Mechanisms of cytokine release syndrome and neurotoxicity of CAR T-cell therapy and associated prevention and management strategies

After the recognition of tumor antigens, CAR T-cells release massive amounts of perforin/granzymes and cytokines, including TNF-α and IFN-γ, resulting in tumor pyroptosis [9‐12]. Pyroptosis is a type of programmed cell death that differs from apoptosis [13], and is characterized by cellular swelling, lysis and subsequent cell content and proinflammatory factor release. It is believed that pyroptosis of the target cell represents the onset of CRS. Two pathways are likely to be involved, which are mediated by granzyme B (GZMB) and granzyme A (GZMA) (Fig. 1B). GZMA and GZMB can both enter cells through pores formed by perforin [9, 12]. Subsequently, GZMB cleaves gasdermin E (GSDME) directly or by activating caspase-3 [9], while GZMA directly cleaves gasdermin B (GSDMB) for its activated form [12]. Then, the N-domains of gasdermin, which are veiled by the C-terminus, can be released and oligomerize on the cell membranes to form pores, causing decreased cell viability, bubbles blowing from the plasma membrane, cell swelling and finally cell lysis [9, 13, 14]. The different types of gasdermin and their pyroptotic pathways differ among tumor cells. GSDME widely exists in hematologic malignances [9], while GSDMB is found more frequently in bladder cancer, skin cancer and renal clear cell carcinoma, and its expression can be upregulated by cytokines, such as IFN-γ [12].

Cell death through either apoptosis or pyroptosis mainly depends on the amount of gasdermin expression [9, 12, 15]. Low levels of gasdermin induce apoptosis, while high levels of gasdermin switch apoptosis to pyroptosis [9]. Cytotoxic T lymphocytes (CTLs) mediate apoptosis in tumor cells via a low level of perforin/granzyme release, consequently activating little gasdermin and producing a few pores on the cytoplasmic membrane [16, 17]. Notably, cells are capable of repairing perforin pore formation to a certain degree, protecting cells from pyroptosis [18]. In contrast, CAR T-cells release a large amount of perforin/granzymes to induce massive gasdermin release, surpassing their self-repair capability and leading to pyroptosis [9]. Clinically, GSDME is widely expressed in B leukemic cells after CD19 CAR T-cell infusion, and the severity of CRS is positively associated with an increase in GSDME [9].

An increasing number of studies indicate that monocyte and macrophage lineages are the key origin of inflammatory cytokines in relation to CRS [19‐21]. Macrophages can be activated by damage-associated molecular patterns (DAMPs), including high-mobility group box 1 (HMGB1), adenosine 5′-triphosphate (ATP) and double-stranded DNA (dsDNA) [9, 22], which are pyroptotic products of tumor cells (Fig. 1C). HMGB1 can bind Toll-like receptor 2 (TLR2) and TLR4 on the surface of macrophages. Then, the adaptor proteins myeloid differentiation primary-response 88 (MyD88) and TIR-domain-containing adaptor inducing IFNβ (TRIF) recruit and activate mitogen-activated protein kinases (MAPKs) and IκB kinase (IKK). MAPKs and IKK regulate the release of a wide range of cytokines, including IL-6, via transcription factors activator protein 1 (AP-1) and nuclear factor κB (NF-κB) [9, 22]. In addition, TLR2 induces soluble IL-6R (sIL-6R) secretion, which can combine with IL-6 and facilitate the proinflammatory effects of IL-6 [23]. ATP, which is recognized by its receptor P2X7 on macrophages, can induce the activation of NOD-, LRR- and pyrin domain-containing 3 (NLRP3) in the cytoplasm and recruit apoptosis-associated speck-like protein (ASC) and pro-caspase 1 to form the NLRP3 inflammasome, subsequently leading to the maturation of caspase 1 [22]. On the one hand, caspase 1 is responsible for pro-IL-1β cleavage and the secretion of IL-1β; on the other hand, caspase 1 transforms GSDMD into the active form and leads to pyroptosis in macrophages [9, 22]. Pyroptotic macrophages consequently produce more DAMPs and proinflammatory factors, creating a vicious cycle and leading to the further activation of macrophages [22]. Additionally, the dsDNA-mediated absent in melanoma 2 (AIM2) inflammasome pathway could be involved in caspase 1 formation [24]. Macrophages phagocytize dsDNA released by pyroptotic tumor cells and activate the AIM2 inflammasome, which is a dsDNA sensor, in the cytoplasm. Similar to the NLRP3 inflammasome pathway, activated AIM2 forms an AIM2/ASC-pro-caspase 1 complex to trigger caspase 1-dependent IL-1β maturation [24], thereby causing more IL-1β release and aggravation of CRS.

