NK cells are most often confused with T cells because they may have similar morphologies, express several cell surface molecules in common,^9,10 and share functional capabilities. While this confusion was frequent before the molecular description of the T-cell receptor (TCR)/cluster of differentiation (CD)3 complex, their similarities remain a potential source of uncertainty. However, mature NK cells are clearly not T cells by several criteria.¹¹ Conventional NK cells do not require a thymus for development and are normal in athymic nude mice (though this is not the case for the newly described “thymic” NK cell subset, discussed subsequently). NK cells do not express the TCR on the cell surface, do not produce mature transcripts for TCR chains, and do not rearrange TCR genes.^12,13 Mice with the scid mutation or deficiencies in Rag1 or Rag2 lack TCR gene rearrangements and mature T cells but possess NK cells with apparently normal function.^14,15,16,17 Several CD3 components may be found in the cytoplasm of NK cells, particularly immature NK cells, but they are not displayed on the cell surface¹⁸ with the exception of CD3ζ. But CD3ζ is expressed in association with FcγRIII (CD16) and other NK cell activation receptors instead of the TCR/CD3 complex.^19,20 Whereas mice lacking CD3ζ lack most T cells, NK cell number and function are minimally affected.²¹ On the other hand, NK cells are completely absent in mice with only partial defects in T-cell subsets, such as in mice lacking components of the IL-15R (see following discussion). NK cells do not require the presence of major histocompatibility complex (MHC) class I (MHC-I) molecules on their targets for lysis in an important functional distinction with CD8+ MHC-I-restricted T cells. Instead, NK cells kill more efficiently when their targets lack MHC-I expression. Thus, NK cells can be clearly distinguished from T cells, even from so-called CD3+ NKT cells that express NK cell markers (see following discussion).

Another area of overlap between NK and T cells concerns cytokine responses. When mouse splenocytes or human peripheral mononuclear cells are exposed to high concentrations of interleukin (IL)-2 (800 to 1000 U/mL), robust lymphocyte proliferation ensues (ie, LAK cells are generated).^22,23,24,25 Although most are CD3- NK cells, TCR/CD3+ T cells are also produced. To distinguish NK cells within this population, they are sometimes called “CD3^– LAK” cells or “IL-2-activated NK cells.”

In a related phenomenon, NK cells are activated when mice are injected with polyinosinic-polycytidylic acid (poly-lic) or other agents that trigger through Toll-like receptors (TLRs), often on plasmacytoid dendritic cells (DCs).²⁶ NK cells can also be activated in vitro with interferon (IFN)α/β, IFNγ, or low concentrations of IL-2 that are insufficient to induce proliferation.^3,4,27

NK cells activated in these various ways, with or without proliferation, display enhanced killing of typical NKsensitive targets. They also kill a broader panel of targets, including those that are generally resistant to freshly isolated NK cells, such as the murine P815 mastocytoma cells and
freshly explanted tumors. Many agents that enhance killing by NK cells may also activate T cells, such that even T-cell clones may display promiscuous killing of targets that is no longer MHC-restricted.²⁸ This phenomenon was a frequent source of confusion between NK cells and T cells during the initial characterization of both cell types.

Why cytokine activation leads to enhanced killing is incompletely understood. Interestingly, mouse NK cells constitutively express messenger ribonucleic acid (mRNA) for cytotoxicity components, perforin, and granzymes, but no protein.²⁹ Cytokine activation enhances expression of mRNA for perforin and granzymes,³⁰ and translation into expressed proteins that contributes to more lytic capacity,^29,31 but this effect may not explain the capacity of LAK cells to kill a broader panel of targets. Activated NK cells express additional receptors that may deliver stimulatory signals,^32,33 but their contribution to the LAK phenomenon remains unclear. In mice, cytokine activation of NK cells results in expression of an alternatively spliced form of a receptor termed NKG2D (see subsequent discussion), which may play a role,^34,35 and IL-15 contributes to LAK-like activity of CTLs, although how much this applies to conventional NK cells is not understood.³⁶ Enhanced killing may also be due to effects on adhesion molecules.^37,38 Nonetheless, it remains unclear how each of these factors contribute to the LAK phenomenon of enhanced and broader killing capacity as compared to resting NK cells.

It seems unlikely that high concentrations of IL-2 can be achieved, even locally, to stimulate NK cells in vivo. Furthermore, NK cells tend to be early responders in immune responses whereas the prime reservoir of large amounts of IL-2 is the activated T cell that produces it somewhat later. Although naïve T cells can make IL-2 soon after stimulation^39,40 and DCs can produce IL-2 to enhance NK lytic activity,⁴¹ whether the resultant IL-2 concentrations are sufficient to generate LAK cells in vivo is unclear.

Interestingly, NK cells are apparently normal in mice with a targeted mutation in the IL-2 gene or the IL-2Rα chain,^42,43 indicating that IL-2 itself is not required for normal NK cell development. Paradoxically, NK cells are deficient in mice with a mutation in either IL-2Rβ or IL-2Rγ.^44,45,46 The discrepancy in NK cell dependence on IL-2 versus IL-2 receptor (IL-2R), as well as the LAK cell phenomenon, may be best understood by comparing the components of the IL-2 and IL-15 receptors.

In brief, the high affinity IL-2R is a heterotrimeric receptor complex comprised of α (p55), β (p75), and γ (p64) chains.⁴⁷ Though individual components may bind IL-2 with low affinity, only the intermediate-affinity βγ receptor (K_d ˜1 nM) and the high-affinity αβγ receptor (K_d ˜10 pM) are capable of signaling. Resting NK cells constitutively express IL-2Rβγ^47,48 and upon activation, may induce IL-2Rα and further upregulate IL-2Rγ chain expression.⁴⁷ In contrast, resting T cells generally do not express any functional IL-2 receptors, and most naïve T cells do not respond to high concentrations of IL-2.⁴⁹ The IL-2Rγ chain is also termed the common γ subunit (γc) because it is a required component of the multimeric receptor complexes for other cytokines, including IL-15,⁵⁰ that is particularly relevant to NK cells. IL-15 does not bind to IL-2Rα but instead utilizes a unique IL-15Rα chain to form a high-affinity complex with IL-2Rβγ.^51,52 The IL-15Rα chain does not directly signal. Its distribution is widespread on numerous cell and tissue lineages including NK cells.

