Small-molecule control of CAR T cells

Abstract

Chimeric antigen receptor (CAR) T cell therapy is a ‘living drug’ in which the T cells of patients are genetically engineered with an artificial receptor that directs them to attack diseased cells. CAR T cell therapies have had remarkable impact, curing subsets of patients with previously untreatable, late-stage cancers. However, limitations persist, including severe toxicities, limited survival of engineered cells, and therapeutic resistance. Genetically encoded small-molecule control systems have been developed to address these limitations. They can halt toxicities by eliminating CAR T cells or switching off their function. Furthermore, they can enhance therapy by directly targeting antigens or broadening cell killing ability through cytotoxic pro-drug activation. Small-molecule controllers include protease inhibitors, protein dimerizers, protein degraders, bi-specific adaptors and conditionally activated chemotherapeutics. Here, we outline small-molecule-based control approaches, categorizing them by function and detailing their molecular mechanisms. We emphasize systems in the clinic and highlight emerging applications and unmet areas.

Graphical Abstract

Introduction

Engineered cell therapies are an emerging class of therapeutics capable of sophisticated functions transcending those of traditional molecular drugs. Cells are naturally equipped and can be further engineered to sense various biological and chemical input stimuli and respond with complex output behaviours. Immune cells are of particular interest as they can migrate throughout the body and naturally have a role in combating disease¹. Indeed, the first engineered cell therapies to achieve FDA approval were chimeric antigen receptor (CAR) T cell therapeutics^2–5 (Fig. 1a, Box 1). CAR T therapies targeting B cell leukaemias and lymphomas and multiple myeloma have transformed the treatment of these cancers^6–8. Patients who did not respond to standard treatments and had a dismal prognosis are now showing high rates of clinical response (40–74% complete responses) including durable cures^6–8. Additionally, as a ‘living therapy’, CAR T cells can persist and exert therapeutic activity >10 years after treatment in some patients⁹. Beyond approved agents, CAR T cell therapies are in development to target a wide variety of diseases including solid tumours, autoimmunity, viral infections and heart failure^10–13.

Fig. 1 |. Introduction to CAR T cells, shortcomings and small-molecule control solutions.

a, Chimeric antigen receptor (CAR) T cells are manufactured by viral gene transfer of the CAR gene. The CAR consists of a single-chain variable fragment (scFv) antigen binding domain fused to T cell signalling domains. The patient is infused with CAR T cells. Upon binding to the antigen, the CAR mediates antitumour function. b, Challenges to CAR T cell therapy broadly include unwanted toxicity and limited therapeutic efficacy. Many of these challenges are being addressed by one or more small-molecule control methodologies as noted by the connecting lines. CD3, cluster of differentiation 3.

Box 1 |. Basics of CAR T cell therapy.

Chimeric antigen receptor (CAR) T cell therapy is an immune cellular therapy in which T cells are genetically modified to express the CAR and are infused into a patient to mount an immune attack against diseased cells^2–5. CAR T cell therapies are of high interest as they have shown strong antitumour effects against advanced haematological cancers wherein other therapies such as chemotherapy and targeted drugs have failed^6–8. Additionally, they tap into the natural way of the body to eliminate diseased cells and have the potential for ‘immunological memory’ to provide continued protection against cancers from re-occurring. This behaviour is similar to the ability of our bodies to eliminate infections more readily following previous infection or vaccination.

CARs are engineered cell surface receptor proteins that direct immune cell (T cells for CAR T cell therapy) signalling and effector functions in response to a cell surface antigen on a neighbouring target cell (Fig. 1a). CARs are chimeric in that they consist of disparate parts fused together. The most distal part is the extracellular antigen binding domain responsible for recognizing the antigen on the target cell. This domain is fused via an extracellular spacer region responsible for establishing the distance of the binding domain from the T cell membrane, a key parameter for signalling¹⁸⁶, to the transmembrane domain that situates the receptor in the membrane. On the intracellular side are cytoplasmic signalling domains. These protein domains are responsible for activating the T cell receptor (TCR) and co-stimulatory signalling pathways^5,187,188. Most CAR designs use a single-chain variable fragment (scFv) from an antibody as the antigen targeting domain and the ‘CD3ζ’ cytoplasmic domain from the TCR as the signalling domain. Most designs also contain a co-stimulatory cytoplasmic signalling domain, for example, CD28 or 4–1BB, which provides pro-survival and expansion signals to the T cells^189,190. Binding to the target antigen on a neighbouring cell leads to mechanical force generation and CAR clustering on the surface of the T cell¹⁹¹. This clustering triggers a kinase phosphorylation cascade of T cell receptor and co-stimulatory signalling proteins¹⁹². These events lead to calcium flux, target cell lysis, and proliferation of the activated CAR T cell. It also induces a gene expression programme including expression of inflammatory cytokines such as interferon-γ and tumour necrosis factor¹⁹². T cell receptors recognize their target antigens in the context of the major histocompatibility complexes requiring human leukocyte antigen (HLA)-type matching — similar to donor matching of recipients for organ transplantation. By contrast, CARs directly bind to the cell surface antigen and, thus, do not depend on HLA matching for their function^11,193.

CAR T cell therapy is referred to as an ‘adoptive cell therapy’as CAR T cells are manufactured outside of the body and then adoptively transferred back into a patient^194,195. Different strategies for CAR T manufacturing are in development, but current approved therapies all undergo a related process. This process involves leukapheresis harvesting of the blood of a patient and isolation of T cells. Then, the CAR gene is transferred into the cells of the patient by retrovirus or lentivirus transduction. Cells then undergo ex vivo expansion (~100–1,000-fold) and are infused into the patient^196,197.

Currently, there are seven FDA-approved CAR T cell therapies. Five of these therapies target CD19 for treating relapsed or refractory B cell leukaemias and lymphomas, and two of them target BCMA for treating relapsed or refractory multiple myeloma¹⁹⁸. There is a major interest in applying CAR T cell approaches to solid tumours as they make up the majority of cancer cases¹⁹⁹. There are also several clinical studies showing remarkable clinical responses in severe autoimmune diseases^176,177.

Despite the strengths and promise of CAR T cell therapy, several challenges limit the effectiveness of approved treatments and hamper expansion to additional disease indications (Fig. 1b). Broadly, these roadblocks can be divided into the two categories of safety and therapeutic efficacy. Toxicities and side effects of CAR T cell therapy remain a major hurdle to overcome³. Systemic toxicities resulting from high levels of immune activation include the following: cytokine release syndrome, a systemic inflammatory response that can lead to multi-organ system toxicity; cytokine storm, mediated by nonspecific T cell activation leading to interferon-γ and tumour necrosis factor release^14,15; neurotoxicity^14,15, including CAR T-related encephalopathy syndrome^16,17; hemophagocytic lymphohistiocytosis^18,19; and cytopenias^20,21. Agents targeting interleukin-6 (IL-6) and its receptor, siltuximab or tocilizumab, respectively, are routinely used to block cytokine-release syndrome²². However, many toxicities can be fatal and remain a major concern.

