An Updated Model for the Epigenetic Regulation of Effector and Memory CD8⁺ T Cell Differentiation

CD8⁺ T cells, also known as CTLs, have an essential role in protection against intracellular pathogens and in antitumor immunity. Upon encountering their cognate Ag, naive CD8⁺ T cells become activated and undergo clonal expansion, resulting in a heterogenous population of CTLs that have distinct phenotypic and functional properties. Conventionally, these activated CD8⁺ T cells at the peak of the expansion phase have been characterized into two major populations: 1) short-lived effector cells (SLECs), which express high levels of the killer cell lectin-like receptor G1 (KLRG1), CD25 (IL-2Rα), and have high cytotoxic activity, and 2) memory precursor cells (MPECs), which have high expression of IL-7Rα (also known as CD127) (1, 2), CD62L, CXCR3, and CD27, among other receptors (3–6). Once the infection is resolved, most of the CD8⁺ T cells die by apoptosis, and ∼10% of those Ag-specific CTLs survive and differentiate into long-lived memory CD8⁺ T cells, providing durable protection (7).

Several models have been proposed to explain the plausible generation of effector and memory T cell heterogeneity (8). The separate precursor model emphasizes that naive CD8⁺ T cells are already preconditioned to differentiate into effector or memory CD8⁺ T cells. This view is different from the decreasing potential, signal strength, and asymmetric cell fate models, which consider that the time of exposure to Ag and inflammatory signals, the strength of TCR/costimulation/cytokine signaling, and asymmetric distribution of cellular components upon T cell activation drive CTL differentiation, respectively. Despite evidence supporting these different models, none can fully reconcile with all the experimental findings.

It has been thought that naive CD8⁺ T cells contain little heterogeneity, suggesting that the differentiation is not likely due to cells predestined to become effector or memory cells (separate precursor model), but instead due to the dynamic changes happening postactivation (9–11). However, in support of the separate precursor model, it has been recently suggested that naive CD8⁺ T cells that originated at different developmental stages differentiate distinctively into effector or memory cells (12). Using fate-mapping approaches, Smith et al. (12) were able to label naive CD8⁺ T cells that originated from birth to adulthood and demonstrated that fetal naive cells have a tendency to differentiate into effector-like cells upon antigenic exposure in adulthood, whereas cells generated during adulthood preferentially gave rise to the memory pool. Consistently, the transcriptome as well as chromatin accessibility were distinct between naive cells that developed at different stages through life, suggesting naive cells are heterogenous. New tools, including single cell chromatin immunoprecipitation sequencing, scDNA methylome sequencing, and the single-cell assay for transposase-accessible chromatin using sequencing, can properly assess naive CD8⁺ T cell heterogeneity at the epigenetic level (13–17).

The asymmetric cell division has been initially suggested by Reiner and colleagues (18) and has since been observed in some models (19–22). The immune synapse between APC and CD8⁺ T cells provides clues for asymmetric cell division that affects effector and memory fate and is required for maintaining this asymmetry via regulating polarity network proteins (19). Chang et al. (20) observed that distribution and unequal degradation of T-BET by proteasome drives the asymmetric division within 48 h of T cell activation. Similarly, Pollizzi et al. (23) showed the asymmetric partitioning of mammalian target of rapamycin complex 1 (mTORC1) activity in activated naive T cells on the first division, with one daughter cell showing high mTORC1 activity and expressing more effector-associated molecules, whereas the other daughter cell shows low mTORC1 activity but enhanced long-term survival capability. In line with this report, Borsa et al. (22) further demonstrated that mTOR inhibition by rapamycin treatment enhances memory population differentiation and increases asymmetric cell division rate.

