Developmental pathways that generate natural-killer-cell diversity in mice and humans

Key Points

• Natural killer (NK) cells are a subset of lymphocytes that participate in innate immunity and may also influence the outcome of adaptive immune responses. Multiple functional activities have been associated with NK cells, including cytotoxicity to remove infected or transformed cells and the production of cytokines and chemokines that recruit and sustain inflammatory cells. Diverse NK-cell subsets might have dedicated roles in eliciting these effector functions.

• NK-cell development occurs largely in the bone marrow, where committed NK-cell precursors acquire activating and inhibitory receptors, engage self MHC molecules, and mature before emerging as functional NK cells. The soluble factors that drive this process and the immature NK-cell stages that characterize it have been identified in both mice and humans.

• Most mature NK cells in humans or mice uniformly express NK group 2, member D (NKG2D), CD161, NK-cell protein 46 (NKp46) and CD122. However, the differential expression of Ly49 family members, CD127, CD27 and KLRG1 (killer-cell lectin-like receptor subfamily G, member 1) in mice, and of KIRs (killer-cell immunoglobulin-like receptors), CD56 and CD16 in humans suggests heterogeneity within the mature NK-cell pool. In both humans and mice, several well-characterized NK-cell subsets have been identified that demonstrate skewed effector functions (cytokine production and cytolysis) and may have unique biological roles.

• Evidence for NK-cell development outside the bone marrow has been provided in both mice and humans, in the thymus and lymph nodes, respectively. NK cells generated at these sites appear to have distinct phenotypic and functional attributes, suggesting that the tissue in which NK cells develop might have an important role in the diversification of NK-cell subsets. In a similar manner, uterine NK cells may be conditioned by the unique environment afforded by the uterus during pregnancy.

• Ethical and technical difficulties limit our understanding of NK-cell development in humans. The generation of humanized immune systems in immunodeficient mice provide a novel, robust and accessible way to further dissect human NK-cell development.

Abstract

Natural killer (NK) cells are large granular lymphocytes capable of producing inflammatory cytokines and spontaneously killing malignant, infected or 'stressed' cells. These NK-cell functions are controlled by cell-surface receptors that titrate stimulatory and inhibitory signals. However, we remain puzzled about where and when NK cells develop and differentiate, and this has fuelled the debate over the diversification of the peripheral NK-cell pool: are NK cells functionally homogeneous or are there subsets with specialized effector functions? In this Review, we consider the developmental relationships and biological significance of the diverse NK-cell subsets in mice and humans, and discuss how new humanized mouse models may help to characterize them further.

Main

Natural killer (NK) cells are a subset of lymphocytes that principally participate in innate immunity but may also have important roles in determining the outcome of the adaptive immune responses. Diverse functional activities have been associated with NK cells, including 'natural' (immediate) cytotoxicity and the capacity to promptly produce cytokines (tumour-necrosis factor (TNF) and interferon-γ (IFNγ)) and chemokines that help to generate and sustain an inflammatory environment (reviewed in Ref. 1). By virtue of these different effector functions, NK cells are poised to participate in an early defence system that can directly eliminate virally infected or transformed cells through target-cell lysis, and/or recruit nonspecific haematopoietic cells (neutrophils and macrophages), activate dendritic cells (DCs) and prime adaptive lymphocytes (T cells and B cells) through the release of soluble amplifying factors. Understanding how NK cells develop into functional effector cells that are equipped with the potential to accomplish distinct roles in immunity remains a challenge, but it will be important to harness the therapeutic potential of these innate lymphocytes².

The functional pleiotropy exhibited by NK cells provides the immune system with a 'multitasking' cell that could serve different biological roles under different conditions. Maintaining this property of NK cells would appear advantageous to the organism; however, programming NK cells (at the level of gene transcription) so that they are capable of a 'one cell does all' endpoint but that at the same time are still able to generate adequate effector functions under diverse assaults might prove difficult. So, rather than by having a single multifunctional cell, diverse NK-cell effector functions could be achieved in an alternative way through NK-cell diversification. In this way, the functional pleiotropy exhibited by NK cells during immune responses would be governed by distinct NK-cell subsets. Over the years, several reports have suggested that there are NK-cell subsets with specialized functions in humans and mice (including 'helper' NK cells with type 1 and 2 cytokine profiles, regulatory NK cells, cytotoxic NK cells and antigen-presenting NK cells^3,4,5).

This notion of NK-cell diversity raises several fundamental questions, the most obvious being the definition of what constitutes an NK-cell subset in the first place. But once defined, it is important to know how and when NK-cell subsets arise. Do NK cells develop and differentiate in multiple sites? Does the site of NK-cell differentiation impart particular characteristics to the NK cells that develop there, resulting in the generation of NK-cell subsets? If NK-cell subsets exist, are their phenotypes and functions stable? What distinction needs to be made between activation and resting states for NK cells? Finally, is there a biological necessity for NK-cell diversity? In this Review, we examine the evidence for functional NK-cell subsets in mice and humans.

States and sites of NK-cell differentiation

Prior to describing the phenotype and functional attributes of immature NK cells and their differentiated progeny, some generalities of NK-cell development should be considered (Fig. 1). For any given tissue localization, we can distinguish 'stages' in the life of an NK cell. Multiple factors condition this process, including cell-intrinsic signals (transcription factors; for a recent review, see Ref. 6) and environmental cues. The first stage is defined by the NK-cell precursor (NKP) that is generated from multipotent haematopoietic precursors (including haematopoietic stem cells (HSCs) and early lymphoid precursors (ELPs)). By definition, NKPs have the potential to develop into mature NK cells but not into any other haematopoietic lineage. NKPs give rise to immature NK cells, which express some (but not all) NK-cell-specific markers; immature NK cells are considered to be the precursors to more mature NK cells. Whether immature NK cells have functional roles in immunity is not clear, although some studies have shown that NK cells with an immature phenotype have functional attributes (for examples, see Ref. 6).

Figure 1: Multiple sites for NK-cell development.

