Natural killer–cell differentiation by myeloid progenitors

Hematopoietic stem cells (HSCs) give rise to all blood lineages.¹  As HSCs differentiate along one lineage, they gradually lose the ability to develop into other lineages.²  Hematopoietic differentiation involves lineage commitment, defined here as the initiation of developmental program(s) that lead to a particular cell fate. The accompanying inability to differentiate into other lineages has been referred to as lineage maintenance.³  Lineage commitment and lineage maintenance are complementary processes that guide cell fate decisions. Thus, cells committed to a particular lineage have alternative developmental choices until lineage maintenance is complete.⁴ 

Hematopoietic differentiation has been schematically depicted as a “tree of hematopoiesis,”¹  outlining the possible developmental choices. According to this prevailing schema, the decision between lymphoid and myeloid lineages occurs very early. However, alternative views have been proposed, including the existence of a common myelo-lymphoid progenitor.^5,6  Elucidation of hematopoietic developmental pathways and extrinsic stimuli that influence them is instrumental to understanding both normal and malignant hematopoiesis. In particular, factors that favor natural killer (NK)–cell development could be used to exploit their activity against malignancies.

NK cells are innate immune effector cells. Their derivation from either lymphoid or myeloid lineages was debated early in their discovery.⁷  Further research showed that NK cells can be derived from common lymphoid progenitors (CLPs) and hence have been considered separate from myeloid lineage.^8,9  However, some studies question this and have shown that progenitors expressing myeloid antigens can develop into NK cells.^10,11  NK-cell differentiation from hematopoietic progenitor cells (HPCs) can be studied in vitro.^12,13  This process depends on cytokines, notably interleukin-2 (IL-2) or IL-15, whereas other factors (stem cell factor [SCF], fms-like tyrosine kinase-3 ligand [FLT-3L]) induce early HPC expansion and responsiveness to IL-2 and IL-15 signaling.¹⁴ 

CD34⁺ HPCs are heterogeneous and include cells at various levels of differentiation. Multipotent HPCs with long-term repopulation potential are contained within the CD34⁺CD38⁻ subset.¹⁵  More advanced lineage precursors, included in the CD34⁺CD38⁺ fraction,¹⁵  have been categorized as common myeloid progenitor (CMP; CD34⁺CD38⁺CD123⁺CD45RA⁻), granulocytic-monocytic precursor (GMP; (CD34⁺CD38⁺CD123⁺CD45RA⁺), and megakaryocyte-erythroid precursor (CD34⁺CD38⁺CD123⁻CD45RA⁻).¹⁶  Subsets of CD34⁺ precursors have also been distinguished by their ability to readily differentiate into NK cells. Surface receptors that define NK precursors include CD7,¹³  CD122,¹⁷  CD161,¹⁸  integrin β7, and CD45RA^high.¹⁹ 

Stromal cell layers have been used to differentiate HPCs into NK cells. Sources of stroma include bone marrow,^13,20  murine fetal liver cell lines,²¹  and human splenic fibroblasts.²²  Stroma increase the efficiency of NK-cell generation and advance the maturational status of HPC-derived NK cells.²¹  Physiologic concentrations of hydrocortisone (HDC) also advances NK-cell development from CD34⁺ HPCs.¹⁰ 

We previously described discrete stages of human NK-cell differentiation by studying CD34⁺ cells cultured on the murine fetal liver stromal cell line, EL08.1D2.²³  Notably, the stages we defined in vitro closely resemble those in human lymph nodes.²⁴  This culture system results in a strikingly high efficiency of NK-cell differentiation from CD34⁺ HPCs.²³  Here, we investigate the mechanism of stroma-induced NK-cell differentiation to better understand the origins of NK cells. We demonstrate that fetal liver stroma and HDC enhance NK differentiation by recruiting CMP and GMP to the NK lineage. Likewise, precursors at more advanced stages of myeloid differentiation retain NK cell–generating capacity. The concept that NK cells can be derived from the myeloid lineage advances our understanding of NK ontogeny and the relationship to other blood lineages, creating the potential for therapeutic applications.

CD34⁺ cells were positively selected from umbilical cord blood (UCB) mononuclear cells (Miltenyi Biotec). Selected cells were > 97% pure and lacked CD56⁺ cells (not shown). When specified, CD34⁺ cells underwent fluorescence-activated cell sorting (FACS) into CD34⁺NKlin⁺, CD34⁺CD38⁻NKlin⁻, and CD34⁺CD38⁺NKlin⁻ (NK lineage markers were CD161, CD122, CD7, integrin β7; and CD45RA^high; cells negative for all were considered NKlin⁻). Granulo-monocytic precursors (CD34⁺CD38⁺CD64⁺),²⁵  CMPs (CD34⁺CD38⁺CD123⁺CD45RA⁻), and GMPs (CD34⁺CD38⁺CD123⁺CD45RA⁺) were FACS purified from UCB CD34⁺ cells.¹⁶  After the initial sort, the CMP and GMP fractions were resorted and deposited directly into culture wells at 1 or 10 cells per well.