In addition to the “pyroptosis–DAMPs–macrophage” pattern, macrophages can be recruited and activated by cytokines produced by CAR T-cells, such as TNF-α, IL-2, GM-CSF and IFN-γ [25, 26]. Contact-dependent CD40 ligand (CD40L)-CD40 interactions between CAR T-cells and macrophages may also play a role in triggering IL-6 and IL-1 release and higher inducible nitric oxide synthase (iNOS) expression [21]. iNOS stimulates NO production, resulting in potentiated vasodilatation and hemodynamic instability, which are common clinical features of CRS [27]. In addition, catecholamines are produced after the coculture of CAR T-cells and malignant cells, which may be involved in cytokine release [24, 28]. By binding α1-adrenergic receptors (ARs) on the surface of macrophages, catecholamines can enhance the AIM2/ASC-caspase-1 pathway and further promote IL-1β production and macrophage pyroptosis [22, 24]. In addition, catecholamines lead to a self-amplifying feed-forward loop in macrophages, promoting the release of catecholamines, IL-6 and other cytokines, including macrophage inflammatory protein (MIP)-1α, IFN-γ, IL-2 and TNF [28]. Monocytes are also likely to be the key mediators and the main source of IL-6 and IL-1 in CRS, as Norelli et al. reported that monocyte depletion prior to CAR T-cell therapy can protect mice from lethal CRS [20].

Among the cytokines released by monocytes/macrophages, including IL-6, IL-1, IL-2, TNF-α, GM-CSF and IFN-γ, IL-6 plays a central role [29‐31]. Notably, in addition to macrophages and monocytes, dendritic cells, endothelial cells, and even CAR T-cells are considered to participate in IL-6 production [21, 32]. IL-6 activates its downstream Janus kinase (JAK) and signal transducer and activator of transcription 3 (STAT3) mainly by binding the membrane-bound IL-6 receptor (mIL-6R) (classic signaling) or sIL-6R (trans-signaling) and another membrane protein, gp130. gp130 is ubiquitously expressed, whereas mIL-6R is mainly expressed on hepatocytes and immune cells. Cells that do not express mIL-6R, such as endothelial cells, are activated by trans-signaling in which IL-6 binds sIL-6R and forms a complex in the serum, triggering the dimerization of gp130 on the cell membrane [33‐35]. As a result, activated endothelial cells secrete additional IL-6 and other proinflammatory factors, such as vascular endothelial growth factor (VEGF), IL-8, monocyte chemoattractant protein-1 (MCP-1) and coagulation cascade activator plasminogen activator inhibitor-1 (PAI-1) [36‐38], leading to a positive loop of cytokine release and amplified inflammatory responses. In addition, endothelial cells are injured by cytokines and contribute to increased vascular permeability and leakage, edema, organ hypoperfusion, coagulopathy and organ dysfunction [39]. In CRS-associated coagulopathy, tissue factor (TF), factor VIIa, factor Xa, thrombin and platelets can also play roles as proinflammatory factors, upregulating cytokine synthesis in endothelial cells [40‐42].

In addition to the most common systemic CRS, it is reported that local CRS (L-CRS) is the earliest toxicity in non-Hodgkin lymphoma (NHL) and mainly manifests as local swelling and redness [43]. Initially, CAR T-cells distribute locally in the tumor mass, triggering L-CRS. Then, with CAR T-cell and cytokine accumulation, they overflow into the circulation and cause systemic CRS [43]. However, the mechanisms of L-CRS are likely to be unique since the inhibition of IL-6 may aggravate L-CRS. More research investigating its specific mechanisms is warranted.

