The broad spectrum of human natural killer cell diversity

SUMMARY

Natural killer (NK) cells provide protection against infectious pathogens and cancer. For decades it has been appreciated that two major NK cell subsets (CD56^bright and CD56^dim) exist in humans and have distinct anatomical localization patterns, phenotypes, and functions in immunity. In light of this traditional NK cell dichotomy, it is now clear that the spectrum of human NK cell diversity is much broader than originally appreciated as a result of variegated surface receptor, intracellular signaling molecule, and transcription factor expression; tissue-specific imprinting; and foreign antigen exposure. The recent discoveries of tissue-resident NK cell developmental intermediates, non-NK innate lymphoid cells, and the capacity for NK cells to adapt and differentiate into long-lived memory cells has added further complexity to this field. Here we review our current understanding of the breadth and generation of human NK cell diversity.

eTOC blurb for Freud et al

Recent advances in the field of human natural killer cell biology have revealed that there is a remarkably high amount of cellular diversity within different tissues. Freud et al review these advances and provide insight into the generation of natural killer cell diversity and its roles in innate immunity.

INTRODUCTION

Natural killer (NK) cells are large granular lymphocytes endowed with the inherent capacities to recognize and kill foreign, infected, and malignant cells and also to modulate other aspects of the immune system through their rapid production of numerous cytokines and chemokines (Caligiuri, 2008; Orr and Lanier, 2010). NK cells constitute approximately 5–15% of circulating lymphocytes in healthy adults and therefore represent one of the three major human lymphocyte lineages including B cells and T cells. There are many functional and phenotypic similarities between NK cells and T cells, particularly CD8⁺ T cells (Sun and Lanier, 2011). However, the ways in which these two cell types develop in the body, detect infected or malignantly transformed cells, and become activated are distinct. T cells develop in the thymus and become activated when their somatically rearranged T cell receptors (TCRs) encounter foreign antigen in the context of self major histocompatibility complex (MHC) molecules and costimulatory ligands expressed on antigen presenting cells (Halle et al., 2017). In contrast, NK cells primarily develop outside of the thymus in various other tissues, and they do not express a rearranged TCR (Ritz et al., 1985; Yu et al., 2013). Rather, NK cells are regulated by numerous types of germline-encoded, non-rearranged activating and inhibitory receptors, including two major types of MHC class I-binding receptors: the evolutionarily conserved and non-polymorphic, heterodimeric, C-type lectin-like receptors formed primarily by the combination of CD94 with either NKG2A (inhibitory) or NKG2C (activating); and the large polygenic and highly polymorphic family of killer immunoglobulin-like receptors (KIRs) (Colonna et al., 1999). Whereas CD94/NKG2 heterodimeric receptors bind non-classical MHC class IB molecules such as HLA-E, KIRs bind to classical MHC class IA molecules HLA-A, -B, and -C. These MHC class I-binding receptors regulate NK cell function in an antigen-independent fashion through binding to conserved amino acid residues located outside of the peptide-binding pockets of MHC class I molecules (Das and Khakoo, 2015).

Given the distinct ways that T cells and NK cells are designed to respond to MHC class I molecule expression (i.e. T cell activation through the TCR; NK cell regulation through MHC class I-binding receptors), it is likely that T cells and NK cells provide complementary immunity against infection and cancer, in which MHC molecules may or may not be downregulated (Garrido et al., 2017; Griffin et al., 2010). Moreover, because it takes days to mount a robust T cell response in the immunologically naïve setting, T cell (i.e. adaptive) immunity is complemented by a much more rapid innate response in part mediated by NK cells (Deguine and Bousso, 2013; Jain and Pasare, 2017). However, this is an overly simplified view of T cell and NK cell responses and functions, because T cells can express many NK cell-associated receptors including MHC class I-binding receptors (Davis et al., 2015; Strauss-Albee et al., 2014). In addition, NK cells can adapt through epigenetic remodeling in response to environmental exposures and can even form long-lasting immunological memory (O’Sullivan et al., 2015; Tesi et al., 2016).

In light of the functional and phenotypic overlap of T cells and NK cells, the specific requirement for adequate NK cell function in humans is highlighted by the identification and characterization of patients with selective NK cell deficiencies and who succumb to uncontrolled viral infections, particularly those belonging to the herpes family of viruses (Mace and Orange, 2016). Moreover, from the ground-breaking translational work of Velardi, Ravetch, Levy, and several other scientists, it is clear that human NK cell effector function has a critical role in the direct elimination of malignancy (Clynes et al., 2000; Ruggeri et al., 2002; Weng and Levy, 2003). This was first highlighted in a seminal study by Ruggeri et al who observed that donor CD34⁺ progenitor cell-derived NK cells are critical to successful outcomes following T cell-depleted, MHC haploidentical, allogeneic, hematopoietic stem cell transplantation (allo-HSCT) for acute myeloid leukemia (AML) (Ruggeri et al., 2002). Patients whose donor-differentiated NK cells demonstrate alloreactivity in the graft-versus-leukemia direction have a significantly superior survival compared to those patients whose donor-differentiated NK cells lacked alloreactivity. Notably, this is also not associated with an increased risk of graft-versus-host disease. A likely mechanism to account for the NK cell-mediated alloreactivity is based on the “missing self” hypothesis, whereby NK cells can recognize and specifically destroy cells that have downregulated MHC class I molecules (Ljunggren and Karre, 1990). In the setting of MHC haploidentical allo-HSCT, disparate MHC class IA and inhibitory MHC class I-binding receptor genotypes between the donor and recipient can in some instances result in a “mismatch” whereby a donor NK cell MHC class I-binding receptor does not engage with a cognate ligand on the recipient target cell and therefore sees it as “missing self” (Caligiuri, 2008). In this scenario, the MHC class I-binding receptor-mediated inhibitory signal is absent allowing the uninhibited donor NK cell to kill its AML target. NK cell killing can be further promoted in the presence of AML-directed monoclonal antibodies and/or if the leukemic blasts express stress-induced ligands that can engage surface NK cell activating receptors such as NKG2D (Vasu et al., 2016).

