Main

“The intensity of the conviction that a hypothesis is true has no bearing on whether it is true or not.”

Peter Brian Medawar

Cells communicate by producing soluble molecules or by expressing cell surface ligands. These signals are recognized by receptors on separate cells and transformed into biochemical information by intracellular adaptors, molecular scaffolds, second messengers and enzymes that regulate cellular responses to the signals. Natural killer (NK) cells are specialized in detecting aberrant cells in the body. Cells that have been infected, transformed or “stressed” can pass this biological information onto NK cells, which respond by killing them or by generating cytokines and chemokines, which activate other components of the immune system. Thus, NK cells act as a first line of defense before the development of adaptive immunity by B and T lymphocytes. Ironically, investigators have focused more on the major mechanisms that can prevent NK cell activation1,2, rather than on the mechanisms that activate them. Therefore, the receptors and intracellular pathways that activate NK cells are still incompletely known. In addition, only a few of the molecular motifs that can signal the presence of infection or malignancy to NK cells have been identified. Beside these intellectual challenges, we must also face up to our frustrating inability to manipulate NK cells for effective therapeutic use. When NK cells were discovered in 19753,4, they were soon thought to be the perfect cellular weapons in the fight against cancer. To date, however, there have been few clinical successes5,6,7, mainly because of our incomplete understanding of the biology of NK cell activation.

The most commonly used techniques for dissecting the roles of a molecule involved in signal transduction are overexpression of either the wild-type molecule or specific mutants, inactivation of the gene that encodes the wild-type molecule, and pharmacological perturbation of the molecule's activity. As is the case for all experimental designs, these three approaches each have their own strengths and limitations8. Overexpression technologies are well suited to dissect the functions of different domains of the signaling molecules. However, for any given pathway being studied, functional domains may be shared by other proteins capable of modulating that pathway. Therefore, potential loss of specificity is a major limitation. Pharmacologic inhibitors may selectively block enzymatically active second messengers in a dose-dependent manner, but their specificity decreases as the dose is further increased. Specificity is the strength of classical genetic manipulations in laboratory animals such as flies, worms and mice. These genetic approaches are successful when they target an essential component of a pathway, but the presence of functionally redundant gene products—a common theme in vertebrates—can obscure the role of the targeted gene. In addition, the organisms may adapt to the functional loss of one pathway with the use of parallel ones.

Studies of intracellular signaling in NK cells have to contend not only with these technical limitations, but also with the complexity of NK cell biology. The heterogeneous nature of the stimuli that can activate NK cells, the absence of a single clonotypic antigen receptor and the diversity of ligands and receptors involved in different NK-target interactions make it difficult to directly compare results obtained with different techniques. To complicate matters further, the overexpression of signaling molecules is generally done in human NK cells, whereas gene inactivation is often done in mice. When the two approaches are used to study the same molecule in parallel, they may result in disconcordant conclusions9,10,11,12, tempting the investigator to solve the controversy by assuming species differences exist. However the differences may be deceptive and they could reflect instead a bias in the experimental approaches. The intrinsic complexity of signal transduction is best tackled when data obtained with the various techniques are integrated8.

We propose that, despite some well documented differences between mouse and human NK cells—for example major histocompatibility complex (MHC)-recognizing C-type lectin Ly49 receptors in mice versus killer cell immunoglobulin (Ig)-like (KIR) receptors in humans1—the intracellular signaling pathways that activate NK cells are generally conserved. In line with the quote from Medawar, we will critically review here the literature on the subject, discussing evidence that validates or disproves our hypothesis. For a summary of the known activating receptors on mouse and human NK cells see Table 1 and Fig. 1a.

Table 1 Activating receptors and their ligands in human and mouse NK cells
Full size table
Figure 1: Molecules that regulate NK cell activation.
figure 1

