Rethinking peripheral T cell tolerance: checkpoints across a T cell’s journey

Quiescence

Quiescence is a peripheral tolerance mechanism that can limit the responsiveness of naive T cells to tonic signals. When naive T cells exit the thymus, the quiescence programme, regulated by quiescence mediators, maintains these cells at a lower basal metabolic state than their thymocyte counterparts⁹. Quiescence is an active process that maintains T cells in the G0 stage of the cell cycle, sustains a small cell size, ensures low cellular metabolism and prevents the development of effector functions in response to tonic signals, including self-antigens and other mediators^10–12. It is an important mechanism to prevent the expansion of the numbers of naive T cells in the steady state and thereby maintain their relative numbers. The mechanisms controlling quiescence also set the threshold for activation of naive T cells. As such, quiescence plays a critical role in clonal selection, ensuring that most of the repertoire that binds a particular antigen is restrained and limiting clonal expansion to high-affinity clones. However, it is unclear whether the threshold for quiescence is fixed or exists within a dynamic spectrum. It is possible that the mediators of quiescence can be downregulated to allow naive T cell responses to lower-affinity antigens or upregulated to restrict responses more stringently. Given that the signalling molecules and networks that control T cell quiescence are dynamic and respond to outside cues, one may surmise that the threshold for quiescence changes in different environmental contexts and that these changes determine the set point for a productive response to an antigen, the development of autoimmunity or the response to neoantigens in cancer.

In mice, many of the molecular effectors and networks involved in T cell quiescence have been identified and have been shown to control a diverse array of biological pathways. TGFβ1 and TOB1, two effectors that are part of the transforming growth factor-β superfamily signalling network, have been identified as quiescence mediators^13–15. They signal via the SMAD family of signal transducers and downregulate the production of IL-2 and T cell activation in general. TOB1 is a member of the APRO family of genes, which encodes proteins with antiproliferative functions and also includes BTG1 and BTG2 (REF.¹⁶). Recently, it was shown that all these genes are selectively and highly expressed in naive T cells and that BTG1 and BGT2 play an important role in promoting the deadenylation and degradation of mRNA in naive T cells, thereby actively maintaining a low rate of translation^17,18.

Transcriptional and proteomic analyses of the naive T cell repertoire have also revealed a high expression of the DNA-binding protein KLF2 by quiescent T cells, qualifying it as a marker for T cell quiescence^17,19. KLF2 impacts on T cell quiescence partly through the suppression of the transcription factor MYC and by affecting cell cycle progression through p21 (REFS^20–23). KLF2 also appears to regulate thymocyte egress and the expression of S1PR1 and CD62L, two receptors that are highly expressed by naive T cells^24,25. Whether this lymphocyte recirculation regulatory mechanism by KLF2 plays an important role in the maintenance of quiescence remains to be determined. In T cells, the expression of KLF2 is induced and maintained by the transcription factor FOXO1, which has a well-established role in maintaining T cell quiescence^26–28. Indeed, the expression of FOXO1 and KLF2, as well as the expression of the transcription factor FOXP1, are all directly upregulated by the transcription factor RUNX1, which also has an important regulatory role in maintaining T cell quiescence, as evidenced by the observation that loss of RUNX1 leads to T cell hyperactivation^29,30. As mentioned above, part of the quiescence phenotype is a low basal metabolic programme. The tumour suppressor proteins TSC1 and TSC2 play an important role in this process by suppressing the activation of mechanistic target of rapamycin complex 1 (mTORC1), an important nutrient, energy and redox sensor^31,32. This results in a downregulation of cellular metabolism, thereby maintaining low nutrient demands by the large number of peripheral naive T cells.

Insights into how these interconnected, multidimensional networks of quiescence regulators control the threshold of responsiveness of naive T cells in health and disease are evolving³³. For all documented quiescence regulators (TABLE 1), their genetic deletion results in an increased percentage of T cells with a memory phenotype, with enhanced activation and pro-inflammatory polarization states^17,22,27,32. Moreover, there is ample evidence that alterations in the expression of quiescence regulators result in a breakdown in peripheral tolerance and the development of autoimmune disease^28,30,34. For example, several studies have shown that abrogation of TGFβ1 signalling in T cells precipitates aggressive and fatal autoimmune disease^35,36, likely due to disruption of the suppressive activities of SMAD and TOB1 on T cell activation. In addition, Foxo1 deletion in T cells in mice results in the development of profound colitis, pancreatitis and multi-organ lymphocyte infiltration; however, this may be partly caused by T_reg cell dysfunction in addition to lowering of the threshold of naive T cell quiescence. Mice with Runx1^−/− T cells experience a cytokine-release syndrome and fatal pneumonitis³⁰. Collectively, these findings suggest that a breakdown of T cell quiescence lowers the threshold of naive T cell responsiveness to self-antigens, resulting in profound autoimmunity.

