NK Cell Responses Redefine Immunological Memory

The immune system has classically been divided into two arms, innate and adaptive immunity. Innate immunity is poised for swift, short-lived effector responses mediated through recognition of evolutionarily conserved signals via germline-encoded receptors (1). Although initially slow in onset, adaptive immunity is considered highly specialized based on the ability to somatically rearrange Ag receptor genes to generate a diverse repertoire of T and B cells that can amplify Ag-specific responses through prolific clonal expansion (2–4). Because adaptive immune cells can persist long-term following recognition of cognate Ag and execute a quantitatively and qualitatively more robust response following rechallenge with the same Ag, T and B cells were thought to be the only immune population capable of generating memory.

The emergence of immunological memory in the adaptive immune system can be traced to lower vertebrates, including the jawless fish such as lamprey and hagfish. Early studies demonstrated that lampreys immunized with the killed bacterium Brucella abortus produce long-lived “antibody” capable of agglutinating Brucella cells upon rechallenge but not capable of agglutinating typhoid-paratyphoid cells, underscoring the Ag-specific nature of these Ab titers (5). The lamprey can also mediate delayed-type hypersensitivity reactions following stimulation with Mycobacterium tuberculosis–fortified CFA (5). The basis for these phenomena may be attributable to the recently discovered lymphocyte-like populations and variable lymphocyte receptors, akin to primordial T and B lymphocytes and their Ag receptors, respectively (6, 7).

However, evidence also exists for immunological memory in invertebrates devoid of an adaptive immune system. Protection of the American cockroach Periplaneta americana against the bacterium Pseudomonas aeruginosa was enhanced by prior immunization with killed P. aeruginosa but not with saline or immunization with an array of other Gram-negative organisms (8). Interestingly, this protection against P. aeruginosa rechallenge persisted for 14 d postimmunization (8), demonstrating both specificity and memory. Similar findings were demonstrated in the copepod Macrocyclops albidus, in which exposure to larvae of their natural parasite, the tapeworm Schistocephalus solidus, resulted in fewer infections from sibling but not unrelated parasites, the first evidence of innate immune memory in crustaceans (9). Similar protective memory responses against bacteria, parasites, and fungi in other invertebrates add to the mounting evidence for innate immunity’s capacity for memory responses in lower organisms (10–16), suggesting that the ability to remember encountered pathogens may be evolutionarily conserved across both innate and adaptive immunity.

NK cells were first described in 1975, when several groups identified a lymphocyte population in athymic (nude) mice capable of mediating “natural” (i.e., without the requirement of prior target exposure) cytotoxicity against both syngeneic and allogeneic tumor cell lines (17–20). Cytotoxic activity was maintained despite filtration of splenocytes through an anti-immunoglobulin column or treatment with carbonyl iron/magnet, excluding a contribution from T and B lymphocytes or macrophages (18, 20). Because of their capacity to rapidly secrete lytic molecules and the proinflammatory cytokine IFN-γ upon sensing pathogen-derived or host stress ligands through a repertoire of germline-encoded receptors, and because they lack RAG-mediated rearrangement of receptor genes, NK cells were characterized as a component of innate immunity (21–25).

However, current evidence has revealed striking similarities between NK cells and adaptive immune cells. NK cells develop from the common lymphoid progenitor, from which T and B cells and the newly described lineage of innate lymphoid cells (ILCs) are also derived (26). Similar to T and B cells, NK cell development, homeostasis, survival, and function require common γ-chain–dependent cytokine signaling, particularly IL-15 (27–29). Although NK cells do not require expression of the RAG recombinase for development or generation of their receptor repertoire, nearly a third of peripheral NK cells have a history of RAG expression during ontogeny (30, 31). Furthermore, analogous to thymic T cell education, NK cells acquire functional competence through a “licensing” or “arming” process; during development, NK cells become self-tolerant (and thus gain effector function) due to engagement of self-MHC by inhibitory receptors (32–34).

