Clinical implications of T cell exhaustion for cancer immunotherapy

Abstract

Immunotherapy has been a remarkable clinical advance in the treatment of cancer. T cells are pivotal to the efficacy of current cancer immunotherapies, including immune-checkpoint inhibitors and adoptive cell therapies. However, cancer is associated with T cell exhaustion, a hypofunctional state characterized by progressive loss of T cell effector functions and self-renewal capacity. ‘Un-exhausting’ T cells in the tumour microenvironment is commonly regarded as a key mechanism of action for immune-checkpoint inhibitors, and T cell exhaustion is considered a pathway of resistance for cellular immunotherapies. Several elegant studies have provided important insights into the transcriptional and epigenetic programmes that govern T cell exhaustion. In this Review, we highlight recent discoveries related to the immunobiology of T cell exhaustion that offer a more nuanced perspective beyond this hypofunctional state being entirely undesirable. We review evidence that T cell exhaustion might be as much a reflection, as it is the cause, of poor tumour control. Furthermore, we hypothesize that, in certain contexts of chronic antigen stimulation, interruption of the exhaustion programme might impair T cell persistence. Therefore, prioritization of interventions that mitigate the development of T cell exhaustion, including orthogonal cytoreduction therapies and novel cellular engineering strategies, might ultimately confer superior clinical outcomes and the greatest advances in cancer immunotherapy.

ToC blurb

T cells are key effectors of immunotherapies that have revolutionized the treatment of cancer; however, chronic exposure to tumour-associated antigens can result in progressive loss of T cell effector functions and self-renewal capacity, a hypofunctional state termed ‘T cell exhaustion’ that is believed to limit the efficacy of immunotherapy. This Review synthesizes the new immunobiological insights that present a more nuanced view beyond T cell exhaustion being entirely undesirable and indicate that this hypofunctional state might be a reflection — rather than the cause — of poor tumour control. Hence, the authors describe how, in certain contexts, interruption of this programme could impair T cell persistence and discuss interventions to mitigate the development of T cell exhaustion that might ultimately improve clinical outcomes.

Introduction

Immunotherapies, particularly immune-checkpoint inhibitors (ICIs) and adoptive cell transfer, have revolutionized the treatment of cancer^1,2. These advances have led to substantial research and clinical interest in T cell exhaustion, which is a cellular differentiation pathway associated with progressive declines in T cell effector functions in the setting of chronic antigen exposure (BOX 1). In this context, persistent stimulation of the T cell receptor (TCR) and downstream Ca²⁺ signalling activates the transcription factor NFAT, which in turn drives expression of transcription factors from the TOX^3,4 and NR4A⁵ families. Thus, NFAT signalling in T cells leads to transcriptional upregulation of multiple inhibitory immune-checkpoint proteins, including PD-1, CTLA4, CD39 and LAG-3⁶. The process of T cell exhaustion spans a spectrum from highly proliferative T cells with stem-like properties (termed ‘precursor exhausted T cells’) to T cells with complete loss of effector function and replicative capacity (frequently termed ‘terminally exhausted T cells’). While the precise definition of an exhausted T cell remains a subject of healthy debate⁷, we herein define T cell exhaustion as a state characterized by reduced effector function at the time of tumour cell encounter that is associated with expression of co-inhibitory markers and reduced production of inflammatory cytokines (such as IFNγ and TNF). Although parallels can be drawn with other hypofunctional T cell states, such as ignorance, anergy and tolerance (BOX 1)⁶, T cell exhaustion typically relates to T cells that have already been activated (primed) by antigen-presenting cells in secondary lymphoid organs and are undergoing secondary antigen encounter in the tumour microenvironment (TME). T cell exhaustion, as assessed by PD-1 expression on intratumoural lymphocytes, is frequently associated with poor patient survival across a number of different cancer types^8–15; therefore, a broader understanding of factors underlying this T cell state could enable further improvement of cancer immunotherapies.

Box 1 |. Hypofunctional states of T cells.

Multiple mechanisms can result in T cell hypofunction, including ignorance, anergy, tolerance and exhaustion, defined as follows.

• Ignorance: A lack of T cell response owing to a low level of antigen expression, usually early during tumorigenesis.

• Tolerance: Central or peripheral inactivation of self-reactive T cells.

• Anergy: A hypofunctional state of T cells following antigen-induced T cell receptor (TCR) and downstream NFAT signalling in the absence of co-stimulatory signals from professional antigen-presenting cells (APCs) and/or CD4⁺ T cell help. Anergic T cells can express the inhibitory immune-checkpoint receptors PD-1 and CTLA4, and thus generally are not distinguishable from exhausted T cells.

• Exhaustion: Progressively declining T cell function owing to chronic TCR stimulation in the setting of persistent antigen exposure. What distinguishes this process from anergy is that these T cells have undergone productive initial activation (for example, with co-stimulation by professional APCs in secondary lymphoid organs). Exhausted T cells are characterized by reduced cytokine production and the expression of inhibitory receptors, such as PD-1, CTLA4 and LAG-3, as well as the immunosuppressive enzyme CD39. These receptors are also expressed by functional T cells following physiological, APC-mediated priming; however, when expressed intratumourally (particularly in patients with cancer in whom tumours typically form over the course of months to years), these markers denote exhausted T cells. A spectrum of T cell exhaustion states exists, ranging from precursor exhausted to terminally exhausted, with concordant reductions in effector function and proliferative capacity.

Notably, these mechanisms were first defined under specific experimental conditions. In a clinical context, no definitive phenotype has been identified that can distinguish between these T cell states. Moreover, rather than manifesting a specific T cell phenotype, each of these states can result in an absence of the hypofunctional T cell clones intratumorally, further complicating identification of the relevant mechanism. T cells found in the tumour might be hypofunctional owing to anergy or exhaustion, or both. Hence, the composite term ‘dysfunctional’ has been favoured by some groups^6,7. Other hypofunctional states of T cells have been described that overlap with the four typical categories described above, including terminal differentiation²³⁴, activation-induced cell death²³⁵, senescence²³⁶, burned-out²³⁷ and impaired metabolism²³⁸. Indeed, the degree of overlap among these hypofunctional T cell states is an area of ongoing research, and further clarification of the transcriptional, epigenetic and metabolic pathways that underlie each of these cell programmes will lead to a better understanding of their relative contributions to inadequate tumour control.

The initial report of T cell exhaustion in the presence of high levels of antigen exposure dates back to the 1960s, prior to the development of immunological techniques enabling longitudinal analyses of T cell specificities (reviewed in¹⁶). In 1998, the groups of Rafi Ahmed and Rolf Zinkernagel used fluorochrome-labelled peptide–MHC tetramers to track lymphocytic choriomeningitis virus (LCMV)-specific CD8⁺ T cells following infection of mice with this virus^17,18. This innovation facilitated the first description of a hypofunctional CD8⁺ T cell state associated with an inability to clear a viral pathogen^17,18. In 2006, Ahmed’s group and their collaborators further demonstrated that even the markedly impaired responses of ‘helpless’ CD8⁺ T cell against LCMV (in mice depleted of CD4⁺ T helper cells) could be rescued through antibody-mediated blockade of the PD-1/PD-L1 immune checkpoint¹⁹. Concurrent with these discoveries of hypofunctional T cell states in the setting of chronic viral infections, the groups of James Allison and Jeffrey Bluestone discovered in the mid-1990s that CTLA4 acts as an co-inhibitory receptor in T cells^20,21. Allison and colleagues also demonstrate that antibody-mediated blockade of CTLA4 enhances antitumour immunity in mice²⁰. Work by the groups of Tasuku Honjo, Arlene Sharpe and Gordon Freeman subsequently identified the PD-1–PD-L1 axis as another immune checkpoint limiting T cell function^22,23. Studies from the laboratories of Honjo and Lieping Chen demonstrated that abrogation of PD-L1 enhanced antitumour responses in mice^24,25. In 2011, a gene-expression profiling study by the Speiser group showed that human melanoma-reactive tumour-infiltrating lymphocytes (TILs) have an ‘exhaustion profile’ similar to that previously reported for chronic viral infection, including expression of CTLA4 and PD-1²⁶. Pivotal preclinical and translational studies such as these paved the way to the first approved immunotherapies targeting CTLA4 and PD-1, initially for metastatic melanoma^27–29 followed rapidly by an expanding list of approvals across a range of other cancer types². Seminal work by multiple groups has demonstrated that T cell exhaustion is epigenetically encoded, for example, through reduced chromatin accessibility of genes involved in effector functions and increased accessibility of the genes encoding TOX transcription factors and co-inhibitory receptors, and that PD-1 inhibition can only partially reverse the transcriptional phenotype associated with exhaustion^30–32. Taking these remarkable discoveries together, a straightforward interpretation emerges in which removal of the molecular ‘brakes’ from T cells using ICIs can reinvigorate clonally expanded CD8⁺ T cells and thereby result in clinical benefit, albeit only certain T cell subsets can be reinvigorated. Notwithstanding, logical hypotheses that extend from this conceptual understanding of immunological exhaustion include: (1) abrogation of additional immune checkpoints (such as LAG-3, TIM-3 and/or TIGIT) at the tumour–T cell interface might improve on the success of CTLA4 and PD-(L)1 inhibitors; and (2) interruption of the molecular exhaustion programme can yield therapeutic benefit in the setting of T cell-mediated immunotherapy.