Using single-cell transcriptome analysis in mice during acute infection has revealed hundreds of genes are differentially expressed between pre-SLEC and pre-MPEC CTLs upon the first division of CTLs (9–11). Arsenio et al. (10) performed single-cell sequencing and, using 18 hidden Markov models, concluded that the asymmetric division bifurcation led to pre-T effector and pre-T memory cells. However, this view was challenged by Flossdorf et al. (11), who reanalyzed the same data using the unsupervised machine learning algorithm Monocle and found that naive CD8⁺ T cells follow a progressive differentiation path shared between short-lived effector and memory precursor CTLs (24). Despite the controversy surrounding how asymmetry cell division steers T cell fate, evidence suggests that the unequal division promotes CTL differentiation heterogeneity and is at least one of the mechanisms driving differentiation.

There has been more support in the models proposing that TCR signal strength and/or its duration, together with costimulatory molecules and cytokines, drive the commitment toward effector and memory CTLs. CD8⁺ T cells that receive adequate but not excessive signals become memory CTLs, whereas any excess of these signals leads to effector CTL fate during priming (25–27). Several reports have confirmed the crucial role of multiple cytokines, including IL-2, IL-7, IL-15, IL-21, type-I IFNs, IL-1β, IL-12, and IL-10 in balancing the differentiation of effector and memory CTLs (27–36). These cytokines signal through Jak–STAT pathways, highlighting the importance of STAT transcription factors in activation, proliferation, and/or differentiation of CD8⁺ T cells (37). More recently, Solouki et al. (38) took advantage of Itk^−/− Rag^−/− OT-1 TCR transgenic CD8⁺ T cells that have blunted TCR signaling and demonstrated that the TCR strength and Ag affinity independently regulate memory CD8⁺ T cell formation that can be modulated by an inflammatory milieu. Contrary to these observations, Li et al. (39) used a Nur77^GFP Bim^mCherry double-reporter mouse model and identified that central memory CTLs showed high expression of Bim, which correlated with high TCR signal strength. Future studies are needed to clarify these discrepancies and further understand how TCR signal strength, Ag affinity/avidity, as well as the inflammatory environment concomitantly regulate CTL differentiation.

In recent years, a higher complexity in the heterogeneity of effector CTLs has been uncovered. Flavell’s group (40) has found that during the expansion phase KLRG1⁺ CD127⁺ effector CTLs can lose KLRG1 and differentiate into memory CTLs during the contraction phase in a CX3CR1- and transcription factor BACH2-dependent manner. During the contraction phase, CX3CR1-expressing effector CTLs preserved stemness and had the potential to become functional, robust memory cells. Other reports have characterized a long-lived effector cell or terminally differentiated effector memory cell population found at memory time points during infection that possesses the characteristics of both effector and memory T cells and also express high levels of CX3CR1, although this population differs from the conventional central memory and effector memory CTLs (41–43). Youngblood et al. (44) showed that memory CTLs do pass through an effector state, indicating that CTL differentiation is not likely to be linear and that commitment toward memory cells may require a transition through an effector state.

Altogether, these models only partially reveal different aspects of CD8⁺ T cell differentiation, and a more comprehensive view is required. A unified model that integrates observations from all of these aspects is desired to conceptualize the major cell fate decision-making process during differentiation.

CD8⁺ T cell differentiation and subsets of transcriptional identity are governed by several fate-determining transcription factors. Expression and/or activation of these transcription factors is driven by signaling downstream of TCR, costimulatory molecules, and cytokine/chemokine receptors (45). Transcription factors BATF, IRF4, T-BET (encoded by Tbx21), ID2, BLIMP-1 (encoded by Prdm1), STAT4, ZEB2, and NFAT1 have been shown to regulate effector CTL differentiation, whereas memory CD8⁺ T cell formation is driven by the combined activity of transcription factors EOMES, ID3, ZEB1, BCL-6, TCF-1 (encoded by Tcf7), RUNX3, STAT3, and NFAT2 (reviewed in 8, 46, 47). During the effector and memory CTL differentiation, drastic chromatin accessibility changes are accompanied by dynamic transcription factor–regulated gene expression among different subsets (48–54). Understanding the modulation of the chromatin accessibility profile during CD8⁺ T cell differentiation is crucial for harnessing these cells in immunotherapies and vaccine development.

