Chimeric antigen receptor (CAR)-T cell therapy has been clinically validated as a curative treatment for the difficult to treat malignancies of relapsed/refractory B-cell acute lymphoblastic leukaemia and lymphoma. Here, the CAR-T cells are re-directed towards a single antigen, CD19, which is recognised as a virtually ideal CAR target antigen because it has strong, uniform expression on cancer cells, and is otherwise expressed only on healthy B cells, which are ‘dispensable’. Notwithstanding the clinical success of CD19-CAR-T cell therapy, its single specificity has driven therapeutic resistance in 30% or more of cases with CD19-negative leukaemic relapses. Immune checkpoint blockade is also a highly successful cancer immunotherapeutic approach, but it will be less useful for many patients whose malignancies either lack a substantial somatic mutation load or whose tumours are intrinsically resistant. Although CAR-T cell therapy could serve this unmet medical need, it is beset by several major limitations. There is a lack of candidate antigens that would satisfy the requirements for ideal CAR targets. Biological properties such as clonal heterogeneity and micro-environmental conditions hostile to T cells are inherent to many solid tumours. Past clinical studies indicate that on-target, off-tumour toxicities of CAR-T cell therapy may severely hamper its application. Therefore, re-designing CARs to increase the number of antigen specificities recognised by CAR-T cells will broaden tumour antigen coverage, potentially overcoming tumour heterogeneity and limiting tumour antigen escape. Tuning the balance of signalling within bi-specific CAR-T cells may enable tumour targeting while sparing normal tissues, and thus minimise on-target, off-tumour toxicities.
Recent US FDA approvals of tisagenlecleucel (Kymriah™) and axicabtagene ciloleucel (Yescarta™) for relapsed/refractory cases of acute B-cell lymphoblastic leukaemia (B-ALL) and adult large B-cell lymphoma, respectively, mean that chimeric antigen receptor (CAR)-T Cell technology has joined mainstream cancer treatment along with immune checkpoint inhibitory antibodies. However, unlike these antibodies, which empower endogenous cancer antigen-specific T cells with cancer fighting ability, CAR-T cells are allogeneic or autologous T cells, which are genetically engineered to have cancer-selective cytotoxicity. Indeed, CAR-T cell therapy may be particularly useful for the many cancers lacking a high somatic tumour mutation burden that is thought to confer susceptibility to immune checkpoint blockade .
The typical CAR is a transgenic construct, which encodes a cancer-selective ectodomain fused via a transmembrane domain to a T-cell-activating endodomain. The CAR transgene is expressed by a viral or non-viral vector, which is used to genetically modify a polyclonal population of T cells, each bearing its own unique T-cell receptor (TCR). Generally, a single-chain variable fragment (scFv) derived from a cancer antigen-specific monoclonal antibody confers cancer selectivity on the CAR-T cell ectodomain, although other target-recognition systems can also be used. The CD3ζ chain derived from the TCR is critically required for CAR-T cell activation (‘Signal 1’) and comprises the first-generation CAR endodomain . However, robust expansion and persistence in vivo of CAR-T cells requires the endodomain to additionally include either one or two co-stimulatory signalling domains (such as CD28, OX40 or 4-1BB; ‘Signal 2’), to produce second-generation [3–5] or third-generation CARs , respectively.
Hence, CAR-T cells unite the exquisite specificity of monoclonal antibodies with the cytotoxic potency of T cells and enable otherwise non-cancer-targeting T cells to be re-directed towards cancer cell targets. Several important consequences flow from this use of a ‘living drug’. Unlike conventional T-cell recognition of cancer cells, which depends on MHC expression and protein antigen processing often lost during immune evasion by cancer cells, CAR-T cells can bind target antigens independently of MHC, as well as antigens not normally recognised by T cells, such as the GD2 carbohydrate antigen . Like TCR-mediated cytotoxicity, CAR-T cells can serially engage and kill numerous target cells via the CAR [8–10]. And similarly to T-cell responses to infectious pathogens, a population of activated CAR-T cells expands and then contracts, leaving behind a (presumably) self-renewing sub-population of memory CAR-T cells [8,11]. However, CAR-T cells require a higher antigen density than native T cells to elicit a full set of effector functions .