IL-15 has a number of effects on NK cell biology. It is required for NK cell development; mice lacking IL-15 or any component of the trimeric IL-15R complex lack NK cells.^{44,45,46,53,54} Not surprisingly, mice deficient in other components of the IL-15R complex and its signaling pathway (IL-2Rβ, Jak3, and STAT5α/β) exhibit similar defects in NK cell development.^44,55,56,57 Depending on its relative concentration, IL-15 has an antiapoptotic or proproliferative effect.^58,59 When NK cells are transferred to NK cell-deficient mice, they undergo “homeostatic” proliferation,^60,61 akin to T-cell homeostatic proliferation.⁶² Like memory CD8+ T-cell homeostasis, NKcell homeostasis is IL-15-dependent,^60,61 to a more or less degree.⁶³ Finally, LAK cells can be generated with IL-15.⁶⁴ These studies strongly suggest that LAK cells are generated because high-dose IL-2 acts through the IL-2Rβγ that is normally expressed with IL-15Rα as components of the constitutively expressed trimeric IL-15R complex on resting NK cells. Thus, the LAK cell phenomenon is related to the role of IL-15 and its receptor in NK cell biology.

Interestingly, IL-15 is expressed at very low levels and is difficult to detect in vivo.⁶⁵ The IL-15Rα chain can present IL-15 in trans to NK cells that can respond through IL-2/15Rβγ alone.⁶⁶ For example, IL-15Rα-deficient NK cells develop in bone marrow (BM) chimeric mice in which IL-15Rα-deficient BM was used to reconstitute IL-15Rα-sufficient animals,⁶⁷ indicating that IL-15Rα on another cell can allow IL-15Rα-deficient NK-cell development. In certain inflammatory situations in vivo, DC presentation of IL-15 in trans can enhance NK cell responses (also known as priming).^31,68 In DCs expressing both IL-15 and IL-15Ra, the IL-15Ra chain provides a chaperone function to stabilize receptor-cytokine complexes on the cell surface.⁶⁹ Taken together, trans presentation of IL-15 may be physiologically important to NK cell function.

The constitutive expression on NK cells of IL-15R complex with IL-2Rβ has practical usefulness because anti-IL-2Rβ (CD122) is sometimes used to identify naïve CD3- NK cells or deplete them in mice,⁷⁰ but anti-CD122 is less useful during an ongoing immune response and CD122 is expressed on regulatory T cells. Anti-IL-15Rα antibodies have not been widely used. Other markers have proven to be more useful for analysis of NK cells.

In the mouse, the NK1.1 molecule is an especially important marker on NK cells in C57BL strains.¹¹ NK1.1 is an activation receptor encoded by Nkrp1c (Klrb1c),⁷¹ a member of the Nkrp1 gene family (see following discussion). In FACS sorting experiments, the NK1.1+ fraction contained all of the natural killing activity in the spleen.⁷² In vivo administration of the anti-NK1.1 mAb PK136⁷³ completely abrogated natural killing but did not affect adaptive immune responses⁷⁴
(mAb PK136 is available from the American Type Culture Collection [ATCC], Manassas, VA [HB-191] and is an IgG2a isotype [ATCC, and data not shown], not IgG2b as originally described⁷³). While mAb PK136 is very efficient at NK-cell depletion and is widely used for this purpose, unfortunately it recognizes an epitope on NK1.1 that is confined to C57BL/6, C57BL/10, and a few other strains.⁷³ Moreover, in Swiss, NIH and SJL/J mice, mAb PK136 recognizes another NKRP1 family member, NKRP1B.^75,76 However, there are now available NK1.1+ congenic strains, such as BALB.B6- Cmv1^r (catalogued as C.B6-Klra8^Cmv1-r /UwaJ, stock number 002936 at The Jackson Laboratory, Bar Harbor, ME) in which the C57BL/6 allele of NK1.1 has been genetically bred onto the BALB/c background that otherwise lacks the NK1.1 epitope.⁷⁷ Similarly, the NK1.1 allele has been introgressed onto the nonobese diabetic (NOD) background.⁷⁸

A subpopulation of T cells expresses NK1.1, described in detail in another chapter 18. These “natural killer T (NKT) cells” express the TCR/CD3 complex and typically are restricted by the nonclassical MHC-I molecule, CD1, which presents glycolipid antigens to NKT cells. NKT cells respond early during the course of an immune response and may potently activate conventional NK cells.⁷⁹ Nonetheless, NKT cells can be distinguished from conventional NK cells by expression of the CD3 complex (ie, conventional NK cells are NK1.1+ CD3-).

The NKG2D (Klrk1) activation receptor is expressed on all NK cells in human and all strains of mice examined.⁸⁰ In humans, NKG2D is also expressed on all γδTCR+ and CD8+ T cells, whereas in mice, NKG2D is expressed on most NKT and γδTCR+ T cells but not on resting CD8+ T cells.^81,82,83 However, essentially all activated mouse CD8+ T cells express NKG2D. In both humans and mice, CD4+ T cells do not express NKG2D, but it is found on a subset CD4+CD28- T cells in patients with rheumatoid arthritis.⁸⁴ Blocking anti-NKG2D mAbs and NKG2D-deficient mice have been described.^80,85,86,87 Regardless, conventional NK cells are NKG2D+ CD3-.