Epigenetic regulators are factors that conduct heritable cell changes, including differentiation, that do not involve alterations in the DNA sequence. Epigenetic regulation includes DNA methylation, long noncoding RNA interaction, and covalent modifications of histone residues. Although chromatin dynamics of naive T cell activation are directly associated with transcriptional control of gene expression and T cell lineage commitment (50, 55), the exact molecular mechanisms governing this process are still elusive.

Two models of CD8⁺ T cell differentiation incorporating the epigenetic interplay have been proposed: 1) the circular model, and 2) the developmental differentiation model or linear model (48). The circular model considers epigenetic modification changes with the cyclical burst of proliferation, proposing that there is an on-off-on or off-on-off pattern of transcriptional and epigenetic changes over time that enhances effector or memory differentiation (48). The developmental differentiation model depicts a linear, progressive acquisition of effector-associated transcriptional and epigenetic profile during CD8⁺ T cell differentiation that depends on signal strength and duration. Naive cells activated on the lower end of the signaling strength spectrum during this period commit toward a memory population (48). Empirical evidence suggests that these models might be operating simultaneously (22, 40, 56–60), and therefore a unified view needs to be reconsidered. In this article, we will discuss the recent advances in epigenetic regulation of CD8⁺ T cell biology and propose a model contemplating these recent studies.

A CpG island is formed when a cytosine nucleotide followed by a guanine nucleotide appear in high frequency in a particular DNA region. In humans, ∼72% of all promoters contain CpG islands (61). Furthermore, nearly 70–80% of the CpG islands are methylated in mammalian cells (62). The methylation of the CpG island near promoters is associated with repression of transcription (63–65). 5-methylcytosine (5mC) is catalyzed by DNA methyltransferase (DNMT) proteins. DNMT3a/b are known for establishing de novo methylation of cytosine nucleotides in CpG islands, whereas DNMT1 is necessary for maintaining methylation patterns during DNA replication (66, 67). During CD8⁺ T cell differentiation, peripheral CTLs lacking DNMT3a have been shown to generate lower percentages of KLRG1⁺ CD127⁻ effector cells and higher percentages of KLRG1⁻ CD127⁺ MPECs when responding to several different acute viral infections (68). DNMT3a deficiency in T cells results in an increased commitment toward a central memory population with no defect in effector function and no significant changes in the expansion of Ag-specific cells during the effector phase of acute viral infections (68). Mechanistically, DNMT3a represses TCF-1 expression through methylation of CpG islands at the Tcf7 (which codes for TCF-1) promoter region in early effector cells (68). A more recent study revealed that DNMT3a is responsible for methylation of many other naive/memory–associated genes, including Ccr7 and Sell, which code for CCR7 and CD62L respectively, in early effector cells (44). Furthermore, wild-type control has ∼50% more de novo methylated regions than DNMT3a-deficient effector CTLs. Altogether, these results suggest that these newly methylated genomic regions are crucial for effector cell differentiation (44). Similarly, deficiency in DNMT1a results in reduced effector CD8⁺ T cell generation, with generation of defective memory CTLs that do not respond properly upon rechallenge (69). It is surprising, however, that CD8⁺ T cells deficient in the methyl-CpG reader MBD2 show reduced memory CTL generation (70). Thus, how methylated CpG islands are recognized by readers to drive effector and memory CTL differentiation remains unclear.