The clinical data reported by Ruggeri et al have inspired a wave of investigations to evaluate the potential efficacy of augmenting NK cell numbers and/or functions in the post-transplant setting for AML as well as in other malignancies (Moretta et al., 2011; Velardi, 2012). In order to circumvent the inherent clinical and immunological challenges related to allo-HSCT, alternative NK cell-based approaches to the treatment of AML and other forms of cancer have been developed. These include providing patients with recombinant interleukin (IL)-2 or IL-15 to augment endogenous NK cell function; adoptive transfer of fresh or cytokine-activated allogeneic NK cells derived from healthy umbilical cord blood (UCB) or adult peripheral blood (PB) donors; and infusing genetically modified chimeric antigen receptor NK cells into patients (Chen et al., 2016b; Fehniger and Cooper, 2016; Han et al., 2015; Knorr et al., 2014; Mehta et al., 2015; Moretta et al., 2014; Vasu and Blum, 2013).

Overshadowing all of the above is the somewhat recent realization that in any given individual and in any given tissue the total NK cell population is much more diverse than previously appreciated in terms of developmental, phenotypic, and functional parameters. Indeed as we learn more about NK cells in experimental animal models and in humans, it is now clear that the traditional view of the NK cell lineage as comprised of a relatively homogeneous population of cells with similar functions and longevity is not at all accurate. Rather, the NK cell lineage is remarkably diverse. It goes without saying that this diversity invariably impacts many areas of medicine, including cancer immunotherapy, infection control, and vaccine development. Here we provide a working definition of the human NK cell lineage and then discuss our current understanding of the breadth and generation of human NK cell diversity.

DEFINING THE NK CELL LINEAGE IN HUMANS

Before discussing human NK cell diversity it is useful to first define an NK cell as well as the NK cell lineage. The traditional definition of a mature NK cell is that of a non-T, non-B lymphocyte that can rapidly produce interferon gamma (IFN-γ) following stimulation and that can also mediate cellular cytotoxicity (Caligiuri, 2008). As mentioned earlier NK cells can perform a myriad of functions, yet these two attributes – IFN-γ production and cytotoxicity - are widely accepted in the field as the defining functions of mature NK cells. Here we refer to these two functions as defining features of conventional NK cells (cNK), noting that a developmentally immature NK cell (iNK) or a specialized tissue-resident NK cell (trNK) may lack one or both of these functions and yet still be considered part of the NK cell lineage (Luetke-Eversloh et al., 2013). This then begs the question: what defines the NK cell lineage? The answer to this question is much more complex as it must take into account multiple layers of developmental, immunophenotypic, and functional diversity that continue to be elucidated and refined. This is particularly challenging for the NK cell lineage, because to date there are no known surface markers that can specifically and unequivocally identify NK cells in the same way that, for instance, surface CD3 expression specifically identifies T cells. Rather, NK cells have been traditionally identified immunophenotypically (e.g. by flow cytometry) by first excluding other leukocytes expressing “lineage” (Lin) markers, including those specifying T cells (e.g. CD3 and CD5), B cells (e.g. CD19 and CD20), and myelomonocytic cells (e.g. CD13 and CD14), and then by including CD45⁺ lymphocytes expressing the neural cell adhesion molecule-1 (CD56) and/or the low affinity Fc gamma receptor 3A (CD16) that mediates antibody-dependent cellular cytotoxicity (ADCC) (Lanier et al., 1986). Again, neither of these two antigens (CD56 and CD16) is specific to the NK cell lineage.

It is important to note that NK cells are not the only kind of Lin⁻ lymphocyte. It is now known that NK cells represent one subtype of a diverse and recently discovered innate lymphoid cell (ILC) family (Artis and Spits, 2015). All ILCs share a close developmental relationship that is reliant on key transcription factors (TFs), such as ID2, and on cytokines that signal through the common gamma-chain (γc) shared by IL-2, IL-7, and IL-15, among others (Lim et al., 2017b; Montaldo et al., 2016). Like NK cells the other ILCs share many overlapping features with T cells, but unlike NK cells they are non-cytotoxic. In addition, whereas NK cells comprise the overwhelming majority of ILCs in the PB, spleen, and BM, non-NK ILCs are relatively enriched and more highly represented in other tissues, particularly other secondary lymphoid tissues (SLTs) including mucosa-associated lymphoid tissues (MALTs) in the oral cavity and gut (Mjosberg and Spits, 2016).

All ILC subsets lack the aforementioned Lin-specifying antigens and may be further distinguished from each other ex vivo using combinations of additional ILC subset-associated markers. In particular, the TF EOMES, cytolytic granules (i.e. perforin and granzymes), MHC class I-binding receptors, and CD16 are currently widely accepted as NK cell-specific markers among ILCs (Spits et al., 2013). In addition, expression of the surface activating receptor, NKp80, is detected on virtually all cNK cells in healthy PB and closely correlates with EOMES, perforin, and MHC class I-binding receptors during NK cell development in vivo (Freud et al., 2016; Vitale et al., 2001). Therefore, NKp80 also appears to be an NK cell-specific marker among human ILCs. In contrast, the traditional pan-NK cell marker, CD56, is not restricted to NK cells even among ILC subsets as it can also be expressed by Group 3 ILCs (Cupedo et al., 2009). Given all of the above, for the purposes of this review we refer to the human NK cell lineage as a diverse population of developmentally-related cell types that share common immunophenotypic and functional features and that to the best of our knowledge can be distinguished from traditionally described Lin⁺ leukocytes as well as from other recently described Lin⁻ ILCs based on the expression of EOMES, cytolytic granules, MHC class I-binding receptors, NKp80, and/or CD16 by NK lineage cells.

THE BREADTH OF HUMAN NK CELL DIVERSITY

In the following sections we discuss the various levels of human NK cell diversity: traditional PB cNK cell subsets; non-conventional trNK cells; population diversity associated with variegated receptor expression and NK cell function; and environmental-induced diversity including adaptive NK cells. These facets of human NK cell diversity are represented schematically in Figure 1.

Figure 1. Facets of human NK cell phenotypic diversity.

Traditionally, two major subsets of NK cells have been characterized in PB based on their differential expression of CD56 and CD16, termed CD56^bright and CD56^dim. Recent studies have expanded on this definition by characterizing NK cells within additional human tissues each possessing distinct phenotypic profiles as indicated. The CD56^bright cNK cell in the blood shares some common features with several of these non-conventional trNK cell populations, but it is becoming clear that trNK are actually quite distinct functionally as well as immunophenotypically (Björkström et al., 2016; Freud et al., 2014; Melsen et al., 2016). Likewise, we are now appreciating the high amounts of diversity within the CD56^dim cNK cell population, as by the generation of adaptive NK cells in response to environmental stimuli such as virally infected cells or cancer (Björkström et al., 2016; Schlums et al., 2015). While many of these populations share a common CD56^brightCD16⁻ immunophenotype they are not necessarily identical to each other, likely due to the influence of a unique microenvironment in each tissue.