For an individual tissue site, one can distinguish committed natural killer (NK)-cell precursors (NKPs); in the bone marrow and thymus these differentiate from haematopoietic stem cells (HSCs) or early lymphoid precursors (ELPs). Immature NK cells provide the substrate for the process of NK-cell education, which results in mature NK cells with functional competence. Mature NK cells may also be considered according to their activation state (resting or activated, for example; not shown). NKPs, immature NK cells and mature NK cells may recirculate between different tissue sites (bone marrow, thymus, liver, spleen and lymph nodes).

As immature NK cells differentiate further, NK-cell receptors specific for MHC class I molecules are expressed in a stochastic, variegated and overlapping manner⁷, and are tested in a process that has been referred to as 'NK-cell education'. NK-cell education results in two types of mature NK cell: the type that expresses at least one inhibitory NK-cell receptor that has been engaged by self MHC molecules present in the environment in which the education occured, and the type that expresses self-specific NK-cell receptors that have not been engaged by self MHC molecules. Recent studies have shown that the engagement of inhibitory NK-cell receptors by MHC ligands is critical to allow developing NK cells to achieve their full functional competence^8,9,10,11. Yokoyama and colleagues refer to this education process as 'licensing'¹², whereas Raulet and colleagues favour a 'disarming' model¹³. While the terminology debate rages, the precise molecular mechanism by which education occurs remains unknown. Educated or licensed NK cells have a mature phenotype, are fully responsive to triggering by activating receptors and are tolerant to self. Mature NK cells that develop in the absence of interactions with self MHC molecules are 'hyporesponsive' to activation signals.

All NK cells may leave the site of their generation and take up residence in other tissues (such as the secondary lymphoid organs, lungs, liver and gut). In these peripheral sites, mature NK cells can be further distinguished by their activation state: 'resting' NK cells are mostly a default definition (lacking signs of recent activation), whereas 'activated' NK cells are larger in size, bear markers of recent stimulation (such as CD69) and can demonstrate heightened functional properties. It is also possible that chronically stimulated NK cells that are analogous to 'exhausted' T cells exist¹⁴. The complex process of NK-cell differentiation occurs at several distinct tissue sites, including the bone marrow, liver, thymus, spleen and lymph nodes, and may involve the circulation of NK cells at different stages of maturation between these sites (Fig. 1). The comparison of human and mouse molecules expressed by NKPs, immature NK cells and mature NK cells reveals a remarkable conservation of phenotypes and functions (Fig. 2), although there are some differences in the expression pattern of certain receptors (for example, Ly49 expression in mice versus KIR (killer-cell immunoglobulin-like receptor) expression in humans or lack of CD56 expression in mice).

Figure 2: Phenotypic and functional characteristics of human and mouse NK cells.

The traditional view of natural killer (NK)-cell development is represented by a linear pathway from haematopoetic stem cells (HSCs) in the bone marrow to fully functional NK cells in the periphery. This diagram depicts the expression pattern of several molecular markers in humans and mice throughout the stages of NK-cell maturation. NK-cell precursors (NKPs) lack typical NK-cell markers but give rise to NK cells. Immature NK cells express CD161 and natural-killer group 2, member D (NKG2D) in both mice and humans, as well as receptors required for growth and survival. NK-cell education via self MHC class I molecules probably begins at the immature NK-cell stage. More mature NK cells leave the bone marrow and populate peripheral lymphoid organs (spleen, liver and lymph nodes). Peripheral NK cells can be activated following detection of missing or altered self MHC class I molecules or by an inflammatory environment. Activated NK cells have increased functional capacity compared to resting NK cells. Peripheral NK-cell subsets (NK1 and NK2 cells) have also been proposed in mice and humans. Some phenotypic and functional properties of the NK2-cell subsets differ from the NK1-cell subset; for example, human NK2 cells are CD56^lowCCR7⁻, whereas human NK1 cells are CD56^hiCCR7⁺. Although most evidence suggests that NK2 cells arise from NK1 cells, it is also possible that they both arise from a common precursor (an immature NK cell). ^*Heterogeneous expression. ^‡Heterogeneous expression in NK2 cells. ^§Thymus derived NK cells in mice. ^∥Tumour-necrosis-factor-related apoptosis-inducing ligand (TRAIL)-mediated cytotoxicity is more prominent in neonatal mice. CCR7, CC-chemokine receptor 7; CD62L, CD62 ligand; IL, interleukin; KIR, killer-cell immunoglobulin-like receptor; KLRG1, killer-cell lectin-like receptor subfamily G, member 1; NKp46, NK-cell protein 46.

NK-cell precursors in the bone marrow. The bone marrow has been considered the main site for NK-cell development, as ablation of the bone-marrow microenvironment was associated with dramatic defects in NK-cell homeostasis and function^15,16. Several (and perhaps all) aspects of NK-cell development can occur in the bone marrow, including the generation of NKPs from haematopoietic precursors, the acquisition of expression of the NK-cell surface receptors involved in target-cell recognition, NK-cell education, and the emergence of differentiated NK cells with full functional competence (reviewed in Refs 17–19). By analysing immature and mature NK cells in the bone marrow, models have emerged proposing several distinct stages for developing NK cells (reviewed in Ref. 18).

NKPs in mice were initially identified in the asialo-GM1⁻ bone-marrow fraction, although this cell population was not homogeneous as it could still give rise to both myeloid cells and other lymphocytes²⁰. Although more lymphoid-restricted bone-marrow subsets that have NK-cell potential have been recently characterized²¹, none of these represent true NKPs, as all can differentiate into CD19⁺ B cells and can colonize or reconstitute the thymic population. The only NK-cell-restricted murine bone-marrow subset that has been described to date is a lineage (LIN)⁻CD49b⁻CD161⁻CD122⁺ population that can differentiate into mature NK cells in vitro or in fetal thymic organ cultures (FTOCs)²². Although enriched in NKPs (about 1 in 10 cells gives rise to a NK cell in vitro), this population is still heterogeneous for the expression of natural-killer group 2, member D (NKG2D), suggesting that true NKPs might be NKG2D positive^22,23.

