Molecular signatures and transcriptional regulatory networks of human immature decidual NK and mature peripheral NK cells

Authors

  • Fuyan Wang,

    1. Institute of Immunology, School of Life Sciences and Hefei National Laboratory for Physical Sciences at the Microscale, University of Science and Technology of China, Hefei, China
    Search for more papers by this author
  • Yonggang Zhou,

    1. Institute of Immunology, School of Life Sciences and Hefei National Laboratory for Physical Sciences at the Microscale, University of Science and Technology of China, Hefei, China
    Search for more papers by this author
  • Binqing Fu,

    1. Institute of Immunology, School of Life Sciences and Hefei National Laboratory for Physical Sciences at the Microscale, University of Science and Technology of China, Hefei, China
    Search for more papers by this author
  • Yang Wu,

    1. Institute of Immunology, School of Life Sciences and Hefei National Laboratory for Physical Sciences at the Microscale, University of Science and Technology of China, Hefei, China
    Search for more papers by this author
  • Ruya Zhang,

    1. Institute of Immunology, School of Life Sciences and Hefei National Laboratory for Physical Sciences at the Microscale, University of Science and Technology of China, Hefei, China
    Search for more papers by this author
  • Rui Sun,

    1. Institute of Immunology, School of Life Sciences and Hefei National Laboratory for Physical Sciences at the Microscale, University of Science and Technology of China, Hefei, China
    Search for more papers by this author
  • Zhigang Tian,

    Corresponding author
    1. Institute of Immunology, School of Life Sciences and Hefei National Laboratory for Physical Sciences at the Microscale, University of Science and Technology of China, Hefei, China
    • Full correspondence Prof. Haiming Wei, Institute of Immunology, School of Life Sciences and Hefei National Laboratory for Physical Sciences at the Microscale, University of Science and Technology of China, Hefei, 230027, China

      Fax: +86-551-6360-6783

      e-mail: ustcwhm@ustc.edu.cn

      Additional correspondence Dr. Zhigang Tian, Institute of Immunology, School of Life Sciences and Hefei National Laboratory for Physical Sciences at the Microscale, University of Science and Technology of China, Hefei, 230027, China

      Fax: +86-551-6360-6783

      e-mail: tzg@ustc.edu.cn

    Search for more papers by this author
  • Haiming Wei

    Corresponding author
    1. Institute of Immunology, School of Life Sciences and Hefei National Laboratory for Physical Sciences at the Microscale, University of Science and Technology of China, Hefei, China
    • Full correspondence Prof. Haiming Wei, Institute of Immunology, School of Life Sciences and Hefei National Laboratory for Physical Sciences at the Microscale, University of Science and Technology of China, Hefei, 230027, China

      Fax: +86-551-6360-6783

      e-mail: ustcwhm@ustc.edu.cn

      Additional correspondence Dr. Zhigang Tian, Institute of Immunology, School of Life Sciences and Hefei National Laboratory for Physical Sciences at the Microscale, University of Science and Technology of China, Hefei, 230027, China

      Fax: +86-551-6360-6783

      e-mail: tzg@ustc.edu.cn

    Search for more papers by this author

Abstract

Many differences exist between human immature and mature natural killer (NK) cells, but their respective molecular signatures and transcriptional regulators are relatively unknown. To gain new insights into the diversity and developmental regulation of human NK cells, we used data from high-resolution microarrays with independent verification to describe a comprehensive comparative analysis between immature decidual NK (idNK) cells with a CD56brightCD16T-bet phenotype and mature peripheral NK (mpNK) cells with a CD56dimCD16+T-bet+ phenotype. This study shows that many novel growth factors, cytokines, and chemokines are expressed by NK cells, and they may regulate NK-cell development or function in an autocrine manner. Notably, we present that idNK and mpNK cells are enriched for homeobox and zinc-finger transcription factors (TFs), respectively. Additionally, many novel candidate transcriptional regulators are common to both idNK and mpNK cells. We further describe the transcriptional regulatory networks of NK cells and show that the endogenous growth factors, cytokines, and TFs enriched in idNK cells regulate each other and may contribute to idNK-cell immaturity. Together, these findings provide novel molecular signatures for immature and mature NK cells, and the novel candidate regulators identified here can be used to describe and further understand NK-cell differentiation and function.

Introduction

Human natural killer (NK) cells can be classified into immature CD56brightCD16 cells and mature CD56dimCD16+ cells. CD56bright NK cells are generally believed to be the precursors of mature CD56dim NK cells [1-3]. However, the molecular signature differences between immature and mature NK cells are mostly unknown. Compared with mature NK cells, immature NK cells have a stronger capacity to proliferate and to secrete cytokines and chemokines. Phenotypic and functional changes occur during NK-cell maturation, including downregulating NKG2A expression; acquiring killer immunoglobulin-like receptors (KIRs), CD57, and CD11b; altering the expression patterns of homing molecules; displaying a gradual decline in proliferative capacity; and enhancing cell-mediated cytotoxicity [4, 5]. Decidual NK (dNK) cells are mostly a CD56bright immature population with immunomodulatory potential, whereas peripheral NK (pNK) cells are generally a functionally mature CD56dim population with high cytotoxic activity [6, 7]. Our previous work showed that most dNK cells belong to the CD27CD11b subset, which has an immature phenotype, displays developmental potential in response to IL-15, and increases cytotoxic potential in response to insulin-like growth factor 1 (IGF-1) [8, 9].