Another source of toxicity is the CAR T cell attack of normal cells expressing the target antigen, referred to as ON-target–OFF-disease toxicity²³. As an example, approved anti-CD19 CAR T therapies target normal B cells that express CD19 in addition to tumour cells. This activity leads to B cell aplasia, a transient loss of B cells that can be managed with pooled immunoglobulin administration^3,24. Unfortunately, other cases of CAR T cell-mediated ON-target–OFF-disease toxicity can be less benign and have led to organ dysfunction and death^15,25–28.

Lastly, there has been recent concern for secondary malignancies resulting from CAR T cell therapy including potential malignant transformation of engineered cells²⁹. Observation of secondary malignancies in patients who received CAR T cell treatment led to an FDA box warning on approved treatments. However, several large studies have now found no evidence of CAR T cell transformation^30–33. Furthermore, recent data suggest that CAR treatment may not lead to increased incidence of secondary malignancy³³. Notably, one therapy delivered via a non-viral transposon-based vector has led to malignant transformation³⁴. This result highlights the importance of safety evaluations for new delivery modalities.

The challenge of attaining sufficient therapeutic potency is intertwined with the issue of toxicity. A primary variable for potency is the number of CAR T cells present at the diseased site(s)³⁵. This number is affected by the administered cell dose and by the proliferation and survival of CAR T cells in a patient. Although CAR T cells traditionally proliferate upon activation, cells that are chronically activated, over-activated or activated without the proper accessory co-signalling can undergo programmed cell death or cell exhaustion^36,37. One major source of chronic activation is ‘tonic signalling’ in which CARs cluster at low levels in the absence of antigen. This leads to basal activity of the receptors and early exhaustion of the cells^38,39. Additionally, immunosuppressive signalling from the diseased cells or from accessory immunosuppressive cells can directly inhibit the functions of CAR T cells⁴⁰. However, another major challenge affecting CAR T therapeutic efficacy is antigen loss or heterogeneity within tumour tissue. These issues lead to incomplete eradication of diseased cells. Much like the therapeutic failure observed with targeted small-molecule therapeutics, resistance mechanisms resulting from target gene mutations and underlying heterogeneity have been frequently observed in patients who experience disease recurrence after CAR T cell therapy^41,42.

The unique one-time administration of CAR T cell therapy further complicates the ability of a physician to address CAR T cell toxicity. Unlike conventional pharmacologic agents that afford granular pharmacokinetic control through variable dosing and predictable metabolism, CAR T cell therapies are ‘living drugs’ that expand in vivo and can have variable cell persistence between patients^43,44. On top of this, once a patient receives a CAR T cell infusion, there is no currently approved way for physicians to selectively modulate the behaviour of the engineered cells in vivo or to remove them.

Consequently, a major focus of next-generation CAR T cell designs is the development of technologies that afford prompt and tunable control over CAR T cell function following infusion⁴⁵. In general, these systems consist of one or more proteins in the CAR T cells that are activated or inhibited by an input stimulus provided by a clinician. ‘Remote control’ systems responsive to various input stimuli have been developed, including small molecules, mechanical forces^46,47, magnetism⁴⁸, heat^49–53 and light^54–57. These broad approaches have been recently reviewed⁵⁸.

Small-molecule control systems have gained particular interest owing to their many favourable characteristics. Small molecules are the cornerstone of modern pharmacology with predictable kinetics and biodistribution. A catalogue of small molecules exists that can carry out diverse functions including viral protease inhibitors, protein dimerizers, proteolysis-targeted chimeras (PROTACs), heterobifunctional adaptors, and conditionally activated chemotherapeutics and kinase inhibitors. There are also many already FDA-approved small-molecule drugs that can be re-purposed for CAR T cell control by incorporating known interacting protein domains into system designs. Compared to other input modalities such as light, heat or magnetism, small-molecule drug administration is relatively simple. There is no need for complex apparatuses, and there is the possibility for oral administration. The breadth of the small-molecule toolbox also enables the possibility for multi-input and orthogonal control of multiple cell behaviours through combinatorial application. Figure 2 highlights several of the small molecules used to date, including their structure, molecular function, target protein and application. Supplementary Table 1 contains a list of compound names with PubChem compound identifiers and Chemical Abstracts Service Registry Numbers. The goals of small-molecule systems that have been developed can largely be broken into four major categories: CAR T cell elimination, control of CAR presence or assembly, antigen targeting, and control of accessory gene function (Fig. 1b).

Fig. 2 |. Chemical structures of small molecules used to control CAR T cells.

Structures are shown for select small molecules discussed in this Review. Small molecules are grouped by function. The following additional information is provided: compound name (year of FDA approval), interacting protein domain(s), and the type of small-molecule chimeric antigen receptor (CAR) system application (indicated by coloured asterisks as defined in the figure key). This information is also provided (in light grey text) for molecules whose structures are not shown because of their similarity with other displayed molecules (in black text). Bcl-XL, B cell lymphoma extra-large protein; Bim, Bcl-2 interacting mediator of cell death; BRD4, bromodomain-containing protein 4; CRBN, cereblon protein; DUPA, 2-[3-(1,3-dicarboxypropyl)ureido]pentanedioic acid; eDHFR, Escherichia coli dihydrofolate reductase; ERT2, tamoxifen-responsive oestrogen receptor domain; FITC, fluorescein isothiocyanate; FKBP12, 12-kDa FK506-binding and rapamycin-binding protein; FOLR1, Folate Receptor 1; FRB, FKBP-rapamycin-binding domain; GAI, Gibberellic acid insensitive protein; GID1, Gibberellin Insensitive Dwarf1 protein; HCV, hepatitis C virus; HSV, herpes simplex virus; IZF3, IKAROS family zinc finger 3; NS3, nonstructural protein 3; POM, pomalidomide; PSMA, prostate-specific membrane antigen; scFv, single-chain variable fragment; sdAb, single-domain antibody; ^tBu, tert-butyl; TetR, tetracycline repressor protein; TetRB, tetracycline repressor protein B; TMP, trimethoprim.

Ideal systems share some common characteristics. They incorporate physiologically inert small-molecule inducers with favourable pharmacokinetics and pharmacodynamics, are titratable and reversible, and demonstrate effector functions at least as well as standard CAR T cells. Small-molecule CAR T cell control technologies are rapidly advancing, with several approaches being evaluated in the clinic (including NCT03373097, NCT03016377 and NCT02744287)^59–61. Supplementary Table 2 contains a list of registered clinical studies at the time of publication.

Here, we review current and emerging technologies for small-molecule control of CAR T cells. This Review is organized based on system goals. In each section, we cover the molecular system designs, the mechanisms of small-molecule-based control, and technological capabilities of each approach. We also discuss suitability for clinical translation including current limitations. We cover emerging areas wherein current small-molecule systems can be applied in the future, and key areas of need for new synthetic chemical development.

