Transcriptional and epigenetic regulation of memory NK cell responses

SUMMARY

Natural killer (NK) cells are cytotoxic innate lymphocytes with key roles in host protection against viruses and malignancy. Notwithstanding their historical classification as innate immune cells, NK cells are now understood to have some capacity to mount memory or memory-like immune responses in which effector cells undergo antigen-driven expansion and give rise to long-lived memory cells with enhanced functionality. Understanding how antigen-specific effector and memory NK responses are regulated is an important and active area of research in the field. Here, we discuss key transcription factors and epigenetic processes involved in antigen-specific effector and memory NK cell differentiation.

Summary:

Here, we review key transcription factors and epigenetic process that regulate antigen-specific effector and memory responses by NK cells.

I. Introduction

NK cells are cytotoxic innate lymphocytes that play critical roles in host defense against viral pathogens and malignancy. Prototypical NK cell effector responses involve the release of cytotoxic molecules such as granzymes and perforin, which lyse target cells. Activated NK cells are also an important source of the proinflammatory cytokine, IFN-γ, which serves to amplify and tailor the overall immune response. NK cell activation and function are controlled by signaling through germline-encoded receptors, including families of activating and inhibitory NK receptors such as natural cytotoxicity receptors (NCRs), NKG2 receptors (e.g. NKG2D and NKG2A), killer cell immunoglobulin-like receptors (KIRs; in humans), and the Ly49 family receptors (in mice), among others^{1, 2}. NK cells also express Fc receptors that facilitate antibody-dependent cellular cytotoxicity (ADCC), as well as a variety of cytokine receptors and costimulatory receptors that further contribute to NK cell activation. In general, inhibitory NK receptors recognize major histocompatibility class I (MHC-I) and MHC-I-like molecules expressed on host cells, which serves to inhibit NK cell-mediated killing of healthy tissues. In contrast, activating NK receptors typically recognize stress-related or microbe-derived molecules that are upregulated on host cells in response to infection, malignancy, or cellular stress^{1, 2}.

NK cell activation is tightly regulated by the integrated balance of activating and inhibitory inputs from NK receptors, cytokines, and costimulatory receptors^{1, 2}. In addition, environmental signals such as inflammation and educational processes during development further fine-tune the activation threshold of an NK cell^3-6. For example, target cells with low MHC class I expression may be tolerated by NK cells in uninflammed tissues, but become rapidly targeted for killing under conditions of inflammation or infection^{5, 7, 8}.

Historically, NK cells have been classified as innate immune cells, owing to their rapid effector functionality even in the absence of prior antigen exposure and their reliance on a fixed repertoire of germline-encoded receptors for activation. However, recent studies showing that NK cells are capable of adaptive immune responses in certain settings have refined the field’s understanding of NK cell biology. Similar to memory T cell responses, memory NK responses are long-lived, exhibit antigen-specificity, and are enhanced or more protective than responses by naïve NK cells^9-11. Memory or memory-like NK responses have now been described in mice and non-human primates in a variety of immunological settings, including viral and bacterial infections, vaccination protocols, and delayed type hypersensitivity reactions^9-13. In addition, multiple studies have provided convincing evidence of memory-like NK cell populations in humans with a history of cytomegalovirus (CMV) infection^14-16.

Understanding the molecular signals that regulate antigen-specific effector and memory NK cell responses remains an active area of research in the field. The majority of studies on the subject have relied on an experimental model of mouse cytomegalovirus (MCMV) infection, which drives robust, antigen-specific memory NK responses in certain mouse strains (e.g. C57BL/6) that express the activating NK receptor, Ly49H⁹. In this model, Ly49H⁺ NK cells are activated by the MCMV-encoded glycoprotein, m157, which is expressed on the surface of infected host cells and serves as a ligand for Ly49H^17-20. The overall NK cell response to MCMV in Ly49H-bearing mice is broadly reminiscent of an anti-viral CD8⁺ T cell response. Infection drives the robust proliferative expansion of Ly49H⁺ effector NK cells, which reach peak numbers at ~day 7 of infection⁹. This effector phase is followed by a distinct contraction phase from ~day 10 to ~day 21 post-infection, during which most effector cells undergo apoptosis. The small pool of MCMV-experienced NK cells that survive past the contraction phase exhibit key features of immunological memory, including an ability to mount recall responses that are enhanced, long-lived, and antigen-specific⁹.

