Natural killer (NK) cells have several important roles in innate immune responses. Their best-defined effector functions are their ability to kill haematopoietic tumour cells and virally infected cells and to produce pro-inflammatory cytokines, such as interferon-γ (IFN-γ) and tumour-necrosis factor, that activate other immune cells. As cells of the innate immune system, NK cells have an intrinsic and rapid ability to respond to damaged cells. Unlike the adaptive immune cells B cells and T cells, NK cells do not need to be primed, and the molecular mechanisms for this heightened state of alert are now being elucidated1,2.

The function of NK cells in the elimination of transformed, infected and otherwise stressed cells is an important component of the early immune response. This ability to combat foreign invaders occurs together with tolerance to normal cells (that is, tolerance to self), a property of NK cells that is essential to prevent autoimmunity or destruction of bystander cells. The mechanisms of self-tolerance of NK cells are clearly effective, because there is no known autoimmune state that results from NK-cell dysfunction. Both B and T cells undergo rigorous scrutiny during development, which prevents the maturation of most autoreactive lymphocytes. Any leakage in these CENTRAL-TOLERANCE mechanisms is addressed by PERIPHERAL-TOLERANCE mechanisms. These peripheral mechanisms include monitoring of the way in which antigen is recognized: for example, autoreactive T cells can be tolerized by regulatory T cells or by tolerogenic dendritic cells expressing low levels of co-stimulatory ligands3. NK cells are also educated during development to create a population of mature NK cells that are self-tolerant but sensitive to foreign threats4. In the periphery, NK cells maintain non-responsiveness to self by being equipped with a large number of inhibitory receptors that recognize normal (self) cells.

Until recently, it was thought that NK-cell inhibitory receptors mainly recognize MHC class I molecules. However, there have now been several reports of NK-cell inhibitory receptors that recognize ligands that are not related to MHC class I molecules (Table 1). These ligands are broadly expressed by normal cells, indicating that they have a role in self-tolerance. As we discuss, these findings indicate a novel system that might explain how NK-cell self-tolerance is maintained in MHC-class-I-deficient individuals. Furthermore, novel inhibitory-receptor–ligand pairs provide new targets for the therapeutic manipulation of NK-cell responses. For example, interrupting self-tolerance could improve NK-cell activity to haematopoietic malignancies. This overview discusses newly characterized receptor–ligand pairs, focusing on NK-cell inhibitory receptors that recognize widely distributed non-MHC-molecule ligands.

Table 1 NK-cell inhibitory receptors that have non-MHC-molecule ligands

The 'missing-self' hypothesis and MHC receptors

Early studies indicated that NK-cell responses were induced by the absence of expression of MHC class I molecules at the surface of target cells, as described by the missing-self hypothesis5 (Fig. 1). This hypothesis states that cells that express host MHC class I alleles are protected from NK cells; however, if a target cell fails to express MHC class I molecules, then NK cells kill this cell. The missing-self hypothesis is supported by findings that tumour cells or virally infected cells that have decreased expression of MHC class I molecules become susceptible to NK-cell killing5,6. A corollary of this idea is that healthy allogeneic cells are targets for NK cells because they also lack self-MHC alleles7. Because MHC class I molecules are expressed by almost all nucleated cells, they are an ideal universal marker of self. In some situations, activation signals can overcome MHC-dependent inhibition8. This occurs when cell-surface expression of activating ligands, such as ligands for NKG2D (NK group 2, member D), is induced on target cells by stress, infection or transformation (reviewed in Ref. 8).

Figure 1: NK-cell self-tolerance and the 'missing-self' hypothesis.
figure 1

a | On interacting with a normal, autologous target cell, a natural killer (NK) cell might receive activating signals; however, because the target cell expresses the appropriate self-MHC class I alleles, the NK cell does not lyse the target cell. This is a consequence of inhibitory signals from the ligated MHC-binding receptor at the surface of the NK cell. b | If the target cell loses expression of MHC class I molecules, owing to viral infection or transformation, then the MHC-binding inhibitory receptor at the surface of the NK cell is not engaged. In this way, the NK cell does not receive inhibitory signals and therefore lyses the target cell. In this case, the target cell is perceived by the NK cell to be missing self. c | In the setting of an allogeneic transplant, host NK cells interact with donor target cells that express foreign MHC class I alleles. In most cases, the foreign, non-self-MHC class I molecules do not engage all of the inhibitory receptors at the surface of a host NK cell. This leads to lysis of the allogeneic cells by host NK cells.

To monitor MHC class I molecules, NK cells express numerous receptors that recognize polymorphic epitopes of MHC molecules. The inhibitory receptors block NK-cell activation by recruiting protein tyrosine phosphatases that bind IMMUNORECEPTOR TYROSINE-BASED INHIBITORY MOTIFS (ITIMs) in the cytoplasmic domain of the receptors8. The human killer-cell immunoglobulin-like receptor (KIR) family and the murine Ly49 C-TYPE-LECTIN-like receptor family evolved to carry out similar functions8. The CD94–NKG2A heterodimer of lectin-like molecules is another inhibitory receptor, and it recognizes human HLA-E and mouse Qa1b, which present MHC class I LEADER PEPTIDES9,10,11. Human NK cells also express the inhibitory receptor leukocyte immunoglobulin-like receptor 1 (LIR1), which engages almost all MHC class I molecules12. All of these receptors are expressed by subsets of NK cells, but it is thought that, to remain tolerant to self, each NK cell must express at least one inhibitory receptor that recognizes a self-MHC class I allele4.

