016; 24:1135–1149.

13.

Sanmamed

, Chester

, Melero

, et al. Defining the optimal murine models to investigate immune checkpoint blockers and their combination with other immunotherapies. Ann Oncol, 2016; 27:1190–1198.

14.

Holzapfel

, Wagner

, Thibaudeau

, et al. Concise review: humanized models of tumor immunology in the 21st century: convergence of cancer research and tissue engineering. Stem Cells, 2015; 33:1696–1704.

15.

Kochenderfer

, Yu

, Frasheri

, et al. Adoptive transfer of syngeneic T cells transduced with a chimeric antigen receptor that recognizes murine CD19 can eradicate lymphoma and normal B cells. Blood, 2010; 116:3875–3886.

16.

Davila

, Kloss

, Gunset

, et al. CD19 CAR-targeted T cells induce long-term remission and B cell aplasia in an immunocompetent mouse model of B cell acute lymphoblastic leukemia. PLoS One, 2013; 8:e61338.

17.

Cheadle

, Hawkins

, Batha

, et al. Natural expression of the CD19 antigen impacts the long-term engraftment but not antitumor activity of CD19-specific engineered T cells. J Immunol, 2010; 184:1885–1896.

18.

Cheadle

, Sheard

, Rothwell

, et al. Differential role of Th1 and Th2 cytokines in autotoxicity driven by CD19-specific second-generation chimeric antigen receptor T cells in a mouse model. J Immunol, 2014; 192:3654–3665.

19.

Burga

, Thorn

, Point

, et al. Liver myeloid-derived suppressor cells expand in response to liver metastases in mice and inhibit the anti-tumor efficacy of anti-CEA CAR-T. Cancer Immunol Immunother, 2015; 64:817–829.

20.

VanSeggalen

, Hammill

, Dvorkin-Gheva

, et al. T cells engineered with chimeric antigen receptors targeting NKG2D ligands display lethal toxicity in mice. Mol Ther, 2015; 23:1600–1610.

21.

Spear

, Barber

, Rynda-Apple

, et al. Chimeric antigen receptor T cells shape myeloid cell function within the tumor microenvironment through IFN-γ and GM-CSF. J Immunol, 2012; 188:6389–6398.

22.

Chinnasamy

, Yu

, Theoret

, et al. Gene therapy using genetically modified lymphocytes targeting VEGFR-2 inhibits the growth of vascularized syngenic tumors in mice. J Clin Investigation, 2010; 120:3953–3968.

23.

Chinnasamy

, Tran

, Yu

, et al. Simultaneous targeting of tumor antigens and the tumor vasculature using T lymphocyte transfer synergize to induce regression of established tumors in mice. Cancer Res, 2013; 73:3371–3380.

24.

Tran

, Chinnasamny

, Yu

, et al. Immune targeting of fibroblast activation protein triggers recognition of multipotent bone marrow stromal cells and cachexia. J Exp Med, 2013; 210:1125–1135.

25.

Wang

, Lo

, Scholler

, et al. Targeting fibroblast activation protein in tumor stroma with chimeric antigen receptor T cells can inhibit tumor growth and augment host immunity without severe toxicity. Immunol Res, 2013; 2:154–166.

26.

Flanagan

. “Nude,” a new hairless gene with pleiotropic effects in the mouse. Genet Res, 1966; 8:295–309.

27.

Goldman

, Blundell

, Lopes

, et al. Enhanced human cell engraftment in mice deficient in RAG2 and the common cytokine receptor gamma chain. Br J Haematol, 1998; 103:335–342.

28.

Walsh

, Kenney

, Jangalwe

, et al. Humanized mouse models of clinical disease. Annu Rev Pathol, 2017; 24:187–215.

29.

Kakarla

, Chow

, Mata

, et al. Antitumor effects of chimeric receptor engineered human T cells directed to tumor stroma. Mol Ther, 2013; 21:1611–1620.

30.

van der Stegen

, Davies

, Wilkie

, et al. Preclinical in vivo modeling of cytokine release syndrome induced by ErbB-retargeted human T cells: identifiying a window of therapeutic opportunity?. J Immunol, 2013; 191:4589–4598.

31.

Zah

, Lin

, Silva-Benedict

, et al. T cells expressing CD19/CD20 bi-specific chimeric antigen receptors prevent antigen escape by malignant B cells. Cancer Immunol Res, 2016; 4:498–508.

32.

Han

, Bryson

, Zhao

, et al. Masked chimeric antigen receptor for tumor-specific activation. Mol Ther, 2017; 25:274–284.

33.

