A simple but effective way of regulating CAR T cells consists of transiently transfecting T cells with CAR-encoding mRNA (18–23). Due to the lack of genomic integration, CAR expression is limited by the degradation of the CAR-encoding mRNA and dilution following each T-cell division (18). The result is a steady decrease in CAR-expressing T-cell numbers, unless new cells are infused. Repeated infusions are however associated with a higher risk of an anaphylactic reaction due to the CAR T cells (24). While the inherently limited persistence of these CAR T cells might compromise continued anti-leukemic effect (25), it also limits long-term hematologic toxicities and off-target effects.

Lowering the binding domain's affinity toward the targeted antigen aims to prevent on-target, off-tumor toxicities from arising in the first place (26, 27). While affinity-tuned CARs retain the ability to bind to cancer cells with a high antigen expression, healthy tissues with lower expression are spared (28). The use of low-affinity CARs is therefore especially interesting when targeting antigens known to be expressed on healthy tissue in low amounts, e.g., HER2 or EGFR (26, 27). This, however, might also lead to cancer cell escape variants with low antigen expression (29). Additionally, both promoter usage and transduction level of T cells might result in heterogeneous expression of the CAR protein, making it hard to ensure consistent behavior among individual CAR T cells as their avidity toward the antigen can vary. One promising strategy to overcome heterogeneous CAR expression is to instead integrate the CAR construct into the endogenous TCR alpha chain (TRAC) locus using the CRISPR/Cas9 system (30).

Depletion of CAR T cells can be achieved by designing CAR constructs that also express a suicide gene, such as inducible Caspase 9 (iCasp9) (31–39), herpes simplex virus tyrosine kinase (HSV-TK) (40, 41) or human thymidylate kinase (TMPK) (42). In cells expressing iCasp9 and TMPK, elimination is achieved through activation of the caspase 3 apoptotic pathway when a small molecule is administered. The iCasp9 system has successfully been validated in patients receiving haploidentical stem-cell transplants (HSCT) in which iCasp9-expressing T cells were rapidly removed at onset of graft-versus-host disease (GvHD) (31). Likewise, administration of ganciclovir to T cells co-expressing HSV-TK causes formation of a toxic metabolite but cell death may take up to several days as it depends on cell proliferation (40, 41). The use of HSV-TK is severely limited by the high immunogenicity of the virally-derived protein (43). Furthermore, the HSV-TK suicide system is complicated by the fact that ganciclovir is used as a first-line treatment against cytomegalovirus (CMV) infection, a virus which is often reactivated in HSCT and other immunocompromised patients (44). As TMPK and iCasp9 are of human origin, the risk of immunogenicity is low. Indeed long-term engraftment up to several years in patients infused with iCasp9 expressing cells has been reported (45).

Co-expression of a cell-surface elimination marker, not normally present on T cells, allows for antibody-mediated degradation and control of the CAR T cells (22, 46–52). By utilizing clinically approved antibodies, e.g., rituximab targeting CD20 (48–50) or cetuximab targeting EGFR (51, 52), complement- or antibody-dependent cytotoxicity (CDC/ADCC) can be achieved toward the CAR T cells (51). Choosing a marker that is co-expressed on cancer cells, allows this method to create additional tumor killing, with the caveat that further collateral toxicity might arise. Cell-surface markers also allow for positive selection of transduced T cells in the manufacturing process, and subsequent monitoring of CAR T-cell levels in vivo. However, the efficacy of the strategy can be compromised by the fact that CDC/ADCC capacity is limited in patients treated with chemotherapy prior to CAR T infusion (53). In addition, antibodies can have limited biodistribution and tissue penetration, especially in poorly vascularized tumors (54). In order to address these problems, researchers have instead created anti-idiotype CARs recognizing murine CD19-specific CARs (55) or incorporated a short peptide epitope, called an E-tag, into the extracellular domain of the CAR, and created anti-E-tag CARs which could then be used to eliminate the anti-tumor CARs (56).