In humans, bone-marrow NKPs have been identified that give rise to CD3⁻CD56⁺ NK cells following in vitro culture with soluble factors (including interleukin-2 (IL-2) or IL-15). Human NKPs form part of the CD34⁺CD38⁺CD45RA⁺ subset^24,25,26,27, and further selection for CD7⁺ cells enriches for NKPs that express transcripts of CD122 (Refs 28, 29). A more differentiated CD7⁺CD34⁻LIN⁻ subset also generates functional NK cells in vitro and may represent a more restricted NKP population³⁰. So, in both mice and humans, there are bone-marrow cells with the functional properties of NKPs (Fig. 2), and these provide a starting point for understanding the transcriptional regulation that takes place during commitment to the NK-cell lineage.

Immature NK cells. Unlike NKPs, mouse immature NK cells express CD161 (a family of C-type lectin receptors that includes NKRP1C (also known as CD161c), which is recognized by NK1.1 antibodies in C57BL/6 and C57BL/10 mouse strains). Although CD161 (NKRP1C) is expressed by all immature and mature NK cells in C57BL/6 mice, it is not an NK-cell-specific marker as it can also be detected on NKT cells and some γδ T cells. A family of natural cytotoxicity receptors (NCRs), including NKp30 (NK-cell protein 30), NKp44 and NKp46 (also known as NRC1), that are expressed in an NK-cell-specific manner (reviewed in Ref. 31) has been described in humans. NKp46 is highly expressed in NK cells, and mice that express green fluorescent protein (GFP) reporter genes under the control of NKp46 regulatory sequences show an NK-cell-specific transcription programme that starts at the immature NK-cell stage in the bone marrow^32,33 (J.P.D. and O. Mandelboim, unpublished observations). By contrast, immature NK cells express neither CD49b (also known as α₂-integrin; which is recognized by the monoclonal antibody DX5) nor CD43 and only low levels of CD11b (also known as α_M-integrin) and CD11c (also known as α_X-integrin). Immature NK cells begin to express CD94 and a few express the Ly49 receptor^22,34 (C.A.J.V. and J.P.D., unpublished observations). The adoptive transfer of immature NK cells in vivo results in the generation of Ly49⁺ NK cells, indicating that immature NK cells are the precursors of more mature NK cells³⁵. Immature NK cells are not cytotoxic towards YAC-1 cell targets (an NK-cell-sensitive murine thymoma cell line) in vitro and fail to secrete IFNγ²².

In humans, antibodies specific for CD161 (NKRP1A) have also proved useful in characterizing immature NK cells. A population of CD3⁻CD56⁻CD161⁺ cells has been detected in umbilical cord blood, and these cells can be generated in vitro from CD34⁺LIN⁻ cord blood cells, suggesting that initial expression of CD161 identifies human immature NK cells^36,37,38,39. These cells lack expression of KIR-family proteins and NKp46 (Refs 27,37). In addition, these CD3⁻CD56⁻CD161⁺ cells are poorly cytolytic compared with the more mature CD56⁺CD161⁺ cells, but they can acquire CD56 expression and high levels of natural cytotoxicity after in vitro culture^37,39. Interestingly, CD56⁻CD161⁺ cells were able to mediate cytotoxicity through TNF-related apoptosis-inducing ligand (TRAIL), but this was not the case for the more mature CD56⁺CD161⁺ NK cells³⁹. Although TRAIL-dependent cytotoxicity has not been described for mouse immature NK cells in the bone marrow, these cells express TRAIL ex vivo (C.A.J.V. and J.P.D., unpublished observations). Collectively, the available data argue convincingly for the existence of immature NK cells in the bone marrow of mice and humans that represent an intermediate stage in NK-cell development towards maturity (Fig. 2).

Are NKPs and immature NK cells unique to the bone marrow? The available data from mice studies suggest that this is not the case as immature NK cells have been found at other sites, including a prominent population in the peri-natal liver (Box 1), and a small but discernible population in the spleen^23,40. Human NKPs are also found in the lymph nodes^27,41. These observations are consistent with the idea that multiple sites can support NK-cell differentiation. Alternatively, it suggests that once generated in the bone marrow, NKPs and immature NK cells might have access to the circulation.

NK-cell education: linking functional competence and maturation. The process of education generates NK cells that are tolerant to self and have the capacity to 'sense' MHC class I expression levels on target cells that may vary during infection, transformation or stress (reviewed in Ref. 13). Several reports have provided evidence that self MHC interactions are important for human or mouse NK cells to attain full functional competence^8,9,10,11. However, the stage at which this process occurs in the bone marrow (or at other sites) has not yet been defined. In the periphery, both educated NK cells (inhibitory NK-cell receptors engaged by self MHC molecules) and hyporesponsive NK cells (no engagement by self MHC molecules) have a mature phenotype (CD11b^hiCD49b⁺)¹⁰. As the expression of NK-cell receptors specific for self MHC are first detected on immature NK cells²², it is plausible that education is initiated at this stage and, when completed, results in mature CD49b⁺ NK cells that are either educated or hyporesponsive. Alternatively, immature NK cells might not be competent for the educational process, either owing to intrinsic or environmental constraints, so that only more mature NK cells (expressing CD49b) would be educated. Whatever the stage, triggering of NK-cell receptors specific for self MHC would transmit a (yet ill-defined) signal to the maturing cells and represent a quality control checkpoint in the differentiation process. Self MHC recognition can activate an ITIM (immunoreceptor tyrosine-based inhibitory motif)-dependent phosphatase reaction within NK cells, and although this pathway may be involved in NK-cell education, the precise protein targets involved are not known. Last, the cell type that provides the self MHC signal (and potentially other essential membrane or soluble factors) is likewise undefined.