Recent studies have established that the naive T cell repertoire in mice (defined by CD62L^hi and CD44^low) is not uniform or monomorphic but displays remarkable steady-state heterogeneity with multiple cellular subsets¹⁷. The role of specific networks of quiescence mediators in controlling the steady-state distribution of these subsets is not currently understood. Whereas most naive T cells exhibit a quiescent phenotype and coexpress high levels of genes encoding quiescence regulators, there are several subsets within the naive T cell compartment¹⁷, including T cells with high levels of expression of genes encoding the type I interferon response module, others with enhanced early T cell receptor (TCR) signalling activity and subsets with augmented expression of genes encoding cytoskeleton components¹⁷. We argue that the current phenotypic definition of naive T cells needs to be expanded beyond the use of conventional markers such as CD44, CD62L and CD45RB to those that are more relevant to quiescence such as KLF2 and TOB1 to more appropriately capture the heterogeneous landscape within the naive repertoire.

Importantly, it appears that the expression of the inhibitory checkpoint receptor VISTA is critical for the intrinsic, steady-state maintenance of these heterogeneous naive T cell subsets and for the maintenance of quiescence^17,37. It was shown that VISTA^−/− T cells (VISTA is also known as VSIR) express reduced levels of important quiescence mediators such as KLF2, BTG1 and BTG2 (REF.¹⁷). Further analysis revealed that the expression of VISTA and KLF2 is co-regulated in T cells and suggests a direct relationship between VISTA expression and the maintenance of quiescence networks. The loss of key quiescence regulators in VISTA^−/− T cells results in hyper-responsiveness to co-stimulation-deficient T cell activation and precipitates several autoimmune manifestations, including experimental autoimmume encephalomyelitis, splenomegaly and enhanced infiltration of activated T cells into non-lymphoid tissues in mice^38–40. These findings link tonic VISTA signalling, elements of the quiescence networks and the prevention of autoimmunity.

At present, the steady-state heterogeneity of the human naive T cell compartment is not well explored⁴¹. There are indications that some factors that are critical for maintaining mouse T cell quiescence and their naive phenotype are functionally conserved across mouse and human T cells. For example, it has been reported that TOB1 and KLF2 are quiescence regulators that are common to both mouse and human T cells^42,43. A recent study showed that the same transcription factors implicated in naive T cell quiescence in mice (FOXO1, FOXP1 and KLF2) are also highly expressed in human naive T cells⁴⁴. These transcription factors have a constitutively rapid turnover, enabling their depletion on T cell activation to facilitate the transition from a quiescence state to effector T cell (T_eff cell) differentiation⁴⁴. Although we map quiescence as a tolerance checkpoint to naive T cells, there are also indications that memory T cells exhibit quiescence features (discussed in BOX 1).

Ignorance

Ignorance is another tolerance checkpoint at the naive T cell stage that has been observed in multiple systems, yet its mechanisms remain poorly understood⁴⁵. Simply put, self-reactive T cells can fail to activate and provoke autoimmune disease despite the presence of the specific self-antigen. These T cells remain in a naive, responsive state. Mechanisms that control ignorance can include intrinsic and extrinsic mechanisms. Intrinsic mechanisms include TCR affinity for an antigen where TCR affinity is too low to elicit a T cell response. Extrinsic mechanisms likely comprise lack of T cell stimulation owing to low antigen density and/or the restriction of antigen recognition owing to its anatomical location⁴⁶. This was demonstrated in two recent studies using elegant mouse models of neo-self-antigen expression restricted to specific organs based on tissue-specific promoters, which allowed the tracking of the response of antigen-specific endogenous T cells in vivo^47,48. In both cases, the investigators identified ‘ignorance’ of a self-antigen as the main mechanism curtailing reactivity and immunopathology to an antigen expressed in the pancreas but not in the lung or intestines, where thymically derived T_reg cells instead emerged^47,48. They concluded that the anatomical location and abundance of the self-antigen are the deciding factors, as low antigen dose in the pancreas maintained ignorance by not provoking T cell priming, whereas high doses elsewhere resulted in deletional peripheral tolerance⁴⁹.