Lastly, NK cells can undergo clonal expansion and initiate Ag-specific recall responses. Perhaps the earliest evidence suggesting the possibility of NK cell memory was reported in 1964 in a model of F₁ hybrid resistance. Adult F₁ hybrid mice (B10 × B10.D2) reject parental (B10 or B10.D2) bone marrow grafts (35). When primed with a bone marrow graft from one parent (B10), F₁ hybrid mice more rapidly rejected a second graft from that same parent (B10) compared with a graft derived from the other parent (B10.D2) (35). Conversely, priming the F₁ hybrids with a B10.D2 or allogeneic graft did not accelerate B10 graft rejection (35). Taken together, these findings suggested that a non–T or non–B cell responded in a qualitatively different manner upon re-exposure to previously encountered Ags. The later discovery of the NK cell and the “missing-self” hypothesis (that NK cells selectively kill cells lacking self–MHC class I) implicated NK cells as the cell type mediating F₁ hybrid resistance (36). Since then, NK cells have been found to possess a number of adaptive features that have redefined their role in immunity.

The first evidence of anamnestic NK cell responses was in the setting of contact hypersensitivity responses to chemical haptens. Rag2^−/− mice lacking both T and B lymphocytes exhibited a severe inflammatory reaction when sensitized and rechallenged with the same hapten (either 2,4-dinitro-1-fluorobenzene [DNFB] or oxazolone) (37). Depletion of NK cells abrogated the contact hypersensitivity, suggesting that NK cells either directly or indirectly were responsible for the observed recall response (37). Adoptive transfer of DNFB-sensitized NK cells into Rag2^−/−Il2rg^−/− mice was also sufficient to drive contact hypersensitivity upon recipient challenge with DNFB, although transferable hapten-specific recall was retained only in hepatic, but not splenic, NK cells (37, 38). Specifically, contact hypersensitivity depended on a subset of hepatic NK cells expressing the chemokine receptor CXCR6, which was required for the homeostasis but not Ag recognition of these cells (38). Thus, hepatic NK cells can generate Ag-specific recall responses to haptens, although whether these cells are truly mature NK cells or a distinct subpopulation of the type 1 ILC family is unresolved (39).

NK cells can also undergo recall responses against viral pathogens (40). During the expansion phase of the NK cell response to mouse CMV (MCMV) infection that peaks at day 7 postinfection, the Ag-specific NK cell compartment has been measured to undergo ∼100- to 1000-fold growth in size (40). This proliferative burst is driven by Ag-specific interactions between the activating receptor Ly49H on NK cells and the MHC class I–like viral glycoprotein m157 on the surface of infected cells (41–43). Following robust expansion of Ly49H⁺ NK cells after MCMV infection, these effector cells contract and form a long-lived pool of memory NK cells in both lymphoid and nonlymphoid tissues that is detectable at least 70 d later (40). These memory NK cells exhibit enhanced IFN-γ production and degranulation compared with naive NK cells (40). The response of memory NK cells rechallenged with MCMV was found to be comparable in both kinetics and magnitude to that of naive NK cells, yet memory NK cells conferred greater protection against MCMV challenge to susceptible neonate mice (40). Thus, MCMV-experienced NK cells are capable of recall responses, enhanced functionality, and protection against repeated MCMV exposure.

Evidence for secondary NK cell responses against different viral pathogens continues to build. Analogous to hapten-specific memory NK cell–mediated contact hypersensitivity responses, adoptively transferred hepatic, but not splenic, NK cells from Rag1^−/− mice immunized with virus-like particles expressing influenza A virus, vesicular stomatitis virus, or HIV-1 Ags afforded protection to Rag2^−/−Il2rg^−/− hosts challenged with the immunizing, but not unrelated, virus (38). Similar immunization-dependent and virus-specific NK cell protective responses in the absence of adaptive immunity have been described for HSV-2, vaccinia virus, and influenza virus (44–46).