Molecules shared by humans and mice are shown in black, whereas species-specific molecules are indicated in blue (human) or red (mouse). (a) NK cells can be activated upon contact with normal, stressed or cancer cells and during infection. These situations are separated for clarity in this figure, but the boundaries in real-life are not so well defined. For example, cytokines may activate cells during antitumor immunity as well as during infection. Some viral products, such as MCMV gp40, can prevent interaction of the activating NKG2D receptor with the H-60 ligand expressed by infected cells. Similarly, HCMV-derived UL16 prevents the interaction of NKG2D with the ULBP1, ULBP2 and ULBP3 ligands expressed on infected cells. (b) Classical MHC class I molecules (HLA in humans, H-2 in mice) and nonclassical MHC molecules (HLA-E in humans, Qa1b in mice) inhibit NK cells upon contact with normal or diseased cells. As well as the MHC molecules, there are a number of molecules that can be expressed by infected cells, which allow the pathogen to escape the immune system by inhibiting NK cells. For example, the product of the MCMV m157 gene can bind to the Ly49I inhibitory receptor. m144 is a class I–like protein produced by MCMV that may act similarly. UL18 is a class I–like protein produced by HCMV that binds to the inhibitory receptor ILT2. HCMV UL40 increases the expression of nonclassical HLA-E class I molecules on infected cells; it binds to NKGA, resulting in the inhibition of NK cells.

Full size image

Strategy of recognition by NK cells

NK cells are considered effector cells in innate immunity13,14. They are called into action once disease processes have been initiated. Because of their capacity for cytotoxicity and pro-inflammatory cytokine release, they help to eliminate “altered” host cells. Strategies for NK cell recognition of diseased cells during infection, cancer or cellular “stress” involve the activating and inhibitory receptors that recognize “self”, “missing self” and “altered or stressed self”. NK cells have also the peculiar capacity to be potentially activated upon interaction with normal cells (Fig. 1).

An emerging concept is that NK cells can also directly recognize pathogen-derived products. One of the best examples of this is the role played by distinct NK cell subsets in resistance to murine cytomegalovirus (MCMV) infection. Early after MCMV infection, viral replication is controlled in the liver by recruitment of NK cells, and resistance to infection is mediated—at least in some strains of mice—by NK cells bearing the activating Ly49H molecule15,16,17,18. Recently, it was shown that a viral gene product (m157) expressed on the cell surface of MCMV-infected cells was capable of binding to an inhibitory NK cell receptor (Ly49I) in MCMV-susceptible strains, whereas m157 bound and activated Ly49H-bearing NK cells in MCMV-resistant strains18. These results define the receptor-ligand pairs that can critically control NK cell responses to infected target cells. One results in viral escape, whereas the other leads to viral containment. Both human CMV (HCMV) and MCMV have evolved multiple strategies for escaping NK cells (Fig. 1), including MHC class I orthologs m144, gp40 (encoded by m152), UL-18 and UL-16, which prevent NK cell activation by different means19,20,21. Perhaps the struggle that viruses have had to come to terms with NK cells gives us a hint of the importance of these cells in antiviral immunity.

Another example of direct pathogen recognition is the specific binding of NKp46 and NKp44 molecules to hemagglutinin derived from influenza and sendai viruses, at least in an in vitro system22. This interaction requires expression of sialic acid residues on NKp44 and NKp46. However the role that this recognition plays in vivo is still unknown.

If NK cells can directly recognize pathogens, do they play a decisive role in the initiation of immune responses, distinct from their roles as effector cells? In the case of MCMV, it is clear that NK cells are actively recruited into the sites of viral replication through locally produced chemokines and after up-regulation of adhesion molecules on inflamed vasculature. Toll-like receptors (TLRs) are broadly expressed on leukocytes and act as molecular receptors for a variety of pathogen-encoded products. The engagement of TLRs on macrophages and dendritic cells (DCs) plays a critical role in the initiation and orientation of immune responses. The expression of TLR1, TLR6 and TLR9 on human NK cells23 indicates that they may participate in triggering immune responses. Alternatively, activation of TLRs on NK cells may serve to amplify NK cell effector functions directly or through the paracrine activation of macrophages and DCs.