A significant observation from these studies is that self-reactive ignorant T cells do exist in the periphery of these mice^47,49–51. Similarly, ignorant self-reactive T cells exist in human peripheral blood and can be as frequent in healthy individuals as in patients with autoimmune disease. The difference is that self-reactive T cells in healthy individuals are maintained in a naive state, whereas these T cells have an activated phenotype in individuals with autoimmune conditions^52–54. In light of the substantial steady-state heterogeneity now known to exist within the naive T cell compartment, a reassessment of T cell ignorance in humans and mice at much higher resolution is warranted. Studies are currently under way to evaluate whether naive, ignorant self-reactive T cells may differ with regard to their quiescence mediators, as highly resolved phenotypic analyses (for example, single-cell RNA sequencing) have yet to be reported.

A noteworthy distinction between quiescence and ignorance is that quiescence is a general tolerance hallmark of all naive T cells, irrespective of their antigen specificity, whereas ‘ignorance’ refers to the avoidance of activation of specific self-reactive T cells, with the host tissue and the location of the self-antigen being determining factors. Much more is understood about the regulators of naive T cell quiescence than the molecular regulators of T cell ignorance. It is currently unknown whether some of these regulators overlap.

Like quiescent T cells, and unlike anergic T cells as discussed later, ‘ignorant T cells’ are not dysfunctional, as when appropriately stimulated in an inflammatory context, such as with viral pathogens^50,51 or inflammatory cytokines^55,56, they are capable of overriding the ignorance checkpoint and inducing autoimmune disease^57,58.

One of the unanswered questions regarding T cell ignorance is whether there is an evolutionary benefit to imperfect central tolerance that results in a significant number of ignorant self-reactive T cells reaching the periphery. Mouse models in which central tolerance is titrated to eliminate all self-reactive clones may be of value here. Despite the lack of empirical investigation, we speculate that these low-avidity self-reactive T cells are likely reactive and may have high specificity for some pathogens and that their potential protective benefit in an infection setting may outweigh the risk they pose with regard to autoimmunity.

Anergy

The mechanisms of quiescence, ignorance and anergy all serve to limit the responses of naive T cells to an antigen. However, quiescence is constitutively maintained in a manner that is agnostic to TCR stimulation, ignorance is a result of the antigen being hidden away or presented at extremely low levels, whereas anergy is a direct result of ‘defective (imbalanced)’ TCR stimulation in naive T cells. Anergy is the most proximal, non-deletional tolerance mechanism following TCR engagement and is functionally defined as T cell hyporesponsiveness to restimulation under robust stimulatory conditions. Functionally, it serves as an early checkpoint during T cell priming to prevent potential T cell pathogenicity before the onset of the T cell effector stage.

At the molecular level, anergy is induced by co-stimulation-deficient (‘tolerogenic’) TCR activation. In T cells that receive TCR signalling with productive co-stimulation (for example, CD28), this induces the transcription factor NFAT1 together with AP-1 (a dimeric transcription factor composed of FOS and JUN family subunits) to induce T cell differentiation and effector functions. By contrast, tolerogenic TCR activation leads to a defect in RAS–mitogen-activated protein kinase (MAPK) signalling^59–62, which in turn impairs translocation of AP-1 into the nucleus⁶³. In this instance, TCR signalling is now shifted towards the transcription factor NFAT1 in the absence of AP-1 activity. This imbalanced shift in downstream TCR signalling towards NFAT1 activation results in the induction of several genes that encode proteins involved in establishing the anergic state⁶⁴, such as diacylglycerol kinase-α (DGKα)⁶⁴, an enzyme critical for anergy induction through inhibition of RAS activation via depletion of diacylglycerol^62,65. Immediate transcriptional targets of NFAT1 also include the transcription factors EGR2 and EGR3, which suppress IL-2 transcription and upregulate the regulatory ubiquitin ligase CBLB^66–68.

