Adoptive cell transfer using chimeric antigen receptors (CAR) has emerged as one of the most promising new therapeutic modalities for patients with relapsed or refractory B-cell malignancies. Thus far, results in patients with advanced solid tumors have proven disappointing. Constitutive tonic signaling in the absence of ligand is an increasingly recognized complication when deploying these synthetic fusion receptors and can be a cause of poor antitumor efficacy, impaired survival, and reduced persistence in vivo. In parallel, ligand-dependent tonic signaling can mediate toxicity and promote T-cell anergy, exhaustion, and activation-induced cell death. Here, we review the mechanisms underpinning CAR tonic signaling and highlight the wide variety of effects that can emerge after making subtle structural changes or altering the methodology of CAR transduction. We highlight strategies to prevent unconstrained tonic signaling and address its deleterious consequences. We also frame this phenomenon in the context of endogenous TCR tonic signaling, which has been shown to regulate peripheral tolerance, facilitate the targeting of foreign antigens, and suggest opportunities to coopt ligand-dependent CAR tonic signaling to facilitate in vivo persistence and efficacy. Mol Cancer Ther; 17(9); 1795–815. ©2018 AACR.
This article is featured in Highlights of This Issue, p. 1793
Adoptive cell transfer (ACT), utilizing autologous T cells engineered to express chimeric antigen receptors (CAR), has proven to be a highly efficacious strategy for the management of patients with relapsed or refractory B-cell malignancies (1–3). Indeed, following the recent FDA approvals of the second-generation CD19-directed autologous CAR T-cell products: tisagenlecleucel (tradename KYMRIAH) for the management of pediatric and young adult patients with B-cell acute lymphoblastic leukemia (ALL; refs. 4, 5) and axicabtagene ciloleucel (tradename YESCARTA) for adult patients with relapsed or refractory large B-cell lymphoma following two or more lines of systemic therapy (6), CAR T-cell therapy is now a standard of care and can no longer be regarded as a purely experimental therapeutic modality. However, the field remains in its infancy and these great strides are yet to be replicated in patients with advanced solid tumors (7–9). Much work remains to be undertaken to more fully appreciate how CAR structure determines function and delineate the complexity of CAR intracellular signaling as well the web of interactions between CAR T cells and other protagonist cells within the tumor microenvironment (TME) in vivo. Considerable effort continues to be applied to the optimization of the CAR construct itself to enhance antitumor potency, metabolism, proliferative capacity, and persistence (10, 11). It is becoming increasingly apparent that subtle differences in CAR design can have amplified effects both in vitro and particularly in vivo and that the optimal selection of the CAR's extracellular targeting moiety, hinge, spacer, transmembrane domain (TMD), and intracellular costimulatory domain(s) (ICD) is crucial.
It has become evident since the 1990s that nonactivated basal state T cells (and indeed B cells) exhibit low-level constitutive tonic signaling that is able to regulate their function and survival in a homeostatic manner (12–14). More specifically, it is now understood that T-cell receptor (TCR)–mediated tonic signaling in nonengineered naïve endogenous T cells, mediated by routine nonantigen-specific interactions with mature antigen-presenting dendritic cells (DC), is able to enhance their subsequent ability to react to foreign peptides (such as tumor neoantigens; refs. 12, 13). This is controlled, at least in part, by interactions between the TCRs of naïve T cells and self-peptide presented on MHC molecules expressed on the surface of DCs and appears to be an important physiologic mechanism to ensure the homeostatic control of T-cell tolerance in the periphery (15, 16). Despite considerable progress in understanding the molecular events involved in B-cell receptor (BCR)-mediated tonic signaling, which is a regulator of B-cell maturation and survival (14, 17), our understanding of TCR-mediated T-cell tonic signaling, which shares many of the hallmarks of the former, remains poorly defined (14).
CAR tonic signaling, however, may be defined as a noncoordinated and sustained activation of the T cells in either a ligand-independent or dependent fashion. In the absence of spatial and/or temporal control of CAR cell surface expression, this constitutive or chronic cell signaling may have a substantial deleterious impact on CAR T-cell effector function and survival, and may lead to a significant disparity between in vitro cytolytic capacity and in vivo antitumor efficacy (18–21). This review highlights the current research being undertaken to identify and address CAR tonic signaling in all its forms, drawing attention to data that are at times conflicting and hypothesis-generating. At least four major overlapping patterns of ligand-independent CAR tonic signaling are presented and a variety of strategies designed to ameliorate the negative consequences of these are expounded. Finally, through the prism of endogenous T-cell tonic signaling and its important regulatory role in immune tolerance and cell-mediated adaptive immunity, we posit a number of hypothetical strategies designed to harness the potential benefits of CAR tonic signaling to improve CAR T-cell antitumor efficacy and in vivo persistence.
Conventionally designed CARs exploit the specificity of an antibody-derived extracellular binding domain while harnessing the effector and memory capacity of T cells to target tumors (22). CAR T cells may thus deliver the promise of “living drugs,” capable of targeting tumor-associated or tumor-specific antigens (TAAs or TSAs) over a prolonged period of time (23). Given that CARs function in the absence of HLA/TCR interactions, they have considerable applicability across patient groups and are ideally placed to address the growing problem of acquired resistance to immune checkpoint inhibition due to disrupted antigen processing and/or presentation (24). Furthermore, with the advent of allogeneic HLA and TCR-edited CAR T cells, the potential exists for scalable “off the shelf” delivery, potentially in combination for optimized TAA pattern recognition (25, 26).
CAR design has undergone a number of iterative developments over the last two decades, with the aim of optimizing CAR T-cell effector function and persistence (27). First-generation CARs or “T-bodies” linked an extracellular antibody-derived recognition moiety to a lymphocyte-stimulating domain, such as the signal-transducing subunit of either the immunoglobulin receptor (FcγR) or the TCR CD3ϵ or CD3ζ chains (28). First-generation CAR T cells tended to elicit only weak antitumor activity and were highly prone to anergy (29). The fusion of costimulatory ICDs with the cytoplasmic tail of CD3ζ-containing first-generation constructs has led to the emergence of second-generation [comprising a single costimulatory ICD such as CD28 (30), 4-1BB (CD137; ref.31), inducible T-cell costimulator (ICOS); ref. 32, OX40 (33), CD27 (34), or DNAX-activating protein 10 (DAP10); ref. 35] and third-generation CARs [comprising multiple costimulatory ICDs, aligned in cis (36, 37)]. Incorporation of costimulatory ICDs can recapitulate signal 2 required for T-cell activation, leading to enhanced effector function, proliferation, survival, and ultimately enhanced tumor killing (38). Fourth-generation CAR T cells (termed “TRUCKs”) containing CAR-inducible transgenes and “armored CARs” capable of constitutively producing cytokines (such as IL12, IL15, and IL18) in secreted or membrane-tethered form have been engineered to recapitulate signal 3 in an autocrine and paracrine manner (39–42). These designs are illustrated in Fig. 1. Further modifications have been explored with respect to the CAR TMD (43) and hinge/spacer region (20, 44). The extracellular targeting moiety, which has typically constituted an antibody-derived single-chain variable fragment (scFv), may alternatively comprise an endogenous receptor or ligand (9). Antitumor efficacy relies upon optimal CAR binding to the target epitope and the formation of a cytolytic immune synapse between the CAR T cell and the target cell. Spacer length, which impacts upon both the flexibility of the CAR (45) and the distance (46) between the target cell and the CAR T-cell membrane, is increasingly seen as critical in ensuring optimal immune synapse formation, particularly with regard to membrane-proximal epitopes (44).
Endogenous TCR tonic signaling
Maintenance of naïve T cells in the periphery following their release from the thymus is maintained by tonic signaling via the TCR and common γ chain cytokine receptors (15, 16). Specifically, the survival of naïve CD4+ and CD8+ T cells in the periphery relies upon a combination of low- to intermediate-affinity binding of the TCR to MHCs loaded with self-peptides presented on the surface of DCs and the presence of IL7 on the surface of fibroblastic reticular cells (FRC) in the T-cell zone of secondary lymphoid organs (16). Naïve CD8+ T cells are also partly reliant on JAK-STAT signaling mediated by the engagement of IL15 receptors with DC-derived IL15 (47). Steady-state DC-mediated TCR tonic signaling enhances T-cell responsiveness to MHC-associated foreign antigen (12), but does not necessarily induce a transition to a central memory phenotype (48). T-cell hyporesponsiveness has been shown to be associated with lower baseline phosphorylation in proximal TCR events, for example, reduced basal phosphorylation of zeta-chain-associated protein kinase 70 (ZAP70)–associated CD3ζ (13). Some baseline tonic signaling in naïve T cells may reflect constitutive activation of Lck, a member of the SRC family kinase (SFK) that plays a pivotal role in TCR signaling (49), maintaining a basal level of phosphorylation on TCR-associated CD3ζ-chain immunoreceptor tyrosine-based activation motifs (ITAMs; ref.50). This process is also regulated by a highly dynamic interplay between the receptor-like tyrosine phosphatase CD45 and the protein tyrosine kinase Csk (51, 52). The importance of this interaction is illustrated by the fact that DC depletion results in rapid loss of T-cell responsiveness to cognate antigen, rapidly reversed with the restoration of T-cell/DC interactions (13). Similarly, following the exposure of mice to MHC class II blocking antibodies, a loss of basal CD3ζ chain phosphorylation is observed (53). In addition to complementary effects mediated by the engagement of leukocyte β integrins, such as lymphocyte function–associated antigen 1 (LFA-1) with cell-adhesion molecules, such as intercellular adhesion molecule 1 (ICAM-1; ref.54, 55), low-affinity interactions between the TCR and MHC (including monomeric MHC) appear to lower T-cell activation threshold by replenishing intracellular Ca2+ stores and increasing plasma membrane phosphatidylinositol 4,5-bisphosphate (PIP2; ref. 56).