Studies over the past decade have shed light on the broader signaling pathways and specific molecules that regulate the effector, contraction, and memory phases of the NK cell response during MCMV infection. Optimal effector and memory NK cell responses appear to be broadly governed by the “3 Signal Model” defined in T cell studies, with signal 1 provided by antigen receptor engagement, signal 2 by costimulation, and signal 3 by inflammation²¹. In the context of MCMV-driven NK cell responses, recognition of MCMV-encoded m157 by the Ly49H receptor provides signal 1; costimulatory receptors such as DNAM1 provide signal 2; and various proinflammatory cytokines provide signal 3^{21, 22}. With respect to signal 3, optimal effector NK cell expansion is highly dependent on IL-12 and to a lesser extent on IL-18, IL-33, and type 1 interferon signaling; optimal memory cell responses, on the other hand, require IL-15 and IL-12, but not IL-18, IL-33, or type 1 interferons^23-27.

II. Transcription factors involved in antigen-specific memory NK responses

Signals 1, 2, and 3 initiate intracellular signaling cascades that act, in part, to regulate expression of genes involved in processes related to effector and memory cell differentiation, including proliferation, survival, and effector function. Transcriptional changes are orchestrated by the individual and cooperative activities of specific transcription factors acting at each stage of the response. These include both “signal-regulated” transcription factors (e.g. STAT4, STAT1, Zbtb32, and IRF8), which are upregulated in NK cells upon activation, as well as “lineage-defining” transcription factors (e.g. T-bet, Eomes, and Runx transcription factors), which play broader roles in NK cell development and lineage-identity^{23, 24, 27-31}. Here, we discuss key transcription factors with known roles in regulation of antigen-specific effector and memory NK responses (Figure 1).

Figure 1.

Schematic showing transcription factors known to regulate antigen-specific effector cell expansion and/or memory cell differentiation by Ly49H⁺ NK cells during MCMV infection in mice.

STAT4

NK cells are highly responsive to the proinflammatory cytokine, IL-12, which is produced by myeloid cells such as dendritic cells and macrophages in response to diverse stimuli, including infections involving intracellular pathogens³². Engagement of the IL-12 receptor (IL-12R) on NK cells initiates an intracellular signaling cascade involving Janus kinase (JAK)-mediated phosphorylation of signal transducer of activated T cells 4 (STAT4), which dimerizes and translocates to the nucleus to regulate transcription of thousands of gene involved in NK cell activation and immune function³¹.

Although IL-12R signaling is dispensable for NK cell development, it is essential for both effector and memory responses by Ly49H⁺ NK cells in the context of MCMV infection²³. NK cells lacking either IL-12R or STAT4 exhibit severe defects in effector cell expansion during the effector phase of the MCMV response, owing to a requirement for IL-12R-STAT4 signaling in proliferation, but not survival, by effector cells²³. At a transcriptional level, cell-intrinsic loss of STAT4 signaling in NK cells led to dysregulated expression of >1700 genes by day 2 post-infection. Many of these genes were confirmed to be direct targets of STAT4 in cytokine-stimulated NK cells in vitro, including Ifng and a number of transcription factors (e.g. Zbtb32, Tbx21, Stat1, Runx1, and Runx3) that are independently required for effector and/or memory NK cell responses, as discussed below^{31, 33}.