Although most developing NK cells become educated, around 10–20% of human or mouse NK cells do not express any known inhibitory receptors for self MHC class I molecules^8,9,10. These cells would be potentially harmful as they would not be subject to inhibition by self MHC class I molecules. One curious property of these NK cells is their reduced response to stimulation via activating receptors. This hyporesponsiveness could explain how these NK cells fail to provoke pathology under normal circumstances. Despite their functional differences, hyporesponsive and educated NK cells have an almost indistinguishable cell-surface phenotype (including similar levels of NKp46, NKG2D, CD11b and CD16) and both express perforin and granzyme B (in mice and humans). However, the presence of hyporesponsive NK cells poses questions about the efficiency of the educational procedure, which may be inadequate to assure that all NK cells are educated. Hyporesponsive mouse NK cells secreted the same amount of IFNγ in response to infection in vivo when compared to educated NK cells, and hyporesponsive NK-cell function could be corrected by culture with IL-2 in vitro¹⁰. So, hyporesponsive NK cells may have biological roles and utility beyond 'missing-self-MHC' recognition. Interactions between C-type lectin-related ligands and stimulatory or inhibitory NKPR1 receptors^42,43 might parallel or complement the process of education and help to explain the development of functional competence by hyporesponsive NK cells in the absence of classical MHC molecules. Defining the stage at which these different receptors are triggered might shed light on the mechanisms underlying the educational process. Finally, functions that do not involve receptors for self MHC class I molecules and instead involve the detection of viral or microbial infections might be independent of the process of education, although the acquisition of self-MHC-receptor-independent functions might require an equivalent educational process. So, development of complete functional capacities by NK cells might necessitate an ensemble of educational processes.

Mature NK cells: further complexities in phenotype and function. NK cells that have completed their education and attained full functional competence leave their generation site to take up residence in the periphery. In mice, the main peripheral sites where mature NK cells are found include the spleen, liver, lungs and blood and, to a lesser extent, the bone marrow, lymph nodes and thymus. Mature NK cells in mice express CD49b, CD122, CD161, NKG2D, NKp46 and a range of Ly49 receptors, as well as high levels of CD11b and CD43. These NK cells lyse YAC-1 targets and produce cytokines (IFNγ and TNF) after stimulation with target cells or exposure to a combination of IL-12 and IL-18 (reviewed in Ref. 18). Although these observations suggest that most mature NK cells are homogeneous, other evidence points to the contrary. For example, only a fraction of splenic NK cells expresses IFNγ after stimulation, and analysis of CD107a expression (as a marker of granule exocytosis) suggests that only a subset of NK cells are involved in natural cytotoxicity⁴⁴. So it is likely that the mature NK-cell pool in mice is composed of NK-cell subsets with different functional properties.

Hayakawa and Smyth identified CD27 as a new marker that subdivides the NK-cell pool in mice⁴⁵. On the basis of CD27 and CD11b expression, these authors characterized three NK-cell subsets and proposed a linear differentiation model of CD11b^lowCD27⁺ to CD11b^hiCD27⁺ to CD11b^hiCD27⁻ NK cells. This scheme is based on the early appearance of CD11b^low CD27⁺ NK cells after bone-marrow transplantation and the observation that purified CD11b^lowCD27⁺ NK cells give rise to CD11b^hiCD27⁺ NK cells after adoptive transfer^45,46. Phenotypically, CD11b^lowCD27⁺ NK cells express higher levels of the CXC-chemokine receptor 3 (CXCR3) and slightly lower levels of Ly49 receptors, whereas the CD11b^hiCD27⁻ subset lacks CXCR3 expression but expresses more Ly49 receptors. The CD11b^hiCD27⁻ subset has increased expression of KLRG1 (killer-cell lectin-like receptor subfamily G, member 1), an inhibitory receptor that is induced after chronic stimulation and proliferation⁴⁷. Functionally, CD11b^lowCD27⁺ NK cells are potent cytokine producers but are less cytotoxic than the intermediate CD11b^hiCD27⁺ subset, and both of these subsets are more functional than CD11b^hiCD27⁻ NK cells. Interestingly, CD11b^lowCD27⁺ NK cells are prevalent in the lymph nodes and bone marrow, whereas the CD11b^hiCD27⁻ NK cells are enriched in the blood, spleen and lungs. Collectively, these results provide strong evidence for an activation-induced differentiation of peripheral NK cells in mice, which modifies their functional capacities (Fig. 2). However, whether subsets defined by the CD27 marker represent developmentally distinct NK-cell subsets has not been clearly demonstrated. As the recently described IFN-producing killer DC (IKDC) subset has similarities to activated NK cells (Box 2, Table 1), it would be interesting to investigate whether the expression of MHC class II molecules and co-stimulatory receptors (CD80 and CD86) are enriched in the CD11b^hiCD27⁺ NK-cell subset.

Table 1 Comparison of dendritic-cell and natural-killer-cell subsets

Mature human NK cells represent a substantial fraction of circulating lymphocytes in humans (up to 20%), but they are typically less frequent in the spleen and bone marrow (5–10%), and are uncommon in uninfected lymphatics (< 1%)^1,48,49. Lanier and colleagues first proposed that human peripheral blood NK-cell subsets existed⁵⁰, and subsequently an impressive collection of literature detailing the functional and phenotypic differences between mature human peripheral NK-cell subsets has been compiled (reviewed in Refs 51,52). Two human NK-cell subsets were defined by the differential cell-surface expression of CD16 (also known as FcγRIII) and CD56 on circulating CD3⁻ lymphocytes; CD16⁺CD56^low NK cells represent ∼90% of NK cells in the blood, whereas the remainder are CD16⁻CD56^hi NK cells⁵³. CD16⁻CD56^hi NK cells express CD94–NKG2A, CCR7, CD25 and CD117 (also known as KIT) but lack expression of KIRs, whereas CD16⁺CD56^lowNK cells are CD94–NKG2A^+/−CCR7⁻CD25⁻CD62L⁻CD117⁻ but express KIRs^{54,55,56,57,58,59,60,61}. These two NK-cell subsets also show differential functions: CD16⁻CD56^hi NK cells produce greater amounts and a wider range of cytokines (IFNγ, TNF, granulocyte/macrophage colony-stimulating factor (GM-CSF), IL-13 and IL-10) than CD16⁺CD56^lowNK cells^54,62,63. By contrast, CD16⁺CD56^lowNK cells have abundant intracellular perforin and display enhanced cytotoxicity^53,61,63. Whether these two NK-cell subsets in humans share a precursor–product relationship or are the result of independent pathways of development is not known. Recent molecular characterization of the thymic NK-cell developmental pathway in mice (see later) may provide a means to address the second issue.