Several exogenous cytokines are involved in NK-cell development and function, including IL-2, IL-15, and IL-21, and novel endogenous cytokines have more recently been found, including bone morphogenetic protein 4 (BMP4) and IGF-1 [9-11]. Although many transcription factors (TFs) have been identified to affect NK-cell differentiation or function in mice, including ID2, NFIL3 (E4BP4), EOMES, ETS-1, and T-bet, little is known about human NK-cell transcriptional regulators [12-14]. Since the molecular signature of NK cells has been well-described in mice [15], we sought to compare the results from mouse models to humans to see if developmental mechanisms are conserved.

Using human whole-genome microarray data sets with subsequent verification, we provide novel molecular descriptions for human immature and mature NK cells; most importantly, our findings offer new insights into the transcriptional regulators that govern NK-cell development and function. Our study thus provides a valuable resource for further investigations into NK-cell biology.

Results

idNK cells exhibit immature characteristics compared with mpNK cells

NK cells in the peripheral blood account for a small fraction of total lymphocytes (∼10%) and are composed of two different subsets: the predominant CD56dimCD16+ mature subset (∼95%) and the much smaller CD56brightCD16 immature subset (∼5%). In contrast, NK cells are the dominant lymphocyte in the decidua during normal pregnancy, comprising up to ∼70% of the total lymphocytes and approximately 90% of dNK cells were of the CD56brightCD16 immature phenotype [6, 16] (Fig. 1A and B, and Supporting Information Fig. 1A and C).

Figure 1.

idNK cells exhibit immature characteristics relative to mpNK cells. (A) Representative density plots for CD16 and intracellular T-bet expression, as assessed by flow cytometry, are shown for the gated CD3CD56+ NK cells. Plots are representative of one out of seven independent experiments. (B) The percentage of cells expressing CD16 and T-bet was determined for each NK population (n = 14 for CD16; n = 12 for T-bet). The data show mean + SEM and are pooled from seven independent experiments. (C) Western blot for ID2 in idNK and mpNK cells. GAPDH was used as the loading control. The data are representative of more than three experiments. (D–G) Comparative transcriptome analyses were based on Whole Human Genome Oligo Microarray data of purified dNK, cNK, pNK, CD56+ T, and T cells (three donors were pooled for each cell type). Data shown are derived from a single experiment. (D) Hierarchical clustering of idNK, mcNK, mpNK, CD56+ T, and T cells, as reported by Cluster 3.0. (E) Heat map of mRNA transcripts up or downregulated at least twofold in one type of NK population relative to the other two NK populations, as reported by MEV 4.9 software. (F) The scatter plot of all valid probe signals for gene expressed in idNK, mcNK, and mpNK cells is presented as normalized log2 values for genes, and points outside the two thin red lines indicate genes exhibiting greater than twofold changes. R values represent the overall distribution of central tendency. (G) Distribution by functional category of genes exhibiting a minimum twofold change in expression among each NK cell type. I, immune response-related integral membrane proteins; II, cytokines; III, chemokines; IV, transcription factors; V, cytoskeleton or structural; VI, signal transducers; VII, cell proliferation or cell cycle; and VIII, apoptosis.

T-bet regulates the terminal maturation and function of murine NK cells during the final stage of development [17]. ID2 is expressed in NK-cell progenitors and regulates their early developmental processes [13, 18]. To further confirm that dNK cells are more immature than pNK cells, we analyzed T-bet and ID2 expression. Interestingly, we found that dNK cells were mostly T-bet, while >90% of the pNK cells were T-bet+ (Fig. 1A and B). Additionally, western blot analysis showed that ID2 was exclusively expressed by dNK cells (Fig. 1C). Furthermore, we detected many known cell-surface markers related to NK-cell maturation. Consistent with published observations [5, 12, 19], dNK cells expressed many immature cell markers, including CD27 and CD94, whereas pNK cells displayed a more mature phenotype with higher expression of CD57 and CD11b (Supporting Information Fig. 2). Overall, these results demonstrate that dNK cells exhibit immature characteristics relative to pNK cells.

To investigate novel molecular signatures and transcriptional regulators of immature and mature human NK cells, we performed whole-genome microarray analysis on purified dNK, cord-blood NK (cNK), and pNK cells, peripheral CD56+ T cells, and CD3+CD56 T cells were used as controls (Supporting Information Fig. 1). Notably, hierarchical clustering algorithm of variable genes indicated that pNK cells clustered far more closely with cNK cells than with dNK cells (Fig. 1D). We observed a similar transcriptional relatedness among dNK, cNK, and pNK cells by heat map and scatter plot analysis, where dNK cells notably differed from cNK and pNK cells (Fig. 1E and F). The expression profiles of cNK and pNK cells were similar, suggesting that these populations were at similar stages of maturity. We therefore designated these cells as immature dNK (idNK), mature cNK (mcNK), and mature pNK (mpNK).

To investigate novel molecular signatures and transcriptional regulators of immature and mature human NK cells, we classified the differentially expressed genes between idNK and mpNK cells into eight categories by function. idNK cells were enriched for cytokine and chemokine genes as well as genes related to cell cycle and proliferation (Fig. 1G). We found that idNK cells had higher expression levels of 30 genes encoding cell cycle- or proliferation-related proteins, including cyclins, CDKs, E2f TFs, and PCNA, relative to mpNK cells. In contrast, only two genes (encoding CCNB3 and CDC25B) were highly expressed in mpNK relative to idNK cells (Supporting Information Table 1). This result indicated that idNK cells possess a higher proliferative capacity than mpNK cells.