MHC class II interactions, in particular, have been linked to maintaining T-cell reactivity and proliferative capacity following activation to cognate antigen (50, 57). This may occur through a two-step process in the lymphoid tissue, whereby naïve T cells with TCRs exhibiting low to intermediate affinity for self-peptides presented on steady-state DCs may induce TCR tonic signaling. This is enhanced further by cross-presentation of foreign peptide by activated DCs (accompanied by costimulatory interactions between B7/CD28 and CD70/CD27) leading to enhanced T-cell effector function (12). This may be leveraged further by the “pseudodimer” effect (postulated for CD4+ T-cell/MHC class II interactions), whereby the concomitant recognition of MHC-loaded self-peptides and foreign peptides enhances T-cell responsiveness against the latter (58). In parallel, peripheral tolerance is maintained by high-affinity binding of TCR to self MHCs leading to unconstrained tonic signaling, T-cell tolerance, and exhaustion [mediated by upregulation of inhibitory checkpoints such as CTL-associated protein 4 (CTLA-4) and programmed death 1 (PD-1)], anergy, apoptosis, and/or enhanced regulatory T-cell (Treg) functionality. This model is illustrated in Fig. 2 and demonstrates how TCR tonic signaling may contribute to a dynamic equilibrium between positive and negative regulators of T-cell efficacy and autoimmunity. Such a paradigm is analogous to thymic T-cell selection, whereby intermediate TCR and MHC affinity is positively selected and induces TCR tonic signaling required for subsequent reactivity to foreign peptides in the periphery (12).
In contrast to CAR-mediated tonic signaling, low-level TCR tonic signaling appears to be spatially compartmentalized to the lymphoid tissue. T cells isolated from peripheral blood fail to demonstrate the basal CD3ζ chain phosphorylation (53) and it has been proposed that TCR tonic signaling may occur in a cyclical fashion as T cells enter and exit lymphoid organs from the circulation (14). Furthermore, TCR tonic signaling appears to have a differential impact upon maintenance and survival of CD4+ versus CD8+ T cells. Specifically, self-recognition of MHC class I by peripheral CD8+ T cells appears to be far more crucial for their survival [supported by their significantly reduced half-life following TCR ablation (59)] than recognition of MHC class II by peripheral CD4+ T cells, which are able to undergo homeostatic proliferation as measured by bromodeoxyuridine (BrdU) incorporation in recombination activating gene 2 (Rag2)−/− class II−/− transgenic mice (60). With the emergence of validated surrogate markers of TCR tonic signaling, we are gaining greater insight into the mechanistic basis of this process, illuminating the multitude of downstream pathways that impact upon T-cell function, differentiation, and survival (14). Examples include T-cell surface expression of CD5 (a negative regulator of TCR signaling), which correlates with self–MHC interactions and basal TCR tonic signaling; or expression of nuclear receptor subfamily 4 group A member 1 (Nr4a1) [an immediate-early transcription factor encoding nuclear hormone receptor 77 (Nur77)], which is rapidly upregulated by TCR stimulation in thymocytes and T cells. Similar approaches are likely to prove insightful when applied to CAR tonic signaling pathways.
Revealing a further layer of complexity to T-cell tonic signaling, it has been shown that TCR expression (specifically the TCRα chain) may itself be subject to tonic signaling mediated by basal activity through the linker for activation of T cells (LAT) - diacylglycerol (DAG) - RAS guanyl-releasing protein 1 (Rasgrp1) pathway (61). Studies involving transgenic mice with Rasgrp1 deficiency or mutation (Rasgrp1Anaef) have revealed evidence of basal tonic signaling via the mTOR pathway in the absence of TCR-ligand binding (14). Rasgrp1Anaef mice were found to uniformly express elevated levels of the activation marker CD44 on all CD4+ T cells [irrespective of CD62L expression, a marker of naïve, stem cell memory (TSCM), and central memory T cells (TCM)] and exhibited enhanced basal phosphorylation of the ribosomal protein S6, a downstream target of mTOR. These animals exhibited enhanced T-cell autoreactivity and autoimmunity and it is interesting to speculate whether this impact of elevated tonic signaling can be coopted for the development of effective CAR T cells with a nonterminally differentiated phenotype.
Ligand-independent CAR tonic signaling
A number of CAR constructs have been shown to elicit prolonged exponential expansion, constitutive cytokine release, and progressive differentiation to an effector phenotype in the absence of ligand, exogenous cytokines, or feeder cells (21). This appears, at least in part, due to the level of CAR surface expression achieved as well as the specific characteristics of the individual scFvs utilized, with those designed to target the disialoganglioside GD2, c-mesenchymal–epithelial transition (c-Met), and mesothelin featuring repeatedly in the literature concerning ligand-independent expansion (18–21), whereas CD19-targeting scFvs, such as FMC63, appear to be relatively resistant to this phenomenon (18, 21). In a study by Frigault and colleagues, ligand-independent tonic signaling leading to continuous T-cell expansion ex vivo was shown to be dependent upon the integration of the CD28 transmembrane and cytosolic domain within the CAR construct (21). Other members of the CD28 immunoglobulin superfamily, such as ICOS, did not appear capable of inducing constitutive expansion when substituted for CD28 in otherwise similar CAR constructs, and while utilization of a 4-1BB costimulatory ICD appears to confer enhanced ligand-independent proliferation (62), continuous expansion and constitutive cytokine release have not been demonstrated (11), although more recent reports described later in this review muddy the water somewhat by highlighting alternative mechanisms for 4-1BB–mediated tonic signaling, characterized by cell death rather than proliferation (19, 63).
Frigault and colleagues evaluated a set of 12 CARs designed to target c-Met, mesothelin, and CD19 (21). These contained either an immunoglobulin G4 (IgG4) hinge or CD8α stalk, coupled with CD28, ICOS, or CD8α transmembrane domains. The intracellular signaling domains comprised either CD28, 4-1BB, or ICOS bound in cis with CD3ζ. A lentiviral vector was utilized with an elongation factor 1 alpha (EF-1α) promoter. As expected, following in vitro activation with anti-CD3/anti-CD28–loaded beads and subsequent viral transduction, the majority of CAR T cells demonstrated a predictable pattern of rapid initial proliferation followed by a return to a resting state in the absence of exogenous IL2. Intriguingly, however, certain CAR constructs demonstrated continued expansion for up to 60 to 90 days in the absence of IL2 or target ligand. These included a c-Met–directed IgG4 28ζ CAR and both mesothelin-directed SS1 IgG4 and CD8α 28ζ CARs. Of note, of the c-Met–directed CARs neither the CD8α 28ζ, IgG4 BBζ, IgG4 ICOSζ, or first-generation IgG4 CD3ζ CARs exhibited this continuous activation phenotype. Continuous expansion of both CD4+ and CD8+ T cells transduced with c-Met IgG4 28ζ CARs was observed and was associated with a 100- to 1,000-fold increase in various cytokines [including IFNγ, TNFα, IL2, IL4, IL13, IL3, and granulocyte-macrophage colony-stimulating factor (GM-CSF)] as well as elevated levels of granzyme B and perforin. “Continuous” CARs were also characterised by significantly enhanced expression of the master transcription factors T-box transcription factor 21 (TBX21; encoding T-bet), EOMES (encoding eomesodermin), and GATA-3, as well as early enhanced expression of antiapoptotic proteins such as B-cell lymphoma-extra large (Bcl-xL). A pattern of sustained signal transduction protein activation was also identified, involving Akt (pS473), ERK1/2 (pT202 and pY204), and nuclear factor (NF)-κB [p65 (RelA) pS529] as well as reciprocal downregulation of endogenous CD28. High CAR surface expression also appeared key, as use of a cytomegalovirus (CMV) or variably truncated phospho-glycerate kinase (PGK) promoters led to both reduced CAR surface expression and a noncontinuous phenotype. While alloreactivity may have been a confounding factor in their in vivo model, c-Met IgG4 28ζ CARs encoded downstream of the shortest PGK promoter (PGK100) outperformed their more highly expressed EF-1α counterparts in terms of antitumor efficacy and persistence in NOD SCID γcnull (NSG) mice implanted with human ovarian cancer cell line (SK-OV3)-derived xenografts. This is reminiscent of later results published by Eyquem and colleagues (64) and Hale and colleagues (65) regarding the targeted expression of CARs to the T-cell receptor alpha constant (TRAC) locus. This phenotype of CAR tonic signaling is summarized in Fig. 3A.
Long and colleagues have subsequently shown that antigen-independent clustering of CAR scFvs is seen in second-generation γ-retrovirally transduced GD2 28ζ CARs incorporating a 14g2a-derived scFv with an IgG1-derived hinge and CH2-CH3 spacer, leading to chronic CAR CD3ζ domain phosphorylation, CAR T-cell exhaustion, and increased rates of apoptosis (18). This was shown to occur during anti-CD3/anti-CD28 bead-based ex vivo CAR T-cell expansion and was associated with an increase in cellular volume, CD25 upregulation, and an exhausted phenotype indistinguishable from exhausted nonengineered T cells in the context of chronic viral infection and cancer. An important mechanism appears to relate to the propensity for 14g2a (and, to a greater or lesser degree, other scFvs, and antibody fragments studied, e.g., targeting CD22 and ErbB2) to oligomerize, resulting in cell surface CAR clustering, visualized using functional CAR-fluorescent protein fusion constructs. The effect was found to be related specifically to the nonantigen-binding framework regions within the 14g2a scFv rather than the CAR's linker peptide or spacer domain. GD2-directed CAR tonic signaling-mediated T-cell exhaustion was found to be associated with a transcriptional profile favoring the expression of numerous inhibitory receptors, including PD-1, lymphocyte-activation gene 3 (LAG-3), T-cell immunoglobulin and mucin-domain containing-3 (TIM-3), CTLA-4, B-, and T-lymphocyte attenuator (BTLA) and 2B4 (CD244); the helix-loop-helix (HLH) apoptosis–associated protein ID-1; as well as recognized exhaustion-associated transcription factors, such as T-bet, EOMES, Blimp-1, and Helios. These CAR T cells demonstrated poor proliferative capacity, cytokine production, and antitumor efficacy in vivo. Elaborating on the data of Frigault and colleagues (21), Long and colleagues were able to demonstrate that costimulation with CD28 augmented CAR T-cell exhaustion, mediated by tonic signaling, whereas 4-1BB costimulation was able to limit it (18). This finding, alongside data highlighting differences between CD28 and 4-1BB ICDs with regard to CAR T-cell responsiveness to hypoxia, oxidative metabolism, and negative regulation of apoptosis as well as data generated by a number of groups highlighting 4-1BB–mediated mitochondrial biogenesis, persistence, and central memory differentiation (10, 66, 67), have direct relevance to future optimal CAR design. Indeed, analogous differences in CAR persistence have already been demonstrated in clinical trials evaluating CD19-directed CARs containing CD28 versus 4-1BB (68–70). Recent data, however, from Klein Geltink and colleagues highlight the complexity of CD28-mediated costimulation, which, at least during the initial phase of T-cell activation, has been shown to prime mitochondria with latent metabolic capacity that is essential for future T-cell responses (71). Thus, the timing and duration of CD28 and 4-1BB signaling may be crucial to optimize CAR T-cell metabolism and differentiation. Of relevance to subsequently described studies evaluating the contribution of the hinge and spacer to tonic signaling, modified GD2 BBζ CARs lacked the IgG1 hinge-CH2-CH3 spacer and utilized a CD8α TMD with an scFv peptide linker derived from the CD19 FMC63 scFv. The relative contribution of these changes to the amelioration of CAR exhaustion appears limited, based upon subsequent experiments utilizing a GD2 28ζ CAR incorporating a CD19 FMC63-derived peptide linker and lacking the IgG1 hinge-CH2-CH3 spacer. This adapted CAR demonstrated no improvement in exhaustion and no antitumor efficacy in vivo. This exhaustion-predominant CAR tonic signaling phenotype is illustrated in Fig. 3B.