Elimination of CAR T cells

The first small-molecule control systems applied to CAR T cells were ‘suicide switches’ with the goal of specifically eliminating CAR T cells upon the administration of a pharmacologic agent⁴⁵ (Fig. 3a). The impetus for this approach is to serve as a safety switch to allow for an abortive remedy of life-threatening CAR T cell-mediated toxicities. An ideal suicide switch would be rapid, allowing for immediate cessation of further toxic effects after initial toxicity is detected, specific, in that it eliminates CAR T cells without affecting other cells, and titratable, potentially allowing for partial cessation of effects to tune activity allowing for some CAR T cells to remain. Overall, given the permanent destruction of CAR T cells by the suicide switch approach, which will likely lead to irreversible attenuation of the antitumour response, it is a rescue mechanism best reserved for catastrophic events. Thus, suicide switches have gained special interest as fail-safes for testing novel CAR T cell approaches with unknown toxicity profiles.

Fig. 3. Small-molecule approaches to eliminate CAR T cells.

| a, Concept of targeted CAR T cell depletion. CAR T cells are engineered with a gene construct that makes them sensitive to killing by a small-molecule drug while non-engineered endogenous cells are spared. b, Three different systems for small-molecule-directed cell elimination. HSV-TK phosphorylates ganciclovir, which is then further phosphorylated by cellular kinases into a potent DNA polymerase inhibitor⁶⁵. Rimiducid dimerizes iCasp9 (refs. 59,73,77) and rapamycin dimerizes RapaCasp9^80,89 to initiate the caspase cascade that triggers apoptosis. CAR, chimeric antigen receptor; Casp9, caspase-9; FKBP12, 12-kDa FK506-binding and rapamycin-binding protein; FRB, FKBP-rapamycin-binding domain; HSV-TK, herpes simplex virus thymidine kinase; iCasp9, inducible caspase-9; RapaCasp9, rapamycin-activated caspase-9.

The earliest suicide switch developed was the herpes simplex type 1 virus thymidine kinase (HSV-TK) system⁶² (Fig. 3b, left). This system is controlled by ganciclovir, an anti-viral guanosine nucleoside analogue. HSV-TK, which is the target gene of ganciclovir, is inserted in CAR T cells wherein it confers lethal sensitivity. Upon systemic administration of ganciclovir, HSV-TK phosphorylates it to a nucleoside monophosphate that is further phosphorylated by endogenous cellular kinases to a nucleoside triphosphate. This metabolite competitively inhibits guanosine incorporation into DNA and disrupts DNA synthesis, leading to cell death^63,64. CAR T cells expressing the HSV-TK system showed antitumour efficacy in human leukaemia and solid tumour xenograft mouse models^65,66. Importantly, the solid tumour study demonstrated effective ganciclovir-mediated depletion of CAR T cells and reduction of their activity⁶⁶. Although the HSV-TK system was investigated in several clinical trials, its activation was not required owing to a lack of CAR T toxicity, and thus, its activation has not been studied⁴⁵. The HSV-TK system has some limitations, such as the requirement for active cell replication to induce cell death that leads to incomplete cell eradication and slow cell ablation kinetics (3–5 days)^64,67–69. Immunogenicity against the virally derived HSV-TK gene has also been observed, which could eliminate cells before drug ablation^62,70. However, notably, no immunogenicity was observed in the CAR T cell trials. HSV-TK has been used in clinical trials for a second purpose, as a reporter gene for positron emission tomography (PET) imaging of CAR T cells⁷¹. For this application, a fluorine-18-radiolabelled analogue of penciclovir, 9-[4-[¹⁸F]fluoro-3-(hydroxymethyl)butyl]guanine ([¹⁸F]FHBG), was applied to patients with glioblastoma. The system was found to be safe and effective for longitudinal tracking of CAR T cell viability and localization⁷¹.

The most advanced CAR T elimination switch developed to date is the inducible caspase-9 (iCasp9) system^72,73. It uses the chemical inducer of dimerization (CID) rimiducid (AP1903) or its analogue AP20187 to activate the apoptosis-triggering enzyme caspase-9. Rimiducid is a cell-permeable tacrolimus analogue developed to specifically bind and homodimerize the F36V mutant of FK506-binding protein (FKBP12). It has 1,000-fold weaker binding for wild-type FKBP12 and, thus, does not inhibit the mechanistic target of rapamycin (mTOR) kinase, providing physiological orthogonality⁷⁴. The iCasp9 protein, expressed in CAR T cells, consists of the caspase-9 protein with a replacement of its caspase recruitment domain by FKBP12^F36V. Upon CID addition, iCasp9 homodimerizes, triggering the apoptotic signalling cascade^69,72,75,76 (Fig. 3b, middle). The iCasp9 system has been the most widely used of all small-molecule CAR T cell control systems in human clinical trials (38 out of 46 trials; Supplementary Table 2). The system has shown remarkable effectiveness with both rapid (within 30 min) and robust cell ablation (up to 95%) after a single administration of rimiducid or AP20187 (refs. 72,73,77–80). It was capable of eliminating CAR T cells in the blood and central nervous system of patients and alleviated toxicities (NCT01494103)⁸¹. In several other studies, the iCasp9 switch was present but not used. These studies have demonstrated safety and lack of a negative impact of the iCasp9 system on CAR T cell therapeutic efficacy (for example, NCT04196413, NCT02414269 and NCT01822652)^82–84. In another study (NCT03373097), the iCasp9 system was also shown to be titratable, as low rimiducid doses remediated toxicity and allowed for later CAR T cell expansion and restarting of antitumour functions⁷⁷. Being of human origin, iCasp9 is also largely non-immunogenic^85–87. Although the level of ablation will be effective for alleviating many toxicities, it may not be sufficient for severe cases of ON-target–OFF-tumour toxicity⁸⁸.

One limitation is that neither rimiducid nor AP20187 are FDA-approved agents. ‘RapaCasp9’, an alternative iCasp system, uses rapamycin, a macrocyclic natural product that is FDA-approved as an immunosuppressant⁸⁹ (Fig. 3b, right). RapaCasp9 showed potent ablative activities in vitro and in preclinical models⁸⁹. The ability of rapamycin to interact with its native target mTOR could be a potential concern. However, the single-dose administration required for cell ablation is probably to be well-tolerated, and immunosuppressive activity is potentially advantageous in the context of remediating immune toxicity. Thus, this system has high clinical potential.

Control of CAR presence or assembly

An alternative strategy to mitigate CAR-related toxicities is the use of small molecules to control the presence or assembly of the CAR protein^90–116 (Fig. 4). Unlike suicide switches, this approach preserves therapeutic cells and allows for reversible and temporal control of CAR T cell activation. It also adds potential benefits for the fitness of the T cells. Chronic activation of T cells by antigen or tonic receptor signalling can lead to T cell exhaustion and ultimately apoptotic cell death^37,39. Resting cells by pausing CAR signalling through small-molecule control has been established as an effective way to remediate this problem and to boost antitumour efficacy^117,118.