Since the use of suicide genes and elimination markers result in irreversible depletion of this complex treatment, researchers have developed a number of reversible methods to control CAR T cells as well.

Current methods for controlling CAR T cells include systemic immunosuppressive agents, e.g., corticosteroids (57). The lymphocytotoxic anti-CD52 antibody alemtuzumab, has also been proposed as a method of depleting CD4- and CD123-specific CAR T cells (47, 58), as targeting these proteins might cause hematological aplasia and toxicity. More recently, it was shown that CAR signaling could be inhibited using the tyrosine kinase inhibitor dasatinib. Dasatinib inhibits phosphorylation of lymphocyte-specific protein tyrosine kinase (LCK), a critical component in the T-cell signaling pathway. Preclinical studies suggest that treatment with dasatinib can reversibly inhibit CAR T-cell proliferation and cytokine production without negatively affecting viability (59, 60). Although dasatinib cannot adequately inhibit already activated CAR T cells, limiting its usage against acutely arising CRS or neurotoxicity, the drug was shown to be superior to dexamethasone in inhibiting further activation in a preclinical study (59). Finally, while corticosteroids and alemtuzumab cause widespread inhibition or complete elimination of both CAR T cells and healthy lymphocytes, dasatinib has the advantage of acting as a faster on/off switch, due to its short half-life of 4 h (61).

Aiming to specifically control CAR T-cell activity toward the antigen, several models of adapter-mediated CARs, also known as universal CARs, have been developed (62–71). A shared feature is their method of tumor recognition, which is achieved by linking an adaptor, a molecule recognized by the CAR, to an antibody or ligand that recognizes the tumor antigen. While current clinically approved CARs are designed to be constitutively active, adapter-dependent CAR T cells can only recognize and kill when the adapter is administered, allowing for titratable and reversible control of the CAR T cells. A major advantage of this approach is the ability to target different antigens without the need to re-engineer and re-transfuse T cells. Adapters have also been designed to redirect anti-CD19-specific CARs to another target, using CD19-fusion proteins, suggesting that adapter proteins might be used with current clinically approved CAR T cells (72). However, large differences in adapter kinetics and subsequent effects on CAR T cells have been reported, probably reflecting differences in models used, affinity differences between the adapter and both target and CAR T cells, and biodistribution of the adapter molecules.

Instead of directly regulating CAR binding to antigen, pharmacological inducers can also be used to control the activity of CAR T cells themselves, by splitting the CAR's extracellular antigen-binding domain from its intracellular signaling domains (73, 74). Assembling of the fully functional CAR is therefore dependent on administration of a dimerizing drug, limiting the CAR activity by the half-life of the drug. Consequently, CAR T-cell activation requires two inputs: the tumor antigen and the dimerizing drug. The split-CAR can thus also be considered an AND gate CAR (see below). The split-CAR allows temporal and reversible control over the number of functional CARs, but the design does not prevent on-target, off-tumor toxicities as no spatial control is achieved due to a lack of control over the distribution of the drug.

Strategic incorporation of the autocleaving hepatitis-C-derived NS3 protease in the CAR construct has also been suggested as a way to control CAR T-cell activity. Juillerat et al. incorporated the NS3 protease between the CAR and a degradation moiety, thereby tagging the CAR construct for degradation when a NS3 protease inhibitor was administered (75). This construct showed reversible as well as tunable control over CAR T-cell cytotoxicity in vitro. As the NS3 protease is virally derived, it holds immunogenic potential which could potentially limit CAR T-cell persistence.

It has also been suggested to regulate CARs at the transcriptional level. Several groups have shown that a drug-inducible CAR system can be generated by controlling CAR transcription using the TET-on system, allowing reversible control of CAR T cells (76–78). CAR mRNA is thus only produced in the presence of doxycycline, although some background CAR expression was observed (76). As the TET-on system is derived from both bacteria and virus, it holds significant immunogenic potential with the risk of host mediated elimination of the CAR T cells. Another drawback is the lack of rapid control, should life-threatening side effects occur, due to the control occurring on a transcriptional level. However, for highly proliferative CAR T cells that are constantly diluting the CAR protein amongst daughter cells, transcriptional regulation may be sufficient to limit the quantity of functional CAR complexes.