Among the differentially expressed genes, we next analyzed the integral membrane proteins involved in immune responses. To verify the microarray data, a representative set of cell-surface antigens was evaluated by FACS (fluorescence-activated cell sorter). Whereas idNK cells expressed more inhibitory receptors, including NKG2A, CD158b, and GITR, (Supporting Information Fig. 3A, D, and F), mpNK cells expressed more activating receptors and co-stimulatory factor genes, including CD2, CD8, and CD226 (Supporting Information Fig. 3B, E, and G). Additionally, integrin subunits CD29, CD49d, and CD11a were more highly expressed on mpNK than idNK cells (Supporting Information Fig. 3E and G). Tnfsf10 (CD253/TRAIL) and Klrg1 (KLRG), which are known as immature and mature NK-cell markers, respectively, in mice [12], were also, overexpressed in human idNK and mpNK cells, respectively (Supporting Information Fig. 3A and B).

In addition to the differences between idNK and mpNK cells, we also studied the common cell-surface molecules among different types of NK cells relative to T cells. The Kir (KIR), Klr (KLR), Ncr (NCR), and Hladr (HLADR) genes were all more highly expressed in NK and CD56+ T cells than in T cells (Supporting Information Fig. 3C). Overall, our data confirm that idNK cells exhibit immature characteristics compared with mpNK cells.

Numerous growth factor and cytokine genes are expressed in NK cells, especially idNK cells

In response to stimulation, immature CD56bright NK cells have a stronger ability to produce cytokines, such as IFN-γ, TNF-α, IL-8/CXCL8, and GM-CSF, than mature CD56dim NK cells [20, 21]. In addition to these well-known molecules, our microarray data revealed that two members of the transforming growth factor beta (TGF-β) superfamily, Tgfb2 and Bmp2, were significantly overexpressed in idNK relative to mpNK cells (Fig. 2A and D). TGF-β2 suppresses the generation of cytotoxic NK cells and plays an important role in immunoregulation [22]. BMP2 is a multifunctional growth factor that regulates cell proliferation, differentiation, and apoptosis. Deletion of uterine BMP2 receptor type 2 (BMPR2) in conditional knockout mice induces a deficiency in uterine NK cells [23]. Here, we found that idNK cells also highly express mRNA for the BMP2 receptors BMPR2 and BMPR1A (data not shown), indicating that endogenous BMP2 may also promote the early differentiation of NK cells in the decidua.

Figure 2.

Growth factor, cytokine, and cytokine receptor profiles of idNK and mpNK cells. (A–C) Comparative transcriptome analyses of growth factors, immune effector molecules, cytokines, and cytokine receptors were based on Whole Human Genome Oligo Microarray data of purified dNK, cNK, pNK, CD56+ T, and T cells (three donors were pooled for each cell type). Data represent the mean gene expression (or fold-change) values from three samples in each group and are derived from a single experiment. (A) Genes upregulated more than twofold or (B) downregulated by a minimum of 50% in idNK relative to mpNK cells. (C) Heat map of transcript expression upregulated at least twofold in both idNK and mpNK cells relative to T cells. (D) Verification of the transcripts identified in (A) by real-time PCR. Results were normalized to the housekeeping gene Actb and presented relative to mpNK cells. Data are shown as mean + SEM (n ≥ 4) and are pooled from more than three experiments. (E and F) Cytokine receptor expression was assessed by flow cytometry on idNK and mpNK cells. (E) Representative histograms and (F) statistical analysis for cytokine receptors are shown for the gated CD3CD56+ NK cells (n ≥ 5). Gray shading in (E) indicates staining by isotype-matched control antibody. (F) Data are shown as mean + SEM and are pooled from more than three experiments.

In addition to bmp2, idNK cells overexpressed several other growth factors crucial for bone formation and function, including Ptn, Tnfrsf11b, Spp1, Tnfsf11, and Ogn. Among these, Spp1 was notably upregulated more than 2000-fold in idNK relative to mpNK cells (Fig. 2D). Osteopontin (OPN, encoded by Spp1), which is secreted by hematopoietic stem cells (HSCs) and stromal cells, promotes NK-cell development [24]. Interestingly, mRNAs for two OPN receptors, CD44 and integrin αVβ3 (CD51/CD61) [25], were also highly expressed in idNK cells (data not shown). We therefore speculate that endogenous OPN may induce early differentiation of idNK cells. The Notch ligand Jagged2 (encoded by Jag2) promotes the development of NK cells [26]. Here, we found that Jag2 was significantly overexpressed in all three types of tested NK cells relative to T cells, especially in idNK cells (Fig. 2A and C). Taken together, our study identifies many growth factors that were reported to promote the development of NK cells can be secreted by NK cells. We infer that these NK cell-derived growth factors may execute immunoregulatory functions as well as regulate NK-cell self-renewal, homeostasis, and early differentiation. Compared with idNK cells, mpNK cells overexpressed a relatively small number of cytokine transcripts, including Ltb and Tnf (Fig. 2B). Compared with T cells, several growth factors, cytokine, and immune effector genes were upregulated both in idNK and mpNK cells; those upregulated more than twofold are detailed in the heat map shown in Figure 2C.

