The heterogeneity of Tex cells mainly depends on their distinguished phenotypes, effector molecules, key transcription factors and inhibitory receptors. Four heterogenous Tex subpopulations with developmental relationships were identified during chronic viral infections and tumors (18). There existed two progenitor Tex (Tex^prog) subpopulations, Tex^prog1 and Tex^prog2, which shared related open-chromatin landscapes but differed in phenotypes and functions. Tex^prog1 upregulated genes associated with stem-cell biologies, such as Tcf7, Myb, Il7r and Sell. While genes preferentially up-regulated in Tex^prog2 were related to cell cycle (Cdk and Mki67) and cell migration (Anxa2, Itgb7 and Alcam). Tex^prog1 was resident and capable of transforming into Tex^prog2, which could re-localize from blood-inaccessible lymphoid niches to blood-accessible locales. The Tex^prog2 subpopulation underwent a proliferation-driven transition to the intermediate Tex (Tex^int) cells, which presented as a transitional population with some cytolytic and migratory capacities. The transcriptional characteristic of the Tex^int population contained genes linked to cytotoxic functions, such as Gzma, Gzmb, Prf1, Klrk1 and Cx3cr1. Tex^int also expressed higher TF genes like Tbx1, Zeb2 and Id2. The Tex^term subpopulation exhibited the most severe exhausted profile and the weakest cytotoxicity. Tex^term up-regulated multiple inhibitory receptor genes (e.g., Pdcd1, Lag3, Tigit and Cd244) and molecule genes related to terminal depletion (Entpd1, Cd101 and Cd38). Meanwhile, the Tex^term showed elevated TCR signalings with highly expressed Zap70 and Nfatc1. Moreover, although the key exhaustion regulator TOX (19, 20) was expressed by all Tex cells, the Tex^term population exhibited the highest expression.

The differentiation hierarchy of Tex cells confirms the progressive development of T cell exhaustion. Within the spectrum of T cell exhaustion, numerous Tex cells are identified. Zhang et al. identified five main pan-cancer Tex clusters across 30 bulk solid tumor types (21). These five Tex clusters in the TME showed Tex-specific profiles in glycolysis, chemokinesis and cytotoxic functions. The transcriptional factor TCF-1 and T-bet gradually decreased among the five clusters from Tex^prog to Tex^term. TOX had a decreasing trend from Tex^prog to Tex^int but was highest in the Tex^term. Besides, the Tex^term subset displayed reduced gene mutations and DNA methylations when comparing the genomic alterations. Similarly, Bengsch et al. also uncovered nine distinct Tex clusters in human HIV infection and lung cancer by using a comprehensive mass cytometry approach (22). Validated by a functional exhaustion score, the Tex clusters with higher scores had elevated co-expression of IRs and TF Eomesodermin (Eomes). Conversely, a lower score owned Tex clusters expressed higher levels of functional markers like CD127 and TCF-1 but lower levels of IRs like PD-1, CD38 and 2B4. Additional observations in patients with B-ALL also illustrated that Tex cells are remarkably heterogeneous in cell cycle and expression patterns of cytotoxic molecules (23).

Totally, the expression of IRs like PD-1, CTLA-4, LAG-3 and TIM-3 is considered a crucial factor for categorizing Tex cells. The PD-1^lo subset expressed fewer or no IRs such as TIM-3 and 2B4, and can differentiate into the PD-1^hi subset (24). TCF-1 plays a crucial role as a central transcription factor in the PD-1^lo Tex cluster (12, 25–27). The TCF-1⁺PD-1^lo Tex cluster exhibits a self-renewing capacity and maintains persistent T cell responses (18, 21, 22, 28, 29) in chronic infections and cancers. Based on the expression intensity of IRs and TCF-1, Tex cells are generally divided into three stages (1): the Tpex, which is the stem-like PD-1^loTCF-1⁺ Tex precursors (2); the Tex^term, which is the terminally differentiated PD-1^hiTCF-1^- Tex cluster with the highest exhaustion profile; and (3) the Tex^int, which is at a transitional stage between the Tpex and Tex^term, characterized by intermediate PD-1 expression but without TCF-1 and has the highest effector functions.