ITAM-initiated signals in NK cells

The activation of many cell types within the immune system occurs through multisubunit immune recognition receptors expressed on the cell surface, which are noncovalently associated with one or more transmembrane adaptor molecules. A common feature of these adaptor molecules is that they contain an aspartic residue within their transmembrane region that is involved in noncovalent association with the ligand-binding receptor subunit. In addition, the cytoplasmic tail of these adaptors contains one or more copies of the semiconserved peptide sequence YxxL(I)x6–8YxxL(I)(where x denotes any amino acid), which is commonly referred to as an immunoreceptor tyrosine-based activation motif (ITAM)24. The transmembrane adaptor proteins on NK cells that contain ITAMs include CD3ζ, FcεRIγ and DAP1225,26. They participate in the signal transduction that emanates from a variety of activating receptors on NK cells (Fig. 2). Some receptor-adaptor associations are conserved, whereas others are species-specific. FcγRIII (also known as CD16) triggers antibody-dependent cellular cytotoxicity (ADCC) in both human and mouse NK cells. In human NK cells it associates with homodimers and heterodimers of FcεRIγ and CD3ζ, whereas in mouse NK cells it can only associate with FcεRIγ homo-dimers. Natural cytotoxicity receptors (NCRs), NKp46, NKp44 and NKp30, activate natural cytotoxicity upon contact with tumor cells, normal cells and possibly microbial molecules26. Their signal transduction has been studied in human NK cells, where they associate with CD3ζ homodimers (NKp46 and NKp30), FcεRIγ-CD3ζ heterodimers (NKp46) or DAP12 homodimers (NKp44)26. Although NKp46 is expressed by mouse NK cells and the gene encoding NKp30 is present in the mouse genome (whereas the gene encoding NKp44 is absent), their signal transduction in mouse NK cells is unknown. Finally, some receptors for MHC can trigger natural cytotoxicity by associating with DAP12 homodimers (CD94-NKG2C in both humans and mice, which activate KIRs in humans and Ly49s in mice).

Figure 2: Major biochemical pathways for NK cell activation.
figure 2

NK cells have multiple choices for activation70. Molecules shared by humans and mice are shown in black, whereas species-specific molecules are shown in blue (human) or red (mouse). The figure presents an oversimplified view of the signaling molecules that can activate NK cells. Cytokines, immune complexes, adhesion molecules, classical and nonclassical MHC molecules, pathogen-associated ligands and stress-induced ligands initiate multiple and possibly redundant pathways that culminate in activation of cytotoxicity and/or production of cytokines. The different pathways may converge on shared signaling modules. For simplicity, some receptors are grouped together (for example, NKp46 and KIR2DS), regardless of whether they form heterodimers or not. NKp46 has mostly been studied in human NK cells, but mouse NK cells also express this receptor. The gene encoding NKp30 is present in the mouse genome but its expression pattern in mouse NK cells is unknown.

Full size image

The prototypic FcγRIII receptor initiates proximal signal events that are characteristic of the ITAM-containing multisubunit immune recognition receptors27. Biochemical and pharmacologic analyses suggest that ligation of the FcγRIII complex on NK cells rapidly initiates a protein tyrosine kinase (PTK) cascade (Fig. 2) that includes proximal activation of Src family PTKs (which can include Lck, Fyn, Yes and Lyn), followed by the recruitment and activation of Syk family PTKs (ZAP-70 and Syk). These early signals couple to a series of critical downstream regulators that include intracellular adaptor molecules (such as linker for activation of T cells (LAT), SLP-76 and 3BP2), phospholipase C-γ (PLC-γ), phosphatidylinositol 3-kinase (PI3K), extracellular signal–regulated kinase (Erk) kinases, Vav family guanine nucleotide exchange factor and Rho-Rac low molecular weight GTP-binding proteins and their effectors.

The above described biochemical and pharmacologic analyses would suggest that the mechanisms described for ITAM-dependent activation of NK cells are similar to those regulatory events that control ITAM-dependent activation of a variety of hematopoietic cells (for example B and T lymphocytes or mast cells). However, additional genetic studies have highlighted a number of distinct features of signal transduction in NK cells. First, the ITAM-containing subunits utilized for cellular activation in NK cells can differ between species. For example, whereas the human FcγRIII complex is associated with FcεRIγ or CD3ζ homodimers and heterodimers, murine FcγRIII only uses FcεRIγ homodimers, with the resulting abrogation of ADCC in FcεRI-γ−/− mice but not in CD3ζ−/− mice28. This difference in the composition of the FcR complex may result in qualitative differences in human versus murine FcR-initiated signaling, as different ITAMs can clearly couple to different intracellular signaling molecules. In addition, despite clear biochemical and pharmacologic observations that suggest a necessary role for Src family PTKs in FcR-initiated NK cell activation, murine models with single (Lck−/−) or multiple (Lck−/− Fyn−/−) gene deficiencies show intact ADCC29. This may be due to the potential functional redundancy between different Src family members; however, this explanation can only be validated by demonstrating defective ADCC in mice where each of the potentially redundant Src family members have been deleted. Alternatively, findings obtained in hematopoietic cell types suggest that Syk (but not ZAP-70) has the potential to tyrosine-phosphorylate ITAM-containing receptors and to initiate downstream activation in a Src family PTK-independent manner30. A proximal Syk-dependent but Src-independent signaling mechanism in mice would also be consistent with the observation that CD45-deficient mice have intact FcR-initiated killing31.