Current limitations of clinical CAR-T cell technology
Factors affecting target antigen selection — tumour heterogeneity and antigen loss variants
The selection of an appropriate cancer antigen (or antigens) for the targeting of CAR-T cells is an important requirement for developing effective therapies. In point of fact, the remarkable success of CD19-targeting CAR-T cells for the treatment of B-cell malignancies in large part stems from the use of CD19 as a target antigen . This surface molecule is an excellent CAR-T cell target, because it is strongly and uniformly expressed by the cancer cells but shows very limited expression on healthy cells, being strictly limited to the B-cell lineage of the haematopoietic system. Hence, normal tissue toxicity is limited to a ‘dispensable’ cell type and the B-cell aplasia of treated patients is reversible and manageable with replacement immunoglobulin administered intravenously or subcutaneously. Other excellent CAR-T cell-target antigens may soon be validated for some additional haematological malignancies such as multiple myeloma, because of the tightly regulated patterns of surface marker expression within the haematopoietic system, which are exemplified by the cluster of differentiation (CD) marker panel. Nonetheless, for other haematological malignancies such as acute myeloid leukaemia, a single antigenic target may not suffice, and a specialised combinatorial targeting strategy may be required .
The example of CD19 teaches us that the ideal CAR target antigen is expressed at high level by all tumour cells in a large fraction of patients. At the same time, vital normal tissues should not express this antigen nor resting or activated T cells (ATCs) to avoid fratricide of the CAR-T cell population . By these criteria, ideal CAR target antigens are yet to be identified and clinically validated for most solid tumours, making the development of CAR T therapy for these diseases significantly more challenging . Indeed, tissue and genetic heterogeneity is an inherent feature of many solid cancers [15–17], and this is reflected in heterogeneity of tumour antigen expression; a significant factor limiting endogenous anti-cancer immunity . Thus, CAR-T cells directed towards one of these antigens will generally only recognise a sub-population of cancer cells, theoretically leaving the remainder to continue growing. However, evidence is emerging that the immune-mediated destruction and ensuing inflammation induced by CAR-T cells may lead to an ‘in situ vaccination’ effect, resulting in the priming of new immune responses and subsequent attack of any cancer cells spared by the original CAR-T cell preparation [19,20].
Although the single antigen specificity of clinically approved CD19-specific CAR-T cell therapies is a major advantage, it is also the current main limitation of clinical CAR-T cell technology. Hence, ∼30% of B-ALL patients who received CD19-specific CAR-T cells experienced relapse with variants of disease not expressing CD19 [21,22]. CAR-T cell therapy applies a strong immune selection pressure, which can cause cancer cells to evolve to lose expression of the targeted antigen and enable resistance to CAR-T cell therapy of single antigen specificity to develop [22–24]. This therapeutic resistance is not only an issue for CD19-targeting clinical CAR-T cell therapies, as noted above, but also hampers the development of new CAR-T cell therapies for other cancers. For example, a complete remission was induced in a patient with recurrent glioblastoma by intraventricular and intracavity administration of CAR-T cells specific for IL13Rα2 [23,25]. However, after 8 months, the patient relapsed with a glioblastoma variant lacking IL13Rα2 expression, suggesting that immune escape had occurred.
On-target, off-tumour toxicities
An additional challenge is that most potential CAR-T cell-target antigens for solid tumours show some level of expression on normal body tissues. This can lead to CAR-T cell-mediated attack of healthy tissues, resulting in on-target but off-tumour toxicity. For example, mesothelin is being actively explored as a CAR-T cell target in several solid tumours, but is also basally expressed on normal mesothelial cells of the peritoneal and pleural cavities and the pericardium . However, neither significant anti-tumour activity nor on-target, off-tumour toxicity have been observed so far in clinical trials of mesothelin-specific CAR-T cell therapy in patients with malignant pleural mesothelioma . Similarly, the CAR-T cell-target fibroblast activation protein (FAP) is strongly expressed in cancer-associated fibroblasts , but also shows lower level expression on bone marrow stromal cells and within other normal tissues such as skeletal muscle, adipose tissue and pancreas [29,30]. The significance of FAP expression by healthy tissues for CAR-T cell targeting is a matter of debate, with mouse tumour xenograft studies showing either minimal toxicity [31,32] or fatal cachexia  following administration of CAR-T cells specific for murine FAP. The difficulty of predicting off-tumour toxicity of CAR-T cells is exemplified by an early clinical trial of CAR-T cells targeting carbonic anhydrase-IX (CAIX) in patients with metastatic renal cell carcinoma, in which grade 2–4 liver toxicity developed following repeated CAR-T cell infusions . Biopsy of the affected patients' livers revealed that CAIX was expressed by bile duct epithelial cells, and the bile ducts were infiltrated with T cells, presumed to be transferred CAR-T cells.