Interestingly, the Penn group, utilizing the same anti-CD19 FMC63 scFv had previously demonstrated that incorporation of a 4-1BB (rather than CD28) costimulatory ICD could mediate ligand-independent tonic signaling and enhanced proliferation during ex vivo expansion using anti-CD3/anti-CD28–coated beads (62). However, as these cells later lost their proliferative advantage (albeit not their persistence) following removal from the bead-containing culture medium, this is markedly different from the continuous expansion phenotype described by Frigault and colleagues with regard to their c-Met and mesothelin-directed 28ζ CARs (21). Milone and colleagues compared a number of CD19-directed lentiviral-transduced CARs utilizing the potent EF-1α promoter. FMC63 scFvs were fused to a CD8α stalk/TMD and various costimulatory ICDs (namely 4-1BB, CD28, and both in tandem). As mentioned, 4-1BB–containing single ICD CARs, unlike the other constructs, continued to proliferate during in vitro expansion with anti-CD3/anti-CD28–coated beads in the absence of CD19 antigen or exogenous 4-1BBL. This enhanced proliferation was observed in both CD4+ and CD8+ T cells and was associated with a prolonged period of increased cellular volume akin to a more durable blast phase. This initial period of enhanced proliferation appears to recapitulate the findings seen in other models with the continuous administration of a 4-1BB agonist antibody or the ectopic in trans expression of 4-1BBL (72). The authors postulated that CAR oligomerization or impaired dephosphorphylation of CD3ζ ITAMs [by SRC homology region 2 domain-containing phosphatase 1 (SHP-1) and protein tyrosine phosphatase 1 (PTPH1), for example] may be playing a role in this regard (summarized in Fig. 3C). The absence of tonic signaling seen with both the 28ζ and third-generation 28BBζ CARs would imply that scFv oligomerization may not be playing a particular role here, particularly as Long and colleagues have clearly demonstrated that anti-GD2 scFv oligomerization can induce CD28 ICD–mediated ligand-independent proliferation, whereas this was not witnessed with CD19-directed CARs (18). It is conceivable that differences between lentiviral and retroviral promoters used by the two groups may also have been relevant, particularly with regard to promoter strength, CAR surface expression, and potential unforeseen interactions between CAR intracellular signaling and the promoter itself (see Discussion below regarding recent work published by Gomes-Silva and colleagues; ref. 73). Furthermore, tonic signaling has been rarely reported with anti-CD19 CARs other than those containing a single 4-1BB ICD (11, 19, 63). Indeed, the fact that 4-1BB–mediated tonic signaling was only seen during endogenous TCR-mediated activation implies a direct interaction with TCR-related adapter proteins and/or signaling molecules. Given that the introduction of the CD28 TMD and ICD upstream of 4-1BB was able to abrogate ligand-independent tonic signaling (also seen more recently in a third-generation ICOSBB construct; ref. 74), this implies that the aforementioned interaction between the 4-1BB ICD and endogenous TCR activation may depend upon the relative position of the 4-1BB ICD with respect to the cell membrane or that CD28-associated proteins may block this interaction. Indeed, geometric constraints that emerge during the trimeric engagement of TNF receptor family members with their corresponding ligands are thought to facilitate recruitment of signal adapter proteins such as TNF receptor–associated factors (TRAF) that activate downstream signaling pathways (75, 76). It is intriguing to posit that the fusion of the 4–1BB costimulatory ICD into a dimerizing synthetic receptor may alter the natural recruitment and/or disengagement of TRAFs involved in downstream signaling (11). Importantly, in this model, CAR tonic signaling did not appear to compromise in vitro or in vivo efficacy and was, in fact, associated with considerable efficacy and persistence, with anti-CD19 scFvs being detectable in the splenic tissue of mice at 6 months (62).
Intriguingly, recent reports from the Baylor group highlight some important parallels with regard to 4-1BB costimulation, but also reveal some key differences (19, 63, 73). While structurally their CAR is identical to the Penn and colleagues' CD19-directed BBζ CAR (comprising the FMC63 scFv, CD8α stalk, and TMD; refs. 21, 62), Gomes-Silva and colleagues use a non-self–inactivating (non-SIN) γ-retrovirus with an long terminal repeats (LTR) promoter for CAR transduction (73), whereas the Penn group have utilized lentiviral transduction and a variety of promoters with EF-1α most commonly associated with tonic signaling. During ex vivo expansion, Gomes-Silva and colleagues have demonstrated that CARs containing the 4-1BB ICD alone proliferated 70% more slowly, exhibited a 4-fold increase in apoptosis and were characterized by a gradual downregulation of CAR expression (63). Further analysis revealed evidence of constitutive CD3ζ ITAM phosphorylation as well as 4-1BB–associated tonic signaling via TRAF2, leading to phosphorylation of the IκB kinase (IKK) complex (containing IKKα/β subunits), noncanonical NFκB pathway activation, upregulation of Fas (CD95), and Fas ligand (CD95L; which were seen to colocalize on the surface membrane) and, ultimately, caspase 8–dependent activation-induced cell death (AICD); ref.73. Tonic signaling appeared to be further increased via a self-amplifying positive feedback loop acting at the level of the retroviral LTR promoter, which is positively regulated by host NFκB. There was also an upregulation of cell surface ICAM-1, which is also known to be activated by NFκB and the authors postulate that ICAM-1 overexpression facilitated the cell clustering seen in their model, causing trans-engagement of Fas and Fas ligand between neighboring CAR T cells. This 4-1BB–dependent tonic signaling phenotype is illustrated in Fig. 3D. Subsequent work revealed that by disrupting the TRAF2-binding site in the 4-1BB domain, Fas upregulation could be prevented, restoring T-cell function, albeit at the expense of costimulation (19, 73). While the finding that 4-1BB costimulation could induce AICD appears to contradict the data from Penn and other groups (not least in the clinical domain, where Penn's 4-1BB–containing CD19 CAR, tisagenlecleucel, is now FDA-approved), the different viral vectors and promoters may confer different levels of CAR surface expression, which appears to be a crucial factor for ligand-independent tonic signaling. The interaction between 4-1BB–mediated tonic signaling and the retroviral LTR promoter also appears to be important. Similar outcomes were noted using 14G2a GD2-directed CARs and therefore the results differ markedly from those seen by Long and colleagues. Although exhaustion markers were not evaluated by Gomes-Silva and colleagues, these differences are likely to be occurring at the level of the CAR promoter. While both groups made use of retroviral vectors and LTR sequences, Gomes–Silva utilized an SFG vector, whereas Long and colleagues transduced with a murine stem cell virus–based splice-gag (MSVG) vector, which utilizes the murine stem cell virus LTR with an extended gag region and Kozak sequence (77) and may not be regulated by host NFκB in the same manner. Indeed, enforced reduction of CAR expression using an internal ribosome entry site (IRES) element upstream of the CAR transgene reduces tonic signaling in Gomes–Silva and colleagues's model. A similar restorative effect was seen using lentiviral transduction and an EF-1α promoter, exactly replicating the experimental model utilized by Frigault and colleagues Of note, the use of the IRES element does not appear to have inhibited the continuous expansion of CD19 and GD2 CD28ζ CARs, which is also redolent of the findings seen by Frigault and colleagues using non-CD19 CARs. However, differences in ex vivo expansion may have played a role with the former being expanded in the presence of continuous IL7 and IL15.
Intriguingly, a third-generation construct combining a CD28 ICD/TMD upstream of 4-1BB was able to overcome or avoid this deleterious 4-1BB tonic signaling despite utilizing the same γ-retrovirus and LTR promoter and, following delivery to a small number of patients in combination with a 28ζ second-generation CAR, was able to demonstrate a 23-fold greater level of expansion and correspondingly longer persistence in vivo (63). The differing effects may parallel models of acute viral infection, whereby 4-1BB appears to have a biphasic role (78). Early 4-1BB activation has been shown to have a deleterious impact on antiviral T-cell effector function by inducing AICD through prolonged upregulation of TNF and Fas (79). Thus, the precise timing and duration of 4-1BB costimulation may be key (11), and it is interesting to speculate that the relative position of the 4-1BB ICD and its preferential access to TRAF2 rather than TRAF1 or TRAF3, which are both known to exert a negative regulatory role on noncanonical NFκB activation by preventing activation of the NFκB-inducing kinase (NIK; refs. 80, 81), may be playing a role. Indeed, TRAF1 is also known to activate ERK, upregulate Bcl-xL, and downregulate the proapoptotic protein BIM (82) and loss of TRAF1 has been associated with CD8+ T-cell dysfunction during human and murine chronic infection (83). These data, while currently only hypothesis generating, appear to emphasize the considerable importance of optimally positioning costimulatory ICD(s) to facilitate interactions with cell membrane-localized adapter and signal transduction molecules (such as members of the TRAF family, which are likely to have pleiotropic roles in different contexts) in the CAR's activated conformational state, as well as the hitherto relatively underexplored impact of using different hinges, spacers, and TMDs.
