Epigenetic regulation of natural killer cell memory

Abstract

Immunological memory is the underlying mechanism by which the immune system remembers previous encounters with pathogens to produce an enhanced secondary response upon re-encounter. It stands as the hallmark feature of the adaptive immune system and the cornerstone of vaccine development. Classic recall responses are executed by conventional T and B cells, which undergo somatic recombination and modify their receptor repertoire to ensure recognition of a vast number of antigens. However, recent evidence has challenged the dogma that memory responses are restricted to the adaptive immune system, which has prompted a reevaluation of what delineates “immune memory.” Natural killer (NK) cells of the innate immune system have been at the forefront of these pushed boundaries, and have proved to be more “adaptable” than previously thought. Like T cells, we now appreciate that their “natural” abilities actually require a myriad of signals for optimal responses. In this review, we discuss the many signals required for effector and memory NK cell responses and the epigenetic mechanisms that ultimately endow their enhanced features.

1 ∣. INTRODUCTION

Natural killer (NK) cells are cytotoxic lymphocytes capable of rapidly responding to pathogens and tumors. Their most canonical responses have been studied and extensively defined during infection with viruses, where NK cells contribute an essential role in early innate defense.^1-3 Such functions most reputably include the release of cytolytic granules that contain the pore-forming protein perforin and proteolytic granzymes, as well as the secretion of the key antiviral cytokine interferon-gamma (IFN-γ).⁴

Whereas T and B cells constitute adaptive lymphocytes, NK cells are often presented as the prototypical innate lymphocyte, and vacillate between the innate and adaptive immune system in a context-dependent manner. Across different pathogens, NK cells can exhibit various adaptive qualities, including antigen specificity, clonal expansion, contraction, enhanced effector responses upon a second encounter, and longevity relative to their usual turnover of about 2 weeks^5-7 (Figure 1, top). These adaptive features are manifested most clearly during specific viral infections, but traces of NK cell memory have been discovered in other contexts. Although NK cell activity was originally described as “spontaneous,”⁸ we now appreciate that NK cells actually require a plethora of signals to properly prime them for their peak effector potential. For the generation of effector and memory NK cells, these signals draw many parallels to T-cell signaling and can be broadly categorized into the canonical cues required for proper T-cell activation: antigen-receptor signaling (signal 1), costimulation (signal 2), and cytokine signaling (signal 3). In some cases, these signals initiate a cascade of events that eventually lead to long-lived potential for enhanced responses. Given the inability of NK cells to undergo somatic rearrangement, retention of their enhanced capabilities and longevity in lineage-committed cells are likely endowed through epigenetic mechanisms that can be sustained or passed onto progeny.

FIGURE 1.

Stable changes underlie epigenetic reprogramming during lymphocyte memory responses. Top panel shows schematic of NK and CD8 T-cell memory responses over time, highlighting stages of population expansion, contraction, and memory formation. Bottom panel depicts a schematic of transient vs stable epigenetic changes (including DNA methylation, histone modifications, and chromatin accessibility) that may be associated with indicated stages. Not depicted are potential transcription factors and histone modifiers during the memory phase that are retained at open chromatin regions. These proteins may either actively maintain the open chromatin state, and/or are recruited so that they are poised to enable rapid transcription. Note that these epigenetic changes are not limited to open conformations and may also lead to transient and stable closed conformations

This review provides an overview of our current understanding of how these exogenous signals affect the epigenetic landscape of NK cells. We will first summarize the most well-studied models of NK cell memory and their epigenetic features, highlighting the multiple molecular signals involved. Then, we delve deeper into these different pathways of signal transduction, and describe how these signals affect the epigenetic landscape. Ultimately, elucidating the epigenetic strategies that NK cells utilize to stably augment their potential will pave the way for opportunities to exploit these mechanisms for therapeutic use.

2 ∣. DEFINING “EPIGENETICS” IN THE CONTEXT OF NON-STEADY-STATE CONDITIONS

Since its coinage by Conrad Waddington, the definition of “epigenetics” has evolved through several iterations, not without controversy.⁹ For the purposes of this review, we utilize the broader term of epigenetics to include a change underlying the state of expression of a gene that does not arise from an altered genetic code. These epigenetic changes can thus be found in many forms, from precise changes on DNA to global topological changes in chromatin structure and accessibility. Below, we describe some of the common epigenetic features utilized to define epigenetic regulation:

2.1 ∣. DNA methylation

Perhaps one of the oldest and most well-studied epigenetic “mark” is DNA methylation. Deposition of methyl groups on the 5-carbon of cysteine (5-mC) are mediated by DNA methyltransferases and occurs under two contexts: (1) de novo methylation, which is mediated by DNMT3a and DNMT3b, and (2) maintenance methylation, which is mediated by DNMT1 and responsible for postreplicative inheritance. Active removal of DNA methyl groups is initiated by Ten-eleven translocation (TET) proteins. TET proteins catalyze 5-mC to 5-hydroxymethylcytosine, which is often found at active enhancer sites.^10,11 While a majority of DNA methylation occurs at CpG sites, a smaller but substantial proportion does not. Generally, DNA methylation is associated with gene repression; however, some genes are both highly methylated and highly transcribed, suggesting that DNA methylation may play a role in promoting transcription.¹² In lymphocytes, the impact of DNA methylation has been well-documented to impact homeostasis, effector function, longevity, and even exhaustion in settings of host immunity against infection and cancer.¹³

2.2 ∣. Histone modifications

Histone modifications are covalent, post-translational modifications found on the tails of histones.¹⁴ An array of molecular modifications has been described, which includes acetylation, methylation, phosphorylation, ubiquitinylation, sumoylation, ADP-ribosylation, deamination, propionylation, and butyrylation.¹⁴ Among the most commonly described are histone acetylation and methylation, which are deposited and removed via histone modifying enzymes known as “writers” and “erasers,” respectively. In general, histone acetylation is associated with gene activation with recruitment of histone acetyltransferases promoting transcription. In contrast, the recruitment of histone deacetylases (HDACs) generally represses transcription. Histone methylation can be either permissive or repressive, and can be altered by the activity of histone methylases and demethylases. Some of the most common histone marks include H3K4me3 (which marks active transcription and is enriched at promoter regions), H3K4me1, and H3K27Ac (which are concentrated at enhancer regions), and H3K27me3 (which is a repressive mark).^11,15 In lymphocytes, these global histone modifications can direct the expression of specific genes critical to their development and effector function.¹⁶

2.3 ∣. Chromatin accessibility

Although not an epigenetic “mark” per se, chromatin accessibility is a reflection and consequence of the combinatorial activity of epigenetic modifications and output. High-throughput sequencing paired with global enzymatic cleavage of exposed DNA as a result of chromatin decompaction has become a powerful tool to globally assay the epigenome in different cell types and cell states. Among the most commonly used is Assay for Transposase-Accessible Chromatin using sequencing, or ATAC-seq, which incorporates the more sensitive Tn5 transposase technology that both cleaves and tags open chromatin regions.¹⁷ Using this method, many studies are able to reproducibly define genome-wide regions of accessibility, revealing precise putative enhancer regions and areas of active epigenetic activity. Within the innate and adaptive immune system, such chromatin remodeling leading to accessibility during cellular activation controls both fate and function of immune cells.^18,19

In lymphocytes responding to an exogenous stimulus, we noted that these epigenetic changes can be further distinguished as either transient or stable.²⁰ Transient epigenetic changes return back to baseline in the absence of this stimulus. Stable epigenetic changes, on the other hand, differ from baseline and are maintained long term in the absence of this stimulus, either through the longevity of the cell, or through inheritance to its cell progeny (Figure 1, bottom). We believe the latter are reflective of true cellular reprogramming, and likely underlie the processes that endow immunological memory.²¹ Therefore, while epigenetic reprogramming certainly requires epigenetic changes, epigenetic change does not denote epigenetic reprogramming unless it is sustained. This review aims to highlight literature that encompasses these epigenetic changes in lymphocytes regardless of whether long-term maintenance has been formally tested since ultimately these epigenetic changes have the potential to lead to reprogramming.