Finally, a death after treatment with erbB2-specific CAR-T cells was attributed to on-target, off-tumour effects of the CAR-T cells on lung epithelial cells expressing low levels of erbB2, but a multi-factorial causation may also be possible. Before CAR-T cell administration, the patient, who had high-level expression of erbB2 on pulmonary and liver metastases of colorectal cancer, was conditioned with fludarabine and cyclophosphamide chemotherapy. Then, she was given an IV dose of 1010 T cells (79% CAR-T cells) concurrently with low-dose IL-2. Other cited studies had shown that 90% of infused T cells accumulate almost immediately in the lungs. Within 15 min of the CAR-T cell infusion, the patient developed acute respiratory distress syndrome, and 4 h later, cytokine release was demonstrated with acute high-level elevation of inflammatory cytokines including IL-6. The patient succumbed 5 days later to multi-organ failure . The risk of pulmonary toxicity with erbB2-targeted trastuzumab is significantly worsened in the presence of pulmonary metastases and may be aggravated further by concomitant IL-2 administration . Consequently, it would be expected that the IV administered erbB2-specific CAR-T cells, which localise immediately in the lungs, would directly target high erbB2-expressing pulmonary metastases in this patient, and the authors present evidence of such targeting. Conversely, on-target, off-tumour toxicity would depend on CAR-T cell targeting of weak erbB2-expressing lung epithelial cells in an abluminal location.
Solutions using combinatorial CAR-T cell targeting
To surmount the obstacle that ideal cancer antigens will probably not be found for most cancer types, and to address the limitations of current clinical CAR-T cell technology, investigators have often sought to add another antigen receptor to uni-specific CAR-T cells. Some of the proposed solutions adopt the rational design approach of using Boolean logic to ‘gate’ the operability of CAR-T cells according to ‘AND’, ‘OR’ and ‘NOT’ rules. Two separate CAR molecules, which target two distinct antigens, are usually assembled in the one T cell to facilitate this approach. Consequently, the arrangements could include two fully functional CAR molecules that both have CD3ζ and a co-stimulatory domain (OR gate); two CAR molecules that lack one or the other signalling domains (and so must complement each other for full T-cell function; AND gate) or a pair of CAR molecules with one inhibitory CAR molecule that antagonises the function of the other (NOT gate). An OR gate can also be achieved by pooling CAR-T cells of multiple specificities (a CARpool). These commonly used approaches may overcome the problems of tumour antigen heterogeneity and antigen escape, as well as on-target, off-tumour toxicity (Figures 1 and 2).
Types of bi-specific CAR-T cells designed to mitigate tumour antigen heterogeneity and escape.
Types of bi-specific CAR-T cells designed to mitigate on-target, off-tumour toxicity.
Dual CAR-T cells: bi-specific T cells transduced with two CARs, each mediating Signals 1 and 2
Dual CAR-T cells contain two individual CAR constructs, each recognising a different target and each containing both Signals 1 and 2 (Figure 1A). These T cells are engineered to be activated when either antigen is present (OR gate) and can be employed to broaden cancer recognition. This, in turn, can help to overcome antigen heterogeneity and also limit the emergence of tumour antigen escape variants. For example, dual CAR-T cells simultaneously targeting HER2 and IL-13Rα2 were shown to offset antigen escape in vitro and showed improved tumour control in a U373/SCID xenograft mouse model of glioblastoma compared with either single targeting CAR-T cells or pooled CAR-T cells . More recently, members of the same research group showed that, in comparison with uni-specific and bi-specific CAR-T cells, tri-specific CAR-T cells extend tumour antigen coverage to almost 100% of patient-derived xenografts from their glioblastoma cohort with corresponding improvements in survival of tumour-bearing mice . Similarly, dual CAR-T cells combining CD19 and CD123 specificity provided superior in vivo activity compared with pooled CAR-T cells in a mouse model engrafted with either a B-ALL cell line or primary B-ALL blasts .