The mainstay of available data with regard to the impact of the CAR hinge and spacer domain relates to potential FcγR-mediated interactions with immune cells causing ligand-independent CAR tonic signaling, chronic activation, and AICD (20, 44, 84). A commonly utilized spacer domain comprises an IgG-derived hinge (usually IgG1 or IgG4), and a variable length IgG Fc CH2-CH3 domain. However, CARs comprising an IgG1 Fc spacer domain are prone to ligand-independent activation by binding to bystander immune cells expressing FcγR. Substitution of an IgG1-derived CH2 sequence with that from IgG2 (which has a lower affinity for FcγR) has been shown reduce this effect in vitro (85). IgG4 has been shown to bind to FcγRI and other FcγRs with an equivalent or lower affinity than IgG2. However, Hudecek and colleagues have shown that the use of a full-length IgG4 Fc motif (containing the hinge, CH2, and CH3 modules) in CD19 and receptor tyrosine kinase-like orphan receptor 1 (ROR1)-directed CARs was associated with significant tumor-independent trapping of CAR T cells in the lungs of NSG mice, and reduced antitumor efficacy and persistence compared with CARs with a truncated IgG4 spacer lacking CH2 and CH3 (44). The authors postulate that CARs with a full-length IgG4 Fc spacer are sequestered by lung-resident Ly6C+ mononuclear cells expressing FcγR, highlighting the finding that the few CAR-T cells able to escape to the periphery have a highly activated phenotype with a significant propensity to undergo AICD. For patients, this may be particularly relevant in cases of B-cell lymphodepletion or hypogammaglobulinemia where immune cell FcγRs may be relatively under-occupied, accentuating the interaction with IgG Fc-containing CARs. Aside from causing AICD, the cross-activation of FcγR+ immune cells may activate the innate immune system contributing to macrophage activation syndrome (MAS) and/or cytokine release syndrome (CRS). Targeting of the myeloid compartment and/or natural killer (NK) cells (depending upon the spacer's IgG subclass) would also be liable to have repercussions for antitumor efficacy and, with regard to myeloid cells, may have positive or negative effects in different tumor models.
The contribution of the hinge/spacer domain to ligand-independent CAR tonic signaling has been investigated further by Watanabe and colleagues (20). Starting with a γ-retrovirus–transduced second-generation prostate stem cell antigen (PSCA)-directed CAR comprising an IgG1-derived hinge and CH2-CH3 spacer bound to a CD28 TMD/endodomain and CD3ζ chain (termed P1.CAR), they proceeded to evaluate how modifications to the spacer could impact in vitro expansion and cytotoxicity as well as CAR performance in vivo using NSG mice engrafted with human PSCA-expressing tumor cell lines. In keeping with other reports of FcγR-mediated pulmonary trapping, intravenous delivery of P1.CARs resulted in poor trafficking to the tumor or lymphoid tissue and significant accumulation in the lungs. This was found to be mediated by interactions with monocytes and macrophages expressing FcγR I and II and could be abrogated by making residue alterations to the IgG1 CH2 region or, optimally, by substituting the IgG1 framework for IgG2. Despite far superior migration and an absence of significant pulmonary trapping these modified CARs continued to perform poorly in vivo. Subsequent analysis revealed that all of these CAR T-cells (except cells that expressed a truncated CAR control) exhibited continuous expansion and cytokine production in vitro in the absence of ligand, consistent with other reports of constitutive tonic signaling. Continuous CAR expansion was associated with progressive differentiation toward a terminal effector phenotype with elevated expression of EOMES, FASL (encoding Fas ligand) and GZMB (encoding granzyme B) and loss of CD27, CD28, and CD62L (encoded by SELL; illustrated in Fig. 3E). However, unlike the exhausted phenotype identified by Long and colleagues (18), these CARs did not exhibit an upregulation of PD-1 or other inhibitory molecules. Accelerated cell senescence, however, was a feature, although telomere length following expansion was not evaluated in this study. Deletion of the IgG2 CH2-CH3 spacer prevented tonic signaling, allowing CARs to maintain an undifferentiated phenotype (high CCR7:CD45RO ratio), but at the expense of cytotoxicity (particularly in the face of low target surface expression). Reinsertion of an IgG2-derived hinge and CH3 domain to create an intermediate length spacer (X32.CAR) could restore cytolytic capacity without the reemergence of tonic signaling and demonstrated significantly improved in vivo performance. Interestingly, alteration of spacer length in both a first-generation MUC1 CAR and a second-generation CD19 CAR resulted in a similarly undifferentiated phenotype. Taken together, these data suggest that a different pattern of tonic signaling can occur with different hinge/spacer domains and that this is likely to be occurring at the level of scFv oligomerization, which may be facilitated by the flexibility and length of these domains.
CARs containing murine scFvs have, unsurprisingly, been found to be immunogenic when used in humans (86) and can cause anaphylaxis (87). While the former would not be anticipated to be a significant long-term problem with CARs targeting B-cell antigens (such as CD19, CD20 or CD22) or indeed in the aftermath following the administration of a lymphodepleting conditioning regimen; however, in the longer-term, this is anticipated to be a problem, particularly if using these scFvs to target solid tumors. Indeed, antibodies directed to a murine scFv targeting carbonic anhydrase 9 (CAIX) were detected in patients receiving a first-generation CAR (86). While binding of host immunoglobulin to murine scFVs in this manner would be expected to elicit CAR cross-linking and cell-surface clustering in the absence of ligand, the detrimental impact of antibody-induced tonic signaling is likely to be considerably outweighed by the targeting of antibody-bound CAR T cells for destruction by antibody-dependent cell-mediated cytotoxicity (ADCC) and complement-dependent cytotoxicity (CDC).
Ligand-independent tonic signaling may also induce constitutive systemic production of cytokines outside of the TME, with potentially deleterious effects, including CRS, MAS, multiorgan toxicity, and the expansion of immunosuppressive cells. This issue may be further magnified when utilizing TRUCKs or armored CARs, capable of secreting transgenic cytokines at a high level in an inducible or constitutive manner. Indeed, a phase I clinical trial evaluating ACT with inducible IL12-engineered tumor-infiltrating lymphocytes (TIL) in patients with advanced melanoma revealed high serum levels of IL12 and significant hepatotoxicity (88). CAR tonic signaling acting on the nuclear factor of the activated T-cell (NFAT) promoter may exacerbate this further. A method of potentially constraining this, albeit without addressing tonic signaling itself, would be to link cytokine production to an inducible switch promoter and a background reduction signal (BRS). Uchibori and colleagues have developed such a system by delivering a switch cassette comprised of two modified Simian virus 40 early polyA sequences (serving as a BRS), four NFAT-responsive elements, a minimal IL2 promoter, a ZsGreen1 reporter, and a bovine growth hormone polyadenylation (BGH polyA) sequence, to CD19-directed CAR Jurkat cells (89). Jurkat cell ZsGreen1 expression was only seen when cocultured with CD19-positive target cells in this model.
Finally, unconstrained activation caused by tonic signaling may lead to impaired trafficking of CAR T cells into the TME, mediated by the downregulated expression of relevant chemokine receptors. It has been shown, for example that ex vivo activation of CAR T cells using anti-CD3 and anti-CD28 antibodies can lead to a concomitant reduction in the surface expression of both C-C motif chemokine receptor 9 (CCR9) and α4β7 integrin (90), which may, for example, be predicted to impair trafficking to the small intestine (91). However, this area remains relatively underexplored and as such is subject to conflicting reports. For example, in Watanabe and colleagues' model of CAR tonic signaling, significant upregulation of the chemokine receptors CCR2 and CCR5 is noted (20). In a non-CAR context, CCR2+ CCR5+ CD4+ T cells derived from healthy human donors have been shown to harbor a central memory and effector memory phenotype and are capable of migrating to a number of inflammatory chemokines (including the C-X-C motif chemokine ligands CXCL-9 and CXCL-12; and the C-C motif chemokine ligands CCL-2 and CCL-20; ref.92). Furthermore, ligand-dependent CAR tonic signaling may be anticipated to enhance cell adhesion molecule interactions, diapedesis, and trafficking via inside-out signaling to integrins (such as LFA-1). This is known to be dependent, in part, on the activation of TCR downstream signaling shared with CARs, and specifically the activation of membrane-derived DAG via the LAT-phospholipase Cγ1 (PLCγ1) signalosome complex causing knock-on activation of the small GTPase Rap1 (93).
Ligand-dependent CAR tonic signaling
Thus far, CARs designed to target TAAs in solid tumors have been reliant upon the existence of an expression differential between tumor cells and normal tissue (22, 94). While adjustments to CAR expression levels and target binding affinity and avidity can help discriminate between low-level and high-level antigen expression (95, 96), difficulties with physiologic low-level antigen expression have been encountered with both first and second-generation constructs evaluated in clinical trials (27). Unfortunately, outcomes can be dire, as illustrated by the case of a patient with colorectal cancer who developed an acute respiratory distress syndrome (ARDS)–type picture followed by fatal multiorgan failure after the intravenous delivery of 1010 CD8+ T cells expressing a potent third-generation HER2-directed CAR containing both CD28 and 4-1BB costimulatory ICDs (97). Subsequent post-mortem analysis provided credence to the hypothesis that physiologic low-level HER2 expression on lung epithelium and/or microvasculature resulted in rapid cytokine release syndrome (CRS), exacerbated by first pass sequestration of CARs in the lungs following intravenous administration. On-target, off-tumor toxicity has been witnessed in other models, including a first-generation CAR targeting CAIX, a TAA commonly expressed by clear cell renal cell carcinoma. However, clinical trials revealed multiple cases of cholangitis due to the targeting of low-level CAIX expression on biliary epithelium (98). Below the threshold of cytotoxicity, however, chronic engagement of CARs with low-level off-tumor antigen may induce chronic ligand-dependent tonic signaling, anergy (particularly with first-generation constructs) and exhaustion prior to their entry into the TME.