Fig. 4 |. Small-molecule control mechanisms of CAR presence and assembly.

a, Concept of a CAR T cell ON switch and system designs. A small-molecule drug induces the presence of a functional CAR and can be used to titrate its expression level. Notably, antigen is also required for CAR activation and the systems signal according to [antigen] AND [small molecule] Boolean logic. System designs include the following: assembly, in which a dimerizer brings together antigen binding and signalling receptor fragments^93,97; degron stabilization, in which a ligand stabilizes a protein degron to preclude CAR degradation¹¹⁸; proteolytic control, in which a protease inhibitor blocks destructive proteolytic cleavage of the CAR^104,105; transcriptional control, in which a ligand binds to a transcriptional activator to direct its DNA binding or nuclear localization^{108–110,116}; and recombinase control, in which ligand binding localizes an engineered DNA recombinase to the nucleus to invert the CAR gene to allow for its expression¹¹². b, Concept of a CAR T cell OFF switch and system designs. A small-molecule drug reduces the level of functional CAR and can be used to titrate its level. These systems signal according to [antigen] AND NOT [small molecule] Boolean logic. System designs include the following: degradation control, wherein a PROTAC or molecular glue binds to a specific protein domain on the CAR recruiting an E3 ubiquitin ligase to initiate CAR degradation^90,92–95; proteolytic control, wherein a protease inhibitor blocks the proteolytic cleavage of a degron domain off of the CAR, leading to CAR degradation^91,104,105; a PPI disruptor, which blocks a protein fragment with CAR signalling domains from associating with the receptor fragment containing the antigen binding domain^96,105–107; and recombinase control, in which ligand binding localizes an engineered DNA recombinase to the nucleus to excise the CAR gene, turning OFF its expression¹¹². CAR, chimeric antigen receptor; mRNA, messenger RNA; PPI, protein–protein interaction; PROTAC, proteolysis-targeted chimera.

Owing to these advantages, small-molecule systems directly controlling the CAR have made up the bulk of preclinical development efforts to date. Researchers have created ON switches in which the small molecules induce the presence of the CAR (Fig. 4a) and OFF switches that eliminate or inactivate the CAR (Fig. 4b). Many of these systems afford fine-tuned control. However, it is notable that long-term activity also requires frequent and repeated small-molecule dosing. As detailed below, a multitude of unique system designs for CAR switches have been created. These systems share many features, but their key nuances make them suited to specific use cases.

Direct induction of CAR protein degradation

Small-molecule control of CAR protein degradation provides a robust and rapid means to regulate CAR activity. Several CAR OFF switch and ON switch designs use this mechanism. One embodiment, using the FDA-approved molecular glue lenalidomide (and its analogue pomalidomide), was developed by two independent groups^93,94. Lenalidomide is a thalidomide analogue that belongs to a class of imide-containing immunomodulatory drugs. It is a ‘molecular glue’ that binds to the cereblon E3 ubiquitin ligase and targets it to Ikaros zinc finger proteins, leading to Ikaros degradation¹¹⁹. In both CAR designs, the target domain from the Ikaros family zinc finger protein 3 (IKZF3) is fused to the C terminus of the CAR. Binding of the molecular glue to this domain leads to polyubiquitination and degradation of the CAR fusion protein (Fig. 4b, ‘Degradation’, left). Both systems displayed dose-dependent degradation with a high dynamic range of CAR activity. One design was optimized to overcome limited degradation with the initial construct⁹³ by developing ‘super-degron’ variants through screening of previously identified zinc finger binders to thalidomide analogues. The optimized system showed improved sub-nanomolar sensitivity and in vivo control of CAR expression and function in animal models⁹³. The FDA-approved status and oral drug formulations make these systems ideal for clinical translation.

A related approach uses PROTACs to mediate CAR degradation (Fig. 4b, ‘Degradation’, right). PROTACs are heterobifunctional small molecules that consist of a ligand for a target protein fused by a linker to an E3 ubiquitin ligase binding moiety (most often cereblon or von Hippel–Lindau)¹²⁰. Like molecular glues, binding leads to ubiquitination of the target protein and its degradation^121,122. PROTAC–target pairs used for CAR T regulation include ARV-771 and ARV-825 targeted to the bromodomain (BD)BRD4 fused to the CAR^90,121. Administration of these drugs resulted in near-complete and dose-dependent reduction in CAR T cell effector functions¹⁰⁴. However, some toxicity was observed, probably owing to the degradation of endogenous BRD4. This toxicity could limit the clinical viability of this system. Towards the goal of an orthogonal approach, another design involved a protein tag derived from the bacterial enzyme Escherichia coli dihydrofolate reductase (eDHFR). This protein can be used to specifically bind to the aminopyrimidine antibiotic trimethoprim (TMP) and its derivatives⁹⁵. The PROTAC trimethoprim–pomalidomide 7c (TMP-POM-7c) led to rapid CAR downregulation (80% reduction within 4 h). This system was effective both in vitro and in vivo in a preclinical mouse model⁹⁵. Although no toxicity was observed, immunogenicity could be a concern for clinical translation owing to the bacterial eDHFR. However, preliminary analysis has shown low reactivity in immunogenicity assays. Finally, creation of an ¹⁸F-radiolabelled TMP enabled in vivo tracking of CAR T cells in mice and characterization of their biodistribution using PET or CT imaging⁹⁵.

Degron stabilization

Another CAR control strategy relies on drug-stabilized protein degrons. In this ON switch design, a destabilized protein degron is fused to the CAR, leading to its degradation until the small-molecule ligand is administered to bind to the degron to stabilize it and the CAR protein (Fig. 4a, ‘Degron stabilization’). In one design, the CAR was fused to the FKBP12^{F36V, L106P} domain. Addition of Shield-1, an analogue of the macrocyclic natural product FK506, leads to tunable control of CAR levels and function in vitro¹¹⁸. An alternative system using an eDHFR-based degron with TMP as the drug inducer was also created and tested in vivo. It showed robust activity in a mouse model by attenuating CAR T cell-induced systemic inflammation¹¹⁸. Additionally, this construct was used to develop the concept of ‘resting’ CAR T cells to improve their function. Periods of time without the inducer led to increased T cell persistence, decreased cell exhaustion and more memory-like CAR T cells. Antitumour activity was significantly improved. Interestingly, an OFF-switch system termed CAR-LID (ligand-induced degradation) used Shield-1 for a reverse role in which binding exposed a cryptic degron in the FKBP12^F36V mutant domain. This approach showed robust activity in vitro and in mouse models⁹², but it is limited by Shield-1 not being FDA-approved.