The embryonic liver cell line EL08.1D2 was cultured to confluence and was irradiated (30 Gy [3000 rad]).²³  CD34⁺ cells or specified subsets were cultured with/without a monolayer of EL08.1D2 cells in Ham F12 plus Dulbecco modified Eagle medium (1:2 ratio) with 20% human AB sera, ethanolamine (50μM), ascorbic acid (20 mg/L), 5 μg/L sodium selenite (Na₂SeO₃), β-mercaptoethanol (24μM), and penicillin (100 U/mL)/streptomycin (100 U/mL). At the start of cultures, IL-3 (5 ng/mL), IL-7 (20 ng/mL), IL-15 (10 ng/mL), SCF (20 ng/mL), and FLT-3L (10 ng/mL) were added. Where specified, HDC (10⁻⁶M; Stem Cell Technologies), macrophage colony-stimulating factor (M-CSF), or Dickkopf-1 (R&D Systems) were added. Cultures were refed weekly by 50% volume change of media, supplemented with the above-listed cytokines (except IL-3), with or without HDC. After 14, 21, and 28 days of culture, 3 replicate wells were harvested, counted, and analyzed.

CD34⁺ HPC or specified subsets were plated at 200, 100, 50, and 25 cells/well, 24 replicates/dilution. Conditions included media with or without stroma (EL08.1D2) and with or without HDC (10⁻⁶M). After 4 weeks cellular outgrowth was scored, and split-well analysis was performed in which one-half of each well was tested for K562 cytotoxicity. If cytotoxicity exceeded a predetermined cutoff value (mean spontaneous release + 3× SD), wells were considered positive. The fraction of negative wells at each dilution was used to calculate the NK-cell precursor (NKp) frequency as determined by Poisson kinetics.²⁶ 

CMP and GMP (n = 5) were double-sorted and deposited at 1 and 10 cells per well (n = 60 replicates per condition). Conditions included (1) cytokines, (2) cytokines and stroma, (3) cytokines and stroma and HDC, and (4) methylcellulose enriched with growth factors (MethoCult; StemCell Technologies). The latter condition was included to control for their colony-generating potential (CMP or GMP). Cultures were maintained for 17-21 days then tested for expression of NK-cell markers (CD56, CD94) in combination with myeloid markers (CD15, CD14, CD33). The colony-forming potential of sorted fractions was assessed after 14-21 days of methylocellulose culture by 2 researchers, including an independent (blinded) investigator not otherwise involved in the study.

CD34⁺ HPCs were plated at 1 cell/well in methylocellulose enriched with growth factors (MethoCult GF+H4435; StemCell Technologies). After 20-21 days individual granulocyte-macrophage colony-forming units (CFU-GMs) were isolated and plated into culture in medium as for NK differentiation (IL-15, IL-7, SCF, FLT-3L) with stromal cells and HDC (10⁻⁶M).

NK-differentiation cultures were harvested after 14 days and depleted of CD56⁺ cells with the use of using magnetically activated cell-sorting beads. This CD56⁻ population underwent FASC into CD56⁻CD33^+/−CD13^low/−, CD56⁻CD33⁺CD13^intermediate, and CD56⁻CD33⁺CD13^high fractions. Sorted populations were cultured (in triplicates) at 10 000 cells/well in 96-well plates in NK-differentiation medium with/without stroma and with/without HDC. After 18-21 days cells were analyzed. Similarly, cultures started with CD34⁺ HPCs were harvested at day 18, CD56⁺ cells were depleted, and CD117⁺ cells were isolated from CD56⁻ fraction with the use of magnetically activated cell-sorting beads. The resulting populations were then sorted into CD56⁻CD94⁻CD117⁺M-CSFR⁺ (myeloid precursors) and CD56⁻CD94⁻CD117⁺M-CSFR⁻ (nonmyeloid, presumed to include lymphoid precursors). Sorted fractions were cultured as described earlier. In selected experiments M-CSF was added.