Theoretically, antigen shedding [e.g., soluble carcinoembryonic antigen (CEA)] could induce low-level CAR tonic signaling in the TME or systemic vasculature. However, there is little evidence in the published literature to support this phenomenon in vivo. In vitro studies, thus far, have demonstrated that CEA-directed CARs are not inhibited by high concentrations of soluble CEA (up to level 10-fold higher than usually found in the sera of patients with cancer; ref. 99).
Constitutional ligand-dependent tonic signaling is a problem that might be anticipated in the specific scenario of targeting T-cell lineage leukemias or lymphomas. CD5- and CD7-directed CAR T cells have been developed and a variable degree of fratricide has been noted (100, 101). Following CAR engagement, surface expression of CD5 on transduced T cells is lost due to complete ligand-dependent internalization, leading to acceptable levels of fratricide. The loss of CD5, which is known to exert inhibitory effects on TCR activation (ref. 102; at least partly due to negative regulation of ZAP70 via a reduction in the kinase activity of Fyn; ref. 103), may also confer a beneficial effect on CAR T-cell effector function. In the case of CD7, however, there is incomplete loss of expression leading to high levels of fratricide that was capable of impairing CAR expansion. This could be abrogated by using CRISPR/Cas9-mediated targeted disruption of the CD7 gene prior to CAR expression. Both unedited CD5 and edited CD7 CAR T-cells have been successfully expanded long-term ex vivo and both were able to effectively eliminate malignant T-cell acute lymphoblastic leukemia (T-ALL) and T-cell lymphoma cell lines in vitro, as well as inhibited disease progression in xenograft mouse models (100, 101). Separately, we have found that the transmembrane glycoprotein mucin 1 (MUC1) may be expressed on activated T cells, giving rise to ligand-dependent tonic signaling by MUC1-directed CAR T cells comprising scFvs derived from either the SM3 or HMFG2 antibodies (104). Ex vivo expansion was associated with increased activation, cytokine release, and fratricide. A possible method to circumvent this may be to culture the CAR T cells in the presence of a peptide epitope capable of blocking the CAR without downregulating it.
Finally, both ligand-independent and ligand-dependent tonic signaling may theoretically be enhanced by nonspecific T-cell adhesion, for example, via ICAM-1/LFA-1 interactions, which are able to lower the threshold for TCR-mediated T-cell activation (56). This process is known to contribute to the priming of naïve T cells in lymphoid structures by mature DCs, which readily express and modulate cell-surface adhesion molecules (105). These interactions may also facilitate TCR-mediated tonic signaling following binding of self-peptide presenting MHC and, due to the impact on downstream signal transduction pathways shared with CARs, it is plausible that a similar phenomenon may occur following CAR ACT in vivo. As ICAM-1, ICAM-2, VCAM-1, and other adhesion molecules are often overexpressed on both tumor cells and the TME (106), the impact of these interactions on CAR signaling may be particularly beneficial when targeting solid tumors. Interestingly, tumor cell surface expression of adhesion molecules can be induced by exposure to activated CAR T cells. For example, following exposure of low mesothelin, expressing A549 lung cancer cells, to cytokines secreted by activated mesothelin-directed 28ζ CARs, A549 cells upregulated ICAM-1 and were susceptible to enhanced bystander cytotoxicity by CAR T cells that also demonstrated upregulation of cell surface LFA-1 (107).
Potential strategies to address CAR tonic signaling
Engineering or altering the targeting moiety.
With the absence of structural support provided by the IgG constant regions, scFv stability and/or folding properties can render CARs susceptible to oligomerization, clustering, and ligand-independent tonic signaling. Variable heavy (VH) and light (VL) immunoglobulin chains are typically joined by a flexible peptide linker resistant to endopeptidase degradation (108). Nevertheless, most employed scFvs still demonstrate a tendency to unfold at the VH:VL interface, leading to suboptimal stability of the two immunoglobulin domains. This type of unfolding can permit “protein domain-swapping,” whereby complementary domains from adjacent scFv molecules can interact with one another leading to scFv oligomerization (109). Depending upon the length of the peptide linker, which may impede the rotation of the complementary Ig domains, oligomers comprising two, three, or even four scFv molecules can form (illustrated in Fig. 4A; ref. 108). Allowing for steric hindrance caused by the attached CARs, such oligomers can also be envisaged on the cell surface, potentially causing tonic signaling. The use of highly flexible and/or long extracellular spacers may be expected to facilitate oligomerization of intrinsically unstable scFvs. Likewise, the construction of tandem CARs with two scFvs per CAR monomer (110) may be at greater risk of oligomerization, clustering, and tonic signaling.
These issues may be addressed by optimizing the orientation of heavy and light chains; selecting VH and VL consensus master gene sequences (111); by engineering disulfide bonds between the VH and VL domains, either in the absence of a peptide linker or in combination to ensure maximal stability (112); by introducing charged mutations within the VH and VL domains (113); by complementarity-determining region (CDR) grafting (114); or by using a combination of these strategies. Longer peptide linkers may reduce the likelihood of multivalent oligomerization, although potentially at the cost of increased proteolysis or weak domain association. A linker of 15–20 residues is generally regarded as thermodynamically most stable (109). In the absence of covalent bonds, VH and VL interactions are dependent upon electrostatic interactions, hydrophobic repulsion, hydrogen bonds, and van der Waals forces. They are thus, subjected to local temperature, protein concentration, ionic strength, and above all pH, which in the TME may be detrimental to stable folding (108). These issues and the impact of scFv engineering upon antigen-binding affinity and avidity, as well as the immune synapse formation remain to be characterized.
Single-chain variable fragment aggregation may also occur in the absence of domain swapping due to the hydrophobic nature of residues within their CDRs, which mediate binding to target antigens. Various techniques have been deployed to resist aggregation without reducing binding affinity. Examples include inserting two or more negatively charged residues at each edge of the scFv's third CDR (CDR3; ref. 115) or introducing a glycosylation site inside the second CDR to compensate for the presence of hydrophobic residues within the third CDR (116).
Further improvements in scFv stability, particularly at the VH:VL interface can also be achieved using advanced computational modeling. Ultimately, the biophysical characteristics of scFvs are determined by their germline sequence but influenced by somatic hypermutations in the framework regions. Computational modeling has been used to revert these hypermutations to germline consensus and optimize stability at the VH:VL interface. For example, outside of the CDRs additional hydrogen bonds can be introduced between the VH and VL domains by replacing phenylalanine with tyrosine residues and by filling in pockets within the topological structure of the VH domain by substituting phenylalanine (Phe) with tryptophan (Trp; ref. 117). Improvements in CAR stability translated to improved CAR surface expression and enhanced in vitro cytotoxicity. Furthermore, a reduction in tonic signaling was noted in comparison to the original scFv-derived CARs. Beyond dissociation at the VH:VL interface, disparities may exist between the relative thermal stability of the VH and VL domains. Such a scenario may lead to an accumulation in equilibrium of an unfolding intermediate, where one domain completely unfolds and the other remains native, leading to enhanced aggregation. By simulating the molecular dynamics in silico, particularly with regard to the less stable of the two domains, one can systematically engineer scFvs to improve intrinsic stability and minimize aggregation (118). These strategies are summarized in Fig. 4B.
Alternative strategies may include selecting a target epitope localized on the membrane-anchored part of antigen to avoid tonic signaling mediated by shed, oligomerization of soluble antigen, or by utilizing scFvs with low- to intermediate-binding affinity to enhance discrimination between membrane-bound and soluble antigens (119), and, in general terms, the use of a low-affinity scFv is likely not only to provide a means of discriminating between tumor cells with high-level antigen expression and normal cells with low-level expression (thus improving safety; ref. 120) but may also minimize ligand-dependent tonic signaling caused by the presence of more widespread low-level antigen expression on normal tissues.
Alternative targeting moieties, such as camelid single-domain antibodies (VHHs) termed “nanobodies,” which share a high degree of homology with human VH sequences, and are the smallest known single-chain antibodies (121) may also avoid tonic signaling by being intrinsically unable to domain swap, although their efficacy in CARs remains to be fully elucidated in experimental models. Interestingly, due to their small size and the length of their CDRs, which form extended loops, they are able to access cryptic epitopes (such as catalytic sites in enzymes) or large structures that typically escape immunosurveillance (122). Like murine scFvs, the potential immunogenicity of camelid nanobodies is being addressed using sequence humanization techniques.
In addition, centyrin-based CARs have also been designed, with properties that may limit ligand-independent tonic signaling. Centyrins represent a novel class of alternative scaffold protein on the basis of a consensus tenascin fibronectin domain. They are smaller than scFvs and are monomeric. A human B-cell maturation antigen (BCMA)–directed centyrin CAR transduced using the Super PiggyBac transposon/transposase system has shown excellent in vitro cytotoxicity with a predominantly stem cell memory phenotype (123).
The issue of scFv clustering may also be circumvented by utilizing endogenous receptors or ligands as CAR extracellular targeting moieties. A wide variety of such constructs have been designed with several already undergoing clinical evaluation (9, 124). Those that have progressed furthest along clinical development include IL13-zetakine CARs, incorporating membrane-tethered IL13 to target the IL13 receptor subunit alpha-2 (IL13Rα2) decoy receptor, a glioma-restricted cell surface epitope (125, 126); CARs armed with an EGF/TGFα fusion molecule capable of targeting pan-ErbB homo- and heterodimers expressed on a plethora of solid tumors (127, 128); CARs armed with the natural killer group 2D (NKG2D) protein fused to CD3ζ alone (129) or in combination with an intracellular costimulatory domain (130) to target a wide variety of hematologic malignancies and solid tumors overexpressing NKG2D stress–inducible ligands [such as MHC class I chain-related protein A (MICA), MHC class I chain-related protein B (MICB), and UL16 binding proteins 1 to 6 (ULBP1-6) in humans; ref. 131]; and CARs utilizing CD27 to target CD70 (132), an antigen aberrantly expressed by a broad range of hematologic malignancies and some solid tumors including RCC and glioma. Thus far, no target-independent tonic signaling has been reported, but due to the nonrestricted expression of many of their targets, these constructs may be liable to encounter chronic low-level ligand-dependent tonic signaling, which may have positive or negative effects in different contexts.