In contrast, DNA methylation is removed by a family of methylcytosine dioxygenases called ten-eleven translocation (TET). In mammalian cells, TET1, TET2, and TET3 are responsible for the 5mC demethylation (71). Mechanistically, TET dioxygenases catalyze the oxidation of 5mC to 5-hydroxymethylcytosine (5hmC), then to 5-formylcytosine, and 5-carboxylcytosine. The 5-formylcytosine and 5-carboxylcytosine are substrates for thymine DNA glycosylase (TDG) to convert them into unmodified cytosines (71–74). Another event resulting in loss of DNA methylation could include a lack of remethylation of hemimethylated cytosines after replication or a combination of this together with TET enzymes only converting 5mC into an intermediate, such as 5hmC, then the methyl mark would be diluted after subsequent replication rounds. TET proteins have been shown to play important roles during T cell development. TET2 and TET3 are expressed at higher levels than TET1 during T cell development in the thymus and peripheral T cells (75, 76). Mixed bone marrow chimera and deletion of TET2 using CD2-Cre at the DN4 stage does not show altered thymocyte development or peripheral T cell populations, suggesting that TET3 might compensate for TET2 function (75, 77). Besides their role in development, TET proteins have been shown to differentially regulate Th differentiation (75, 78, 79).

In CD8⁺ T cells, TET2 deficiency alone skews CTLs toward the memory population without compromising the survival and proliferation, or the cytokine production capability of Ag-specific cells at the peak of CTL response during acute infection (80). This is similar to what has been reported for DNMT3a, although these two epigenetic modulators display opposing methylation/demethylation functions. TET2 depletion also leads to increased central memory population at the memory phase with higher EOMES expression. One or multiple regions within the Tbx21, Prdm1, Irf4, and Runx3 loci have enriched DNA methylation upon TET2 deficiency (80), suggesting that those sites might be crucial for TET2 to demethylate and to promote a proper effector program. TET2 deficiency results in enhanced effector function, as shown by transfer of central memory CD8⁺ T cells in a Listeria protection model (80).

In conclusion, these data indicate that the DNA methylation via DNMT3a is crucial for the repression of genes maintaining naive/memory identity, whereas demethylation by TET2 is indispensable for activating many effector transcription factor–controlled programs. The specific redundancy between TET family members, how DNMT1 and DNMT3 might overlap in their functions, and the specific role of methyl-CpG readers in this process still needs to be better comprehended.

Histone modifications, such as acetylation, methylation, sumoylation, phosphorylation, and ubiquitination play important roles in regulating gene expression (48, 81). These modifications affect the nucleosome positioning that grant or deny the accessibility of transcription factors (TFs) required for gene expression (81). The histone modifications that promote active transcription are known as permissive histone marks. Histone 3 lysine 4 monomethylation (H3K4me) and histone 3 lysine 4 dimethylation (H3K4me2), enriched at enhancer regions, and H3K4 trimethylation (H3K4me3), usually enriched at active enhancers or gene promoters, are associated with less dense nucleosome structures, characteristic of a transcriptionally open or poised (containing also repressive marks and therefore known as bivalent) chromatin in human CD4⁺ T cells (82). To better understand the role of H3K4 modification in CD8⁺ T cell differentiation, Russ et al. (83) performed chromatin immunoprecipitation sequencing using CD8⁺ T cells that respond to acute influenza A infection. The authors found that naive CD8⁺ T cells already showed H3K4me1/2 marks at some terminally effector-associated genes, such as Prdm1, Tbx21, and Zeb2, suggesting that these genes are poised for transcriptional activation, whereas some terminal enhancers also acquired H3K4me3 marks upon differentiation (83).

Other modifications associated with open chromatin accessibility and active transcription are acetylated H3 lysine 9 and lysine 27 residues (H3K9Ac and H3K27Ac, respectively) at either promoter and/or enhancer regions (48, 84). Araki et al. (85) found that H3K9Ac is enriched at the Eomes, Prf1, and Gzmb loci in in vitro–differentiated memory CTLs but not naive cells, and these genes in memory cells either are consistently expressed at a higher level or can be rapidly expressed (Gzmb), suggesting that H3K9Ac correlates with gene expression in CD8⁺ T cells.