Evidence for MHC-independent inhibition

The missing-self hypothesis elegantly explains much of the biology of NK cells, and it made important predictions, such as the existence of MHC-binding receptors at the surface of NK cells. But there are some aspects of NK-cell biology that cannot be adequately explained solely on the basis of this premise (Box 1). Most astonishing is the finding that humans and mice with abnormally low levels of MHC class I molecules, owing to a mutation in the TRANSPORTER ASSOCIATED WITH ANTIGEN PROCESSING 2 (TAP2) protein, develop normal numbers of NK cells. These NK cells can kill tumour cells, although suboptimally, and remarkably, they are tolerant of autologous MHC class Ilow cells13,14,15,16. These findings were initially thought to be explained by the low levels of residual MHC class I molecules expressed by the hosts, thereby allowing tolerance to be maintained. To address this, investigators analysed mice that were deficient in both TAP1 and β2-MICROGLOBULIN2m), which express undetectable amounts of MHC class I molecules at the cell surface. NK cells from these mice were found to function similarly to those from mice that are deficient in either TAP or β2m, indicating that there are non-MHC-dependent mechanisms of NK-cell self-tolerance17,18.

To account for these observations, two new explanations for NK-cell self-tolerance have been proposed: first, NK cells from MHC-class-I-deficient hosts have a lower activation potential, owing to decreased activating-receptor expression and/or function; or second, NK cells are kept self-tolerant by interactions between non-MHC-dependent receptor–ligand pairs. Considering the first possibility, there is both evidence that supports it and evidence that repudiates it19,20,21. In some TAP2-deficient patients, the expression of activating receptors is lower than in healthy control individuals, and of those NK-cell clones that express activating receptors, the receptors function normally19. In other patients and in β2m-deficient mice, activating receptors are expressed at normal levels, and these receptors are also functional20,21. So, it is not clear to what extent activating-receptor modulation accounts for NK-cell self-tolerance in MHC-class-I-deficient hosts. As will become apparent here, there is an increasing body of evidence that supports the second possibility, that non-MHC-mediated inhibition regulates NK cells.

Another independent piece of evidence that supports the idea of non-MHC-dependent self-tolerance is the finding that not all NK cells express an inhibitory receptor for a self-MHC class I allele22. In the C57BL/6 strain of inbred mouse, which is of the haplotype H–2b, the known H–2b-binding NK-cell inhibitory receptors are Ly49C, Ly49I and CD94–NKG2A9,23,24. So, it would be expected that every NK cell in a C57BL/6 mouse would express at least one of these receptors, yet 10% of NK cells in these mice do not express any of these receptors22. It has not been ruled out that other H–2b-binding inhibitory receptors exist, but an alternative explanation for how these NK cells remain self-tolerant is that they express non-MHC-binding inhibitory receptors.

An additional observation that points to the existence of non-MHC-binding inhibitory receptors is that, during NK-cell development, some effector functions are acquired before the appearance of MHC-binding receptors1,25,26,27. This implies that there are unregulated and armed NK cells in the bone marrow, which could potentially be autoaggressive. Again, this paradox could be resolved if immature NK cells express an inhibitory receptor that recognizes a non-MHC-molecule self-ligand, and there is evidence for this28,29.

Non-MHC-binding NK-cell inhibitory receptors

Inhibition of NK cells by 2B4. Of the non-MHC-binding receptors, 2B4 (also known as CD244) is one of the best characterized (Fig. 2). 2B4 is expressed by all human and mouse NK cells, as well as by memory αβ T cells, γδ T cells, monocytes, granulocytes and mast cells30,31,32. It is a member of the SIGNALLING LYMPHOCYTIC ACTIVATION MOLECULE (SLAM; also known as CD150) subfamily of the CD2 family of immunoglobulin receptors. Members of the SLAM subfamily have two or more cytoplasmic IMMUNORECEPTOR TYROSINE-BASED SWITCH MOTIFS (ITSMs). Signalling through molecules that contain ITSMs involves many of the same SRC HOMOLOGY 2 (SH2)-DOMAIN-containing molecules as signalling through molecules that contain ITIMs or IMMUNORECEPTOR TYROSINE-BASED ACTIVATION MOTIFS (ITAMs), including the protein tyrosine phosphatases SHP1 (SH2-domain-containing protein tyrosine phosphatase 1) and SHP2 and kinases such as LCK and FYN33,34 (Fig. 3). ITSMs are implicated in the pathogenesis of X-LINKED LYMPHOPROLIFERATIVE SYNDROME (XLP), which is caused by a mutation in the ITSM-binding protein SLAM-associated protein (SAP)34. SAP binds ITSMs in human 2B4 and mediates activating signals by recruiting FYN; in the absence of functional SAP, the activating signals are abrogated, and human 2B4 either fails to signal or recruits phosphatases that lead to inhibitory outcomes33.