The following antibodies were used: CD7-FITC (fluorescein isothiocyanate), CD13-PE (phycoerythrin), CD1aFITC, CD14-FITC or PE, CD16 (FITC, PE, or peridinin chlorophyll protein and cyanine 5.5 [PerCp-Cy5.5]), CD33 PerCp-Cy5.5, CD34 (PerCp, PE, and allophycocyanin [APC]), CD38-APC, CD56-APC, CD94-FITC, CD117 (PE or PerCp-Cy5.5), CD122-PE, CD161 (FITC or PE), CD158a-PE, CD158b-PE, CD158e-PE (DX-9), CD64-PE, Granzyme-B–FITC, Perforin-FITC, CD68 (macrosialin)–PE, and myeloperoxidase-FITC. For CMP, GMP, and megakaryocyte-erythroid precursor subsets included CD123-PE, CD34–PerCp-Cy5.5, CD45RA-Horizon450 (all from BD Biosciences), CD38-APC, and Lineage cocktail, including FITC-conjugated CD2, CD3, CD4, CD7, CD8, CD10, CD14, CD16, CD19, CD11b, CD20, CD56, GPA (e-Bioscience). Additional antibodies included CD115-PE, CD45RA-FITC (Biolegend), CD13-FITC (Serotec), lysozyme-FITC (Caltag), NKp30-PE, NKp44-PE, and NKp46-PE (Beckman Coulter). Intracellular staining was performed with cytofix/cytoperm (BD Biosciences). FACS was performed on a FACSCalibur. Data were analyzed with WinMDI (FACS Core Facility, The Scripps Research Institute) or FlowJo (TreeStar Inc). Sorting was performed on FACSAria, and postsort reanalysis usually showed > 99% purity (not shown).

K562 or B721.221 lymphoblastoid cell line targets were used in killing assays.²³ 

For grouped analysis, experimental values were normalized by log transformation before 1-way (or 2-way) analysis of variance, as indicated. Individual groups were compared with Bonferroni posttest. The χ² analysis was used to compare fractions of methylcellulose (CFU-GM) colonies that gave rise to NK cells.

We investigated how stroma, HDC, and the combination of the two affect NK differentiation. CD34⁺ cells were cultured in media containing cytokines (defined here as IL-3 for the first week, SCF, FLT-3L, IL-7, and IL-15), and, for comparison, cells were cultured with (1) cytokines and stroma, (2) cytokines and HDC, or (3) the combination of all three. The addition of stroma or HDC or both significantly increased cell numbers and the percentage of NK cells (CD56⁺ cells) at 2 (Figure 1A,D), 3 (Figure 1B,E), and 4 (Figure 1C,F) weeks (P < .05). The 2 factors (stroma and HDC) were additive because the combination was significantly better than either one alone (P < .05 at 4 weeks).

With the addition of either HDC or stroma, all wells contained NK cells (not shown). However, when CD34⁺ HPCs were cultured without stroma or HDC, we observed considerable variability between replicates. For instance, there were individual wells in which NK cells developed, whereas in others they did not. This suggested that the CD34⁺ HPC fraction contained precursors capable of NK differentiation with cytokines alone (ie, without HDC or stroma). Given the well-to-well variability with initial input of 500 CD34⁺ cells/well, such NK precursors appear to be rare. At least 2 nonexclusive mechanisms can explain these findings. First, HDC and stroma might increase the efficiency of NK-cell generation from progenitors already “committed” to the NK lineage. Second, they might act on progenitors that might not otherwise become NK cells. One method to verify if the latter possibility was operational is to measure the NKp frequency with the use of a limiting dilution assay. If stroma or HDC or both increase the frequency of CD34⁺ HPCs giving rise to NK cells in limiting dilution assays, then the logical explanation would be that these factors act on CD34⁺ HPCs to recruit them to the NK lineage.

To measure the NKp frequency serial, 2-fold dilutions of CD34⁺ cells (25-200 cells/well, 24 replicates/dilution) were plated as in Figure 1 (ie, cytokines alone, cytokines and stroma, cytokines and HDC, and cytokines and HDC and stroma). Because the hallmark NK characteristic is unprimed effector function (eg, K562 cytotoxicity), we defined an NKp as a CD34⁺ HPC that could give rise to effectors with K562 cytotoxicity at 4 weeks of culture. Wells showing no cytotoxicity were assumed to contain no NKps (ie, precursors capable of NK-cell development in the given condition). The fractions of wells showing no killing (Figure 2A) were used to calculate the NKp frequency (Figure 2B).²⁶  Compared with CD34⁺ HPCs cultured in cytokines alone, the same cell suspension cultured with stroma, HDC, or the combination of both showed a higher NKp frequency (P < .001; n = 8; Figure 2B).