The long-term TCR-induced hyporesponsive state observed in anergic T cells shows substantial evidence of coordinated epigenetic^69,70 and post-transcriptional⁷¹ programming that silences effector cytokine expression. Recent work identified the TCR-induced gene Nr4a1 (also known as Nur77), which encodes a nuclear receptor that acts as an inhibitor of AP-1 function, to mediate the epigenetic reprogramming of mouse T cells towards an anergic state⁷². Hallmark functional changes that define anergy include a profound reduction in the levels of IL-2, interferon-γ and tumour necrosis factor (TNF) in response to TCR stimulation. This acquired refractory state is associated with growth arrest and defects in cell cycle progression. Although anergy can be long-lasting, it is reversible, and in vivo studies show that anergic T cells slowly recover functional responsiveness in the absence of the antigen, indicating that the maintenance of anergy requires prolonged antigen exposure^73–75. The reversal of anergy is also observed on adoptive transfer of anergic cells into a lymphopenic environment. In this case, the absence of a cognate antigen and the abundance of homeostatic cytokines result in the establishment of an effector cell state^76,77.

It appears that TCR stimulation of quiescent T cells results in the downregulation of factors that enforce quiescence, whereas TCR stimulation of anergic T cells upregulates mediators of anergy. Transcriptional profiles of naturally anergic CD4⁺ T cells (defined as CD44^hiCD73^hiFR4^hiFOXP3⁻)⁷⁸ revealed that, following TCR engagement, most quiescence regulators, with the exception of TOB1 (REF.¹⁵), are downregulated, whereas several factors involved in the acquisition of the anergic state, including MAF, NFAT1, NRP1 and NR4A1, are upregulated^17,79,80. One key similarity between quiescent and anergic T cells is that mTORC1 activity is suppressed in both states. The metabolic consequences of mTORC1 suppression include reduced protein synthesis, as well as nutrient acquisition via suppressed expression of amino acid and glucose transporters, all key features for the energy charge necessary for the acquisition of T cell effector functions^81,82. In anergic T cells, the absence of co-stimulation prevents the full mobilization and upregulation of the metabolic machinery, even by subsequent productive stimuli^82,83. However, a striking difference between quiescent and anergic T cells is that the former are maintained in the non-proliferative G0 stage of the cell cycle, whereas the induction of anergy involves early proliferation, which is followed by an arrest of cell cycle progression in the G1 to S phase⁸⁴. Thus, there are substantial differences but also similarities in the tolerance mechanisms of naive T cells before and after T cell activation.

Distinctive markers for anergy across T cell subsets and across species have yet to be clearly identified, and most studies resort to an operational definition (a state of tolerance induced by defective TCR stimulation) to classify T cells as anergic. In mice, recent studies defined anergic CD4⁺ T cells by their higher expression of the 5′ nucleotidase CD73 and the folate receptor (FR4) than T_eff cells, and they are distinguished from T_reg cells by their lack of FOXP3 expression^77,78. Whether CD73 and FR4 are functionally involved in the induction or maintenance of anergy, and whether these are exclusive markers of CD4⁺ anergic cells, remains to be determined. In several mouse tumour models, anergic CD8⁺ T cells were found to express the inhibitory receptors LAG3 and 4–1BB, which are both regulated by the transcription factor EGR2 (REF.⁸⁵). It is currently unknown whether human anergic T cells display similar molecular markers. Human anergic CD8⁺ T cells have been defined as expressing both the co-inhibitory receptor CTLA4 and the chemokine receptor CCR7, which are not co-expressed by activated or naive T cells. Surprisingly, these studies reported no differences in the expression of the inhibitory receptor PD1 or the anergy-related genes EGR2, GRAIL (also known as RNF128) and CBLB⁸⁶. Efforts are ongoing to establish the common phenotypes that define anergic T cells in both mouse and human T cell subsets and to examine the role of anergy in anticancer immune response (BOX 2).

Exhaustion

T cell exhaustion represents the predominant non-deletional tolerance mechanism at the T cell effector stage. The term ‘T cell exhaustion’ was coined by viral immunologists who described desensitized T cells in the context of chronic viral infections^87,88. During an acute immune response such as the response to an acute infection, functional central memory T cell (T_CM cell) and effector memory T cell (T_EM cell) subsets arise from T_eff cells. By contrast, when antigen stimulation persists, such as during chronic infection and in certain types of cancer, memory T cells can fail to develop properly^89–91, and functionally compromised, persistently ‘exhausted’ T cells (T_ex cells) dominate the antigen-specific repertoire. Compared with fully functional primed T_eff cells, T_ex cells display reduced but not absent responses to antigens at multiple levels. Characteristic features of T_ex cells include (1) decreased cytokine production, (2) sustained, high levels of inhibitory receptor expression, (3) altered epigenetic, metabolic and transcriptional states and, importantly, (4) an inability to transition to the quiescence-like cell state observed in memory T cells (BOX 1). However, recent findings suggest that there may be differences in the T_ex cell state in chronic viral infections and in cancer (BOX 3).