Figure 2: NK-cell inhibitory receptors that have non-MHC-molecule ligands.
figure 2

As well as MHC class I molecules, there are several other types of molecule that inhibit natural killer (NK) cells. As depicted, these ligands are structurally diverse, and they are widely expressed by both haematopoietic and non-haematopoietic cells. Unlike killer-cell immunoglobulin-like receptors (KIRs) and Ly49-family members, which are expressed by subpopulations of NK cells, the non-MHC-binding inhibitory receptors are expressed by most (and in some cases, all) NK cells. The human form of 2B4 is shown; in mice, 2B4 also has a short isoform with two immunoreceptor tyrosine-based switch motifs (ITSMs). Both forms interact with CD48. Both NK-cell receptor protein 1 (NKR-P1)-family members and their ligands C-type-lectin-related (CLR)-family members form homodimers and have lectin-like extracellular domains. Sialic-acid-binding immunoglobulin-like lectin 7 (SIGLEC7) binds sialic-acid linkages that are present on the exposed portion of its ligand sialic acid. Glycoprotein 49B1 (gp49B1) is found in mice and is structurally similar to KIRs. It interacts with αVβ3-integrins. Carcinoembryonic-antigen-related cell-adhesion molecule 1 (CEACAM1) has many isoforms, which are generated by alternative splicing. These have varying numbers of immunoglobulin domains, and immunoreceptor tyrosine-based inhibitory motifs (ITIMs) can be either present or absent from their cytoplasmic domain. The most common form expressed by leukocytes is shown. GPI, glycosylphosphatidylinositol.

Figure 3: Proximal signalling pathways of 2B4: activating and inhibitory.
figure 3

Immunoreceptor tyrosine-based switch motifs (ITSMs) are more versatile than traditional immunoreceptor tyrosine-based inhibitory motifs (ITIMs) and immunoreceptor tyrosine-based activation motifs (ITAMs), because ITSMs recruit kinases in some situations and phosphatases in others. a | Ligation of 2B4 leads to activation of mature human natural killer (NK) cells. When engaged by CD48, ITSMs in the cytoplasmic tail of 2B4 bind the SRC homology 2 (SH2) domain of SAP (signalling lymphocytic activation molecule (SLAM)-associated protein). SAP then recruits the kinase FYN, through the SH3 domain of FYN. Linker for activation of T cells (LAT) present in lipid-raft domains also associates with 2B4, and in conjunction with FYN, this interaction leads to the activation of NK cells. b | In mouse NK cells, as well as in NK cells from patients with X-linked lymphoproliferative syndrome (XLP) and immature human NK cells, 2B4 transmits inhibitory signals. In humans, this inhibition occurs in the absence of functional SAP, as occurs in patients with XLP and in immature NK cells. When 2B4 and an activating receptor are co-engaged by a potential target cell, the ITSMs of 2B4 are phosphorylated by the SRC-family kinases LCK and/or FYN. This leads to the recruitment of phosphatases, such as SHP1 (SH2-domain-containing protein tyrosine phosphatase 1) and SHP2, which in turn inhibit activation by dephosphorylating molecules downstream of the activating receptor. In mice, 2B4 transmits inhibitory signals regardless of the presence or absence of SAP. ITAM, immunoreceptor tyrosine-based activation motif.

The ligand for 2B4 is CD48, a GLYCOSYLPHOSPHATIDYL-INOSITOL (GPI)-LINKED member of the CD2 family35. CD48 is expressed by all nucleated haematopoietic cells and by human endothelial cells; this broad expression provides numerous opportunities for interaction between 2B4-expressing immune cells and CD48-expressing immune cells30. In fact, many of these interactions are beginning to be characterized in normal and diseased states. In mice, 2B4 has been considered to be an activating receptor, but recent evidence indicates that it can also inhibit NK cells29,36,37,38,39. The initial characterization in mice used 2B4-specific antibodies37; when added to cytotoxicity assays, these antibodies increase target-cell lysis, indicating that 2B4 is an activating receptor37. More recent studies indicate that such 'activation' is a consequence of interrupting negative signals from CD48 expressed by target cells29,38,39. Consistent with this, antibody-mediated enhancement of lysis is seen only when CD48-expressing target cells are used29,37. The most unambiguous data regarding the function of 2B4 are provided by recent studies of 2B4-deficient mice29,38,39. NK cells from 2B4-deficient mice show higher levels of killing of CD48+ tumour cells than wild-type NK cells29,38,39. Moreover, in vivo, 2B4-deficient mice eliminate CD48+ tumour cells more efficiently than wild-type mice29,39. Together, these studies indicate that mouse 2B4 mainly functions as an inhibitory receptor. In addition to inhibiting cytotoxicity, 2B4 also inhibits NK-cell production of IFN-γ when it is engaged by targets that express CD48 (Ref. 29). Furthermore, unlike the human form of 2B4, inhibition by mouse 2B4 occurs regardless of the presence or absence of SAP29,38. Despite the dominant role of 2B4 being inhibition of NK-cell function, there are specific situations in which 2B4 can activate mouse NK cells. These include the crosslinking of 2B4 at the surface of NK cells, using plate-bound antibodies37, homotypic 2B4–CD48 interactions among NK cells and among CD8+ T cells, and 2B4–CD48 interactions among NK cells and CD8+ T cells40,41,42.