Parallel to the rise of ICI therapy for solid tumours, adoptive transfer of genetically engineered chimeric antigen receptor (CAR) T cells has revolutionized therapy for patients with treatment-refractory lymphoid malignancies³³. CAR T cells are autologous (or potentially allogeneic) T cell products with redirected antigen specificity, achieved through expression of a exogenous CAR construct consisting of an extracellular tumour-targeting motif (typically an antibody-derived single-chain variable fragment) coupled to intracellular signalling domains derived from the TCR CD3ζ chain and co-stimulatory receptors^33,34. Initial clinical response rates are typically high with CAR T cells targeting CD19-expressing and BCMA-expressing B cell malignancies^{35–38,39,40}, and these therapies have proven effective and even curative for a subset of patients^41,42.

In spite of these important advances, most patients receiving the currently approved CAR T cell therapies will ultimately have disease relapse. As representative examples, the ZUMA-1 study of axicabtagene ciloleucel in adults with multiply-relapsed large B cell lymphoma (LBCL) showed a 42% continuing response rate at a median follow-up duration of 15 months³⁵, and the ELIANA study of tisagenlecleucel in paediatric and young-adult patients with B cell acute lymphoblastic leukaemia (B-ALL) revealed relapse-free survival of <50% even among responders and despite a high rate of consolidation with allogeneic haematopoietic stem cell transplantation⁴³. Disease relapse eventually occurs in almost all patients receiving BCMA-directed CAR T cell products for multiple myeloma³⁹, and durable response rates with most investigational CAR T cell products for solid tumour types are low⁴⁴. These data emphasize the need for a better understanding of disease relapse following CAR T cell therapy, including the potential contribution of T cell exhaustion, with the ultimate aim of enhancing the existing therapies and expanding this treatment modality to new diseases, including solid malignancies.

Herein, we review the important immunological insights into T cell exhaustion gleaned over the past decade. In light of these findings, we propose that T cell exhaustion is a reflection that the totality of antineoplastic efforts is insufficient to successfully reduce tumour antigen burden, rather than simply a cause of inadequate tumour control. Moreover, we discuss how abrogation of the exhaustion programme could, in certain contexts, be counterproductive for T cell-mediated antitumour immunity. We do not intend to imply that T cell exhaustion is not clinically actionable; instead, we propose that strategies focused on mitigating (or avoiding), as opposed to interrupting, T cell exhaustion might yield greater clinical efficacy. This shift in thinking will be informative for the design of next-generation immunotherapies that build on the clinical successes of ICIs and CAR T cell therapies. While CD4⁺ T cells also undergo T cell exhaustion in cancer^45–47, CD8⁺ T cell exhaustion is better characterized, particularly in the setting of ICI treatment. Thus, the early sections of this Review are focused on CD8⁺ T cell exhaustion specifically, whereas the later sections on adoptive T cell therapies are subset-agnostic given that infusion of both CD4⁺ T cells and CD8⁺ T cells is crucial for the efficacy of such therapies⁴⁸.

A balanced view of T cell exhaustion

Exhausted CD8⁺ T cells are enriched for tumour-reactive clones.

In 2014, Gros et al.⁴⁹ reported that in patients with melanoma, CD8⁺ TILs expressing PD-1, LAG-3 and/or TIM-3 comprise an oligoclonal T cell population enriched for clonotypes capable of recognizing autologous tumour cells. Subsequent work by the same group demonstrated that circulating PD-1⁺ lymphocytes in patients with melanoma and gastrointestinal cancers are also enriched for tumour-reactive and neoantigen-specific T cells^50,51. The preferential tumour-antigen reactivity of intratumoural PD-1^highCD8⁺ T cells was also demonstrated in patients with non-small-cell lung cancer (NSCLC)⁵². Moreover, tumour antigen-linked MHC tetramers were reported to preferentially bind to CD39-expressing CD8⁺ TILs in lung and colorectal cancers (as opposed to bystander, mostly virus-specific CD8⁺ T cells lacking CD39 expression)⁵³. Concordantly, CD39⁺CD103⁺CD8⁺ TILs from patients with melanoma or head and neck cancer demonstrated selective reactivity to autologous tumour cells and, for the one patient analysed, enhanced cytotoxicity compared with CD103⁻ and/or CD39⁻ CD8⁺ TILs from the same patient⁵⁴. Of note, the CD39⁺CD103⁺CD8⁺ TILs also expressed high levels of PD-1, CTLA4 and TIM-3⁵⁴.

A subsequent single-cell RNA sequencing study of human melanomas revealed that greater than one-third of CD8⁺ TILs has a T cell dysfunction-related gene signature, characterized by high expression of PDCD1 (encoding PD-1), ENTPD1 (encoding CD39), LAG3, HAVCR2 (encoding TIM-3), ITGAE (encoding CD103), CXCL13, TNFRSF9 (encoding 4-1BB) and a number of cellular proliferation-related genes, which surprisingly suggested enhanced rather than reduced replicative capacity⁵⁵. Moreover, CD8⁺ TILs obtained from patient biospecimens with enrichment for this transcriptional signature of T cell dysfunction preferentially demonstrated reactivity against autologous tumour digests⁵⁵. In a Review published in 2020, the authors summarized these findings and identified PD-1, CD39, TIM-3, LAG-3, CXCL13, 4-1BB, GITR, proliferation markers and an oligoclonal TCR repertoire as hallmarks of tumour-reactive CD8⁺ T cells⁵⁶. In several more-recent studies across multiple solid tumours, CD8⁺ TILs with empirically validated tumour reactivity were all marked by expression of PDCD1, ENTPD1 and CXCL13^57–60. Notably, many of these markers of tumour reactivity are also generally consider to be markers of CD8⁺ T cell exhaustion. By contrast, bystander T cells, including those with specificity for viral pathogens, typically lack expression of these markers⁶¹, and thus have low expression of tumour-reactivity signatures. Also notable is that a high pre-treatment or early on-treatment abundance of CD8⁺ TILs expressing exhaustion markers such as PD-1, CD39 and/or CTLA4 is predictive of a clinical benefit from ICIs across tumour types^52,62–68. Thus, exhausted CD8⁺ T cells are enriched for tumour reactivity and their presence intratumourally prior to or early after ICI treatment is associated with improved outcomes.

T cell exhaustion is inevitable in the setting of persistent tumour.

One interpretation of the aforementioned findings is that exhaustion of CD8⁺ TILs is inevitable in the context of chronic antigen exposure. Consistent with the ‘decreasing potential’ model posited two decades ago⁶⁹, the picture that emerges is a linear diagram of progressive T cell exhaustion owing to successive stimulation with a persistent tumour antigen (FIG. 1a)^70,71. Importantly, this model does not preclude a role for other factors, such as resistance of target tumour cells to cytotoxicity⁷², hypoxia^73–75, immune contexture⁷⁶ or TCR signal quality and/or quantity, in modifying the tempo of exhaustion.