Initial studies of the T_ex cell phenotype found that these cells express multiple inhibitory receptors, including PD1, LAG3, TIGIT, CD38, CD39 and TIM3 (REF.⁹²). Indeed, subsequent work defined a stage of T cell exhaustion based on these markers⁹³. However, we now know that these receptors are not exclusive features of T_ex cells as (1) highly functional T_eff cells can also express inhibitory receptors and mediators of exhaustion such as TOX^94–96, (2) human T cells express some of these markers in the steady state and (3) their expression dynamically varies according to localization and differentiation state^97,98. Another challenge is the difficulty in distinguishing T_ex and anergic T cells on the basis of surface markers as these two states of T cell dysfunction overlap with regard to the expression of several of the inhibitory receptors. TCR stimulation is the central driver of T cell hyporesponsiveness and the resultant dysfunctional state for both T_ex cells and anergic T cells. However, the key distinction between anergy and exhaustion is that anergic T cells arise early after T cell activation and priming, whereas T_ex cells arise from T_eff cells that have undergone productive activation, at the effector (memory-precursor) stage⁹⁹. Moreover, the natures of the signals that induce these cell stages differ. Anergy is the product of co-stimulation-deficient T cell activation, whereas exhaustion occurs owing to chronic TCR stimulation in the presence of appropriate co-stimulatory signals.

On the molecular level, both anergic T cells and T_ex cells express NFAT1 as an important driver of tolerance. However, the gene expression profile typical for T_ex cells also appears to be determined by a complex transcription factor profile that includes reduced expression of TCF7 and increased expression of TOX, NR4A, BATF, IRF4, BLIMP1 and other transcription factors^100,101. High-throughput transcriptional and epigenetic analyses identified several transcription factors induced by TCR stimulation that are involved in the induction of the T_ex cell state and in allowing the T_ex cells to survive beyond the T_eff cell contraction phase. These transcription factors include NFAT, IRF4, BATF¹⁰², NR4A1 (REF.¹⁰³) and TOX^103–106. However, an important observation from these studies is the identification of a ‘progenitor’ or ‘precursor’ population of predysfunctional T_ex cells that have self-renewal (stem cell-like) properties and re-expansion potential and are defined by the expression of the transcription factor TCF1 (also known as TCF7)^107–109. Of note, this cell state is still inferior in effector function in comparison with fully functional T_eff cells but is likely responsive to checkpoint blockade¹⁰⁸. With prolonged activation, these progenitor T_ex cells ultimately give rise to the terminally differentiated dysfunctional subset called ‘terminal T_ex cells’.

Some consensus and controversies exist in defining the mediators and markers of exhaustion in mouse and human T cells. In both mice and humans, TCF7 has emerged as a transcription factor critical for defining the progenitor subset of self-renewing T cells that can re-expand even in settings of chronic infection or cancer¹¹⁰. A T_ex cell immune signature predicted favourable prognosis in multiple human autoimmune and inflammatory diseases¹¹¹. In transplantation, kidney transplantation from a CMV-positive donor into a CMV-negative recipient leads to an T_ex cell state that may play a role in graft tolerance¹¹². This is supported by mouse models where a T_ex cell state is correlated with reduced graft rejection of heart and liver transplants^113–115. Despite these similarities, there remains a great deal of mechanistic investigation to understand human T cell exhaustion.

Our understanding of T cell tolerance suggests that T cell exhaustion has several evolutionary benefits. From a systems perspective, T_ex cells persist beyond the T_eff cell lifespan and contribute to the containment of chronic viral infections. Second, exhaustion permits pathogen-specific T cells to curtail their activity to avoid persistent inflammation, tissue damage and chronic autoreactivity. This is important for host survival, as failure in the induction of exhaustion^113–115 or reinvigoration of T_ex cells in some viral models can induce immunopathology^116,117.