Proteolytic control

Another creative CAR control design uses viral proteases and small-molecule inhibitors to generate switches. The general ON-switch design (Fig. 4a, ‘Proteolytic control’) includes expression of a protease and a CAR containing a proteolytic cleavage site, which dismantles the CAR upon protease exposure. Administration of a protease inhibitor prevents cleavage, protecting the CAR. Two systems, VIPER-CAR and SNIP-CAR, were developed using this approach^104,105. Both use the Hepatitis C Virus Nonstructural Protein 3 Protease. Although conceptually similar, the two system designs differ in the location of the protease. VIPER has a direct fusion of the protease to the CAR, and SNIP-CAR incorporates the protease as a separate membrane-bound protein^104,105. They are controlled by three different FDA-approved inhibitors: grazoprevir¹²³, danoprevir¹²⁴ and asunaprevir¹²⁵. All are peptidomimetics containing sulfonamide motifs that competitively inhibit activity through non-covalent binding to the protease active site. Both VIPER and SNIP-CAR showed robust control of the CAR levels with a high dynamic range of signalling. Tuning the grazoprevir dose reduced toxicities and cytokine production while retaining antitumour activity in vitro and in mouse models^104,105. There were also some unique features for the systems. SNIP-CAR was especially robust in combating tonic activation and promoting memory T cell formation¹¹⁸. VIPER was expanded to be combined with a lenalidomide-control construct. This dual system enabled orthogonal small-molecule control of two CARs targeting different antigens¹¹⁹. In another study, an OFF switch system was created, termed ‘SWIFF-CAR’⁹¹ (Fig. 4b, ‘Proteolytic control’). The CAR was fused to both a protease and a protein degron. In the base state, the active protease cleaves off the degron domain, leading to stable CAR expression. Administration of asunaprevir inhibited cleavage, leading to CAR degradation⁹¹. These systems have many favourable features for translation, including oral drug formulations and well-understood pharmacokinetics and pharmacodynamics. The only limitation is potential immunogenicity of the virally derived protease.

Control of CAR assembly

Splitting the CAR protein and using a CID to re-assemble CAR parts constitute another ON switch strategy (Fig. 4a, ‘Assembly’). Most commonly, the CAR is divided into antigen binding and signalling domain parts, and each CAR fragment is fused to a CID binding domain. Several CID-based systems have been reported with titratable control of CAR T cell activation. One approach, termed ‘split-CAR’, divided the CAR into two distinct membrane-bound peptides and fused them to FRB^T2089L and FKBP12 domains⁹⁷. Heterodimerization was induced by the rapamycin analogue AP21967. A second split-CAR system with the DID1 and GAI domains could be dimerized upon treatment with the diterpenoid plant hormone gibberellin⁹⁷. These systems demonstrated titratable cytokine production and cell proliferation comparable with conventional CAR T cells in vitro. The rapalog design has additionally showed in vivo functionality in mouse models^97,98. The tunability and use of human components in the rapalog systems present advantages for clinical translation. Some potential limitations include the short half-life of the CID, the need for repeated drug administration to maintain CAR activity and intravenous administration.

A related design, termed the dimerizing agent-regulated immunoreceptor complex (DARIC), places the CID interaction domains extracellularly. This positioning allows for dimerization and subsequent CAR induction at significantly lower CID concentrations⁹⁹. Generated systems were functional using both rapamycin and AP21967 at concentrations as low as 100 pM. Importantly, given the immunostimulatory role of these systems, these concentrations are far below the required dose for the immunosuppressive activities of the drugs. Testing in human tumour xenograft models has demonstrated robust CAR activity that was highly responsive to pauses and re-administration of the CID⁹⁹. A related strategy, termed ‘AvidCAR’, uses AP21967 to dimerize CARs with low antigen binding affinity, triggering signalling by inducing higher avidity¹²⁶. Another design, termed ‘multi-chain CAR’¹⁰⁰, was based on the high-affinity Fc receptor for the IgE (FcεR1) oligomeric structure. This system also showed titratable CAR T effector function at a low dose with an EC₅₀ of 12 nM (ref. 100). In another clever design, the previously discussed lenalidomide degron-based OFF-switch was converted into an ON switch assembly system by using lenalidomide as a CID for binding proteins (IKZF3 and CRBN) fused to CAR fragments⁹³. Key to this approach was mutation of intracellular lysine residues that could otherwise serve as ubiquitin conjugation sites and lead to degradation⁹³. The use of human protein components and FDA-approved small molecules with oral formulations, rapamycin and lenalidomide, makes these designs ideal for clinical translation.

A counterpart to CID assembly, protein–protein interaction (PPI) disruptors have been coopted to create CAR OFF switches through CAR disassembly (Fig. 4b, ‘Disassembly’). Although PPIs have been notoriously difficult to target with small molecules, recent advances through computational drug design have led to the development of several PPI disruptors under investigation^96,106,107. For the ‘STOP-CAR’ system, researchers used computational design to create chemically disruptable heterodimer proteins⁹⁶. They chose to target the B cell lymphoma extra-large protein (Bcl-XL) owing to the long half-life, high affinity and high tolerability in humans of the known inhibitors A-1155463 and A-1331852. These compounds are potent and selective BH3 mimetic inhibitors that bind to the hydrophobic groove in Bcl-XL. Although Bcl-XL is a globular domain appropriate for CAR fusion, its interacting protein Bcl-2 interacting mediator (BIM) is unstructured. Computational design produced a human-derived globular protein scaffold grafted with the BIM binding motif, termed LD3. Fusing LD3 and Bcl-XL domains to CAR fragments led to CAR assembly driven by the Bcl-XL–LD3 interaction. STOP-CAR T cells demonstrated CAR effector functions comparable to standard CAR T cells. Titratable downregulation of CAR activity was observed at A-1155463 concentrations as low as 10 nM. Activity could be re-started after 48 h of wash-out⁹⁶. A related system was generated using the tetracycline repressor protein class B (TetRB), known to bind to the short peptide TIP. The tetracycline analogue minocycline could be used to disrupt CAR activity¹⁰⁶. Minocycline is an FDA-approved generic antibiotic that is orally bioavailable. However, the bacterially derived TetRB protein is potentially immunogenic¹⁰⁶. To overcome this, another PPI system was developed based on a single domain antibody that binds minocycline. A peptide library was screened to identify an antibody-binding peptide that could be displaced by minocycyline¹⁰⁷. These binders were used to create an OFF switch CAR ‘MinoCAR’ and several gene control systems¹⁰⁷. Finally, a fourth disruptor system was created as a variant of the VIPER system¹⁰⁵. A mutated NS3 protease that can bind its target peptide but cannot cleave it formed one binder. The target peptide was used as the other binder. Grazoprevir administration successfully disrupted the interaction to control CAR activity¹⁰⁵. Overall, the human-derived components and favourable pharmacokinetics of the STOP-CAR and MinoCAR systems may give them a translational advantage⁹⁶.