In contrast to mouse 2B4, human 2B4 seems to be mainly an activating receptor36. This conclusion is based on studies in which 2B4 was crosslinked by antibodies, as well as studies that involved CD48+ and CD48 target cells36,43,44. In general, CD48-expressing target cells (unlike those in mice) are more susceptible to NK-cell killing than cells that do not express CD48 (Refs 43,44). 2B4 can also be inhibitory in humans but only in the absence of functional SAP, as occurs in patients with XLP and during the development of human NK cells28,45.

In mice, the inhibitory effect of 2B4 extends beyond tumour-cell targets. 2B4-deficient mouse NK cells show increased killing of CD48+ allogeneic and syngeneic splenocytes, indicating that, even in the presence of the powerful inhibitory effect of self-MHC class I molecules, 2B4 has an important role in the inhibition of NK cells29. These data show that the mouse 2B4–CD48 pair provides a system for NK-cell self-tolerance that is independent of MHC class I molecules. So, could these findings resolve the paradoxes (Box 1) of MHC-class-I-dependent inhibition? For the question of how NK-cell self-tolerance is maintained during development, when MHC-binding receptors are absent, the answer is yes; human and mouse 2B4 function to protect autologous cells from autoaggressive, developing NK cells28,29. Concerning the other mysteries of MHC molecules and NK-cell self-tolerance, emerging data indicate that 2B4–CD48 interactions maintain self-tolerance in the absence of regulation by MHC class I molecules. NK cells from β2m-deficient mice are prevented from killing syngeneic cells by CD48 (V.K. and M.E.M., unpublished observations). Furthermore, mature NK cells that lack self-MHC-binding inhibitory receptors are kept self-tolerant by 2B4 (V.K. and M.E.M., unpublished observations). These findings strongly support the idea that, both in the presence and in the absence of MHC-class-I-mediated regulation, 2B4–CD48 interactions restrict NK-cell autoreactivity.

Why are there differences in the function of 2B4 expressed by mouse and human NK cells? One explanation is that the studies of human and mouse NK cells were carried out differently. The studies of mouse cells examined 2B4-deficient NK cells in vivo, whereas the studies of human cells used antibody crosslinking of 2B4 in vitro. A second potential explanation is that human NK cells have evolved a different requirement for SAP, which provides an activating signal in human NK cells.

Inhibitory receptors of the NK-cell-receptor protein 1 family. In mice, the NK-cell-receptor protein 1 (NKR-P1; also known as killer-cell lectin-like receptor B1, KLRB1) family of receptors has five members: NKR-P1A, NKR-P1B, NKR-P1C, NKR-P1D and NKR-P1F. Similar to the Ly49 family, NKR-P1-family members are homodimeric C-type-lectin-like molecules. NKR-P1C was the first to be identified and is also known as NK1.1, which is a commonly used marker for identifying the NK cells of C57BL/6 mice46,47. NKR-P1C has a charged transmembrane residue that associates with the γ-chain of the high-affinity receptor for IgE (FcεRI), and ligation of NKR-P1C activates mouse NK cells48. NKR-P1A and NKR-P1F have similar sequences and are thereby thought to be activating receptors. By contrast, NKR-P1B and NKR-P1D contain ITIMs, and ligation of these receptors inhibits NK cells in functional assays49,50,51,52. In humans, only one NKR-P1 molecule, NKR-P1A, has been identified, and its function is unclear, because it lacks an ITIM or a charged transmembrane residue53.

Recently, ligands for NKR-P1 molecules have been identified: NKR-P1B and NKR-P1D recognize C-type-lectin-related B (CLR-B; also known as OCIL), and NKR-P1F binds CLR-G (also known as OCILrP2)51,52. The CLR family has three members: CLR-B, CLR-F and CLR-G. The genes encoding these are located, together with the genes encoding NKR-P1-family members, on human chromosome 12 and mouse chromosome 6, in the NK complex of lectin-like molecules54. Similar to 2B4 and CD48, the NKR-P1 family and the CLR family are tightly genetically linked, indicating that co-inheritance of the receptor and self-ligand might be biologically important30,54.

Similar to MHC class I molecules, CLR-B protein is broadly expressed, by all nucleated haematopoietic cells and by some non-haematopoietic cells52. This finding is consistent with the observation that Clr-b mRNA is widely expressed55,56. NKR-P1D protein is expressed by most NK cells, and it inhibits NK-cell lysis of CLR-B-expressing target cells51. Interaction between NKR-P1D and CLR-B also inhibits the lytic activity of NK cells against syngeneic MHC class I+ and MHC class Ilow tumour cells51, showing that, similar to the 2B4–CD48 system, NKR-P1D–CLR-B-mediated inhibition occurs in parallel with MHC-class-I-mediated inhibition. Intriguingly, NK cells from MHC class Ilow, β2m-deficient mice are inhibited by NKR-P1D, indicating that NKR-P1D can regulate NK cells in the absence of normal levels of MHC class I molecules. Despite this in vitro data, the in vivo significance of NKR-P1B- and NKR-P1D-mediated inhibition has yet to be determined.