For decades it has been appreciated that the human cNK cell population in PB is functionally heterogeneous. Two major PB cNK cell subsets were first identified in the 1980s according to the differential expression of CD56 and CD16, namely CD56^brightCD16^lo/− and CD56^dimCD16⁺ (hereafter referred to simply as CD56^bright and CD56^dim, respectively) (Lanier et al., 1986; Lanier et al., 1983). Most cNK cells in PB are CD56^dim, whereas CD56^bright cNK cells represent the minority of cNK cells in healthy donor PB, usually constituting ≤15% of the total circulating cNK cell pool (defined as Lin⁻CD56⁺). In healthy individuals, both PB cNK cell subsets constitutively express similar amounts of various intracellular and surface factors including T-BET, EOMES, the common IL-2 and IL-15 receptor beta chain (IL-2/15Rβ, CD122), 2B4, DNAM-1, NKG2D, NKp30, and NKp80 (Caligiuri, 2008; Collins et al., 2017; Lanier, 1998; Moretta et al., 2014; Simonetta et al., 2016). Moreover, in a general sense both subsets show the characteristic functions of cNK cells; i.e. they are capable of producing cytokines and mediating cellular cytotoxicity (Cooper et al., 2001a). However, there are also many very distinct phenotypic and functional differences between these two PB cNK cell subsets. For example, a significant fraction of CD56^bright cNK cells uniquely and constitutively expresses low amounts of c-kit (CD117), IL-2Rα (CD25), and IL-7Rα (CD127), suggesting that the respective receptor ligands, c-kit ligand (KL), IL-2, and IL-7, selectively influence the survival and/or function of CD56^bright cNK cells in vivo (Caligiuri et al., 1990; Carson et al., 1994; Romagnani et al., 2007; Vosshenrich et al., 2006). PB CD56^bright cNK cells are also characteristically uniformly positive for the inhibitory MHC class I-binding receptor, CD94/NKG2A, as well as for the activating receptor, NKp46, yet they mostly lack expression of KIRs and the activating MHC class I-binding receptor, CD94/NKG2C (Caligiuri, 2008). In terms of function, CD56^bright cNK cells can rapidly produce large amounts of immunomodulatory cytokines and chemokines in response to IL-1β, IL-2, IL-12, IL-15, and/or IL-18 produced by activated monocytes, dendritic cells (DCs), and T cells (Cooper et al., 2001b; Fehniger et al., 1999). However, in line with their constitutively low expression of intracellular perforin and granzymes A and B and their low or undetectable surface expression of CD16, CD56^bright NK cells are poor mediators of direct cytotoxicity and ADCC, although these functions can be augmented and/or induced following in vitro stimulation (Michel et al., 2016).

In contrast to the above, PB CD56^dim cNK cells show low or variable expression of NKp46 and CD94/NKG2A and yet more brightly express CD16, perforin, and granzymes as well as more KIRs and CD94/NKG2C. In addition, certain other markers, such as CD57, KLRG1, and PEN5, are only expressed (albeit variably) by CD56^dim cNKs (Cooper et al., 2001a). Predictably, CD56^dim cNK cells are more highly cytotoxic compared to CD56^bright NK cells ex vivo. However, when stimulated by IL-12, IL-15, and IL-18 CD56^dim cNK cells produce much lower amounts of cytokines, such as IFN-γ, in comparison to CD56^bright cNK cells (Fehniger et al., 1999). Nonetheless, CD56^dim cNK cells can produce abundant cytokines upon direct engagement with target cells that express receptor-triggering ligands and/or are coated with antibodies that can trigger CD16 (Anfossi et al., 2006; Bjorkstrom et al., 2010; Strauss-Albee et al., 2015).

Given these distinct functional capabilities and responses to extrinsic stimuli, it is likely that CD56^bright and CD56^dim cNK cells fulfill distinct roles in immunity, with CD56^bright cNK cells serving more of an immunomodulatory role and CD56^dim cNK cells serving more of a cytotoxic effector role (Cichocki et al., 2016b; Cooper et al., 2001a). Consistent with this notion, CD56^bright cNK cells represent the predominant NK cell subset in SLTs, including LNs, where they reside in the parafollicular regions in close proximity to DCs and T cells and can reciprocally interact with and modulate the activity of these cell types (Cooper et al., 2004; Fehniger et al., 2003; Ferlazzo et al., 2004a; Ferlazzo et al., 2004b). CD56^bright cNK cells were also recently shown to specifically respond to and protect against infection by the Epstein-Barr virus (EBV) in SLTs (Hatton et al., 2016; Lunemann et al., 2013).

Interestingly, the relative proportions of CD56^bright and CD56^dim cNK subsets in many other tissues are also very different from what is observed in PB. Notably, in healthy individuals CD56^dim cNK cells normally predominate in tissues such as BM, lung, spleen, subcutaneous adipose tissue, and breast tissue, whereas CD56^bright cNK cells constitute much higher proportions of total NK lineage cells in MALTs (e.g. gastric and intestinal mucosa), liver, uterus, visceral adipose tissue, adrenal gland, and kidney (Carrega et al., 2014; Freud et al., 2006). Not surprisingly, these distinct patterns of tissue localization correspond with markedly different patterns of chemokine receptor expression. For example, although both cNK cell subsets express CXCR4, PB CD56^bright cNK cells uniquely express CCR7 and CXCR3, and they also constitutively express high amounts of L-selectin (CD62L); these molecules facilitate homing and/or entry into SLTs and other tissues where the reciprocal ligands are abundantly expressed by the local cellular milieu. In contrast, PB CD56^dim cNK cells lack or show relatively low expression of these receptors yet instead express CXCR1, CXCR2, CX3CR1, sphingosine-1-phosphate receptor 5 (S1P5), and the receptor for chemerin (Carrega et al., 2014; Cichocki et al., 2016b; Maghazachi, 2010). Given their presence and relatively high frequencies in so many tissues, including the numerous SLTs throughout the body, it is intriguing that, contrary to original belief, CD56^bright cNK cells are more abundant than CD56^dim cNK cells. Nonetheless, CD56^bright cNK cells may not be the most abundant NK cell subset overall, as numerous recent reports reveal marked heterogeneity within tissues due to the presence of specialized trNK cells that share some immunophenotypic and functional features with CD56^bright cNK cells but that are otherwise distinct (Melsen et al., 2016).