Senescence

Senescence is defined as a growth and proliferative arrest stage that is induced when “cells reach the end of their replicative potential or are exposed to various stressors”¹¹⁸. As discussed already, chronic antigen stimulation can cause the functional exhaustion of T_eff cells. However, repeated TCR stimulation or lymphocyte ageing can also induce telomere erosion and/or irreparable DNA damage, leading to a loss of the T cell replicative capacity on further antigen encounters, which occurs at the effector or memory stage of T cell differentiation. Telomere-dependent senescence occurs during extensive replication and can be the result of repeated clonal expansion and antigen recall (where pre-existing memory T cells respond to previously encountered antigens). However, it is now known that there are factors that can inhibit telomerase and accelerate senescence, such as interferon-α, which directly inhibits telomerase activity¹¹⁹. On the other hand, telomere-independent senescence can be induced by DNA-damaging agents such as reactive oxygen species (ROS) and ionizing radiation or by activation of the p53 pathway or other stress pathways in response to growth factor deprivation¹²⁰. Strong TCR stimulation and oxidative phosphorylation as a result of high-affinity antigen encounter can enhance the production of ROS^121,122, which can affect T cell function and induce cell death¹²³. However, little is known about the role of telomere-independent senescence in T cell tolerance.

Studies on primary human T cells identified senescent T cells as poorly proliferative highly differentiated T cells that display the surface markers CD45RA, KLRG1 and CD57 but do not express the co-stimulatory receptors CD27 and CD28 (REF.¹²⁴). Whereas CD27⁺CD28⁺ T cells have long telomeres, senescent T cells with the shortest telomeres lose expression of CD27 and CD28 but re-express CD45RA^124–126. Unlike T_ex cells, senescent T cells are not compromised with regard to effector function but acquire a senescence-associated secretory phenotype that is characterized by the production of high levels of pro-inflammatory and suppressive cytokines, despite their proliferative block^124,125,127. Senescent T cells have constitutively activated p38 MAPK activity, and the inhibition of this activity can reconstitute proliferation and telomerase activity¹²⁸. However, this constitutive p38 activity is not mediated by the canonical pathway of p38 activation through upstream MAPK, or the alternative pathway of p38 activation induced by the TCR, but rather is mediated through the intracellular metabolic sensor AMPK, which triggers p38 autophosphorylation via the scaffold protein TAB1 (REF.¹²⁹). These findings suggest that senescence is actively maintained in T cells, as has been shown for other cell types. They also implicate DNA damage and nutrient sensing in the induction of T cell senescence.

Senescent T cells markedly accumulate with ageing, during chronic viral infections, in individuals with autoimmune disorders and in individuals with particular types of cancer^{125,130–132}. From an evolutionary standpoint, senescence may protect against T cell lymphoma development by preventing the excessive proliferation of T cells with damaged DNA. Another benefit may be the local control of excessive inflammation during chronic autoimmunity or infections. It is unclear whether T cell senescence has any benefit during ageing, where instead it appears detrimental to T cell responses and productive immunity. Recently, a study showed that senescent CD8⁺ T cells lost TCR signalling in aged individuals and were reprogrammed to acquire an innate-like cytotoxic activity via the upregulation of the natural killer cell receptor NKG2D pathway. This CD8⁺ T cell to natural killer cell-like transformation was mediated via the stress-sensing protein Sestrin 2 (REF.¹³³). However, T cell senescence can also lead to a loss of immune functions. For example, reports show a loss of memory T cell proliferative responses to antigens in senescent T cells, and the loss of clonal expansion in response to antigenic rechallenge¹³⁴. These observations are important in the setting of vaccination of elderly people, where senescence may hamper the clonal expansion of T cells necessary for primary immune responses^134–136.

One major challenge in advancing our understanding of T cell senescence is the almost complete absence of studies of T cell senescence in mouse models. This limits our knowledge of how reversible senescence is and whether therapeutic interventions to reverse senescence in T cells can enhance T cell memory and immunity. The senescence tolerance checkpoint may well become increasingly relevant as human life expectancy is increasing¹³⁷, given that ageing individuals develop enhanced susceptibility to infections to which they were previously immune^134,138. Immunity clearly diminishes with age, and a major factor in this decline is compromised T cell function^139–141. Therefore, more mechanistic in vivo studies in multiple settings are needed to qualify and illuminate the relevance and setting of this tolerance checkpoint.

Peripheral deletional tolerance (death)

Multiple programmed cell death checkpoints exist at several stages of the T cell’s journey. However, the signalling cues that induce death in T cells vary according to stage of differentiation. Overall, deletional tolerance serves a central role in pruning the repertoire of peripheral T cells as well as in terminating the immune response after a productive immune challenge (such as during the contraction phase). However, unlike the other cell-intrinsic mechanisms of tolerance discussed in previous sections, this mechanism irrevocably eliminates clones from the repertoire. Although the role of clonal deletion in central tolerance is well studied, the role of T cell death in maintaining peripheral tolerance has been largely overlooked.