3 ∣. EPIGENETIC FEATURES OF MEMORY NK CELL RESPONSES

Memory NK cells have been described in a wide array of non-pathogenic and pathogenic contexts, most notably in viral infections, but also extending to bacterial and parasitic infections. Although these responses have been well-documented,²² deeper studies delving into the mechanistic consequences of these pathogenic challenges are still in their infancy. Therefore, we will focus on the memory NK cell responses that have been described to undergo some form of epigenetic modulation.

3.1 ∣. Memory responses to cytomegalovirus

Most of our understanding of how memory NK cells form has been drawn from studies on cytomegalovirus (CMV) infection in mice and humans. CMV is a double-stranded DNA beta herpesvirus that infects approximately 40% to 90% of the human population, depending on country.²³ Humans infected with CMV are largely asymptomatic, but infection can lead to adverse outcomes in immunocompromised settings. Several aspects of human CMV (HCMV) can be recapitulated in the mouse system using mouse CMV (MCMV), which has proven to be instrumental in unveiling many mechanisms of host responses.²⁴ CMV exposure can initiate an acute systemic infection, affecting such organs as the spleen, liver, lung, and salivary glands. Upon resolution, CMV becomes latent in most peripheral organs where virions cannot be detected, with the exception of the salivary gland, where it becomes locally persistent.²³

Early observations that host resistance to MCMV varies across mouse strains later led to the discovery that proliferating NK cells bearing the activating receptor Ly49H were crucial to control viral infection.^25-29 Shortly thereafter, the MCMV viral glycoprotein m157 was identified as the ligand for Ly49H that activates NK cells to kill and produce IFN-γ,^30,31 indicating that NK cells could respond to MCMV in an antigen-specific manner. Based on these seminal findings, Ly49H⁺ NK cells were then shown to exhibit several key qualities of immunological memory, including effector cell expansion and contraction, and formation of a long-lived pool of memory cells more protective on a per-cell basis against a secondary MCMV encounter.^32,33 Importantly, Ly49H⁺ NK cells underwent extensive epigenetic remodeling over time, as assessed by chromatin accessibility²⁰ and H3K4me3 signal.³⁴ In both CMV-specific NK and CD8⁺ T cells, some of these changes were transient and mostly peaked during early infection in coordination with transcriptional changes.²⁰ Other changes were stable, resulting in both open and closed chromatin configurations at certain loci in the memory NK and CD8⁺ T cells.²⁰ In the differentiating NK cells, most of the epigenetic rewiring that was detected had occurred by day 14 after infection, suggesting the programming of NK cell memory may be occurring earlier than previously thought. Whether such kinetics are similar for T cells remains to be determined, as early transcriptomic studies suggested optimal T-cell memory is not established until at least a month after initial priming.³⁵ Given distinct models of lymphocyte differentiation where subsets of early effector cells may possess a high degree of heterogeneity during clonal expansion,³⁶ including precursor subsets poised to give rise to the eventual memory population, future single cell ATAC-seq studies on T and NK cells will reveal whether the epigenetic poising occurs even earlier than day 14, or even day 7 following infection. Finally, these changes that resulted in memory NK cells, which were epigenetically distinct from their naive Ly49H⁺ counterparts, persisted for at least a month after infection. Among regions that were significantly modulated during infection, naive NK cells most resembled memory CD8⁺ T cells; however, upon differentiation during viral infection, NK and CD8⁺ T cells followed a similar epigenetic trajectory relative to their respective naive states, which revealed a common transcriptional and epigenetic signature.²⁰

Soon after researchers established an MCMV antigen for the mouse Ly49H receptor, a similarly expanded population of human NK cells expressing the activating receptor NKG2C was reported in HCMV seropositive individuals, but not in those that were seropositive for other herpesviruses.³⁷ Hypotheses that this population represented a human “adaptive” NK cell corollary to the mouse Ly49H⁺ NK cell were eventually validated by subsequent studies reporting the expansion of NKG2C⁺ NK cells in response to HCMV reactivation following solid-organ³⁸ and hematopoietic^39,40 transplants. Importantly, these NKG2C⁺ NK cells exhibited enhanced IFN-γ production, which was retained upon transplantation into new patients⁴¹ and could recognize UL40 peptides derived from HCMV through NKG2C.⁴² Global methylome and chromatin accessibility profiling revealed that these adaptive NKG2C⁺ NK cells displayed distinct epigenetic profiles compared to conventional NK cells, and displayed parallel epigenetic features as observed in differentiating CD8⁺ T cells.^43-46 Overall methylation status was correlated with transcriptional output, with hypermethylated regions associated with reduced transcription, and hypomethylated regions associated with increased transcription.⁴⁴ Epigenetic features of HCMV adaptive NK cells include the hypermethylation of genes encoding for activating-receptor signaling molecules, such as tyrosine kinase SYK, adaptor molecules EAT-2 and FcεRIγ, as well as transcription factor PLZF, with associated reduction in expression.^43,44,47 Silencing of PLZF driven by BCL11B activity is believed to contribute to the developmental trajectory of these adaptive NK cells.^44,45 Of note, while adaptive NKG2C⁺ NK cells are prevalent among HCMV seropositive individuals, a small population of those who are genetically deficient for NKG2C were described to also possess NK cells with memory-like properties akin to NKG2C⁺ adaptive NK cells,⁴⁸ suggesting that NKG2C may not be absolutely required. Consistent with this latter population, adaptive NK cells can be alternatively defined independent of NKG2C status by their lack of FcεRIγ, which mark a distinct but overlapping subset that elicits increased antibody-mediated effector functions.^43,49,50 Nevertheless, a common epigenetic characteristic to both adaptive NK cell populations is the hypomethylation of the IFNG locus,^46,48 particularly at a conserved non-coding region upstream of the promoter that is shared with Th1 cells. This enhancer region was shown to be fully functional to promote transcription in response to NKG2C engagement with other activating-receptor engagement, and methylation was sufficient to dampen transcription.⁴⁶ Overall, the environmental cues generated by CMV infection provide a conducive environment for NK cells to acquire robust, long-lived memory features in both mice and humans.

3.2 ∣. Memory responses to contact hypersensitivity

Pioneering studies describing the role of NK cells during experimental hapten-induced contact hypersensitivity provided the first convincing evidence of immune memory within the NK cell compartment. In these studies, researchers observed that Rag2^−/− mice, which lack B cells and T cells, could elicit substantial contact hypersensitivity upon a secondary challenge with the same hapten, but not a different hapten.⁵¹ This response persisted for weeks after initial challenge, and was mediated by NK cells localized to the liver but not the spleen.⁵¹ Transfer of liver NK cells expressing NK cell inhibitory receptors Ly49C/Ly49I elicited a greater response than those that lacked the receptor, suggesting that hapten-specific memory NK cells were concentrated within this subset. Furthermore, blocking signaling through the activating receptor NKG2D significantly suppressed this response.⁵¹

In light of this, a recent study has described a potential human liver-resident counterpart to these hapten-specific responses.⁵² In this study, researchers described a CD49a⁺ CD16⁻ NK cell population that was distinct from more conventional CD49a⁻ CD16⁺ cells. Remarkably, these CD49a⁺ CD16⁻ NK cells in both liver and skin showed specific lytic activity after either an in vivo challenge with nickel, but not cobalt, in patients sensitized to nickel, or those that had a medical history of contact allergy with nickel. Chromatin accessibility profiles comparing conventional CD49a⁻ CD16⁺ cells to these CD49a⁺ CD16⁻ demonstrated chromatin remodeling at genes that may increase effector potential.⁵² Thus, NK cells primed through non-pathogenic means may also undergo epigenetic changes that underlie enhanced effector function. However, the extent and durability of these epigenetic changes induced in non-pathogenic settings compared to CMV infection remain to be determined.