Fig. 1 |. Tumour-reactive CD8⁺ T cells undergo progressive exhaustion in the TME.

a | Naive CD8⁺ T cells (T_n) can differentiate along of spectrum of exhaustion states associated with progressively declining functional and proliferative capacity, spanning from stem-like precursor exhausted T cells (T_PEX) to terminally exhausted T cells (T_EX) that can ultimately undergo overstimulation-induced cell death. In the setting of persistent exposure to tumour antigen, expression of the transcription factor TCF1 in the T cell decreases while expression of TOX family transcription factors increases. As a result, expression of exhaustion markers (such PD-1, TIM-3, LAG-3 and CD39) will increase, effector function will decrease and the T cell will incur metabolic dysfunction. However, the tempo at which an individual T cell becomes exhausted is context-dependent (as indicated in the top right of the figure); the various T cell-intrinsic and T cell-extrinsic factors that modulate this process include the antigenic load, tissue oxygen concentration, immune contexture and target cell sensitivity to T cell-mediated cytotoxicity. b |Although the key mechanism of anti-PD-(L)1 antibody therapy was initially believed to be functional reinvigoration of exhausted T cells in the tumour microenvironment (TME), more recent data indicates that abrogation of the priming interaction between dendritic cells and precursor exhausted CD8⁺ T cells in tumour-draining lymph nodes is the crucial mechanistic target of such immune-checkpoint inhibitors.

A corollary of T cell exhaustion being inevitable in the setting of chronic antigen exposure is that the time of assessment in preclinical experiments and in clinical trials is crucial to the interpretation of exhaustion. This factor might partially explain the seeming paradox that evidence of CD8⁺ T cell exhaustion in pre-treatment and early on-treatment tumour samples predicts clinical benefit from ICIs^52,62–68 (implying that it is indicative of a pre-existing antitumour T cell response that can be enhanced by therapeutic intervention), whereas the same signal observed in late on-treatment and post-treatment readouts portends unfavourable outcomes^57,77 (given that exhaustion as an indicator of persistent cancer is still present despite therapeutic intervention).

One therapeutic implication emerging from this interpretation is that concomitant use of ICIs with orthogonal (that is, non-ICI) interventions that are cytotoxic to tumour cells might confer additive benefits. For example, ICIs combined with chemotherapy now constitute the standard-of-care front-line treatment for patients with advanced-stage lung cancers^78–82. When such combinations were initially being tested in clinical trials, the pervasive belief was that chemotherapy could be detrimental to the benefit of ICIs owing to the elimination of effector T cells, either owing to direct cytotoxicity or potentially indirectly through the immunosuppressive influence of corticosteroid pre-treatment to mitigate the adverse effects of chemotherapy. Although the benefits of chemotherapy and immunotherapy might not be synergistic⁸³, it is mechanistically intuitive that ICIs will have better activity in a setting in which chemotherapy has achieved some degree of tumour cytoreduction. Indeed, a high baseline tumour burden is associated with unfavourable response rates and overall survival in patients with metastatic melanoma or head and neck cancer receiving ICIs^84–86. Using PD-1 or CD39 as markers of exhausted intratumoural CD8⁺ T cells, various cytotoxic chemotherapy agents have been shown to reduce T cell exhaustion in preclinical tumour models^87,88. Similarly, patients receiving neoadjuvant chemotherapy have reductions in the abundance of exhausted intratumoural T cells^89,90. Hence, orthogonal oncological approaches to reduce tumour burden (such as chemotherapy, receptor tyrosine kinase-targeted therapies, radiotherapy and/or surgical resection) might improve the activity of ICI and CAR T cells by reducing the antigenic load, which drives T cell exhaustion. This effect might partially explain why ICIs have shown substantial antitumour activity when added to neoadjuvant chemotherapy in patients with early stage NSCLC, potentially secondary to the cytoreductive effect of chemotherapy^91,92. The common narrative is that T cell exhaustion is an important cause of poor tumour control, but perhaps it is equally true that T cell exhaustion is simultaneously a consequence of poor tumour control.

New insights on the mechanisms of ICIs

PD-L1⁺ cDCs in tumour-draining lymph nodes are a key target of ICIs.

If reversal of the exhaustion programme in tumour-reactive effector CD8⁺ T cells in the TME is the predominant therapeutic mechanism of ICIs, one would expect efficacy to be preserved in preclinical experiments in which tumour infiltration of new T cells is blocked. In multiple mouse models, however, blockade of T cell egress from secondary lymphoid organs using the sphingosine 1-phoshate receptor agonist FTY720 after tumour challenge but before treatment with anti-PD-(L)1 antibodies abrogates the therapeutic benefit^93–98. Given that blockade of lymphoid egress effectively eliminates recruitment of new CD8⁺ T cells into the tumour, these findings suggest that reinvigoration of T cells in the TME alone is insufficient to mediate the preclinical efficacy of anti-PD-(L)1 antibodies. This hypothesis raises the question of which site outside the TME is the key target for blockade of the PD-1 axis.

Profiling of immune cells in the tumour-draining lymph nodes (TDLNs) of mice bearing peritoneal tumours revealed both enrichment for tumour-reactive PD-1⁺ T cells and high levels of PD-L1 expression on myeloid cell populations⁹⁸. Notably, intrapleural infusion of anti-PD-L1 antibodies at a dose titrated to block PD-L1 on myeloid cells only in the mediastinal TDLN was sufficient to confer therapeutic benefit in this model⁹⁸. Moreover, the quantity of interactions between PD-1 and PD-L1 in TDLNs, as opposed to in the primary tumour, was found to predict benefit from ICIs in patients with metastatic melanoma⁹⁸. These findings were recapitulated in a separate study using mouse models of melanoma and breast cancer⁹⁹. In these models, intradermal injection of anti-PD-L1 antibodies resulted in preferential targeting of TDLNs and superior antitumour activity, as compared with typical intraperitoneal (systemic) administration⁹⁹. These results suggest that interactions between T cells and antigen-presenting cell in TDLNs, as opposed to solely in the primary tumour, are crucial for the efficacy of anti-PD-(L)1 antibodies, which implies that lymphoid tissues must harbour a population of PD-L1⁺ cells that can be perturbed by ICIs.

In the aforementioned study in mice bearing peritoneal tumours⁹⁸, PD-1⁺ T cells in TDLNs were found in close proximity to PD-L1⁺ conventional dendritic cells (cDCs), with evidence suggesting that blockade of PD-L1 on these cDCs results in the induction of precursor-exhausted T cells that migrate to the tumour. Moreover, genetic deletion of PD-L1 in Clec9a-expressing cDCs, but not LysM (lysozyme M)-expressing macrophages, has been shown to abrogate the efficacy of anti-PD-L1 antibodies in a PD-L1-deficient MC38 colon adenocarcinoma model¹⁰⁰. Similarly, genetic ablation of PD-L1 in cells expressing CD11c (predominantly found on DCs) impairs the growth of PD-L1-deficient MC38 tumours and abolishes the enhanced antitumour immunity conferred by PD-L1 inhibitors¹⁰¹. Furthermore, loss of PD-L1-inhibitor efficacy in Batf3-deficient mice, which lack cross-presenting cDC1, indicates that ICIs abrogate the co-inhibitory interactions between PD-L1⁺ cDCs and PD-1⁺CD8⁺ T cells, thereby enhancing T cell activation¹⁰¹. Together, these results indicate that cDCs in the TDLNs are a crucial target for blockade of the PD-1 axis.

ICI efficacy is dependent on TCF1⁺ precursor exhausted CD8⁺ T cells.