Analyses of mice deficient in Syk family PTKs highlight both the complexities and strengths of genetic models. Initial studies analyzing either Syk−/− mice10 or ZAP-70−/− mice32 showed FcR-initiated killing and natural cytotoxicity. This might have prompted us to suggest that Syk family PTKs were not essential regulators of ITAM-dependent activation in NK cells. However, functional redundancy among the Syk family members may well account for the lack of major ITAM-signaling defects in Syk−/− and ZAP-70−/− mice, as the signaling from ITAM-containing receptors is profoundly impaired in Syk−/−ZAP-70−/− mice33. Early biochemical studies suggested that both PLC-γ1 and PLC-γ2 were tyrosine-phosphorylated after FcR ligation on NK cells34. However, in studies with PLC-γ2−/− mice there was a block of NK cell activation, which implied preferential utilization of PLC-γ2 rather than a functional redundancy35. The latter example illustrates the importance of integrating results obtained with different approaches. The situation is less clear with regard to the role played by the guanine nucleotide exchange factors of the Vav family in NK cell activation. Biochemical analyses suggest multiple Vav family members are expressed in NK cells (Vav1, Vav2 and Vav3)36, and genetic studies have shown that NK cells from Vav1−/− mice secrete interferon-γ (IFN-γ) normally but show impaired natural cytotoxicity and ADCC37. This leaves open the question of whether different Vav family members (for example, Vav1 and Vav2) are functionally redundant or whether they regulate separate parallel signaling pathways.

Adaptor molecules and ITAM-initiated signaling

A variety of experimental approaches have been used to characterize the roles of specific adaptor molecules (LAT, SLP-76 and 3BP2) in ITAM-initiated NK cell activation. FcR ligation of human NK cells enhances the tyrosine phosphorylation of adaptor molecules such as LAT12 and SLP-7638, and overexpression of wild-type (but not specifically mutated) adaptors enhances FcR-initiated human NK cell activation. Yet LAT−/− mice11 and SLP-76−/− mice39 are capable of mediating robust FcR-initiated NK cell–mediated killing. These observations are consistent with several potential interpretations. There could be a difference between human and murine NK cell signal transduction, and this difference may be related to the previously mentioned differences in the ITAM-containing subunits that compose their FcR complexes. Alternatively, there may be as yet unidentified NK-specific adaptor molecules that have functional redundancy with LAT and SLP-76. Direct comparisons of signaling in human versus murine NK cells, together with additional murine genetic models of signaling-deficient strains of mice, will be needed to assess these potential alternative explanations.

Signals for NK “natural cytotoxicity”

“Natural cytotoxicity” defines the capacity of NK cells to spontaneously kill targets without any prior immunization or stimulation. In contrast, ADCC is mediated by antigen-bound IgG antibodies, which are recognized by the FcγRIII on NK cells. The signaling pathways that activate ADCC have been well characterized and resemble those that regulate activation of B and T cells via their antigen receptors. Because multiple stimuli can activate natural cytotoxicity (Fig. 1a), a long-standing question has been whether the different stimuli use common signaling pathways and whether they resemble those that activate T and B cells. The molecular identification of a variety of structurally different NK cell–activating receptors has suggested that the signaling mechanisms are heterogeneous. For example, NK recognition of MCMV-infected cells in some strains of mice depends on Ly49H-mediated recognition of specific MCMV proteins15,16,17,18 and Ly49H is associated with the signal-transducing, ITAM-containing subunit DAP1240. The NKG2D receptor complex can recognize specific ligands on tumor cells, such as retinoid acid early transcript 1 (Rae-1) and the histocompatibility antigen H-60 in mice, and MICA, MICB and the UL16-binding proteins ULBP1, ULBP2 and ULBP3 in humans. NKG2D activates both human and mouse NK cells in an ITAM-independent, but DAP10-dependent, manner41. DAP10 noncovalently associates with NKG2D and uses a consensus PI3K-binding motif to transduce NKG2D-initiated signals. Therefore, different NK-activating receptors engaged in alternative modes of NK cell–mediated killing utilize distinct signal-transducing subunits that initiate different pathways, which are not necessarily similar to those that activate B and T cells (Fig. 2).