Similar to mouse CD27⁺ NK-cell subsets, human CD56⁺ NK-cell subsets show some degree of plasticity, as cytokine activation can modify their phenotype and function. Cytokine stimulation downregulates CD16 expression and upregulates CD56 and CCR7 expression on CD16⁺CD56^low NK cells in vitro^64,65. Moreover, cytokines can greatly enhance cytotoxicity and cytokine production of CD16⁻CD56^hi and CD16⁺CD56^low NK-cell subsets, respectively^{63,64,65,66,67}. Whether repeated stimulation results in functional exhaustion in human NK cells has not yet been demonstrated, but could be relevant in the case of chronic infections and inflammation.

A new pathway of NK-cell development in the thymus. A new thymic pathway of NK-cell development has recently been characterized in mice^68,69. Thymic NK cells were found to express high levels of CD127 (the IL-7 receptor a-chain) and were functionally distinct from CD127⁻ NK cells from the spleen, blood and liver. Thymic CD127⁺ NK cells had a CD11b^lowCD16⁻CD69⁺Ly49⁻ phenotype and showed reduced lytic potential, but produced high amounts of IFNγ, TNF and GM-CSF when re-stimulated in vitro⁶⁹. Interestingly, thymic CD127⁺ NK cells were exported to the lymph nodes, where they represented a substantial fraction of resident NK cells, and CD127⁺ NK cells were reduced in the lymph nodes of athymic mice⁶⁹. These data suggest that haematopoietic precursor cells that seed the thymus generate NK cells by a different developmental programme than that used to generate NK cells in the bone marrow, although it remains possible that completely different precursors are the cellular substrates for NK-cell development at these two sites.

The zinc-finger transcription factor GATA3 was found to be essential for the development of thymic NK cells⁶⁹. Although GATA3 was previously shown to be redundant for bone-marrow-dependent NK-cell development, GATA3-deficient NK cells appeared less mature and produced less IFNγ than wild-type cells, which is consistent with a general role for this transcription factor (among others¹⁸) in controlling NK-cell cytokine production. Although IL-7 deficiency has no effect on NK-cell development in the bone marrow, it abrogates thymic NK-cell development^69,70. T-cell receptor (TCR) γ-chain rearrangements have been detected in a subset of peripheral NK cells in mice and provide a marker of their thymic origin⁷¹. This finding suggests that thymic NK cells and thymocytes share a common precursor (possibly the bipotent T- and NK-cell progenitor that has initiated TCR gene rearrangements) and use common signals (such as IL-7) to develop. Some thymic γδ T cells are CD3^lowCD127⁺ and express NK-cell markers, which makes them difficult to distinguish from thymic NK cells⁷². Nevertheless, thymic NK cells are clearly detected in mice lacking recombination-activating gene 2 (Rag2), and thymic NK cells from wild-type and Rag2^−/− mice have similar phenotypic and functional characteristics⁶⁹. These results distinguish two subsets of mouse NK cells that rely on different sites of development, different transcriptional regulation and different soluble factors for their generation. Thymic and bone-marrow pathways of NK-cell development in mice generate NK-cell subsets with distinct phenotypic and functional characteristics.

NK cells can develop from human fetal thymic haematopoietic precursors (LIN⁻CD34^hi cells) as revealed by in vitro culture or using murine FTOCs^73,74,75,76. Later stages of thymocyte differentiation (that is, the acquisition of CD1a expression) correlate with a loss of NK-cell potential⁷⁷, suggesting that commitment to the T-cell lineage is incompatible with NK-cell development. In analogy to the GATA3-dependent thymic pathway described in mice, human thymic NK cells were uniformly shown to express CD127. Interestingly, analysis of CD127 and GATA3 expression in human peripheral blood NK-cell subsets showed that CD16⁻CD56^hi NK cells were CD127⁺ and expressed GATA3, whereas CD16⁺CD56^low NK cells were CD127⁻ and lacked GATA3 expression^69,78. Finally, there are functional similarities between mouse CD127⁺ NK cells and the human peripheral blood CD56^hiNK cells, as both populations are better producers of cytokines but are less cytotoxic than their CD127⁻ counterparts.

Lymph nodes: an additional site for NK-cell development? The lymph node contains a small population of NK cells (about 0.5%) that has escaped the interests of immunologists and NK-cell biologists for many years. However, the recent evidence for potential functional interactions between NK cells, DCs and T cells has sparked renewed interest in the biological roles of lymph-node NK cells (reviewed in Ref. 79). Phenotypically, murine NK-cell populations in the lymph node have lower proportions of Ly49⁺ cells and are enriched in CD11b^low cells⁸⁰. This may be due in part to the preferential migration of CD127⁺ thymic NK cells (which are CD11b^lowLy49⁻) to the lymph node⁶⁹. Most NK cells gain access to lymph nodes by L-selectin-mediated passage through high endothelial vessels, and NK cells have been localized to the T-cell zones and the medulla using antibodies specific for CD49b^80,81, as well as using NKp46-GFP reporter mice³³ (J.P.D. and O. Mandelboim, unpublished observations). Unlike most other tissues, the lymph node contains both CD127⁺ and CD127⁻ NK-cell subsets⁶⁹. An outstanding issue is whether thymic CD127⁺ NK cells have a distinct mechanism for localization within the lymph node and whether their tissue localization is associated with any unique biological function.