If the efficacy of anti-PD-(L)1 antibodies is dependent on cellular interactions in the TDLN, as opposed to the TME, another implication is that a lymphoid-resident tumour-reactive PD-1⁺CD8⁺ T cell population must be the key target of invigoration. Work from multiple groups has collectively demonstrated that a CXCR5⁺TCF1⁺PD-1⁺CD8⁺ T cell population residing in lymphoid tissues serves as the source of precursors for terminally exhausted CD8⁺ T cells in the setting of chronic viral infections^102–105. The self-renewal of these precursor exhausted CD8⁺ T cells was found to be dependent on expression of TCF7, which encodes the transcription factor TCF1^102–105. In both mouse and human tumours, infiltrating T cells undergo progressive exhaustion with concomitant loss of TCF1 expression owing to epigenetic silencing of the Tcf7/TCF7 locus (FIG. 1a)³⁰. CXCR5⁺TCF1⁺PD-1⁺CD8⁺ T cells have variously been called ‘progenitor exhausted’, ‘memory-like’, ‘stem-like’ and ‘follicular cytotoxic’, but ‘precursor exhausted’ is likely the most accurate term (reviewed in¹⁰⁶). Indeed, single-cell profiling of tumour-infiltrating CD8⁺ T cell from patients with melanoma or NSCLC revealed the presence of a partially exhausted population identified by expression of either TCF1⁷⁷ or CXCR5¹⁰⁷. Subsequent data from mouse models definitively demonstrated that TCF1⁺PD-1⁺CD8⁺ T cells (that also express SLAMF6, but not TIM-3) served as precursors for exhausted CD8⁺ T cells in tumours, and importantly, it was this precursor population that underwent a proliferative burst after blockade of the PD-1 axis^108–110. The crucial requirement of these precursor exhausted CD8⁺ T cells for the efficacy of ICIs has been demonstrated elegantly in multiple model systems, including through adoptive transfer experiments, depletion of the TCF1⁺ population using an inducible diphtheria toxin receptor-based ablation system (Tcf7^DTR-GFP), or deletion of Tcf7 under the control of the E8i CD8-Cre promoter (E8i-Cre⁺ Tcf7^fl/fl)^108–110. For example, following adoptive transfer of LCMV gp33-specific Tcf7^DTR-GFP CD8⁺ T cells (P14 T cells) into mice, selective depletion of Tcf7-expressing cells via diphtheria toxin administration reduced the total number of P14 T cells (including the abundance of the Tcf1⁻ subset) in gp33-expressing melanomas and impaired tumour control¹⁰⁸. Similarly, OT-1 CD8⁺ T cells (specific for an ovalbumin antigen) were found at a lower frequency in ovalbumin-expressing MC38 tumours from E8i-Cre⁺ Tcf7^fl/fl mice (which lack TCF1 expression in CD8⁺ T cells) compared with their wild-type counterparts, and this deficiency was associated with impaired antitumour immunity¹⁰⁹. Further heterogeneity within this TCF1⁺ precursor exhausted population is likely, with a CD62L⁺KIT⁻ subset emerging as the subpopulation with the greatest progenitor capacity and anti-PD-1 responsiveness in a mouse chronic viral infection model¹¹¹. Subsequent studies will further dissect the TCF1⁺ precursor exhausted population with the greatest progenitor capacity in the human tumour context.

The emergence of paired single-cell RNA and TCR sequencing data has confirmed that clonally-related CD8⁺ T cells can exist in the precursor exhausted and terminally differentiated states in human tumors¹¹². Importantly, in a cohort of 14 patients with metastatic melanoma who derived clinical benefit from nivolumab and ipilimumab, a high proportion of TCF1⁺ cells among PD-1⁺CD8⁺ T cells (above the median of 14.9%) in pre-treatment biopsies was associated with substantially longer progression-free survival (median 681 days versus 392 days)¹¹⁰. Similarly, another analysis of pre-treatment tumour samples from 24 patients with metastatic melanoma demonstrated a higher ratio of TCF1⁺ to TCF1⁻ CD8⁺ T cells in those who had a radiographical response following ICIs therapy⁷⁷. Hence, TCF1⁺PD-1⁺CD8⁺ precursor exhausted T cells are a crucial target of ICIs.

If blockade of the PD-1 axis acts primarily in secondary lymphoid organs, one would expect to find evidence of a proliferative burst after anti-PD-(L)1 antibody treatment with subsequent trafficking of the expanded precursor exhausted cells through the circulation to tumour sites. Indeed, an increase in the abundance of Ki67⁺PD-1⁺CD8⁺ T cells in the peripheral blood of patients with melanoma occurs 3 weeks (and as early as 1 week) after the initiation of anti-PD-1 therapy; at least a subset of these PD-1⁺CD8⁺ T cells were CXCR5⁺ and TCF1^+84,113. Notably, patients with metastatic melanoma with a high circulating Ki67⁺CD8⁺:tumour burden ratio (≥1.94) after one cycle of anti-PD-1 therapy had superior clinical outcomes⁸⁴. Similarly, patients with NSCLC who have an increase in the abundance of peripheral blood Ki67⁺CD8⁺ T cells within 4 weeks of initiating anti-PD-1 therapy are more likely to have a clinical response¹¹⁴. These observations dovetail with reports of extensive clonal replacement of intratumoural CD8⁺ T cell populations after ICI therapy^112,115,116, presumably from proliferating precursor exhausted T cells that originated in the TDLNs (FIG. 1b).

Implications of the new mechanistic insights on ICI efficacy.

Together, these findings challenge the commonly held view that the key target of PD-(L)1 inhibitors is contained exclusively within the TME. Of note, however, that these data are not wholly inconsistent with a direct effect of anti-PD-(L)1 antibodies on PD-1⁺CD8⁺ T cells in the TME. Indeed, CD8⁺ TILs isolated from patients with ovarian cancer retain ex vivo capacity for activation and cytotoxicity after blockade of the PD-1 axis¹¹⁷. Moreover, extensive single-cell RNA and TCR sequencing of longitudinal tumour samples from patients with NSCLC receiving ICIs in combination with chemotherapy revealed that, in addition to robust infiltration of new clonotypes, marked expansion of pre-existing CD8⁺ TIL clonotypes also occurs¹¹⁸. This dual source of antitumour T cells (that is, expansion of pre-existing TILs and clonal replacement by newly infiltrating T cells) was termed ‘clonal revival’, and it was theorized that different cancers have differential contributions from these two sources of tumour-reactive T cells¹¹⁸. We hypothesize that additional temporal and spatial heterogeneity in the contributions of these two T cell sources is likely, even within the same patient. For example, elegant studies using ex vivo tumour-fragment cultures have shown anti-PD-1 treatment can unleash a productive response by pre-existing CD8⁺ TILs that includes the release of cytotoxic markers (granzyme B, perforin, and granulysin), cytokines (IL-2, IFNγ, and TNF) and chemokine ligands (CXCL9, CXCL10, CCL5 and CXCL13)¹¹⁹, which might be crucial for recruitment of the circulating T cells that result in clonal replacement. Of note, this ex vivo response was preferentially observed in pre-treatment tissues obtained from patients who subsequently had a clinical response to ICIs. Consequently, in certain contexts, blockade of the PD-1 axis might modulate the activity of T cells directly in the TME.

Notwithstanding these results, the totality of the contemporary data suggests that the mechanistic ‘heavy lifting’ — at least in the context of ICI therapy — is accomplished by newly infiltrating CD8⁺ T cells. Thus, the picture that emerges is one in which the crucial target of anti-PD-(L)1 antibodies is the interaction between precursor exhausted PD-1⁺CD8⁺ T cells and PD-L1⁺ cDCs in TDLNs, which results in a proliferative burst of the precursor exhausted T cells that can be observed in the circulation and go on to differentiate towards terminally exhausted T cells in the TME (FIG. 1b). This paradigm broadens our understanding of why intratumoural expression of PD-L1 is an imperfect predictor of response to ICIs, and also explains the seeming paradox of why mice^94,120–122 and patients² with PD-L1-deficient tumours can still derive clinical benefit from such agents. Despite the methodological challenges of reliably measuring PD-L1 across pathologists, assays and disease types, NSCLC is one setting in which tumoural PD-L1 expression has consistently been associated with greater benefit from ICIs¹²³. This correlation might reflect a scenario in which PD-L1 expression is induced by IFNγ derived from infiltrating tumour-reactive CD8⁺ T cells. In this scenario, tumoural PD-L1 might be predictive not because it is the crucial mechanistic target of PD-1–PD-L1 blockade, but rather because PD-L1 expression is a ‘canary in the coal mine’ associated with the presence of tumour-reactive CD8⁺ T cells in the periphery that can be mobilized with ICIs². The pathways regulating PD-L1 expression are highly complex and can be influenced by genomic aberrations, transcriptional control, mRNA stability, oncogenic signalling and protein stability¹²⁴. Hence, contexts exist in which tumoural PD-L1 expression is uncoupled from an antitumour CD8⁺ T cell response; therefore, some tumours with high levels of PD-L1 expression might not respond to inhibition of the PD-1 axis.