Epigenetic studies have shown that exhaustion is a unique and enduring stage in T cell differentiation rather than merely a momentary reduction of T cell function. Approximately 6,000 open chromatin regions distinguish CD8⁺ Tex cells from effector and memory T cells (30, 31). The exhaustion program is firmly established in the epigenome of Tex cells. Although the inhibitory receptor has been blocked, the epigenetic remodelings of Tex cells are minorly changed (30). This explains why reinvigorating PD-1^loCD8⁺ Tex cells can merely cause a transient cytotoxic activity in response to ICB therapy.

Tex cells exist within a range of steady and constant epigenetic landscapes. In a general model, exhaustion-regulated genes are accessible in Tex cells and are transcriptionally controlled by exhaustion-related transcription factors. Memory-related genes are inaccessible to the transcriptional machinery, whereas several effector-related genes are still available in exhausted T cells but are not effectively bound by transcription factors. For example, the chromatin-accessible regions at the Ifng and Gzmb locus both emerged in CD8⁺ Teff cells and Tex cells, but the gene expressions were significantly distinct on the basis of functional pathway enrichment (31). Tex cells exhibited significant alterations in the regulation of IFN-γ production and immune effector process pathways (31). However, in reinvigorated Tex cells by PD-L1 blockade, the exhaustion relevant genes like Pdcd1 and Ctla4 are still expressed for DNA demethylation of these gene loci. Although inhibition of PD-1 signaling activated the expression of many effector genes like Gzmb and Ifng, the revived Tex cells eventually returned to be exhausted under consistent high antigen stimulation in chronic LCMV infection. The anti-PD-L1 treated Tex cells were unable to differentiate into Tmem cells even if the antigen was eliminated (30). This suggests that the powerful reinvigoration of checkpoint blockade may not be long-lasting without any epigenetic alterations.

Via transposase-accessible chromatin sequencing (ATAC-Seq), CD8⁺ Tex cells reveal numerous distinctive genomic features. A Nr4a1 binding motif at -22.4 kb of the murine Pdcd1 locus becomes accessible in Tex cells (32), and a de novo enhancer at approximately 23.8 kb upstream the transcription start site of Pdcd1 is also identified (30, 31). Functional analysis of this enhancer has shown that it is necessary for the persistent and integrated expression of PD-1 in Tex cells (31). Epigenetic modifications at the Pdcd1 locus reflect the regulatory differences between Tex and Teff cells. During the early phase of chronic infection, the differentiation of CD8⁺ Teff cells is followed by a temporal loss of DNA methylation at the Pdcd1 locus (33). Following durable TCR signaling, histone acetylation facilitates the opening of Pdcd1 enhancer elements. Some of these gene loci are found to be recognized by T-bet, directly supporting that T-bet also functions to regulate PD-1 expression (34). In exhausted CD8⁺ T cells, the Pdcd1 region is entirely demethylated and it remains in this status even when inflammation is eliminated. The absence of DNA methylation leaves the Pdcd1 locus primed for accelerated expression (35).

It is crucial to identify gene loci that are epigenetically modified and related to severe T cell exhaustion for improving immunotherapies. Zhang et al. demonstrated that treating TILs with valproic acid, a histone deacetylase inhibitor, ex vivo reduced the exhaustion degree of Tex cells and improved the therapeutic effect of ICB treatment (36). Additionally, Ghoneim et al. found that Dnmt3a mediates the DNA methylation reprogramming in the development of terminal T cell exhaustion and targeting Dnmt3a synergizes with PD-1 blockade to improve ICB therapy (13). To develop new therapeutic options, an in-depth understanding of the molecular mechanisms underlying epigenetic modification involved in T cell exhaustion is needed.