STAT1, STAT2, and IRF9

Type I IFN signaling is a well established driver of anti-viral immune responses. Canonical type I interferon receptor signaling involves phosphorylation and dimerization of cytosolic STAT1 and STAT2, which assemble with interferon regulatory factor 9 (IRF9), to form a heterotrimeric transcriptional regulatory complex known as interferon-stimulated gene factor 3 (ISGF3)³⁴. ISGF3 translocates to the nucleus and binds interferon-sensitive response elements (ISRE) at target loci to activate expression of IFN-stimulated genes (ISGs)³⁴. In addition to ISGF3, transcriptional regulatory complexes that lack IRF9 or utilize different combinations of STATs also act downstream of type 1 interferon signaling and may facilitate cell- or environment-specific tailoring of anti-viral immune responses^{34, 35}.

The importance of type I interferon and ISGF3-mediated anti-viral responses is underscored by the increased susceptibility of Ifnar-, Stat1-, or Irf9-deficient mice to diverse viral pathogens^{30, 36}. Indeed, host-protective NK cell responses to MCMV infection are severely compromised in mice lacking either the type 1 interferon receptor or individual ISGF3 complex members^{24, 30}. This susceptibility is due, in part, to the requirement for type 1 interferon-ISGF3 signaling in maintaining the pool of cytotoxic effector NK cells throughout the first week of infection^{24, 30}. Impaired type 1 interferon or ISGF3 signaling renders Ly49H⁺ NK cells highly susceptible to fratricide – i.e. killing by other activated NK cells – resulting in a significantly reduced effector cell pool at the peak of the effector response, although memory cell persistence remains intact^{24, 30}. Additionally, components of the type 1 interferon-ISGF3 signaling axis are required for optimal expression of cytotoxic molecules such as granzyme B in effector NK cells during MCMV infection, indicating that both effector cell survival and cytotoxic functionality are regulated by this critical signaling pathway^{24, 30}.

Consistent with the central role of type I interferon signaling in antiviral immunity and NK cell cytotoxicity, Gzmb and many canonical ISGs were highly upregulated in Ly49H⁺ NK cells in an IFNAR− and STAT1-dependent manner within 36 hours of MCMV infection³⁰. Notably, IFNAR-deficiency and STAT1-deficiency commonly affected only ~60% of differentially expressed genes, possibly reflecting roles for non-ISGF3 regulatory complexes acting downstream of IFNAR and/or roles for STAT1 downstream of other cytokine receptors such as IL-21 or IFN-γ. By day 4 post-infection, a time point coinciding with the onset of antigen-driven proliferation by Ly49H⁺ NK cells, ISGF3 signaling (defined by the requirement for IRF9) was not only required to drive ISG expression, but also to maintain high expression of MHC I genes and to limit expression of genes encoding activating NK receptor ligands³⁰. These findings indicate that ISGF3 acts downstream of type 1 interferon signaling, and possibly other signals, to protect Ly49H⁺ NK cells from fratricide through mechanisms that include promoting and limiting expression of ligands for inhibitory and activating NK receptors, respectively.

Studies examining naïve NK cells stimulated with IFNα in vitro have demonstrated that type I interferon signaling drives the expression of additional transcription factors involved in effector and memory NK responses, including Runx3, Tbx21, and the three ISGF3 genes³⁰. This is likely caused by STAT1 directly regulating these genes because STAT1 associates with the loci for Runx3, Tbx21, Stat1, Stat2, and Irf9 in IFNα-stimulated NK cells in vitro³⁰. Additionally, MCMV-driven upregulation of Stat1, Stat2, and Irf9 in Ly49H⁺ NK cells in vivo was strictly STAT1-dependent³⁰. Taken together, these findings suggest that type 1 interferon-ISGF3 signaling acts as an early initiator of transcriptional networks that regulate later effector and memory NK responses. Moreover, ISGF3 autoregulates expression of its own components to amplify canonical Type I interferon signaling in NK cells during viral infection.