At least on the surface, the crucial importance of a peripherally-derived CD8⁺ T cell population might not mechanistically align with the findings presented above, for example, that the abundance of intratumoural exhausted T cells is predictive of ICI response. To elaborate, why would intratumoural expression of PD-1 and CD39 on infiltrating T cells predict a response if the therapeutic target of PD-(L)1 inhibition is remote from the tumour? One potential explanation is that the abundance of intratumoural tumour-reactive PD-1⁺CD39⁺CD8⁺ T cells is proportional to the number of tumour-reactive T cell clones in the periphery that can be invigorated upon blockade of the PD-1 axis. Indeed, tumour-reactive PD-1⁺ T cell clonotypes detected in peripheral blood are often also present within TILs⁵¹. Moreover, a study in 30 patients with triple-negative breast cancer revealed a correlation between tissue and blood levels of PD-1⁺CD8⁺ T cells and CD39⁺CD8⁺ T cells¹²⁵. Conversely, the presence of few PD-1⁺ and CD39⁺ T cells in the tumour might indicate an overall paucity of CD8⁺ T cells in the patient that are capable of recognizing tumour antigen, and thus fewer tumour-reactive clones in the periphery that could be activated by ICIs.

Broadly, the new mechanistic insights discussed above have two implications for ICI therapy. First, caution should be exercised when the rationale for utilization of ICIs is upregulation of PD-L1 in the TME: certain cancer therapies might result in upregulation of PD-L1 on tumour cells, but this does not necessarily indicate that combination of these therapies with ICIs will yield greater clinical benefit. As stated above, this rationale might only be valid when PD-L1 expression is indicative of a pre-existing antitumour CD8⁺ T cell response, and not in contexts in which PD-L1 is upregulated by other tumour-intrinsic factors. Second, combinatorial ICI-based approaches should focus on candidate partner therapies that will amplify the total numbers of tumour-reactive T cells, rather than ‘un-exhausting’ terminally differentiated cells in the TME (which might provide limited therapeutic benefit given that these cells are in a fixed epigenetic state)^30–32. By extension, further attention should be focused on therapies that lead to increases in the clone sizes of existing tumour-reactive T cell clonotypes and/or recruitment of CD8⁺ T cells with additional tumour-reactive TCR specificities. Approaches that might achieve these goals include vaccination with tumour antigens¹²⁶, in situ vaccination via DC activation¹²⁷, agonism of co-stimulatory receptors^128,129, in vivo cytokine supplementation^130,131 and adoptive T cell therapies.

Evidence of exhaustion in CAR T cells

As described previously, disease relapse after CAR T cell therapy remains a major challenge, motivating a desire to understand CAR T cell exhaustion as a potential mechanism of tumour escape^39,41,132. In contrast to endogenous tumour-reactive T cells, CAR T cells are marked by their genetic modification, facilitating tracking and assessment of these tumour-targeted cells. Moreover, next-generation CAR T cell therapies can be further gene-edited to mitigate or interrupt exhaustion. These factors make CAR T cell therapy both a clinically relevant context and an important model in which to study T cell exhaustion.

Preclinical in vivo models have been crucial to understanding the CAR T cell exhaustion programme^4,133,134. In a mouse model of melanoma, tumour-infiltrating CAR T cells have a PD-1⁺TIM-3^high phenotype associated with increased expression of the canonical transcription factors of exhaustion, Tox and Tox2⁴, and Nr4a1, Nr4a2 and Nr4a3¹³³. Similarly, both chronic antigen-induced signalling¹³⁵ and constitutive, low-level, basal (or ‘tonic’) signalling¹³⁴ have been shown to drive CAR T cell exhaustion. These preclinical findings demonstrate that CAR T cells can have phenotypic, transcriptional and epigenetic profiles characteristic of exhaustion^136,137.

Clinically, analyses of on-treatment or post-relapse biopsy samples would be informative to determine the extent to which CAR T cell exhaustion contributes to loss of disease control in patients. In patients with B-ALL, CD19-targeted 4-1BB-based CAR T cells acquire exhaustion-associated DNA-methylation signatures, including methylation of effector genes and demethylation of TOX, by 4 weeks after infusion¹³⁸. Another study demonstrated persistent TIM-3⁺PD-1⁺ CAR T cells in the blood of patients with CLL several months after infusion, and these cells were enriched in the patients with inferior clinical outcomes¹³⁹. Notwithstanding the comprehensive profiling of circulating exhausted CAR T cells in these two studies, there is a relative dearth of data on CAR T cell exhaustion in the TME, which is typically where exhaustion is most prominent. At least one group could not detect CAR T cells in post-treatment lymphoma biopsy samples by immunohistochemistry or flow cytometry as soon as 10 days after infusion¹⁴⁰. A lack of persistence of the infused CAR T cells might limit the assessment of their exhaustion and explain the paucity of documented evidence of CAR T cell exhaustion intratumourally in patients.

Although the number of clinical reports characterizing post-infusion CAR T cells is low, the literature on in vitro CAR T cell phenotypes after manufacture and prior to infusion is comparatively robust. Products with higher proportions of less differentiated cells^141–144, such as CD45RA⁺CCR7⁺ CAR T cells¹⁴⁵, are associated with improved outcomes in a variety of haematological malignancies. In addition, CAR T cell products with higher levels of inhibitory receptors such as PD-1 and LAG-3 expand less following infusion and are associated with lower response rates in patients with chronic lymphocytic leukaemia (CLL)¹⁴¹, paediatric and young-adult CD19⁺ leukaemias¹⁴⁶, or LBCLs¹⁴⁷. Although T cells can show epigenetic hallmarks of exhaustion as early as 5–8 days post-stimulation¹⁴⁸ (and therefore during CAR T cell manufacturing), co-inhibitory receptors are also expressed on highly functional, activated T cells. These pre-infusion data thus emphasize two important points: (1) other cell states beside exhaustion, including effector and memory differentiation, can explain T cell function and persistence, and (2) T cell exhaustion can be difficult to evaluate based on co-inhibitory receptor expression alone.

Interrupting CAR T cell exhaustion

Pre-infusion and post-infusion CAR T cell phenotypes taken together with data from preclinical in vivo modelling suggest, but do not conclusively demonstrate, a role for CAR T cell exhaustion in limiting clinical responses. If CAR T cells are indeed impaired by exhaustion, one would expect that interrupting this hypofunctional cell state would improve therapeutic efficacy. Interruption of CAR T cell exhaustion programme could potentially be achieved through abrogation of the canonical TOX and NR4A transcription factors or antibody-mediated blockade of co-inhibitory receptors, such as PD-1 (FIG. 2).

Fig. 2 |. Overview of strategies to mitigate or interrupt CAR T cell exhaustion.

Downstream signalling induced by the chimeric antigen receptor (CAR) activates the transcription factor NFAT. Chronic NFAT signalling leads to expression of NR4A and TOX family transcription factors, which are crucial mediators of the epigenetic and transcriptional programme of T cell exhaustion. Cooperation between NR4A and TOX transcription factors results in the expression of genes encoding co-inhibitory receptors, such as PD-1, and also results in reduced cytokine production and effector function (not shown). Potential interventions to improve the function and thus therapeutic efficacy of CAR T cells are indicated in text boxes and are categorized as strategies that either mitigate T cell exhaustion (a) or more directly interrupt the exhaustion cascade downstream of NFAT signalling (b). a | Interventions to optimize effector-to-target ratios in the patient undergoing CAR T cell therapy, culture conditions during CAR T cell manufacturing, CAR signalling, transcriptional and/or epigenetic programmes and mitochondrial fitness might mitigate T cell exhaustion. b |More downstream interventions targeting the canonical transcription factors associated with T cell exhaustion (that is, NR4A and TOX) and co-inhibitory receptors are strategies to interrupt T cell exhaustion. DNR, dominant-negative receptor; ICIs, immune-checkpoint inhibitors; NAC, N-acetylcysteine; ROS, reactive oxygen species.