3.3 ∣. Memory responses to inflammatory cytokines

Inflammatory cytokine exposure represents another setting in which memory-like NK cells with enhanced potential to generate IFN-γ can be generated. With the ex vivo administration of proinflammatory cytokines IL-12 and IL-18 along with homeostatic cytokine IL-15, NK cells transferred into naive hosts were able to retain the ability to produce more IFN-γ in response to a secondary cytokine stimulation or activating-receptor engagement.^53,54 These cytokine-induced memory-like (CIML) NK cells were functionally superior in protecting mice against tumor growth compared to those cultured in IL-15 alone, and importantly had similar enhanced functions in human NK cells.^55-57 Thus, these CIML NK cells have been introduced into the clinic as a promising form of cellular immunotherapy.^58,59 Similar to observations in CMV-induced adaptive NK cells, CIML cells display substantial and stable (for at least 10 days) demethylation near the IFNG region, suggesting that CIML cells have been epigenetically reprogrammed.⁶⁰

More recently, memory-like NK cell responses have been described in models of endotoxemia induced by the in-vivo administration of LPS.⁶¹ In this model, systemic inflammation drives the formation of memory-like NK cells that acquire similar features as those described with CIML NK cells. These NK cells persist up to 9 weeks in vivo, produce greater IFN-γ upon restimulation, and can be moderately protective against a bacterial challenge with E. coli. The increased IFN-γ production was associated with H3K4me1 deposition at a distal enhancer region of the Ifng locus, and systemic inhibition of methyltransferase activity abrogated the enhanced capability for NK cells to produce IFN-γ upon restimulation.⁶¹ Thus, whether in the in-vitro settings of specific cytokine stimulation, or the in-vivo bystander exposure to inflammation, epigenetic programming of NK cells for greater effector function may represent a valuable therapeutic intervention in the treatment of cancer and infectious diseases.

3.4 ∣. Memory in response to other non-CMV pathogens

Recently, a developmentally distinct subset of CD56^hi CD94⁺ NK cells were described in HIV patients with enhanced function.⁶² These cells expressed high levels of TCF7 and were epigenetically distinct compared to either CD94⁻ CD56^dim or CD94⁺ CD56^dim cells, as assayed by H3K4me1, H3K4me3, and chromatin accessibility. Most of these epigenetic readouts were increased in signal or accessibility, which included the TCF7 locus itself as well as other lymphocyte memory-associated genes. Motif analysis of the accessible regions CD56^hi CD94⁺ indicated an enrichment of TCF7, RUNX, NK-κB motifs.⁶² Because HIV⁺ individuals are often coinfected by CMV,⁶³ it is unclear what the contribution of HCMV is to the generation of these NK cells, as serostatus was not reported. Although the features described above do not fully align with observations in HCMV, as HCMV adaptive NK cells are low for TCF7 and CD56^dim,^44,45 co-infection may provide additional signals that drive NK cell memory. Furthermore, it is unclear how long-lived this population is and how long these epigenetic features are maintained. Nevertheless, evidence that memory NK cells are present during experimental Simian immunodeficiency virus⁶⁴ support the possibility of an existing CMV-independent, HIV-induced memory NK cell population.

In addition, memory NK cells have been described in mouse models of Zika virus.⁶⁵ Here, researchers found that adoptive transfer of a Zika-experienced CD27⁺ NK cell population harvested after a month postinfection could reduce viremia in susceptible hosts better than their naive CD27⁺ NK cell counterpart. Chromatin accessibility profiles displayed a distinct landscape compared to naive CD27⁺ NK cells and also implicated TCF7 as promoting a stem-like program.⁶⁵ Again, these features slightly differ from memory NK cells described in CMV, as they are TCF7 low and CD27⁻, which defines a more mature phenotype.^20,66 Taken together, memory NK cell responses to pathogens are not restricted to CMV and may vary in their cell-surface phenotype and differentiation.

4 ∣. EPIGENETIC MECHANISMS OF MEMORY NK CELL RESPONSES

As with T-cell activation and memory formation, NK cells require an integration of signals to properly mediate their effector functions and retain their memory potential. While much progress has been made on the nature of these external stimuli and how these signals are transduced, less is known about how these signals and mechanisms may epigenetically rewire the cell for memory potential. T-cell memory responses are associated with substantial chromatin remodeling as assayed by genome-wide histone modifications, DNA methylation, and chromatin accessibility.^67-72 Likewise, as described above, memory and memory-like NK cells can undergo both local and global epigenetic reprogramming,^{20,43-46,52,60} which in some cases parallel the epigenetic trajectories of CD8⁺ T-cell differentiation.^20,44-46 Given the many signaling parallels, we discuss the epigenetic regulation of NK cell activation using the classic 3-signal paradigm described in T cells.

4.1 ∣. Activating-receptor signaling and costimulation (Signals 1 and 2)

Despite their reliance on a fixed set of germline encoded receptors, NK cells exhibit a population diversity and clonal specificity based on the stochastic combinatorial expression of cell-surface receptors, particularly in humans.^73,74 Much of this diversity is driven through inhibitory receptors (Ly49 receptors in mice and killer cell immunoglobulin-like receptors [KIR] receptors in humans) that recognize self-major histocompatibility complex (MHC), which ultimately shape the responsiveness of a given lineage-committed NK cell, a process known as NK cell education.⁷⁵ These inhibitory signals are often countered by activating receptors, which make up a minority of the NK cell receptor repertoire in mice,⁷⁶ and it is believed that the balance between inhibitory and activating-receptor activity determines how responsive an NK cell is to outside stimuli. Surprisingly, a recent study described almost no significant changes in chromatin accessibility between educated and uneducated NK cells,⁴⁵ consistent with an observed lack of transcriptional changes suggesting that differential responsiveness may be regulated at a post-transcriptional level.^77,78

The events following ligand engagement to activating receptors in NK cells draws many parallels with antigen-specific receptor signaling through the T-cell receptor (TCR) and costimulatory signaling, which have been extensively reviewed elsewhere.^79-82 NK cell receptors (NKRs) that relay through immunoreceptor tyrosine-based activation motifs (ITAM) signaling exist as NKR complexes. In the same way that the TCR utilizes its alpha and beta chains for ligand specificity and CD3 subunits for ITAM-mediated signal transduction, NKR ITAM signaling utilizes C-type leptin-like receptors for ligand specificity and pair with ITAM-bearing adaptor subunits. These include such proteins as DAP12 (TYROBP/Tyrobp, also known as KARAP), FcεRIγ (FCER1G/Fcer1g), and CD3ζ (CD247/Cd247). While the TCR complex has 10 ITAMs across all of its subunits, these aforementioned adaptor proteins exist as dimers and can harbor up to six ITAMs per dimer (DAP12 and FcεRIγ monomers have one ITAM, while CD3ζ monomers have three). These adaptors are required for stable expression of their associated activating receptor on the cell surface.⁸³ ITAM signaling, however, is not the exclusive mechanism of transduction, as other NKRs, exclusively or in combination, utilize adaptor protein DAP10, which instead harbors a YxNM motif commonly found in costimulatory receptors utilized by T cells, such as the prototypical costimulatory molecule CD28. An overview of signal transduction induced by activating-receptor signaling is summarized in Figure 2. Although other activating-receptor signaling has been excellently reviewed elsewhere,^84,85 here we restrict our review to the signaling pathways induced by several key activating receptors involved in NK cell memory responses and look toward parallels in other cell types to guide our understanding on how these signaling pathways may relay epigenetic reprogramming in NK cells.

FIGURE 2.

