Overcoming Intrinsic Resistance of Cancer Cells to CAR T-Cell Killing

In the past several years, adoptive immunotherapies have gained unprecedented interest in oncology with the development of chimeric antigen receptors (CAR) T cells (1, 2). CAR T cells are T cells that are genetically engineered to express a CAR that includes a single-chain variable fragment (scFv) recognition domain specific for a tumor antigen linked to the signaling machinery of the T-cell receptor (TCR; CD3zeta) and a costimulatory domain (3). Unlike natural T cells, which recognize their target through major histocompatibility complex (MHC) presentation, CAR T cells can kill tumor cells in an MHC-independent manner (4). This approach demonstrated remarkable efficacy in relapsed or refractory (R/R) CD19 B-cell malignancies, leading to the FDA approval of four CD19-directed CAR T cells, in pediatric B-cell acute lymphoblastic leukemia (B-ALL; ref. 5), diffuse large B-cell lymphoma (DLBCL; refs. 6, 7), mantle cell lymphoma (8), and more recently follicular lymphoma (9). Despite these remarkable results, a significant number of patients treated with CAR T cells may experience primary (no response) or secondary (initial response followed by relapse/escape) resistance. Efforts are being made to understand the underlying mechanisms of resistance to CAR T-cell therapy to develop alternative strategies that would overcome these limitations. Three main mechanisms of resistance have been described thus far: (i) antigen escape; (ii) defective CAR T-cell function; and (iii) immunosuppressive microenvironment (10). However, recent data suggest that cancer cells may also survive CAR T-cell therapy due to intrinsic resistance mechanisms that render tumor cells insensitive to CAR T-cell killing (Fig. 1; ref. 11). To understand this novel mechanism of resistance, it is important to understand how CAR T cells kill their target. The apoptotic pathways are highly complex and involve a cascade of molecular events (12). These interactions are even more complex due to the tumor heterogeneity. Like natural cytotoxic T lymphocytes (CTL), CAR T cells can engage two distinct pathways to mediate tumor cell death: (i) the perforin/granzyme pathway and (ii) the death receptor pathway. These signaling cascades have in common the same terminal effector pathway, converging to the cleavage of caspase-3, finally triggering cell apoptosis through DNA damage. Moreover, similar to CTLs, CAR T cells are able to secrete cytokines that can act as effectors for target cell killing. The perforin/granzyme axis has been considered as the most important pathway for CD8⁺ and CD4⁺ CTLs cytotoxic functions and for CAR T cells by extension (13). However, recent data suggest that the death receptor axis and cytokine-mediated killing should not be overlooked when deciphering CAR T-cell–mediated killing (Figs. 2 and 3). Here, we review the data that support the critical role of the intrinsic mechanisms of resistance and discuss the therapeutic implications that could be employed to overcome this resistance. CAR T-cell resistance is a complex and dynamic process. Intrinsic resistance as described in this review is one of the many resistance mechanisms potentially involved in CAR T-cell failure.

CTLs can exert their cytotoxic function on tumor cells by secreting preformed granules containing perforin, a pore-forming protein, and granzymes, which are serine proteases (14). Upon T-cell receptor engagement, CTLs link to their target cells, establishing the immune synapse (IS), a highly organized structure that acts as a signaling platform for bidirectional interactions (15). After TCR-mediated target-cell recognition, cytotoxic granule contents are released in a calcium-dependent manner into the IS. When released, perforin induces channel formation on the target cell facilitating proapoptotic granzymes entry. Thus far, granzyme B has received the most attention because it activates the caspase-dependent pathway of apoptosis through (i) caspase-10 activation; (ii) facilitation of cytochrome C release from the mitochondrial intermembrane space; and (iii) direct caspase-3 targeting (16). Moreover, granzyme B–induced cell death is not abolished by complete caspase inhibition, unraveling the ability of granzyme B to induce cell death in a caspase-independent manner (17). Indeed, one important function of granzyme B is to cleave Bid (BH3-interacting domain death agonist), a member of the proapoptotic Bcl-2 (B-cell lymphoma 2) family, generating truncated Bid (tBid). tBid translocates to the mitochondria and activates Bax (Bcl-2 homologous antagonist killer) and Bak (Bcl-2 associated X protein), two proapoptotic members of the Bcl-2 family, which results in the release of cytochrome C. Once released in the cytosol, cytochrome C binds to and activates apoptotic protease activating factor 1 (Apaf-1) and procaspase-9. This molecular platform called “apoptosome” enables the activation of functional caspase 9 (18). Granzyme B–mediated release of cytochrome C is inhibited by Bcl-2 and Bcl-xL, two antiapoptotic mitochondria localized molecules. Granzyme B is also able to directly disrupt the mitochondrial transmembrane potential in a caspase- and Bid-independent manner (19). In parallel, the release of second mitochondrial-derived activator of caspases (Smac) promotes apoptosis by inhibiting the activity of the inhibitor of apoptosis proteins (IAP; refs. 20, 21). Indeed, IAP represent a family of caspase inhibitors that bind to and inhibit caspases 3 and 9, thereby preventing apoptosis (22). Among IAP family members, XIAP (X-linked IAP), cIAP1, and cIAP2 are the most studied. Finally, the granzyme B signaling cascade converges toward the execution pathway through the activation of the effector caspase-3 and leads to cell death. Finally, another granzyme molecule, granzyme A, that is contained in cytotoxic granules induces cell death in a caspase-independent manner (23). This pathway is still not completely understood, as downstream caspase substrates are not involved, and cytochrome C is not released.