The BBB consists of endothelial cells with tight junctions, the endothelial basement membrane (EBM), pericytes surrounding capillaries, the parenchymal basement membrane (PBM) and the endfeet of astrocytes, also called glia limitans [49, 50]; of these components, the endothelium is the most important component. Some characteristics of brain vascular endothelial cells are distinct from those of peripheral vascular endothelial cells, such as tight junctions and specific transport systems, restricting various molecules and cells from entering the CNS [51, 52]. However, inflammation can lead to endothelial activation, reversible contraction, damage to tight junctions, and even cell death [53, 54], disrupting the integrity of the BBB and activating coagulation. The angiopoietin (Ang)-Tie2 axis plays a significant role in balancing the endothelium between quiescence and activation. Ang I is produced by platelets and pericytes constitutively, while Ang II is primarily reserved in Weibel-Palade (W-P) bodies in endothelial cells. Normally, Ang I binds Tie2 on endothelial cells, promoting cell spreading and accumulating vascular endothelial cadherin (VE cadherin), which is an important component of tight junctions. This characteristic ensures the integrity of the endothelial barrier [55, 56]. When stimulated by inflammatory factors, such as VEGF, thrombin and epinephrine, W-P bodies and Ang II are released from the endothelium via Ca²⁺-mediated or cAMP-mediated pathways [57]. As a result, concentrated Ang II in serum replaces Ang I, inhibiting the Tie2 pathway and increasing the permeability of the endothelium [58, 59]. Disorder of the Ang-Tie2 axis is likely to be implicated in CAR T-cell-associated ICANS, and severe ICANS patients have been observed to present a higher ratio of Ang II/Ang I [39, 45]. Interestingly, another study showed that a decrease in Ang I triggered a higher Ang II/Ang I ratio rather than increased Ang II, which is associated with the thrombocytopenia observed in severe ICANS patients [45].

Endothelial activation is likely to contribute to ICANS-associated microthrombi and disseminated intravascular coagulation (DIC). Laboratory markers indicate that patients with severe ICANS suffer DIC, and an autopsy study showed intravascular von Willebrand factor (vWF) binding and platelet microthrombi [39, 45]. vWF can be released from W-P bodies in the form of ultralarge vWF and bind endothelial cells [60]. Then, vWF exposes the binding sites for platelet adhesion and vWF-platelet string formation. The blood flow shearing force induces vWF to expose another binding site for a disintegrin and metalloproteinase with thrombospondin type 1 repeats member 13 (ADAMTS13), which is responsible for vWF cleavage [60, 61]. Cytokines influence this process. IFN-γ, TNF, IL-4 and IL-6 can directly inhibit the production of ADAMTS13 or impede its cleavage [61, 62]. It is assumed that thrombocytopenia and consumptive coagulopathy in severe ICANS patients are due to ADAMTS13 deficiency [39], which is similar to the mechanism responsible for thrombotic thrombocytopenic purpura (TTP) [63]. However, since there is no evidence of other systematic manifestations of thrombotic microangiopathy, such as renal failure and morphological changes in red blood cells, thrombotic microangiopathy in ICANS presents a unique pathogenesis requiring further investigation.

Detrimental cytokines could be accessible to other components of the BBB due to the high permeability of the endothelial barrier, exacerbating the inflammatory response and further damaging the integrity of the BBB. Astrocytes represent a component with endfeet in the inner layer of the BBB and have direct contact with the brain parenchyma, regulating the osmotic pressure of the brain [64, 65]. Autopsy studies of severe ICANS patients have revealed that astrocytes are activated and injured [39, 66, 67], and the astrocyte markers glial fibrillary acidic protein (GFAP) and calcium-binding protein B (S100b) [68, 69] are increased in the CSF of ICANS patients after treatment with CD19 CAR T-cells [66]. Astrocyte injury is likely to play a role in the development of CAR T-cell-associated cerebral edema [66]. Iron channels and aquaporins on astrocytes can be interrupted by extracellular environmental changes, such as glutamate (Glut), hypoxia and stimulation by cytokines, such as IL-6, IL-1β and TNF, leading to cell swelling and abnormal osmotic forces, followed by cerebral edema and neuronal injury [64, 70, 71]. In addition, astrocytes have proinflammatory potential [64]. Triggered by IL-1β, astrocytes can release VEGF-A and destroy the tight junctions of endothelial cells [72, 73]. Astrocytes can also increase BBB permeability through various mechanisms, including the production of apolipoprotein E (APOE) and inhibition of the cyclophilin A (CYA)–NF-κB–matrix metalloproteinase 9 (MMP9) pathway in pericytes [74].