NK cell receptor signaling. Dotted arrows indicate signal transduction in which other signals are required. Green arrows demarcate molecules that have been described to play direct roles in regulating the chromatin landscape. Left pathway shows activating ITAM signaling as represented through Ly49H and DAP12. Activated Src-family kinases, or tyrosine kinase SYK,^203-205 can phosphorylate ITAM tyrosine residues to initiate downstream signaling. Upon phosphorylation, these ITAM domains serve as docking sites to ZAP-70 and/or SYK.^84,206 After docking, activated SYK/ZAP-70 can further activate PI3K, allowing for activation of serine/threonine kinase AKT, and conversion of phosphatidylinositol-3,4-bisphosphate (PIP₂) to phosphatidylinositol-3,4,5-trisphosphate (PIP₃).^85,206,207 Syk also phosphorylates membrane-bound adaptor proteins LAT1²⁰⁸ and/or LAT2,^209,210 adaptor protein SLP-76,^205,211 and Vav proteins.²²⁸ SLP-76 binds to LAT1/2 via GRB2-related adaptor protein (GADS), thereby acting as a platform to recruit Vav proteins and ITK.²¹² ITK then activates SLP-76-bound PLC-γ, which becomes activated and hydrolyzes PIP₂ into inositol 1,4,5-tripohosphate (IP₃) and diacylglycerol (DAG).²¹³ IP₃ promotes calcium flux, which ultimately activates the calcium-sensitive transcription factor NFAT, which can associate with AP-1 transcription factors,²¹⁴ while DAG leads to activation of protein kinase C (PKC), in turn activates the Carma1-BCL10-Malt1 (CBM) complex and thus NF-κB.^215-217 Of note, PKC-θ specifically affects ERK, JNK, and NFAT phosphorylation, but does not affect NF-κB or p38 activation.¹⁰⁷ In addition to NF-κB, JNK and p38 MAP kinases can also be activated by some components of the CBM complex^216,217 to ultimately activate AP-1 factors.^218,219 PI3K can also promote activation of ERK via Rac1 -> PAK -> MEK²⁰⁷ likely through Vav proteins.²²⁰ Alternatively, ERK can be activated by DAG-activated RASGRP1, which activates the RAS -> RAF1 -> MEK -> ERK pathway.^221,222 Right pathway shows activating YxNM signaling as represented through NKG2D and DAP10. Signaling through NKG2D occurs independent of SYK, ZAP-70, and LAT.²²³ Both the p85 subunit of PI3K and GRB2 can bind to the YxNM motif of DAP10.^224-226 GRB2 exists as a complex with VAV1, leading to its activation and subsequent phosphorylation of PLC-γ and SLP-76.²²⁶ PLC-γ can then induce calcium signaling and promote JNK activation.²²⁷ RhoGTPases RHOA and RAC1 are likely activated by VAV1.^220,223,228 NKG2D engagement leads to activation of ERK1/2²²⁹ and AKT and via PI3K,²²⁴ where AKT activity is dependent on VAV1 in mice.²³⁰ Both signaling through PI3K and the GRB2-VAV1 complex are required for full calcium release and cytotoxicity.²²⁶ Both VAV1 -> ERK (perhaps through RAC1) and PI3K -> AKT can lead to NF-κB phosphorylation in the presence of other activating-receptor signals²³¹

4.1.1 ∣. Activating and costimulatory receptors involved in memory responses

Ly49 receptors are homodimeric receptors expressed in mice that generally recognize class I MHC complexes in a peptide-dependent but not peptide-specific manner.^76,86 In humans, KIRs serve as functional analogs to Ly49 receptors, as both play important roles in NK cell education.⁷⁵ More often than not, receptor ligation leads to inhibitory signaling; however, notable exceptions of non-self-ligand recognition induce activating signaling that underlies some of the NK cell memory responses described in mice. In C57BL/6 mice, the only expressed activating Ly49 receptors are Ly49D and Ly49H, and both have been demonstrated to facilitate memory features in response to alloantigen⁸⁷ and MCMV,³² respectively. Ly49D and Ly49H both signal through DAP12 and DAP10, and both adaptors contribute to optimal NK cell responses in vivo.^83,88,89 Nevertheless, based on the more pronounced reduction of cell-surface receptor expression, activating Ly49-DAP12 complexes may be more stable and preferential.⁸⁸

The classic NK cell identification marker NK1.1 (NKR-P1C encoded by Klrb1c) pairs with FcεRIγ⁸⁴ and has been recently demonstrated to recognize the viral glycoprotein m12 encoded by MCMV.⁹⁰ Given that NK1.1 is ubiquitously expressed on all NK cells and ILC1, a recent study suggested that an NK1.1-m12 interaction drives a memory response in liver-resident ILC1, highlighting an additional innate lymphocyte capable of possessing adaptive features.⁹¹ Although ILC1 have previously been shown to be epigenetically poised at specific loci (e.g. Il12rb1 and Ifng) during development to a greater extent than NK cells and CD8⁺ T cells,⁹² ILC1 can continue to undergo changes in their epigenetic landscape following viral infection leading to memory.⁹¹

Activating receptor NKG2C exists as a disulfide-linked heterodimer with glycoprotein CD94, which is also shared with inhibitory receptor NKG2A. NKG2C/CD94 complexes pair with DAP12 and recognize HLA-E in humans^93,94 or H2-Qa1 in mice,⁹⁵ which are non-classical MHC class I molecules that bind to peptides derived from the leader segments of other classical MHC class I proteins.^96,97 While expressed on mouse NK cells, mouse NKG2C does not seem to be required for the development or function of NK cells in response to different pathogens, as NK cells deficient for CD94 develop and respond normally.⁹⁸ In contrast, human NKG2C has been a subject of intense research within human NK cells as a marker of adaptive NK cells in HCMV. Previous work identified UL40 as an HLA-E mimetic derived from HCMV that could signal through NKG2A/CD94.^99-102 Recent studies confirmed that the NKG2C receptor could also recognize UL40 peptides derived from HCMV, finally validating an HCMV antigen for NKG2C.⁴² Signaling through this peptide (along with costimulation and cytokine regulation) recapitulated many of the transcriptional features described above, such as downregulation of FcεRIγ, SYK, and EAT-2, as well as a key epigenetic feature of hypomethylation of the IFNG locus.⁴²

As alluded to before, NKG2C may not be the only activating receptor that adaptive NK cells utilize for their enhanced memory responses. NK cells deficient in adaptor protein FcεRIγ have been shown to have enhanced responses to antibody-mediated cellular cytotoxicity (ADCC), one of the key functions executed by NK cells and the prime reason why monoclonal antibody-mediated cancer therapy has been successful. In NK cells, this process is largely mediated through cell-surface receptor FcγRIII, also known as CD16. CD16 in mice exclusively utilizes FcεRIγ as its paired adaptor; however in humans, CD16 can pair with both FcεRIγ as well as CD3ζ. Thus, in the absence of FcεRIγ, it is thought that CD16 exclusively utilizes CD3ζ, which harbors more ITAMs and thus strengthens signaling.⁴⁹ Interestingly, CD16 preactivation with a high affinity antibody on human NK cells can lead to the enhanced effector function in response to cytokine stimulation or tumor cells. The effects of this preactivation were observed after 5 days of culture, and required IL-2 to maintain the enhanced responses, which is somewhat reminiscent of CIML responses.¹⁰³ Furthermore, CD16A engagement in the presence of IL-12/18/15 improved tumor killing against CD30+ tumors,¹⁰⁴ further highlighting the cooperativity between activating receptor and cytokine receptor signaling. Whether this cooperativity rewires the epigenetic landscape long-term remains to be determined.

DNAX accessory molecule-1 (DNAM-1, also known as CD226) is a costimulatory cell-surface glycoprotein expressed in NK cells as well as CD8⁺ T cells that recognizes poliovirus receptor (PVR, CD155), as well as Nectin-2 (CD112). Antigen-specific NK cell responses during MCMV depended on DNAM-1 signaling to not only proliferate during early infection, but also to generate and/or maintain the memory NK cell pool. Furthermore, lack of this costimulation abrogated memory recall responses and increased viral load.¹⁰⁵ DNAM-1 does not require additional adaptor molecules, as it harbors an immunoreceptor tyrosine tail-like motif and can signal via the recruitment of GRB2, resulting in VAV1, PI3K-AKT, ERK and PLC-γ1 activation.¹⁰⁶ Further studies implicated protein kinase C eta (PKC-η) as a contributing component for both the expansion, generation and recall responses of Ly49H⁺ memory NK cells, while PKC-θ, which is typically downstream of ITAM signaling, was dispensable for control of early MCMV infection.^105,107 Despite the fact that Src-family kinase Fyn was demonstrated to not be required for initial NK cell activation,¹⁰⁶ Fyn was implicated in vivo to play a specific role in late-stage memory NK cell formation,¹⁰⁵ highlighting unique molecular requirements for the persistence of memory in NK cells.