In order for cytokines to execute their intended functions, they must interact with their receptors. Among the many cytokine receptors differentially expressed between idNK and mpNK cells, CD122 (IL-2Rβ, encoded by Il2rb) and ST2L (IL-33R, encoded by Il1rl1) were more highly expressed in idNK relative to mpNK cells (Fig. 2A, E, and F). CD360 (IL-21R), CD218a (IL-18Rα), and TGF-βR III were highly expressed in mpNK relative to idNK cells (Fig. 2B, E, and F), consistent with the known functional effects of these cytokines on mature NK cells [27]. Overall, we found that NK cells overexpress several growth factors and cytokines along with their corresponding receptors, which may be useful for the further study and understanding of NK-cell characteristics.

idNK and mpNK cells overexpress chemokine and chemokine receptor genes, respectively

The chemokine superfamily is extremely important for both innate and adaptive immune responses because of its involvement in recruiting and activating specific leukocyte subsets. During immune responses, NK cells produce many chemokines, including MIP-1α/CCL3, MIP-1β/CCL4, and RANTES/CCL5 [20, 28]. In our microarray study, many chemokine genes were overexpressed in idNK relative to mpNK cells, and the significantly different genes were verified by real-time PCR (Fig. 3A and D). Among these chemokine genes, very high fold changes were particularly evident in Cxcl10, Cxcl1, and Cxcl14 (Fig. 3D). In addition, nine chemokine genes were highly expressed both in idNK and mpNK cells compared with T cells, as shown by heat map analysis (Fig. 3C).

Figure 3.

idNK and mpNK cells overexpress chemokine and chemokine receptor genes, respectively. (A–C) Comparative transcriptome analyses of chemokines and chemokine receptors were based on Whole Human Genome Oligo Microarray data of purified dNK, cNK, pNK, CD56+ T, and T cells (three donors were pooled for each cell type). Data represent the mean gene expression (or fold-change) values of three samples in each group and are derived from a single experiment. (A) Genes upregulated more than twofold or (B) downregulated by a minimum of 50% in idNK relative to mpNK cells. (C) Heat map of transcripts upregulated at least twofold in both idNK and mpNK cells relative to T cells. (D) Verification of the chemokines identified in (A) by real-time PCR. Results were normalized to the housekeeping gene Actb and presented relative to mpNK cells. Data are shown as mean + SEM (n ≥ 5) and are pooled from more than three experiments. (E and F) Chemokine receptor expression was assessed by flow cytometry on idNK and mpNK cells. (E) Representative histograms and (F) statistical analysis for chemokine receptors are shown for the gated CD3CD56+ NK cells (n ≥ 5). Gray shading in (E) indicates staining by the isotype-matched control antibody. Data are shown as mean + SEM and are pooled from more than three experiments.

In contrast to our finding that idNK cells showed enriched expression of many chemokine genes, mpNK cells were instead enriched for many chemokine receptor genes. While 13 chemokine receptor genes were overexpressed in mpNK cells, no chemokine receptor genes were overexpressed in idNK cells (Fig. 3A and B), indicating that mpNK cells may have more migration ability than idNK cells. By FACS analysis, we found that CXCR4, CXCR2, CXCR1, and CCR6 expression was higher on mpNK relative to idNK cells (Fig. 3E and F). However, the higher CXCR3 protein expression on idNK relative to mpNK cells was inconsistent with the corresponding relative mRNA levels (Fig. 3B, E, and F), suggesting a possible role for post-transcriptional regulation. CXCR3 is mainly expressed on activated T cells and NK cells and can be induced by three IFN-γ-inducible ligands, CXCL9/MIG, CXCL10/IP-10, and CXCL11/I-TAC [29]. Interestingly, idNK cells also had higher expression of Cxcl9, Cxcl10, and Cxcl11 (Fig. 3A and D). Many chemokines have been reported to stimulate the migration of activated NK cells, including XCL1, CXCL8, CXCL9, CXCL10, CXCL11, CXCL14, CCL3, CCL4, and CCL5 [30]. Given that these chemokine genes were all highly expressed in NK cells [29] (Fig. 3A–C), we assume that in addition to affecting other lymphocytes, these chemokines can affect NK-cell function in an autocrine manner. Overall, these data suggest a specific signature of chemokines and chemokine receptors for each of the idNK and mpNK-cell populations.

Transcription factor profiles differ between idNK and mpNK cells

Transcription factors are key regulators of gene transcription and have a major effect on immune cell fate and differentiation [31]. To discover the key TFs important for NK-cell development and function, we focused on TFs differentially expressed between idNK and mpNK cells. We broadly classified these TFs into five categories based mainly on their DNA-binding domains: homeobox, C2H2-type zinc finger, multi-cysteine zinc finger, basic DNA binding domain, and others. We found that idNK cells clearly expressed different TFs relative to mpNK cells. idNK cells were enriched for homeobox TFs, while mpNK cells were enriched for zinc-finger proteins (Fig. 4A and B). Whereas only five homeobox TFs were upregulated in mpNK cells (Fig. 4B), approximately 29 were overexpressed in idNK cells, including Hop, Hoxa9, Hoxa5, and Six4 (Fig. 4A). Homeobox TFs share a well-conserved homeodomain and are key regulators of embryonic development. One mechanism by which HOX proteins achieve functional specificity is by cooperating with additional DNA-binding cofactors, including PBX, MEIS, and Smad proteins [32]. We found that idNK cells also expressed higher Meis1 and Pbx1 cofactors relative to mpNK cells (Fig. 4A and C). It was reported that Pbx1-null mice are completely deficient in B and NK cells [33]. Given that Pbx1 is highly homologous between humans and mice, we presume that PBX1 may also be involved in human NK-cell differentiation.