Roybal

, Williams

, Morsut

, et al. Engineering T cells with customized therapeutic response programs using synthetic Notch receptors. Cell, 2016; 167:419–432.

34.

Rodgers

, Mazagova

, Hampton

, et al. Switch-mediated activation and retargeting of CAR-T cells for B-cell malignancies. Proc Natl Acad Sci U S A, 2016; 113:E459–E468.

35.

Siegler

, Li

, Kim

, et al. Designed ankyrin repeat proteins as Her2 targeting domains in chimeric antigen receptor-engineered T cells. Hum Gene Ther, 2017; 28:726–736.

36.

Han

, Cinay

, Zhao

, et al. Adnectin-based design of chimeric antigen receptor for T cell engineering. Mol Ther, 2017; 25:2466–2476.

37.

Maher

, Brentjens

, Gunset

, et al. Human T-lymphocyte cytotoxicity and proliferation directed by a single chimeric TCRζ/CD28 receptor. Nat Biotechnol, 2002; 20:70–75.

38.

Imai

, Mihara

, Andreansky

, et al. Chimeric receptors with 4-1BB signaling capacity provoke potent cytotoxicity against acute lymphoblastic leukemia. Leukemia, 2004; 18:676–684.

39.

Long

, Haso

, Shern

, et al. 4-1BB costimulation ameliorates T cell exhaustion induced by tonic signaling of chimeric antigen receptors. Nat Med, 2015; 21:581–590.

40.

Carpenito

, Milone

, Hassan

, et al. Control of large, established tumor xenografts with genetically retargeted human T cells containing CD28 and CD137 domains. Proc Natl Acad Sci U S A, 2009; 106:3360–3365.

41.

Suarez

, Chang

, Sun

, et al. Chimeric antigen receptor T cells secreting anti-PD-L1 antibodies more effectively regress renal cell carcinoma in a humanized mouse model. Oncotarget, 2016; 7:34341–34355.

42.

Lee

, Hayman

, Pegram

, et al. In vivo inhibition of human CD19 targeted effector T cells by natural T regulatory cells in a xenotransplant murine model of B cell malignancy. Cancer Res, 2011; 71:2871–2881.

43.

Kofler

, Chmiewlewski

, Rapp

, et al. CD28 costimulation impairs the efficacy of a redirected T-cell antitumor attack in the presence of regulatory T cells which can be overcome by preventing Lck activation. Mol Ther, 2011; 19:760–767.

44.

Perna

, Pagliara

, Mahendravada

, et al. Interleukin-7 mediates selective expansion of tumor-redirected cytotoxic T lymphocytes (CTLs) without enhancement of regulatory T-cell inhibition. Clin Cancer Res, 2014; 20:131–139.

45.

Zhao

, Moon

, Carpenito

, et al. Multiple injections of electroporated autologous T cells expressing a chimeric antigen receptor mediate regression of human disseminated tumor. Cancer Res, 2010; 70:9053–9061.

46.

Jiang

, Jiang

, Chen

, et al. Anti-GPC3-CAR T cells suppress the growth of tumor cells in patient-derived xenografts of hepatocellular carcinoma. Front Immunol, 2017; 7:690.

47.

Tasian

, Kenderian

, Shen

, et al. Optimized depletion of chimeric antigen receptor T cells in murine xenograft models of human acute myeloid leukemia. Blood, 2017; 129:2395–2407.

48.

Bhattacharya-Chatterjee

, Saha

, Foon

, et al. Carcinoembryonic antigen transgenic mouse models for immunotherapy and development of cancer vaccines. Curr Protoc Immunol, 2008;chap. 20:unit 20.8.

49.

Pegram

, Lee

, Hayman

, et al. Tumor-targeted T cells modified to secrete IL-12 eradicate systemic tumors without need for prior conditioning. Blood, 2012; 119:4133–4141.

50.

Chmielewski

, Hahn

, Rappl

, et al. T cells that target carcinoembryonic antigen eradicate orthotopic pancreatic carcinomas without inducing autoimmune colitis in mice. Gastroenterology, 2012; 143:1095–1107.

51.

Chmielewski

, Hombach

, Abken

, et al. Antigen-specific T-cell activation independently of the MHC: chimeric antigen receptor-redirected T cells. Front Immunol, 2013; 4: 371.

52.

Blat

, Zigmond

, Alteber

, et al. Suppression of murine colitis and its associated cancer by carcinoembryonic antigen-specific regulatory T cells. Mol Ther, 2014; 22:1018–1028.

53.