Pericytes lining along the endothelium play crucial roles in regulating BBB permeability and neuroinflammation [44, 66, 75, 76]. High concentrations of IFN-γ, TNF-α and other cytokines transmitted from serum to the CSF can cause pericyte stress. These cytokines stimulate pericytes to secrete additional inflammatory factors to further activate endothelial cells as positive feedback [39]. The incubation of primary human brain vascular pericytes with IFN-γ and TNF-α can induce large amounts of IL-6 and VEGF secretion [39]. IFN-γ can also inhibit the platelet-derived growth factor receptor β (PDGFRβ) signaling pathway, which is crucial for the proliferation and migration of pericytes, inducing pericyte stress and disrupting their regulatory role in the BBB [77, 78]. Furthermore, a recent study suggested that the on-target, off-tumor effect of CD19 CAR T-cells is also responsible for the breakdown of the BBB. Through single-cell RNA sequencing analysis (scRNA-seq), Parker et al. [79] found that CD19 is expressed on mural cells, including pericytes and vascular smooth muscle cells. These CD19-positive cells can be recognized by CAR T-cells, resulting in pericyte depletion and further BBB disruption. This phenomenon could also explain the higher incidence of ICANS in CD19 CAR T-cell therapy compared with other targets, such as CD22 and CD30 [80‐83].

Increased BBB permeability allows plasma leakage into the brain parenchyma as confirmed by imaging and histopathologic examinations of patients with CAR T-cell-associated cerebral edema [39, 67, 84‐86]. In addition, immune cells and cytokines can penetrate the brain parenchyma and trigger inflammatory reactions in the CNS, thereby inducing neuronal injury and dysfunction. Accumulating evidence sheds light on the contribution of myeloid cells to the pathophysiology of ICANS. Monocytes and macrophages are likely to be recruited to the CNS and produce IL-1β and IL-6, which are key factors in ICANS [20, 87]. Coincidentally, Deng et al. [88] used scRNA-seq to identify rare but significant ICANS-associated cells (IACs), possibly belonging to the myeloid lineage. Moreover, it was demonstrated that GM-CSF, which is crucial for myeloid cell proliferation and activation, was the factor most significantly associated with ICANS in the ZUMA-1 clinical trial [89]. GM-CSF was also highly increased in the CSF of nonhuman primate ICANS models [90]. The increase in GM-CSF may be driven by the diffusion of increased serum GM-CSF or be produced by CNS-infiltrating T cells or activated endothelium. Consequently, GM-CSF promotes the production of IFN-γ-inducible protein 10 (IP-10), MCP-1, and CC chemokine ligand 1 (CXCL1) and attracts myeloid cells to infiltrate the brain [25, 45, 91]. GM-CSF also plays a role in activating microglia [92]. Microglia, which are brain-resident macrophages, can polarize into a proinflammatory phenotype after exposure to GM-CSF, IL-6, and IFN-γ and amplify the inflammatory cascade [93, 94]. Clinically, microglial activation has also been observed in ICANS induced by CAR T-cells [39, 45, 87, 95]. It is assumed that the inflammatory cytokines in the brain parenchyma, including both those produced locally and those traveling from the bloodstream, cause neuron dysfunction and a series of neuropsychiatric symptoms. In preclinical studies, IL-1β and TNF could change neuronal excitability, inducing delirium and other neuropsychiatric symptoms [96, 97]. Other cytokines, such as IL-6 and IL-8, may also be implicated in neural injury [98].

Moreover, myeloid cells are likely to mediate excitotoxicity in ICANS by producing endogenous excitatory agonists [45]. Stimulated by IFN-α2, IFN-γ, and TNF-α, microglia and macrophages can produce the N-methyl-d-aspartate (NMDA) receptor agonists quinolinic acid (QA) and excitatory neurotransmitter Glut, causing seizures and other excitatory symptoms [99, 100]. In addition, QA can promote the secretion of Glut [101], activate astrocytes to secrete numerous cytokines, such as TNF-α, IL-6, and MCP-1, and change the cohesion of the BBB, representing a feed-forward mechanism exacerbating brain dysfunction [101, 102].