Adjusting the hinge/spacer
Hinge and spacer domains have proved particularly beneficial for the targeting of membrane-proximal epitopes and are able to relieve spatial constraints that may hinder interactions between tumor antigens and CARs (9, 45, 133). The CD8α hinge is typically used with the CD8α TMD and plays an important role in maintaining the flexibility of the CAR-binding domain and the ability to form an efficient immunologic synapse with the target cell. The substitution of cysteine residues normally involved in CD8α/α and CD8α/β dimerization can permit both homo- and hetero-dimerization of the CAR, enhancing its transport out of the endoplasmic reticulum (ER) to the cell surface (134), and increasing the level of productive dimerization resulting in more effective target-cell killing in a transduced NK-cell model (135). In the context of a molar excess of endogenous CD3ζ, enhanced heterodimerization would be expected to lower the threshold for CAR tonic signaling.
Experiments undertaken by Watanabe and colleagues whereby an intermediate length IgG2 hinge/spacer was shown to abrogate CAR tonic signaling without compromising cytolytic capacity have already been discussed previously (20). Separately, experiments conducted with second-generation lentiviral-transduced 28ζ CARs directed to a variety of antigen targets (including CD19, mesothelin, PSCA, HER2, and MUC1) either with or without an IgG4-CH3 hinge/spacer domain have demonstrated that, in all cases, the presence of the hinge conferred increased expansion (and particularly late expansion beyond Day 15) in a ligand-independent manner during in vitro culture following prior exposure to anti-CD3/anti-CD2/anti-CD28–loaded microbeads (136). Enhanced hinge-containing CAR T-cell expansion appeared to depend upon proliferation of the CD4+ subfraction, but was abrogated if CD4+ and CD8+ T cells were cultured separately, suggesting that tonic signaling may have a differential role in CD4+ and CD8+ populations, and that cross-talk between the two lineages may also be occurring. Interestingly, utilizing a chemoattractant assay, the researchers were also able to show that the hinge-containing CAR T cells had inherently enhanced migratory and invasive capabilities, reinforcing the likelihood of tonic signaling playing a decisive role here.
As already discussed, in cases where CARs are utilizing full-length IgG Fc-containing spacers, interactions with FcγR-expressing mononuclear or NK cells are expected to induce “off-target” activation and AICD. Although myelodepleting conditioning regimens may limit these interactions in the immediate period following CAR T-cell infusion, this problem would be expected to reemerge following recovery of the myeloid compartment. Likewise, saturating immune cell FcγRs with exogenous human immunoglobulin prior to CAR T-cell administration provides only a short-term solution. In a ROR1-targeting CAR incorporating the R11 scFv, IgG4 Fc spacer, CD28, and 4-1BB costimulatory ICDs with CD3ζ, Hudacek and colleagues have shown that modification of the spacer to limit FcγR-mediated activation and AICD promotes enhanced effector function and persistence in a NSG mouse model (44). While CARs designed to target nonproximal cell-surface epitopes (such as CD19) can be optimized with shortened spacers that omit the entire IgG4 CH2 domain (thereby eliminating binding by FcγRI), CARs designed to target a transmembrane proximal epitope (such as the ROR1 kringle domain) require a full-length spacer to optimize immune synapse formation and reduce steric hindrance. Hudacek and colleagues were able to maintain function and address pan-FcγR activation, sequestration, and AICD in this model by swapping the CH2 sequences of the IgG4 spacer with those of IgG2 and replacing the crucial N-glycosylation site Asn297 with a conserved residue not amenable to N-linked glycosylation.
Optimal selection of the transmembrane domain
Although few reports exist regarding the role of the CAR TMD in contributing to tonic signaling, it is abundantly clear that the TMD plays a vital role in CAR cell surface expression and stability, as well as its ability to interact with other cell surface molecules that may contribute to signal transduction (137). Utilizing an unedited CD3ζ TMD may facilitate heterodimerization with endogenous CD3ζ chains, potentially lowering the threshold of antigen binding required to elicit a cytotoxic response (138) and, as an anticipated corollary, enhanced tonic signaling. However, cell surface expression of CD3ζ TMD-containing CARs appears to be lower than those containing CD28 or CD8α TMDs (139). The optimal selection of TMD to mitigate tonic signaling remains to be elucidated and is likely to be impacted or subsumed by the many other factors outlined in this review.
Optimal selection of costimulatory intracellular domains
When ligand-independent tonic signaling occurs due to scFv clustering, particularly negative effects appear to be mediated by constitutive CD28 signaling, leading in some scenarios to IL2 gene expression, and a positive feedback loop of unconstrained proliferation and activation (21, 64). Uncontrolled IL2 production may also have the unintended consequence of attracting and enhancing the proliferation of immunosuppressive Tregs (140). Deletion of the CD28 Lck-binding moiety in this model could abrogate enhanced IL2 production, without compromising IFNγ secretion, proliferation, and cytolysis. Greater complexity may exist in certain tumor models due to endogenous receptor interactions [e.g., between CD2 (LFA-2) on CAR T cells and CD58 (LFA-3) on tumor cells] potentially recapitulating CD28-mediated IL2 production (141). In most cases where scFv clustering has been implicated, the CD28 domain appears to be detrimental and the 4-1BB domain beneficial. Nevertheless, there appears to be at least two distinct phenotypes—one that is characterized by continuous expansion, terminal differentiation, and senescence [seen in studies by Frigault and colleagues (21), Watanabe and colleagues (20), and Qin and colleagues (136)]; and one characterized by T-cell exhaustion [seen in studies by Long and colleagues (18)]. In the case of the latter, the production of activating cytokines (such as IL2 and TNFα) appears to be significantly curtailed and the use of a 4-1BB costimulatory domain could rescue these cells from exhaustion and was associated with a memory T-cell metabolic phenotype (18). In the case of the former, a reduction in CAR surface expression or a shortened spacer could ameliorate the negative consequences of tonic signaling both in vitro and in vivo. In Frigault and colleagues' experiments the use of a 4-1BB or ICOS costimulatory domain also could alleviate continuous expansion. Watanabe and colleagues did not explore the use of a 4-1BB costimulatory ICD.
However, when using scFvs that are not typically prone to clustering (e.g., anti-CD19 FMC63 scFv), the use of a 4-1BB costimulatory domain may confer ligand-independent 4-1BB tonic signaling that appears to require T-cell activation (mediated by CD3 and CD28 binding; ref. 62). Again, there appear to be at least two phenotypes—one that is noncontinuous and characterized by improved expansion, in vivo persistence, and antitumor efficacy (Milone and colleagues; ref. 62); and another associated with poor expansion, upregulation of Fas and Fas ligand, and AICD (Mamonkin and colleagues; ref. 19). In the case of the latter, the negative consequences of tonic signaling could be ameliorated by adding a CD28 costimulatory domain upstream of 4-1BB to construct a third-generation CAR (63) or by reducing CAR expression by adding an IRES element between the retroviral promoter and the CAR transgene. A highly vector-specific amplification loop involving the LTR promoter appears to explain this unusual phenomenon. Therefore, the negative or positive consequences of 4-1BB–mediated tonic signaling are likely to result from differences in quantitative and qualitative 4-1BB activation and from differences in the temporal and spatial interaction of the 4-1BB ICD with membrane-associated signal transduction molecules that are also involved more broadly in T-cell activation.
Additional techniques that have been associated with improved CAR surface expression, such as mutating CD28 noncanonical di-leucine internalization motifs (albeit, in a murine CD28 model; ref.142) may also be expected to exacerbate the consequences of tonic signaling occurring in certain CARs.
Finally, there may be scope to utilize alternative strategies to recapitulate the benefits of 4-1BB costimulation, wheresas preventing the possibility of 4-1BB–mediated tonic signaling. Zhao and colleagues have found that the expression of a second-generation SJ25C1 CD19 28ζ CAR in trans with constitutively expressed transgenic 4-1BBL (thus, providing paracrine costimulation following inducible 4-1BB upregulation) resulted in considerably improved performance (and significant IFNβ production) compared with an equivalent third-generation 28BBζ CAR (10). However, care may be needed with this approach based upon reports that 4-1BBL crosslinking in the absence of available 4-1BB may foster suboptimal CD4+ T-cell activation (143). Whether this would have consequences for CAR, rather than TCR activation, remains to be seen as there may be differences in the degree of 4-1BB upregulation, which if more potent following CAR activation, may more easily reverse the suppressive effects of 4-1BBL through T-cell–intrinsic 4-1BB–regulated 4-1BBL internalization.
Controlling CAR expression
With regard to gene transduction using viral vectors, numerous strategies have been adopted to improve safety by minimizing the risk of producing replication-competent virus and reducing the potential to cause insertional mutagenesis. Self-inactivating (SIN) vectors have been developed using both retroviruses and lentiviruses by deleting/replacing LTR elements. Nonintegrative lentiviruses (NILV) have also been designed by mutating the integrase gene or by modifying the attachment sequences of the LTR (144–146). By limiting high-level CAR expression, these methods may reduce the likelihood of tonic signaling being caused by CAR clustering. As discussed previously, Frigault and colleagues have demonstrated continuous ligand-independent CAR T-cell proliferation with lentiviral vectors using the EF-1α promoter, but not when driven by the CMV or variably truncated PGK promoters (21). More recently, Gomes-Silva and colleagues have demonstrated that 4-1BB–mediated tonic signaling was highly dependent upon CAR surface expression and that a γ-retroviral LTR promoter was liable to amplify CAR expression in a positive feedback loop mediated by 4-1BB–induced NFκB activation (73). The use of an IRES element upstream of the LTR promoter could curtail CAR expression, thereby reducing tonic signaling. A similar improvement was also seen following transduction with a SIN lentiviral vector.