New research from the Goldrath laboratory (86) has shown that the chromatin regulator BRD4, which contains a chromodomain that recognizes acetylated histones, is required for proper effector CTL differentiation. BRD4-deficient cells or in vivo inhibition of BRD4 activity resulted in reduced effector cell generation and a concomitant increase in central memory cells. BRD4 overlapped genome-wide with H3K27Ac marks, particularly in the effector-associated gene loci Bhlhe40, Id2, Zeb2, Gzma, Kdm6b, Gzmb, and Klrg1 (86). Although BRD4 levels are similar between distinct CTL populations, these findings clearly indicate that expression profile data at the RNA level by itself may undermine the role of certain epigenetic regulators.

The repressive posttranslational modifications (PTMs) found to regulate CD8⁺ T cell differentiation are deacetylation or methylation of H3K9 and H3K27 residues (48, 84). H3K9 methyltransferase G9a and the histone deacetylase HDAC2 have been identified to coregulate the expression of Cd27 and Il2ra genes in a BLIMP-1–dependent manner at the peak of the antiviral response. In the absence of BLIMP-1, these loci showed increased H3K9 acetylation and reduced methylation, which resulted in an increase in memory generation (87). SUV39H1, another lysine 9 methyltransferase, was shown to repress stemness/memory–associated genes, such as Sell and Il7r in effector cells, further reinforcing the generation of SLECs (88). More recently, HDAC3 was shown to suppress the development of short-lived high cytotoxic CTLs during differentiation upon acute infection (89). Inhibition or deficiency of HDAC3 led to increased T-BET and Granzyme B expression within 48 h of CD8⁺ T cell priming, suggesting HDAC3 negatively regulates cytotoxic response early during activation. Indeed, RNA-sequencing analysis identified that many effector-associated genes are upregulated upon Hdac3^−/− or HDAC3 inhibition in vitro postactivation. Hdac3 knockout CTLs showed a significant increase in global H3K27Ac levels as well as enhanced enrichment of H3K27Ac marks at gene loci associated with effector or cytotoxicity (89).

The polycomb repressive complex 2 (PRC2) mediates methylation of H3K27 residues. Deficiency of enhancer of zeste homolog 2 (EZH2), the PRC2 core component with methyltransferase activity, in activated T cells using Gzmb–Cre-driven deletion results in impaired effector cell differentiation and impaired secondary responses to reinfection (90). Mechanistically, EZH2 targets memory/survival–associated gene loci, including Tcf7, Id3, Bcl6, Bach2, and Bcl2, among others, to induce deposition of H3K27me3 residues, thereby repressing their expression (90). Similar results were also reported by Yeo and colleagues (9), in which they also demonstrate that EZH2 deficiency results in impaired effector CTL differentiation. EZH2 as well as other PRC2 components, including EED, and SUZ12 were highly upregulated upon the first division of cells committed to an effector path in vivo (9). However, conflicting results regarding the proliferation or survival of EZH2-deficient cells in vivo have been reported (9, 90, 91) and require further elucidation.

Bivalent deposition of H3K4me3 and H3K27me3 on the same locus is considered to poise gene expression, allowing for adequate temporal regulation of these genes in the presence of the appropriate signals (92). In naive CD8⁺ T cells, many genes critical for lineage commitment, differentiation, metabolism, and cell proliferation possess bivalent H3K4me3 and H3K27me3 marks. Upon naive T cell activation, the repressive H3K27me3 and permissive H3K4me3 marks are extensively changed, facilitating the differentiation of CTLs into effector and memory cells (83, 90, 93). In particular, demethylation of H3K27me3 is associated with increased chromatin accessibility and further deposition of activating marks to allow for induction of gene expression (94). Demethylation of H3K27me3 residues is mediated by lysine demethylase 6 (KDM6) family members KDM6a and KDM6b (94, 95). We recently showed that KDM6b is required for the proper generation and function of effector CD8⁺ T cells. KDM6b deficiency results in defective open chromatin accessibility early during acute viral infection at sites that require active removal of H3K27me3 residues from naive cells for cells to commit to an effector phenotype (96). Thus, KDM6b deficiency resulted in defective effector-associated gene expression, which led to impaired effector function as assessed by tumor control or upon secondary challenge (96). Similarly, KDM6a has also been shown to regulate effector CTL differentiation by demethylating H3K27me3 residues from the Prdm1 gene, which codes for BLIMP-1 (97). Consistent with these observations, inhibition of H3K27 demethylases using GSK-J4 results in enhanced H3K27me3 deposition in effector-associated genes Tbx21 and Irf4 in vitro (98). Although Yamada et al. (97) observed increased memory formation in wild-type cells treated with this inhibitor, Turner and colleagues (98) observed that naive cells incubated with the inhibitor were unable to expand in vivo and showed inadequate recall responses. In our studies, we have not observed any defects in CD8⁺ T cell expansion upon KDM6b deficiency at the peak of a primary immune response or during memory maintenance. However, reduced expansion of Ag-specific cells was observed upon secondary exposure to Ag (96). Future studies aimed at identifying the possible redundancy or specificity of these family members are needed to fully comprehend their respective function during CD8⁺ T cell activation and differentiation.