So far, a human homologue of these inhibitory NKR-P1 molecules has not been identified. A human CLR-like molecule, lectin-like transcript 1 (LLT1), is known and is closely related to mouse CLR-G57. LLT1 protein is broadly expressed by peripheral-blood cells, including NK cells58. Surprisingly, ligation of LLT1 at the surface of NK cells activates IFN-γ production, indicating that CLR and CLR-like molecules might signal bidirectionally, functioning both as ligands and as signalling molecules themselves58. Identifying the cognate interacting proteins for LLT1 and other orphan lectin-like receptors encoded in the human NK complex54 will shed light on human NK-cell regulation by the NKR-P1 family of receptors.

Carcinoembryonic-antigen-related cell-adhesion molecule 1 mediates immune self-tolerance. The carcinoembryonic antigen (CEA) family is large and multifunctional59. Its founding member, CEA-related cell-adhesion molecule 5 (CEACAM5; also known as CEA), is most renowned for its use as a marker of colon cancer59. CEAs are members of the immunoglobulin superfamily and can be divided into two subgroups: the CEACAM subgroup and the pregnancy-specific glycoprotein subgroup. The CEACAM subgroup has seven members. CEACAM1 (also known as CD66a or BGP) is the only CEACAM that is expressed by NK cells60. The ligand for CEACAM1 is CEACAM1 itself, binding in a homophilic interaction61. CEACAM1 also takes part in heterophilic interactions with the GPI-linked molecule CEACAM5 (Ref. 61). CEACAM1 has several isoforms, which are generated by alternative splicing; these include two cytoplasmic tail derivatives: a long form that contains two ITIMs, and a short form that lacks ITIMs. Humans and mice preferentially express the ITIM-containing form62. In addition to NK cells, CEACAM1 is expressed by granulocytes, dendritic cells, lymphocytes, endothelial cells and epithelial cells59,63.

Recent investigations have established the importance of CEACAM1-mediated negative regulation of NK cells19,64,65. One study examined the role of CEACAM1 in the prevention of NK-cell autoreactivity in TAP2-deficient patients19. Remarkably, whereas 12% of NK-cell clones from healthy donors express CEACAM1, 80% of NK-cell clones from TAP2-deficient patients express CEACAM1; the TAP2-deficient NK cells also express CEACAM1 at higher levels than normal NK cells. So, CEACAM1 expression seems to be upregulated to compensate for abnormal MHC-class-I-mediated NK-cell inhibition. Importantly, blocking the engagement of CEACAM1 at the surface of TAP2-deficient NK cells allows the killing of autologous cells (which express CEACAM1), showing that, in the absence of self-MHC class I molecules, CEACAM1 can prevent NK-cell autoaggression19.

Notably, even in the presence of self-MHC class I molecules, CEACAM1 prevents NK-cell autoreactivity19. Specifically, CEACAM1 inhibits TAP2-deficient NK cells from killing TAP2+ cells. Even NK-cell clones from healthy donors killed autologous cells when CEACAM1 was blocked. Both TAP2+ and TAP2 NK cells showed maximal killing of MHC class I+ target cells only when MHC class I molecules and CEACAM1 were concomitantly disrupted19. Therefore, CEACAM1 and MHC class I molecules at the surface of target cells provide layers of inhibition that are not redundant. Inexplicably, CEACAM1 NK-cell clones from TAP2-deficient patients did not kill autologous cells; these NK cells might be regulated by decreased activating-receptor expression or by expression of another receptor that recognizes self19.

Soluble CEACAM1 is found in normal serum and is present at increased levels in various pathological states66. In comparison with healthy control individuals, TAP2-deficient patients have much less serum CEACAM1 (Ref. 65). Soluble CEACAM1 blocks signalling through cell-surface CEACAM1, as shown by the increase in NK-cell activity in vitro caused by the addition of CEACAM1–immunoglobulin fusion proteins or serum from healthy control individuals. So, although the biological function of soluble CEACAM1 remains unknown, a decreased level of soluble CEACAM1 might ensure that the inhibitory effect of CEACAM1 in the absence of MHC class I molecules is maximized65.

Role of sialic-acid-binding immunoglobulin-like lectins and sugar residues in self-recognition. Sialic-acid-binding immunoglobulin-like lectins (SIGLECs) have a V-SET IMMUNOGLOBULIN DOMAIN, which binds sialic acid, and varying numbers of C2-SET IMMUNOGLOBULIN DOMAINS67. Humans have 11 SIGLECs, and human NK cells express SIGLEC7 (also known as p75 or AIRM1) and SIGLEC9 (Ref. 67). Mice have eight SIGLECs, and mouse NK cells express SIGLEC-E68. SIGLEC9 and SIGLEC-E each have an ITSM (similar to 2B4) and an ITIM, whereas SIGLEC7 has two ITIMs67. Sialic acids are a large, diverse family of nine-carbon α-keto-acid monosaccacharides68. They are ubiquitously expressed and are often bound to the exposed portions of other carbohydrates that are components of cell-surface lipids and proteins, through α2,3- and α2,6-sugar linkages, or bound to each other through α2,8-sugar linkages68.