NON-CONVENTIONAL NK CELLS IN HUMAN TISSUES: SPECIALIZED trNK CELLS

Recently there has been increasing interest in characterizing and studying NK cells in human tissues. For many years it has been appreciated that CD56^brightCD16⁻ NK cells represent the predominant NK cell population in both SLTs and in the uterus where their respective functions likely include promoting T helper-1 (Th1) cell-associated responses and in facilitating decidual changes associated with pregnancy such as vascular remodeling (Gaynor and Colucci, 2017; Melsen et al., 2016). In human LNs and tonsils NK cells are relatively rare and represent up to only 1–2% of total lymphocytes. In contrast, NK cells constitute the majority of lymphocytes in the uterus, primarily in the gravid uterus during the first trimester. Growing evidence indicates that specialized trNK cell populations are present in these tissues and are distinct from their cNK cell counterparts that predominate in PB. For example, trNK cells in human LNs and tonsils have bimodal expression of DNAM-1, have high expression of CD54 (ICAM-1), and have slightly lower expression of CD56 compared to CD56^bright cNK cells in PB (Lugthart et al., 2016; Lunemann et al., 2013). On the other hand, uterine trNK cells are characteristically “super bright” for CD56 and also express high amounts of CD9 and KIRs, neither of which are expressed on most PB CD56^bright cNK cells (Koopman et al., 2003). Uterine trNK cells and PB cNK cells have also recently been shown to express distinctly different splice variants of the NCR2 and NCR3 genes, which encode NKp44 and NKp30, respectively (Siewiera et al., 2015).

More recently, trNK cell populations have also been described in human BM, spleen, lung, and liver (Aw Yeang et al., 2017; Cuff et al., 2016; Hudspeth et al., 2016; Lugthart et al., 2016; Lunemann et al., 2013; Marquardt et al., 2015; Marquardt et al., 2017; Stegmann et al., 2016). These various populations of human trNK cells described to date share a number of common features with each other and are overall quite distinct from PB CD56^bright cNK cells (Melsen et al., 2016). First of all, trNK cells lack expression of CCR7 and CD62L, which are characteristically expressed by PB CD56^bright cNK cells. Rather, trNK cells express surface adhesion molecules and other receptors that serve to retain them within tissues and prevent their egress into the periphery. These molecules include CD69, which was originally identified as an activation-associated marker and is now known to retain cells in tissues by suppressing the function of S1P receptors. In addition, trNK cells express CCR5, the receptor for CCL3, CCL4, and CCL3L1, as well as CXCR6, the receptor for CXCL16. CD103 (integrin α_E), which binds E-cadherin, is also preferentially expressed by trNK cells in tissues such as MALTs that have an epithelial component (Cuff et al., 2016; Hudspeth et al., 2016; Lugthart et al., 2016; Stegmann et al., 2016). In addition to the above, whereas most PB CD56^bright and CD56^dim cNK cells express CD49e (integrin α₅), the absence of this surface protein appears to specifically identify trNK cells, at least in human liver (Aw Yeang et al., 2017). In mice, the expression of CD49a (integrin α₁) is uniquely expressed by trNK cells and/or ILC1s and aids in their identification in multiple tissues including the liver and salivary glands (Cortez and Colonna, 2016; Peng and Tian, 2015). However, in humans CD49 may not be as sensitive or specific a marker for all trNK cells. For example, CD49a is less frequently expressed by trNK cells than is CXCR6, and in one study CD49a⁺ trNK cells were only detected amongst a subset of individuals seropositive for human cytomegalovirus (HCMV) (Marquardt et al., 2015; Stegmann et al., 2016).

In general, trNK cells are capable of cytokine production following stimulation with IL-12, IL-15, and IL-18, but they tend to express low amounts of intracellular perforin and granzymes and are relatively poorly cytotoxic; thus, they functionally appear to be similar to PB CD56^bright cNK cells. Likewise, trNK cells in BM, SLTs, and in the liver are mostly CD94/NKG2A⁺ and NKp46^hi and for the most part lack expression of CD94/NKG2C, KIRs, CD16, and CD57. Interestingly, whereas PB CD56^bright and CD56^dim cNK cells show the same T-BET^hiEOMES^lo TF profile, trNK cells in multiple different fetal and adult tissues express relatively lower amounts of T-BET and relatively higher amounts of EOMES in comparison to PB cNK cells (Collins et al., 2017; Cuff et al., 2016; Harmon et al., 2016). This T-BET^loEOMES^hi profile that is expressed by trNK cells has also been observed among relatively immature CD94⁺CD16⁻ NK cell developmental intermediates (NKDIs) derived in vitro from human CD34⁺ hematopoietic progenitor cells (HPCs) (Collins et al., 2017). The significance of the T-BET^loEOMES^hi profile in trNK cells is not yet clear. While on the one hand, the functional and phenotypic features described above raise the possibility that trNK cells represent immature NKDIs that give rise to PB cNK cells, data from parabiotic mouse experiments and human liver transplantation studies indicate that trNK cells can persist in situ for very long periods (up to 13 years in one human study), do not enter the circulation, show different TF requirements, and are likely terminally differentiated (Cuff et al., 2016; Peng and Tian, 2015; Sojka et al., 2014; Zhang et al., 2016). Taken together, these data suggest that trNK cells collectively comprise a bona fide subset of the human NK cell lineage that is distinct from the two originally described PB cNK cell subsets.

NK CELL POPULATION DIVERSITY AS A FUNCTION OF NATURE AND NURTURE

Apart from the three major NK cell subsets described above (i.e. CD56^bright cNK cells, CD56^dim cNK cells, and trNK cells) there is also remarkable phenotypic and functional heterogeneity within the human NK cell lineage due to a high degree of cell to cell variability in terms of the numbers and relative intensities of different types of functional surface markers that are expressed. Indeed recent technological advances enabling the simultaneous assessment of more than 30 parameters with the aid of mass cytometry, or cytometry by time-of-flight (CyTOF), indicate that in any individual there may be at least 30,000 or more distinct PB NK cell phenotypes (Horowitz et al., 2013). Much of this diversity, particularly within the CD56^dim cNK subset, is imparted by the stochastic and variegated expression patterns of KIRs. Indeed the KIR family is composed of a large, diverse, and polymorphic set of inhibitory and activating receptors that bind conserved MHC class IA epitopes, which in turn are variably expressed on different HLA-A, -B, and –C molecules (Parham, 2005). In any given individual’s genome there may be a dozen or so KIR genes that are located in the leukocyte receptor complex on chromosome 19. The specific number of KIR genes varies markedly among individuals within and between different human ethnic and racial populations and also depends upon the inherited KIR haplotype. Furthermore, within any individual’s total NK cell population, only a subset of the available KIR genes is expressed by any given NK cell and in a stochastic manner. This results in a diverse NK cell repertoire that is capable of sensing minute changes in MHC expression, which impacts cancer immune surveillance and NK cell-based immunotherapy (Moretta et al., 2011).

