Exhaustion of CAR T cells: potential causes and solutions

The idea of combining the function of antibody with T cell receptor (TCR) arose from the fact that the affinity of the TCR for peptide-major histocompatibility complex (MHC) complex is lower than that of the antibody-antigen complex. Therefore, replacement of the TCR variable region with the variable fragment (Fv) of a monoclonal antibody is expected to induce a stronger signal in T cells. Indeed, such chimeric T cell receptors (cTCRs) successfully form TCR complexes, recognize their ligands, and induce IL-2 production and cell lysis in T cell hybridomas [1]. Furthermore, to avoid pairing of cTCR and endogenous TCR, a single chain format was introduced using the single-chain variable region fragment (scFv) of monoclonal antibody as well as a CD3ζ chain as a signaling domain [2], which is designated as a first-generation chimeric antigen receptor (CAR). As this type of chimeric receptor does not require endogenous CD3 molecules (γ, δ, ε, and ζ) in the signaling complex, CAR can be applied to T cells as well as NK cells. However, the first-generation CAR cannot induce the activation of naïve T cells [3] because they require a second signal from costimulatory molecules for activation. Therefore, the CD28 intracellular domain was introduced into the CAR construct to provide both the first and second signals from a single chain (28z-CAR) [4]. This is designated as the second-generation CAR and has been widely utilized in clinical trials and therapeutics (Fig. 1A). Even the second-generation CAR with CD28 co-stimulatory domain, however, cannot mimic TCR signal in effector/memory T cells, which utilize TNF receptor superfamily members such as OX40 and 4-1BB as costimulatory molecule [5]. This has led to the development of the second-generation CAR with 4-1BB costimulatory domain (BBz-CAR), and third-generation CAR, which possesses multiple costimulatory domains. Furthermore, fourth-generation CAR has been designed to modify tumor microenvironment (TME) by secreting cytokines such as IL-12 [6] and IL-15 [7, 8].

Albeit CAR is designed to mimic TCR signaling, there are a number of differences in the activation mechanisms between CAR and TCR. First, 28z-CAR does not induce significant phosphorylation of CD3 components other than ζ chain, because CAR is designed to work independently of endogenous TCR components. Furthermore, 28z-CAR phosphorylates LAT to a lesser extent than TCR [9]. Although CD3ζ is a major component of the CD3 complex, its phosphorylation alone is not sufficient for full activation of the ZAP-70–LAT pathway. In addition, in another second-generation CAR containing the 4-BB costimulatory domain (BBz-CAR), LAT phosphorylation was also not detectable [9]. The reason why CAR can function without phosphorylation of LAT, which is an essential scaffold molecule for TCR signaling, was explained by Dong et al. [10]. They used LAT-deficient cell lines expressing either TCR or CAR and found that only CAR, but not TCR, can form a ligand-induced microcluster. Thus, CAR might have a bypass pathway for activating downstream signals, which has not been revealed yet.

CAR T cells against CD19 are one of the most successful adoptive transfer therapies, with a complete remission rate of 55% at 10.6 months in acute lymphoblastic leukemia (ALL) [11] and 29% at 40 months in chronic lymphocytic leukemia (CLL) [12]. However, effectiveness of CAR T cells in other cancer types, especially in solid tumors, remains to be improved. One of the major problems associated with unsuccessful CAR T cell treatment includes exhaustion of CAR T cells that results in impaired CAR T-cell expansion and persistence in vivo [13, 14]. In this review, we focus on CAR T cell exhaustion, which impair their functions, and discuss strategies to prevent it.

T cell exhaustion is characterized by the expression of checkpoint molecules as well as activation markers and a defect in effector functions, and was first described in chronic lymphocytic choriomeningitis virus infection [15]. It is induced by excessive antigen signaling and is considered to be one of the negative feedback mechanisms of T cell activation [16]. While it would be indispensable to stop immune responses at certain stages to prevent adverse effects such as autoimmune diseases [17], it also prevents the effects of cancer immunotherapy; tumor infiltrating T cells frequently become exhausted by excessive antigen stimulation and/or inhibitory TME [18]. Similarly, exhaustion is a common issue in CAR T cells that devastates their functions. For example, it was reported that anti-mesothelin CAR T cells in an orthotopic mouse model of pleural mesothelioma showed an exhausted phenotype represented by PD-1 upregulation and the defective expression of effector molecules such as IL-2, IFN-γ, TNF-α, and GRZB, resulting in impaired functions [19]. In addition, CD19 CAR T cells infused in patients with CLL [20] or B-cell lymphomas [21] tended to show exhaustion phenotype in non-responding patients. Furthermore, single cell RNA sequencing of CD19 CAR T infusion product also showed enrichment of exhaustion signature in patients with partial response or progressive disease compared to those with complete response [22]. These observations suggest that CAR T cell exhaustion is one of the major obstacles to successful treatment.