The NKp46 (Ncr1) activation receptor appears to be expressed on all CD3- NK cells in humans and all strains of mice. However, recent reports indicate expression of NKp46 on immune cells in the gut that may be developmentally distinct from conventional NK cells.^{88,89,90,91,92} Moreover, depleting anti-NKp46 mAbs have not been described, limiting its usefulness for in vivo functional experiments. Nonetheless, recently developed mice may allow other approaches, such as a mouse where a green fluorescent protein (GFP) cassette was inserted into Nkp46 and two different transgenic (Tg) mice with a Nkp46 promoter contruct for expression of Cre or diptheria toxin receptor.^93,94,95

The mAb DX5 recognizes a molecule that is coexpressed on most NK1.1+ CD3- cells and on small populations of splenocytes in NK1.1- strains, consistent with identification of NK cells in all strains. However, mAb DX5 recognizes the α2 integrin that is widely expressed on other leukocytes, not just NK cells,^96,97 and its expression is regulated.⁹⁸ Nevertheless, the DX5 epitope has been used to identify NK cells in mouse strains that do not express NK1.1, but it has been largely supplanted by other nonpolymorphic markers such as NKp46.

The glycolipid determinant asialo-GM₁ is expressed by most if not all murine NK cells and a subpopulation of T cells.^99,100,101 Although the functional significance of this molecule is unknown, polyclonal rabbit anti-asialo-GM₁ (Wako Chemicals USA, Richmond, VA) has been used to effectively deplete NK cells. In more recent studies, the anti-NK1.1 mAb PK136 has become the reagent of choice for NK-cell depletion because of an available defined mAb and its more restricted reactivity with NK cells.^11,72,73,74 However, anti-asialo-GM₁ remains in use for NK-cell depletion when anti-NK1.1 cannot be employed.⁷³

In addition to NKG2D and NKp46, human NK cells selectively express CD56. Although it is also found on neural tissues and some tumors, CD56 is generally not expressed by other hematopoietic cells or lymphocytes.^102,103,104 This 140 kDa molecule is derived from alternative splicing of the gene encoding neural cell adhesion molecule (NCAM) involved in nervous system development and cell-cell interactions.^105,106 CD56 may be involved in adhesion between NK cells and their targets,¹⁰⁷ but this function is controversial. Curiously, mouse CD56 is not expressed on hematopoietic cells,¹⁰⁸ indicating that its role on NK cells is not conserved. Nevertheless, CD56 is particularly useful as a pan-NK-cell marker in humans.

Human NK cells can be functionally divided according to the level of CD56 expressed.^103,109 Most human peripheral blood NK cells are CD56^dim, a phenotype associated with more cytotoxicity and less cytokine production than a smaller subset of NK cells that express CD56 at higher levels (CD56^bright). These cells also tend to differentially express receptors involved in target recognition as well as CD16. The CD56^bright cells may undergo a maturation process to become CD56^dim cells¹¹⁰ and may be related to a subset of NK cells identified in mice, termed “thymic” NK cells.¹¹¹

Other molecules selectively expressed on NK cells are better discussed below under the general topic of NK cell receptors because they are molecularly defined and their ligands are known.

A hallmark of NK-cell effector function is target killing, mediated by a process termed granule exocytosis that can be initiated by exposure to susceptible targets or cross-linking of specific activation receptors. Like CTLs, NK cells possess preformed cytoplasmic granules that resemble secretory lysosomes with properties of both secretory granules and lysosomes.¹²⁰ Granule formation is affected by Lyst, the molecule defective in humans with Chediak-Higashi syndrome^121,122 in which enlarged lysosomes are observed apparently due to decreased lysosome fission.¹²³ Normal granules contain perforin and granzymes (granule enzymes). Perforin, a pore-forming protein, is rendered inactive by association with calreticulin and serglycin, and is activated by a cysteine protease.¹²⁴ Granzymes are first produced as inactive proenzymes that are activated by N-terminal cleavage by dipeptidyl peptidase I, also known as cathepsin C. However, granzymes are rendered inactive by the acidic pH of the granules. Upon activation by a sensitive target, NK (and T) cells are triggered to rapidly polarize the granules and reposition the microtubule organizing center toward the target in a dynein-dependent manner.¹²⁵ The granule membrane ultimately fuses with the plasma membrane, and externalizes, releasing granule contents. Calcium-dependent polymerization of perforin results in “perforation” of the target cell plasma membrane, and granzyme entry by an as yet incompletely understood process. A recent study suggests that perforin induces a plasma membrane repair process that results in endocytosis of perforin and granzymes into enlarged endosomes, called “gigantosomes.”¹²⁶ Perforin pores in the gigantosome membrane then allow delivery of granzymes that mediate cleavage of caspases and Bid, ultimately leading to target cell apoptosis.¹²⁷

Recently, many details of the granule exocytosis pathway have come from studies of the heterogeneous human disorder, hemophagocytotic lymphohistiocytosis (HLH).^128,129 In particular, genetic studies of heritable HLH, termed familial HLH (FHL), led to identification of the first described FHL mutation in the perforin gene (PRF1), responsible for FHL2. Studies of patients with FHL without PRF1 mutations led to discovery of other genes whose products (MUNC13-4, syntaxin 11, MUNC18-2) affect granule exocytosis by cytotoxic lymphocytes. Fusion of the cytolytic granule with the plasma membrane requires vesicular RAB27a, a member of the small GTPase superfamily. Defects in RAB27a are associated with the human disorder Griscelli syndrome, type 2. Mice have been described with defects in granule exocytosis components including LYST (beige), perforin (Pfn1-/-), Unc13d (equivalent to MUNC13-4, also known as Jinx), and Rab27a (ashen). As highlighted by the names of the mutant mice, many mutations of molecules in the granule exocytosis pathway are associated with skin pigment changes because these molecules also affect melanosomes in melanocytes.¹³⁰

Human T and NK cells also express another pore-forming molecule, granulysin, that is related to a family of saposin-like proteins.¹³¹ Based on crystallographic studies, these molecules appear to be active against bacteria, fungi, and tumor cells by charge association with target membranes and subsequent disruption, leading to target cell lysis.¹³² Granulysin is contained in cytolytic granules containing the other cytolytic proteins, such as granzymes.¹³³ In a perforindependent manner, granulysin causes target apoptosis but is not expressed in mouse cytotoxic lymphocytes.¹³⁴