Before cytotoxicity testing, wells were microscopically scored for outgrowth. Similar to the NKp results, cell density was least when CD34⁺ HPCs were cultured with only cytokines (Figure 2C). Combining the cytotoxicity with the outgrowth data, we observed wells with the following patterns: (1) no (or few) cells and no cytotoxicity, (2) cells present but no cytotoxicity, and (3) cells present and cytotoxicity present. Representative wells were analyzed by FACS. Cytotoxicity was uniformly associated with the presence of NK cells at a functional stage of development (ie, CD56⁺CD94⁺CD117^low/−).^23,27  Frequently, these cells were intermixed with cells at earlier stages of development (CD56⁺CD94⁻CD117^high; not shown). Conversely, wells with no cytotoxicity lacked NK cells and instead contained cells of myeloid origin by forward scatter (FSC) and side scatter (SSC) characteristics (not shown).

Collectively, these studies show that a small but measurable fraction of CD34⁺ HPCs give rise to NK cells under the influence of only cytokines. Conversely, the majority of HPCs cultured with the above cytokine combination do not generate functional NK cells. Thus, the increased NKp frequency in the presence of stroma, HDC, or both, strongly suggest that these agents induce NK differentiation by acting on HPCs (and their progeny), which do not generate NK cells when cultured with cytokines only.

Our aim was to isolate the subset of CD34⁺ HPCs that required stroma or HDC or both for NK differentiation. Several surface receptors are associated with NK lineage commitment, including CD7, CD122, CD161, integrin β7, and CD45RA^high.^13,17-19  These were expressed on nonfully overlapping CD34⁺ subsets (not shown). We hypothesized that CD34⁺ HPCs lacking all of these NK lineage markers (CD34⁺NKlin⁻) might be “uncommitted” to the NK lineage (ie, unable to give rise to NK cells with cytokines alone). Two NKlin⁻ subsets (CD34⁺CD38⁻NKlin⁻, CD34⁺CD38⁺NKlin⁻) were tested in the NKp frequency assay with and without stroma and HDC. CD34⁺ cells expressing any of the above NK lineage markers were used for comparison (CD34⁺NKlin⁺). In the absence of stroma or HDC or a combination, both CD34⁺NKlin⁻ populations (CD38⁻ and CD38⁺) yielded few wells with K562 cytotoxicity, resulting in a flat slope of the semi-log–limiting dilution plot (Figure 3A). In contrast, the CD34⁺NKlin⁺ population cultured in only cytokines produced a noticeable fraction of wells with K562 cytotoxicity, resulting in a steeper slope (Figure 3A).

For the NKlin⁻ populations (CD38⁻ and CD38⁺) the NKp frequency was not significantly different from 0 (CD34⁺CD38⁻NKlin⁻, 1102/10⁶ cells; 95% confidence interval [CI], −598 to 2728; CD34⁺CD38⁺NKlin⁻, 474/10⁶ cells, 95% CI, −637 to 2153). In contrast, CD34⁺NKlin⁺ cells showed a higher NKp frequency (median, 7370/10⁶ cells; 95% CI, 3891-13 178; n = 4; P < .01) (Figure 3B). The addition of HDC or stroma or both induced the capacity of CD34⁺NKlin⁻ populations (both CD38⁻ and CD38⁺) to give rise to functional NK cells (Figure 3D,F), changing the slopes of the semi-log plots (Figure 3C,E). Thus, we defined a subset of HPCs critically dependent on HDC or stroma or both to differentiate into NK cells (ie, CD34⁺NKlin⁻).

To understand the mechanisms of stroma and HDC here, CD34⁺NKlin⁻ cells were cultured in cytokines or in cytokines, stroma, and HDC. Transcriptional profiling studies showed up-regulation of the WNT signaling pathway in response to stroma and HDC (Figure 3G). To verify whether WNT signaling plays a role in the effect of stroma and HDC, we added Dickkopf-1, a canonical WNT pathway inhibitor, resulting in a dose-dependent decrease in the number of NK cells generated (Figure 3H; P = .012).