It has been recognized that nonactivated T and B cells at basal state transmit a low-level constitutive signal, designated as a tonic TCR or B cell receptor (BCR) signal, in the absence of ligands. Tonic signaling from TCR or BCR (including pre-BCR) in lymphocytes are involved in cell differentiation and maintenance of cellular responses to antigen stimulation [23, 24]. Similarly, in the case of CAR, various levels of ligand-independent receptor signaling caused by self-aggregation of CAR, also designated as tonic signal, have been observed [25]. Interestingly, higher levels of tonic signaling have been reported to result in the exhaustion and dysfunction of CAR T cells (Fig. 1B). The structure of CAR extracellular domain is thought to be responsible for tonic signaling. For example, CAR with IgG1 CH2-CH3 region as a spacer between the transmembrane domain and scFv generated a stronger tonic signal compared to that with CH3 only [26].

Some CAR T cells, such as c-Met IgG4 28z CAR T cells, show long-term in vitro proliferation without exogenous antigen stimulation or growth factors, which is designated as “continuous CAR”. In T cells expressing this “continuous CAR”, proliferation is considered to be mediated by self-activating tonic signal originating from CAR; for example, the c-Met IgG4 28z CAR in NFAT Reporter Jurkat cell line showed ligand-independent activation of the NFAT signal [27]. In addition, reduction of c-Met IgG4 28z CAR expression on cell surfaces ameliorated continuous proliferation, suggesting that ligand-independent activation of CAR signal is mediated by self-aggregation. Of note, these CAR T cells might be in the exhausted state in vitro because they show upregulation of T-bet and EOMES, which is associated with T cell exhaustion [28, 29]. Indeed, although c-Met IgG4 28z CAR T cells showed good cytotoxicity in vitro, their antitumor activity was compromised in vivo [27].

Another type of self-activating CAR T cells (GD2.28z) showed impaired proliferation and an exhausted phenotype in vitro [30]. This phenotype was attributed to the tonic CAR signaling activated by ligand-independent aggregation of CAR scFv, which was dependent on the framework region (FR) rather than on the complementarity-determining region (CDR) (Fig. 1A). Although GD2.28z CAR T cells showed strong cytotoxicity against target cells in vitro, their in vivo activity was very poor. In addition, GD2.28z CAR T cells upregulated the expression of checkpoint molecules such as PD-1, TIM-3, and LAG-3, which indicates exhaustion [31].

Interestingly, these two types of self-activating CARs induced different phenotypes in in vitro T cell proliferation. As mentioned above, c-Met IgG4 28z CAR T cells show continuous proliferation in vitro, but fail to persist in vivo, while GD2.28z CAR T cells exhibited poor proliferation both in vitro and in vivo. The differential in vitro proliferation capacity exhibited by the two functionally defective CARs may be explained by the difference in cytokine expression, because c-Met IgG4 28z CAR induced sustained IL-2 levels whereas GD2.28z CAR downregulated IL-2 expression. CAR signaling can upregulate IL2 transcription through the activation of NFAT and NF-κB, but this process is suppressed by TIM-3 [32]. It is possible that differences in the strength of constitutive CAR signaling may lead to differences in the strength and/or duration of the exhaustion phenotype, including TIM-3 expression, which affects the expression of the IL2 gene (Table 1).

As the self-activating CAR T cells spontaneously fall into exhaustion, they might be a convenient and desirable model to examine the molecular mechanisms of CAR T exhaustion. For example, RNA sequencing (RNAseq) analysis of highly exhausted CAR T-cells expressing a high-affinity version of GD2.28z (HA.28z) showed markedly different gene expression patterns compared to those in CD19.28z CAR T cells. Upregulated genes in HA.28z CAR T cells included activation-related genes (IFNG, GZMB, and IL2RA), inhibitory receptors (LAG and CTLA4), and inflammatory factors (CXCL8, IL13, and IL1A), while downregulated genes included memory-related genes (LEF1, TCF7, IL7R, and KLF2) [34]. This gene expression profile overlapped with that observed in exhausted T cells following a chronic viral infection [35], suggesting the feasibility of this model for T cell exhaustion.