NK-cell cytotoxcity can be demonstrated in several related ways. Natural killing refers to the process by which NK cells kill certain tumor targets without need for prior host sensitization with the target. Natural killing was first
assessed with a simple in vitro assay for target membrane integrity that is still used today, the standard ⁵¹Cr-release assay.¹³⁵ The prototypical NK-sensitive tumor target for mouse NK cells is YAC-1 (TIB-160 from ATCC), a thymoma derived from Moloney virus-infected A strain mice, whereas the standard human target is K-562 (CCL-243 from ATCC), an erythroleukemic cell line derived from a human patient with chronic myelogenous leukemia in blast crisis.¹³⁶ Maximal killing by enriched, IL-2-activated NK cells usually occurs with effector:target (E:T) ratios of <10:1 whereas unfractionated, freshly isolated peripheral blood or splenocyte preparations usually require E:T ratios of >100:1. Even at high E:T ratios, not all targets are killed, with percentagespecific cytotoxicity typically ranging from ˜10% with fresh NK cells to ˜80% with activated NK cells. Note that perforin-dependent leakage of ⁵¹Cr from the targets is mostly complete within about an hour; 4-hour assays are standard. Longer periods may reflect other apoptotic processes, such as Fas-induced apoptosis.

While ⁵¹Cr release is still the gold standard, there are also numerous nonradioactive tests for perforin-dependent killing, including release of intracellular enzymes or use of fluorochromes for target labeling.^137,138,139 The release of granule components, including granzymes, into the supernatant can be determined by conversion of an appropriate substrate, such as granzyme A-mediated cleavage of alpha-N-benzyloxy-carbonyl-L-lysinethiobenzyl ester (also known as BLT-esterase activity).¹⁴⁰

A particularly useful new flow cytometric assay exploits the orientation of lysosomal-associated membrane protein-1 (LAMP-1, CD107a) on the luminal side of cytotoxic granules in unactivated NK cells. During granule exocytosis, the granule fuses with the plasma membrane, resulting in externalization of the granule membrane and exposing CD107a on the external surface of the plasma membrane as an indicator of NK-cell activation.^141,142,143 By contrast to other methods, the CD107a assay provides the opportunity for measuring NK-cell responses at the single cell level, isolating triggered NK cells,¹⁴⁴ and possibly simultaneously assessing other NK-cell functions.

Activated NK cells and CTLs also induce perforin-independent target cell killing by expressing Fas ligand (tumor necrosis factor [TNF] superfamily 6) that binds Fas (TNF receptor superfamily 6, TNFRSF6) on the target, triggering apoptosis.^{145,146,147,148,149,150} Similarly, other TNF superfamily members, such as TNF-related apoptosis-inducing ligand (TRAIL TNFSF10), can be involved in related processes.¹⁵¹ However, mice deficient in TNF family members or their receptors may manifest significant alterations in lymphoid organogenesis and splenic architecture, and NK cell number and function,^152,153,154 such that the relative contributions of these pathways to NK-cell function are incompletely understood. Moreover, NK cells from mice deficient in perforin, granzymes, or molecules involved in granule formation or exocytosis demonstrate profound defects in natural killing in vitro.^{155,156,157,158} Similar defects have been found with NK cells derived from patients with deficiencies in granule exocytosis.¹²⁸ Thus, the available data strongly suggest that granule exocytosis is the predominant mechanism for natural killing.

In addition to natural killing, cytotoxicity by NK cells can be triggered by deliberate cross-linking of activation receptors (discussed in greater detail in the “Activation Receptors” section). Plant lectins can also trigger target killing.¹⁵⁹ In general for all stimuli, cytotoxicity occurs via granule exocytosis and can be measured with the same assays for natural killing.

When exposed to NK-sensitive targets or cross-linking of receptors, NK cells also produce cytokines, including IFNγ, TNFα, and granulocyte-macrophage colony stimulating factor (GM-CSF).^160,161,162 They can also be similarly triggered to produce chemokines, such as RANTES, lymphotactin, MIP-1α, and MIP-1β.¹⁶³ Moreover, NK cells produce cytokines in response to other cytokines. For example, in response to IL-12, NK cells produce IFNγ.¹⁶⁴ Similarly, NK cells respond to type I IFNs (IFNα/β) produced by DCs stimulated by in vivo administration of poly-I:C and other ligands for TLR and nucleic acid sensors.¹⁶⁵ Cytokine-stimulated responses may obscure detection of specific activation by activation receptors in vivo.^163,166

While cytokine production can be indirectly measured with RT-PCR for mRNA, it should be noted that resting NK cells typically already express abundant levels of cytokine mRNA even though the proteins are not synthesized,¹⁶⁷ as described previously for granule components in mouse NK cells.²⁹ Enzyme-linked immunosorbent assays (ELISA) of tissue culture supernatants are often used, but recent studies have utilized intracellular staining of cytokines, such as IFN, for analysis of individual NK-cell responses that may be more informative, akin to use of the CD107a degranulation assay.¹⁶⁸

In immune responses, NK-cell production of cytokines should occur relatively early and may thereby influence the subsequent adaptive immune response. Moreover, their responses to cytokines are regulated by complex interacting pathways.¹⁶⁹ A fuller description of NK-cell cytokine responses and production is provided in the following sections on NK cell responses during infections and interactions with DCs.

Whereas initial studies suggested that natural killing was “non-MHC-restricted,”¹¹ substantial progress in understanding NK-cell recognition began with ascertaining the role of MHC-I molecules in natural killing (Fig. 17.1). Kärre and colleagues discovered that MHC-I-deficient tumors remained susceptible to in vivo rejection, apparently by NK cells.¹⁷⁰ Conversely, target cell expression of MHC-I molecules appeared to have a protective effect against NKcell -mediated lysis in vitro. A number of methods, such as IFNγ treatment, to upregulate MHC-I correlated with target protection but other effects could not be excluded.¹⁷¹ There was significant variability in capacity of specific MHC-I molecules to protect targets^172,173; in vitro culture conditions could influence NK-cell specificities,¹⁷⁴ and the specificities of individual human NK cell clones were not easily assignable to specific MHC-I alleles.¹⁷⁵ Thus, the MHC-I effect on natural killing was controversial for some time.