There are other PTMs, such as H3K79 methylation that have recently been identified in regulating the CD8⁺ T cell differentiation and function, but whether it associates with transcriptional activation or repression is still not well understood (99). DOT1L regulates the deposition of mono- to trimethyl groups on H3K79 and governs the cell identity by prohibiting premature differentiation and effector function (99).

Overall, the dynamic and tightly-controlled regulation of PTMs of histone tail residues is fundamental for the proper differentiation and commitment to distinct T cell fates. Although research in the past few years has started to unveil the specific regulators required for these processes, more studies are required to understand the cooperative and/or redundant functions between these epigenetic modulators. In the final section, we attempt to provide a temporal and specific requirement of these regulators in the context of CTL differentiation.

Previous models described CTL differentiation from the phenotypic, transcriptional or epigenetic regulation perspective with an underscore of signal strength and duration (8, 48). In our model, we emphasize the timing and availability of epigenetic regulators in controlling CD8⁺ T cell differentiation through modulating effector- and memory-associated gene accessibly (Fig. 1A). Early T cell activation is almost unidirectional in gaining chromatin accessibility and transcription for early effector- and activation-associated genes required for both effector and memory CD8⁺ T cell differentiation. This is supported by the fact that, whereas short-lived effector and MPECs differ from their transcriptional profile, there is similarity in the open chromatin regions between them, which become open within the first 24 h of T cell activation from naive T cells (50, 55).

Pioneer transcription factors, activated upon TCR/costimulation and cytokine signaling, can bind heterochromatin and recruit chromatin regulators, facilitating nucleosome displacement and initiation of transcription (47, 100). In CD8⁺ T cells, BATF, AP-1 factors JUNB and c-FOS, and RUNX3 have been suggested to act as pioneer factors during T cell differentiation (47, 55, 101, 102). BATF in cooperation with IRF4 was shown to directly bind to chromatin regions that are not transposase accessible (103). The pioneering activity of AP-1 transcription factors c-FOS and JUNB was also suggested in activated human T cells (102). BATF, together with IRF4 and JUN, has been shown to positively regulate the expression of early genes required for proper activation of T cells and differentiation into early effector cells, further reinforcing their roles as pioneer factors (101). Similarly, Pipkin and colleagues (55) have also suggested that RUNX3 was required for inducing chromatin accessibility of regions containing binding site motifs for transcription factors belonging to IRF, Blimp-1–like, and bZIP families. Given that NFATs are rapidly translocated into the nucleus upon naive T cell activation (104, 105) and they differentially regulate effector and memory CTL differentiation (106), we propose that NFAT transcription factors could also act as pioneer factors. However, the exact genes regulated by distinct NFAT members and how they affect CTL differentiation still remain to be elucidated.