CAR-T cell targeting of multiple tumour antigen combinations
Another method that may minimise the development of tumour antigen escape has been described and is being evaluated clinically in head and neck cancer patients (clinicaltrials.gov: NCT01818323). In this approach, a novel fusion protein of the epidermal growth factor ligand, which is called T1E, has been used as the ectodomain of a CAR with a CD28- and CD3ζ-containing endodomain. The T1E ligand of this CAR can bind multiple homo- and heterodimeric forms of erbB family members, including erbB1 and erbB2/erbB3, which have significant oncogenic properties . It is known that therapeutic monoclonal antibody targeting of single erbB family members such as erbB2 can be circumvented because of recruitment of other erbB family members into the intratumoral signalling network. Therefore, the ability of a single CAR to target multiple tumour antigens simultaneously may allay the development of therapeutic resistance.
Pooling CAR-T cells of single specificities (a CARpool)
Pooled single-specificity CAR-T cells can be particularly useful for targeting two antigens, which are expressed on distinct cell types within cancers to enhance the anti-cancer effect. For example, CAR-T cells targeting FAP have been mixed with those targeting EphA2 to enable simultaneous attack of A549 lung cancer cells (EphA2+) and their supporting stroma (FAP+ cancer-associated fibroblasts), resulting in significantly enhanced anti-cancer efficacy compared with either CAR-T cell. Nonetheless, a CARpool of two antigen specificities was less effective than dual CAR-T cells having the same specificities .
Tandem CAR-T cells: bi-specific CAR-T cells expressing scFvs in tandem
Tandem CAR (TanCAR) T cells, similar to dual CAR-T cells, also provide OR gate logic and broaden cancer recognition. However, in the case of TanCAR T cells, the T cells are engineered with a single CAR, which contains two scFvs in tandem; binding of either scFv to its cognate target is sufficient to activate the T cells (Figure 1B). The first TanCAR was developed by Nabil Ahmed's group, who showed that effector functions of the engineered T cells were synergistically enhanced when both targets were encountered simultaneously, highlighting the advantage of this approach . In support, TanCAR targeting of HER2 and IL13Rα2 could mitigate cancer antigen escape and showed superior anti-cancer activity compared with the equivalent dual CAR in the U373/SCID mouse model mentioned above . Moreover, TanCAR T cells simultaneously recognising both cancer antigens killed tumour cells more effectively than dual CAR-T cells recognising the same antigens. The increased anti-tumour efficacy of TanCAR T cells may be related to at least two factors. Compared with the merely additive cytokine secretion by dual CAR-T cells, TanCAR T cells displayed super-additive cytokine production and also had more efficient immunological synapse formation marked by higher density bi-specific clustering of the target antigens . A recently developed CD19/CD20-specific TanCAR was shown to effectively eradicate tumours and prevent antigen escape in a Raji B-cell tumour model, thus highlighting the potential of this approach for improving clinical CD19-targeting CAR-T cell therapies .
Split-receptor CAR-T cells: bi-specific CAR-T cells with Signals 1 and 2 engineered into different CARs
A major concern of CAR-T cell therapy is on-target, off-tumour toxicity resulting from the recognition of targets on normal tissues. Although incorporation of a suicide gene in CAR-T cells  or loco-regional administration of CAR-T cells [23,25,44] may both reduce the risk of normal tissue toxicity, specific CAR designs employing a more specific activation threshold could also obviate this problem. Split-receptor CAR-T cells meet this requirement (Figure 2A). Similar to dual CAR-T cells, split-receptor CAR-T cells also contain two CAR constructs recognising different antigens. However, in the case of split-receptor CAR-T cells, Signals 1 and 2 are split across the CARs such that one CAR contains the CD3ζ endodomain, while the other contains the co-stimulatory endodomain; this latter CAR is sometimes referred to as a co-stimulatory CAR. Thus, the T cells are fully activated only when both antigens are present (AND gate). This improves cancer specificity and minimises on-target, off-tumour toxicity against normal tissues, which may express one, but not both, of the antigens. Split-receptor CAR design is particularly relevant for non-haematological malignancies, which generally do not provide their own co-stimulatory signals [45,46].