Figure 4.

Transcription factor profiles differ between idNK and mpNK cells. Transcription factor gene expression (A) upregulated more than twofold or (B) downregulated by a minimum of 50% in idNK relative to mpNK cells based on analyzing the Whole Human Genome Oligo Microarray. Data represent the mean gene expression signal values of three samples in each group and are derived from a single experiment. (C and D) Verification of the TFs identified in (A) and (B) by real-time PCR, respectively. Data are shown as mean + SEM (n ≥ 4) and are pooled from more than three experiments.

Whereas only six zinc-finger protein genes (including Klf5, Znf215, and Nr2f2) were upregulated in idNK relative to mpNK cells, 20 C2H2-type zinc-finger protein genes (including Klf9, Znf831, and Znf483) and five multi-cysteine zinc-finger protein genes (including Nr5a1, Pparg, and Nr3c2) were overexpressed in mpNK relative to idNK cells (Fig. 4A and B). The zinc-finger proteins, characterized by the common Znf domain, comprise the largest family of DNA-binding TFs in humans and other mammals, but few studies have explored their function in NK cells [34]. In addition to the homeobox and zinc-finger TFs, the basic DNA-binding domain protein genes Maf, Olig2, and Tbx21 were also upregulated in mpNK relative to idNK cells (Fig. 4B and D).

In summary, we found that homeobox and zinc-finger TFs were enriched in idNK and mpNK cells, respectively. Notably, most of the TF genes that were found to be differentially expressed between idNK and mpNK cells have not yet been reported, and their functions in NK cells are worthy of further study.

Transcription factor profiles common in different NK-cell populations

Despite the diversity of NK-cell populations, we hypothesized that common, dominant transcriptional regulators may exist in immature and mature NK cells to determine development and function. Using our microarray data, we identified TFs that were significantly overexpressed in the three tested NK-cell populations — idNK, mcNK, mpNK — relative to T cells. The transcripts upregulated twofold or more in two (Fig. 5A), or all three (Fig. 6A) types of the tested NK populations were chosen for further analysis, and we verified the significantly changed TFs by real-time PCR. Compared with T cells, Etv5, Nfe2, Erg, Mycn, Pbx1, Hoxa3, and Nr2f2 were all significantly upregulated in NK cells, and Nfib was overexpressed specifically in idNK and mcNK cells (Fig. 5B).

Figure 5.

Identification of unique transcription factors in two types of NK cells. (A) Heat map of the mRNA transcripts upregulated at least threefold in two of the tested NK populations relative to T cells. Analyses are based on the microarray data of purified dNK, cNK, pNK, CD56+ T, and T cells (three donors were pooled for each cell type). Data represent the mean fold-change values of three samples in each group and are derived from a single experiment. (B) Verification of the TFs identified in (A) by real-time PCR. Results were normalized to the housekeeping gene Actb and presented relative to T cells. Data are shown as mean + SEM (n ≥ 4) and are pooled from more than three experiments.

Figure 6.

Identification of unique transcription factors in all three types of NK cells. (A) Heat map of the transcripts upregulated at least twofold in all three tested NK-cell populations relative to T cells. Analyses are based on the microarray data of purified dNK, cNK, pNK, CD56+ T, and T cells (three donors were pooled for each cell type). Data represent the mean fold-change values of three samples in each group and are derived from a single experiment. (B) Verification of the TFs identified in (A) by real-time PCR. Results were normalized to the housekeeping gene Actb and presented relative to T cells. Data are shown as mean + SEM (n ≥ 4) and are pooled from more than three experiments.

TF genes, including Id2, Nfil3, Eomes, Mitf, and Klf4, were all highly expressed in the three tested NK-cell populations and the CD56+ T-cell lineage relative to T cells (Fig. 6A), which was consistent with studies showing a requirement for these TFs in the development and function of NK cells in mice [14]. In addition to the above-mentioned TFs, many previously unknown transcriptional regulators were also identified, including Hoxa10, Fosl2, and Creb5 (Fig. 6A and B).

As many diverse NK-cell populations have been described, we isolated other NK-cell populations, including immature CD56bright pNK cells, mature CD56dim pNK cells, and relatively mature CD34+ cell-derived NK populations (Supporting Information Fig. 4A and B). We then analyzed the expression levels of the identified TFs among those NK-cell populations. We found that idNK cells have some different TF profiles (such as Hoxa9, Six4, and Nr2f2) than immature CD56bright pNK cells. However, several TF mRNA transcript levels (such as Creb5, Nr4a1, and Pbx1) in idNK cells are comparable to CD56bright pNK cells. Furthermore, we found that most of these common TFs (such as Fosl2, Erg, and Cenpf) were similarly expressed in these NK-cell populations. Additionally, several TF mRNA transcript levels in CD56dim pNK cells are comparable to CD56bright pNK cells (Supporting Information Fig. 4C).

Overall, we found numerous novel TFs expressed in both immature and mature NK cells that may be involved in NK-cell maturation and function.

Proposed self-regulatory network of NK cells

Our data suggest that a number of NK cell-derived growth factors, cytokines, and chemokines may be essential for NK-cell differentiation, maturation, or function in an autocrine manner. However, their precise mechanism of action is currently unclear. Additionally, we found that NK cells highly expressed many TFs, but identifying and understanding their downstream targets or upstream regulators remains challenging. In view of the respective co-expression of these growth factors, cytokines, chemokines, cell-surface receptors, and TFs within each distinct NK population, we hypothesize that they form a regulatory network together.