Transcriptional control

Another important approach is small-molecule control of CAR transcription. Transcriptional gene control systems for mammalian cells were first developed >30 years ago and are mainstay research tools for cell culture and animal models^127–129. They are capable of tight gene regulation and a high dynamic range, as they can restrict gene expression at the mRNA level before protein is produced. However, they also face limitations. Kinetics of gene regulation are slower relative to direct control of protein function, and inducible gene promoters can be susceptible to silencing that can limit their activation potential¹³⁰. The first transcriptional inducer systems applied to CAR T cells were based on transcription factors used in early research tools (Fig. 4a, ‘Transcriptional control’). One system used the bacterially derived tetracycline transcriptional repressor (TetR) and reverse trans-activator (rTTA) controlled by the tetracycline analogue doxycycline. Upon binding to doxycycline, the rTTA undergoes a conformational change that allows it to bind to a promoter region, activating CAR expression. Using the TetR system, CAR levels were shown to be tightly controlled with 20-fold induction¹⁰⁹. CAR T activity was specific to the presence of both the drug and the antigen. However, a low number of cells were transduced with the system (~15%)¹⁰⁹. This result was probably owing to gene silencing. A later variant achieved ~30% transduction efficiency using a two virus system¹¹⁰. Another system uses an oestrogen receptor-based transcription factor regulated by the selective oestrogen modulator tamoxifen and its active metabolite 4-hydroxy tamoxifen (4-OHT)¹⁰⁸. The transcription factor consists of a synthetic zinc finger transcription factor fused the oestrogen binding domain¹¹¹. Binding of the metabolite 4-OHT leads to nuclear localization of the transcription factor and DNA binding and activation of CAR transcription. In a mouse model the titration of tamoxifen fine-tuned antitumour function¹¹¹. In a recent report, drug-inducible zinc finger transcription factors termed ‘SynZiFTR’ were generated to be activated by grazoprevir or 4-OHT–tamoxifen¹¹⁶. The DNA binding domains in this study recognized 18 base pair DNA sequences, making them improbable to activate unintended genes. The grazoprevir synZiFTR contains the NS3 protease, which self-cleaves and destroys the transcription factor unless inhibited by a small molecule. Similar to a previous work, the 4-OHT–tamoxifen system functioned through nuclear localization. These factors displayed robust levels of small-molecule-specific activity in vitro and in human tumour mouse models. The authors also took advantage of the ability to co-express an alternative transgene, super-IL-2, with the system. This factor promoted the expansion of CAR T cells and their antitumour activity¹¹⁶. Immunogenicity remains a concern for translation of the TetR and NS3 systems. Although the tamoxifen-inducible system may have an advantage, the required continuous application of the drug could lead to unwanted toxicities¹³¹.

Recombinase control

Memory switches provide an alternative design that does not require the continual presence of the small-molecule inducer. For these systems, CAR expression is maintained or turned OFF after a single dose of the small molecule¹¹². Drug-inducible recombinases that have been critical tools for mouse genetics form the basis of these systems¹³². One design, termed flip-excision ‘FLEx’, used an ERT2-FLPO recombinase fusion that translocates to the nucleus upon interaction with 4-OHT¹¹². It acts to flip the CAR gene into an orientation to turn CAR gene expression ON or OFF (Fig. 4a,b, ‘Recombinase control’). Although the system showed some leakiness before 4-OHT induction, a 24-h exposure to 4-OHT led to strong and titratable CAR induction in up to 75% of cells. Excitingly, CAR expression was maintained 15 days after initial 4-OHT induction¹¹². This system has favourable characteristics for translation, including the use of an FDA-approved inducer and functionality with a single drug dose. However, immunogenicity of the bacterially derived recombinase could be a limitation¹¹².

Finally, an alternative OFF switch strategy is to apply the FDA-approved tyrosine kinase inhibitor dasatinib¹³³. Although not specific to CAR T cells, it inhibits CAR phosphorylation and can be used to rest cells and tune activity^117,118. Dasatinib has the downside of potentially regulating kinase signalling pathways in endogenous cells but requires no additional CAR engineering for activity. It is being tested in clinical trials (Supplementary Table 2).

Antigen targeting

To generate CAR T cells capable of targeting multiple antigens, several groups have created universal ‘adaptor’ CARs. Instead of binding directly to tumour antigens, adaptor CARs exert function by binding to a tag on one or more bifunctional adaptor molecules that bind to the tumour cells. These systems have three potential advantages. Advantages include tuning by the dose of the adaptor to avoid toxicities, simultaneous targeting of multiple antigens to avoid cancer relapse, and universality as they could be applied to patients with different types of cancers by administering different adaptors. Most systems have been implemented with antibody adaptors containing chemical or peptide CAR-recognition tags. Tags include fluorescein, biotin, short peptide neoepitopes, and leucine zippers^134–137. These systems have broadly shown tunability and robust targeting to a wide variety of antigens in preclinical mouse models. A recent clinical study has highlighted the feasibility and promise of the universal CAR approach showing comparable efficacy and faster resolution of adverse events to traditional CAR therapy (NCT04450069).

New efforts have focused on the development of heterobifunctional small-molecule adaptors to mediate antigen targeting by universal CARs. For these adaptors, the tumour binding antibody is replaced with a small-molecule binder and fused via a chemical spacer such as polyethylene glycol to a chemical CAR recognition tag (Fig. 5a). Compared to antibody-based adaptors, small molecules have advantages of dramatically lower production cost, homogeneity and potential for enhanced tissue penetration and oral bioavailability. The first system created used an anti-fluorescein universal CAR and a fluorescein-folate adaptor (EC17) that recognizes folate receptors¹⁰². Folate receptors consist of FOLR1, a surface-bound glycoprotein preferentially expressed on an estimated 40% of all human cancers, and FOLR2 found at high levels on some immunosuppressive cell populations^138,139. EC17 had previously been used clinically as a fluorescent probe imaging agent for detection of tumour margins during surgery¹⁴⁰. Thus, it translates well for use as a CAR targeting adaptor and could potentially serve a dual role as a theranostic agent^103,141. In animal testing, EC17 demonstrated inducible and curative antitumour activity in several human tumour xenograft mouse models^103,141,142. Cytokine release syndrome was noted upon high dose administration of EC17. However, toxicity could be averted without compromising tumour clearance by gradually increasing EC17 dosage^103,141. Additionally, free fluoresceine or supraphysiologic levels of folate could be used to swiftly block CAR T cell-directed tumour lysis by competing with EC17 binding¹⁰². EC17 also exhibited favourable pharmacokinetic characteristics. It was able to efficiently penetrate tumours, and unbound molecules were rapidly eliminated from the bloodstream within 90 min of administration^140,143. Further expanding the scope of antigen targeting via small-molecule adaptors, the heterobifunctional adaptors FITC–DUPA, which targets prostate-specific membrane antigen (PSMA), and FITC–acetazolamide, which targets carbonic anhydrase IX, were developed and characterized to show tumour regression in human tumour xenograft mouse models¹⁴². Rapid adaptor clearance could be favourable for limiting toxicities; however, it also necessitates repeat injections for activity. Pharmacokinetic enhancement or the creation of oral formulations may be essential for successful clinical translation.