CD2 is a costimulatory and adhesion molecule that promotes immunological synapse formation, actin cytoskeleton rearrangement, and TCR activation.¹⁰⁸ CD2 signaling depends on CD3ζ and requires Src-family kinases Lck and Fyn. CD2 also participates in Lck-mediated RAS-MAPK signaling and in downstream PLC-γ signaling, which includes nuclear factor of activated T cells (NFAT) and JNK activation.¹⁰⁸ In the context of HCMV, CD2 was implicated in playing a critical role in promoting the enhanced IFN-γ production seen in adaptive NK cells in response to CD16 signaling, agnostic to NKG2C expression on the NK cells.^48,109 Consistent with this, CD2 expression was increased in adaptive NK cells,^43,44 and its locus became hypomethylated.⁴⁴ Blocking CD2 or its ligand LFA-3 (also known as CD58) abrogated the enhanced IFN-γ production in response to antibody triggering.¹⁰⁹ Furthermore, CD2 was also implicated in enhancing NKG2C signaling via weaker UL40 peptides.⁴²

Finally, like T cells, chronic stimulation through activating-receptor signaling may lead to exhaustion in NK cells, which may lead to long-term reprogramming. Using NKG2C⁺ NK cells from CMV seropositive individuals, constant stimulation using antibodies that engaged NKG2A/C or NKG2C for 7 days in culture generated NK cells that had high proliferative potential concomitant with decreased IFN-γ production against tumor targets. Genome-wide profiling on these dysfunctional NK cells demonstrated DNA hypomethylation at many sites previously shown also to be epigenetically modified in exhausted T cells,¹¹⁰ including genes encoding for inhibitory checkpoint molecules LAG3, PDCD1, and TIGIT. Thus, continuous activating signals can result in long-term epigenetic effects and may underlie settings where NK cell responses fail to control viral pathogenesis or tumorigenesis.

4.1.2 ∣. Epigenetic regulation induced by activating receptor and costimulatory signaling

NKR signaling initiates a cascade of events that ultimately activate a number of important molecules that govern the direct interface between extracellular signaling and epigenetic modifications. Signaling across different receptors converges to activate at least one of three key transcription factors in NK cells: NFAT, AP-1, and NF-κB. A dearth of studies exist for how NKR signaling translates into long-term epigenetic changes via these transcription factors; however in T cells, the epigenetic regulation induced by these transcription factors has been better-characterized and can shed light on the potential of conserved epigenetic mechanisms. Based on these T cells studies, these transcription factors alone or collectively promote significant modifications to the epigenetic landscape, providing a stable backdrop for differentiation.

The NFAT family of transcription factors consists of five members, encoded by the following genes: NFATC1 (NFATc or NFAT2), NFATC2 (NFATp or NFAT1), NFATC3 (NFAT4), NFATC4 (NFAT3), and NFAT5 (NFAT5).¹¹¹ All members (except NFAT5) depend on the classic signaling, where IP₃ induces the release of calcium from intracellular stores, which further triggers calcium release through the plasma membrane to maintain intracellular calcium levels. Calcium then binds to the intermediate calcium sensor calmodulin, which in turn activates the calmodulin-dependent phosphatase calcineurin. NFAT proteins are dephosphorylated by calcineurin, thereby promoting their nuclear translocation and transcriptional activity.¹¹¹ NFAT has been previously shown to induce permissive histone marks by recruiting the histone acetyltransferases p300 and CREB binding protein (CBP) to target genes.¹¹² During Th1 and Th2 differentiation, NFAT binding to target genes in response to TCR activation is believed to be generally non-specific. But in coordination with lineage-defining cytokines that shape the permissive landscape, calcineurin/NFAT can promote a feed-forward loop in which specification is amplified in a loci/cell-type specific manner to maintain histone acetylation and open chromatin accessibility.^113-115 In the case of Th17 development, an alternate model has been proposed where NFAT is the initial primer of the epigenetic permissive landscape assayed by H3K4me3, H4 hyperacetylation, and H3K27Ac, and in coordination with NF-κB, can promote expression of RORγt.¹¹⁶

Of the many binding partners of NFAT, the most notable is its cooperative interaction with AP-1 factors. The AP-1 family consists of several members that broadly fall into four subfamilies: FOS, JUN, ATF, and Maf. Downstream signaling via ITAM receptor engagement primarily activates MAP kinases ERK, JNK, and p38, which in turn activate AP-1 factors. The AP-1 transcription factor complex exists as hetero- or homodimers of AP-1 family members, with Fos and Jun combinations exhibiting potent transactivation potential.¹¹⁷ During T-cell activation, NFAT can bind to DNA either as a complex with AP-1 or before AP-1 recruitment and promote chromatin accessibility.^118,119 Following recruitment and initial nucleosome repositioning, epigenetic modifiers like the SWI/SNF chromatin remodeling complex Brg1 or CBP may bind to further modify the local epigenetic landscape to promote transcription.^112,119 More recently, the cooperation between NFAT and AP-1 factors has received increasing attention due to the fact that NFAT signaling in the absence of AP-1 cooperation can drive T-cell exhaustion,¹²⁰ leading several groups to report widespread epigenetic reprogramming in these exhausted T cells driven by the transcription factor Tox.^121-125

The NF-κB family of transcription factors comprises five members: NF-κB1 (also named p50), NF-κB2 (also named p52), RelA (also named p65), RelB and c-Rel. NF-κB1 and NF-κB2 can exist as precursor proteins p105 and p100 that are processed into their active forms p50 and p52, respectively. NF-κB family members are found in both hetero- and homodimeric forms, and are sequestered in the cytoplasm by IκB proteins until activation. NF-κB signaling falls into two major pathways: the canonical and non-canonical pathway. ITAM signaling is believed to primarily operate through the canonical NF-κB pathway, which consists of proteasomal degradation of IκB proteins triggered by the IKK complex (IKKα, IKKβ, NEMO), releasing active dimers (usually p50/RelA and p50/c-Rel) for translocation into the nucleus. On the contrary, non-canonical NF-κB signaling occurs independently of IκB proteins, and instead utilizes NIK and IKKα to initiate the processing of NF-κB2 p100 into p52. This processed protein then pairs with RelB to form p52/RelB dimers as the active form of the NF-κB complex.¹²⁶ Like NFAT signaling, the interplay between the epigenetic landscape and NF-κB activity has been well-documented across many cell types.¹²⁷ Establishing the proper chromatin environment is required to prime NF-κB for DNA binding and initiation of its transcriptional program, and NF-κB itself may in turn direct epigenetic changes to facilitate signaling,¹²⁷ which certainly holds true for ITAM signaling. For instance, epigenetic regulation as assayed by chromatin accessibility and/or permissive histone marks of the GM-CSF and IL-17 loci have been reported to be dependent on NF-κB or upstream activators of NF-κB during T cell activation or TCR-dependent TCR polarization.^128,129 In the case of GM-CSF, NF-κB was capable of recruiting Brg1.¹²⁹ Furthermore, NF-κB subunits have the ability to directly interact with p300 and HDACs.^130-132 Despite the ubiquity of NF-κB signaling within the immune system and ITAM signaling in particular, the direct implications of NF-κB signaling on the activated epigenome and its long-term maintenance are still understudied. TCR-dependent canonical NF-κB signaling is required to maintain the survival of memory T cells, independent of its early activating signals,¹³³ and so it is possible that similar ITAM-mediated signaling occurs via NKRs in NK cells. More recently, NF-κB signatures were uncovered from distinct chromatin accessibility profiles of stem-like CD8⁺ T cells with memory properties that give rise to exhausted T cells during chronic infection, providing a link between NF-κB signaling and long-term epigenetic reprograming during memory responses.¹³⁴

We have previously shown that NF-κB motifs are enriched in regions that become less accessible in memory Ly49H⁺ NK cells, whereas AP-1 factor motifs are enriched in regions that become more accessible in memory NK cells.²⁰ Given the key roles of both NF-κB and AP-1 during activating-receptor signaling, one could speculate that these changes in the memory NK cell chromatin landscape may affect a functional poising in response to antigen-receptor signaling. In fact, we have also shown that AP-1 motifs are enriched among regions that are commonly open in memory NK and memory CD8⁺ T cells, suggesting that epigenetic regulation via AP-1 signaling may be a common feature of memory lymphocyte responses. In agreement, collated transcription factor motif analysis across multiple cell types and stimuli (from ATAC-seq analyses) highlighted enrichment of Fos and Jun “footprints” and suggested AP-1 factors as a universal mediator of “inflammatory memory.”¹³⁵ It should be noted that induction of these transcriptional programs is not exclusive to activating-receptor signaling, as IL-18 also utilizes NF-κB signaling and thus may contribute to this chromatin remodeling (as discussed below). Therefore, in depth studies dissecting out these pathways are warranted.