T cell exhaustion is initially driven by persistent and high amounts of TCR stimulation in chronic infections or tumors (39). The calcium influx from TCR signaling promotes the activation of TFs NFAT, BATF and IRF4 (40). This TCR-responsive TF network is crucial for maintaining T cell expansion and function (41–43). Associated with the antigen loads, T cells express higher levels of these TFs in chronic infections than in acutely solved infections (44). At the early stage of T cell activation, NFAT translocates into the nucleus to form heterodimers with AP-1 for fostering the expression of T cell effector genes (45, 46). However, at the late effector stage, the ratio of nuclear NFAT : AP-1 was abnormally risen (46, 47). On the one hand, the consistent TCR signaling recruits superfluous NFAT to relocate into the nucleus, resulting in a higher ratio of NFAT : AP-1 and leaving many NFATs free. On the other hand, the increased BATF signaling antagonizes the formation of classic AP-1 complex, leading to less AP-1 and a much higher NFAT : AP-1 ratio (42, 48). NFAT is reported to function as an effective PD-1 promoter (46) and elevate the expression of NR4A and TOX (49). As a result, the excessively free NFATs primarily promote the expression of inhibitory receptor genes of T cells, and gradually drive the acquisition of exhausted phenotypes (11, 49, 50)

What’s more, the TCR signal strength determines the differentiation of Tex cells. It has been observed that the T cell cluster specific for antigen with lower affinity, compared with another cluster recognizing antigen with higher affinity, is more abundant in Tpex cells in the same tumor (51). Besides, Daniel et al. found that high TCR signaling correlates with more terminal Tex cells, whereas low TCR signal strength generates the effector-like Tex cell cluster with a killer cell lectin-like receptor (KLR)-expressed cytotoxic profile (52). Shakiba et al. also demonstrated that tumor-specific CD8⁺ T cells, whether undergoing low or high TCR signal strength, both upregulated inhibitory receptors and couldn’t mediate tumor control. However, high TCR signaling leads to an effector function loss and a dysfunctional state of CD8⁺ T cells while the low TCR signal restrains them in a functional state (53). These findings all illustrate that TCR signal strength is a key determinant of T cell response and Tex cell formation.

Costimulatory molecules play pivotal roles in regulating T cell proliferation, activation and differentiation. However, inappropriately dampened or enhanced co-stimulatory signals contribute to T cell exhaustion. The activation of T cells relies on both TCR signaling and co-stimulatory signaling from CD28. The effects of CD28 signal are compromised during persistent inflammation, where a significant level of CTLA-4 is expressed. Moreover, the engagement of PD-1 directly inhibits CD28 signaling and prevents T cell activation (88). Critically, CD28 is necessary for successful PD-1 blockade during chronic LCMV infection. Conditional deletion of CD28 has been found to diminish the efficacy of anti-PD-1 therapy in mouse tumor models (89). Continual CD27 signaling has also been reported to result in more severe T cell exhaustion due to excessive T cell stimulation (90). However, another member in the TNF receptor family, 4-1BB, may synergize with anti-PD-L1 to rejuvenate Tex cells (91). The immunoglobulin superfamily member CD226 also plays a role in T cell exhaustion. As a co-stimulatory receptor that mediates T cell adhesion and execution of cytotoxicity (92), CD226 competes with immune checkpoints like TIGIT for the same ligands expressed by cancer cells to execute the anti-tumor effects (93–95). Chronic stimulations on tumor-infiltrating Tex cells induce high expression of Eomes, which downregulates CD226 (67, 96, 97). The Eomes-dependent loss of CD226 limits TCR signaling and anti-tumor effects (98–101). Moreover, the loss of CD226 has been found to impair T cell response to PD-1 blockade (102–104). In a mouse melanoma model, anti-PD-1 therapy was unable to rescue effector functions of Tex cells lacking CD226, implying that the CD226^- Tex group displayed a more terminally exhausted profile (105).