Adding further complexity to this field is the fact that not all NK cells that express MHC class I-binding receptors are created equally. This is due to a process commonly referred to as NK cell “licensing” or “education” (Anfossi et al., 2006; Kim et al., 2005). Specifically, education means that for an NK cell to become fully responsive to surface activating receptor ligation and to be capable of efficiently killing MHC class I-deficient target cells, that NK cell must engage its inhibitory MHC class I-binding receptor(s) with cognate MHC class I ligand(s) during the developmental process (Elliott and Yokoyama, 2011; Kadri et al., 2016). Otherwise, the NK cell that is produced is hyporesponsive to receptor ligation and demonstrates poor or absent killing of MHC class I-deficient target cells. This is the scenario for essentially all NK cells that develop in MHC-deficient mice or humans (Kim et al., 2005; Zimmer et al., 1998). While it is not yet known how during development signaling through inhibitory receptors results in a poised state of NK cell responsiveness, the process of NK cell education likely ensures that each NK cell that is produced in the body maintains self-tolerance to other cells that are expressing MHC class I molecules. Notably, the availability of KIR-specific MHC class IA ligands is dependent on the HLA haplotype of the individual, because different HLA molecules contain different KIR-binding epitopes. Therefore, different KIR⁺ NK cells are educated in different individuals (Horowitz et al., 2016), and this directly impacts whether or not NK cell alloreactivity exists in the setting of haploidentical allo-HSCT for AML (Velardi et al., 2012).

Interestingly, it has been estimated that approximately 20% of CD56^dim cNK cells in healthy adult human PB lack all inhibitory MHC class I-binding receptors, including CD94/NKG2A and inhibitory KIRs (Cooley et al., 2007). This represents an unexpectedly large number of theoretically uneducated NK cells and would at first glance seem like a high amount of inefficiency in how the body produces effector NK cells. However, recent studies in both mice and humans indicate that other receptors, such as the SLAM family receptors 2B4 and SLAMF6, can also mediate NK cell education in addition to KIRs and CD94/NKG2A (Chen et al., 2016a; He and Tian, 2017; Meazza et al., 2017; Wu et al., 2016). Furthermore, there is also recent evidence that educated and uneducated NK cells can serve distinct roles in the body, with educated NK cells providing immunity against MHC-deficient tumors and uneducated NK cells providing immunity against some viral infections and tumors that retain MHC expression (Orr et al., 2010; Tu et al., 2016; Zamora et al., 2017). Of clinical importance, uneducated NK cells can also mediate enhanced cytotoxicity in the setting of antibody therapy (Tarek et al., 2012). Uneducated human cNK cells can also be induced with IL-12, IL-15, and IL-18 to overcome their baseline hyporesponsiveness and have increased functionality in response to CD16 triggering and AML target cells (Wagner et al., 2017). Therefore, it is most likely that after millions of years of evolution all NK cells produced by the human body can participate in immunity and can also likely be targeted for immunotherapeutic benefit in patients with infectious diseases and cancer.

Given the influence of genetics on human NK cell diversification and education as discussed above, one might predict that genetically identical individuals who share the same KIR and HLA genes would contain the exact same amounts of functional and phenotypic diversity (i.e. diversity “indices”) within their pools of circulating PB NK cells. However, this is not entirely true. Indeed it has been recently reported that while diversity indices are more similar between genetically identical adult twins than between unrelated adults, even twins’ NK cell populations are distinct from each other (Horowitz et al., 2013). This is largely due to differences in the expression of surface activating and costimulatory receptors such as NKG2D, NKp30, NKp46, and some activating KIRs, which can rapidly change following activation (Strauss-Albee et al., 2015). Interestingly, newborn NK cell populations have relatively low diversity indices compared to those observed in adults despite the absence of any major age-related differences in NK cell repertoire structural components (e.g. phenotypically defined cNK cell subsets such as CD56^brightCD16^lo/− and CD56^dimCD16⁺) (Strauss-Albee et al., 2017). Taken together, these data indicate that genetics largely influences the expression patterns of inhibitory receptors including inhibitory MHC class I-binding receptors, whereas environmental exposures such as infection or cancer can have a substantial impact on increasing NK cell population diversity due to changes in activating receptor expression (Beziat et al., 2013).

ADAPTIVE NK CELLS

The NK cell population diversity studies described above are in line with numerous other reports that changes in NK cell surface marker expression and relative KIR⁺ NK cell frequencies occur in the settings of cancer and infection (Beziat et al., 2012; Beziat et al., 2013; Davis et al., 2015; Fauriat et al., 2007; Freud et al., 2013; Guma et al., 2004; Guma et al., 2006; Sun et al., 2012). Many such studies were originally performed and interpreted in the conceptual framework of the traditional CD56^bright and CD56^dim cNK dichotomy. However, a number of mouse and human studies have highlighted the potential for NK cells to form immunological memory and differentiate into specialized memory-like cells now commonly referred to as adaptive NK cells (Cooper et al., 2009; O’Leary et al., 2006; Romee et al., 2012; Sun et al., 2009). These adaptive NK cell populations have been mostly extensively characterized in the settings of murine cytomegalovirus (MCMV) and HCMV (Hammer and Romagnani, 2017; Tesi et al., 2016). Indeed adaptive NK cells show an array of functional, phenotypic, epigenetic, and homeostatic differences from cNK cells that warrant their classification as a subset distinct from cNK cells (Corat et al., 2017; Hwang et al., 2012; Schlums et al., 2015; Schlums et al., 2017).