Several groups, however, observed that MHC-I-expressing parental targets were resistant to natural killing, whereas mutants selected for absence of MHC-I expression became susceptible. The parental (resistant) phenotype could be restored by reconstitution of MHC-I expression by transfected expression of molecules to correct the defect, be it β₂-microglobulin (β2m)¹⁷⁶ or transporter associated with processing (TAP).^177,178 Studies utilizing mice with a targeted mutation in the β2m gene added substantial support to the MHC-I protective effect, as normal expression of MHC-I heavy chains requires β2m.^179,180 β2m-deficient lymphoblasts were susceptible to lysis by normal NK cells. Moreover, β2m-/- BM transplanted into otherwise syngeneic normal hosts was rejected by recipient NK cells.^180,181 These results resembled hybrid resistance whereby NK cells in irradiated F₁ hybrid mice reject parental BM transplants.¹⁸² Hybrid resistance is regulated by parental determinants that are genetically linked to the MHC-I region, H-2D.¹⁸³ Thus, in several distinct NK-cell recognition systems, the target cell expression of certain MHC-I molecules correlated with resistance to
natural killing whereas absence of MHC-I was associated with susceptibility to NK cells.

NK cells, therefore, have a different relationship to target cell MHC-I molecules than MHC-I-restricted CTLs (see Fig. 17.1). Strictly speaking, NK cell lysis is “non-MHC-restricted,”¹¹ at least as far as MHC restriction is precisely defined for T cells having a requirement for specific self-MHC molecules presenting a given peptide antigen.¹⁸⁴ However, the term “non-MHC-restricted” (and its synonyms) is now somewhat outdated because it implies, when viewed in a broader sense, that MHC plays no role in NK-cell cytotoxicity. Avoidance of these terms will minimize confusion concerning the relationship of target cell MHC-I molecules with NK-cell specificity.

As initially observed and discussed by Kärre in the “missing-self” hypothesis, the relationship between target expression of MHC-I and resistance to natural killing highlights a fundamental distinction between NK and T cells¹⁸⁵ (see Fig. 17.1). Whereas T cells are triggered by detection of “foreign” epitopes, Kärre proposed that NK cells are equipped to detect the absence of “self” epitopes. The [missing-self] hypothesis suggests that NK cells survey tissues for expression of MHC-I molecules that are normally ubiquitously expressed and that somehow prevent NK-cell activity. If MHC-I molecules are downregulated or mutated, NK cells can then lyse the target. The generally opposite requirements of NK and T cells for target cell MHC-I expression may be physiologically important. Several pathogens, including herpes viruses, possess mechanisms that prevent the normal expression of MHC-I molecules on infected cells, providing means to avoid MHC-I-restricted T cells.¹⁸⁶ Moreover, tumorigenesis is frequently associated with alterations in MHC molecules, either mutation in structural genes or decreased expression, again leading to escape from T-cell surveillance.^187,188,189 In either case, however, the MHC-I-deficient cells should become more susceptible to natural killing. The host, therefore, is endowed with two components (T and NK cells) with opposing requirements for self-MHC-I expression. This fail-safe system should eliminate pathologic processes that might otherwise evade immune responses by any alteration of MHC-I expression (either increased to avoid NK cells or decreased to avoid T cells). The missing-self hypothesis thus provided a tentative physiologic explanation for MHC-I-associated resistance, creating a framework for initial attempts to define NK-cell recognition of their targets.

The MHC-I-associated resistance to natural killing inspired a panoply of models and their variants to explain not only resistance but also natural killing.¹⁷² The target interference or masking model predicted that a single NK-cell receptor activates natural killing when it engages its putative target cell ligand.¹⁸⁵ MHC-I molecules mask the putative target cell ligand and block its recognition by the NK-cell receptor. This hypothesis was initially favored because it was the simplest.¹⁹⁰ Moreover, it made the most sense if one considered that NK cells should have only one defined receptor analogous to the TCR. The effector inhibition or inhibitory receptor model suggested that NK cells are inhibited from natural killing by an NK-cell receptor that binds MHC-I on the target and delivers negative signals overriding a default pathway of activation.¹⁸⁵

Although the target interference model has not been refuted, it is now known that NK cells express inhibitory receptors that physically bind MHC-I in a manner that may be influenced by MHC-bound peptides¹⁹¹ (Fig. 17.2).
All known inhibitory receptors contain cytoplasmic immunoreceptor tyrosine-based inhibitory motifs (ITIMs) consisting of V/I/L/SxYxxL/V (single amino acid code where x is any amino acid).¹⁹² Ligand engagement leads to phosphorylation of the ITIM, which leads to inhibition, traditionally thought to be due to recruitment and activation of intracellular phosphatases, although recent evidence suggests more complexity.

MHC-I inhibitory receptors on NK cells have either of two general structures¹⁹³: 1) C-type lectin-like receptors that are disulfide-linked dimers with type II transmembrane topology (extracellular carboxyl termini). These receptors are encoded in the NK gene complex (NKC) and were first described in mice. 2) Immunoglobulin (Ig)-superfamily receptors that have type I transmembrane orientation. These molecules are encoded in a different genetic region, termed the leukocyte receptor complex (LRC), and were first described in humans. Although ongoing studies indicate that both structural types of receptors are expressed on mouse and human NK cells, the lectin-like receptors (Ly49 receptors) are the major MHC-specific inhibitory receptors in mouse whereas the Ig-like receptors (killer Ig-like receptors [KIRs]) predominate in human.

The absence of MHC-I does not always result in killing, indicating that release from inhibition does not result in activation by default (Fig. 17.2B). Instead, it was suggested that NK cells express two functionally different receptors for target cell ligands.^194,195 In this two receptor model, one receptor triggers activation upon ligand binding (Fig. 17.2C) whereas the MHC-I-specific receptor inhibits activation by negative signaling. In many circumstances, the inhibitory receptor effect dominates over the activation receptor (Fig. 17.2D), but the outcome usually reflects the integration of signals from both types of receptors, which can be affected by ligand expression or affinities (not shown).