Fig. 5 |. Antigen targeting of universal CAR T cells by small-molecule adaptors.

a, Universal CARs are targeted to one or more antigens of interest by co-administered bifunctional small-molecule adaptors^{102,141,142,168,169}. b, Universal CARs containing a CID (chemical inducer of dimerization) binding protein recognize target antigens via co-administered single-chain variable fragment (scFv)-CID binding protein fusions and a small-molecule heterodimerizer¹⁴⁴. c, Covalent small-molecule adaptors target CAR T cells against antigens by forming covalent bonds with the SNAP-tag and 38C2 and SPE7 scFvs^147,149,150. Reactions that form covalent bonds are shown, which include a nucleophilic substitution, imine formation and displacement of a fluoride. Lys, lysine; scFv, single-chain variable fragment; Ser, serine; Tyr, tyrosine.

In a related system, the universal CAR was further split into a secreted antibody binding domain (labelled as scFv) fused to a CID binding domain. This is brought into contact with a CID binding domain on the CAR via binding a heterobifunctional CID small molecule¹⁴⁴ (Fig. 5b). A1120, an orally bioavailable investigational non-retinoid RBP4 antagonist, was used as the inducer. This molecule heterodimerizes a lipocalin-based binding domain with a second binder bringing together antigen binding and CAR signalling fragments. The system showed robust in vitro activity, although the increased complexity of a three-component system could be a limitation. Another study has created a universal CAR OFF switch system by using a reactive small molecule to cleave the CAR recognition tag from the antibody adaptor¹⁴⁵. The system used an anti-biotin universal CAR with biotin-conjugated antibody adaptors. These adaptors were fitted with cleavable N3 linker, and the phosphine induced a Staudinger reduction, transforming an electron-poor azido group into an electron-rich amino group. This induces a 1,6-elimination to cleave biotin and prevent CAR function¹⁴⁵. In the multi-antigen targeting context, this approach enabled deactivation of targeting one antigen while maintaining targeting of a second one. Such control could allow for selective OFF-switching of an antigen while allowing targeting of a safe antigen to continue. The bioavailability of the phosphine is a concern for in vivo applications owing to its susceptibility to oxidation, which could be addressed through the application of other cleavage reactions, such as tetrazine-triggered bio-orthogonal ‘click and release’ chemistry¹⁴⁶.

One advance to universal CAR engineering has been the development of systems with adaptors capable of covalent attachment to the CARs (Fig. 5c). These systems have advantages of higher potency and the ability to pre-assemble CARs with adaptors before infusion. Covalent bond formation could also lead to more favourable pharmacokinetics with the adaptor being preserved for a longer time on the CAR T cell surface. The first systems were created using antibody adaptors using SNAP-tag–benzylguanine (Fig. 5c, left) and SpyCatcher/SpyTag reactions^147,148. A recent study has created a universal adaptor system that forms a covalent bond with small-molecule adaptors¹⁴⁹. Here, the catalytic 38C2 antibody was used as the targeting part of the CAR and paired with adaptors containing 1,3-diketone tags¹⁴⁹. The antibody 38C2 has a reactive lysine residue capable of forming a reversible covalent bond in the catalysis of aldol and retro-aldol reactions (Fig. 5c, middle). Folate receptor-targeting folate–diketone and PSMA-targeting DUPA–diketones with various spacer lengths were assembled and tested. They exhibited strong in vitro and in vivo antitumour activity. Furthermore, a DNA-encoded small-molecule library approach was used to identify a human epidermal growth factor receptor 2 binding ligand and was converted into an adaptor¹⁴⁹. Another recent system used electrophilic proximity-inducing synthetic adaptors with a dinitrophenol CAR recognition tag¹⁵⁰. These adaptors were designed to covalently bind to an anti-dinitrophenol CAR via acyl-imidazole ester or aryl sulfonyl fluoride electrophilic chemistries (Fig. 5c, right). They were capable of potent in vitro re-targeting of CAR activity and were more effective than non-covalent molecular counterparts. These systems have high translational potential owing to multi-antigen targeting capabilities. However, there is some immunogenicity concern from the protein adduct formation.

Control of accessory gene function

Beyond directly controlling CAR activity, there is tremendous therapeutic promise in using small molecules to control accessory genes co-expressed by CAR T cells (Fig. 6a). Although there is a limit to the genetic payload transferred by viral vectors expressing the CAR (length of total DNA inserted), there is often room for additional gene transfer¹⁵¹. Many accessory genes are most beneficial or safe when expressed in certain contexts. Thus, they could benefit from or even require conditional control. For example, expressing a cytotoxic gene only in the tumour site could yield potent antitumour activity while minimizing systemic toxicities. Fortunately, many of the same control systems regulating CAR activity described in the previous sections can be applied to accessory genes. The goals of accessory gene strategies to date fall into two major categories: improving antitumour function through enhancing the CAR T cells or exerting distinct antitumour effects and using CAR T cells as a delivery vehicle.

Fig. 6 |. Strategies to control accessory genes and cytotoxic pro-drug activation.

a, ON switch and OFF switch control systems can be applied to regulate accessory transgenes of interest to express proteins of interest (POIs) to further enhance CAR T cell therapy¹¹⁶. b, Dimerization systems to control accessory signalling proteins can be activated via chemically induced protein dimerization^153,154. c, CAR T cells can be engineered to secrete pro-drug activating enzymes to mediate activation of cytotoxic pro-drugs and enhanced cell killing¹⁵⁹. d, Chemical structures of small-molecule pro-drugs with the following information provided: compound name and interacting protein. The caging group is shown in blue font. AMS, 5′-O-sulfamoyladenosine; APdMG, 7-O-aminopropyl-7-O-des[morpholinopropyl]gefitinib; ceph, cephalosporin; CPG2, carboxypeptidase G2; Glu, glutamate.

Activating alternative signalling pathways in CAR T cells via small-molecule control has been one major focus. These approaches aim to directly combat defects in CAR T cell functions (Fig. 1b), especially CAR T cell survival and persistence. Drug control is key as constitutive expression of signalling genes can lead to overactivation of the CAR T cells, leading to cellular anergy, a state in which they are unable to respond to further antigen encounter, unwanted toxicities or uncontrollable T cell expansion with risk for leukaemogenesis. The first signalling system was an inducible co-stimulatory receptor (iCO) containing the myeloid differentiation primary response 88 (MyD88) and CD40 cytoplasmic domains fused to FKBP12^F36V152. Addition of rimiducid leads to homo-dimerization and signalling activation (Fig. 6b). When induced in CAR T cells, it led to improvements in proliferation, cytokine production and serial target cell killing ability. It also enhanced antitumour activity in murine models¹⁵³. An inducible MyD88/CD40–anti-prostate stem cell antigen (PSCA) CAR plus rimiducid was recently tested in a phase I clinical trial for patients with advanced abdominal and pelvic cancers (NCT02744287)¹⁵⁴. CAR T cells were observed to infiltrate the solid tumours and expanded and persisted in the peripheral blood. Promisingly, cytokine and chemokine production increased with rimiducid¹⁵⁴. A MyD88/CD40 OFF-switch was also created using the aforementioned sdAb/cyclic peptide and minocycline PPI disruptor system¹⁰⁷.