SIGLEC7 was originally identified as an inhibitory receptor expressed by all NK cells and monocytes and some CD8+ T cells69,70. The cytoplasmic region of SIGLEC7 binds SHP1, and SIGLEC7 inhibits NK-cell killing when its extracellular domain is crosslinked by antibody69. One ligand that has been identified for SIGLEC7 is the ganglioside GD3, a glycosphingolipid that contains an α2,8-linked disialic acid71,72, and NK cells do not lyse GD3-synthase-transfected P815 tumour cells, as a result of inhibition by SIGLEC7 (Ref. 73). Interestingly, GD3 is expressed by central-nervous-system cells, by melanoma cells and by some T cells74,75. Of the other SIGLECs expressed by NK cells, SIGLEC9 inhibits both T-cell lines and basophil cell lines76,77, and it is expressed by 50% of NK cells78. SIGLEC-E, which is expressed by mouse NK cells, is similar to SIGLEC7 and SIGLEC9; 50% of mouse NK cells express SIGLEC-E79, and it is also inhibitory80. There is little known about the function of these receptors; however, as discussed later, SIGLECs might have a role in discrimination between self and non-self in pathogenic states.

Other non-MHC-binding NK-cell inhibitory receptors. Five signal-regulatory proteins (SIRPs) have been identified in humans, and one, SIRP-α81, has been identified in mice. SIRPs are immunoglobulin-superfamily members; they are polymorphic and are also generated by variations in splicing. SIRPs are thought to be the primordial adaptive immune receptor, because they have joined V–J immunoglobulin domains81. SIRP-α and SIRP-β2 recognize the ubiquitously expressed molecule CD47 (also known as integrin-associated protein, IAP)81,82,83. SIRP-α inhibits phagocytes from eliminating normal, CD47+ self-cells; CD47-deficient haematopoietic cells are promptly cleared by phagocytes when transplanted into normal, syngeneic hosts84,85. This innate system of self-recognition resembles the regulation of NK cells by MHC class I molecules, so it is not surprising that NK cells express SIRP-β2 (Refs 82,83). SIRP-β2 is expressed by all activated NK cells and by most T cells. What is surprising is that SIRP-β2-specific antibody does not influence NK-cell lysis82, but this is probably a consequence of the lack of ITIMs or charged transmembrane residues in the cytoplasmic region of the SIRP-β2 protein. The presence of SIRP-β2 at the surface of T cells promotes their adhesion to CD47+ cells, and it is possible that SIRP-β2 expressed by NK cells might function in a similar manner82.

Glycoprotein 49 B1 (gp49B1) is an inhibitory receptor in mice, and it has two C2-set immunoglobulin domains and two ITIMs86. No human homologue has been identified as yet, but it is structurally similar to the killer-cell immunoglobulin-like receptors (KIRs)87. gp49B1 is expressed by most activated NK cells, as well as by mast cells, macrophages and activated T cells86,87,88. gp49B1 binds αVβ3-integrin (a heterodimer of CD51 and CD61), which is expressed at low levels in most tissues and at high levels by osteoclasts, macrophages, some types of granulocyte, platelets, endothelial cells, endometrial tissue, inflammatory sites and invasive tumour cells89,90. The NK cells of gp49B1-deficient mice are mostly functional91, but after in vivo infection with virus, gp49B1-deficient NK cells produce more IFN-γ ex vivo92. In light of the identification of the ligand for gp49B1, it will be important to readdress the role of gp49B1 in NK-cell killing of αVβ3-integrin+ target cells.

KLRG1 (also known as MAFA) is expressed by NK cells and other leukocytes93. It is thought to be inhibitory, as it contains an ITIM and as crosslinking with specific antibody decreases cytokine production and lytic activity by an NK-cell line94. Its ligand is unknown, but interestingly, in contrast to Ly49 molecules, KLRG1 expression is decreased in MHC-class-I-deficient mice95.

Leukocyte-associated immunoglobulin-like receptor 1 (LAIR1) is also expressed by most human leukocytes, including NK cells96. LAIR1 contains two ITIMs, binds SHP1 and SHP2, and inhibits NK cells when crosslinked with specific antibody96. Its ligand is unknown, but it is thought to have a non-MHC molecule as a ligand96.

Pathogens and non-MHC inhibitory ligands

It is clear that pathogens exploit MHC class I molecules to their advantage6. Some microorganisms have evolved mechanisms to inhibit the presentation of peptides by MHC class I molecules so that they remain concealed from the adaptive immune system. Furthermore, some pathogens express MHC-class-I-like molecules to ensure their protection from host NK-cell responses. Given the influence of non-MHC-binding inhibitory receptors, non-MHC-molecule markers of self are apt to be exploited by pathogens for immune protection (Table 2).