Many, but not all, NK-cell activation receptors are encoded in the NKC and LRC, having similar structural properties as their inhibitory receptor counterparts except for absence of cytoplasmic ITIMs. The activation receptors typically do not have signaling motifs in their cytoplasmic domains but contain charged transmembrane residues that facilitate association with reciprocally charged residues in the transmembrane domains of signaling chains having immunoreceptor tyrosine-based activation motifs (ITAMs) analogous to ITAMs in TCR and B-cell receptor (BCR) complexes (D/ExxYxxL/Ix_6-8YxxL/I). NK cells express three ITAM-containing signaling chains: CD3ζ, FcεRIγ, and DAP12 (DNAX associated protein of 12 kDa, also known as killer activating receptor-associated protein, KARAP; Ly83; tyrosine kinase binding protein, Tyrobp). NK cells also express DAP10 (hematopoietic cell signal transducer, Hcst) that lacks ITAMs and instead contains a motif for recruitment of phosphatidylinositol 3-kinase (PI3K) and Grb2. The signaling chains typically provide two major functions: facilitate cell surface expression of the associated activation receptor, and transduce signals.

To date, the ligands for activation receptors fall into several major groups. One group is encoded by the host and is expressed normally. Presumably, NK-cell attack against cells expressing these ligands is limited by inhibitory receptors (Fig. 17.2D). Another group of ligands is characterized by their relatively low expression on normal tissues and induced expression under “stress” conditions (Fig. 17.2E). Other ligands become exposed when tissue architecture is altered (Fig. 17.2F,G). Because the ligands are encoded in the normal host genome, they would be recognized by the NK cell as indicators of pathologic conditions, either as “induced-self” or “exposed-self,” respectively. Another group of ligands is found on infected cells and is encoded by the pathogen (not shown but similar to Fig. 17.2C or E).

Finally, many other NK-cell receptors have been discovered that do not fall neatly into the categories described here. Some appear to have similar inhibitory function as the MHC-specific inhibitory receptors but bind non-MHC ligands, strongly suggesting MHC-independent self-recognition. The function of these and other receptors remains under intense investigation.

The mouse and human MHC-specific inhibitory receptors are remarkably different in protein structure. Each will be discussed separately.

The Ly49A receptor was the first inhibitory MHC-I-specific NK-cell receptor to be described in molecular terms.^191,196 Ly49A was originally identified as a molecule of unknown function on a T-cell tumor.^197,198 It is a disulfide-linked homodimer (44 kDa subunits) with type II membrane orientation, and C-type lectin superfamily homology.^199,200 Previously termed Ly49, it is now appreciated that Ly49A (Klra1) belongs to a family of highly related molecules.^{195,201,202,203} Indeed, genetic analysis revealed that the genes for Ly49A and NK1.1 are linked in the NKC (Fig. 17.3), leading to studies indicating that Ly49A is constitutively expressed on a distinct subpopulation (20%) of NK cells in C57BL/6 mice.²⁰³

Multiple lines of evidence indicate that Ly49A is an inhibitory receptor specific for MHC-I, particularly H2D^d: 1) Functional analysis: the Ly49A+ NK-cell subset were equivalent to Ly49A- NK cells in killing several targets, but they could not lyse a large panel of targets that were readily lysed by Ly49A- NK cells. This phenotype was related to MHC-I expression of certain H2 haplotypes on target cells.^191,204,205 Transfected expression of H-2D^d selectively rendered a susceptible target resistant to natural killing by Ly49A+ NK cells. Moreover, killing through disparate stimuli by Ly49A+ NK cells was also inhibited. 2) Cell binding: Ly49A+ tumor cells bound specifically to immobilized MHC-I molecules²⁰⁶ and to H-2D^d-transfectants.²⁰⁷ 3) Antibody blocking: F(ab′)₂ fragments of mAb directed against either Ly49A or the α1/α2 (but not the α3 domain) of H-2D^d reversed resistance in killing experiments (permitted lysis) and blocked the cell binding assay.^{191,204,206,207} 4) In vivo expression: the apparent level of Ly49A expressed per NK cell was downregulated in MHC congenic and Tg mice expressing H-2D^d.^208,209,210 This was not due to negative selection because the percentage of Ly49A+ NK cells was unchanged. 5) Gene transfer: primary NK cells and T cells expressing a Ly49A transgene and a Ly49A-transfected NK cell line were specifically inhibited by H-2D^d.^211,212 6) Inhibition by Ly49A is ITIM-dependent, based on gene transfer of mutant Ly49A molecules.²¹² 7) H2D^d tetramers bind Ly49A transfectants.²¹³ 8) Ly49A tetramers bind H2D^d on transfected cells.^214,215 9) Biophysical studies: recombinant Ly49A binds recombinant H2D^d in surface plasmon resonance (SPR) studies with K_D = ˜2.0 µM.²¹⁶
10) Crystallography: the structure of Ly49A complexed with H2D^d was determined.²¹⁷ 11) Less extensive studies also indicate that Ly49A recognizes H-2D^k, H2D^p, and alleles in H2^r and H2^q.^{191,208,213,218,219} Therefore, Ly49A is an MHC-I-specific receptor for H-2D^d, H-2D^k, and H2D^p, and alleles in H2^r and H2^q.