In addition, by ensuring transient or self-limiting CAR expression utilizing plasmid or mRNA electroporation, the risks of both genotoxicity and tonic signaling –induced T-cell exhaustion may be addressed. Following RNA transfection, the transgene is typically expressed for approximately one week (147). Such a system, therefore, is likely to require repeated CAR T-cell administration at multiple time points (148). Constitutive T-cell proliferation caused by tonic signaling has not yet been reported when CARs are expressed by electroporation of mRNA or plasmids encoding the Sleeping Beauty transposon/transposase system (149, 150), in contrast to lentiviral transduction (21). The impact on tonic signaling of newer NILVs (e.g., those containing a scaffold/matrix attachment region (S/MAR) element) with the capacity to confer long-lasting episomal CAR expression on par with that of integrative lentiviral vectors (151) remains to be seen.
Regulated on/off switches, designed primarily to mitigate CAR toxicity, may also have a dual role in reducing tonic signaling by ensuring that CAR surface expression is tightly controlled in a temporal manner following antigen exposure. One such setup incorporates a single vector tetracycline (Tet)-On inducible gene expression system, whereby the CAR gene is located downstream of a reverse Tet transactivator (rtTA) fusion protein, which is able to activate its promotor only in the presence of doxycycline (152). Extrapolating from the supposition that CAR tonic signaling, terminal differentiation and/or exhaustion are, at least in part, due to unconstrained CAR cell-surface expression, one may hypothesize that the intermittent withdrawal of doxycycline using this model (particularly during the initial phase following CAR T-cell delivery) may avert these negative consequences and enhance antitumor efficacy. Naturally, this approach would rely upon the pharmacokinetics of doxycycline being conducive to ensuring a kinetically optimal CAR transcription profile that could minimize tonic signaling. Separately, Mamonkin and colleagues have reported that the negative consequences of 4-1BB–mediated tonic signaling could be prevented by utilizing a small molecule to regulate CAR expression at the level of their γ-retroviral promoter (19).
Likewise, separating the CAR into two functional entities and/or utilizing a dimerizing agent such as a rapamycin analogue (rapalog; ref. 153) could limit the likelihood of both ligand-independent and ligand-dependent tonic signaling. The latter could also be limited by divorcing the CAR scFv from the intended epitope by utilizing an exogenous targeting module as per the UniCAR system (154). Wu and colleagues have constructed a split ON-switch CAR triggered only in the presence of target ligand and a small-molecule dimerizing agent (using either the rapalog AP21967 or the plant hormone gibberellin; ref. 155). For the rapalog-gated CAR, one component comprises the scFv extracellular targeting moiety linked to a CD8α hinge/TMD, a 4-1BB costimulatory ICD, and a distal FK506 binding protein (FKBP) domain; the second component comprises a DAP10 ectodomain, CD8α hinge/TMD, 4-1BB ICD, mutant FKBP-rapamycin–binding domain (FRB*), and CD3ζ ICD. The DAP10 ectodomain was selected to aid homodimerization, doubling the potential number of CD3ζ ITAM domains per assembled CAR. For the gibberellin-gated CAR, FKBP, and FRB* were substituted with gibberellin insensitive dwarf 1 (GID1) and gibberellic-acid insensitive (GAI). These CARs demonstrated titratable cytotoxicity in the presence of the dimerizing agent and similar in vivo efficacy.
The design of customizable logic-gated circuits may also alleviate the negative consequences of unconstrained tonic signaling by rendering CAR surface expression amenable to both temporal and spatial control. The synNotch system couples CAR transcription to the signaling of a synthetic Notch receptor, engineered to engage with a second TAA (156–158). Subsequent proteolytic cleavage of the receptor induces the release of a synthetic transcription factor able to induce (or suppress) CAR transcription. Because of the orthogonal nature of these synthetic gene circuits, it is conceivable that a single CAR T-cell may be controlled by multiple synNotch receptors. The authors have not commented on the potential for tonic signaling to occur in this model. However, tonic signaling of endogenous Notch has been reported in a variety of contexts, such as in mouse myoblasts (159) and epidermal keratinocytes, where the metalloprotease ADAM17 has been implicated in maintaining a basal level of Notch1 activity in a ligand-independent fashion (160). In addition, in T-cell Notch has been shown to undergo spontaneous cleavage in the absence of Notch ligands following TCR engagement, where it may augment signal 1 and 2-induced proliferation (161). Separately, the CAR product may exhibit tonic signaling due to the choice of promoter or due to intrinsic structural issues. SynNotch may also be liable to induce immunogenicity, as well as theoretical off-target effects due to persistent CAR transcription in synNotch-controlled T cells that have exited the TME.
Eyquem and colleagues have demonstrated that targeting a CD19-specific 28ζ CAR to the T-cell receptor α constant (TRAC) locus using CRISPR/Cas9 results in superior performance and persistence, compared with conventionally generated CAR T cells using a SFG γ-retroviral vector, independent of TCR disruption (64). Interestingly, targeting the CAR to the TRAC locus reduced tonic signaling and enhanced CAR internalization and reexpression following repeated exposure to antigen, delaying effector T-cell differentiation and exhaustion. Unlike CD19-specific CARs utilizing the FMC63 scFv, γ-retrovirally transfected SJ25C1 scFv CARs (RV CAR) demonstrated constitutive activation and tonic signaling in the absence of ligand evidenced by baseline ITAM phosphorylation. Furthermore, by localizing the CAR to the β2-microglobulin (B2M) locus or to TRAC using either the EF-1α constitutive promoter or the LTR retroviral promoter, they were able to show that the degree of antigen-independent tonic signaling correlated with CAR surface expression. In contrast to endogenous promotor TRAC CAR T-cells, following repeated stimulation by antigen, RV CARs (and TRAC EF-1α CARs) rapidly differentiated into an effector T-cell phenotype with loss of CD62L expression, potent secretion of IL2, and expression of T-bet, EOMES and GATA3. A critical difference was identified in the level of CAR expression following repeated antigen exposure in endogenous promotor TRAC CARs versus EF-1α or RV LTR CARs. While the latter group demonstrated a rapid step-wise increase in CAR surface expression following each antigen exposure, endogenous promotor TRAC CAR cell-surface expression conversely reduced following each exposure and remained below baseline after 48 hours, mediated by CAR internalization and degradation. The implication of this work is that by constraining both baseline and dynamic CAR tonic signaling (and mimicking endogenous TCR expression), CAR T-cell exhaustion can be delayed or avoided and antitumor efficacy enhanced.
Certainly, one may conceive of using pharmacologic agents to inhibit or reshape the negative consequences of CAR tonic signaling such as exhaustion and terminal differentiation. The former can potentially be reversed utilizing mAb inhibitors of upregulated immune checkpoints, such as PD-1, LAG-3, or TIM-3 (162). Attempts have been made to address the latter using various strategies. These include activating the canonical Wnt/β-catenin pathway (which has been shown to promote a TSCM phenotype) by inhibiting glycogen synthase kinase 3 beta (GSK3β), a serine/threonine kinase implicated in β-catenin degradation, either alone (163) or during ex vivo culture with IL7, IL21, and CD8+/CD62L+/CD45RA+ streptamer–based serial-positive selection (164); inhibiting glycolysis using 2-deoxyglucose (165); and remodeling mitochondrial function to replicate a memory T-cell metabolic phenotype characterized by oxidative phosphorylation (OXPHOS) and fatty acid oxidation (FAO; ref. 166). Separately, the phosphatidylinositol-3-kinase (PI3K)/Akt pathway has been implicated in T-cell memory formation. For example, Akt has been shown to phosphorylate and sequester the Forkhead box O (FOXO) transcription factors, blocking the transcription of molecules associated with less differentiated T cells [such as CD62L, CCR7, and IL7 receptor-α (IL7Rα or CD127); ref.167]. In parallel, Akt inhibitors have been demonstrated to improve the in vitro expansion of minor histocompatibility antigen-specific CD8+ T cells with a minimally differentiated early memory phenotype, correlating with improved long-term persistence, and a superior graft-versus-tumor effect in mice following adoptive transfer (168). More recent work by Klebanoff and colleagues has demonstrated that the inhibition of Akt using an allosteric kinase inhibitor during the ex vivo expansion of CAR and TCR retroviral-transduced T cells decouples differentiation from expansion, enhancing the intranuclear localization of FOXO1. These cells exhibited a CD62L+ early memory phenotype, suppressed glycolysis, and superior antitumor efficacy (169). Likewise, inhibiting the PI3Kδ catalytic subunit p110δ using the small-molecule selective inhibitor Idelalisib (formerly CAL-101), can promote a strong undifferentiated memory phenotype in both murine and human CD8+ mesothelin-directed CAR T cells (as well as pmel-1–directed transgenic TCR T cells) characterized by the upregulation of transcription factor 7 (Tcf7) and elevated surface expression of CD62L, CCR7, and CD127 (170). In vivo, idelalisib-exposed CAR and transgenic TCR T cells persisted longer following ACT and induced greater tumor regression compared with the traditionally expanded CD8+ controls. Inhibition of Akt may also be anticipated to reverse metabolic dysfunction caused by CAR tonic signaling by blocking the negative regulation of peroxisome proliferator-activated receptor gamma coactivator 1-alpha (PGC-1α) caused by chronic Akt activation (171). PGC-1α is known to enhance T-cell mitochondrial oxidative metabolism and promote mitochondrial biogenesis. Indeed, inhibiting Akt during ex vivo TIL expansion has also been shown to confer a memory T-cell metabolic profile with increased rates of OXPHOS and FAO, leading to enhanced in vivo persistence and improved antitumor immunity (172). Engineering T cells to overexpress PGC-1α may be an alternative strategy to optimize its beneficial effects on effector T-cell metabolism (171). Finally, CAR activation has been shown to induce expression of the adenosine A2A receptor (A2AR); (173), which is able to exert potent negative feedback on CAR function via an interaction with adenosine, which is upregulated in the TME of many tumors (174). Inhibition of A2AR either with a small-molecule antagonist or using a short hairpin RNA (shRNA) could reverse CAR suppression, and was found to be particularly synergistic with anti-PD-1 therapy in a HER2+ breast cancer model (173).