Although the above NKp frequency assay has a functional endpoint (ie, cytotoxicity), it offers no insight into the developmental events during the 4-week culture. Prior studies show that HDC increases NK expansion,²⁸  perhaps through proliferation or antiapoptotic activity. Thus, HDC or stroma or both might simply induce proliferation or maintain survival of immature NK cells. To address this, cultures were evaluated at days 14 and 21. CD34⁺NKlin⁺ cells cultured in only cytokines (ie, no stroma or HDC) generated first CD56⁺ NK cells within 14 days, whereas CD34⁺NKlin⁻ cells did not (supplemental Figure 1, available on the Blood Web site; see the Supplemental Materials link at the top of the online article). In an alternative approach, small numbers of CD34⁺NKlin⁻ cells (100 cells/well, 12 replicates) were tested at days 14 and 21. In the presence of cytokines alone, CD56⁺ cells were not observed. In contrast, HDC or stroma or both induced CD56⁺ development after 14 and 21 days (Figure 4A and B, respectively). Collectively these results show that CD34⁺NKlin⁻ cells did not develop into CD56⁺ cells throughout the culture with cytokines alone, excluding the possibility that NK cells merely failed to expand or survive or both in the absence of HDC or stroma or both.

CD34⁺NKlin⁻ cells cultured with only cytokines showed growth after 14 and 21 days, but they did not differentiate into NK cells (Figure 4C; supplemental Figure 1). The majority of these cells expressed markers associated with the myeloid lineage (CD13 and CD33) (Figure 4C). When the same CD34⁺NKlin⁻ cells were cultured with HDC or stroma or both, some CD56⁺ NK cells coexpressed CD13 and CD33 (Figure 4C; supplemental Figure 1). Thus, CD56⁺ cells might arise from CD13⁺CD33⁺ precursors, prompting us to speculate that stroma and HDC could recruit myeloid progenitors to differentiate into NK cells.

CD13⁺CD33⁺ cells in these cultures were tested for characteristics unique to the myeloid lineage. CD13⁺ cells expressed intracellular enzymes (myeloperoxidase, lysozyme, macrosialin [CD68]) and surface receptors (CD14 and CD1a), consistent with myeloid maturation (supplemental Figure 2). We next tested whether precursors with various amounts of CD13 could give rise to NK cells. Not only the CD13^low/− but also the CD13^int and CD13^high subsets differentiated into NK cells after culture with cytokines and stroma (Figure 5A). The efficiency of NK-cell generation decreased with increasing CD13 expression (P < .05) (Figure 5B). Thus, cells following myeloid differentiation as indicated by increasing CD13 staining intensity²⁹  gradually lose the ability to give rise to NK cells.

To further test whether myeloid cells could give rise to NK cells, CD34⁺ cells were plated in methylcellulose. Individual CFU-GM colonies were isolated and cultured in NK-supporting conditions. A fraction of CFU-GM colonies gave rise to NK cells (CD56⁺ lymphocytes by FACS) after culture with cytokines, stroma, and HDC but not with cytokines alone (9 of 54 vs 0 of 36; P = .01; supplemental Figure 3).

Prior studies have identified human hematopoietic progenitors with the capacity for myeloid and erythroid differentiation (CMPs, CD34⁺CD38⁺IL-3R^lowCD45RA⁻), as well as descendants that are restricted to the myeloid lineage (GMP, CD34⁺CD38⁺IL-3R^lowCD45RA⁺).¹⁶  We examined whether freshly isolated CMPs and GMPs could develop into NK cells. Because purity was critical, cells were double-sorted directly into 96-well plates. As a control, CMPs and GMPs were also deposited into plates containing methylcellulose. As previously described,¹⁶  single CMPs gave rise to myeloid, erythroid, or mixed (granulocyte, erythroid, megakaryocyte, and macrophage) colonies (cloning efficiency, 61.7%-91.7%; average, 76.7%), whereas GMPs yielded exclusively myeloid colonies (cloning efficiency, 43.3%-82.3%; average, 62.7%) (Table 1), and none of the wells showed 2 separate colonies. GMPs plated at 10 cells/well gave virtually no erythroid or mixed colonies (1 mixed colony in 300 wells; n = 5 donors). When cultured in NK-supporting cytokines (ie, IL-3, SCF, FLT-3L, IL-7, and IL-15), CMPs and GMPs very rarely yielded CD56⁺ cells (0.7% of wells; Table 1). These CD56⁺ cells were at an immature stage of NK differentiation (CD56⁺CD117^highCD94⁻)^19,23  (not shown). When the same populations were cultured with cytokines and stroma with/without HDC, a significant fraction generated CD56⁺ cells (Table 1). These CD56⁺ cells fell in the lymphocytic gate (by FSC/SSC) and coexpressed CD94, corresponding to mature, functional NK cells.^19,23  Myeloid cells (CD15⁺) coexisted with NK cells in some cultures (Figure 5C), showing that a single CMP or GMP could produce both myeloid cells and NK cells. As in the above-described experiments, single CFU-GM colonies (derived from the control-sorted CMPs and GMPs cultured in methylocellulose) could give rise to NK cells if isolated and further cultured in cytokines and stroma and HDC (2 of 30 colonies; not shown). These results show that on a single-cell level, myeloid progenitors (CMPs and GMPs) can produce NK cells.