In humans, recent evidence suggests that the lymph node may provide a unique site for the development of CD16⁻CD56^hiNK cells^27,41. Lymph-node-resident cells with the LIN⁻CD34^lowCD117^hiα₄β₇-integrin⁺ phenotype can develop in a stepwise manner into CD56^hi NK cells with classical NK-cell markers (CD122, CD11b, NKG2D and NKp46)⁴¹. One intermediate in this process included a subset of CD94⁻CD117^hiCD161⁺ immature NK cells⁴¹; a similar population has been identified following culture of cord blood CD34⁺LIN⁻ cells with IL-3, IL-7, IL-15, stem-cell factor (SCF) and FMS-like tyrosine kinase 3 ligand (FLT3L)⁸². Both in vitro derived and ex vivo sorted CD94⁻CD117^hi cells could differentiate into CD56⁺CD94⁺CD117^low cells in vitro^41,82, indicating they have NK-cell precursor activity. Evidence suggests that these precursors derive from the bone marrow^26,28,29, although it is possible that they have a thymic origin. It is not known whether lymph-node LIN⁻CD34^lowCD117^hi α₄β₇-integrin⁺ cells can give rise to other cell types.

Interestingly, the ratio of CD16⁺CD56^low to CD16⁻CD56^hi NK cells in the lymph node contrasts that of the peripheral blood, with most lymph-node NK cells being CD16⁻CD56^hi (Ref. 83). CD16⁻CD56^hi NK cells express higher levels of CD62L, CCR7 and CXCR3, which may mediate their preferential recruitment to the lymph nodes^84,85,86. Enhanced cytokine production by CD16⁻CD56^hi NK cells following interactions with activated macrophages or DCs^62,87 may have a role in priming T helper 1 (T_H1)-cell responses in the draining lymph nodes^85,88. In addition to inducing CCR7 expression, IL-18 has been shown to reduce NK-cell cytotoxicity against activated DCs, promoting their production of IL-12 and IFNγ (Ref. 64), and thereby enhancing T_H1-cell responses. Whether lymph-node-resident CD127⁺ NK cells in mice have an analogous role in T-cell differentiation is not known.

Environmental 'shaping' of NK-cell diversity

A large body of evidence supports the classification of phenotypically and functionally distinct NK-cell subsets in mice and humans. Recently, cell-surface markers (such as CD127 and TRAIL) have been used in combination with CD16 to provide a means to classify the main NK-cell subsets in both of these species (hepatic immature NK cells: TRAIL⁺; thymic NK cells: CD127⁺; splenic and blood NK cells: CD16⁺). It will be interesting to know whether these markers can be used to define NK-cell subsets in other species (including rats, cows or chimpanzees). An unambiguous identification of NK-cell subsets will provide a means to further characterize their specific functions and to define their roles in immune responses.

Several questions remain as to how these different NK-cell subsets arise, including the nature of the NK-cell precursors and the mechanisms that control the developmental pathway. Although NKPs have been identified in the bone marrow²², the presence of NKPs in the thymus and the developmental pathway that ultimately generates thymic NK cells remains to be elucidated. Although it is likely that the thymic microenvironment has a role in establishing the 'thymic' phenotype (CD127 expression and TCR γ-chain rearrangements), the intercellular and soluble signals that control this process are not defined, but could include Notch signalling, which is instrumental in specifying CD127⁺ early thymocyte progenitors⁸⁹. A similar process of environmental conditioning of NK-cell subsets could occur in the placenta during pregnancy (Box 3). Although the genetic, phenotypic and functional evidence suggests that bone-marrow and thymic NK-cell developmental pathways are clearly distinct⁶⁸, it remains possible that the thymic pathway is an extended version of a small developmental window that normally occurs in the bone marrow (for example, induced by Notch signals).

Another important issue is whether any NK-cell developmental intermediates found in different tissue sites are equivalent (or identical). This includes the nature of the thymus-derived cell that ultimately seeds the lymph node (Fig. 1). Circulation in the blood has already been described for several types of multipotent and more lineage-restricted haematopoietic precursors. Previous studies have also shown that the thymus can export early T-cell progenitors that can seed the gut and develop locally⁹⁰. Whether thymus-derived early T-cell precursors provide the cellular substrate for lymph-node CD127⁺ NK cells should be assessed. The possibility that NKPs and immature NK cells could reach the circulation and home to other tissue sites would have several consequences. First, this would provide multiple opportunities for NK-cell development to occur if one developmental site were to be rendered non-functional. Second, developmental sites would not be fixed anatomically and 'neo-NK'-cell development would be possible at any site that provides the appropriate cellular and soluble factors.

Although there are plenty of data on the phenotype and function of NK-cell subsets, we still lack knowledge of their ontogenic relationships. For example, evidence suggests that CD16⁻CD56^hiNK cells are precursors of CD16⁺CD56^lowNK cells. First, the proportion of CD16⁻CD56^hi NK cells in the blood is increased in patients shortly after bone-marrow transplantation and declines over time in these patients, when the CD16⁺CD56^low NK-cell population becomes more prominent^91,92,93. Second, telomere length, which progressively diminishes with cell division and age, is significantly shorter in CD16⁺CD56^lowNK cells compared with CD16⁻CD56^hi NK cells from the same donor⁹⁴. Third, CD16⁻CD56^hi NK cells can differentiate into CD16⁺CD56^lowNK cells following transfer to alymphoid NOD/SCID mice (non-obese diabetic (NOD)/severe combined immunodeficiency (SCID) mice)⁹⁵. However, if CD16⁻CD56^hiNK cells are analogous to mouse thymus-derived NK cells, there might be alternative explanations. For example, the same data are consistent with thymic NK cells arising early after bone-marrow grafting and having a shorter replicative history. Thymus transplantation in mice can be used to assess the precursor–product relationships of thymic NK cells with other defined NK-cell subsets. Comparison of NK-cell-mediated immune responses in wild-type mice, congenitally athymic nude mice and mice receiving thymus grafts should allow a better characterization of the biological role of thymic NK cells in vivo.