Similar to the expansion of NK cells in MCMV-infected mice, human NK cells expressing the activating heterodimeric CD94/NKG2C receptor are preferentially found in the peripheral blood of healthy individuals seropositive for human CMV (HCMV), compared with donors who were HCMV-seronegative or -seropositive for other herpesvirus infections (47–49). This CD94/NKG2C⁺ subset commonly coexpresses the maturation marker CD57 and lacks expression of the inhibitory NKG2A receptor (49–52). Although the ligand driving expansion of CD94/NKG2C⁺ NK cells in vivo has yet to be elucidated, in vitro studies demonstrated that short hairpin RNA–mediated knockdown of HLA-E on HCMV-infected fibroblasts abrogated this expansion (53). HCMV reactivation in patients receiving allogeneic hematopoietic stem cell grafts, umbilical cord blood grafts, or solid-organ transplants also precipitates expansion of CD94/NKG2C⁺ NK cells, followed by persistence of these cells for months to years (49, 54–57). Furthermore, the transfer of NKG2C⁺ NK cells in grafts from HCMV-seropositive donors resulted in enhanced target cell–induced IFN-γ production upon secondary CMV exposure in the recipient compared with grafts from HCMV-seronegative donors (50). In combination, these studies highlight the Ag specificity, longevity, and transplantability of human memory NK cell responses to HCMV. A recent study also identified a subset of CD16⁺CD56⁺FcεRIγ⁻ NK cells associated with prior HCMV and HSV-1 infection that persisted upwards of 9 mo and was capable of mediating enhanced Ab-dependent cellular cytotoxicity in the presence of HCMV- and HSV-1–infected cells coated by virus-specific Abs (58, 59). Thus, human NK cells appear to have evolved redundant mechanisms to generate immunological memory.

Several studies have also demonstrated expansion of long-lived CD94/NKG2C⁺ human NK cells in response to HIV-1, hantavirus, Chikungunya virus, hepatitis B virus, or hepatitis C virus, although this expansion occurred only in individuals previously infected with HCMV (60–64), suggesting that superinfection with these viruses may trigger HCMV reactivity. Interestingly, a recent study in rhesus macaques established that primate NK cells can achieve pathogen-specific memory independent of HCMV exposure (65). Infection of rhesus macaques with simian HIV or SIV elicited splenic and hepatic memory NK cells capable of lysing Gag- and Env-pulsed dendritic cells in an NKG2A- and NKG2C-dependent fashion for as long as 5 y postinfection (65). Vaccinating the macaques with recombinant adenovirus expressing HIV-1 Env or SIV Gag likewise produced robust, stable, and Ag-specific NK cell memory (65).

The benefit of pathogen-specific NK cell memory is clear in the case of persistent or repeated encounter with the same virus. However, there is some evidence that suggests there may be a costly trade-off associated with NK cell memory. A recent study highlighted that human NK cell receptor diversity increases with age (66). CD57⁺NKG2C⁺ NK cells that expand during HCMV infection are far from clonal, displaying substantial heterogeneity for other receptors (67). However, high NK cell receptor diversity was associated with greater risk of HIV-1 acquisition in Kenyan women (66). Given that viral challenge may enhance NK cell receptor diversity, human NK cells appear to risk unresponsiveness to novel Ags to better protect against previously encountered pathogens (66). The consequence of NK cell receptor diversity for human health and disease thus requires further exploration.

Analogous to the generation of T cell memory against pathogens (4), NK cells progress through three phases during their response to MCMV: expansion, contraction, and memory maintenance (68). During each stage, both intracellular and extracellular cues are necessary for establishing a long-lived memory NK cell pool (Fig. 1). In addition to Ag engagement by activating receptors (analogous to TCR engagement for T cell activation), NK cells require proinflammatory cytokine signaling for robust expansion (4). The proinflammatory cytokine IL-12, through a STAT4-dependent but IFN-γ–independent mechanism, is indispensable for optimal MCMV-specific NK cell clonal expansion as well as memory NK cell formation (69). IL-12 and STAT4 may be responsible for programming activated NK cells early during MCMV infection for memory formation (69). Members of the IL-1 family of cytokines, particularly IL-33 and IL-18, are similarly necessary for amplifying NK cell proliferation in response to MCMV, but are dispensable for recall responses (70, 71). A recent study also identified a role for type I IFN and downstream STAT1 signaling in shielding proliferating NK cells from fratricide (killing by other NK cells) by modulating their cell surface expression of NKG2D ligands (72).