To reaffirm this conclusion, granulo-monocytic precursors were purified by FACS according to an alternative definition.²⁵  CD34⁺CD38⁺CD64⁺ myeloid precursors also differentiated into NK cells less than the influence of cytokines and stroma with a frequency similar to the remaining CD34⁺CD38⁺NKlin⁻ precursors (supplemental Figure 4).

We next isolated myeloid progenitors on the basis of M-CSF receptor expression. The M-CSF receptor (M-CSFR, CSF-1R, CD115) is expressed by granulo-monocytic precursors³⁰  and renders hematopoietic cells responsive to M-CSF, a growth factor promoting monocytic differentiation.³¹  Therefore, M-CSFR is not merely a phenotypic marker, but it is also functionally linked to the monocytic lineage. We previously determined that CD34⁺ HPCs cultured on stroma continue to generate NK cells when depleted of CD56⁺ cells after 2-3 weeks of culture. The cells responsible for this are contained in the CD56⁻CD117⁺ subset²³  (not shown). These cells variably express M-CSFR (Figure 6A). Therefore, CD56⁻CD117⁺M-CSFR⁺ and CD56⁻CD117⁺M-CSFR⁻ cells were isolated from NK differentiation cultures to test their potential to generate NK cells. Both populations gave rise to CD56⁺ cells on further culture (Figure 6A). To examine their maturation status, cells were tested for CD94 expression.^23,24  CD56⁻CD117⁺M-CSFR⁻ precursors cultured in only cytokines readily differentiated into functional NK cells (CD56⁺CD94⁺). In contrast, CD56⁻CD117⁺M-CSFR⁺ precursors cultured in cytokines alone gave rise to CD56⁺ cells lacking CD94 (Figure 6A). The addition of HDC or stroma or both to CD56⁻CD117⁺M-CSFR⁺ myeloid precursors increased the CD94 and killer immunoglobulin–like receptor (KIR)–expressing cells (Figure 6B). Thus, CD56⁻CD117⁺M-CSFR⁺ myeloid precursors require HDC or stroma or both to advance to a functional stage of NK differentiation, whereas the M-CSFR⁻ counterparts do not.

To investigate whether myeloid and NK differentiation represent 2 alternative developmental pathways for CD56⁻CD117⁺M-CSFR⁺ precursors, M-CSF was added to NK developmental cultures. M-CSF increased the generation of CD14⁺ (CD56⁻) monocytes and decreased NK-cell development in a dose-dependent manner (P < .001) (Figure 6C and D, respectively). High concentrations of M-CSF (≥ 50 ng/mL) abolished NK-cell development from M-CSFR⁺ precursors, despite conditions optimal for NK-cell development (cytokines, stromal cells, and HDC) (Figure 6D). Thus, development of NK cells (CD56⁺) or monocytes (CD14⁺CD56⁻) from M-CSFR–expressing myeloid precursors are alternative outcomes, influenced by M-CSF. Interestingly at lower concentrations (1-20 ng/mL), M-CSF induced low-level CD14 coexpression on a fraction of developing CD56⁺ lymphocytes (Figure 6E). This provides complementary evidence that the cells that give rise to NK cells also respond to M-CSF. Collectively, these studies show that myeloid progenitors can generate NK cells, strongly supporting a myeloid NK-cell differentiation pathway.

We tested whether myeloid precursor–derived NK cells differ from the primarily lymphoid (M-CSFR⁻) fraction. Both CD56⁻CD117⁺M-CSFR⁺ and CD56⁻CD117⁺M-CSFR⁻ subsets were cultured in conditions optimal for NK-cell generation (cytokines, stroma and HDC). A higher percentage of NK cells derived from CD56⁻CD117⁺M-CSFR⁺ progenitors expressed KIR compared with NK cells derived from CD56⁻CD117⁺M-CSFR⁻ precursors (Figure 7A-B; P < .0001). Interestingly, NK cells derived from CD56⁻CD117⁺M-CSFR⁺ progenitors also contained a fraction of NK cells that were NKG2A⁻KIR⁺, whereas this fraction was absent in the CD56⁻CD117⁺M-CSFR⁻ precursor–derived NK cells (Figure 7C). Both populations showed K562 killing; however, M-CSFR⁺ precursors consistently showed marginally higher cytotoxicity (Figure 7D). Human leukocyte antigen–deficient Epstein-Barr virus–transformed B-cell targets (721.221) showed a more pronounced difference. NK cells derived from M-CSFR⁻ precursors showed minimal cytotoxicity, whereas M-CSFR⁺–derived cells readily killed these targets (Figure 7E). There were no consistent differences in perforin and granzyme B protein or in the expression of activating receptors (2B4, NKp30, NKp46, and NKG2D) that would sufficiently explain these results (not shown). Thus, NK cells developing from distinct precursors have diverse phenotypic and functional properties.