Following the initial activation of pioneer factors, we propose three stages of regulation in which epigenetic modulators (Fig. 1A) in concert with fate-determining transcription factors (Fig. 1B) drive CTL differentiation. In the first stage, which occurs early during T cell activation, epigenetic regulators that can be recruited by pioneer transcription factors facilitate the opening of chromatin in activation/effector–associated gene loci, and closing of naive/memory–associated gene loci. In the second stage, epigenetic regulators modulate the accessibility of effector- and memory-associated gene loci that results in the commitment of cells to effector or memory cell fates. In the final stage, we propose a stabilization of the epigenetic landscape that results in the full commitment toward generation of short-lived effector CTLs, whereas cells that commit to a memory fate retain the ability to dedifferentiate into other CTL subsets upon secondary Ag exposure.

The first stage of epigenetic regulation consists of gaining and repressing chromatin accessibility and gene expression in effector/activation–associated and naive/memory–associated genes, respectively (Fig. 1A). Our findings, together with two other recent publications, show that removal of H3K27me3 marks from effector-associated genes by KDM6b and KDM6a is essential in promoting effector cell differentiation (96–98). Upon removal of this repressive mark, rapid acetylation of H3K27 residues also occur (93, 98). In support of this, we found that as early as day 4 postinfection, KDM6b-deficient CTLs already showed defective induction of effector-associated genes and a concomitant increase in memory-associated and early effector-associated gene signatures in vivo (96). Consistent with our observations, inhibition of H3K27 demethylases activity in vitro resulted in immediate defective gene expression as early as 3 h postactivation (98). Similarly, the activity of TET2 in demethylation of effector-associated transcription factors has also been shown critical for reinforcing effector CTLs differentiation (80). In vitro activation of naive CD8⁺ T cells revealed upregulation of TET2 expression within 2 h of priming coupled with 5hmC accumulation (80). Therefore, KDM6 family members and TET2 act early during T cell activation/differentiation, allowing for increased accessibility and expression of effector-determining TFs, such as T-BET and BLIMP-1 (80, 96–98). However, there are differences in the phenotype of cells lacking KDM6 or TET2. For instance, we showed that KDM6b deficiency results in increased generation of memory CTLs, which were not functional upon recall responses (96). In contrast, deficiency in KDM6a or TET2 in T cells skewed the population toward higher memory cells with intact effector function and enhanced recall responses using similar experimental models (80, 97). Further investigation is required to determine whether other TET family members can compensate for TET2 function and whether the DNA methylation pattern is consistent before and after rechallenge as well as whether KDM6b compensates for KDM6a deficiency.

KDM6b-deficient CTLs not only have reduced expression of genes associated with effector the transcriptional program, but also have increased early effector/memory–associated gene expression (96). Although this could be an indirect effect of reduced effector gene expression, it also points to the need of repressing naive/memory–associated genes, at least temporarily (Fig. 1A). SUV39H1 has an indispensable role in promoting effector CTL differentiation through methylation of H3K9 residues as early as day 3 postactivation, which are often associated with heterochromatin to silence naive/memory genes (88). DNMT3a and to a lesser extent DNMT1 are crucial for DNA methylation and repression of genes maintaining naive/memory identity early during T cell activation (44, 68). Youngblood and colleagues (44) have demonstrated that memory genes are also initially silenced (before day 4.5 postinfection), even in cells that commit to a memory cell later during the differentiation process. However, losing DNMT3a does not affect cytokine IFN-γ and IL-2 production, even when commitment toward effector populations is reduced (107).