Co-stimulation-deficient T cell activation by either an antigen or a superantigen elicits naive T cell activation, limited proliferation anergy and/or death^73,142. The precise conditions that result in the induction of anergy versus death under co-stimulation-deficient conditions are yet to be resolved. However, after the initial phase of T cell activation and proliferation in response to a tolerizing stimulus (co-stimulation-deficient antigen), a large proportion of T cells are rapidly lost by apoptosis, and the surviving minor population develop an anergic profile^73,143,144. It has been presumed that the sole engagement of the TCR (signal 1) in the absence of co-stimulation (signal 2) governs the induction of cell death. However, recent studies have shown that genetic deficiency of VISTA on resting T cells spares T cells from TCR-induced death but does not affect the induction of anergy, whereas triggering through VISTA enhances tolerogen-induced death¹⁷. Therefore, there may be factors, in addition to TCR engagement by antigens, that regulate the magnitude of tolerogen-induced T cell death in the absence of co-stimulation.

A series of elegant studies in mouse models have defined the molecular aspects of death induced by a tolerogen of naive T cells. These studies demonstrated that cell death was mediated via the intrinsic proapoptotic family member BIM^145,146, which is a distinct pathway from the extrinsic pathway of apoptosis mediated by death receptors such as FAS. Of importance, these findings support earlier work highlighting how co-stimulation via CD28 or γC cytokines (IL-7 and IL-4) augments T cell survival after activation via the upregulation of the antiapoptotic factors BCL-X_L (REFS^147–149) and BCL-2, respectively^150,151. Highly resolved transcriptional profiling of mouse T cells under tolerizing versus immunizing conditions revealed a unique molecular signature of cells before apoptosis. This included the upregulation of genes such as Rankl (also known as Tnfsf11), Bim (also known as Bcl2l11) and the Nr4a family members Nr4a1, Nr4a2 and Nr4a3, coupled with the downregulation of the cytokine receptor IL-7Rα¹⁵², all changes that can lead to T cell death^153,154. Of note, the expression and role of NR4A1 in multiple tolerance checkpoints (anergy⁷², exhaustion¹⁰³ and death¹⁵²) warrants further studies that elucidate the contribution of this regulator to each of these settings.

An important mechanistic finding was that the balance between the proapoptotic mediator BIM and the antiapoptotic mediator BCL-2 can determine the eventual fate of the tolerized T cell. It was shown in a model of antigen expression and tolerization of transgenic antigen-specific T cells that BIM-deficient T cells are resistant to deletion in response to tolerogenic stimulation but become equally anergic to their wild-type counterparts¹⁵⁵. Additional studies are needed to define the exact mechanisms that lead tolerogenic stimulation to drive peripheral deletion versus anergy.

A second peripheral death checkpoint occurs when activated T_eff cells are restimulated after activation, leading to activation-induced cell death (AICD)^156,157 (more recently described as restimulation-induced cell death¹⁵⁸). This cell death mechanism is induced by the extrinsic pathway of apoptosis via death receptors (the most well described is FAS (also known as CD95)^159,160, but they also include TNFR1, TRAILR1 and TRAILR2¹⁶¹), which signal through caspase 8 to trigger downstream executioner caspases. The intriguing finding that mouse and human CD4⁺ T cells are selectively susceptible to FAS-mediated AICD^162,163 whereas human CD8⁺ are more susceptible to TNFR1-mediated AICD¹⁶⁴ is noteworthy and warrants further investigation. This peripheral deletion mechanism serves as a self-limiting feedback process to control T cell expansion and is essential for the process of clonal contraction, wherein antigen-specific T_eff cells are eventually eliminated during the termination of an immune response. Co-stimulation is important in this context because signalling through CD28 helps in reducing AICD in T cells in the clonal expansion phase through the strong upregulation of the caspase 8 inhibitor cFLIP¹⁶⁵ and the upregulation of antiapoptotic factor BCL-X_L (REF.¹⁴⁷). T cells are sensitive to FAS-induced AICD only after BCL-X_L downregulation, which occurs later in the effector response. Therefore, T cells in the initial clonal expansion phase are spared from FAS-induced deletion by AICD. At present it is unclear how memory T cells survive this clonal contraction phase, although it is thought that their dependence on homeostatic cytokines and the upregulation of prosurvival molecules may play a central role¹⁶⁰. For example, effector cells destined for memory cell differentiation exhibit higher IL-7R expression, although enforced IL-7R expression does not skew the cells towards a memory fate¹⁶⁶. It has also been demonstrated that the sensitivity to IL-2 can determine the survival of these T cells, as IL-2RA^hi T_eff cells are more prone to apoptosis, whereas less-sensitive IL2RA^low cells give rise to long-lived memory T cells¹⁶⁷. The findings regarding IL-2 may appear counterintuitive at first glance but are appreciated through examination of the dual role of IL-2 in T cell survival at different stages. During the initial phase of T cell activation, IL-2 promotes survival and clonal expansion. By contrast, during the down phase (clonal contraction at the end of a response), IL-2 promotes sensitization to restimulation-induced cell death by enhancing FASL expression and suppressing cFLIP expression^168,169. In addition to these active mechanisms of T cell death induction, there remains substantial evidence that the absence of survival signals mediated by cytokines contributes to T cell death, termed ‘cytokine withdrawal-induced cell death’ or ‘activated cell autonomous death’. In this case, the absence of cytokine signalling reduces the levels of the antiapoptotic factors BCL-2 and BCL-X_L, leading to increased expression and activity of BIM, which ultimately results in T cell apoptosis^158,161,170.