Studying human NK-cell biology in mice

Understanding the full complexity of human NK-cell development is hampered by our reliance on in vitro differentiation assays. Furthermore, ethical and technical considerations restrict access to fetal and newborn tissues, and adult samples are often limited to peripheral blood or diseased tissue samples. For over 20 years, immunodeficient mice have been used to create 'humanized' mouse models that are capable of maintaining transplanted, functional human cells, including lymphocytes^96,97. Therefore, humanized mice should provide a valuable tool for studying human lymphocyte development and human immune-mediated diseases, and for vaccine development.

Recently, several groups have shown that mice deficient in T and B cells (using SCID or RAG-deficient mice) and NK cells (using mice deficient in γ_c) with a suitable genetic background (BALB/c or NOD) provide the appropriate conditions for human HSC development into T-cell and B-cell lineages^98,99,100. The human immune system (HIS) mice also develop CD16⁻CD56^hi and CD16⁺CD56^low human NK-cell subsets that are present in the bone marrow, thymus, blood, liver and spleen, although their relative representation (of the total lymphocyte pool) is clearly less than that observed in humans^99,101,102 (N.D.H. and J.P.D., unpublished observations). Administration of IL-15 in vivo¹⁰¹ increased the absolute numbers of human NK cells in HIS mice (N.D.H. and J.P.D., unpublished observations), suggesting an intimate relationship between IL-15 and NK-cell homeostasis in this model. Human NK cells generated in HIS mice were shown to express CD94, NKp46, CD62L, CD7, CD11b and KIR2DL2/KIR2DL3¹⁰¹ (N.D.H. and J.P.D., unpublished observations). In addition, NK cells derived from HIS mice can lyse K562 tumour cells (an NK-cell-sensitive human leukaemia cell line) in vitro and in vivo, and produce IFNγ after stimulation with IL-12 and IL-18 (Refs 101,102).

The ability to generate human NK-cell subsets in HIS mice and the ability of these mice to maintain human lymphocytes (for up to 6 months)⁹⁸ provides a unique opportunity to begin dissecting the process of human NK-cell development, albeit with some caveats. Given the observations in MHC-class-I-deficient mice and humans^103,104, the efficiency of NK-cell differentiation in HIS mice may depend on adequate interaction with self MHC molecules during NK-cell education; which may be suboptimal given that both thymic and bone-marrow stromal cells are of mouse origin. Whether mouse cytokines and chemokines trigger the corresponding human receptors on developing NK cells is unclear, and inappropriate binding to or signalling in human NK cells could impair their development, survival, migration and function.

Still, the HIS mouse model offers a new and exciting opportunity to investigate human NK-cell development and function. It will be particularly interesting to investigate the role of human transcription factors, signalling proteins and receptors during human lymphocyte development by modulating the expression of these genes in CD34⁺ human stem cells before engraftment, as shown using viral expression of short hairpin RNA^99,105.

Concluding remarks

Diversification of functionally distinct NK-cell subsets seems to have been faithfully conserved during evolution, as shown by their existence in mice and humans. Evolutionary conservation is usually interpreted as a proof of 'usefulness' that translates to a survival advantage for the individual. What is the evidence that NK-cell diversification provides a biological advantage?

Although NK cells are characterized by an immediate response to a given stimulus (transformed cells or cytokines), not all NK cells respond equally, either quantitatively or qualitatively. This phenomenon is clearly demonstrated using studies of NK-cell function in vitro, and probably underestimates the heterogeneous nature of the response that occurs in vivo. It is likely that an even larger diversity of NK-cell responses (and subsets) is yet to be discovered. Developing new methods that allow for in situ analysis of NK-cell function coupled with multiparameter cell-surface phenotyping will help to address this issue. Nevertheless, the potential for a multitude of NK-cell subsets provides the organism with a diversified innate arsenal to protect itself. As such, the NK-cell population as a whole has the ability to respond rapidly to diverse assaults, relying on different NK-cell subsets to react quickly with their respective, specific stimulus and thereby rapidly clearing a diseased cell or sounding the alarm of the adaptive immune system. Certain NK-cell subsets are found at specific anatomical locations, which provide an increased probability to encounter their specific targets. Other locations may contain combinations of NK-cell subsets, generating a repertoire at the population level with the capacity to handle multiple types of danger.

Box 1 | Liver immature NK cells: a window of opportunity for NK-cell development?

The presence of immature natural killer (NK) cells in the peri-natal mouse liver suggests that it is a site of NK-cell development, in addition to the bone marrow. These liver NK cells have an immature phenotype (CD11b^lowCD49b⁻Ly49⁻)^40,106. Although liver immature NK cells have little spontaneous cytotoxic activity against YAC-1 cell targets, they express TRAIL (tumour-necrosis-factor-related apoptosis-inducing ligand) and can eliminate TRAIL-sensitive targets in vitro and in vivo^40,106. In this way, immature NK cells in the liver of newborn mice seem to be analogous to human immature NK cells in human cord blood, which can also kill their targets in a TRAIL-dependent manner³⁹. Mouse TRAIL⁺ immature NK cells can develop into more mature TRAIL⁻Ly49⁺ NK cells following adoptive transfer in vivo⁴⁰, a property that is reminiscent of the precursor capacity of human cord blood CD56⁻CD161⁺ TRAIL⁺ immature NK cells, which generate mature CD56⁺CD161⁺ TRAIL⁻ NK cells when cultured in vitro³⁹. TRAIL⁺ immature NK cells are markedly reduced in the liver of adult mice compared with newborn mice, suggesting that NK-cell development in the liver is a transient phenomenon and coincides with the haematopoietic capacity of this organ during the peri-natal period.

NK cells also represent about half of the infiltrating lymphocytes in the human liver^107,108. Interestingly, human liver NK cells have a CD56^low phenotype, but lack CD16 expression and therefore do not correspond to the typical mature circulating CD16⁺CD56^low NK cells^107,108,109. One clue to the biological role for TRAIL⁺ NK cells is provided by studies of patients with chronic infection with hepatitis B virus. Infected hepatocytes express the TRAIL receptor and are susceptible to TRAIL-dependent apoptosis by CD56^hiTRAIL⁺ NK cells, which are enriched in the liver of these patients¹¹⁰. TRAIL expression by human liver NK cells could be induced by interferon-α (IFNα)¹¹⁰; by contrast, TRAIL expression by mouse liver NK cells seems to be under the control of IFNγ (Ref. 40). Based on these findings, one hypothesis is that TRAIL-dependent killing by NK cells could serve as an inducible defence mechanism that protects the organism from infection and/or cellular transformation. This mechanism could also be relevant during the activation of mature NK cells in response to immune stimulation.