A small-molecule drug inducible IL-2 receptor (IL2R) was developed using a CID mechanism¹⁵⁵. The receptor chains IL2RB and IL2RG were fused to FRB and FKPB12, respectively, to allow for activation via rapamycin or AP21967. IL-2 is applied ex vivo when manufacturing CAR T cells and is often administered to patients to promote T cell expansion¹⁵⁶. The small-molecule system is advantageous as IL-2 is a major manufacturing expense and can affect non-engineered cells in patients, leading to major toxicities¹⁵⁷. Another promising strategy is the control of membrane-bound cytokines, including IL-2, IL-12, IL-15, IL-21 and IL-23 via an acetazolamide-stabilized carbonic anhydrase 2 (CA2) protein degron¹⁵⁸. These cytokines are key mediators of T cell function and proliferation but can be toxic or determinantal in excess. Acetazolamide is an FDA-approved sulfonamide-based non-competitive inhibitor of carbonic anhydrases. CAR T cells armored with degron-fused surface IL-12 showed both acetazolamide-mediated regulation and antitumour function in a mouse model¹⁵⁸. Ongoing clinical trials evaluating degron control of membrane-bound IL-15 on tumour-infiltrating T cells in melanoma (NCT05470283) and other solid tumours (NCT06060613) further highlight the clinical promise for this technology for CAR T applications.

To augment the killing ability of CAR T cells, a system called SEAKER (‘Synthetic Enzyme-Armed KillER’) that activates cytotoxic pro-drugs was created¹⁵⁹ (Fig. 6c). This design was inspired by gene-directed pro-drug therapy approaches¹⁶⁰. The CAR T cells are engineered to secrete an enzyme that can hydrolyse a chemical caging group from a cytotoxic small-molecule pro-drug, triggering its killing ability. Two enzymes were tested. Carboxypeptidase G2 (CPG2) hydrolyses a C-terminal glutamate caging element, and Enterobacter cloacae β-lactamase (B-Lac) cleaves a cephalosporin-based caging group. Three small molecules were chosen for testing: 5′-O-sulfamoyladenosine (AMS), a cytotoxic natural product, ZD2767, a nitrogen mustard, and APdMG (7-O-aminopropyl-7-O-des(morpholinopropyl)gefitinib), an analogue of the EGFR kinase inhibitor gefitinib. Pro-drugs with glutamate or cephalosporin masking groups were synthesized. AMS–glutamate and AMS–cephalosporin were chosen for further testing owing to their high selectivity and cytotoxicity (Fig. 6d). These pro-drugs were further tested in human tumour mouse models and displayed enhanced antitumour activity compared to the CAR T cells alone, including killing of antigen-negative tumour cells¹⁵⁹. This system has exciting potential in addressing resistance to CAR T cell therapy from antigen loss and immunosuppression. These approaches may be susceptible to drug efflux resistance; however, the combination with CAR T cells should be an improvement over either technology alone. Future systems using human enzymes could reduce immunogenicity issues that plagued early gene-directed pro-drug approaches¹⁶¹.

Conclusion and outlook

Innovations to future CAR T cell controller systems will be important in several areas. Immunogenicity of transgenic proteins remains a broad concern. Artificial intelligence approaches for protein immunogenicity prediction coupled with protein design (in this case, for small-moleculeinteracting proteins) could ultimately provide a solution¹⁶². Alternatively, future designs could use RNA-based switches, such as engineered riboswitches, that will not be subjected to T cell immunity¹⁶³. Given the limitation of gene payload size, the creation of compact multi-functional systems (that is, ON switch plus augmentation) responsive to multiple small molecules would be advantageous. Issues with affinity, poor pharmacokinetics and drug efflux could be assisted by the application of covalent chemistries¹⁶⁴. This is particularly true for covalent cell surface modifications, as both the pharmacokinetic and pharmacodynamic properties will change to resemble those of a cell therapy product, including multivalent ligand display.

Synthesis of novel compounds could offer solutions to the outstanding limitations in the CAR T cell field. Multi-targeting using universal CAR T cells with multiple small-molecule adaptors provides an attractive and scalable solution to address resistance owing to target antigen loss or heterogeneity. However, there is a dearth of small molecules that recognize tumour-associated surface antigens to use in heterobifunctional small-molecule adaptors, and the discovery of new targeting compounds is needed. Technologies such as DNA-barcoded libraries and artificial intelligence-aided drug design could accelerate these approaches^165,166. To overcome immune suppression, universal CAR targeting could also be applied to attacking immunosuppressive cell populations^138,167. Indeed, new reports targeting cancer-associated fibroblasts suggest promise for this strategy^168,169. Identifying compounds to target additional immunosuppressive populations is needed. Another major limitation is the lack of true tumour-specific targets that are shared among large patient populations¹⁷⁰. Alternatively, there are many shared chemical signatures associated with tumour cells and the tumour microenvironment, including hypoxia, acidosis, proteases and metabolic enzymes^171–174. Synthesis of small-molecule controllers that are conditionally activated in the presence of these chemical and biochemical signals could provide the needed specificity gate to safely target nonspecific antigens¹⁷⁵.

Innovations in the CAR T field raise new challenges that could benefit from the use of small-molecule control systems. Approved CAR T products are now being applied to treat autoimmune diseases^176,177, which comes with heightened concern about the safety risks and a limited understanding of the level of activity needed for disease remission¹⁷⁸. Suicide switches or CAR control systems stand to aid these efforts. Another major focus is the in vivo production of CAR T cells by injecting a CAR-encoding virus directly into a patient. Erroneous delivery of the CAR to tumour cells could lead to resistance, which could be alleviated by a suicide switch. Additionally, there is the challenge of producing sufficient numbers of engineered cells in patients via viral delivery^179,180. Small-molecule-controlled signalling approaches could be used to enable in vivo CAR T cell expansion. Treatment of solid tumours with CAR T cells has only resulted in temporary responses to date^181–183. Here, persistence is a major issue and could possibly be enhanced by small-molecule-controlled signalling and gene activation. Additionally, several genes derived from tumour cells have been shown to greatly increase the proliferation and persistence of CAR T cells¹⁸⁴. To ensure safety, these genes may benefit from accessory gene control or suicide switch strategies.

Finally, a hallmark of natural immunity is memory and persistence of immune cells to prevent and combat disease states over long periods of time, including the entire course of human lifespan¹⁸⁵. Although it will take many advancements, one can envision future development of synthetic immune cells that could be administered prophylactically. These ‘sentinel’ cells would stand dormant in times of health but could be expanded, contracted and armed by an arsenal of small-molecule controllers to battle emerging disease states.

Supplementary Material

Supplementary information The online version contains supplementary material available at https://doi.org/10.1038/s41570-025-00768-6.