Other T-cell engineering strategies
In addition to the insertion of a CAR, T cells can be further engineered to optimize downstream signaling, express costimulatory molecules, or secrete cytokines. All these approaches may be utilized to ameliorate the negative consequences of tonic signaling. Possible strategies include overexpressing 4-1BBL in combination with a 28ζ CAR (10); knocking out Cbl-b, an E3 ubiquitin-protein ligase that promotes anergy by regulating PI3K access to CD28 (175, 176), in combination with a BBζ CAR; overexpressing PGC-1α to enhance OXPHOS and mitochondrial biogenesis (171); or expressing a cell-surface tethered IL15/IL15R fusion protein, which has been shown to encourage a CD45RO− CCR7+ CD95+ TSCM phenotype (42).
Providing ligand for ligand-dependent tonic signaling
Finally, a common finding when utilizing CARs in vivo (particularly in the case of solid tumors) is that they fail to persist over time. Aside from issues of sequestration, AICD, or exhaustion, one reason is that, unlike the case in hematologic malignancies where CARs and target antigen are in close proximity, they may simply fail to encounter ligand in sufficient quantity or frequency to expand. While repetitive CAR delivery can be feasible in certain circumstances, issues with immunogenicity and/or the toxicities associated with lymphodepleting conditioning can complicate matters. Regional or intratumoral CAR T-cell injection (127, 177) may also be tried but is unlikely to foster a systemic response in the case of metastatic disease. Various groups have attempted to utilize virus-specific (e.g., CMV or EBV) T cells for CAR transduction (178, 179), with the dual aims of reducing GvHD following allogeneic use and enhancing in vivo expansion and persistence following interactions with DCs presenting viral epitopes in previously infected patients. However, T cells dually activated via a CAR and their endogenous TCR may be liable to become anergic and exhausted, undergo AICD, and exhibit poor persistence in vivo. This has been highlighted in a murine allogeneic hematopoietic stem cell transplantation model using CD28-costimulated CD19-directed CAR T cells (180) and more recently in an immunocompetent syngeneic mouse model of CD19+ B-cell ALL, where engagement of the TCR with target antigen was found to have a deleterious impact on CD8+ (but not CD4+) CAR T-cell efficacy, mediated by exhaustion and apoptosis (181).
Other groups are exploring the delivery of autologous T-cell antigen-presenting cells (T-APC) expressing truncated ligands (such as CD19) to mimic antigen presentation and engender persistence. This technique is already being evaluated within the phase I PLAT-02 trial at Seattle Children's Hospital (Seattle, WA; ref. 182). Conceivably, autologous DCs or irradiated engineered autologous tumor cells (EATCs; ref. 183) could be loaded with ligand and introduced intradermally or intranodally to enhance CAR expansion and persistence. APCs are particularly promising as they provide a whole gamut of signals that could stimulate and coordinate CAR (or indeed endogenous nonengineered) T-cell antitumor efficacy. Indeed, one could potentially conceive of CAR “service stations” utilizing APCs modified to upregulate surface adhesion molecules (e.g., ICAM-1) or costimulatory ligands (CD80/86, 4-1BBL, or CD40L) and secrete cytokines (IL7 or IL15) or chemokines (such as CXCL-9 or CXCL-10). All these strategies are summarized together in Fig. 5.
Outstanding Questions and Conclusions
The preclinical data regarding CAR tonic signaling is at times conflicting and contradictory. While ligand-dependent tonic signaling can potentially be coopted to mimic endogenous T-cell/DC interactions and improve in vivo expansion and persistence, ligand-independent signaling appears to be far less benevolent. However, is the latter always detrimental? Certainly, data generated from the majority of experiments using non-4-1BB–containing CARs would suggest so (18, 20, 21). However, the picture is undoubtedly more complex with at least two reports suggesting that 4-1BB or CD28-mediated tonic signaling may confer improved ex vivo expansion and enhanced in vivo efficacy and persistence (62, 136). With regard to 4-1BB, several other reports suggest the contrary (19, 63, 73) and it appears that at least a portion of the blame can be attributed to the viral vector and promoter. It may also be the case that 4-1BB tonic signaling is occurring at different time points and/or spatial compartments. Indeed, other models have suggested that early acute 4-1BB signaling can have a deleterious impact on T-cell function and survival.
In addition, while scFv domain swapping is highly likely to be a common initial event in almost all cases of ligand-independent tonic signaling, it remains to be seen whether this may be triggering both phenotypes of 4-1BB–mediated tonic signaling. Indeed, numerous experiments utilizing identical anti-CD19 (FMC63) scFvs with either a CD28 or 4-1BB costimulatory domain have not demonstrated ligand-independent tonic signaling (18, 63).
Based upon the available literature, we have highlighted four overlapping models of ligand-independent CAR tonic signaling. Model (i) is characterized by continuous proliferation, terminal effector differentiation, and cell senescence and appears to rely upon high CAR surface expression (see Fig. 3A and E). Changes to the promoter or the spacer have reversed this phenotype. Model (ii) is characterized by CAR T-cell exhaustion and may be reversed by the substitution of CD28 with a 4-1BB ICD (see Fig. 3B). The role of the CAR spacer and TMD is less clear here. Model (iii) appears to be mediated by 4-1BB costimulation and is characterized by enhanced (but not continuous) proliferation during ex vivo expansion and greater persistence and efficacy in vivo, without evidence of AICD (see Fig. 3C). Model (iv) also appears 4-1BB–mediated, but is characterized by poor expansion ex vivo, the upregulation of death receptors and their ligands, and enhanced AICD (see Fig. 3D). This phenotype appears primarily to be due to a 4-1BB–mediated positive feedback loop occurring at the level of the γ-retroviral LTR promoter. Interestingly, however, reversal was also seen following the introduction of a CD28 ICD upstream of 4-1BB suggesting that localization of the latter with respect to the cell membrane surface and associated signaling molecules may also be playing a role.
Clearly, considerable work remains to be untaken to uncover the precise function of second- and third-generation CARs at the molecular level, with emphasis placed upon determining how factors such as the costimulatory ICD's relative distance to the cell membrane and their accessibility to downstream adapter and signaling transduction proteins may impact function. The precise contribution of other structural components (such as the spacer and TMD) to tonic signaling also remains to be elucidated, as does the relative importance of scFv instability, promoter strength, and dysfunctional CAR recycling. Furthermore, the use of multiple CARs in parallel for logic gated signaling or the combined use of chimeric costimulatory receptors (CCR) may be anticipated to increase the relative risk of encountering ligand-independent tonic signaling. A further question is whether CARs targeting membrane-proximal epitopes confer greater risk of tonic signaling due to the requirement for longer spacers, increasing the promiscuity of scFv domain swapping. Protein engineering using computational modelling may prove highly effective at stabilizing the scFv VH:VL interface. Alternative strategies employing endogenous ligands, using camelid-derived nanobodies or fibronectin-based targeting moieties may also prevent ligand-independent oligomerization.
If low-level ligand-dependent tonic signaling can be beneficial, it may be beneficial in a CAR context to recapitulate the model whereby endogenous naïve peripheral T-cell TCR/DC self-peptide MHC interactions are able to elicit more efficacious killing of foreign peptide-containing target cells. An important question is how can this be optimized to minimize toxicity, CAR T-cell exhaustion, and/or AICD? If the latter is unavoidable when targeting solid tumors due to a lack of available TSAs, how might this process be minimized or reversed? Certainly, the use of a constitutively expressed 4-1BBL may provide a degree of tonic signaling for activated CAR T-cells in a juxtacrine manner. Likewise, the use of blocking antibodies targeting inhibitory molecules, dominant negative inhibitory receptors (184), or switch CARs (185) to reverse inhibitory signaling may temper tonic signaling–induced exhaustion. Metabolic dysfunction may be targeted with small-molecule inhibitors of signal transduction proteins or by engineering the CAR T-cell itself to function with a TCM or TSCM metabolic phenotype. Recapitulating endogenous TCR expression and recycling by targeting CAR expression to the TRAC locus appears to be a highly promising and efficacious method of CAR transduction. It also facilitates the clean engineering of allogenic CARs lacking functional TCR. These, however, may perform less well in the absence of endogenous self-peptide MHC/TCR interactions conferring a lower activation threshold, particularly when targeting epitopes with low cell surface density. However, by utilizing scFvs, the majority of CARs currently in development have binding affinities several orders of magnitude greater than TCR/MHC binding (186), so although more CARs are required to bind to mediate cytotoxicity, basal tonic signaling may not be so crucial for CAR versus TCR-mediated killing and may worsen on-target, off-tumor toxicity. Other interactions with APCs (e.g., via CD40L/CD40, ICAM-1/LFA-1) will also likely prove useful for CAR efficacy and persistence in vivo and could be recapitulated using T-APCs, engineered APCs, or irradiated EATCs. All these hypotheses remain to be supported by empirical evidence.
The use of engineered cellular therapies to target cancer is rapidly evolving yet clearly, we remain in a nascent period of development. With the emergence of powerful gene editing techniques that can help uncover the hidden mechanisms of CAR function and aid precise engineering, this process is likely to accelerate considerably.
Disclosure of Potential Conflicts of Interest
No potential conflicts of interest were disclosed.
The views expressed are those of the authors and not necessarily those of the NHS, the NIHR or the Department of Health.
A.A. Ajina is supported by the Medical Research Council (grant reference number: MR/R001936/1). J. Maher is supported by the Wellcome Trust (grant reference number: 104802/Z/14/Z); the Medical Research Council (grant reference number: MR/M024733/1); Cancer Research UK (Grant reference number: C11499/A21623); the British Lung Foundation (grant reference number: MPG16-6); Pancreatic Cancer UK, the Experimental Cancer Medicine Centre at King's College London, the King's Health Partners Cancer Research UK Cancer Centre and by the National Institute for Health Research (NIHR) Biomedical Research Centre based at Guy's and St Thomas' NHS Foundation Trust and King's College London.