Recent observations suggest that the signaling paradigms used to characterize regulation by ITAM-containing receptor complexes are insufficient to explain signaling by non-ITAM–containing NK receptors such as NKG2D and 2B4. As previously mentioned, ITAM-containing NK receptors appear to be strictly dependent on the tyrosine kinases ZAP-70 and Syk33. Yet NK cells from Syk−/−ZAP-70−/− mice are able to generate natural cytotoxicity against a large variety of tumor cells, including those that fail to express NKG2D ligands33. Natural cytotoxicity is sensitive to Src family PTK inhibitors, suggesting a critical Src PTK–dependent but Syk PTK–independent regulatory mechanism. How proximal Src family PTK activation couples to PLC-γ tyrosine phosphorylation and calcium signaling in a Syk-independent manner remains to be determined.

Downstream from proximal PTK activation, PI3K can be a potent regulator of certain forms of NK cell–mediated cytotoxicity42. For example, there is clear consensus that PI3K regulates ADCC mediated by ITAM-containing CD3ζ and FcεRIγ, which associate with FcγRIII. However, opposite results have been reported for the role played by PI3K in natural cytotoxicity43,44. Differences in the experimental models may account for these controversies. For example, commonly used human effector NK cells include diverse cell populations such as freshly isolated large granular lymphocytes (LGLs), which are enriched in NK cells, nontransformed and interleukin 2 (IL-2)–dependent clonal NK cell lines and transformed NK-like cell lines. These effectors of natural cytotoxicity may not share the same signaling requirements. In addition, the tumor targets used—for example, K56243 versus Raji44—may engage distinct receptors on the NK cells. It is unclear which state of NK cell activation (freshly isolated LGLs versus IL-2–activated NK cell lines) or which mode of natural cytotoxicity most closely recapitulates the physiologic state of NK cells or their targets during an in vivo antiviral or antitumor immune response. Also, the relative sensitivity of ADCC and natural cytotoxicity to PI3K inhibitors may vary among different NK effector populations. Again, these observations highlight the importance of clearly defining the experimental parameters for each model system and of validating conclusions with the use of multiple approaches.

The cytoplasmic tail of the signal-transducing DAP10 subunit contains a consensus motif, YINM, which can bind the p85 subunit of PI3K45. Consistent with this observation, pharmacologic inhibition of PI3K potently blocks NKG2D-DAP10–initiated killing. Yet, the functional role of the YINM motif appears to go beyond just the direct regulation of PI3K catalytic activity. NKG2D-DAP10 ligation induces the tyrosine phosphorylation of multiple substrates (including PLC-γ), and these tyrosine phosphorylation events are insensitive to PI3K inhibitors (P. J. Leibson, unpublished data). A Y→F mutation in the YINM consensus motif blocks these PTK-dependent phosphorylation events. Therefore, additional experiments are needed to determine how the DAP10 cytoplasmic tail couples to proximal PTK activation.

The receptor 2B4 has also been implicated in certain forms of natural cytotoxicity and in the activation of IFN-γ production by NK cells46. The cytoplasmic tail of 2B4 contains a consensus motif, TxYxxV(I), for the binding of the cytoplasmic Src homology 2 (SH2)-containing adaptor protein (SLAM-signaling lymphocyte activation molecule–associated protein) SAP47. SAP was originally identified by its ability to bind SLAM, a self-ligand that promotes proliferation and IFN-γ production in activated T cells48, but it also binds other receptors containing the TxYxxV(I) consensus motif, including 2B4 and NTB-A49. Humans with X-linked lymphoproliferative (XLP) disease have a loss-of-function mutation in the gene encoding SAP that results in an NK cell defect, predisposition to severe Epstein-Barr virus (EBV) infection and EBV-induced B cell malignancies. SAP was originally thought to act as an adaptor protein and function by blocking the binding of other signal-transducing proteins. Additional reports suggest that SAP in T cells is directly involved in coupling cell-surface receptors (that is, SLAM in T cells) to proximal PTK activation50. Little is known about the 2B4-SAP–initiated signaling pathways in NK cells and their role in NK function. Although CD48 has been identified as a ligand for 2B4, it still remains unclear whether 2B4 functions as an independent activating receptor or more as a costimulatory molecule. NK cell function in 2B4−/− mice remains unknown.