Box 2 | IKDCs: a population containing activated NK cells?

Some mouse B220⁺CD11c^intLy6C⁻ cells co-express both natural killer (NK)-cell and dendritic cell (DC) markers. These B220⁺CD11c^int cells are highly cytolytic, can be stimulated to produce both type I and type II interferons (IFNs) and can present peptides to antigenic-specific T-cell clones. Based on these properties, this cell subset was named IFN-producing killer DCs (IKDCs)^111,112.

So far, the function of IKDCs has been mostly assessed at the population level. It would appear important, however, to rigorously demonstrate that single cells behave similarly, as multifunctional IKDCs may simply reflect heterogeneity in the B220⁺CD11c^intLy6C⁻ cell pool. IKDCs were originally distinguished from NK cells by their higher level of expression of B220 and CD11c. However, B220 expression is not bimodal on peripheral NK cells and therefore the comparison of B220⁺ IKDCs with B220⁻ NK cells is arbitrary. Other than IFNα production (which might be explained by contamination of IKDCs with plasmacytoid DCs^111,113), the functional activities of IKDCs and NK cells in mice are quite similar (Table 1).

What then distinguishes IKDCs from NK cells? One report indicated that IKDCs develop independently of the common cytokine-receptor γ-chain (γ_c), whereas conventional NK cells are completely absent in these mice^111,113. Curiously, CD122 (the interleukin-2 receptor b-chain) was required for IKDC development¹¹². Although a dependence on CD122 expression would be consistent with the known ability of interleukin-2 to synergize with imatinib mesylate to enhance TRAIL-mediated cytotoxicity by IKDCs¹¹¹, it is not clear how a CD122-dependent, yet γ_c-independent, cell would be generated (as all known CD122-dependent ligands are also γ_c dependent).

A second distinguishing feature of IKDCs is their capacity to present (but not process) antigen, which is probably related to their high level of expression of MHC class II molecules¹¹². Most resting mouse NK cells lack expression of MHC class II molecules¹¹², although these molecules can be induced after activation in vitro and in vivo (C.A.J.V. and J.P.D., unpublished observations). Similarly, most resting human NK cells are HLA-DR negative, although inducible expression was observed in vitro in a mixed-lymphocyte reaction¹¹⁴ and is seen in activated human NK cells isolated from the tonsils following chronic inflammation or from the decidual tissue of pregnant women infected with cytomegalovirus¹¹⁵. HLA-DR⁺ human NK cells have a large granular morphology, express increased levels of CD86, CD80 and CD70 molecules, as well as MHC class II molecules, are highly cytotoxic, and, similar to their murine equivalents, can present antigen to CD4⁺ T cells^114,115. So, increased MHC class II expression appears to characterize activated human and mouse NK cells that have enhanced functional capabilities. Should IKDCs be considered a distinct cell subset? Based on the available evidence, we propose that B220⁺CD11c^intLy6C⁻ cells (IKDCs) in mice are simply activated NK cells.

Box 3 | Uterine NK cells: a unique subset or NK cells shaped by the environment

During pregnancy, the uterine decidua is dominated by the presence of CD16⁻CD56^hi natural killer (NK) cells¹¹⁶, known as uterine NK cells. Uterine NK cells are not identical to the CD16⁻CD56^hi NK cells found in the blood, as they express CD9, CD103 and a substantially greater proportion express one or more members of the KIR (killer-cell immunoglobulin-like receptor) family¹¹⁷. CD9, CD103 and KIR expression may be directly induced by transforming growth factor-β (TGFβ), produced by decidual stromal cells. The regulatory activity of TGFβ can also reduce CD16 expression on CD16⁺CD56^low NK cells¹¹⁷, thereby impairing NK-cell-mediated antibody-dependent cell-mediated cytotoxicity (ADCC). Finally, although uterine NK cells express similar levels of NKp46 (NK-cell protein 46), lymphocyte function-associated antigen 1 (LFA1) and NK group 2, member D (NKG2D) as peripheral blood NK cells, they express higher levels of NKp44 and fail to polarize perforin-containing granules required for target-cell lysis¹¹⁸. So, uterine NK cells seem to be phenotypically and functionally distinct from other human NK-cell subsets, which probably results from the unique cytokine environment present in the pregnant uterus.

Uterine NK cells are thought to be recruited from the peripheral blood to the decidua; indeed, CD16⁻CD56^hi NK cells from both the peripheral blood and the decidua preferentially express CXC-chemokine receptor 4 (CXCR4) and migrate towards soluble CXC-chemokine ligand 12 (CXCL12) found in the decidua, particularly on extravillous trophoblasts¹¹⁹. However, uterine NK cells could develop in situ, as the decidua contains lineage⁻CD34⁺ cells that differentiate into CD16⁻CD56^hi cells in the presence of decidual-stroma-conditioned medium, interleukin-15 and stem-cell factor in vitro¹¹⁷. Irrespective of their origin, the function of uterine NK cells seems to be highly specialized, as the extra-embryonic placenta is allogeneic yet is not rejected by the maternal NK cells. Extravillous trophoblasts invade the uterus and are involved in tissue remodelling, to ensure adequate maternal blood flow to the intervillous space for the fetus and placenta¹¹⁶. Decidual NK cells produce CXCL8 and CXCL10, which attract extravillous trophoblasts via CXCR1 and CXCR3, respectively. In addition, vascular endothelial growth factor (VEGF) and placental growth factor (PLGF) are produced at significantly higher levels by uterine NK cells compared with peripheral blood NK cells and could assist vascular growth and placental development^116,120.