The nature of the Ly49A interaction with MHC-I, however, is fundamentally different from TCR/MHC-I interactions because the former appears to be relatively independent of the specific peptide bound by H2D^d.^220,221 However, bound peptides are required for appropriately folded MHC-I molecules that can be recognized. Despite its structural homology to C-type lectins that are carbohydrate-binding proteins,^222,223 Ly49A does not have the residues for coordinate binding to calcium that is required for lectin binding. Moreover, Ly49A binding to its MHC ligands is not carbohydrate-dependent based on functional, SPR, and crystallographic analyses.^216,224 In the crystallographic structure of Ly49A complexed to H2D^d (2.3Å resolution), the lectin-like structure of Ly49A was confirmed²¹⁷ (Fig. 17.4). Two interaction sites were seen between the lectin-like domain of Ly49A and H2D^d: site 1 involved the “left” side of the peptide-binding cleft of H2D^d and a wedge-like site 2 involved the undersurface of the peptide-binding cleft. The residues in Ly49A involved in binding either ligand site are overlapping. Mutational analysis revealed that Ly49A binds site 2 where it contacts α1, α2, and α3 of H2D^d and β2m.^214,216,225 This site is near Asn80, an Asn-linked glycosylation site conserved in all MHC-I molecules, leaving open the issue of whether carbohydrates could affect the interaction, such as affinities or kinetic parameters, but this has not been studied in depth. These studies also provide a structural explanation for species-specific β2m requirements as revealed by functional studies.²²⁶ Thus, Ly49A recognizes site 2 in the MHC molecule in terms of trans recognition between the NK-cell receptor and target cell MHC-I molecule.

Most but not all other Ly49 receptors are MHCI-specific inhibitory receptors. First noted by Southern blot analysis and cDNA cloning, genome sequence analysis revealed 15 complete Ly49 genes in C57BL/6 mice.^{199,200,201,202,227} There is evidence for alternative splicing and alternative transcriptional start sites for the Ly49 genes, though their importance has not been elucidated.^202,228,229 Notably, Ly49C has broad specificity for H2 alleles, as revealed by tetramer staining, and is the only known inhibitory NKcell receptor specific for an H2^b haplotype allele (H2K^b) in C57BL/6 mice, notwithstanding unconfirmed reports that Ly49I also binds H2K^b.^168,213,230 Ly49C is recognized by two mAbs: 5E6, which also binds Ly49I, and 4LO3311, which has exquisite specificity for Ly49C.^231,232,233 X-ray crystallographic studies have indicated that Ly49C binds H2K^b in a manner similar to Ly49A interaction with H2D^d (see Fig. 17.4).^234,235 Only a site 2 interaction was seen with the contact residues, showing a similar but distinct topology to the Ly49A-H2D^d interaction. Interestingly, however, peptides bound to H2K^b clearly affect functional interactions with Ly49C²³⁶ and affinities as measured by SPR.²³⁴ However, Ly49C does not directly engage the peptide, indicating long-range effects. Finally, both Ly49A and Ly49C
appear to undergo conformational changes upon ligand binding, and a Ly49 dimer can engage two MHC-I molecules.^235,237 Thus, Ly49 receptors bind their MHC ligands in a structurally related manner.

Recent studies suggest that Ly49 molecules also can bind MHC in a cis interaction between receptor and ligand on the NK cell itself.^168,238 For example, an MHC ligand for Ly49A or Ly49C on the same cell prevents binding of MHC tetramer.^168,238 If the cells are briefly exposed to mild acidic conditions, MHC-I expression is lost (due to disruption of the noncovalently linked MHC-I heterotrimer), and Ly49A binding to cognate MHC-tetramers is restored. This cis interaction is dependent on site 2 residues in the MHC molecule. These findings may help explain the observation that the presence of self-MHC ligands leads to downregulation of Ly49 expression, as previously noted on primary NK cells in MHC congenic mice.^208,209,210 Cis interactions may also explain functional differences in NK cells in MHC-congenic mice or NK cells that do or do not express MHC ligands in Tg mice that are mosaic for MHC expression.^239,240,241 Finally, there are biophysical data supporting a role for the relatively long, flexible stalk region of Ly49 receptors in allowing either trans or cis interactions.²⁴² At the moment, the physiologic importance of cis interactions is incompletely understood but may be relevant to NK-cell tolerance and education, as discussed below.

Although they have not been studied as extensively as Ly49A and Ly49C, other inhibitory Ly49 receptors and their ligands have been identified^202,230 and characterized with other MHC allele specificities, as detailed in a recent review.²⁴³ Moreover, they are structurally related.^235,244 Individual NK cells may express multiple Ly49 receptors simultaneously,^210,233,245 often (but not always) two or more, suggesting that individual NK cells may be inhibited by more than one MHC-I molecule.

Ontogenetic studies demonstrate that the total repertoire of Ly49 expression does not reach adult levels until sometime after 3 weeks of age, concomitant with attainment of full NK-cell cytolytic activity.²⁴⁶ Thereafter, the expression of Ly49 receptors is generally thought to be fixed and stable on an individual NK cell. Ly49E is expressed only on fetal NK cells, but NK cells in mice deficient in Ly49E are otherwise normal.²⁴⁷ In adult mice, developmental studies indicate that Ly49 receptors are first expressed on immature NK cells in the BM, before a phase of constitutive proliferation⁹⁸ that appears to be modestly affected by MHC haplotype.¹⁶⁸ There are only modest effects on the final “repertoire” of MHCspecific receptors expressed by splenic NK cells in different MHC environments.^210,248

The expression of inhibitory Ly49 receptors appears to occur in a stochastic manner. There is evidence for monoallelic expression of Ly49 receptors (expression from one chromosome), initially described as “allelic exclusion,” a term that has fallen out of favor because it has a specific meaning and mechanism for TCRs and BCRs.²⁴⁹ At least some Ly49 genes possess bidirectional, overlapping promoters directed in opposite orientations.²⁵⁰ Transcription factors driving transcription in one direction prevent binding of other factors driving transcription in the opposite direction. Directionality and monoallelic expression may also be controlled by DNA methylation.²⁵¹ A “probabilistic” model has been proposed to explain these findings that may also explain the stochastic expression of Ly49 genes and their stable expression. However, recent studies of Ly49 indicate highly variable transcriptional start sites, suggesting that the probabilistic model may not be correct.²²⁹ Other data indicate that TCF-1 but not LEF-1 in the T-cell factor/lymphoid enhancer family of DNA-binding proteins affects some but not all Ly49 receptor expression.^252,253,254 Thus, the elements controlling Ly49 gene expression are incompletely understood.