Other functions of microglia include synaptic pruning and scavenging damaged cellular debris to maintain hemostasis of the brain and normal cognitive function [103‐105]. Abnormalities in these functions contribute to cognitive disorders, which are the main pathophysiology of neurodegenerative diseases and aging and may be implicated in ICANS. Microglia may be depleted through off-tumor effects since low levels of CD19 were detected in human microglia by scRNA-seq and another single-cell transcriptomics database [79, 106]. Moreover, microglia were lost in a murine model treated with CD19 CAR T-cells, which was possibly associated with CD19 targeting [106]. CD22 is also expressed in microglia in the human brain [107], whereas there is no evidence indicating that the severity of ICANS is increased in CD22 CAR T-cell therapy. In summary, the possible on-target off-tumor mechanism of cognitive disorder induced by microglial injury in the ICANS remains to be clarified in further studies.

In addition to the infiltration of myeloid cells and resident microglial activation, T cells and CAR T-cells can also enter the CNS [30, 39, 45, 66, 88, 90]. It was assumed that the cytokines secreted by infiltrated CAR T-cells in the brain constitute another crucial factor causing ICANS [30, 108]. However, CAR T-cells can also be found in CSF from patients without ICANS [45, 66], suggesting that they may play a less important role in neurotoxicity.

Because the infusion dose of CAR T-cells and the disease burden are strongly associated with CRS and ICANS [44], Turtle et al. [84] proposed low-dose infusions for patients with high tumor burden, which achieved a high complete remission (CR) rate with no CRS and less ICANS. However, there was a high relapse rate in this study, raising the concern that a reduction in CAR T-cells could impair the long-term prognosis [84]. Recently, split dosing, also called fractionated dosing, has been recommended. This approach delivers CAR T-cells several times in the form of dose escalation, and subsequent infusions can be stopped if early clinical CRS is found [109]. Individualized dose modifications are considered superior to single-dose infusion since they can achieve a balance between the efficacy and safety of CAR T-cell therapy. In the trial reported by Frey et al. [109], there were three schemes of CAR T-cell infusion as follows: high-dose single-infusion (HDS; 5 × 10⁸ CAR T-cells; n = 6), low-dose single or fractionated infusion (LD; 5 × 10⁷ CAR T-cells; n = 9) and high-dose fractionated infusion (HDF; 5 × 10⁸ CAR T-cells; n = 20). In the HDF group, CAR T-cells were planned to be infused within 3 days by split dosing (Day 1, 10%; Day 2, 30%; and Day 3, 60%); however, 9 patients had a high tumor burden; among these patients, 3 received Day 1 only, 3 received Day 1 and Day 2, and the other 3 received all 3 doses. Eighteen patients in the HDF group achieved CR (18/20, 90%), and only 1 patient developed grade 4 CRS (1/20, 5%). In contrast, in the HDS group, only 3 patients achieved CR (3/6, 50%), while the other 3 patients died from CRS and concurrent infections (3/6, 50%). In the LD group, 3 patients (3/9; 33%) achieved CR, and 2 patients experienced CRS > grade 4 (2/9, 22%). The 2-year survival rate was also improved in the HDF group (73% in the HDF group, 22% in the LD group, and 17% in the HDS group). The fractionated dosing scheme was also applied in another clinical trial [110]. The rates of severe CRS and ICANS were decreased in the fractionated infusion cohort, while no remarkable differences in the 1-year progression-free survival (PFS) or 1-year overall survival (OS) were found. Collectively, these data preliminarily demonstrate that split dosing and individualized modification represent promising strategies to alleviate CRS and ICANS. It is more practical for clinicians to adapt infusion dosing than explore drugs or novel CAR T-cell products. More studies are required to validate whether this strategy dampens the efficacy and long-term outcome of CAR T-cell therapy and optimize the specific timing and dosing of infusion.

The basic components of CAR usually include an antibody-derived single chain variable fragment (scFv), hinge domain (HD), transmembrane domain (TMD), and intracellular domain, which consists of one or more costimulatory domains and a CD3-zeta (CD3ζ) domain. These components all contribute to the transmission of signals for CAR T-cell activation, and optimizing the structure for a safer CAR is a feasible strategy.