Zbtb32

The BTB-ZF family protein, Zbtb32, is another signal-regulated transcription factor that is critically required for effector NK cell expansion during MCMV infection in vivo²⁸. A study from our group has shown that Zbtb32 expression is undetectable in naïve NK cells, but is rapidly and highly induced in activated NK cells in response to MCMV infection in vivo. Although signals from multiple cytokines – including IL-12, IL-18, and type 1 interferons – were shown to contribute to optimal Zbtb32 expression in effector NK cells in vivo, Ly49H receptor expression was notably dispensable²⁸. Moreover, stimulation of naïve NK cells with IL-12 and IL-18 in vitro was sufficient to induce STAT4 binding at a conserved sequence upstream of the Zbtb32 promoter, confirming that Zbtb32 is likely a direct transcriptional target of STAT4 in cytokine-activated NK cells²⁸. These findings indicate that Zbtb32 expression in effector NK cells is highly dependent on diverse inputs for signal 3, but is largely independent of signal 1.

Although the function of Zbtb32 in NK cell biology remains only partially understood, functional studies suggest that it plays a permissive rather than instructive role in effector cell proliferation. NK cells lacking Zbtb32 exhibit marked defects in effector cell proliferation following viral challenge²⁸. However, genetic co-deletion experiments suggested that Zbtb32 principally acts to antagonize another transcription factor, Blimp-1, which is typically associated with terminal differentiation and limited proliferative potential in other lymphocyte populations^{28, 37}. Exactly how interactions between Zbtb32 and Blimp-1 support the proliferative expansion of effector NK cells during infection remains largely unknown. Possible mechanisms for future investigation might include transcriptional repression of Prdm1 (encodes Blimp-1) by Zbtb32 and/or physical interactions between Zbtb32 and Blimp-1 that transiently disrupt Blimp-1 function.

IRF8

IRF8 (also known as ICSBP) is another IRF family transcription factor that has been implicated in antigen-specific effector responses by NK cells. IRF8 deficiencies are associated with increased susceptibility to viral and intracellular bacterial pathogens, consistent with severe defects in the IL-12-IFNγ signaling axis^{38, 39}. In addition, specific familial IRF8 mutations have been linked to defects in host-protective immune responses by NK cells in patients suffering from severe viral infections³⁸.

Similar to Zbtb32, Irf8 expression was shown to be rapidly, but transiently, induced in activated NK cells during MCMV infection²⁹. Upregulation required cell-intrinsic expression of IL-12R and STAT4, but not IFNAR, IFNGR, IL-18R, or STAT1. These findings indicate that IL-12 is the key inflammatory signal driving IRF8 expression in NK cells during MCMV infection²⁹. In addition, IRF8 expression was shown to be significantly higher in Ly49H⁺ NK cells relative to Ly49H⁻ NK cells, consistent with a role for antigen receptor signaling in Irf8 induction in antigen-specific NK cells, similar to findings in antigen-specific CD8⁺ T cells⁴⁰. Stimulation of naïve NK cells with IL-12 and IL-18 in vitro was sufficient to drive STAT4 binding at an open chromatin region in the Irf8 promoter, and this was required for the induction of H3K4me3 and promoter activation²⁹. Thus, antigen receptor and cytokine signaling act cooperatively to promote optimal IRF8 expression in effector NK cells, through a mechanism that includes STAT4-mediated chromatin remodeling at the Irf8 locus²⁹.

Studies involving cell-specific or inducible deletion of Irf8 in NK cells revealed a functional requirement for IRF8 in the proliferative expansion of effector NK cells during the first week of MCMV infection²⁹. In contrast, IRF8 was dispensable for memory cell formation, maintenance, and function in both the MCMV system and in a mouse model of NK cell-mediated contact hypersensitivity to haptens²⁹. Cell-specific Irf8-deficiency did not impact NK cell development, IFN-γ production or cytotoxicity, but was nevertheless required for controlling MCMV infection in vivo because of its role in effector cell expansion²⁹.