Analysis of the Ly49 receptors thus far is largely based on examination of the C57BL/6 alleles, but the Ly49 receptors display extensive polymorphism. The Ly49 family is encoded in the NKC located on mouse chromosome 6 with the syntenic human region being chromosome 12p13.2^{195,203,255,256} (see Fig. 17.3). While the NKC also contains genes for other lectin-like receptors, the Ly49 genes are clustered with the exception of Ly49b. Corresponding to restriction fragment length polymorphic (RFLP) variants originally detected with the Ly49A cDNA,²⁰³ there is significant allelic polymorphism of the Ly49 cluster between inbred mouse strains with differences in gene number as well as alleles for the Ly49 genes.^{227,257,258,259} In contrast to C57BL/6J mice, genomic sequence analysis shows 8 putative Ly49 genes in BALB/c mice, 19 in 129 mice (of which at least 9 appear to be pseudogenes) and 22 in NOD mice. Array-based comparative genomic hybridization analysis of 21 mouse strains compared to the reference C57BL/6J strain indicated that these mice could be grouped into five clusters that correspond to or are predictive of restriction fragment length polymorphic patterns on Southern blot analysis.^203,260 There are also multiple alleles for individual Ly49 family members.^{227,249,257,258,261} Thus, there is significant polymorphism of the Ly49 molecules at both the haplotype (gene numbers) and individual gene (alleles) levels, not unexpected because the Ly49 molecules bind highly polymorphic MHC-I molecules.

The MHC-I specificities have generally been well characterized for only a few Ly49 alleles. Interestingly, mAbs specific for one Ly49 allele may bind another molecule with a different function or specificity in another mouse strain,^262,263,264 similar to what was recognized for mAb reactivity with different MHC alleles.²⁶⁵ Thus, the polymorphisms also raise practical issues when studying Ly49 molecules in different mouse strains.

Finally, it should be noted that Ly49 receptors may be expressed by other cells; some are selectively expressed on non-NK cells and some may have specificities for non-MHC ligands. NKT and other T-cell subsets may express Ly49 receptors but they have not been thoroughly studied.^266,267 Ly49B and Ly49Q are not expressed on NK cells; rather, they are expressed on myeloid cells.^268,269 Interestingly, Ly49Q recognizes H2K^b and positively regulates TLR signaling.^270,271 On the other hand, Ly49E appears to recognize urokinase plasminogen activator, though physical binding has not been established.²⁷² Interestingly, Ly49B, Ly49E, and Ly49Q are predicted to be distinct in fine structure from the
known MHC-I-specific Ly49 receptors on NK cells.²³⁵ Thus, while Ly49 receptors are predominantly NK-cell inhibitory receptors for MHC class I, they may have other roles that are less well understood; still other Ly49 receptors have activation function, as discussed subsequently.

As already mentioned, NK cells also express inhibitory receptors for non-MHC ligands, such as mouse gp49b, which binds the αvβ3 integrin,³²³ and there is a growing list of other such molecules.⁴¹⁶ Some of these receptors will be discussed in a more appropriate context in the following sections. Most of these receptors contain cytoplasmic ITIMs so their inhibitory function can be predicted even if not directly tested, though some caution is required because the motifs may be involved in other signaling processes.

Human and mouse NK cells express the ITIM-bearing leukocyte-associated Ig-like receptor 1 (LAIR-1, CD305), which is an Ig-like molecule broadly expressed by most leukocytes and encoded in the LRC.^417,418 Initial reports indicating that LAIR-1 binds epithelial cellular adhesion molecule were irreproducible.^419,420 Instead, LAIR-1 binds multiple forms of collagen,⁴²¹ which has been validated in crystallographic and biochemical studies.⁴²² Interestingly, LAIR-1 mediates inhibition that is independent of Src homology 2 (SH2)-domain-containing phosphatases, and instead recruits C-terminal Src kinase (Csk),⁴²³ suggesting that Csk may be involved in inhibitory signaling. Although the in vivo context for functional interaction awaits further characterization, it is reminiscent of the broader reactivity of the Siglecs.

The CD33-related sialic acid binding Ig-like lectins (CD33rSiglecs) are type I receptors with varying numbers of Ig-like domains expressed on a broad array of cells and encoded in the “extended” LRC^312,424,425 (see Fig. 17.3). Despite having sialic acid recognition in common, the Siglecs appear to show differences in carbohydrate recognition, depending on the specific glycan context.⁴²⁴ Human NK cells express Siglec-7 (p75, adhesion inhibitory receptor 1 [AIRM1]) and Siglec-10, whereas some mouse NK cells express a related Siglec-E.⁴²⁵ Siglec-7 has been most extensively studied. As expected, its cytoplasmic ITIM can recruit SHP-1 and inhibit NK-cell functions.^426,427 Moreover, expression of its ligand on targets inhibits NK cells in a Siglec-dependent manner.⁴²⁸ However, the effects appear to be modulated by cis interactions between the Siglec receptor and its carbohydrate ligands on the NK cell itself,^424,428,429 reminiscent of cis interactions between Ly49 and MHC ligands, as previously discussed.

The Ig-like receptors, termed paired Ig-like receptor (PILR)α and β, are not encoded in the LRC, rather on human chromosome 7 (mouse chromosome 5).^430,431 Mouse PILRα is an ITIM-containing inhibitory receptor, whereas PILRβ is an activation receptor that couples to DAP12.^430,432 Both receptors are expressed on NK (and other immune) cells and recognize CD99. Interestingly, sialylated O-linked glycans on CD99 are involved in recognition by PILRs.⁴³³

Thus, NK cells (and other leukocytes) express multiple inhibitory receptors that are capable of MHC-independent recognition. How they participate in NK-cell responses and contribute to MHC-dependent effects are beginning to be elucidated, and some appear to play a specific role in the context of activation receptors, as discussed in the following.

NK cells clearly kill MHC-I-deficient targets more efficiently than MHC-I-sufficient targets. However, this enhanced killing does not occur simply because a nonspecific default pathway is unleashed when MHC-I is absent. Instead, it is clear that susceptible targets express ligands for NK-cell activation receptors. In general, these receptors and their ligands were defined following description of the inhibitory receptors.