Interestingly, IL-12, IL-18, and type I IFN signaling act synergistically to drive maximal expression of the BTB-ZF family transcription factor Zbtb32, which controls the proliferative burst of virus-specific NK cells by antagonizing the antiproliferative factor Blimp-1 (73). In parallel, IL-12– and IL-18–mediated induction of miR-155 regulates effector and memory NK cell numbers during MCMV infection by regulating targets such as Noxa and suppressor of cytokine signaling 1 (SOCS1) (74). Although the mechanism by which SOCS1 impairs the development of effector NK cells is unclear, the potent restraint that constitutive SOCS1 activity places on STAT signaling may have some influence. Lastly, akin to the necessity of costimulation for T cell activation (signal 2), NK cells require the costimulatory molecule DNAX accessory molecule 1 and downstream signaling through the Src family tyrosine kinase Fyn and the serine-threonine protein kinase C isoform η for optimal expansion of effector NK cells and their differentiation into memory NK cells (75).

Following viral infection, contraction of effector lymphocytes serves to eliminate activated CD8⁺ T and NK cells to stave off immunopathology (76, 77). Mitochondrial apoptosis mediated by the proapoptotic factor Bim shapes the size, maturity, and functionality of the memory NK cell pool in response to MCMV (78), similar to CD8⁺ T cells (79). The accumulation of depolarized mitochondria and mitochondrial-released reactive oxygen species in effector NK cells after virus-driven expansion results in either cell death or clearance of damaged mitochondria, resulting in NK cell survival (80). Analogous to the autophagy-dependent survival and memory formation of virus-specific effector CD8⁺ T cells (81–83), surviving NK cells undergo the self-catabolic process of mitophagy during the contraction-to-memory phase transition, requiring the autophagosome machinery component Atg3 and the mitophagy-specific receptors BCL2/adenovirus E1B 19-kDa interacting protein 3 (BNIP3) and BNIP3-like (or Nix) to promote their survival (80). Mitophagy in contracting NK cells is induced by mechanistic target of rapamycin inhibition or AMP-activated protein kinase activation (80), suggesting that mitophagy may also link other cellular metabolic processes, similar to the catabolic processes memory T cells employ to fuel oxidative phosphorylation during nonproliferative states (84).

In addition to intracellular mechanisms, extracellular cues can also promote the maintenance of NK cell memory. Adoptive transfer of Ly49H⁺ NK cells isolated from MCMV-infected hosts at day 7 or day 21 postinfection into IL-15–deficient recipients led to decreased persistence of the transferred cells, supporting a role for IL-15–dependent maintenance of NK cells during contraction (85). IL-15 was previously shown to promote NK cell survival via Mcl-1 (86). miR-155–mediated suppression of Noxa may also aid long-term survival of memory NK cells (74). Thus, generating long-lived, MCMV-specific memory NK cells requires a complex combination of intracellular and extracellular signals during both the early and late phases of the antiviral response.

During viral infection, two subsets of effector CD8⁺ T cells have been described to develop: terminally differentiated killer cell lectin-like receptor G1 (KLRG1)^hi short-lived effector cells (SLECs) that die after infection, and KLRG1^lo memory precursor effector cells (MPECs) that are long-lived and participate in secondary responses (87). Analogous to CD8⁺ T cells, recent evidence supports the idea of heterogeneity within antiviral NK cell populations that can dictate memory potential. KLRG1^loLy49H⁺ NK cells preferentially generate memory NK cells compared with KLRG1^hi cells, which have a limited capacity for MCMV-driven expansion (88). KLRG1 is associated with NK cell maturation, as indicated by the greater percentage of KLRG1^hi NK cells that are also of the most mature CD11b⁺CD27⁻ phenotype, suggesting that NK cell maturation is antagonistic to their memory potential (88). However, because memory NK cells are themselves KLRG1^hiCD11b⁺CD27⁻, yet still competent to undergo robust secondary expansion following MCMV challenge, KLRG1 alone does not dictate proliferative potential (40, 88).

Interestingly, KLRG1 expression on naive NK cells may be dictated during development by the activity of the RAG recombinase. NK cells with a history of RAG activity during ontogeny preferentially expand and persist as memory cells following MCMV infection (Fig. 2A) owing to an enhanced overall cellular fitness measured by the ability to repair DNA breaks, which can occur during stresses such as rapid proliferation or exposure to ionizing radiation (30). In contrast, the absence of RAG expression in developing NK cells results in diminished expression of the DNA damage repair machinery and a subsequent impairment in DNA double-strand break resolution following DNA damage (30). Thus, although the underlying mechanisms remain to be elucidated, RAG activity during ontogeny shields NK cells from global genomic instability and apoptosis. Furthermore, Rag2^−/− NK cells display a transcriptional signature markedly similar to that observed in SLECs rather than MPECs (SLEC and MPEC signatures published in Ref. 89) (Fig. 2B), supporting the hypothesis that RAG dictates functional heterogeneity within the NK cell compartment. It will be of interest to determine whether the degree of genomic integrity similarly specifies SLEC versus MPEC fate.