It has been speculated that the difference in gene expression between exhausted and non-exhausted CAR T cells may reflect differential chromatin accessibility. For example, an assay for transposase-accessible chromatin using sequencing (ATAC-seq) analysis of HA.28z CAR T cells showed enrichment of AP-1-bZIP and bZIP-IRF binding motifs, in accordance with upregulated expression of JunB, IRF4, and BATF3, which antagonize the function of classical AP-1 [34]. In addition, when a time course of chromatin accessibility status of exhausted and non-exhausted CAR T cells during ex vivo expansion was examined using the same model, differential chromatin accessibility was detected earlier than when differential expression of exhaustion markers was observed, suggesting a causal role of chromatin remodeling for the CAR T cell exhaustion phenotype [36]. Tumor-specific exhausted T cells were reported to show two distinct chromatin states: a plastic dysfunctional state from which T cells can be rescued, and a fixed dysfunctional state in which the cells are resistant to reprogramming [37]. Similarly, CAR T-cells with tonic signaling show two-stage exhaustion-related chromatin remodeling [38], and the exhaustion-imprinted epigenome in CAR T cells can be reversed by termination of the tonic signal [38] as discussed below.

In addition to self-activation by tonic signaling, other cellular or environmental factors have been reported to induce exhaustion in CAR T cells. Some strategies for the prevention of exhaustion are presented and discussed below.

Despite the development and success of the CD19 CAR T cell therapy against leukemias, this approach has not met similar success in other tumor types, especially solid tumors. This may be explained, at least in part, by the possibility that CD19 CARs, specifically consisting of scFv derived from FMC63 clone, is a rare exception that does not generate tonic signals by itself. Such a characteristic of CD19 CAR might be a great advantage over other CARs, which show variable levels of tonic signaling and lead to exhaustion, not only during the ex vivo expansion but also in tumor tissues in vivo [30]. Even with CD19 CAR, however, exhaustion following continuous encounter with CD19 molecules on tumor cells is an obstacle for achieving long-term remission [20]. Therefore, substantial efforts are required to understand the mechanisms underlying CAR T cell exhaustion and overcome it with technological innovation.

Recently, new technologies such as RNA sequencing and global chromatin landscape mapping by ATAC-seq have significantly contributed to the understanding of T-cell exhaustion. Reprogramming of the exhaustion-associated epigenome can be induced following cessation of CAR tonic signaling through several approaches [38] (Fig. 2). Selection of 4-1BB co-stimulatory domain might be a better approach for prevention of CAR T cell exhaustion when CAR has self-activating potency [27, 30, 44]. In addition, reduction of tonic signaling by modifying the scFv structure or transient termination of CAR expression can be considered as a promising approach to fine-tune CAR signaling. Thus, when scFv is modified, absence of tonic signaling should be included as one of the critical parameters when selecting suitable human framework region. In addition, identification of amino acid residues critical for tonic signaling may help fine-tune the framework structure. Furthermore, revealing the molecular mechanisms underlying exhaustion-related epigenetic remodeling and identification of molecular or pharmaceutical approaches to inhibit them will be a powerful measure for countering CAR T cell exhaustion.

At present, various mechanisms of CAR T cell exhaustion have been identified independently by different approaches. Although each study found responsible molecules, general mechanisms have not been identified. Therefore, there is no guarantee that one strategy effective for certain exhausted CAR is also effective for different one. Among various strategies to attenuate exhaustion of self-activated CAR T cells we have reviewed, which is the first choice when CAR T cells are found to be self-activated? First of all, use of the 4-1BB costimulatory domain might be promising way because this method has been verified in multiple self-activating CARs [27, 30, 44], although there has been no consensus for superiority of 4-1BB-derived domains over others yet. Alternatively, other methods, such as use of dasatinib in ex vivo expansion [38] and blockade of inhibitory receptors [63], could be attempted because these approaches are easier to perform without need to modify CAR constructs and has major merit especially when modification of CAR construct changes the stability of CAR protein, resulting in loss of its cell surface expression. Nevertheless, at this moment these approaches, including others, could not be actively recommended, because most of them were demonstrated in a single study, but have not been repeatedly verified in other studies. It would thus be future task to analyze various kinds of exhausted CARs in unified analysis protocol to clarify the differences and similarities among them. Such attempt will provide further improvement in effectiveness of CAR therapies.