Wang

, Westwood

, Moeller

, et al. Tumor ablation by gene-modified T cells in the absebse of autoimmunity. Cancer Res, 2010; 70:9591–9598.

54.

John

, Devaud

, Duong

, et al. Anti-PD-1 antibody therapy potently enhances the eradication of established tumors by gene-modified T cells. Clin Cancer Res, 2013; 19:5636–5646.

55.

Beavis

, Henderson

, Guiffrida

, et al. Targeting the adenosine 2A receptor enhances chimeric antigen receptor T cell efficacy. J Clinical Investigation, 2017; 127:929–941.

56.

Globerson-Levin

, Waks

, Eshhar

, et al. Elimination of progressive mammary cancer by repeated administrations of chimeric antigen receptor-modified T cells. Mol Ther, 2014; 22:1029–1038.

57.

Brentjens

, Riviere

, Park

, et al. Safety and persistence of adoptively transferred autologous CD19-targeted T cells in patients with relapsed or chemotherapy refractory B-cell leukemias. Blood, 2011; 118:4817–4828.

58.

Parkhurst

, Yang

, Langan

, et al. T cells targeting carcinoembryonic antigen can mediate regression of metastatic colorectal cancer but induce severe transient colitis. Mol Ther, 2011; 19:620–626.

59.

Diaconu

, Ballard

, Zhang

, et al. Inducible caspase-9 selectively modulates the toxicities of CD19-specific chimeric antigen receptor–modified T cells. Mol Ther, 2017; 25:580–592.

60.

Casucci

, di Robilant

, Falcone

, et al. CD44v6-targeted T cells mediate potent antitumor effects against acute myeloid leukemia and multiple myeloma. Blood, 2013; 122:3461–3472.

61.

Pizzitola

, Anjos-Afonso

, Rouault-Pierre

, et al. Chimeric antigen receptors against CD33/CD123 antigens efficiently target primary acute myeloid leukemia cells in vivo . Leukemia, 2014; 28:1596–1605.

62.

Gill

, Tasian

, Ruella

, et al. Preclinical targeting of human acute myeloid leukemia and myeloablation using chimeric antigen receptor–modified T cells. Blood, 2014; 123:2343–2354.

63.

Hombach

, Gorgens

, Chmielewski

, et al. Superior therapeutic index in lymphoma therapy: CD30+ CD34+ hematopoietic stem cells resist a chimeric antigen receptor T-cell attack. Mol Ther, 2016; 24:1423–1434.

64.

Zhen

, Kamata

, Rezek

, et al. HIV-specific immunity derived from chimeric antigen receptor-engineered stem cells. Mol Ther, 2015; 23:1358–1367.

65.

Kunkele

, Taraseviciute

, Finn

, et al. Preclinical assessment of CD171-directed CAR T-cell adoptive therapy for childhood neuroblastoma: CD7 epitope target safety and product manufacturing feasibility. Clin Cancer Res, 2017; 23:466–477.

66.

Taraseviciute

, Kean

, Jensen

, et al. Creation of the first non-human primate model that faithfully recapitulates chimeric antigen receptor (CAR) T cell-mediated cytokine release syndrome (CRS) and neurologic toxicity following B cell-directed CAR-T cell therapy. Blood, 2016; 128:651.

67.

Werner-Klein

, Proske

, Werno

, et al. Immune humanization of immunodeficient mice using diagnostic bone marrow aspirates from carcinoma patients. PLoS One, 2014; 9:e97860.

68.

Rongvaux

, Takizawa

, Strowig

, et al. Human hemato-lymphoid system mice: current use and future potential for medicine. Annu Rev Immunol, 2013; 31:635–674.

Preclinical Models in Chimeric Antigen Receptor–Engineered T-Cell Therapy

Abstract

Introduction

Syngeneic (immunocompetent Allograft) Mouse Models

Syngeneic models targeting hematologic and solid-tumor TAAs

Syngeneic models targeting TME markers

Human Xenograft Mouse Models

Xenograft models with dual murine and human TAA recognition

Xenograft models for the validation of proof-of-concept studies

Xenograft models with limited human immune cell interactions

Xenograft models for the study of patient-derived tumors

Immunocompetent Transgenic Mouse Models

Transgenic models for anti-CD19 CARs

Transgenic models for anti-CEA CARs

Transgenic models for anti-Her2 CARs

Humanized Transgenic Mouse Models

Humanized models with transplanted CD34+ cells

Humanized model using fetal bone marrow, liver, and thymus

Primate Models

Conclusions and Future Directions

Footnotes

Acknowledgments

Author Disclosure

References

Humanized models with transplanted CD34⁺ cells