In addition to transcription factors, other components of activating signaling pathways may have parallel roles in shaping the epigenetic landscape. The phosphatidylinositol 3-kinase (PI3K)/AKT pathway is a major signaling pathway that has pleiotropic effects on a wide variety of cell types.¹³⁶ Mammals express three classes of PI3K enzymes, with class I PI3K as the most well-characterized. The class I enzymes consists of four isoforms of the catalytic domain (p110α, β, γ, and δ encoded by PIK3CA, PIK3CB, PIK3CG, and PIK3CD), and five isoforms of the regulatory domain (p85α, p55α, p50α, encoded by PIK3R1; p85β, PIK3R2; p55γ, PIK3R3). AKT is a serine/threonine kinase that is coupled tightly downstream of PI3K¹³⁶ and can control the activity of many proteins via phosphorylation. AKT has been shown to directly phosphorylate and attenuate the activity of several histone modifiers, including histone methyltransferases KMT2D¹³⁷ and EZH2,¹³⁸ demethylase KDM5A,¹³⁹ and deacetylase p300.¹⁴⁰ AKT can also affect maintenance of DNA methylation by phosphorylating DNMT1, rendering it more stable.¹⁴¹ Interestingly, PKC has emerged from its role as a traditional cytoplasmic signaling molecule to a more novel mediator of chromatin structure. PKC-β has been previously shown to phosphorylate histone H3 at threonine 6 (H3T6) and prevent LSD1 from demethylating H3K4, and thus promoting transcription.¹⁴² In addition, PKC-θ has been shown to localize in the nucleosome and directly associate with chromatin to recruit a complex of Pol II and histone modifiers to regulate chromatin accessibility at promoter sites.¹⁴³ Given the importance of PKC-θ in NKR signaling, it remains to be determined whether PKC-θ may also play these more novel roles in NK cells.

While very little is known about the epigenetic mechanisms employed by costimulatory molecules in NK cells, many parallels can be drawn based on the more extensive knowledge of these mechanisms in T cells. NKG2D is often considered a costimulatory molecule in CD8⁺ T cells, and has previously been demonstrated to be required for CD8⁺ T-cell memory responses.¹⁴⁴ In mouse CD8⁺ T cells, blocking NKG2D in activated cells prior to memory formation led to alterations in several epigenetic modifiers, including Dnmt3a.¹⁴⁵ In addition, CD28 represents a prototypical costimulatory molecule in T cells and its signaling bears much resemblance to NKG2D-DAP10-mediated signaling in that both harbor a YxNM motif in their cytoplasmic domains.⁸² Accordingly, costimulation through CD28 was required to also promote DNA hypomethylation and chromatin accessibility at the IL-2 locus, while TCR signaling alone was insufficient to promote chromatin accessibility.¹⁴⁶ CD28 can also increase histone acetylation at the IL-2 locus¹⁴⁶ and the IL-5 locus in an NF-κB dependent manner.¹⁴⁷ Furthermore, these epigenetic modifications seem to be context-dependent, as CD28 deprivation, in contrast, was sufficient to lead to DNA hypomethylation at Treg-related genes upon conversion from conventional T cells to Tregs in vitro.¹⁴⁸ Moreover, both TCR signaling and costimulation are required for the full activating potential of AP-1 factors.¹⁴⁹ Accordingly, recent global profiling of chromatin accessibility in human CD4⁺ T cells has identified AP-1 factors as a major driving force of early T-cell activation induced by costimulation by CD28.¹⁵⁰ The authors confirmed that a majority of the newly open regions were AP-1 bound, and that interference of AP-1 activity or anergy induced by TCR signaling in the absence of CD28 abrogated these early changes.¹⁵⁰ Overall, many of the studies investigating the epigenetic mechanisms induced by signal 1 and signal 2 in T cells can likely be extrapolated to their innate counterparts, but validation will require further investigation.

4.2 ∣. Cytokine regulation (Signal 3)

Cytokine regulation comprises the third signal for T-cell activation and differentiation, and also plays a crucial role for NK cell development, differentiation, and effector function.¹⁵¹ But unlike T cells, NK cells can respond robustly to many cytokines without the requirement of activating-receptor signaling, highlighting the traditional innate function of NK cells. As such, these cytokines are being exploited in the clinic to enhance NK cell cellular therapy.¹⁵² During NK cell responses, they play key roles in shaping optimal effector function and long-lived memory, and in some cases are sufficient to endow memory-like features, as described above. Many of the known cytokines required for memory formation signal through the conserved Janus kinase-signal transducer and activator of transcription (JAK-STAT) pathway. JAKs are tyrosine kinases that associate with cytoplasmic domains of cytokine receptors.¹⁵³ Upon cytokine recognition, JAKs phosphorylate the tyrosine residues on the receptor, creating docking sites for homo- or heterodimers of STAT proteins. Following tyrosine phosphorylation by JAKs, STAT complexes can translocate into the nucleus and initiate their respective programs. Mammalian systems consist of four JAKs (JAK1, JAK2, JAK3, and TYK3) and seven STATs (STAT1, STAT2, STAT3, STAT4, STAT5A, STAT5B, and STAT6). Different cytokines and cytokine receptors utilize different sets of JAK and STAT proteins, and while certain cytokines have preferential JAK-STAT pairings, these pairings are not exclusive.¹⁵³

4.2.1 ∣. Cytokine signaling during memory responses

The IL-12 receptor (IL-12R) pairs with JAK2 and TYK2 to mediate signaling primarily through STAT4. Given that resting NK cells show the highest transcriptional abundance of STAT4 compared to all the other STAT proteins (Figure 3A), it is not surprising that steady-state NK cells are highly sensitive to IL-12 signaling. Consistent with this, we have shown that Ly49H⁺ NK cells deficient for the IL-12R subunit (encoded by the gene Il12rb2) fail to become long-lived memory NK cells, and are virtually undetectable after 2 weeks of adoptive transfer.¹⁵⁴ IL-12 signaling was required not only for the initial proliferation of antigen-specific NK cells, but also for their memory formation, as MCMV-experienced Il12rb2^−/− Ly49H⁺ NK cells, unlike WT counterparts, were unable to generate a pool of memory NK cells in naive hosts. Accordingly, STAT4-deficiency similarly abrogated the ability for Ly49H⁺ NK cells to form memory. Thus, IL-12 signaling is not only crucial for IFN-γ production,¹⁵⁵ but also the proliferation and formation of memory NK cells.¹⁵⁴ Furthermore, IL-12-STAT4 signaling initiates a cascade of several downstream programs that rely on such molecules as microRNA-155, Zbtb32, Myd88, Runx factors, IRF8, T-bet, and Adrb2.^156-162

FIGURE 3.

Epigenetic regulation of cytokine signaling in NK cells. (A) Bar plots show relative transcript abundance of indicated STAT genes in naive splenic NK cells as measured by transcripts-per-million. (B) Graphical abstract depicts model of pathway crosstalk between cytokine-induced signals in NK cells, and feedback inhibition when all three pathways are activated. (C) Through (E) Genomic tracks show histograms of signal assayed for H3K4me3 ChIP-seq, H3K27Ac ChIP-seq, chromatin accessibility, and indicated STAT ChIP-seq using indicated conditions. Shown are the loci for (C) Stat2, (D) Csf2, and (E) Atf3. Yellow boxes highlight regions of differential signal compared to unstimulated conditions or significant binding. X-axis displays genomic coordinates, while y-axis shows signal as normalized reads scaled to 1x for ChIP-seq or normalized counts calculated by the DESEq2 software²³² for ATAC-seq

Another key family of proinflammatory cytokines shown to shape the NK cell memory response are type I IFNs, consisting of IFN-α and IFN-β. The type I IFN receptor utilizes JAK1 and TYK2, which activate STAT1 and STAT2 to recruit IFN regulatory factor 9 (IRF9), ultimately forming the ISGF3 complex. This complex translocates into the nucleus to bind to interferon-stimulated response element genes and activate their transcription.¹⁵³ Absence of type I IFN signaling via the loss of receptor subunit IFNAR1 also abrogates robust antigen-specific NK cell responses and memory formation after MCMV infection via distinct mechanisms from IL-12 signaling.¹⁶³ IFNAR1-deficient NK cells showed impaired activation and cytotoxicity, as assayed by CD69 and granzyme B expression during MCMV infection. Unlike IL-12R-deficient NK cells, IFNAR1-deficiency did not affect proliferation, but instead rendered NK cells susceptible to NK cell fratricide, leading to a loss of these cells. Absence of STAT1, STAT2, and IRF9 phenocopied many of these impairments, highlighting their non-redundant roles in NK cell memory responses.¹⁶⁴