Consistent with the dual role of IRF8 as a transcriptional activator and repressor in other cells, similar numbers of genes were aberrantly up- and downregulated in Ly49H⁺ NK cells on day 4 post-infection, including many genes encoding proteins involved in metabolic, cell cycle, and DNA replication processes²⁹. The suite of downregulated genes included Zbtb32, which was shown to be a direct IRF8 target in ChIP experiments performed in cytokine-stimulated NK cells in vitro. These data suggest that IRF8 may support effector cell proliferation, at least in part, by inducing expression of Zbtb32²⁹. Taken together, these studies support a model in which IRF8 acts downstream of IL-12 and antigen receptor signaling to regulate the expression of genes involved in proliferation and metabolism in effector NK cells during viral infection.

T-bet and Eomes

The T-box transcription factors T-bet and Eomesodermin (Eomes) are important lineage-defining transcription factors with key roles in NK cell development^41-43. Recent studies indicate that both factors are also important for antigen-specific NK responses during MCMV infection²⁷. Experiments involving inducible deletion of Tbx21 (encodes T-bet) or Eomes in mature NK cells demonstrated that both factors were independently required for the differentiation and proliferative expansion of Ly49H⁺ effector NK cells during the first week of infection²⁷. Moreover, deletion of T-bet, but not Eomes, in NK cells during the contraction phase revealed a specific requirement for T-bet in memory cell generation or persistence²⁷.

Consistent with their distinct requirements in effector and memory cells, T-bet and Eomes exhibited different expression patterns in Ly49H⁺ NK cells throughout the course of MCMV infection. Although both factors were expressed at baseline in naïve NK cells, T-bet was further upregulated in effector cells as early as day 2 post-infection and was maintained at elevated levels in memory cells. In contrast, Eomes expression was lower in effector and memory NK cells compared to naïve cells²⁷. Similar to findings in CD8⁺ T cells⁴⁴, the induction of T-bet in effector NK cells during MCMV infection was dependent on IL-12R signaling²⁷. This observation was consistent with the finding that IL-12 and IL-18 co-treatment induced STAT4 binding at a putative enhancer upstream of the Tbx21 locus, and loss of STAT4 led to a reduction in the permissive histone H3K4me3 modifications at the Tbx21 promoter in in vitro activated NK cells²⁷. These findings suggest that STAT4-mediated chromatin remodeling at the Tbx21 locus promotes the elevated T-bet levels that underlie optimal effector and memory NK responses.

Runx 1 and Runx 3

Runx transcription factors are evolutionarily conserved proteins with diverse roles in cellular differentiation processes in a variety of cells, including immune cells. For example, Runx factors are known to play important roles in T cell differentiation and function⁴⁵, and Runx 3 has been shown to promote NK cell and helper ILC1 development by regulating expression of proliferation and cell survival genes downstream of IL-15 signaling^{46, 47}. All Runx factors possess a common Runt domain, which mediates heterodimerization with the shared co-factor, core-binding factor β (CBF-β). CBF-β is a non-DNA binding regulatory protein that enhances DNA binding by its Runx partners and facilitates recruitment of other co-activators or co-repressors to the Runx transcriptional regulatory complex.

Beyond their critical activities in development, Runx factors were recently shown to act as key regulators of effector and memory NK cells responses during infection. Cell-specific deletion of Runx3 or Runx1 in NK cells impaired effector cell expansion during the effector phase, resulting in a smaller memory cell compartment at later time points, although the defect was most pronounced in the absence of Runx3³¹. A similar defect in effector cell expansion was also observed when Cbfb was inducibly deleted in mature NK cells prior MCMV challenge³¹. In Runx1-deficient NK cells, impaired expansion was attributed to a reduction in proliferative capacity, which correlated with reduced expression of cell cycle-related genes at day 3 post-infection. Notwithstanding their respective roles in effector cell expansion, both Runx 1 and Runx 3 were shown to be dispensable for IFN-γ production and degranulation by memory NK cells upon restimulation with cytokines or mitogens ex vivo³¹.