Recent studies have indicated that modifying the affinity of scFv for tumor antigens has an impact on toxicities. Ghorashian et al. [111] designed a novel CD19 CAR (CAT) with a lower affinity but more robust cytotoxicity than FMC63, which showed a high CR rate (12/14, 86%) with a low toxicity rate (no severe CRS; 1 severe ICANS) in the treatment of R/R pediatric B-ALL. In addition, the replacement of murine-derived FMC63 with humanized scFv [112] or the addition of an anti-IL-6 antibody to the scFv sequences to neutralize macrophage-derived IL-6 [113] are regarded as possible strategies to mitigate CRS.

The origin, length and flexibility of the HD and TMD also have effects on CAR function and the occurrence of CRS [114‐116]. CD8α-HD/TMD and CD28-HD/TMD are most frequently applied in clinical practice, and CD8α-HD/TMD CARs present a lower ICANS incidence [45, 87, 117, 118]. Additionally, including humanized scFv and CD8α-HD/TMD in one CAR molecule may exert synergistic effects on reducing toxicities [119]. In addition to the HD/TMD types, changes in length may impact toxicities. A CD19 CAR T-cell product containing a longer CD8α-HD/TMD was found to be associated with lower cytokine production, inducing no severe CRS or ICANS in 25 patients with refractory B cell lymphoma [116].

Several studies have indicated that CAR T-cells with 4-1BB costimulatory domains are associated with less toxicity than CAR T-cells with CD28 costimulatory domains [45, 120, 121]. A possible explanation for this finding is that CAR T-cells with 4-1BB costimulation have a mild but persistent killing effect mediated by the THEMIS-SHP1 complex, which can counteract the phosphorylation of the CD3ζ domain and consequently downregulate the expression of IFN-γ and other cytokines [122]. In comparison, CD28 CAR T-cells with no THEMIS-SHP1 complex have more rapid and intense phosphoprotein signaling, inducing large amounts of cytokine production and powerful, rapid, but transient killing effects [122]. Based on this mechanism, Sun et al. [122] added the small molecule AP21967 to a modified SHP1 on the CD28 domain to reduce cytokine release and ameliorate CRS without compromising the antitumor effects. However, a retrospective analysis showed that the incidences of CRS and ICANS are likely to be more related to HD and TMD than the costimulatory domain [123]. The role of the costimulatory domain is still unclear and remains to be clarified in further research.

Usually, CD3ζ with 3 immunoreceptor tyrosine-based activation motif (ITAM) domains is contained in the intracellular domain of the CAR and transmits activating signals through ζ-associated protein of 70 kDa (ZAP70) [124]. Reducing the number of ITAMs on CD3ζ or using other CD3 subunits with one ITAM, such as CD3ε, may be effective in inhibiting excessive CAR T-cell activation and decreasing CRS [125‐128]. Other optimization strategies include knocking down genes of key cytokines in CAR T-cells that contribute to toxicities, such as GM-CSF [25, 91] and IL-6 [32, 129].

Adding switches on CAR T-cells to regulate the inactivation of CAR signaling is being actively explored to eliminate toxicities. Most switches are controlled by small-molecule agents and can be divided into reversible and irreversible switches. The former can reactivate CAR T-cells when the inducers are removed, while irreversible switches induce permanent CAR T-cell damage.