Table 2 Numerous pathogens encode potential ligands for non-MHC-binding NK-cell inhibitory receptors

For example, the ligand for 2B4, CD48, was originally identified as a protein that is induced by infection of B cells with Epstein–Barr virus (EBV)97. Perhaps, increased expression of CD48 originally inhibited NK-cell antiviral activity. EBV infections occur worldwide, and according to the Centers for Disease Control and Prevention (United States), EBV infects at least 95% of people in the United States. It is conceivable that, under such pressure from EBV, SAP or SAP-like molecules evolved to convert 2B4 from an inhibitory receptor that prevents responses to self in mice to an activating receptor that recognizes pathogens in humans. Supporting this hypothesis, patients with abnormal signalling through 2B4 or with insufficient NK-cell activity are susceptible to fulminant infections with EBV and other herpesviruses34,98. Moreover, this occurrence might have a precedent in the evolution of mouse resistance to mouse cytomegalovirus (MCMV)99. Ly49I is an inhibitory receptor for MHC class I molecules, and the MCMV-encoded protein m157 engages Ly49I, thereby inhibiting NK-cell antiviral function. Some strains of mice resist infection with MCMV by expressing Ly49H, which transmits activating signals when bound to m157 (Ref. 99). Many NK-cell receptors have similar activating and inhibitory counterparts, which might have evolved from a pathogen–host 'arms race'.

C-type-lectin-like molecules are encoded by several viruses, and it would be interesting to determine whether these engage inhibitory NKR-P1-family members100,101,102,103,104,105. In addition, CD47 homologues are expressed by the poxviruses variola virus, vaccinia virus and myxoma virus103,106,107. Could these virally encoded proteins mimic ligands for SIRPs, the inhibitory receptors that are expressed by NK cells, T cells and other leukocytes? The role of these proteins in the pathogenesis of viral diseases is unknown, but the range of microorganisms that have independently acquired genes encoding C-type-lectin-like molecules or CD47-like proteins indicates that this area is worth pursuing. One example of this possibility has been documented: the hepatitis C virus envelope protein E2 binds CD81 expressed by NK cells and, in this way, inhibits NK-cell antiviral activity108,109.

Numerous microorganisms produce or acquire sialic acid, including Neisseria meningitidis, Haemophilus influenzae, Escherichia coli and Trypanosoma cruzi68. It will be important to determine whether sialic acid also protects bacteria by inhibiting SIGLEC-expressing NK cells. SIGLECs can also be masked by the binding of sialic acid in cis73, and they can be unmasked by cellular activation110,111. Numerous pathogens express sialidases68, which could unmask SIGLECs at a local area of infection, thereby providing another method of increasing inhibitory signalling.

In one example, a pathogen receptor for host-cell entry also contributes to immunosuppression112. CEACAM1 is used as a receptor for host-cell entry by Neisseria spp., Salmonella typhimurium, H. influenzae, Moraxella catarrhalis and mouse hepatitis virus113,114,115,116,117. The binding of Neisseria gonorrhoeae to CEACAM1 at the surface of CD4+ T cells also downregulates T-cell activation. Whether Neisseria spp. or other CEACAM1-binding microorganisms similarly inhibit NK cells remains to be investigated112.

Non-MHC molecules and tumour immunity

Tumour cells often have decreased expression of MHC class I molecules118, but it has been found that MHC allotypes that inhibit NK cells are preferentially maintained by transformed cells119. One might predict that, in the same way, tumour variants that induce or upregulate expression of non-MHC-molecule inhibitory ligands would be selected for. This is true for SIGLEC7 ligands, which are highly expressed by renal-cell carcinomas71 and melanomas74, and in the case of the former, higher expression of SIGLEC7 ligands correlates with increased metastatic potential120. Increased metastasis could be a consequence of a non-immune mechanism, such as increased cell adhesion to a SIGLEC-expressing metastatic site, and/or a consequence of immunosuppression of NK cells, although the latter has not been tested.

CEACAM1, which can interact with CEACAM1 at the surface of NK cells, is expressed at higher levels in several tumour types than in their normal tissue counterparts, including in cancer of the breast, prostate, colon and endometrium121. Melanoma cells also express CEACAM1, and increased CEACAM1 expression by melanoma and NK cells correlates with poor prognosis64 and increased metastasis122. CEACAM1 expression is also a prognostic factor in adenocarcinoma of the lung123. It is expressed at only low levels by renal-cell carcinomas, but it is upregulated by these cells after interaction with lymphocytes or exposure to IFN-γ124. Although CEACAM1 has well-characterized roles in carcinogenesis in vivo, including tumour-suppressor activity and angiogenesis- and metastasis-promoting activity121, CEACAM1 does block NK-cell killing of tumour cells in vitro19,64. Its dampening effect on the immune system in vivo deserves further attention.

In humans, leukaemic cells have decreased expression of CD48 compared with non-transformed lymphocytes125. This might reflect immune selection as a result of the activating nature of 2B4 expressed by human NK cells. Moreover, patients with XLP are prone to lymphomas; this could stem, in part, from dysregulated 2B4 signalling and NK-cell tumour surveillance.