Expression of Runx3 and Runx1, but not Runx2 or Cbfb, is upregulated in effector NK cells during MCMV infection in a STAT4-dependent manner³¹. In addition, stimulation of naive NK cells with IL-12 and IL-18 in vitro drives STAT4 binding at the promoter for Runx1 and at a putative enhancer upstream of the Runx3 locus. STAT4 binding at these sites correlated with a STAT4-dependent increase in H3K4me3 marks at the promoters of all three Runx genes and the Cbfb gene³¹. These findings suggest that Runx 1 and Runx 3 act, at least in part, downstream of IL-12R-STAT4 signaling in activated NK cells. It is worth noting that cytokines alone were sufficient to drive Runx3 upregulation in NK cells in vitro and both Ly49H⁺ and Ly49H⁻ NK cells upregulated Runx3 to a similar extent during MCMV infection in vivo³¹. In contrast, cytokine treatment was insufficient to induce Runx1 expression in NK cells in vitro, and only Ly49H⁺ but not Ly49H⁻ NK cells upregulated Runx1 in response to MCMV in vivo. Thus, while inflammation (signal 3) may be the key driver of Runx 3 upregulation in effector NK cells, induction of Runx 1 expression requires both antigen receptor signaling (signal 1) and inflammation (signal 3)³¹.

Other factors

In addition to the factors discussed above, several other transcription factors have been implicated in effector NK cell expansion during MCMV infection, including ID2 and KLF12^{48, 49}. In addition, deficiencies in the aryl hydrocarbon receptor (Ahr) impair liver-resident NK cell development, resulting in defective memory NK responses to haptens⁵⁰. Although a number of transcription factors have now been implicated in antigen-specific effector responses by NK cells, only a few have been shown to specifically regulate memory cell differentiation (Figure 1). Future work is needed to address this gap in knowledge and insights from studies on CD8⁺ T cell memory may be helpful in this regard in light of the extensive similarities between these two cell types.

III. Epigenetic regulation of effector and memory NK cell responses

Recent studies indicate that epigenetic remodeling likely plays a central role in programming durable phenotypic and functional changes in effector and memory NK populations. In humans, the memory-like NK population that expands and mediates host protection during HCMV infection or reactivation exhibits features of epigenetic remodeling at genes involved in effector function and intracellular signaling^51-53. For example, enhanced IFN-γ responses by these memory-like cells correlated with reduced DNA methylation at the IFNG locus, whereas reduced expression of the signaling molecules FcεRIγ and EAT-2 correlated with increased DNA methylation at the gene loci^{52, 53}.

Recent ATAC-seq and ChIP-seq studies indicate that extensive and highly dynamic epigenetic remodeling occurs in Ly49H⁺ NK cells as they progress through the effector, contraction, and memory phases in response to MCMV infection³³. For example, markedly different chromatin accessibility patterns were observed in NK cells isolated at early effector, late effector, and memory time points ³³. Overall, the largest changes in accessibility occurred during the first week of infection, with the majority occurring at putative regulatory elements in intergenic and intronic regions³³. In contrast, promoter regions exhibited less activation-induced remodeling, but were generally more accessible at baseline. These observations are consistent with recent studies showing that p300, a histone acetyltransferase found at enhancers and in high density at ‘super enhancers’, is rapidly recruited to enhancers for highly upregulated genes in NK cells stimulated with cytokines in vitro⁵⁴. These findings suggest that enhancers undergo rapid and extensive epigenetic remodeling in activated NK cells and may serve as the principal target for epigenetic processes that support durable effector and memory cell phenotypes.