Apart from apoptosis, there are multiple other mechanisms of T cell death that may contribute to peripheral T cell tolerance. One example is necroptosis (also called ‘programmed necrosis’, a type of programmed cell death that is caspase-independent and kinase RIPK3-dependent), which has been observed in T cells that are deficient in caspase 8 on TNF receptor signalling, impyling that caspase 8 is an inhibitor of necroptosis¹⁷¹. Ferroptosis (a type of programmed cell death that is dependent on iron) is induced through the accumulation of ROS that lead to membrane lipid peroxidation and subsequent cell death. It was recently shown that the ROS scavenger GPX4 is a critical survival mediator expressed by peripheral effector and memory T cells but not by thymocytes¹⁷². More importantly, several studies demonstrated that T follicular helper cells (T_FH cells) are uniquely susceptible to caspase-dependent pyroptosis (a type of pro-inflammatory lytic programmed cell death that is triggered by inflammasome activation and mediated by caspase 1 (or caspase 11)) by the ionotropic ATP-gated receptor P2X₇ in response to ATP^173,174. Two very recent reports demonstrated pyroptosis in resting (unactivated) human T cells, where it was induced by the activation of caspase 1 through the inflammasome sensor CARD8. This is an exciting finding as this mechanism likely depends on danger signals and not TCR signalling^175,176. This death pathway may therefore control the number of T_FH cells and may restrain the pathogenic T_FH cell states that are observed in conditions such as lupus erythematosus¹⁷³.

Despite the complex multilayered networks of death pathways in T cell biology, there is an appreciation that this tolerance checkpoint mechanism remains the most efficient and most reliable means to restrain T cell expansion at almost every stage of T cell activity. This enables the control of the magnitude and timing of the immune response and thereby prevents the onset of T cell-mediated immunopathology.

Conclusion

The identification of numerous genes and networks that regulate T cell tolerance has yielded clues about the mechanisms that protect us from hyperinflammation and autoimmunity. Here, we argue that intrinsic T cell tolerance is regulated by multiple mechanisms that work in harmony to achieve both constitutive and negative-feedback regulatory mechanisms and sets a strong barrier against pathologic inflammation. Nevertheless, several outstanding questions remain about the relative contribution of these mechanisms to protection versus immune dysregulation. For example, it remains unclear how the constitutive regulators of naive T cell quiescence cooperate in maintaining this state, and the molecular activities of most of the factors involved remain elusive. Another ill-defined aspect is the impact of anergy versus death versus exhaustion in the induction of T cell tolerance in cancer. Given the lack of mouse studies on T cell senescence, the contribution of this mechanism to T cell tolerance in different settings (for example, viral infections) and its effector molecules remain to be better defined. Peripheral deletional tolerance has been appreciated as a tolerance regulatory mechanism for decades. Despite this, we remain ignorant of the potential contribution of cell death mechanisms such as necroptosis and ferroptosis to T cell fate and regulation. Finally, we have reached the stage where complex phenotyping of immune cells allows unprecedented resolution of cell states, including multiple T cell tolerance states in the same model system, with exquisite temporal resolution. This enables us to ask the question of what tolerance mechanisms predominate in each setting and what mediators are at play.