Unlike CTLs, CAR T cells can recognize antigens without the need for MHC presentation. Although CAR T cells utilize, at least in part, the conventional TCR signaling machinery, substantial differences in the IS structure have been observed (24). This nonclassical CAR T-cell target cell IS is capable of accelerating the delivery of lytic granules resulting in faster killing of the target cells (25). In vitro coculture settings of CAR T cells with targeted antigen-expressing cells demonstrated upregulation of CD107a, a degranulation reporter, and secretion of perforin and granzyme B, suggesting the involvement of the exocytosis pathway (25–27). Moreover, blocking perforin release with egtazic acid (EGTA), a calcium ion chelator, or using perforin-deficient CAR T cells was shown to significantly decrease CAR T-cell mediated killing in vitro (28, 29). In vivo experiments reported an involvement of the exocytosis pathway for CAR T cells mediated killing and tumor eradication that was only decreased and not abolished using perforin-deficient CAR T cells (30–32). Lack of perforin negatively affected both CD8⁺ and CD4⁺ CD19-directed CAR T-cell cytotoxicity and resulted in incomplete B-cell aplasia (31). These observations suggest that CAR T-cell killing could be mediated by other signaling pathways, including the death receptor pathway (Figs. 2 and 3; Table 1).

Together with calcium-dependent granule exocytosis, calcium-independent death receptor-mediated killing is a major axis by which target cells are lysed by CTLs. The death receptor signaling pathway of apoptosis involves interactions between death receptors expressed on the target cells and their ligands on effector cells (33). These death receptors are upregulated upon T-cell activation caused by target cell recognition. They are members of the TNF receptor superfamily and are characterized by their intracytoplasmic death domains (34). These domains are responsible for the death signaling transmission from the cell surface to the intracellular signaling pathways upon receptor trimerization by its ligands. Multiple ligand/death receptor couples have been described. The best-characterized are FasL/Fas, TRAIL (TNF-related apoptosis including ligand)/TRAIL-R1/2, and TNFα/TNFR1 (35–37). Interaction between death receptors and their ligands results in the activation of caspase-8, mediated by the adapter protein FADD (Fas-associated protein) and pro–caspase-8, which form the death-inducing signaling complex (DISC; refs. 38–40). Ultimately, caspase-8 activation leads to the execution phase of apoptosis through caspase-3 activation.

Recently, Upadhyay and colleagues established an in vitro model in which GFP-specific CTLs were cocultured with both GFP-expressing (GFP⁺) and GFP-nonexpressing (GFP⁻) lymphoma cells (11). They showed that GFP⁺ cells eradication by GFP-specific CTLs was abrogated under Fas knockout in target cells despite CTL proliferation, granzyme B production, and degranulation. Moreover, CTLs exerted bystander killing of GFP⁻ lymphoma cells in a Fas-dependent manner. Consistent with these results, they observed earlier recurrence of Fas^−/− GFP⁺ tumors compared with the WT counterparts and blunted response to the therapy for tumors with Fas^−/− GFP⁻ bystander cells upon GFP-specific CTLs adoptive transfer in vivo. They demonstrated that bystander cytotoxicity was geographically restricted and dependent on colocalized GFP⁺ and GFP⁻ tumor cells. Fas-mediated on-target and bystander apoptosis was increased after treatment with inhibitors of proteins with known modulatory activity within the Fas pathway, the Smac mimetic birinapant and the Bcl-2/xL inhibitor ABT-737. In line with these results, death receptor–mediated apoptosis was shown to be crucial for tumor cell sensitivity to T-cell killing in patients with cancer treated with immune checkpoint blockers (41, 42).