CLR-B expression by tumour cells inhibits NK-cell cytotoxicity51,52, yet counter-intuitively, it is downregulated by mouse tumour cell lines52. The researchers who made this finding ventured that downregulation might stem from tumour overexpression of casein kinase 2 (CK2), which promotes both growth of cells and internalization of receptors that have sites for phosphorylation by CK2 (as does CLR-B). They also proposed an alternative explanation, for which there was some evidence: that CLR-B is pro-apoptotic and thereby unfavourable to tumour cells. In both cases, downregulation of CLR-B expression might be a novel signal of cell stress that is perceived by NK cells as missing self52.

Regulation of NK-cell alloreactivity

One hope of NK-cell researchers is that alloreactive NK cells could be used for the treatment of haematopoietic malignancies, although this research area is still developing. Donor-derived alloreactive NK cells can increase graft-versus-leukaemia responses, decrease graft-versus-host disease and decrease host-versus-graft rejection in the setting of an allogeneic bone-marrow transplant for the treatment of leukaemia126. These functions arise from mismatches between donor KIRs and host MHC class I molecules126. Could blocking non-MHC-binding inhibitory receptors further increase the activity of donor NK cells?

As yet, the function of non-MHC-binding receptors in alloreactivity has not been studied in detail. The maternal–fetal interface is one example of an allograft in which the self-tolerance of NK cells is important, and this tolerance is delicately balanced127. Unlike peripheral-blood NK cells, CEACAM1 is found to be expressed by most activated decidual NK cells128. Extravillous trophoblasts, which are derived from fetal tissue and are therefore allogeneic, also express CEACAM1, and in vitro, CEACAM1 ligation is inhibitory for decidual NK cells and other lymphocytes128. In conjunction with the other reports on CEACAM1-mediated regulation of NK cells, this supports the idea that CEACAM1 dampens NK-cell alloreactivity.

2B4-deficient mouse NK cells show increased alloreactivity in vitro29. In vivo, CD48-specific antibody blockade decreases the engraftment of allogeneic bone-marrow transplants, which is consistent with the idea that host NK cells reject transplants in the absence of 2B4 ligation129. Indeed, 2B4-deficient mice show greater rejection of allogeneic bone-marrow transplants (V.K. and M.E.M., unpublished observations). This might be relevant in the case of haematopoietic-stem-cell transplantation, in which donor-derived human NK-cell alloreactivity is desirable, but at present, developing donor NK cells are inhibited by 2B4 during the first month after transplantation130.

Regulation of NK-cell autoreactivity

Several non-MHC-dependent NK-cell receptors have been associated with autoimmunity. Clearly, 2B4 has a role in the self-tolerance of autologous cells in vitro: 2B4 present at the surface of mouse cells inhibits the killing of syngeneic cells by NK cells29, and at the surface of human cells, 2B4 inhibits the killing of syngeneic cells by immature NK cells28. In vivo, soluble CD48 is detectable at high levels in patients with rheumatoid arthritis131. The association between soluble CD48 and human arthritis is worthy of further research, particularly because polymorphisms in 2B4-family members have also been associated with a mouse model of systemic lupus erythematosus132.

A genetic susceptibility locus for psoriasis has recently been mapped in humans133. One gene on this locus encodes inhibitory receptor protein 60 (IRp60), a protein that is expressed by all NK cells and by monocytes, granulocytes and a subset of T cells133,134. IRp60 is an inhibitory receptor that is thought to have a non-MHC molecule as a ligand134. It will be interesting to determine whether polymorphisms in IRp60 decrease its function and therefore result in decreased inhibition in an autoimmune setting. In a different model, increased inhibitory signalling prevents pathological inflammation: in vivo crosslinking of CEACAM1 with specific antibody inhibits the development of T-helper-1-cell-mediated colitis in mice135.

Finally, cells of the central nervous system express low levels of MHC class I molecules136 but express GD3 (Ref. 137), which inhibits NK cells through binding SIGLEC7 (Ref. 73). We speculate that a ligand that is not an MHC molecule, such as GD3, might contribute to the differential sensitivity of neurons to syngeneic NK cells138 and might contribute, in part, to the immune privilege that neural tissues are thought to have.


Analysis of MHC-class-I-deficient hosts has yielded valuable information for understanding NK cells; however, this line of study is also fraught with unexplained phenomena and unanswered questions. The accumulated data that are reviewed here strongly indicate that another, previously underappreciated system regulates NK cells. Certainly, further understanding of non-MHC-dependent inhibition of NK cells will yield valuable insights into NK-cell biology and therapeutic opportunities. Unlike the various types of receptor that recognize MHC class I molecules, which are expressed by subsets of NK cells, many of the non-MHC-binding receptors are expressed by most NK cells and might be more conserved between individuals, potentially providing a unique therapeutic opportunity to manipulate an entire NK-cell population. Many of these receptors are also expressed by T cells and myeloid cells, so further knowledge in this field will impact on our understanding of many aspects of the immune system.