In mice, adaptive NK cells can be produced following primary infection with MCMV (Sun et al., 2009). During the infection, precursors to adaptive NK cells [likely a subset of KLRG1⁻ cNK cells (Kamimura and Lanier, 2015) – see below] that bear the Ly49H receptor become activated by MCMV-infected cells expressing the MCMV-encoded m157 molecule, which directly binds and triggers Ly49H. The activated Ly49H⁺ NK cells then rapidly expand and a small percentage differentiate into adaptive NK cells that can survive for many months if not years, although most of the originally expanded Ly49H⁺ NK cells undergo apoptosis. The small pool of persistent Ly49H⁺ adaptive NK cells can again rapidly proliferate and also show increased cytokine production when re-exposed to virus (Sun et al., 2009). In addition, they can protect newborn mice challenged with MCMV. Notably, these features are very reminiscent of how memory CD8⁺ T cells are generated and function (Sun and Lanier, 2011). Together with other murine data showing that CXCR6⁺ liver-derived NK cells can mediate antigen-specific contact hypersensitivity (O’Leary et al., 2006; Paust et al., 2010), the studies of MCMV-associated NK cells provide convincing evidence that NK cells, like T cells, can provide bona fide immunological memory (O’Sullivan et al., 2015).

Similar to what has been observed in mice, humans that have been previously infected with HCMV often contain unique and expanded PB adaptive NK cell populations bearing activating MHC class I-binding receptors, typically CD94/NKG2C (Beziat et al., 2012; Guma et al., 2006; Hwang et al., 2012; Lee et al., 2015; Schlums et al., 2015). The HCMV-associated adaptive NK cells described to date are CD56^dimCD16⁺, yet in comparison to CD56^dim cNK cells, adaptive NK cells tend to show decreased expression of surface CD7, CD161, NKp30, NKp46, and SIGLEC-7 but show retained or even higher expression of CD2, CD57, and CD85j (ILT2, LILRB1). Of note, none of these surface marker expression patterns are in and of themselves specific for adaptive NK cells, but collectively they can aid in the identification of discrete adaptive NK cell populations. In contrast, what seems to be more specific to adaptive NK cells is the downregulation of the TFs PLZF and IKZF2 as well as the variable loss of some intracellular adapter signaling molecules [FcεR1γ (often referred to simply as FcRγ), DAB2, SYK, and/or EAT-2] but not others (CD3ζ, ZAP-70, and SAP) (Hwang et al., 2012; Lee et al., 2015; Schlums et al., 2015). As recently simultaneously reported by the Kim and Bryceson laboratories the downregulation of these intracellular molecules is associated with pronounced changes in DNA methylation patterns that confer an overall epigenetic profile very similar to CD57^bright effector CD8⁺ T cells (Tesi et al., 2016). In particular, the promoter regions of FcRγ, EAT-2, SYK, and PLZF are highly methylated in adaptive NK cells compared to in CD56^dim cNK cells. Likewise, the promoter regions of IL-12 and IL-18 receptor subunits, which are regulated by PLZF, are also highly methylated in human adaptive NK cells, and this accounts for their near complete lack of IFN-γ production in response to IL-12 and IL-18 stimulation in vitro. Human adaptive NK cells also show reduced activation and degranulation in response to activated autologous T cells (Schlums et al., 2015). Interestingly, despite these data, the IFN-γ promoter region is actually hypomethylated in adaptive NK cells, and when stimulated through CD16 ligation adaptive NK cells produce large amounts of IFN-γ and also extensively proliferate (Hwang et al., 2012; Lee et al., 2015; Schlums et al., 2015). Cytotoxic capacity of adaptive NK cells is an ongoing question within the field, with multiple studies indicating similar or reduced CD107a degranulation as compared to cNK cells following CD16 ligation or stimulation with antibody-coated tumor targets, (Hwang et al., 2012; Lee et al., 2015; Schlums et al., 2015). It remains to be seen if in vivo co-stimulation alters adaptive NK cell functionality (Liu et al., 2016). Lastly, recent studies involving patients with either paroxysmal nocturnal hemoglobinuria or GATA2 deficiency indicate that adaptive NK cells selectively persist in these patients whereas cNK cells are reduced, indicative of unique homeostatic regulatory mechanisms at play (Corat et al., 2017; Schlums et al., 2017). Therefore, adaptive NK cells are very distinct in multiple aspects of their basic biology, implying that they fulfill distinct role(s) in immunity.

It is clear from the work of Lanier and colleagues that Ly49H⁺ adaptive NK cells in mice are protective against MCMV infection in an antigen-specific manner (Sun et al., 2009). In fact, in comparison to cNK cells, murine adaptive NK cells generated following MCMV infection show reduced bystander (i.e. cytokine-dependent) activation following infection with influenza or Listeria monocytogenes and vice versa (Min-Oo and Lanier, 2014). Some degree of antigen specificity is likely also the case in humans given the fact that adaptive NK cell populations often show pseudo-clonal or pseudo-oligoclonal expansions of distinct KIR⁺ subsets (Beziat et al., 2013; Davis et al., 2015). However, it has not yet been definitely shown that human MHC class I-binding receptor positive NK cells can respond to HCMV in an antigen-specific manner just as murine Ly49H⁺ NK cells respond to MCMV-encoded m157 expression (Hammer and Romagnani, 2017). Nonetheless, human adaptive NK cells can kill HCMV infected targets via ADCC in vitro when in the presence of patient sera containing anti-HCMV antibodies (Lee et al., 2015), and therefore they likely provide protective immunity against HCMV in vivo. What remains to be determined is whether human adaptive NK cells, at least as defined above, are specific to HCMV infection and/or can also show effector activity against non-HCMV targets. Interestingly, while NKG2C⁺CD57⁺ adaptive NK cell-like expansions have not been detected in the settings of isolated EBV, influenza, and HSV-2 infections, expansions of NKG2C⁺CD57⁺ NK cells (putative adaptive NK cells) have been reported in the settings of hepatitis B, hepatitis C, human immunodeficiency virus (HIV), hantavirus, and chikungunya virus, although these expansions occur primarily in individuals who are also HCMV seropositive (Beziat et al., 2012; Bjorkstrom et al., 2011a; Bjorkstrom et al., 2011b; Goodier et al., 2016; Guma et al., 2006; Hendricks et al., 2014; Petitdemange et al., 2011). Notably, expansions of NKG2C⁺CD57⁺ adaptive NK cells have also been observed in AML patients who reactivate with HCMV following allo-HSCT, and these adaptive NK cell expansions have been associated with significantly reduced rates of AML relapse (Cichocki et al., 2016a; Foley et al., 2012; Schlums et al., 2015). These clinical data highlight a possible and important therapeutic role for heterologous activity against allogeneic AML targets following HCMV reactivation in vivo. This is indeed an area of ongoing investigation (Nabekura and Lanier, 2014).