In vitro cytotoxic assays with death receptor pathway disruption using anti-FasL antibodies or Fas-deficient target cells resulted in reduced CAR T-cell antigen-dependent triggered apoptosis (11, 29, 43). Thus, this apoptosis pathway is required for full CAR T-cell effector functions. Interestingly, it has been shown that CAR T cells are also able to mediate tumor cell bystander killing in an antigen-independent manner via a contact-dependent Fas/FasL interaction or via the TRAIL axis (11, 44, 45). Preactivation of CAR T cells upon antigen exposure was necessary to observe such alternative killing mechanism. Indeed, CAR engagement induces upregulation of FasL and TRAIL expression enabling the CAR T cells to kill tumor cells using the death receptor pathway in both an antigen-dependent and independent manner. Antigen-independent killing offers a path that could be exploited to circumvent relapses due to antigen loss or when degranulation is poor or hampered (46). In vivo, CD19 CAR T-cell efficacy was similar against homogeneous CD19⁺ tumors and heterogeneous tumors composed with both CD19⁺ and CD19⁻ cells (11). In these murine models, FasL blockade severely impaired CAR T-cell functions reaffirming death receptor pathway dependence of on-target CAR T-cell cytotoxicity. Moreover, upon FasL blockade, mice bearing heterogeneous tumors had a significantly worse survival than their counterparts bearing homogeneous tumors, suggesting that the Fas–FasL axis was critical for CAR T-cell bystander cytotoxicity against CD19-loss variants.

Recently, Singh and colleagues addressed the question of tumor-intrinsic resistance to CAR T cells by performing an unbiased genome-wide loss-of-function screen in B-ALL cell lines under CD19 CAR T-cell pressure (47). Using CRISPR–Cas9 technology, they found that disruption of genes associated with proapoptotic death receptor signaling pathway such as FADD, BID, CASP8, and TNFRSF10B conferred resistance to CAR T-cell killing. Conversely, knock out of antiapoptotic genes such as CFLAR, TRAF2, and BIRC2 (encoding for cIAP1) led to an increased susceptibility to CAR T-cell killing. Moreover, CRISPR targeting of the genes encoding for FasL or TRAIL in CAR T cells resulted in a decreased killing, confirming that the death receptor pathway is crucial for CAR T-cell cytotoxicity. In their model, impairing apoptosis by targeting death receptors or their downstream signaling partners led to CAR T-cell dysfunction. In fact, loss of death receptor signaling is responsible for decreased tumor killing and persistent antigen exposure, which in turn causes CAR T-cell exhaustion through a global reprogramming of CAR T cells at the transcriptomic and epigenetic levels. Using samples from two multi-center clinical trials for relapsed pediatric B-ALL, they found a significantly higher death receptor signaling signature in samples from patients who had achieved complete remissions (CR) compared with patients who did not respond although they retained CD19 expression (NR). No difference was seen between CRs and NRs in the intrinsic apoptosis signature. This study unravels a novel mechanism of resistance to CAR T cells that is antigen-independent and relies on death receptor pathway disruption in tumor cells.

Dufva and colleagues validated death receptor genes FADD and TNFRSF10B as major contributors of tumor-intrinsic resistance to CAR T-cell cytotoxicity on leukemia and lymphoma cell lines (48). In agreement with Singh and colleagues, disruption of these genes resulted in decreased tumor sensitivity to CAR T-cell cytotoxicity. To further identify crucial pathways for CAR T-cell function and candidate molecules to modulate cytotoxicity, they carried out a high-throughput drug screen using a coculture assay with CD19-directed CAR T cells CD19 NALM6 B-ALL. Strikingly, the three drugs that most significantly enhanced CAR T-cell cytotoxicity all belonged to the same class, namely Smac mimetics or IAPi (birinapant, AT-406, and LCL-161), consistent with results obtained with CTLs (11).