In general, changes in chromatin accessibility, particularly at promoter regions, were shown to directly correlate with changes in transcript abundance at early time points after MCMV challenge, with increased transcripts mirroring increased chromatin accessibility at or near a gene locus³³. This phenomenon was particularly evident for genes implicated in type I and type II interferon signaling, such as Ifng and various ISGs³³. In addition, loci for a number of transcription factors (e.g. Tbx21, Ifr9, Irf8, and Zbtb32) and cell surface receptors (e.g. Klrg1) associated with effector cell differentiation were more accessible during the effector phase, although accessibility for most returned to “naïve-like” levels by day 14³³. In contrast, chromatin accessibility generally decreased surrounding genes with negative regulatory roles in NK immunity, such as Tgfbr1, Cish, Socs3, and Pdcd1, as well as at genes associated with NK cell immaturity, such as Cd27³³. Moreover, these loci typically remained closed in NK cells into memory time points, providing a possible mechanism by which memory cells become less sensitive to inhibitory signals that would otherwise limit recall responses. Tcf7 (encoding TCF-1) and Nfkb1 also became less accessible at later time points, correlating with lower expression in memory cells. This is consistent with the observation that TCF-LEF and NF-κB motifs were generally less accessible in memory cells compared to naïve cells³³. These findings suggest that TCF-1 and NF-κB may principally function to support naïve rather than memory NK cell function during MCMV infection. However, TCF-1 is known to support memory cell differentiation in CD8⁺ T cells and was recently implicated in the function of memory-like NK cells elicited by inflammation during HIV infection^{55, 56}. Thus, functional studies involving Tcf7-deficient NK cells will be important for clarifying the specific role of TCF-1 in effector and memory NK cell generation in the context of MCMV infection.

Among the changes in chromatin accessibility that persisted into the memory phase, most were already established by the early contraction phase, suggesting the epigenetic changes that underlie memory cell differentiation may emerge early in the response, just after or possibly during the effector phase ³³. Genes for which chromatin became durably more accessible in effector and memory NK cells included the anti-inflammatory cytokine, Il10, and the DNA methyltransferase, Dnmt3a³³. Dnmt3a has been shown to restrict memory precursor and promote terminal effector cell differentiation in CD8⁺ T cells through a mechanism that involves repression of Tcf7 expression. These findings suggest that Dnmt3a might be a particularly interesting target for follow-up studies on effector and memory cell differentiation in NK cells⁵⁷.

NK cell and CD8⁺ T cell responses exhibit many similarities during viral infection, at both a functional and transcriptional level²¹. These similarities extend to, and likely reflect, common changes in chromatin accessibility at specific gene loci that are induced by activating signals³³. Loci commonly altered in NK cells and CD8⁺ T cells during MCMV infection include a number of transcription factors with known roles in effector or memory CD8⁺ T cell responses, including Bach2, Tcf7, Tox, and Zeb2³³. For example, ZEB2 has been shown to promote CD8⁺ T effector (T_eff) and KLRG1^hi T effector memory (T_em) cell differentiation, but it represses the differentiation of central memory (T_cm) cells^{58, 59}. BACH2 and, as discussed above, TCF-1 support memory cell generation^{56, 60}, whereas TOX is required for CD8⁺ T cell exhaustion^61-63. Currently, little is known about how these transcription factors regulate antigen-specific effector and memory NK responses and each would be an interesting target for future studies in the field.

To date, studies aimed at understanding how specific transcription factors impact epigenetic remodeling in effector and memory NK cells have principally focused on roles for the ‘signal-regulated’ transcription factors, STAT4 and STAT1. Indeed, many changes in chromatin accessibility in Ly49H⁺ NK cells are broadly dependent on STAT1 and STAT4 at early time points after MCMV infection, although the specific changes orchestrated by each factor were often distinct³³. In general, STAT4-bound loci exhibited greater and more rapid changes in chromatin accessibility, whereas changes at STAT1-bound loci were milder and slower to evolve³³. In addition, STAT4 has been shown to preferentially bind at enhancer rather than promoter regions in activated NK cells^{31, 33, 54}, and this preference mirrored the broad requirement for STAT4 in activation-induced chromatin opening at enhancers in Ly49H⁺ NK cells following MCMV infection³³. These findings are consistent with reports that STAT4 orchestrates deposition of H3K4me3 marks at promoter and non-promoter regions and recruits p300 to enhancers in cytokine-activated NK cells in vitro^{31, 54}. Indeed, STAT4 has been shown to act as a pioneering factor at ‘de novo’ enhancers that are inaccessible in naïve cells, but become bound by STAT4 in activated NK cells, which promotes their accessibility for other transcription factors such as T-bet⁵⁴.