Wilkie et al. developed a CAR-T cell that co-expresses erbB2- and MUC1-specific CARs with CD3ζ and CD28 signalling domains, respectively. They found that the bi-specific CAR-T cells proliferated in a manner that required co-expression of MUC1 and erbB2 by target cells. However, the cytolytic activity was only erbB2 dependent, irrespective of MUC1 expression . Lanitis et al. developed a bi-specific CAR T approach targeting mesothelin and α-folate receptor with complementary CD28 and CD3ζ signalling. The animal model showed that the bi-specific CAR-T cells killed double-positive tumour cells as efficiently as second-generation uni-specific CAR-T cells, but were more tolerant to single-positive tumours in vivo, suggesting improved specificity . Of note, in both studies, the bi-specific CAR-T cells were not strictly limited to AND gate killing. Thus, some level of cytotoxic activity was observed against targets expressing only the antigen recognised by the CAR with CD3ζ signalling. However, a third study by Kloss et al.  showed that by tuning the binding affinity of the CD3ζ-signalling CAR, true AND gate specificity can be achieved in split-receptor CAR-T cells, suggesting that the balance between Signals 1 and 2 is critical in this situation.
The split-receptor NOT-gated CAR-T cell represents a complementary approach to the AND-gated design described above and is also intended to prevent on-target, off-tumour toxicity to normal tissues in an antigen-restricted manner. To demonstrate preclinical proof of concept for this strategy, Sadelain and colleagues employ a model antigen system, which is not directly relevant to human disease. In the present study, the CAR-T cell, which is engineered using a NOT gate, also incorporates two CARs each recognising different antigens. But one CAR contains the T-cell-activating CD3ζ endodomain, whereas the second CAR contains a co-inhibitory rather than a co-stimulatory endodomain. Hence, the CAR-T cells will only attack tumour cells bearing a single antigen but not normal cells co-expressing a second antigen recognised by the paired inhibitory CAR. The normal cells are protected because the co-inhibitory endodomain of the second CAR derives from the native immune checkpoint molecule, PD1, which negatively regulates the first CAR molecule but only for the duration of T-cell-target cell contact. This feature of the NOT-gated CAR-T cell means that signalling through the second inhibitory CAR is temporary and reversible, thus allowing the CAR-T cell to become re-activated in the presence of tumour cells  (Figure 2B).
CAR-T cell targeting in combination with native TCRs or as a regulatable synthetic receptor
Other AND-gated ways to avoid on-target, off-tumour effects of a CAR-T cell include coupling either a single T-cell activation domain or a single co-stimulatory domain of a CAR with the T cell's own TCR, which separately binds its cognate antigen. Three such split-receptor approaches have been used with varying success.
In the first, allogeneic T cells were genetically modified with a CAR transgene, which encoded an ectodomain comprising an scFv specific for the α-folate receptor expressed by ovarian cancer cells and a T-cell endodomain consisting of the signalling chain of the Fc receptor γ chain. After adoptive cell transfer to ovarian cancer patients, the allogeneic TCRs of the transferred T cells were further stimulated in vivo by immunising the patients with allogeneic PBMCs of the same donor origin. Although tumour localisation by the CAR-T cells was demonstrated, there was little if any anti-tumour activity apparent using this approach .