In addition to cytokine regulation induced by JAK-STAT signaling, other proinflammatory cytokines can influence the optimal response of memory NK cells. Signaling through IL-1-like cytokines IL-18 and IL-33 both have been demonstrated to contribute to the primary proliferative burst during early antigen-specific NK cell responses. In the case of IL-33, memory NK cells still formed despite the early proliferation defect, but IL-33 signaling was again required for proliferation of memory NK cells, which consequently diminished protective recall responses to MCMV.¹⁶⁵ In the case of IL-18, IFN-γ but not granzyme B was selectively impaired in the absence of IL-18-mediated NK cell signaling,¹⁵⁸ in accordance to its role in priming IL-12-induced IFN-γ production.¹⁶⁶ In contrast to IL-33 signaling, absence of IL-18 signaling generated fewer memory cells, and IL-18 signaling was not required for the maintenance nor recall responses of memory NK cells.¹⁵⁸ IL-18 and IL-33 do not require JAK-STAT signaling and instead rely on activation of NK-κB via Myd88 to transduce signaling.¹⁶⁷ In agreement, absence of Myd88 also impairs proliferation and memory formation of Ly49H⁺ NK cells.¹⁵⁸ Furthermore, the Myd88 locus was shown to be a direct target of STAT4, and STAT4 can potently induce its transcription via IL-12,^34,158 revealing a cooperative mechanism to potentiate IL-18 signaling.

As demonstrated by the lack of NK cells in mice deficient for the common gamma chain receptor gene Il2rg, common gamma chain cytokines are essential for the development and maintenance of the NK cell pool during steady-state conditions. IL-15 and IL-2 support the survival and proliferation of NK cells, respectively,^168-171 and are commonly used to maintain both mouse and human NK cells in culture. In addition to homeostatic conditions, these cytokines act through STAT5 to drive NK cell responses during viral infection, with their contribution to optimal memory NK cell responses temporally regulated.¹⁷⁰ Early during MCMV infection, both IL-2 and IL-15 drive the clonal expansion of Ly49H⁺ NK cells, but only IL-15 is required for memory NK cell maintenance after the effector phase.¹⁷⁰ Thus, both proinflammatory and homeostatic cytokines, along with their respective downstream STAT factors, are critical in regulating NK cells throughout their lifespan and responses against various insults and stimuli.

4.2.2 ∣. Epigenetic regulation via cytokine signaling in NK cells

Studies in mice during MCMV infection revealed that signaling pathways that dominated early transcriptional and epigenetic coordination in NK cells were highly enriched for JAK-STAT pathways.²⁰ Given the plethora of signals that NK cells see during infection, recent studies have dissected the contribution of different cytokine combinations that most potently affected the JAK-STAT signaling pathways that memory NK cells relied on.³⁴ Extensive global profiling of NK cells in response to several key combinations of the cytokines mentioned above revealed a complex interplay of cytokine-STAT signaling pathways that enacted distinct modes of epigenetic regulation. The combination of all cytokines best recapitulated what occurred in vivo during early MCMV infection, and all pathways converged to regulate a set of common negative regulators of cytokine signaling, suggesting a negative feedback loop (Figure 3B) to dampen the responses of these potent effector NK cells once they are activated. Among those assayed, the combination of IL-12 + IL-18 stood out as promoting the most changes in chromatin accessibility, compared to IL-2 + IL-15 or to IFN-α. We found that the combination of IL-12 + IL-18 with IL-2 + IL-15 promoted a global transcriptional and epigenetic coordination, where many co-regulated genes were modulated in the same manner (Figure 3B). Interestingly, IL-12, IL-18 and IL-15 all represent cytokines capable of conferring memory-like properties in NK cells,⁵³ suggesting that coordinated efforts of STAT4, STAT5, and NF-κB may play key roles in modifying the epigenetic landscape such that it is conducive for memory responses.

In contrast, type I IFN signaling behaved differently relative to these other cytokine signaling networks. IL-12 + IL-18 appeared to antagonize IFN-α signals (Figure 3B), where more than half of the genes regulated by both IL-12 + IL-18 and IFN-α were expressed in opposite manners. This antagonism was also reflected in vivo where absence of STAT4 signaling during MCMV lead to increases in many type I IFN-inducible genes.²⁰ As represented by the IFN-α-inducible Stat2 gene^34,164 (Figure 3C), we further observed that unlike the other pro-inflammatory and homeostatic cytokines (Figure 3D,E), IFN-α signaling induced very few changes in chromatin accessibility, in contrast to the substantial number of transcriptional changes it induced. Instead, we found that IFN-α and downstream STAT1 induced epigenetic changes at the promoter regions as assayed by the transcription-permissive histone modification H3K4me3, highlighting a divergent mode of epigenetic regulation induced by these distinct proinflammatory cytokine signaling pathways (Figure 3C). Whether these changes reflect transient or stable changes during memory formation have yet to be explored.

A recent study revealed an elegant mechanism in which STAT4 promoted de novo accessibility in IL-12 + IL-2-activated NK cells by recruiting p300 at highly transcribed genes and redeployed lineage-defining factors such as T-bet to carry out their downstream activity in the absence of canonical T-bet binding sites.¹⁷² In this study, highly induced genes in response to IL-12 + IL-2 were concentrated in areas of induced chromatin remodeling as exhibited by changes in the enhancer landscape. STAT4 was required to establish accessibility and thus acted like a pioneer factor for these highly induced regions. Consistent with this principle, we can show that a potential super-enhancer exists near the locus for Csf2, a gene that was one of the most highly expressed genes in response to IL-12 + IL-18, and conserved between mouse and human NK cells.³⁴ Stimulation produced a substantial increase in H3K27Ac signal across the entire locus as well as farther downstream, and was associated with significantly increased accessibility at multiple sites and induced STAT4 binding (Figure 3D). Given its substantial increase in gene expression in human NK cells to the same cytokine conditions, it is possible that a similar epigenetic mechanism at this enhancer may be conserved in humans.

In NK cells, IL-18 induction of IFN-γ is dependent on MyD88 signaling.¹⁶⁶ Interestingly, MyD88 and IκBζ have been shown to be required for complete epigenetic remodeling of “late” genes induced by NF-κB signaling in macrophages.¹⁷³ Of further interest is whether NF-κB acts in response to pre-accessible sites or directly to modify the chromatin landscape. Computational methods have suggested the latter, that NF-κB may work as a pioneer factor.¹⁷⁴ In light of this finding, IL-18 was observed to induce extensive chromatin remodeling in NK cells, as assessed by ATAC-seq shortly after stimulation. These differential sites were highly enriched for NF-κB motifs, whereas IL-12-stimulated NK cells demonstrated chromatin remodeling with sites enriched for STAT motifs.³⁴ Thus, IL-18 via NF-κB may also play a significant role in shaping the epigenetic landscape of NK cells early during infection.

Curiously, although IL-18 signaling is required for optimal antigen-specific NK cell expansion during early MCMV infection, the recall response of memory NK cells was largely unaffected, suggesting that memory NK cells are no longer reliant on IL-18.¹⁵⁸ Consistent with this observation, both mouse and human CMV-specific memory NK cells show reduced IFN-γ production in response to IL-18 or IL-12+IL-18 compared to naive NK cells.^44,48,175 The differential requirement for gene regulation in memory NK cells may also be reflected at an epigenetic level, as NF-κB motifs and cytokine-induced regions (which were primarily driven by IL-12 and IL-18) were enriched in regions that decreased in accessibility.^20,34 This finding may represent a mechanism by which memory NK cells hone in on their activating-receptor specific responses while decreasing their sensitivity to bystander activation via cytokines. Future studies will delineate the precise contribution of these and additional cytokine signals, and their cooperation with both activating and costimulatory signaling.