These studies unravel a critical role for the death receptor pathway of apoptosis and highlight the link between a death receptor signaling signature and sensitivity to CAR T-cell therapy. These observations suggest the potential benefit of strategies aimed to sensitize tumor cells to this type of apoptosis. Lowering the threshold necessary to induce death receptor-mediated apoptosis may indeed overcome intrinsic tumor cell resistance to CAR T-cell therapy. It is well known that cancer therapeutic agents can modulate death receptor expression by tumor cells in p53-dependent and p53-independent manners (49). Multiple cancer therapeutic families can increase Fas, TRAIL-R1/2, or both on tumor cells, including DNA-damaging agents (50), histone deacetylase (HDAC) inhibitors (51), proteasome inhibitors (52), and cyclooxygenase-2 (COX2) inhibitors (53). HDAC inhibitors (SAHA and LBH589) and COX2 inhibitors (celexoxib) were able to partially reverse CD19 CAR T-cell resistance of non-Hodgkin lymphoma cell lines due to altered death receptor pathway machinery (54). In these cell lines, pretreatment with HDAC inhibitors or COX2 inhibitors enhanced tumor cell sensitivity to TRAIL-mediated apoptosis to overcome acquired intrinsic resistance despite CD19 expression. Similarly, low-dose radiation could sensitize tumor cells to CAR T-cell killing by targeting gene sets associated with sensitivity to TRAIL-mediated death, without increasing target antigen expression (44). This strategy increased both on-target and bystander CAR T-cell cytotoxicity, enabling tumoral control even in the case of target antigen loss variants. In vivo, the combination of IAP inhibition by birinapant with CAR T-cell therapy increased tumor cell death in a perforin-independent and TNF-dependent manner (55). In a human-HER2 transgenic mouse model, this combination was required to achieve antitumor activity, whereas each treatment alone failed to control tumor growth, demonstrating that the addition of birinapant enhanced CAR T-cell antitumor efficacy. Under these conditions, CAR T cells expanded >100-fold upon Smac mimetic exposure in vitro, showing lack of overt toxicity of IAP inhibition on CAR T cells.

Bcl-2 is an inhibitor of the intrinsic apoptosis pathway by blocking Bax and Bak and thus cytochrome C release, potentially causing treatment resistance. Thus, the combination of Bcl-2 inhibitors with CAR T cells appears to be an attractive strategy. In a B-ALL and CD19 CAR T-cell coculture model, it has been shown that CAR T-cell mediated caspase-3 targeting in tumor cells could be improved through Bcl-2 inhibition. Bcl-2 inhibition was responsible for a change of tumor cell phenotype with increased levels of death receptors. Moreover, Bcl-2 inhibitors are responsible for the upregulation of TRAIL-R2 by tumor cells and render them more sensitive to death receptor–mediated apoptosis (56). Thus, Bcl-2 inhibition decreases the apoptosis threshold and enhances CAR T-cell cytotoxic functions. However, one may keep in mind that, physiologically, Bcl-2 protects T cells from activation-induced cell death (AICD) by inhibiting death receptor-mediated apoptosis (57). This raises the concern Bcl-2 inhibitors may negatively impact CAR T-cell survival. First, Mamonkin and colleagues showed high expression levels of Bcl-2 in CAR T cells, which seemed sufficient to protect them from apoptosis by Bcl-2 inhibitors. Second, Bcl-2 inhibitors were able to presensitize tumor cells to CAR T-cell therapy, enabling their use only before the adoptive transfer.

CTLs also produce several cytokines, such as TNFα and IFNs that have cytotoxic function when secreted in the vicinity of target cells. TNFα is the soluble ligand of the death receptor TNFR1 and is implicated in the death receptor pathway of apoptosis as described above. IFNs are naturally secreted glycoproteins produced by almost every cell type as a mechanism of defense, possibly in response to tumor cells (58). IFNs are classified into three subfamilies. Type I IFNs composed of multiple members, including IFNα, type II IFNs contain IFNγ as the only member, and type III IFNs (59, 60). Functionally, when IFNs bind to the extracellular domain of their receptor, the intracellular portion triggers activation of the Janus kinase (JAK) through its phosphorylation. In turn, activated JAK phosphorylates signal transducer and activator of transcription proteins (STAT). Activated STATs dimers localize from the cytosol to the nucleus and initiate transcription of IFN-stimulated genes (ISG; refs. 61, 62). As IFNs are able to regulate the expression of approximately 2,000 ISGs, their biological effects are pleiotropic and depend on the IFN transcriptional signature expressed by the target cell in response to IFNs exposure, which can be modulated by other signaling pathways. Thus, IFNs are able to sensitize some cells to apoptosis by lowering the threshold of the intrinsic death pathway (63). However, not all tumor cells are susceptible to the apoptotic effects of IFNs (64). Conversely, some ISGs have prosurvival functions. For example, unphosphorylated STAT1 (U-STAT1) can drive the expression of ISGs with survival functions and be responsible for cancer resistance to cancer therapies (65, 66). Studies in tumor-bearing mice indicate that the IFNγ signaling plays an essential role in antitumor immune response mediated by immune checkpoint blockers (67, 68). Consistent with this observation, melanoma patients who are refractory to immune checkpoint blockers harbor a much higher rate of genomic defects in the IFN-γ signaling pathway as compared with responders (68, 69). Interestingly, the cytokines TNFα and IFNγ produced by CTL increase the sensitivity of tumor cells to CTL killing by both granule- and FasL-mediated death pathways, unravelling synergistic effect of cytotoxic mechanisms (70).