IV. Summary and open questions in the field

Our knowledge of the transcriptional networks that underlie antigen-specific effector and memory NK cell responses has expanded significantly in the past decade. Indeed, recent studies have led to the identification of more than a dozen signal-regulated and lineage-defining transcription factors, most of which act downstream of antigen receptor or cytokine signaling, involved in the regulation of effector and memory NK responses (Figure 1). Moreover, some of these factors, e.g. STAT4, likely effectuate long-term changes in effector or memory cell function by durably altering the chromatin landscape, particularly at existing and de novo enhancer regions. Notwithstanding significant recent advances in the field, much remains to be learned, particularly about regulation of memory cell differentiation and function.

Two critical questions in the field are how and when memory cell fates become programmed in activated NK cells. Possibly analogous to the KLRG1^loCD127⁺ memory precursor effector cell (MPEC) and KLRG1^hiCD127⁻ short-lived effector cell (SLEC) populations that exist within the effector CD8⁺ T cell pool, several observations suggest that the pool of effector NK cells generated early during infection may be a heterogeneous population, comprised of cells with varying degrees of “memory potential”. First, studies involving co-transfer of effector NK cells with a KLRG1⁻ and KLRG1⁺ surface phenotype demonstrated that KLRG1⁻ cells have a competitive advantage in expanding during the effector phase and seeding the memory pool after MCMV infection⁶⁴. These findings may reflect a higher memory potential among cells in the KLRG1⁻ fraction and/or an enrichment of short-lived, terminally differentiated effector cells in the KLRG1⁺ fraction⁶⁴. Second, NK cells with high or low Ly49H expression contribute to the effector and memory pools with different efficiencies. Compared to Ly49H^lo NK cells, Ly49H^hi cells are more proliferative and less apoptotic during the effector phase, and their progeny are present in higher frequencies in the memory pool at later time points^{65, 66}. Nevertheless, the impact of Ly49H signaling on effector function is likely more nuanced, as evidenced by the finding that Ly49H^hi effector cells are more cytotoxic against m157-expressing target cells in vitro, but produce less IFNγ during MCMV infection in vivo⁶⁶. In humans, memory-like NKG2C⁺ NK cells that expand during HCMV reactivation were also observed to express higher levels of NKG2C than NKG2C⁺ NK cells from seronegative individuals, suggesting that antigen receptor signaling may impact memory programming in human NK cells as well⁶⁶. Thus, similar to finding for CD8⁺ T cells, the magnitude or duration of antigen receptor signaling may play a role in programming memory potential in individual NK cells.

An intriguing possibility is that memory potential is established in NK cells very early after activation, possibly through a mechanism involving asymmetric cell division. Indeed, asymmetric partitioning of proteins involved in signaling, transcriptional regulation, and cell fate specification has been reported in activated CD8⁺ T cells undergoing cell division^{67, 68}. Whether asymmetric partitioning of effector- or memory-associated molecules occurs in activated NK cells and contributes to effector or memory cell differentiation is unknown.

Ultimately, many of the aforementioned questions will benefit from recent advances in technology. For example, molecular studies on memory NK cells have been limited by the very low frequency of these cells in mice, making them difficult to isolate in sufficient numbers for bulk-population techniques such as ChIP-seq and RNA-seq. Thus, single cell sequencing and modified ‘ultra-low-input’ ChIP assays, which are ideal for rare populations, may be particularly useful for studying transcriptional and epigenetic phenomenon in memory NK cells. Single cell platforms also provide a means to evaluate the inherent heterogeneity of the effector and memory cell pool at a single-cell level and, when employed in combination with cell barcoding strategies, can be used to track progeny from individual cells at each phase of the response. By using new technologies to overcome the technical difficulties associated with studying memory NK cells, future work will undoubtedly lead to deeper insights into the transcriptional and epigenetic processes that support adaptive NK immunity.