Although RAG appears to determine KLRG1 expression levels in a cell-intrinsic manner, KLRG1 expression can also be influenced in a cell-extrinsic manner. One study demonstrated that T cells can restrain NK cell maturation by limiting the availability of IL-15, driving the preferential generation of KLRG1^lo memory progenitors at the expense of KLRG1^hi NK cells (Fig. 2A) (88). Signals derived from the host commensal microbiota were also shown to control expression of KLRG1 in NK cells (Fig. 2A) (88). Treatment with broad-spectrum antibiotics in the drinking water diminished KLRG1 expression and boosted the frequency of memory NK cells compared with untreated wild-type mice, suggesting that the host microbiota regulates the NK cell pool containing memory potential (88). Thus, the foundation for NK cell memory formation may be laid even prior to encountering virus.

MCMV-specific memory NK cells display greater cell surface expression of not only KLRG1 but also Ly49H compared with naive NK cells, and expression of these receptors is further enhanced in secondary memory NK cells (90). In contrast, naive, memory, and secondary memory NK cells express comparable levels of the activating receptors NK1.1 and Ly49D (90). These data together imply that there may be selection for NK cells with the greatest avidity for ligand during successive rounds of MCMV infection (Fig. 2A). T cells undergo a process of affinity maturation, whereby T cells bearing TCRs with highest affinity for peptide rise to clonal dominance during the response to a pathogen (91–94). A similar process may occur with NK cells responding against infection. There is currently no evidence to support the idea that NK cells undergo somatic mutation of their Ag receptor genes (i.e., the affinity of each Ly49H receptor for m157 is equivalent), indicating that NK cell avidity for viral Ag is determined solely by the number of Ly49H receptors on the surface of a given NK cell. However, it remains to be determined at what stage of the antiviral response activating signals through Ly49H select for memory precursors to constitute the long-lived pool.

Emerging evidence also illustrates the contribution of the activating receptor Ly49D and the inhibitory receptor Ly49A, both of which recognize the MHC class I molecule H-2D^d, during the NK cell memory response to MCMV (95). Adoptive transfer of B10.D2 (H-2D^d–sufficient background) NK cells into MCMV-infected syngeneic Ly49H-deficient recipients demonstrated that Ly49D⁺Ly49A⁻Ly49H⁺ NK cells preferentially differentiated into memory NK cells compared with Ly49D⁻Ly49A⁺Ly49H⁺ NK cells (95). Similar to how MCMV infection can break the anergic state of unlicensed NK cells in B6 mice, it appears that acute viral infection can also breach tolerance of activating receptors for self–MHC class I, a phenomenon that has functional consequences for host protection and NK cell memory (95–98).

Recent human studies have suggested that epigenetic heterogeneity is also associated with differential capacity for NK cell longevity (Fig. 2A). The aforementioned CD16⁺CD56⁺FcεRIγ⁻ NK cells are a subset of memory-like NK cells that can be isolated from HCMV-seropositive individuals, although they can be characterized by either the absence or presence of CD94/NKG2C (58, 59, 99, 100). Interestingly, this population lacks expression of the tyrosine kinase SYK, the signaling adaptors DAB2 and EAT-2, and the transcription factors promyelocytic leukemia zinc finger protein and Helios due to promoter hypermethylation at several of these loci (99, 100). Another study has identified epigenetic imprinting at the Ifng conserved noncoding sequence 1 in NKG2C⁺ NK cells from HCMV-seropositive individuals that is critical for Ifng transcriptional activity in response to stimulation through NKG2C (101). It is unclear whether this epigenetic heterogeneity is a cause or consequence of NK cell memory. Nevertheless, receptor heterogeneity, the microbiota, and history of RAG expression during ontogeny precipitate differential fitness of cells within the naive NK pool. During viral infection, selective pressure may then be acting on the effector pool to preferentially select memory precursors to establish the memory pool.