Numerous preclinical studies and at least 20 active clinical trials¹⁴⁹ have been focused on blocking or eliminating signalling through PD-1 to enhance CAR T cell function. Functional abrogation of PD-1 in CAR T cells can be achieved in a variety of ways, with the most straightforward method being co-treatment with anti-PD-(L)1 antibodies. In patients with disease relapsed after CD19-directed CAR T cell therapy, treatment with such antibodies leads to re-expansion of CAR T cells in the peripheral blood^150,151; however, clinical responses are infrequent, with one complete response and three partial responses in a cohort of 12 patients¹⁵². A larger, single-arm study in 28 patients anti-CD19 CAR T cells (axicabtagene ciloleucel) in combination with early PD-L1 blockade (atezolizumab initiated 1–21 days after cell infusion) for diffuse large B-cell lymphoma resulted in durable response rates similar to those of historical cohorts treated with CAR T cells alone, suggesting no added benefit from concomitant ICI therapy¹⁵³. Data from a phase I trial testing the combination of a mesothelin-targeting CAR T cell product and the anti-PD-1 antibody pembrolizumab in 18 patients with malignant pleural mesothelioma has shown promising efficacy (median and 1-year overall survival of 23.9 months and 83%, respectively)¹⁵⁴, although a subset of patients with mesothelioma will respond to single-agent anti-PD-1 therapy.

One aforementioned caveat of this approach is that blockade of the PD-1 axis is insufficient to epigenetically rewire the T cell exhaustion programme and, therefore, can only partially reverse the transcriptional phenotype of exhaustion^{30,31,155,156}. Hence, whether co-inhibitory receptor blockade following CAR T cell infusion would be most effective early (when it could block inhibitory signalling in less-differentiated effector cells) or late (after the onset of exhaustion) is not clear. Whether CAR T cells develop a precursor exhausted state — which is most responsive to ICIs in other contexts — also remains unclear.

Among the several other methods to suppress signalling by PD-1 or other co-inhibitory receptors (beyond co-treatment with ICIs), one approach is to engineer CAR T cells themselves to secrete antagonistic antibodies (FIG. 2)^157,158. Alternatively, overexpression of PD-1 dominant-negative receptors (DNRs)¹⁵⁹ or gene-editing of the PDCD1 locus in CAR T cells can eliminate PD-1 signalling intrinsically^160–163. Each approach has potential advantages and disadvantages. For example, in situ secretion of anti-PD-1 antibodies by CAR T cells might limit systemic antibody exposure and, therefore, toxicities. T cell-intrinsic gene-editing approaches can efficiently abrogate signalling in the edited cells, but do not potentiate endogenous T cell responses. Overexpression of a PD-1 DNR that lacks intracellular signalling domains can mitigate inhibitory signalling in the modified T cells by competing with endogenous PD-1 for binding to microenvironmental PD-L1. However, this effect of PD-1 DNR can theoretically be overcome by an excess of PD-L1 in the TME, whereas gene-edited T cells with a complete absence of PD-1 are insensitive to any level of PD-L1 expression. Each approach has been demonstrated to enhance the preclinical antitumour efficacy of CAR T cells across a variety of haematological and solid cancers.

More definitive conclusions will come from the multitude of ongoing clinical trials that are incorporating PD-1 abrogation in CAR T cell therapy. However, one early study in three patients who received CRISPR-Cas9-engineered T cells has yielded provocative results¹⁶⁴. In these patients, CRISPR–Cas9-engineered cells expressing an NY-ESO-1-specific TCR and harbouring deletions of PDCD1 as well as TRAC and TRBC (encoding the TCR α-chain and β-chain constant regions, respectively) were tracked longitudinally; T cells harbouring a PD-1 deletion were found to be less persistent than PD-1-non-deleted T cells¹⁶⁴. This finding might have alternate explanations unrelated to PD-1 biology, such as genotoxic stress from multiplex editing; however, it mirrors observations in mice in which Pdcd1-defiency and loss of inhibitory PD-1 signalling results in overstimulation-associated cell death of antigen-specific CD8⁺ T cells responding to a chronic viral infection^165,166. Therefore, expression of co-inhibitory receptors during the effector phase might serve to restrain T cell overstimulation, and further studies will be needed to clarify the optimal method and timing of PD-1 abrogation that preserves T cell persistence and function.

Mitigating CAR T cell exhaustion

Consistent with the aforementioned findings indicating that terminal exhaustion of T cells is epigenetically fixed^30–32 and the general agreement that these latter stages of exhaustion are therefore less amenable to therapeutic intervention⁷, interventions targeting earlier stages of the T cell differentiation pathway might be most effective in improving antitumour immunity. Rather than interrupting the T cell exhaustion programme, strategies to improve cell culture conditions and optimize CAR design and downstream signalling offer the potential to mitigate induction of this programme and thus enhance CAR T cell function (FIG. 2). Most of the strategies discussed below were developed for CAR T cell therapy, but many are also relevant for mitigating T cell exhaustion of TIL-based and TCR-engineered cell therapies (reviewed in¹⁶⁷).

Increasing CAR T cell to tumour cell ratios.

Exhaustion is not purely a T cell-intrinsic process, and can depend on tumour and patient characteristics. T cell exhaustion can be modulated by metabolic⁷³ and stromal¹⁶⁸ factors in the TME, which might differ between cancer histologies. Moreover, features of the malignant cells can dictate the efficacy of CAR T cell therapies. For example, impaired death receptor signalling in leukaemia cells can induce CAR T cell exhaustion⁷², and the efficacy of CAR T cells against solid tumours, in contrast with haematological malignancies, has been found to be dependent on IFNγ receptor signalling¹⁶⁹.

Beyond qualitative characteristics of the cancer cells, quantitative features might also determine the efficacy of CAR T cell therapy. Clinical benefit from CAR T cell therapy is inversely correlated with tumour burden across a variety of diseases, including ALL³⁷ and LBCL¹⁷⁰. One potential explanation for this finding is that higher antigen loads lead to sustained CAR stimulation and thus CAR T cell exhaustion, which further suggests the possibility of cytoreduction prior to CAR T cell treatment as a means of limiting exhaustion. This strategy can come in the form of surgery, radiotherapy, targeted therapies, immune checkpoint inhibitors, or chemotherapy. Notably, bridging chemotherapy might not always be effective¹⁷¹, given that many of the relapsed and/or refractory cancers historically treated with CAR T cells are chemo-refractory, but might be more effective as CAR T cells are used in earlier lines of therapy or novel agents are utilized for cytoreduction.

On the other side of the equation, one can consider increasing the numbers of CAR T cells infused. A retrospective study in 185 paediatric patients infused with CAR T cells found that event-free survival, relapse-free survival and overall survival were all increased with higher doses of tisagenlecleucel¹⁷². However, considering that prior treatment history and disease status can be confounders for CAR T cell yield, the question of whether higher CAR T cell doses would be beneficial needs to be addressed in a prospective clinical trial. Other approaches being studied that can potentially increase CAR T cell numbers in vivo include vaccination using the CAR-targeted antigen¹⁷³.

Optimizing T cell culture conditions.