A variety of adhesion and costimulatory molecules can also trigger natural cytotoxicity via alternative pathways51,52 (Fig. 2). The redistribution of ligands for integrins on target cells explains how NK cells may discriminate between normal and diseased cells53. Finally, cytokines such as IFN-α, IFN-β, IL-12, IL-18 and IL-15, which are produced during infections and antitumor immunity, are also potent activators of mouse and human NK cells (Fig. 1a)54.

Conserved signals for development and activation

Despite differences in certain cell surface receptors expressed by human and mouse NK cells (Table 1), the signals that drive NK cell development from hematopoietic precursors appear to be conserved. A step-wise model of NK cell differentiation has emerged in which stem cell factor (SCF) and/or fetal liver kinase-2 (Flk-2) ligand provide signals that promote the commitment of hematopoietic stem cells to the lymphoid lineage55. Subsequently, common lymphoid or NK cell progenitors receive a second key signal via IL-156,55. The full maturation of NK cells appears to require interactions with stromal cells in the bone marrow, although the nature of the signals provided remains undefined56,57. Supporting evidence for this model derives from two diverse series of observations: genetic approaches such as “experiments of nature” in immunodeficient human patients58 and gene targeting in mice as well as in vitro model systems of NK cell development55.

Once fully differentiated, human and mouse NK cells also show similarities in cytokine responsiveness. In general, IFN-α, IFN-β, IL-12, IL-15 and IL-18 induce the same biological effects in human and mouse NK cells, in part due to the activation of conserved intracellular signaling pathways54. The cellular partners that provide these cytokine signals to NK cells are macrophages and DCs; together these three cells operate in synergy to provide an amplification system involving tumor necrosis factor (TNF) and IFN-γ that primes innate as well as adaptive immunity in response to infection. This suggests that NK cells are part of an evolutionarily ancient cellular and molecular network, which operated before the advent of the components of adaptive immunity.

Heterogeneity of NK cells and their functions

Is the mature NK cell pool phenotypically and functionally homogeneous? The answer, from both human and mouse studies, is no. Stochastic expression and sequential expression of NK cell inhibitory receptors generates a NK cell repertoire, which is conditioned by expression of MHC molecules59. The biological impact of this repertoire likely impinges on the capacity of individuals to respond to various pathogens. Another example of functional NK cell heterogeneity is given by the two “subsets” of peripheral NK cells found in humans. A minor CD56hiCD16lo subset expresses predominantly CD94-NKG2A and c-Kit and shows potent cytokine production, whereas CD56dimCD16hi NK cells show enriched KIR expression and have potent killing activity (both natural cytotoxicity and ADCC)60. The differential expression of chemokine receptors and adhesion molecules on these subsets suggest that they have distinct biological roles61. At present, these precise human NK cell subsets have no equivalents in the mouse, although one would expect a “division of labor” among the different subsets of murine NK cells.

There are examples that suggest that cytokine production and cytotoxicity can be controlled by different biochemical pathways in NK cells. This could provide some degree of “plasticity” for NK cells, which could then select the appropriate response depending on the biological context. Upon 2B4 stimulation of the human NK cell line YT, cytotoxicity is dependent on Erk1 and Erk2 (Erk1/2) and p38, whereas IFN-γ production is p38-dependent but Erk1/2-independent49. Upon 2B4, CD16 and NK1.1 stimulation of Vav1−/− NK cells in the mouse, IFN-γ production is normal, whereas cytotoxicity is impaired37. This suggests that distinct pathways regulate the two effector functions. Such a biochemical framework is important in order to understand how NK cells could select their effector functions. For example, during MCMV infection, mice deficient in IFN-γ efficiently clear infection in the spleen but not the liver62. In contrast, mice deficient in perforin can clear infection in the liver but not the spleen62. This suggests that local signals activate differential pathways. Alternatively, different subsets of specialized NK cells seed different tissues.