Here, we show that a small fraction of CD34⁺ HPCs expressing any of the lineage markers previously associated with NK-cell development (ie, CD122, CD161, integrin β7, and CD45RA^high and CD7)^13,17-19  can develop into NK cells under the influence of cytokines (IL-15, IL-7, SCF, FLT-3L, and IL-3) and do not require HDC or stroma. However, most HPCs lack these markers (ie, NKlin⁻) and are unable to differentiate into NK cells with these cytokines. Rather, NKlin⁻ cells follow myeloid differentiation, as evidenced by surface markers (CD13, CD33, CD14, and CD1a) and intracellular enzymes (lysozyme, myeloperoxidase, and macrosialyn [CD68]). These myeloid precursors could be recruited to the NK lineage by HDC and stroma. Perhaps not surprisingly, they gradually lost the ability to develop into NK cells as they progressed along myeloid differentiation (indicated by increasing CD13 staining). We also show that freshly isolated single CMPs or GMPs or both gave rise to NK cells in the presence of stroma and cytokines but not cytokines alone. The observations that (1) single CMPs and GMPs could generate NK cells, (2) myeloid cells coexisted with NK cells in cultures derived from a single CMP or GMP, and (3) some granulocyte-macrophage colonies from these precursors (grown in methylocellulose) could subsequently differentiate into NK cells provide solid evidence that myeloid precursors can give rise to NK cells. We also show that stroma and HDC mediate these properties, in part, through WNT activation. This is in agreement with the known role of T-cell factor/lymphoid enhancer factor (TCF/LEF) transcription factors in lymphocyte development and the ability of fetal liver stromal cells to trigger this pathway.³² 

We also isolated myeloid precursors (CD56⁻CD117⁺M-CSFR⁺) arising in vitro and demonstrated their potential to generate NK cells. The observation that low concentrations of M-CSF induced CD14 expression by NK cells developing from myeloid precursors further supports the assertion that cells responding to M-CSF give rise to NK cells. These CD56⁺CD14⁺ cells appear to be NK cells (and not monocytes) because they fell in the lymphoid gate (by SSC vs FSC) and expressed other NK antigens, including CD94 and KIR (not shown). Higher doses of M-CSF, however, enforced monocytic differentiation, prohibiting NK-cell development, suggesting a lineage choice between myeloid and NK-cell fates. These choices are influenced by the environment (stroma) and growth factors (cytokines).

Our results also indicate that myeloid precursor–derived NK cells have distinct properties. For instance, M-CSFR⁺ precursors, but not M-CSFR⁻ counterparts, require stroma or HDC for CD94 acquisition, a marker of maturation. Likewise, when these 2 starting populations are cultured with cytokines, stroma, and HDC. the resulting NK cells differ. Myeloid-derived NK cells show a significantly higher KIR expression, a KIR⁺NKG2A⁻ cell fraction and higher cytotoxicity. KIR acquisition has been used as a measure of maturity,²⁷  but the factors that lead to KIR expression are largely unknown. We recently found that myc binds to KIR promoters, driving expression.³³  Interestingly, myc expression is linked to M-CSFR signaling and myeloid development.³⁴  Recently, a thymic pathway of NK-cell development was shown in rodents. Compared with the majority of NK cells, thymic-derived NK cells were uniquely dependent on IL-7 and GATA-3 for development. The thymic-derived NK cells are characterized by low cytotoxicity but high cytokine secretion potential.³⁵  These findings, along with the data presented here, support the concept that various developmental trajectories, governed by specific transcription factors, influence the phenotypic and functional characteristics of the resulting population.