Reversible switches can be achieved through the chemically disruptable heterodimerization of functional chains. Giordano-Attianese et al. [130] designed a STOP CAR (Fig. 3A) that contains the following 2 chains: recognition (R) chain and signaling (S) chain. The extracellular domain of the S chain consists of a c-Myc-tag and DAP10, and DAP10 maintains the expression stability of the S chain. The two chains are spontaneously dimerized through a computationally designed protein pair, apolipoprotein E4 (apoE4) and B cell lymphoma-extra large (Bcl-XL). ApoE4 is expressed at the end of the R chain, and Bcl-XL is located between the CD28 costimulatory domain and the CD3 ζ domain of the S chain. The specific disruption between apoE4 and Bcl-XL is triggered by the Bcl-XL inhibitor A1155463, consequently interfering with CAR signaling and transiently inactivating CAR T-cells. Another off-switch design is combined with proteolysis-targeting chimera (PROTAC) technology and triggered by lenalidomide (Fig. 3B) [131]. The hybrid zinc finger degron tag ZFP91-IKZF3 is added at the C-terminus of the CAR. The presence of lenalidomide can induce CRL4^CRBN E3-mediated ubiquitylation and proteasomal degradation of CAR, while CAR signaling can be restored by decreasing the lenalidomide levels. Although the two forms of switches mentioned above have not been used in clinical settings, they have good clinical suitability since these switches originate from nonimmunogenic human sequences, and the small molecule agents are FDA-approved. Additionally, CAR molecules can be incorporated with cleavable degradation moieties (degrons) called switch-off CARs (SWIFF-CARs) (Fig. 3C) [132]. SWIFF-CARs are controlled by hepatitis C virus (HCV) NS3 protease, and the administration of asunaprevir can inhibit the cleavage activity of HCV NS3 protease, preventing the degradation of protease/degron complex from CAR. As a result, the degron induces the rapid degradation of CAR. However, the strategies mentioned above all require the addition of a switch gene, which is challenging for the vector virus loaded with a CAR gene since the viral payload capacity is limited. Therefore, Park et al. generated a conditional scFv based on a camelid antibody, which is capable of recognizing both tumor-associated antigen (TAA) and methotrexate (MTX) (Fig. 3D) [133]. Upon binding MTX, the conformation of MTX-based scFv changes to disrupt TAA binding, resulting in CAR T-cell inactivation [133]; when MTX is removed, CAR T-cells are reactivated.

Alternatively, irreversible switches also represent a strategy to eliminate the activities and toxicities of CAR T-cells. It can be achieved through complement-dependent cytotoxicity (CDC) and antibody-dependent cell-mediated cytotoxicity (ADCC). CAR T-cells engineered with surface antigens can be neutralized by FDA-approved therapeutic antibodies, such as rituximab (targeting CD20) [134] and cetuximab (targeting truncated epidermal growth factor receptor, EGFRt) [135], and cause irreversible clearance (Fig. 3E). However, this strategy takes effect slowly and, therefore, may be unsuitable for patients with severe cytokine-mediated toxicities requiring urgent treatment. Another feasible solution is to express inducible caspase 9 (iCasp9) in CAR T-cells (Fig. 3F). iCasp9 is an apoptosis-triggering fusion protein, and rimiducid (AP1903) can induce the dimerization of iCasp9 and activate downstream caspase, eliminating the engineered CAR T-cells within 30 min [136, 137]. This strategy induced dramatic and immediate CAR T-cell elimination in a 26-year-old female, effectively alleviating the ICANS symptoms and improving the ICANS grades (from 3 to 1) and the Immune Effector Cell-Associated Encephalopathy (ICE) score (from 0 to 7) within 12 h [138]. The efficacy of iCasp9 CAR T-cells in CD19 B cell malignancies is still being evaluated in clinical trials (NCT03016377 and NCT03696784). Herpes simplex virus thymidine kinase (HSV-tk)/ganciclovir is another safety switch that inhibits DNA synthesis through death-inducing signaling cascades and, therefore, induces CAR T-cell death (Fig. 3G). However, its clinical applications are likely to be limited by its immunogenicity and slow induction of cell death [139, 140]. In addition, transfection with mRNA can construct a ‘biodegradable’ product with the transient expression of CAR molecules. CAR is downregulated along with the degradation of mRNA [141], reducing the risk of supercytokine production from the long-term overactivation of CAR T-cells (Fig. 3H). However, to achieve remission, multiple infusions of such CAR T-cells are required, which may increase the risk of phylaxis [142].

In summary, the off-switch is a promising strategy to eliminate toxicities, especially reversible switches, which avoid the permanent elimination of costly and robust CAR T-cells. More studies are expected to explore the on/off kinetics of such switches and confirm their immediate reactivity in vivo before translation into the clinic. Setting a specific timing for clinicians to turn off the switch is also of great importance. Moreover, safety issues associated with the small-molecule controller should be considered, and using approved agents as controllers may be better choices. In the future, clinical validations of the antitumor effects and toxicity-management ability of these novel products are needed.