Based on our microarray data and PCR-verified results, we selected six autocrine growth factors and more than 30 novel TFs as probable candidate transcriptional regulators of NK cells. We searched for and predicted putative target genes of all the selected TFs by combining information from published reports and three bioinformatic databases; well-known cytokines and TFs important for human and mouse NK-cell development or function were also included (Fig. 7). In the illustration of the resulting molecular network, shown in Figure 7A, molecules appearing in the innermost circle are important growth factors, cytokines, and their corresponding receptors in idNK cells that may be involved in NK-cell development and/or function; TFs appearing in the middle circle — representing transcripts expressed more highly in idNK than in mpNK cells (Fig. 4A and C) — are candidate transcriptional regulators of immature NK cells; ZNF143, PPARG, T-bet, Oligo2, and MAF in the outermost circle — representing transcripts expressed higher in mpNK than in idNK cells (Fig. 4B and D) — are candidate transcriptional regulators of mature NK cells; the other TFs in the outermost circle — representing transcripts expressed higher in at least two of the tested NK populations than in T cells (Fig. 5 and 6) — are candidate transcriptional regulators of NK cells. These growth factors, cytokines, and TFs interact and regulate each other through a series of actions that then affect the expression or activity of other molecules, which finally affect NK-cell development or function.

Figure 7.

Proposed self-regulatory network of NK cells. (A) Putative physical and regulatory interactions between the candidate transcriptional regulators and target genes in NK cells. NK cell–derived growth factors, cytokines, and TFs enriched in NK cells regulate each other, forming a complicated network to activate or promote the expression of molecules important for NK-cell development or function. (B) Self-regulatory network of idNK cells. Autocrine BMP and TGF-β may coordinate to maintain the immature state of idNK cells by promoting CD56 and ID2 expression and inhibiting CD16 and T-bet expression directly, or by impacting the Smad/HOX interaction. BMP2 positively regulates several growth factors including OPN, PTN, and OPG, which are highly expressed in idNK cells. TGF-β inhibits T-bet, NKp30, NKG2D, and CD16 expression, which are involved in NK-cell development and function. (C) Cytokines and TFs highly expressed by idNK cells promote entry into the cell cycle. Endogenous growth factors in idNK cells, such as PTN, OPG, and the TF CREB5, can regulate cell cycle-related proteins cyclin D1 and cyclin A that promote cell proliferation and self-renewal, and may contribute to idNK-cell immaturity.

idNK cells maintain their immature phenotype and possess a stronger proliferative capacity than mature NK cells, but the mechanism of this is unclear. From the established interaction networks, we highlight two important pathways — BMP and TGF-β — that may contribute to the early development and maintenance of idNK-cell immaturity. BMP4, BMP2, and TGF-β can impact TF interactions, growth factor secretion, and receptor expression in idNK cells. The BMP and TGF-β signaling pathways may coordinate to maintain the immature state of idNK cells with a CD56brightCD16T-betID2+ phenotype (Fig. 1A–C and 7B). Additionally, from the speculated network, we found that TFs and endogenous growth factors of idNK cells can regulate cell cycle-related proteins that promote proliferation and self-renewal while inhibiting maturation [35-37] (Fig. 7C). In summary, we found that candidate transcriptional regulators, endogenous growth factors, and cytokines regulate one another in a complicated network within NK cells, which may impact NK-cell phenotype, differentiation, maturation, and function.

Discussion

The molecular definition of NK cells in mice has been well described, but a comprehensive understanding of human NK cells has not yet been achieved. Using high-resolution microarray analyses with independent verification, we provide detailed and comprehensive data regarding the molecular signatures of immature and mature human NK cells, which include novel insights into the regulators of NK-cell development and function.

Human NK cells have many subpopulations with a remarkable degree of repertoire diversity. Using multiparametric mass cytometry to examine expression of many NK cell surface markers, Horowitz et al. provide a framework to understand human NK-cell repertoires [38]. We performed analysis on different types of NK cells, and found that the majority of dNK cells were of the CD56brightCD16T-bet immature type, and most of the pNK cells were of the CD56dimCD16+T-bet+ mature phenotype. Many similarities and differences were identified between idNK and mpNK-cell receptors. As NK-cell diversity and function are closely linked, the differences in cell-surface molecule expression are consistent with the known functions of these cells [38]. The higher expression of inhibitory receptors on idNK cells may relate to their weak cytotoxic ability and maintenance of self-tolerance, while the higher expression of activating receptors and costimulatory receptors on mpNK cells may be required for their strong cytotoxic ability to respond to pathogens and tumors.