The second stage of the epigenetic regulation of CTL differentiation consists of fine-tuning effector/cytotoxicity–associated and memory/survival–associated gene expression (Fig. 1A). We propose that two layers of regulation happen at this stage, mediated by HDAC3 and PRC2, respectively, which can act simultaneously. One layer consists of removal of acetylated residues by HDAC3, resulting in the inhibition of the cytotoxicity and effector program (89). We propose that HDAC3 acts in the second stage during the differentiation because gene loci with bivalent marks, such as the ones observed in effector cells, first require removal of methyl groups in H3K27 residues in a KDM6-dependent manner. Upon removal of these marks, these loci would then become acetylated upon activation (93, 98), which would then require HDAC3 for the removal of acetylated marks at the same lysine 27 positions. Similar to HDAC3, we also found that KDM6b regulates genes associated with the cytotoxicity program, such as Gzma, Gzmb, and Prf1. Further studies need to identify the specific time points at which chromatin modifications and the players responsible for them occur in these loci. The other level of regulation in this second stage corresponds to PRC2-dependent trimethylation of H3K27 residues in memory-associated loci to maintain the commitment toward short-lived effector CTLs. Using H3K27me3 chromatin immunoprecipitation, Kaech and colleagues (90) found that H3K27me3 deposition in Tcf 7 and Bach2 loci was detected at day 10 after lymphocytic choriomeningitis virus Armstrong acute infection, but not on day 4.5, demonstrating a later effect of PRC2 in regulating the commitment toward effector CTL generation.

The last stage of epigenetic regulation consists on an “effector epigenetic commitment,” after which a terminally differentiated SLEC cannot commit to another state. We speculate that effector cells, once they reach a threshold of accumulation of certain modifications, do not have the capability to dedifferentiate into another subset. In contrast, different memory CTL subsets, which have the characteristic of stemness, are able to reacquire an effector fate upon subsequent antigenic exposure (108, 109). In support of this, it has been proposed that memory CTLs preserve the accessibility in memory/survival–associated genes while having poised effector-associated genes (85).

Our proposed model is not in discord with the finding by Herndler-Brandstetter et al. (40) showing that conventional KLRG1⁺ effector cells can lose KLRG1 expression and become memory CTLs. In their article, Flavell and colleagues (40) demonstrated that KLRG1^hi CX3CR1^int CD127⁺ cells are able to differentiate into memory cells during the contraction phase of the immune response. Similar studies have also found effector memory cells at memory time points (41–43). We argue that these newly identified terminally differentiated effector memory cells or long-lived effector cells have more epigenetic plasticity, similar to central memory or effector memory cells, compared with SLECs, given their long-lived nature. Further characterization of the distinct chromatin accessibility and histone/DNA modifications in these different cell types at a single cell level will allow us to comprehend their epigenetic stability and memory potential.

Our model provided a chronological view of the epigenetic and transcriptional regulation of CTL differentiation, with the identification of three stages of epigenetic regulation. We propose an initial stage that occurs upon T cell activation, in which pioneer transcription factors recruit a subset of epigenetic modulators that induce opening of effector-associated gene loci, whereas other epigenetic regulators are required for repression of naive/memory–associated genes. Following this unidirectional activation, a second stage of epigenetic commitment fine-tunes the expression of effector/cytotoxic as well as memory/survival genes. The final stage consists on epigenetic stability for effector CTL generation, whereas MPECs acquire an epigenetic landscape that is poised on effector-associated genes but permissive for memory-associated genes.

As described in this review, recent studies have been focusing on the role of epigenetic regulators in driving T cell differentiation. The potential cooperation between these regulators, how and when exactly they are recruited to chromatin, the specific kinetics in their regulation in vivo, and the redundancy and specificity between family members with similar catalytic activity or domains that recognize specific epigenetic marks remain to be understood. Other areas that need further development are a better comprehension of combinatorial histone PTM marks that control gene accessibility and their dynamic regulation during differentiation. Although we acknowledge that many questions still remain unanswered, this proposed model provides a framework for future research to fill the gaps in our knowledge. With the development of single cell sequencing approaches, temporal deletion of epigenetic regulators at different stages during the immune response in vivo, and gene editing tools like CRISPR/Cas9, some of these questions can be better answered. Understanding how these epigenetic regulators impact CTLs is critical to 1) comprehend the indirect effects of drugs targeting epigenetic regulators on CD8⁺ T cells in clinical applications in which other cell types, such as tumor cells, are the intended targets and 2) target CD8⁺ T cells during pathological contexts or in vaccine development and immunotherapies, which have shown great potential and promising results.

The authors have no financial conflicts of interest.