In light of the clinical implications of HCMV-induced adaptive NK cell expansions, Blish and colleagues recently reported that among a cohort of Kenyan women at risk for HIV-1 infection, those with high NK cell diversity indices were significantly more likely to acquire HIV-1 infection compared to those with low NK cell diversity indices (Strauss-Albee et al., 2015). While it will be important to independently confirm these data in other clinical scenarios, these data suggest that as NK cell populations diversify in response to infection such as HCMV, there may be increased risk to the host in the setting of new infectious challenges. Thus, where one aspect of NK cell-mediated immunity is gained, another may be lost.

GENERATION OF HUMAN NK CELL DIVERSITY

Having reviewed many of the core components of human NK cell diversity known to date, we now ask where this broad spectrum of human NK cell diversity comes from and how the various NK cell subsets are generated in vivo. Are all NK cell subsets derived from a common developmental pathway? While these questions will undoubtedly be investigated in the coming years, numerous recent studies support a linear model for the development of cNK cells and upon which we can begin to overlay and formulate a working comprehensive model for the generation of all human NK cell subsets characterized to date (Figure 2) (Bjorkstrom et al., 2010; Freud et al., 2016; Freud et al., 2006; Grzywacz et al., 2006; Lopez-Verges et al., 2010; Scoville et al., 2016). According to the linear model of cNK cell development (Scoville et al., 2017), BM-resident HSCs give rise to rare multipotent CD34⁺CD45RA⁺ HPCs that then leave the BM, traffic through the PB, and eventually gain entry into SLTs through their preferentially high expression of CD62L, integrin α₄β₇, and LFA-1 (Freud et al., 2005). It is likely that the same multipotent HPCs, which have T cell differentiation potential (Freud et al., 2006), can also seed other tissues including the thymus where in the latter they would be diverted by the thymic microenvironment to differentiate into T cells. Within SLTs the CD34⁺CD45RA⁺ HPCs give rise to NKDIs that progressively lose the ability to differentiate into DCs, T cells, and helper ILCs and then subsequently mature into cNK cells through a series of functionally distinct developmental stages that can be distinguished ex vivo according to the differential expression of CD34, CD117, IL-1R1, CD94/NKG2A, NKp80, CD16, and CD57 (Table 1). CD56 expression is first detected on a fraction of CD34⁺CD117⁺IL-1R1⁺CD94/NKG2A⁻NKp80⁻CD16⁻CD57⁻ stage 2b cells in SLTs, and the relative proportions of CD56⁺ cells progressively increase within each successive developmental stage (Freud et al., 2006; Scoville et al., 2016). However, CD56 has thus far not been shown to correlate with any distinct functional transition in terms of lineage differentiation potential and/or acquisition of mature NK cell functions; thus, it is not included as a stage-defining marker in this model. Nonetheless, CD56 is likely important for the developmental process as CD56 has been recently shown to mediate a required developmental synapse between NKDIs and stromal cells during in vitro NK cell development (Mace et al., 2016).

Figure 2. Comprehensive model for the development and diversification of the human NK cell lineage.

Generation of human NK cell diversity occurs through a multistep process, portrayed here in six steps. (1) BM-resident HSCs give rise to CD34⁺CD45RA⁺ HPCs that leave the marrow (2) and seed SLTs as well as additional tissue niches supporting the generation of multiple lymphoid lineages (3). Within SLTs, these CD34⁺CD45RA⁺ progenitor cells progressively lose their ability to differentiate into DCs, T cells, as well as additional non-NK helper ILC populations through multiple discrete stages of development that have been characterized extensively ex vivo (Scoville et al., 2017). Upon maturation, cNK cells leave the SLTs and accumulate within the PB to create a pool of functionally mature cNK (4). At birth, NK cell diversity is relatively low (Strauss-Albee et al., 2015). However, as an individual ages the cNK cells come in contact with various environment exposures, such as HCMV-infected or malignant cells (5), which drives increased diversity of the overall PB NK cell population and results in the differentiation and expansion of adaptive NK cells (6). The increase in diversity at step 6 is represented by a change in nuclear color from red (cNK) to blue (adaptive NK cell) by a subset of the cells portrayed at step 4.

Table 1.

Stages of human cNK cell development in SLTs

Stage	Surface Immunophenotype							Additional Features
	CD34	CD117	IL-1R1	CD94/NKG2A	NKp80	CD16	CD57
1	+	−	−	−	−	−	−	Multipotent (DC, T, ILC, NK) CD45RA+CD10+*
2a	+	+	−	−	−	−	−	Multipotent (DC, T, ILC, NK)*
2b	+	+	+	−	−	−	−	Common ILC progenitor (ILC, NK)*
3	−	+	+	−	−	−	−	Overlap with ILC3s and ILC/NK restricted precursors**
4a	−	+/lo	+/lo	+	−	−	−	ILC3-like profile (NKp44+/−IL-22+) KIR−, IFN-γ−, non-cytolytic***
4b	−	lo/−	lo/−	+	+	−	−	a.k.a. CD56bright cNK KIR+/−, IFN-γ+, weakly cytolytic ***
5	−	lo/−	lo/−	+/−	+	+	−	a.k.a. “early” CD56dim cNK KIR+/−, IFN-γ+, cytolytic****
6		−	lo/−	+/−	+	+	+	a.k.a. “late” CD56dim cNK KIR+/−, IFN-γ+, cytolytic****

Highlighted boxes denote the immunophenotypic changes that define the stage transitions.

^*(Scoville et al., 2016),

^**(Lim et al., 2017a),

^***(Freud et al., 2016),

^****(Lopez-Verges et al., 2010; Schlums et al., 2015)

According to this developmental model, stage 4b cells are CD56^brightCD16⁻CD57⁻ cNK cells; stage 5 cells are “early” PB CD56^dimCD16⁺CD57⁻ cNK cells; and stage 6 cells, which express the terminal differentiation-associated marker CD57 (Lopez-Verges et al., 2010), are “late” PB CD56^dimCD16⁺CD57⁺ cNK cells (Table 1) (Bjorkstrom et al., 2010). CD56^dim cNK cells are present in SLTs at very low frequencies; however, they have been detected in efferent lymph fluid, suggesting that after CD56^dim cNK cells are produced in SLTs they leave these tissues via the efferent lymph and return to accumulate in the PB (Romagnani et al., 2007). Indeed this model holds that CD56^bright and CD56^dim cNK cell subsets have a progenitor-progeny relationship, respectively. Although some degree of plasticity can be seen following stimulation of these subsets in vitro, most experimental evidence as well as clinical data available to date support a progenitor-progeny relationship, and this has been extensively reviewed elsewhere (Luetke-Eversloh et al., 2013; Yu et al., 2013).