CAR T cells are known for their ability to secrete a high amount of IFNγ when exposed to target cells in vitro (27). In vivo, tumor control after adoptive transfer in colon carcinoma–bearing mice was less effective with IFNγ-deficient CAR T cells compared with wild-type (WT) CAR T cells (30). Recently, Ishii and colleagues established a syngeneic murine model to compare perforin-deficient and WT CD19 CAR T-cell therapy against B-ALL (31). Perforin-deficient CAR T cells demonstrated inferior in vitro cytotoxicity with slower kinetics of leukemia cell killing compared with WT CAR T cells. In parallel, perforin-deficient CAR T cells produced significantly higher amounts of IFNγ and TNFα. Consistent with these in vitro observations, perforin-deficient CAR T cells were less efficient at eradicating leukemia in vivo, although this difference could be overcome with a higher cell dose. Moreover, they showed that IFNγ neutralization negatively impacted the ability of perforin-deficient CAR T cells to eradicate leukemia in vitro and in vivo, suggesting that the IFNγ pathway may take over the exocytosis pathway in case of dysfunction. Finally, a recent study by Arenas and colleagues demonstrated that the disruption of IFNγ signaling, notably through the downmodulation of JAK2, conferred resistance to killing of cancer cells by CAR T cells directed against HER2 in preclinical models (HER2-driven cell lines and patient-derived xenografts; ref. 71). Given the potential contribution of IFNγ in CAR T-cell killing, strategies that aim to increase IFNγ secretion by CAR T cells may enhance their efficacy. This may be achieved by using 4th generation CAR T cells called TRUCKs for “T cells redirected for universal cytokine-mediated killing”, which can secrete cytokines and increase IFNγ production directly or indirectly (72). Avanzi and colleagues designed armored, IL18-secreting CAR T cells, which increased IFNγ secretion and tunned the tumor microenvironment toward an IFNγ signature (73). These IL18-secreting CAR T cells were able to modulate the tumor microenvironment and enhance the endogenous antitumor immune response. IL12 is another proinflammatory cytokine, which has been shown to increase the antitumor efficacy of CAR T cells (74). In a syngeneic tumor model, IL12 promoted IFNγ secretion by CAR T cells in an autocrine fashion. This secretion rendered CAR T cells resistant to the immunosuppression induced by regulatory T cells and myeloid-derived suppressor cells, and thus improved CAR T-cell cytotoxicity. These armored CAR T cells, by shaping the cytokine landscape within the tumor, represent a promising strategy to improve the efficacy of adoptive T-cell therapy. However, these approaches may be limited by the toxicity of high concentrations of IFNγ.

Learning how CAR T cells mediate tumor cell killing is a crucial step toward a better understanding of the mechanisms of primary and acquired resistance. Such understanding offers the opportunity to envision strategies to overcome intrinsic resistance of tumor cells to CAR T-cell killing. For a long time, it was believed that CAR T cells exerted their cytotoxic function mostly through degranulation involving the perforin/granzyme axis. However, recent studies suggest that the death receptor pathway of apoptosis, which is triggered by death receptors and their ligands, is crucial for CAR T-cell functions in vivo. These studies demonstrated that tumor cells harboring aberrant apoptotic machinery that renders them insensitive to the death receptor signaling are intrinsically resistant to CAR T-cell killing. From a clinical point of view, patients experiencing durable clinical responses upon CAR T-cell treatment have a significantly higher sensitivity to the death receptor pathway of apoptosis (6, 11). Thus, lowering the threshold necessary for CAR T cells to trigger tumor cell killing through the death receptor pathway may enhance tumor eradication. Among candidate molecules, IAP inhibitors, Smac mimetics, and Bcl-2 inhibitors appear promising. Furthermore, these approaches may enable the eradication of antigen-defective subsets of tumor cells by CAR T cells through an antigen-independent death receptor–dependent bystander killing. Although intrinsic resistance of cancer cells to immune killing may be important, other mechanisms may also contribute to CAR T-cell resistance, including loss of target antigen and immune suppression of CAR T cells by the tumor microenvironment. These alternative mechanisms may limit the clinical impact of strategies meant to increase or restore the sensitivity of tumor cells to apoptosis.

M. Ruella reports grants and other support from Novartis and Tmunity; grants and personal fees from Abclon; personal fees from GSK, BMS, Bayer, and NanoString; and other support from viTToria Biotherapeutics during the conduct of the study, as well as grants and other support outside the submitted work and a patent issued, licensed, and with royalties paid. No disclosures were reported by the other authors.

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.