IFN-γ produced by uterine NK cells is necessary for vasculature remodeling during pregnancy63. If the same signals that stimulate NK cells to secrete IFN-γ also trigger NK cytotoxicity, this could be detrimental for the fetus. KIR2DL4, which is expressed by all human NK cells, is remarkably conserved among primates; this is in striking contrast to all other KIRs, which appear to be rapidly evolving64. KIR2DL4 binds nonclassical MHC class I HLA-G, which is typically expressed on trophoblasts. By virtue of its distinct structure—one ITIM instead of two and an arginine residue in the transmembrane portion—KIR2DL4 can simultaneously inhibit cytotoxicity and activate IFN-γ; it is therefore an attractive candidate for the regulation of uterine NK cells65. The mechanism of activation by KIR2DL4 has not been identified, as none of the NK adaptor molecules known to date (DAP12, DAP10, CD3ζ and FcεRIγ) serve this function. Whether a functional homolog exists in rodent NK cells is unknown.

Multiple and redundant activation cascades

Although the inhibitory KIR and Ly49 receptors are structurally different, they mediate similar functions. This is done via strikingly similar biochemical pathways. The ITIMs present in the intracellular domains of inhibitory KIRs and Ly49s recruit phosphatases that counterbalance the action of PTKs, blocking the activation of cytotoxicity and cytokine production2. The same biochemical pathway is used by lectin-like inhibitory CD94-NKG2 receptors, which are found in both rodents and humans. In addition, ITIM-containing inhibitory receptors are found in many cell types and across several vertebrate species66,67. These observations suggest that these inhibitory pathways are evolutionarily ancient. One would expect that the activatory pathways, which are the targets of the inhibitory ones, are also conserved during evolution, despite the fact that extracellular portions of some activating receptors have diverged. Clusters of genes encoding NK cell receptors and their ligands are found on syntenic chromosomal regions of mice and humans (Fig. 3), suggesting conserved mechanisms of NK cell regulation in the two species, although divergent strategies have also evolved68.

Figure 3: Chromosomal localization of genes for NK receptors and ligands.
figure 3

This figure emphasizes the presence of gene clusters, which are found in syntenic regions of mouse and human chromosomes. Of special note are the NKC (NK gene cluster) and LRC (leukocyte receptor cluster). Genes that are not shared by the two species are shown in blue (human) and red (mouse).

Full size image

Diverse recognition strategies, tempered by conserved signaling pathways, allow NK cells to respond to a wide range of stimuli. This idea is a logical extension of the “multiple choice” model of NK cell recognition69,70. Thus, NK cells may not rely on single recognition strategies to cope with modifications in self-MHC expression and can be activated by multiple and distinct, yet potentially redundant, signaling pathways (Fig. 2). The distinction between stimulation and costimulation of NK cells appear less clear than that for, say, T cells. Diverse receptors generate positive signals, which are counterbalanced by inhibitory receptors. Once the sum total of activation crosses a given threshold, NK cell activation proceeds. Consistent with this general model of multiple and redundant activation pathways are the rarity of functional NK cell deficiencies in humans and in gene-targeted mice bearing null mutations in signaling molecules, which cause major defects in B and T lymphocytes10,11,29,32,33.

Future directions

Given the differences in the structures of activating and inhibitory receptors expressed by human and mouse NK cells, one may ask whether these cells serve divergent functions in the two species. This is a difficult question to answer because the essential physiological roles played by NK cells in vivo remains to be defined. Human cases of selective NK deficiency are rare. Only one study has suggested that NK cells are essential for protection against herpesvirus71. In the mouse, selective NK cell deficiency has been associated with a failure to reject tumor cells72. Considering the conserved effector functions of NK cells (cytotoxicity and cytokine production), it is plausible that the biological roles of these cells in mice and humans are similar. A better understanding of their signaling pathways should shed light on their biological functions and could eventually help us to manipulate NK cells in therapeutic strategies for the treatment of cancer and infectious diseases.

More ligand-receptor interactions are being identified in human and mouse NK cells. New techniques—such as molecular imaging in living cells, gene microarray analysis, RNA interference, conditional gene ablation, proteomics and computer modeling—are being used in the field of signal transduction. Perhaps a good way to get prepared to integrate new and exciting information is to take advantage of the comparisons between mouse and human NK cells in order to learn from their differences and similarities and to have a global view on the biology of mammalian NK cells.