In the second, autologous CAR-T cells specific for the neuroblastoma antigen, GD2, were generated from two sources: ATCs and Epstein–Barr virus-specific cytotoxic T lymphocytes (EBV-CTLs), which were infused together into paediatric patients with neuroblastoma. The GD2-specific CAR transgene encoded only a CD3ζ T-cell endodomain and also contained one of two different DNA tags, which could be distinguished by PCR depending on whether the CAR transgene was integrated in an ATC or an EBV-CTL . It was hypothesised that the EBV-specific TCR would provide physiological co-stimulation for the GD2-CAR. Although this GD2-CAR-T cell therapy was associated with several tumour regressions and long-term persistence of CAR-T cells with a memory phenotype, the GD2-CAR-expressing EBV-CTLs were not evidently superior to the ATCs in this study [11,52].
In the third, a GD2-specific CAR containing only the NKG2D adaptor, DAP10, as the co-stimulatory molecule was expressed in Vγ9Vδ2+ T cells, which recognise transformed phosphoantigen-expressing cells in an MHC-unrestricted manner via their γδTCR. The authors show that full cytolytic and cytokine effector activity in vitro of the GD2-CAR-bearing γδT cells did not occur unless both the CAR and γδTCR are engaged. The authors anticipate that this precise degree of control of T-cell activation will be achieved in vivo and will be sufficient to target GD2-expressing malignant cells while sparing GD2-expressing normal neural cells that lack phosphoantigen expression .
Finally, the preclinical development of a regulatable, synthetic receptor called synNotch raises genetic engineering of T cells to a new level of sophistication. In addition to the customisation of T-cell signalling inputs provided by the CARs themselves, this technology also enables versatile customisation of T-cell outputs so that, upon activation by CAR signalling, T cells can be made to elaborate specific patterns of cytokines or even therapeutic effector molecules such as bi-specific T-cell engagers. The synNotch receptor couples the antigen-binding element of an antigen targeting scFv to the transmembrane Notch core regulatory region. In turn, this Notch domain controls proteolytic intracellular release of a customised transcription factor, which leads to activation of the in-T-cell engineered transcriptional programme .
The crucial role of signal transduction in CAR-T cells
By the inclusion of additional intracellular signalling domains, second- and third-generation CAR molecules deliver progressively more potent activation signals compared with first-generation CARs and may thus bestow improved effector functions such as proliferation, survival, production of cytokines and anti-cancer activity [5,55,56]. However, the signalling events that follow CAR engagement, and the interaction between elements of the CAR and endogenous signalling molecules, remain poorly understood.
The classical immunological synapse involves clustering of multiple TCRs/ligands at the point of contact between a T cell and an antigen-presenting cell. Within 10 min of TCR engagement, two concentric rings of molecules form, consisting of a central cluster of CD3/TCR molecules surrounded by accessory signalling molecules including CD28 . There is a 24–72 h period of endogenous co-stimulation after initial ligation of the native TCR when cell-surface molecules such as 4-1BB and OX40 are transiently expressed (because of proteolytic shedding) and initiate signalling [58,59]. Furthermore, after TCR engagement, T cells have been shown to be highly sensitive to both the type and duration of co-stimulatory signals received from receptors and cytokines, which are integrated to determine the rate and magnitude of subsequent T-cell expansion .
In contrast, T cells have more rapid kinetics of activation, killing and disengagement when activated through the CAR rather than the native TCR. In addition, CAR activation results in alterations of immunological synapse formation . The accelerated timing of CAR-T cell activation may not be unexpected, given that a second-generation CAR molecule combines in cis configuration the CD3ζ and CD28 or 4-1BB intracellular signalling domains in a single fusion protein [55,61]. In third-generation constructs, co-stimulatory molecules such as 4-1BB or OX40 (CD134) are added to CD3ζ and CD28 and are triggered simultaneously with CAR engagement [5,58–60].