Pertaining to the diversity theme of this piece, one implication of the aforementioned development model is that every NK cell progresses through a CD94/NKG2A⁺ stage 4a intermediate that lacks KIRs as well as CD94/NKG2C expression (Table 1) (Freud et al., 2016; Scoville et al., 2017). Given that many stage 5 and stage 6 CD56^dim cNK cells in PB lack CD94/NKG2A expression and instead express different KIR isoforms and variable amounts of CD94/NKG2C, this implies that as a group cNK cells start out uniform in their MHC class I-binding receptor expression profile (i.e. CD94/NKG2A⁺KIR⁻CD94/NKG2C⁻) and then at some point they diversify to yield NK cell repertoires with the remarkably high degree of heterogeneity detected in the PB of both healthy newborns and adults (Figure 1) (Strauss-Albee et al., 2017). Thus NK cell population diversity is likely a function of developmental stage (Bjorkstrom et al., 2010; Strauss-Albee et al., 2015). In our analyses of SLT NKDIs, KIRs and NKG2C can be detected by flow cytometry within the pool of stage 4b cells (CD34⁻CD117^lo/−IL-1R1^lo/−CD94/NKG2A⁺NKp80⁺CD16⁻CD57⁻) (Freud and Caligiuri, 2006; Freud et al., 2016), suggesting that it is within this developmental stage (4b) that MHC class I-binding receptor-associated NK cell population diversification may be initiated by as yet some unknown mechanism.

Another corollary of this model is that mature cNK cell subsets (i.e. CD56^bright and CD56^dim), which show distinct functional attributes and likely play distinct roles in human immunity, are nonetheless closely developmentally related. That is to say humans have likely evolved to utilize NKDIs to fulfill distinct roles in immunity at different points in their development. However, this does not necessarily mean that all CD56^bright cNK cells are destined to become CD56^dim cNK cells, as it is possible if not likely that individual cells within any of the aforementioned developmental stages may be terminally differentiated. Future studies will be needed to address this interesting possibility. In light of this notion one can also envision a scenario in which trNK cells could derive via a common developmental pathway shared by cNK cells and that at some point a subset of NKDIs diverges, undergoes terminal differentiation, and ultimately takes up permanent residence within tissues as trNK cells (Figure 2). This could occur in response to extrinsic local factors such as transforming growth factor-beta (TGF-β) that has been shown to promote trNK phenotypes in human uterus and mouse salivary gland tissues (Cortez et al., 2016; Keskin et al., 2007). Alternative possibilities are that trNK cells derive through completely distinct developmental pathways, potentially starting from distinct tissue-resident HPC populations or NK cell progenitor populations that have been described (Bozzano et al., 2015; Renoux et al., 2015). In addition, trNK cells may be produced via “tissue-imprinting” (e.g. in response to TGF-β) of cNK cells that traffic to tissues from the PB. These various hypotheses also await further investigation.

In regards to the origin(s) of human adaptive NK cells, it is likely that they can derive from PB cNK cells. In particular, CD56^dim cNK cells represent a probable candidate progenitor cell pool for adaptive NK cells, because the former are more likely to express KIRs and/or CD94/NKG2C that may in turn provide important antigen-sensing signals during infection. This is supported by recent findings from Kamimura and Lanier who showed that MCMV-associated Ly49H⁺ adaptive NK cells can be derived from KLRG1⁻Ly49H⁺ splenic cNK cells (Kamimura and Lanier, 2015). Nonetheless, there may be other pathways for the differentiation of adaptive NK cells, and there may also be distinct types of adaptive NK cells, potentially within tissues. For example, as mentioned earlier, there are liver-resident adaptive NK cells in mice that are characterized by the expression of CXCR6 and that were recently shown to be reliant on expression of the aryl hydrocarbon receptor (Cortez et al., 2016; Keskin et al., 2007; O’Leary et al., 2006; Paust et al., 2010; Zhang et al., 2016). It remains to be determined if such liver-resident adaptive NK cells ultimately rely on a developmental pathway that is distinct from that of MCMV-associated adaptive NK cells, which can derive from splenic KLRG1⁻ cNK cells. Regardless, these data suggest that distinct types of adaptive NK cells may exist in mice. In light of these findings in mice, small populations of HCMV-associated human CD49a⁺NKG2C⁺ liver-resident NK cells have been recently described (Marquardt et al., 2015; Stegmann et al., 2016). These cells are distinct from the predominant population of liver-derived CD49e⁻CD49a⁻ trNK cells (Aw Yeang et al., 2017) and raise the possibility that the former constitute a unique tissue-resident adaptive NK cell population in humans.

Recent studies have begun to elucidate the mechanism(s) of human adaptive NK cell differentiation. While memory-like NK cells with many of the same functional features of HCMV-associated adaptive NK cells can be produced in vitro or in vivo following short-term stimulation by IL-12, IL-15, and IL-18 (Cooper et al., 2009; Romee et al., 2012), human cytokine-induced memory-like NK cells do not show the characteristic downregulation of intracellular adapter signaling molecules that appear to specifically denote HCMV-associated adaptive NK cells derived in vivo (Wagner et al., 2017). Rather, signals propagated through the IL-12 receptor in combination with CD2 and an MHC class I-binding receptor are thought to collectively provide a three-prong stimulation capable of promoting the epigenetic and phenotypic modifications that occur in association with adaptive NK cell differentiation (Hammer and Romagnani, 2017).

CONCLUDING REMARKS

The plethora of recent studies cited here highlight the growing interest in and continued clinical relevance of characterizing and understanding NK cell diversity for the betterment of patients. It is daunting to realize how, despite decades of research by hundreds of teams studying NK cells, we have also only just scratched the surface of this exciting field. Indeed it is critical that the vast breadth of human NK cell diversity be taken into account when considering designing immune-based therapies for cancer and other diseases. Thus far many of the analyses of NK cell population diversity and adaptive NK cells have focused on PB NK cells. However, it will be important in future studies to determine the extent to which trNK cells are also heterogeneous at the population level; how local inflammation or other physiological processes (e.g. pregnancy) may influence NK cell diversity; and how this in turn can impact NK cell mediated function and associated physiologic and pathophysiologic processes.