Manufacturing of clinical-grade CAR T cell products requires a collection of optimized and standardized operating procedures¹⁷⁴. Substantial preclinical research has been focused on modifying this process to reduce exhaustion of CAR T cells. In a mouse xenograft model of B-ALL, CAR T cells that had been cultured for only 3 days as opposed to 9 days (7–14 days is standard) showed greater in vivo expansion and persistence as well as more robust tumour control, despite being infused at a six-fold lower dose¹⁷⁵. Data indicated that the more-advanced differentiation of the 9-day CAR T cells limited their proliferative capacity and effector function¹⁷⁵. The authors theorized that the 9-day CAR T cells were more exhausted¹⁷⁵, suggesting that shorter culture durations might mitigate T cell exhaustion. Changing the cytokine composition of the culture might also limit T cell exhaustion; CAR T cells cultured in the presence of IL-15, as opposed to standard IL-2 supplementation, have reduced expression of exhaustion markers, enhanced antiapoptotic properties and increased proliferative capacity¹⁷⁶. In addition, inhibition of the PI3K–AKT signalling pathway during CAR T cell manufacturing is associated with greater expansion, persistence and antitumour activity in mouse models^177–179. Mechanistically, dual PI3Kδ/γ inhibition using duvelisib results in CAR T cells with dose-dependent decreases in levels of the exhaustion marker TIM-3, greater expression of TCF1 and increased mitochondrial fitness¹⁷⁷. Hence, modifications in culture duration, cytokine supplementation and PI3K/AKT inhibition during CAR T cell manufacturing are promising avenues to test clinically with the aim of mitigating T cell exhaustion and thereby improving the efficacy of CAR T cell therapies.

CAR engineering to optimize signalling.

In endogenous T cells, exhaustion is driven by signals delivered upon repeated stimulation of the TCR¹⁸⁰. In an elegant preclinical study by Shakiba et al.¹⁸¹, three different levels of TCR signal strength were achieved in tumour-specific CD8⁺ T cells through amino acid modifications of the target tumour antigen. Notably, high signal strength was associated with T cell exhaustion, but neither low-signal-strength or high-signal-strength TCRs led to optimal tumour control in mouse models¹⁸¹. Instead, an intermediate-signal-strength TCR resulted in the ‘Goldilocks’¹⁸² level of signalling for optimal in vivo antitumour activity (FIG. 1a)¹⁸¹. T cell persistence has also been correlated with TCR affinity, and thus signal strength, in humans. For example, the persistence of four different clonotypes of adoptively transferred T cells that recognized oncogenic KRAS^G12D in a patient with metastatic colorectal cancer was inversely correlated with the 3D binding affinity between their TCRs and cognate neoantigen–HLA-C complexes^183,184. By analogy, the design and selection of CAR components can affect the signals that drive CAR T cell function, persistence and exhaustion.

The influence of CAR design on T cell exhaustion was first demonstrated for co-stimulatory domains. CARs designed with 4-1BB-based, compared to CD28-based, co-stimulation induce less T cell exhaustion¹⁸⁵, with a corresponding metabolic signature of increased respiratory capacity, fatty acid oxidation and mitochondrial biogenesis (as opposed to enhanced glycolysis in T cells with CD28-based CARs)¹⁸⁶. A phosphoproteomic analysis has demonstrated that both co-stimulatory domains induce similar signalling intermediates, although CD28-based CAR constructs activate faster and with greater potency¹⁸⁷. Further studies identified a CD28 signalling motif (YMNM) responsible for greater VAV1 and PLCγ1 activation and calcium influx, which drives NFAT signalling and thus expression of the canonical transcription factors of exhaustion¹⁸⁸. Modification of the asparagine in this motif to phenylalanine (YMFM) led to decreased exhaustion and enhanced persistence of the CAR T cells (FIG. 2)¹⁸⁸. In line with these findings, 4-1BB-based CAR T cells demonstrate greater persistence than their CD28-based counterparts in some¹⁸⁹ but not all¹⁹⁰ models. However, this enhanced persistence might come at a cost, given that CD19-targeted CAR T cells with a CD28-based domain are more sensitive to low levels of antigen^191,192. Reduction in antigen density has emerged as an important mechanism of immune escape from multiple engineered tumour-targeted agents^191,193,194. Nonetheless, studies in mouse tumour models have demonstrated similar overall efficacy of CD28-based and 4-1BB-based CAR T cell products¹⁹⁵. Furthermore, although direct comparisons are lacking, both co-stimulatory domains are in clinical use with similar efficacy^196,197, despite the longer detectable persistence of 4-1BB-based CAR T cells¹⁹⁸.

The strength of CAR signalling can also be modulated directly through editing of the TCR-derived CD3ζ signalling domain. In comparison with bead-based activation of the endogenous TCR, CAR activation induces more robust and rapid CD3ζ phosphorylation, and thus stronger signalling¹⁹⁹. Modulation of CD3ζ signalling through elimination of two of three immunoreceptor tyrosine-based activation motifs (ITAMs) from this domain in a CD28-based CAR construct (called 1XX) leads to reduced T cell exhaustion and better persistence and tumour control in mouse models (FIG. 2)²⁰⁰.

Regulation of CAR expression can further control and optimize signalling. CRISPR-Cas9 editing of TRAC can enable homology-directed site-specific insertion of the CAR gene construct into this locus, placing CAR expression under the control of the endogenous TRAC promoter and its associated regulatory elements²⁰¹. By contrast, standard retrovirally-transduced CARs integrate at semi-random sites within the genome, resulting in a broad range of expression levels (that is, variegation)²⁰¹. CAR expression from the TRAC is uniform and recapitulates a TCR-like behaviour, such as transient suppression of signalling after stimulation owing to internalization, degradation and re-expression of the CAR²⁰¹. TRAC-driven CAR expression engenders T cells with enhanced in vivo function and a less exhausted phenotype, resulting in greater efficacy compared with conventional CAR T cells in mouse models of B-ALL. Regulated, site-specific insertion of an engineered NY-ESO-1-specific TCR into T cells similarly leads to better tumour control, in association with lower expression of exhaustion markers²⁰².

Tonic, antigen-independent signalling is an alternative source of chronic stimulation that can lead to excessive T cell exhaustion, with certain CAR constructs having an intrinsic propensity to drive constitutive, low-level downstream signalling^185,203. This propensity for tonic signalling is determined by the CAR single-chain variable fragment, hinge and/or transmembrane domains and is thought to be mediated by CAR clustering on the cell membrane^185,204. Tonic signalling manifests as CD3ζ phosphorylation and overexpression of transcription factors implicated in T cell exhaustion even in the absence of target antigen¹³⁴. Notably, CAR constructs can be screened in vitro and in vivo for tonic signalling behaviour, enabling selection of constructs without this undesired behaviour^205–207. The highly effective CD19-directed CAR constructs currently used in clinic have a low propensity for tonic signalling¹⁸⁵, although some degree of tonic signalling might enhance responses in certain contexts²⁰⁸. Thus, a multitude of strategies can be used to engineer the CAR molecule and/or its expression to mitigate excessive signalling, reduce exhaustion and thereby improve the efficacy of CAR T cells.

Given that excessive signalling can result in CAR T cell exhaustion, multiple strategies to pharmacologically ‘turn off’ the signals have also been proposed. For example, the tyrosine-kinase inhibitor dasatinib has been identified as a pharmacological ‘on/off’ switch for CAR T cells owing to its ability to suppress CD3ζ phosphorylation^209,210. In various mouse xenograft models, transient inhibition of CAR signalling with dasatinib was able to restore antitumour functionality in exhausted CAR T cells²¹¹. The investigators concluded that dasatinib treatment enhances CAR T cell efficacy by either mitigating or reversing exhaustion²¹¹. Given that the epigenetic readouts of exhaustion were performed on bulk CAR T cell populations evaluated at different times after dasatinib treatment, whether prevention or bona fide reversal of epigenetic features of exhaustion is the dominant mechanism of transient signalling cessation remains unknown. Nonetheless, this study presents a potential pharmacological approach to mitigate CAR T cell exhaustion and improve the efficacy of a broad range of existing CAR T cell therapies.

Engineering transcriptional programmes.