We need to have a better understanding of the functional contributions of different kinds of intracellular signalling domains and how their incorporation in the unique CAR structure affects CAR-T cell biology. Although CD28 confers potent immediate effector functions and cancer killing, the inclusion of 4-1BB instead of CD28 in CARs results in improved persistence in vivo [56,62]. Distinct differences in metabolic processes have also been reported for the different co-receptor domains. Thus, CD28 promotes glycolysis and 4-1BB promotes fatty acid oxidation, increased respiratory capacity and a central memory phenotype, which may contribute to the improved persistence observed for 4-1BB-containing CARs in vivo [7,63–65]. For third-generation CAR-T cells incorporating CD28 and OX40 domains, constitutive phosphorylation of signalling molecules such as ERK and AKT indicates that CAR expression leads to basal levels of activation in these cells . Although the CD28 and OX40 signalling combination enhances CAR-T cell activation and effector functions, it may also result in T-cell exhaustion or terminal differentiation, which is characterised by constitutive PD1 expression and loss of cytokine production, or deletion via an activation induced cell death pathway [7,64,65].
In some cases, tonic signalling of CAR molecules has been reported whereby the extracellular scFv domains cluster and lead to CD3ζ phosphorylation in the absence of antigen, eventually resulting in functional exhaustion or deletion of CAR-T cells . Initially, this process was noted for GD2-specific, CD28-containing, second-generation CARs ; however, recently, this effect has also been reported for a 41BB-containing CAR with either GD2 or CD19-specific scFvs . In this study, it was found that the co-stimulatory domains and expression system used were responsible for tonic signalling and CAR-T cell apoptosis. In particular, the long terminal repeat (LTR) promoter of the gamma-retroviral vector resulted in high-level CAR surface expression and amplified the tonic signalling effect. Reducing CAR surface expression either by an internal ribosomal entry site or expression from a non-LTR promoter in a lentiviral vector reversed this effect. Others have also reported that a high density of CAR expression on the T-cell surface, regardless of the co-stimulatory molecules selected, may lead to excessive levels of T-cell activation, resulting in exhaustion and deletion of CAR-T cells .
Summary and future perspectives
Given the limitations of current CAR-T cell technology, successful clinical implementation of this promising treatment for solid malignancies, which do not respond well to checkpoint immunotherapy, will depend on broadening tumour antigen coverage to lessen antigen escape while avoiding on-target, off-tumour toxicity. Recent preclinical findings show ways in which the clinical utility of CAR-T cell therapy might be extended. These advances include new CAR designs that broaden the range of simultaneous CAR-T cell targeting to at least two tumour antigens, so that CAR-T cell effectiveness can be increased by targeting more tumour cells while reducing the chance of tumour antigen escape. Conversely, some of these dual CAR designs may reduce the vexatious problem of on-target, off-tumour toxicity of CAR-T cell therapy.
In particular, logic-gated approaches may enable some of these competing clinical demands to be reconciled. For near ideal tumour antigens, multi-targeted CARs, such as TanCARs, the T1E CAR and OR-gated CARs, may extend antigen coverage without comprising safety. For less than ideal tumour antigens, which have significant normal tissue expression, the conditional T-cell signalling afforded by split-receptor CARs may ensure that full T-cell activation only occurs when CAR-T cells encounter both tumour antigens on malignant cells in the case of AND-gated CARs and only the one tumour antigen on malignant cells for NOT-gated CARs.
Notwithstanding the elegance of some of these CAR designs, it is evident that careful and rigorous preclinical testing will be required to achieve the desired balance of signalling, allowing for anti-tumour effects while sparing normal tissues. An ultimate aim of this genetic engineering technology is precise and tuneable control of antigen receptor activation that preserves potency and breadth of targeting while maximising safety.
Nevertheless, many questions remain about optimal dual CAR-T cell designs. Although early data may be informative, the answers may rest on a better understanding of the nature of T-cell activation by CARs versus native TCRs, including their respective activation kinetics, interactions with endogenous signalling moieties and detailed studies of their related immunological synapses. T cells are well suited to integrate multiple inputs so chief among the questions is whether the timing of activation of each CAR in a bi-specific CAR-T cell can have a productive output when their respective cognate antigens are geographically separated; a situation often observed in heterogeneous tumours.
activated T cells
B-cell lymphoblastic leukaemia
chimeric antigen receptor
cluster of differentiation
Epstein–Barr virus-specific cytotoxic T lymphocytes
fibroblast activation protein
long terminal repeat
single-chain variable fragment
We acknowledge support from the South Australian Beat Cancer Project Hospital Research Support Package.
The Authors declare that there are no competing interests associated with the manuscript.