We also identified many growth factor, cytokine, and chemokine transcripts that were distinct between idNK and mpNK cells. In particular, idNK cells contained higher mRNA levels of cytokines linked to osteoclast and osteoblast genesis (e.g., Tnfsf11, Tnfrsf11b, Spp1, and Bmp2) compared with mpNK cells. RANKL (encoded by Tnfsf11) and M-CSF secreted by synovial NK cells can induce monocytes to differentiate into osteoclasts in vitro [39]. Accordingly, further study into the interrelation of NK cells and bone formation/development is desired. Although BMP4 can promote thymic NK-cell development in an autocrine manner [11], whether BMP2 and other autocrine cytokines can affect idNK-cell differentiation or function is still unclear. Furthermore, we found that idNK cells had higher expression of chemokine genes, including Xcl1, Cxcl1, Cxcl10, and Cxcl14, which confer idNK cells with an increased ability to recruit other NK cells or lymphocytes, such as immature dendritic cells (DCs) and neutrophils [40]. Importantly, recent studies suggest that IL-32, OPN, and CXCL14 increase DC maturation and function [40-42], indicating that as producers of these molecules, idNK cells may have an impact on DCs. Some in vivo or in vitro experiments will be necessary to test these inferred functional interactions in future studies. Although the effect of exogenous cytokines on NK-cell development and function has been well described in the past few decades, the effect of endogenous cytokines has generally been ignored. Based on our finding that many growth factors, cytokines, and chemokines are coexpressed with their corresponding receptors, these NK cell-derived molecules may affect NK-cell development or function by a powerful autocrine pathway in addition to the known paracrine pathway.

Notably, our study is the first to describe the probable transcriptional regulators of the different human NK populations, including immature (dNK and CD56bright pNK cells) and mature (CD56dim pNK and CD34+ cell-derived NK cells) NK cells. We found that idNK and mpNK cells are enriched for homeobox family TFs (such as Hoxa5, Hoxa9, Pbx1, and Hop) and zinc-finger proteins (such as Klf9, Znf143, Znf483, and Znf831), respectively. Homeobox family TFs are known as HSCs self-renewal regulators [31]. Distal-less (Dlx) homeobox TFs regulate the differentiation and maturation of NK cells in mice, and persistent expression of Dlx arrests NK-cell development at an immature stage [43]. We therefore presume that the homeobox family TFs highly expressed by idNK cells may contribute to their immaturity. ZNF143 can be induced by IGF-1 [44]. Interestingly, we previously reported that IGF-1 is critical for human NK-cell cytotoxicity [9], which is consistent with the observed higher Znf143 expression in mpNK cells that are known to exert cytotoxic activity. We therefore presume that zinc-finger proteins may regulate genes important for NK-cell cytotoxicity, although further study into the precise role of zinc-finger proteins in NK cells is desired.

Thus far, many transcriptional regulators in the mouse immune system have been identified [15, 31], and we confirm here that many lessons learned from mouse models can also be applied to the human system. In addition to confirming well-known TFs, we identified many novel TFs (including Etv5, Nfe2, Mycn, Nr2f2, and Hoxa10) that potentially regulate NK-cell development and function.

According to our data and the putative genes targeted by the selected TFs, we propose a self-regulatory network for NK cells. From this, we can infer that during HSCs development into NK cells, autocrine and/or paracrine cytokines and chemokines induce or activate TFs that regulate each other and conversely activate or promote the expression of molecules important for NK-cell development and maturation. This may be one feasible mechanism underlying the generation of programmed mature and functional NK cells. Additionally, we highlight that autocrine BMP2 and TGF-β2, in coordination with their regulated cytokines (such as Osteoprotegerin/OPG, OPN, and Pleiotrophin/PTN) and the TFs enriched in idNK cells (such as homeobox TFs, CREB, and MYCN), may contribute to idNK-cell immaturity. It will be necessary to determine the protein levels of these molecules in future studies.

idNK cells display characteristics associated with immature, developing NK cells, and they exhibit many similarities to immature CD56bright pNK cells. For instance, both cell types are CD56bright CD16 CD57−, show relatively higher expression levels of CD94 and CD117, and are known to have large cytokine production capacity and low cytotoxicity. Yet, idNK cells also display several differences compared with immature CD56bright pNK cells. For instance, idNK cells express higher levels of CD27, NKG2A, and CD158b and lower levels of CD62L, CD11a, and CD11c relative to CD56bright pNK cells. In addition to cell surface molecules, there were also several similarities and differences in the TFs expression profiles of these two CD56bright NK cells. Overall, even though the idNK-cell profile delineated in this study may have been influenced by the maternal decidual microenvironment during pregnancy, it was mostly consistent with the reported characteristics of immature NK cells [4, 12, 19, 45, 46]. Our observations on the molecular signatures and transcriptional regulatory networks of idNK and mpNK cells will contribute to further understanding of NK-cell differentiation and function.

A previous study found that human dNK cells are a unique NK-cell subset with immunomodulatory potential [6]. In that study, CD56bright dNK cells were compared with CD56bright and CD56dim pNK cells by microarray analysis. The authors found many genes that were upregulated in dNK cells, including cell-surface proteins (such as CD9, NKG2C, and KIRs) and two secreted proteins (galectin-1 and progestagen-associated protein 14) with immunomodulatory functions. Overall, the previous study by Koopman et al. provided a detailed and comprehensive view of the signatures of dNK and pNK cells. However, at that time, microarray chips only contained probes for ∼10,000 genes. Here, we performed whole-genome microarray analysis on purified dNK, cNK, pNK, CD56+ T, and T cells and the microarray chip included more than 40,000 probes for ∼33,000 genes. Therefore, our research confirms the results of the previous study while expanding the number of examined genes. Additionally, while the previous study well characterized the immunomodulatory potential of idNK cells, we demonstrate our study from the perspective of cell development and differentiation. We analyzed not only the cell-surface receptors but also other NK-cell signatures, including growth factors, cytokines, chemokines, and transcriptional regulators, which may direct NK-cell fate. It will be necessary to determine the function of these molecules in future studies.