Both proinflammatory and immunosuppressive cytokines have regulatory roles in T cell exhaustion. Specifically, high levels of proinflammatory factors trigger the production of unfavorable cytokines which promote T cell exhaustion during long-term infections and malignancies. IL-10 is typically responsive-elevated in inflammatory microenvironments, and blocking IL-10 can prevent and reverse T cell exhaustion (106). The mechanism of reversing exhaustion by blockade of IL-10 may be directly through activating T cells by STAT-3, or indirectly via modulation of suppressive antigen-presenting cells (APCs) (107). In addition, IL-6 and IL-27 are also important cytokines that upregulate co-inhibitory signals on tumor-infiltrating T cells, correlating with T cell exhaustion in both human and mouse tumor models (108–110). The transforming growth factor-β (TGF-β) is another kind of immunosuppressive cytokines that seemingly participated in terminal T cell exhaustion. The expression of TGF-β and the activation of downstream SMAD2 are hallmarks of Tex cells (111, 112). Inhibiting TGF-β1 signaling synergizes with the agonist of bone morphogenetic protein in CD8⁺ T cells in vitro enhanced the efficacy of Tex cells (113). However, it was also reported that therapeutic inhibition of TGF-β alone failed to control the persistent LCMV infection and scarcely enhanced the efficacy of antiviral T cells (114, 115).

In contrast, IL-2 responsiveness is crucial for maintaining the Tex cell pool, as IL-2R deficient Tex rapidly disappear during chronic antigen stimulation (116). In vivo IL-2 therapy after LCMV chronic infection has been shown to increase the number of antigen-specific Tex and achieve excellent viral control (117). Besides IL-2, recent studies have also indicated that supplying adequate amounts of IL-21 may allow TCF-1⁺ Tpex to escape from exhaustion and differentiate into a CX3CR1⁺ Teff subpopulation that exhibits lower PD-1 and displays improved cytotoxicity compared to Tex (16). What’s more, ambiguous in promoting or preventing T cell exhaustion, type I interferons (IFN-I) play complex roles in controlling T lymphocytes. IFN-I is critical for activating and inducing antiviral functions of CD8⁺ T cells (118), which seems beneficial to avoid T cell exhaustion. However, elevated IFN-I in the tumor microenvironment may also induce high levels of IL-10 and PD-L1 (119). What’s more, it has been reported that IFN-I signaling can promote terminal exhaustion by suppressing Tpex cell sustainment through antagonizing TCF-1 (25). On-going IFN-I exposure has also been linked to the promotion of co-inhibitory receptors (including PD-1, TIM-3, LAG-3, and TIGIT) expression in chronic viral infections (120–124).

In addition to the co-stimulatory molecules and cytokines that regulate T cell exhaustion, the TCR-linked phosphoinositide 3 kinase (PI3K)-Akt-mTOR pathway also plays an important role in regulating CD8⁺ T cell exhaustion. Upon T cell activation, PI3Kδ inactivates the downstream transcriptional targets FOXO1 and BACH2 to suppress Tcf7 expression (125–127). This results in the repression of TCF-1-mediated central memorization and the up-regulation of effector genes such as Tbx21 and Gzmb, which execute cytotoxicity functions (128–130). Inhibition of PI3Kδ has been reported to promote the expansion of the TCF-1⁺ Tpex subpopulation and improve the response to ICB therapy (131). Tex cells with reduced PI3Kδ activity preserve TCF-1 and promote the generation of a self-renewing memory Tpex population as well (125, 129, 130). Pre-inhibition of Akt or mTOR is also proven to helpfully maintain the stem-like CD8⁺ T cell pool and enhance the antitumor efficacy (132–134). However, it is noticeable that with the exhaustion progressing, mTOR inhibition may no longer limit the chronic viral infection (134), partly because PD-1 signaling weakens the PI3K-Akt-mTOR signaling and causes metabolic reprogramming including impaired glycolysis and mitochondrial activities in late-stage Tex cells (135). So, a more careful modification of this kinase signaling pathway is necessary to balance the stimulation of TCF-1⁺ Tpex cluster properly.