According to the widely accepted model of hematopoiesis, a multipotent HSC gives rise to CMPs or CLPs, committing the first choice in hematopoiesis between myeloid/erythroid and lymphoid fates. NK cells are believed to be exclusively derived from CLPs, which are, by definition, devoid of myeloid potential.⁸  In disagreement with this, we find that NK cells can be derived from freshly isolated CMPs, GMPs, and CD34⁺CD64⁺ granulo-monocytic precursors²⁵  as well as CD33⁺CD13⁺, and M-CSFR⁺ progeny of cultured CD34⁺ HPCs. Whether this is unique to NK cells or extends to other lymphocyte subsets requires further investigation. However, neither T- nor B-cell differentiation occurred in our cultures (not shown). We hypothesize that CMP and GMP progenitors could give rise to other lymphocyte populations less than appropriate, permissive conditions (DLL-1–expressing stroma and optimal cytokines for T-cell development). In line with these observations, alternative models for lymphoid ontogeny have been proposed. Through the identification of common lympho-myeloid progenitors^5,6  and thymic T-cell progenitors that retain myeloid potential, it has been concluded that alternative schemes of hematopoietic lineage commitment should be considered.^36,37  An important caveat is that the model of hematopoiesis is largely based on in vivo transplantation studies into irradiated mice.³⁸  The in vivo behavior of distinct progenitors is determined, in part, by the environment after transplantation, which is influenced by their trafficking. Although there are clear strengths of the transplantation assay, it does not always show the intrinsic abilities observed in vitro.³⁹ 

We show that myeloid precursors can differentiate into NK cells and that less than certain circumstances NK cells could be derived from seemingly restricted myeloid progenitors. Whether this occurs in vivo and to what extent it contributes to human physiology needs to be addressed in future studies. Given the absence of NK cells in common γ-chain deficiency,⁴⁰  myeloid precursors that develop into NK cells would conceivably acquire (and depend on) this cytokine receptor for NK-cell development.

In humans, 2 NK subsets can be distinguished, CD56^bright and CD56^dim.⁴¹  The CD56^bright subset has low cytotoxicity and readily produces cytokines in response to IL-12 and IL-18.⁴²  Only a small percentage of CD56^bright cells show expression of KIR and CD16, but they do express c-kit receptor (CD117). In comparison, the CD56^dim subset has higher cytotoxicity, abundant CD16 and KIR expression, and lack CD117. The developmental relationship of the 2 subsets has not been unambiguously resolved. The majority of NK cells derived in vitro from CD34⁺ HPCs in our^21,23  as well as in other^12,20,22,43  systems resemble CD56^bright NK cells. It is thus noteworthy that NK cells derived from myeloid precursors are distinguished by higher cytotoxicity and KIR expression, as well a fraction of KIR-expressing cells that lack NKG2A and lower levels of CD117 (not shown), features of the CD56^dim subset.

A myeloid origin of NK cells was formerly considered and debated.^7,44  However, the unequivocal demonstration of a common T/NK precursor resulted in the abandonment of this concept.⁴⁵  More recently, it has been shown that dendritic cells (DCs) and NK cells share a common developmental progenitor,¹¹  whereas others have described a rare CD14⁺ cell in UCB (but not adult blood) that gives rise to NK cells less than the influence of HDC, FLT-3L, and IL-15.¹⁰  These results are in line with the findings present in this study. Here we posit that, similar to DCs,⁴⁶  NK cells can have either a lymphoid or a myeloid origin. The developmental relationship between NK cells and DCs may be closer than previously recognized in that both can be derived from common precursors. Whether a stable cell type exists which constitutively shares the function of NK cells and DCs^47,48  is a matter of debate.^49,50  Nevertheless, in certain conditions DCs can acquire cytotoxicity, characteristic for NK cells.⁵¹  Conversely NK cells can acquire antigen-presenting ability.⁵²  The notion that myeloid precursors previously known to give rise to monocyte/macrophage and DCs⁵³  are also capable of NK-cell differentiation puts the recent findings in a new perspective.⁵⁴  In summary we present evidence that NK cells can be derived from myeloid precursors.

We thank Linda Kluge and BioE for providing some of the UCB units used in these studies.

This work was supported by the American Cancer Society (M.R.V.), Children's Cancer Research Fund (B.G. and M.R.V.), Leukemia Research Foundation (M.R.V.), and the National Institutes of Health (grants P01 CA65493 [J.S.M.], P01 111412 [M.R.V. and J.S.M.], R01 HL55417 [J.S.M.], R01 CA72669 [B.R.B.], and PO1067493 [J.S.M. and B.R.B.]).

National Institutes of Health

Contribution: B.G. designed, performed, and analyzed experiments and wrote the paper; Nandini Kataria and Niketa Kataria performed and analyzed experiments; B.R.B. and J.S.M. analyzed data and contributed to writing the paper; and M.R.V. designed and analyzed experiments and wrote the paper.

Conflict-of-interest disclosure: The authors declare no competing financial interests.

Correspondence: Michael R. Verneris, Department of Pediatrics, Division of Blood and Marrow Transplantation, University of Minnesota, 425 E River Rd, Ste 660, Minneapolis, MN 55455; e-mail: verneris@umn.edu.