In conclusion, our data illustrate the detailed signatures and transcriptional regulatory networks of immature and mature human NK cells. These data enrich our current understanding of NK cells and will allow us to better control NK-cell function in hematopoietic-related diseases and cancer.

Materials and methods

Human sample collection and lymphocyte isolation

Decidual samples from normal donors undergoing elective abortion in the first trimester between 6–12 weeks of gestation and cord-blood samples from full-term, healthy newborns were obtained at the Anhui Provincial Hospital, Hefei. Adult peripheral blood samples were obtained from healthy donors at the Blood Center of Anhui Province. All samples were collected after donors gave informed consent. This study was approved by the Ethics Committee of the University of Science and Technology of China. Lymphocyte isolation was performed as previously described [9, 16].

Flow cytometry

Cells were stained with the following fluorochrome-conjugated human mAbs for FACS: CD3, CD16, CD45, CD56, CD117, CD27, CD94, CD57, CD11b, CD62L, CD38, CD103, CD54, CD158b, CD29, CD11a, CD49d, CD8, CD36, CD226, CD122, CD182, CCR6, and CXCR3 were purchased from BD Biosciences; NKG2A, GITR, and TGF-βR III were purchased from R&D Systems; T-bet, CD218a, CD181, and CXCR4 were purchased from e-Bioscience; CD360 was purchased from Biolegend; ST2L was purchased from Mdbioproducts. Homologous IgG or IgM antibodies were used as isotype controls. FACS staining was performed according to the manufacturer's instructions. The data were analyzed using FlowJo software (Tree Star).

Cell purification

dNK cells (CD45+CD3CD56+) were purified from first-trimester deciduas. cNK cells (CD3CD56+) were purified from cord-blood mononuclear cells. pNK (CD3CD56+), CD56+ T (CD3+CD56+), and T (CD3+CD56) cells were purified from adult peripheral blood mononuclear cells. All cells used for microarray were purified by cell sorting (BD Bioscience). Cell purity was determined to be >95% by post-purification FACS analysis. In some experiments, NK cells and T cells were purified by the MACS isolation system according to the manufacturer's instructions (Miltenyi Biotec). CD56bright and CD56dim pNK cells were purified by cell sorting after MACS.

Immunoblot analysis

Immunoblots were performed as previously described [9]. Antibodies used were as follows: rabbit anti-ID2 (Cell Signaling); rabbit anti-GAPDH (Abmart), goat anti-rabbit HRP-conjugated secondary antibody (Boster).

Gene expression analysis

For analyzing the molecular signatures of human NK cells, purified dNK, cNK, pNK, CD56+ T, and T cells (three donors were pooled for each cell type) were submitted for microarray analysis using the Whole Human Genome Oligo Microarray (Agilent, G4112F). Microarray image analysis was performed using Agilent's Feature-Extraction V9.1.3 software (Agilent Technologies). Expression values were log2-transformed, and subsequent analyses were conducted using SAS statistical software online (http://sas.ebioservice.com/). Hierarchical clustering was exported by Cluster 3.0 (complete linkage clustering), and the heat map and blue–yellow scale schemes were designed by MEV 4.9 software. The microarray data were deposited into the National Center for Biotechnology Information GEO repository under accession number GSE24268.

Generation of human umbilical cord blood CD34+ cell-derived NK cells

Human umbilical cord blood CD34+cell-derived NK cells were prepared as previously described [9].

RNA isolation and real-time PCR analysis

Total RNA was extracted from lymphocytes with TRIzol reagent (Invitrogen) or the Arcturus PicoPure RNA isolation kit (Invitrogen) (Supporting Information Fig. 6). cDNA was amplified using specific primers with SYBR Premix Ex Taq (TaKaRa). The cycling-threshold (CT) value for each gene was normalized to the housekeeping gene β-actin. Primer sequences for β-actin were as follows: 5′-TTGCCGACAGGATGCAGAA-3′ (forward) and 5′-GCCGATCCACACGGAGTACTT-3′ (reverse) for a 101-bp product. The sequences of the other gene-specific primers are listed in Supporting Information Table 2–7. The 2-ΔΔCt method was used to analyze the data.

Prediction of transcription factor target genes and the establishment of interaction networks

The target genes for all selected TFs were identified or predicted by searching published reports and three online bioinformatic databases, including STRING (http://string-db.org), Gene Network Central (http://www.sabiosciences.com), and the Transcriptional Regulatory Element Database (http://rulai.cshl.edu). The interaction networks for these genes were created using Cytoscape v2.8.2 software.

Statistical analyses

We used two-tailed unpaired Student's t-tests to determine statistical significance. Results were considered statistically significant when they achieved values of p < 0.05.

Acknowledgements

This work was supported by grants from the Natural Science Foundation of China (#81330071, #31021061) and the National Basic Research Project (973 Project) (#2012CB519004, 2013CB944902, and 2013CB530506).

Conflicts of Interest

The authors declare no commercial or financial conflict of interest.

Abbreviations
BMP2

bone morphogenetic protein 2

BMP4

bone morphogenetic protein 4

cNK

cord-blood NK

dNK

decidual NK

idNK

immature decidual NK

mcNK

mature cNK

mpNK

mature peripheral NK

pNK

peripheral NK

OPG

osteoprotegerin

OPN

osteopontin

PTN

pleiotrophin

TF

transcription factor

Ancillary