The balance between T cell function and exhaustion occurs downstream of TCR signalling based on an equilibrium between the transcription factor AP-1 and its binding partner NFAT²¹². Upstream TCR²¹³ or CAR signalling¹⁸⁷, including tonic CAR signalling, drives calcium-induced calcineurin activation, which subsequently causes NFAT dephosphorylation and nuclear translocation. Canonical AP-1 is heterodimeric complex consisting of FOS and JUN and is induced both by primary TCR signalling via MAPK and co-stimulatory signalling via CD28²¹⁴. Cooperation between AP-1 with NFAT in T cells leads to robust IL-2 expression and effector function; however, expression of FOS and JUN is lost during sustained T cell signalling, leading to unopposed (or ‘partnerless’) NFAT signalling¹³⁶. Accordingly, T cell exhaustion caused by tonically signalling GD2-specific CARs can be overcome by forced expression of JUN, leading to enhanced CAR T cell efficacy in mouse models¹³⁴. Overexpression of the transcription factor BATF similarly counters NFAT-induced exhaustion, partnering with IRF4 to counter the transcriptional exhaustion programme and enhance CAR T cell function (FIG. 2)¹³⁵.

Unopposed NFAT signalling leads to expression of the exhaustion-associated T cell transcription factors TOX and NR4A, thereby enforcing the epigenetic and transcriptional programme associated with exhaustion^3,4,215. Indeed, murine CAR T cells with short-hairpin RNA (shRNA)-mediated knockdown of Tox are more persistent and have greater therapeutic activity than Tox-wild-type cells, particularly when also genetically deficient in Tox2⁴. Similarly, triple deletion of Nr4a1, Nr4a2 and Nr4a3 restored a functional and epigenetic programme that enhanced CAR T cell efficacy in mice bearing solid tumours¹³³. Intriguingly, other studies indicate that deletion of Tox results in increased apoptosis and impaired persistence of antigen-specific murine T cells in mice bearing tumours or with chronic viral infection^3,215–217. This discrepancy might be explained by differential effects of complete genetic deletion of Tox, as opposed to partial reduction in expression conferred by shRNA-mediated knockdown. [Au: In figure 2, this engineering strategy is shown as a way to interupt, rather than mitigate exhaustion, unlike the previously approach of disrupting ID3 and SOX4. Therefore, the inclusion of TOX/NR4A in part b of figure 2 seems a little inconsistent. Please see my margin comment in response to your previous one.] The transcription factors ID3 and SOX4 have also been shown to be essential for CAR T cell exhaustion, with CRISPR-mediated genetic deletion of either ID3 or SOX4 resulting in improved in vitro cytotoxicity against tumour cells after chronic tumour antigen exposure²¹⁸ (FIG. 2).

As described earlier, exhaustion is enforced in part by transcription factors via a canonical epigenetic programme. Indeed, the first clinical evidence that the disruption of this transcriptional programme could lead to dramatic changes in the functional behaviour of CAR T cells came from a case report of a CAR T cell clone with genetic loss of the epigenetic regulator TET2²¹⁹. TET2 catalyzes the oxidation of 5-methylcytosine in DNA (thereby promoting demethylation), and TET2-disruption occurring through random integration of the CAR vector in a patient with a pre-existing hypomorphic TET2 allele led to a CAR T cell clone that accounted for 94% of circulating T cells at its peak. This preferential expansion of the TET2-deficient CAR⁺ T cell clone coincided with development of a delayed cytokine-release syndrome and tumour regression many months after initial cell infusion²¹⁹. The TET2-disrupted CAR T cells were noted to have an epigenetic profile of reduced exhaustion²¹⁹, suggesting that abrogation of an epigenetic regulator could mitigate T cell exhaustion (FIG. 2). Interestingly, further preclinical work has demonstrated that the effect of TET2 disruption is dependent on CAR signalling properties (for example, occurring with 4–1BB-based but not CD28-based CARs), which has been attributed to inherent variability in the level of effector differentiation induced by each construct²²⁰. Moreover, TET2 disruption eventually uncoupled proliferative capacity and effector function, leading to a hyperproliferative but poorly functional CAR T cells, again demonstrating the complexity of interrupting the broad exhaustion programme²²⁰.

In 2022, an elegant genome-wide CRISPR screen revealed that the INO80 and BAF chromatin remodelling complexes are essential for T cell exhaustion²²¹. Importantly, functional inactivation of Arid1a, a component of the BAF complex, in adoptively transferred T cells resulted in epigenetic and transcriptional changes that impaired their exhaustion and improved their persistence and antitumour activity in mouse models²²¹. Preliminary data from a second study highlighted a similar role for the histone methyl transferase SUV39H1, which has previously been implicated in control of T cell stemness²²², with deletion of SUV39H1 resulting in enhanced anti-CD19 CAR T cell persistence and tumour control in a mouse model of B-ALL²²³. This result was associated with epigenetic changes that curtailed the expression of co-inhibitory receptors.

Mitigating metabolic exhaustion.

Persistent T cell stimulation is associated with substantial mitochondrial dysfunction^73,224–226. Moreover, metabolic and tumour environment conditions such as hypoxia promote T cell exhaustion⁷³ and trigger a dysfunctional metabolic programme driven the transcriptional repressor BLIMP1 (also known as PRDM1)²²⁷, leading to silencing of the transcriptional coactivator PGC1α. Indeed, overexpression of PGC1α in tumour-specific T cells results in enhanced mitochondrial fitness and tumour control in mouse models (FIG. 2)²²⁸.

Notably, impaired mitochondrial biogenesis has been associated with the reduced in vivo persistence of T cells transduced with CARs harbouring CD28, as opposed to 4–1BB, co-stimulatory domains¹⁸⁶. This mitochondrial dysfunction leads to increased generation of reactive oxygen species (ROS), which limits survival and self-renewal of the adoptively transferred CAR T cells^{224–226,229}; importantly, scavenging of ROS using N-acetylcysteine or nicotinamide could restore antitumour T cell responses (FIG. 2). Intriguingly, pre-infusion mitochondrial mass was found be higher in patients who subsequently had a complete response to autologous CD19-targeted CAR T cell therapy compared with those who did not²³⁰. Therefore, the exhaustion programme might have evolved as a mechanism to limit TCR overstimulation and the associated metabolic dysfunction, which if left unmitigated can result in T cell death and thus loss of persistence.

CAR T cell exhaustion in summary

Thus, a diversity of modifications to CAR T cells, from proximal to distal aspects of the signalling, transcriptional, epigenetic and metabolic programmes, have now been shown to potentially mitigate (and in some cases prevent) T cell exhaustion. We have attempted to provide a framework for understanding each of the approaches, and their respective strengths and weaknesses. Modern screening-based approaches will facilitate the identification of other novel candidates to engineer in order to mitigate CAR T cell exhaustion^231–233, and perhaps even enable comparisons between different approaches. More importantly, many promising engineering strategies are being translated into the clinical setting and the upcoming clinical studies will provide information on how to best enhance the therapeutic efficacy of CAR T cells.

Conclusions

The advances discussed in this Review highlight nuances in the concept of T cell exhaustion. Non-exhausted tumour-reactive T cells are undoubtedly preferable to exhausted tumour-reactive T cells for the treatment of patients with cancer; however, this is perhaps a false choice, given the evidence we have highlighted that T cell exhaustion is inevitable in the setting of persistent tumour antigen. Adding further complexity, interruption of the exhaustion programme can impair T cell persistence in certain contexts. Therefore, we advocate that the greatest hope to improve the efficacy of immunotherapy is to focus on strategies that mitigate induction of T cell exhaustion in the first place.

Key points.

• T cell exhaustion is a hypofunctional T cell state that is associated with decreased efficacy of immune-checkpoint inhibitors (ICIs) and adoptive T cell therapies.

• Molecular features of T cell exhaustion are associated with tumour-reactive T cell receptor (TCR) clonotypes and can predict clinical benefit from ICIs.

• The effectiveness of ICI therapy is dependent on peripheral expansion and subsequent infiltration of precursor-exhausted CD8⁺ T cells.

• The T cell exhaustion programme protects CD8⁺ T cells from overstimulation-associated cell death; therefore, interruption of this programme might impair the persistence of tumour-reactive T cells in patients with cancer.

• Engineering optimized receptors and downstream signalling, fine-tuning transcriptional and epigenetic states, and overcoming metabolic dysfunction can mitigate T cell exhaustion (rather than interrupting this programme per se), thereby enhancing the efficacy of chimeric antigen receptor (CAR) and TCR-engineered cell therapies.