Although proinflammatory cytokine signaling is critical in driving the clonal expansion and maintenance of MCMV-specific memory NK cells, proinflammatory cytokines alone were found to be capable of supporting NK cell memory properties in the absence of Ag. Splenic NK cells preactivated with a mixture of IL-12, IL-15, and IL-18 and adoptively transferred into Rag1^−/− mice become long-lived and can be identified in these recipients up to 3 wk following transfer (102). These cytokine-induced memory-like NK cells retained a cell-intrinsic capacity for enhanced IFN-γ production, but not cytotoxicity, when restimulated with cytokine, plate-bound Ab, or target cells (102, 103). Human NK cells preactivated with cytokines also displayed the same properties (104). The progeny of cytokine-induced memory-like NK cells similarly exhibited enhanced effector functions (102, 104), suggesting that NK cell memory properties may be epigenetically inherited. These data may have implications for the Ag-independent self-renewal of memory NK cells following clearance of pathogens, although future studies are necessary to dissect the contribution of cytokine-induced memory-like NK cells during a recall response to a pathogen against which pathogen-specific memory NK cells already exist. Given that Ag-specific NK cells demonstrate diminished bystander activation to heterologous infection (105), cytokine-induced memory-like NK cells may represent a strategy for the host to nonspecifically respond to a new proinflammatory stimulus. Recent studies have shown that NK cells preactivated with IL-12, IL-15, and IL-18 demonstrate enhanced persistence and antitumor activity against established, irradiated mouse tumors compared with naive NK cells, suggesting that harnessing the sustained effector functions of cytokine-induced memory-like NK cells may represent a potential enhancement to adoptive NK cell immunotherapy (106). ILCs are thought to respond exclusively to cytokine cues (107), and it will be of interest to determine whether cytokines can similarly support longevity in other innate lymphocytes.

Lastly, in an NK cell–deficient lymphopenic host, adoptively transferred NK cells undergo a rapid, Ag-independent homeostatic proliferation to fill the empty niche, a process that requires common γ-chain–dependent cytokine signaling (108–110). These same cytokines, as well as TCR-mediated self-peptide/MHC interactions, are known to support the acquisition of memory-like characteristics in naive T cells undergoing lymphopenia-induced proliferation (111). Similarly, following homeostatic proliferation in Rag2^−/−Il2rg^−/− or sublethally irradiated recipients, adoptively transferred NK cells contract to form a long-lived population that persists in both lymphoid and nonlymphoid organs for at least 6 mo (112). These NK cells display an enhanced capacity for IFN-γ production and degranulation when stimulated ex vivo 10 d after transfer, but the functionality of these homeostatically driven NK cells following contraction is unclear (112). Nevertheless, homeostatically expanded NK cells self-renew and are capable of mounting a robust proliferative response when challenged with MCMV 6 mo after transfer, thus sharing some properties with MCMV-specific memory NK cells (112). Emerging evidence demonstrates that Atg5-mediated autophagy is critical for the survival of NK cells during homeostatic proliferation, suggesting that this is one mechanism by which NK cells acquire memory properties following lymphopenia-driven proliferation (113).

Immunological memory represents just one example of the adaptive features NK cells exhibit during their dynamic life span. Mounting evidence in both mice and humans points to the remarkable capacity of NK cells to generate memory responses in both infectious and noninfectious settings. Virus infection models are allowing us to uncover the molecular mechanisms necessary for NK cell memory formation and maintenance, yet the pathways that govern cytokine- and lymphopenia-induced memory-like NK cells remain poorly understood. NK cells have expanded the classical definition of immunological memory found in textbooks. It will now be of interest to determine whether a similar capacity for longevity, self-renewal, and robust recall responses exists within the newly described ILC lineages, which generally respond to proinflammatory cytokine cues. NK cells have a diverse repertoire of cell surface–activating receptors, and future studies are necessary to address whether other activating receptors are sufficient to drive NK cell memory when exposed to cognate ligands. Resolving these questions will facilitate NK cell vaccination strategies and adoptive NK cell immunotherapies for viral infection and malignancy.

We thank members of the Sun Laboratory for helpful discussions and review of this manuscript. We apologize to those whose work we were unable to discuss due to space limitations.

The authors have no financial conflicts of interest.