One approach to enable better decision-making in CARs is the incorporation of logic AND gates, such that a combination of antigens are required for activation. Often this dual CAR design consists of two extracellular domains, with specificities toward different antigens, each coupled to separate components of the intracellular stimulatory apparatus, e.g., CD3ζ and CD28 or 4-1BB (80–83). Such approaches have been tested in preclinical prostate and breast cancer models and might allow for targeting of proteins that are also present in healthy tissue (81, 82). A worry, however, is that even partial signaling through one receptor may generate sufficient T-cell activity to cause off-target damage (81, 83). Another approach has therefore been to use one receptor exclusively as a priming signal, with no activating signaling capacity itself. This was achieved using so-called SynNotch receptors, which were coupled to orthogonal transcription factors that were released upon binding. This “priming” leads to expression of a fully functional CAR targeting another cancer-associated antigen, ensuring localized activity of the CAR T cell (84, 85). Importantly, in mouse models the SynNotch CARs did not migrate, but retained their function only in dual positive tumors, indicating that this approach ensured good spatial control of the CAR (84, 85).

Better discrimination between malignant and healthy cells can also be achieved by designing an inhibitory CAR (iCAR). The iCAR contains a binding domain specific for an antigen expressed on healthy cells fused to the signaling domains of CTLA-4 or PD-1, such that recognition of the healthy antigen leads to an inhibitory signaling cascade that overrides activating signals by dephosphorylating the receptor complex (86). The iCAR should restrict CAR T-cell activity to tumor tissue lacking the healthy antigen, limiting on-target, off tumor activity. Such a system might allow for CAR T cells previously shown to cause lethal off-tumor activity, such as ERBB2-specific designs, to be re-introduced.

While the AND gate and iCARs restrict CAR T-cells' activity spatially, they cannot control the intensity of the CAR T-cell activity nor control them in a temporal manner.

In order to gain better temporal control and limit CAR T-cell activity to the tumor microenvironment (TME), CAR expression can be controlled by incorporating a hypoxia inducible factor (HIF) in the CAR construct (87). The CAR-HIF construct is continuously targeted for degradation when the CAR T cell is present in a normoxic environment, i.e., most healthy tissues, ensuring that CAR expression only occurs under hypoxic conditions, as seen within parts of the TME. Because degradation of the CAR is regulated at the protein level, control is thought to occur quickly, which is favorable to avoid CAR activity outside of the tumor. This method has only been tested in vitro, and may fail to eradicate tumor cells residing in normoxic tissues, e.g., the peripheral parts of the tumor. Moreover, these CAR T cells may show off-target effects in healthy, hypoxic tissues like the bone marrow.

Another method to constrict CAR activity to the TME is by designing a masked CAR, as proposed by Han et al. (88). Here the CARs antigen-binding site is hidden by a masking peptide with a linker sensitive to proteolytic cleavage. Tumor associated proteases present in the TME can then cleave the linker, removing the masking peptide and allowing CAR T cells to target antigen presenting cells. However, a possibility remains that endogenous proteases cleave the masking peptide, paving the way for on-target, off-tumor toxicities.

One of the central cytokines involved in CRS is IL-1. Giavridis et al. recently designed a CAR T cell constitutively producing IL-1 receptor antagonist, protecting mice against CRS associated mortality without affecting the anti-tumor efficacy (9). One major advantage of using IL-1 receptor antagonist is its ability to cross the blood-brain barrier, thereby potentially reducing CAR T-cell-related neurotoxicity (89).

While the above mentioned logic gates and hypoxia sensitive CARs seek to enhance tumor specificity, they do not allow the clinicians control over the CAR T cells and as such offer no solution should life-threatening toxicities arise.

YM and MB have filed a patent relating to a technology presented in this manuscript. The remaining authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.