5 ∣. CHEMICAL TARGETING OF NK CELL EPIGENETICS AND EFFECTOR FUNCTION

Given the observed chromatin remodeling observed in memory NK cells, several studies targeting this epigenetic machinery have provided evidence that such chromatin remodeling is required to shape NK cell effector responses. While some of these enzymes affect NK cell development,^176,177 we will focus on the factors that play active roles in mature NK cell effector function potentially leading to memory.

5.1 ∣. Role of histone methylation on NK cell function

Expression of histone methyltransferases and demethylases become altered in human NK cell lines activated with PMA/ionomycin,¹⁷⁸ suggesting that their active regulation may play a role in NK cell function. Accordingly, treatment with histone methyltransferase EZH2/1 inhibitor UNC1999 led to increased degranulation in response to PMA/ionomycin, while histone demethylase LSD1 inhibitor OG-L002 and histone methyltransferase MLL1 inhibitor MM102 increased IFN-γ and TNF-α production.¹⁷⁸ Furthermore, inhibition of LSD1 complexes decreased viability and impaired NK cell cytotoxicity.¹⁷⁹ In another study, targeting H3K27 demethylases via the chemical inhibitor GSK-J4 or knockdown of demethylase KDM5a/b (JMJD3/UTX) reduced NK cell production of proinflammatory cytokines like IFN-γ, but had no effect on cytotoxicity.¹⁸⁰ Chemical targeting led to extensive alterations in the transcriptome, deregulating several genes associated with activating-receptor signaling. Surprisingly, H3K27 demethylase inhibition did not affect global changes in chromatin accessibility, but did increase global repressive H3K27me3 levels and slightly increased global H3K4me3, regardless of transcriptional activation or repression.¹⁸⁰ Specific changes were seen at the Ifng locus, where increases in H3K27me3 were also observed, in alignment with its transcriptional repression.¹⁸⁰ Consistently, Kdm5a deficiency led to decreased IFN-γ production and was associated with decreased STAT4 activation via increased SOCS1 expression. In this setting, NF-κB subunit p50 was required to recruit Kdm5a to the Socs1 locus, in order to keep the gene in a repressive state.¹⁸¹

5.2 ∣. Role of histone acetylation on NK cell function

Since their initial FDA approval, HDAC inhibitors have been actively used in the clinic for various diseases including cancer; however, their pleiotropic effects warrant a close examination of their effect on the immune system.¹⁸² Administration of histone deacetylase inhibitor entinostat has been shown to increase lytic potential via upregulation of NKG2D on NK cells, as well as their ligands on tumors, and is associated with epigenetic upregulation of genes related to an IFIT1-STING-STAT4 pathway.¹⁸³ Consistent with this, pretreatment of NK cells with entinostat was able to augment ADCC against prostate and non-small cell lung carcinoma.¹⁸⁴ At the same time, administration of dexamethasone, a synthetic glucocorticoid that affects histone acetylation, also affected NK cell effector function in a dose dependent manner. In high, acute concentrations, NK cell lines exhibited decreased cytokine production and cytotoxicity, which could be rescued with HDAC inhibition.¹⁸⁵ At lower concentrations and continuous treatment, NK cells cultured in dexamethasone exhibited reduced cytotoxic capabilities with increased cytokine-producing capabilities. This modulation in function was associated with increased histone acetylation and chromatin accessibility at the IFNG and IL6 loci. Reciprocally, reduced histone acetylation was shown for GZMB and PRF1, corresponding with reduced transcripts.¹⁸⁶

5.3 ∣. Role of DNA methylation on NK cell function

DNA hypomethylating agents, such as decitabine and azacitidine, are cytidine analogs that incorporate into DNA or RNA and cause inhibitory effects on DNA methyltransferases.¹⁸⁷ Like HDAC inhibitors, their effects are widespread, and previous studies have shown that they have variable effects on NK cell function. In some studies, administration of azacitidine impaired NK cell cytotoxicity and IFN-γ production, while decitabine did the opposite depending on concentration.^188-190 On the other hand, others showed that treatment could lead to changes in the KIR-repertoire along with increased IFN-γ and cytotoxic potential in proliferating NK cells.¹⁹¹ Thus, the effect of inhibiting DNA methylation may vary depending on context. Of note, in type 2 diabetic patients, which experience a low-grade inflammation, NK cells and B cells showed specific global profiles in increased DNA methylation.¹⁹² Altogether, these findings highlight an important role for histone modifications and DNA methylation at specific gene loci in controlling the effector function of NK cells, and suggest pharmacological intervention at the histone or DNA level as a means of manipulating NK cells for therapeutic use in the clinic.

6 ∣. ROLE OF METABOLISM IN INFLUENCING EPIGENETIC REGULATION

Recent studies have underscored the growing appreciation that metabolic processes within the immune system contribute to not only the basic needs of cellular homeostasis but also the more specialized processes of cellular differentiation.¹⁹³ Inextricably linked to metabolism is its role in regulating epigenomic activity.¹⁹³ Mechanistically, several metabolites are utilized by many components of the epigenetic machinery in immune cells, including S-adenosylmethionine (SAM) for methyltransferases and acetyl-CoA for acetylation via histone acetyltransferases.¹⁹⁴ Metabolic programs play important roles in memory T-cell fate decisions, as demonstrated by their stage-specific requirements of particular metabolic processes. Seminal studies demonstrated that T cells relied heavily on glycolysis during early activation, whereas they required fatty acid oxidation and oxidative phosphorylation during the memory phase,^194-196 illustrating a causal role between metabolic reprogramming and T-cell differentiation. Furthermore, LDHA-driven aerobic glycolysis in activated T cells was shown to program effector function via maintenance of abundant acetyl-coA levels for histone acetylation at specific gene loci (eg Ifng) and promoting transcription.¹⁹⁷ In NK cells, metabolic processes also play key roles in receptor activation and cytotoxicity.^198,201 During viral infection, elimination of depolarized mitochondria (via mitophagy) to decrease reactive oxygen species in effector NK cells was important to maintain cellular fitness and ultimately promote memory NK cell formation,¹⁹⁹ while both glycolysis and oxidative metabolism was required for the clonal proliferation of Ly49H⁺ NK cells.^200,201 Given the connection between metabolism and epigenetic regulation, it is likely that metabolic processes fueling antiviral NK cell responses may also contribute to the epigenetic rewiring of effector and memory NK cells.

7 ∣. CONCLUDING REMARKS

Although the signals that govern NK cell effector functions have been well-established, we are only beginning to unravel the long-term consequences of these signals in the form of innate immune memory. Innate immune memory in NK cells offers unique opportunities to utilize these potent cytotoxic cells as “off-the-shelf” alternatives for cellular immunotherapy that are renewable and can sustain safe yet effective effector potential. There is clear evidence that cooperation and synergy, as well as antagonism, can occur depending upon different receptor and cytokine signals. How the synergy or antagonism of signals will epigenetically poise the NK cell is still unclear. As depicted in Figure 4, different scenarios could be possible. As shown in model 1, different signals could be independently working in parallel to make up different components of epigenetic memory. On the other hand, these signals may funnel into a key component to amplify that common signal. The third possibility is a combination of both scenarios, which we believe is likely the case. Regardless, understanding how these individual components and their cooperation, synergy, and antagonism contribute to the complexity of memory responses will be imperative to fully exploit the enhanced features of innate immune memory. The application of this concept has recently been translated into clear clinical benefits with the advent of chimeric antigen-receptor (CAR) NK cells, which integrate signal 1 through ITAM signaling, signal 2 through various costimulatory molecules, and signal 3, mostly through IL-15 signaling.²⁰² While clinical trials for these CAR-NK cells are indeed promising, there are many opportunities for improvement to fully unleash the potential of NK cell immunotherapy. These opportunities lie in the epigenetic mechanisms that endow NK cell memory responses, and offer an exciting path to bring NK cells into the new era of cellular immunotherapy.

FIGURE 4.

Models of signal integration for epigenetic memory in lymphocytes

DATA AVAILABILITY STATEMENT

The data that support the findings of this study are openly available in the Gene Expression Omnibus (Super-Series accession numbers GSE106139 and GSE140044).