Human Decidual Natural Killer Cells Are a Unique NK Cell Subset with Immunomodulatory Potential

In pregnancy, hemiallogeneic fetal cells invade the maternal decidua but remain spared from attack by the maternal immune system, posing a great unsolved paradox of immunology (1). Although several factors have been proposed to explain or contribute to maternal tolerance, our understanding of the immunobiology of normal pregnancy and its implications for pregnancy-related pathologies is still limited (2). One prominent feature of the pregnant human decidua is the striking abundance of NK cells, which constitute ∼70% of resident lymphocytes (3). In contrast, NK cells in peripheral blood comprise <15% of circulating lymphocytes (4). Thus, human uterine decidual NK (dNK) cells have been thought to play a role in implantation and pregnancy, at least in early gestation. Although the precise functions of dNK cells in vivo are still unknown, their proximity to the invading trophoblasts, which lack expression of classical HLA-A and -B antigens (5) but selectively express HLA-C and the nonclassical HLA-E, -G, and CD1d molecules (6–9), has led to the proposal that these MHC antigens on trophoblasts interact with NK cell receptors (3, 10). Human NK cell receptors include the C-type lectinlike NKG2 receptors that associate with CD94 and the killer cell Ig-like receptors (KIRs), each of which have both inhibitory and activating isoforms that depend on the presence of immunoreceptor tyrosine-based inhibitory motifs (ITIMs) in their cytoplasmic tails or immunoreceptor tyrosine-based activating motifs in associated molecules, such as DAP12. Moreover, the KIR locus is polygenic and polymorphic, giving rise to variable KIR repertoires that are expressed on subsets of NK cells among individuals. However, it is still unclear if and which receptors on dNK cells interact with trophoblast-expressed HLA molecules, and whether such interactions inhibit NK cell lysis, or lead to production of cytokines that favor normal placental development and maintenance of pregnancy (11). Moreover, the interactions between NK cell receptors and their ligands have been investigated mostly on peripheral blood NK (pNK) cells, but the relationship between dNK and pNK cells is unclear.

pNK cells comprise two different subsets, the predominant CD56^dim NK cell subset and the much smaller CD56^bright NK cell subset, constituting ∼95 and 5% of pNK cells, respectively (4). CD56^dim pNK cells express high levels of CD16 and both CD94-associated lectinlike NKG2 receptors and KIRs, are granular, and known to be cytotoxic. In contrast, CD56^bright pNK cells are mostly devoid of granules, CD16, and KIRs, but express higher amounts of other markers (12–15), such as CD94 (16) and L-selectin (17). Recent papers demonstrate greater cytokine production capacity of CD56^bright than CD56^dim pNK cells and, thus, suggest an immunoregulatory role for the CD56^bright pNK subset (18, 19). The resemblance of dNK cells to the CD56^bright pNK subset, at least in terms of their CD56^bright CD16^neg phenotype, has suggested that dNK cells are derived from CD56^bright pNK cells that are seeded in the uterus and undergo further differentiation in the decidual microenvironment. For instance, they both exhibit low cytotoxicity against the NK cell target K562 (20), and contain cytoplasmic CD3ε mRNA (21), in contrast to CD56^dim pNK cells. On the other hand, like CD56^dim pNK, dNK cells are granulated and express KIRs (22, 23). Yet, other properties seem unique to dNK cells, such as the expression of the activation marker CD69 (24) and the absence of L-selectin (25). Despite some additional characterization of phenotypic and functional differences between dNK and pNK cells (26), a comprehensive comparison of dNK cells in relation to both CD56^bright and CD56^dim pNK cell subsets is lacking.

In this work, patterns of gene expression in freshly isolated CD56^bright pNK, CD56^dim pNK, and dNK cells were compared using oligonucleotide microarrays. The transcriptional profiles provide detailed information on the differences between dNK and pNK cells, as well as between the peripheral CD56^dim and CD56^bright subsets. Using a stringent statistical approach, a large number of differentially expressed genes (197) were identified between dNK and pNK cells, whereas only ∼40 genes distinguished the two subsets of pNK cells from each other. Multiple genes were expressed in dNK cells whose transcripts were completely absent in pNK cells. A number of differentially expressed genes/proteins were validated by RT-PCR or flow cytometry. Several genes differentially and/or uniquely expressed in dNK cells encode proteins that are likely to be involved in maternal-fetal tolerance and provide potential critical targets for further study of dNK cells.

Decidual samples from patients undergoing elective abortion in the first trimester between 6 and 12 wk of gestation were collected at the New York University Medical Center. Decidual tissue was washed extensively in PBS supplemented with 50 μg/ml gentamicin before mincing with sterile scissors. Decidual lymphocytes were released by digesting the tissue with 0.1% collagenase type IV and 0.01% DNase I (both from Sigma-Aldrich) in RPMI 1640 medium for 30 min at 37°C. Decidual stromal cells and macrophages were allowed to adhere to tissue culture plates for 2 h at 37°C in a humidified 5% CO₂ incubator. Lymphocytes from the overlaying solution were subsequently purified by density gradient centrifugation (Ficoll-Hypaque; Amersham Biosciences) and used for flow cytometric staining and cell sorting. Peripheral blood lymphocytes were isolated from peripheral blood from anonymous healthy donors using Ficoll either directly or after enrichment for NK cells using Rosettesep according to the manufacturer's instructions (StemCell Technologies Inc.).

The following murine mAbs, purified or directly conjugated with FITC, PE, or Cychrome, were used in FACS^® analysis: anti-CD3, anti-CD56, anti-CD16, anti-CD9, anti-KIR2DL1/S1 (CD158a; EB6), anti-KIR2DL2/2DS2/2DL3 (CD158b; GL183), and anti-KIR3DL1 (CD158e; NKB1), which were all from BD Biosciences; anti-CD62L (DakoCytomation); anti-KIR2DL4 (KIR2DL4-specific mAbs 33, 36, and 64, supplied by E. Long [National Institutes of Health, Bethesda, MD; reference 27]), and IgG and IgM isotype controls (BD Biosciences). Secondary anti–mouse Ig fluorochrome conjugates used were goat F(ab′)2 anti–mouse IgG-FITC, goat F(ab′)2 anti–mouse IgG + IgM-FITC, and goat anti–mouse IgG-PE (Caltag). For FACS^® staining and sorting, cells were washed in PBS supplemented with 2% FCS and incubated with mAb on ice for 30 min, followed by washing twice with PBS, pH 7.2, and 2% FCS. Cell sorting and fluorescence measurements were performed on a MoFLo high performance cell sorter (DakoCytomation). For fluorescence measurements only, data from 10,000 to 50,000 single cell events were collected using a standard FACScalibur™ flow cytometer (Immunocytometry Systems; Becton Dickinson). Data were analyzed using CELLQuest™ (Becton Dickinson) or Flow Jo (Treestar).

Flow-sorted cells were washed with PBS, and the cell pellets were frozen in TRIzol (GIBCO BRL). Total RNA was isolated according to the TRIzol manufacturer's instructions, with addition of 5 μg linear polyacrylamide (Genelute LPA; Sigma-Aldrich) to aid visualization of RNA pellets. RNA was subjected to the clean-up protocol as per the manufacturer's instructions (Rneasy mini RNA isolation kit; QIAGEN). The subsequent RNA amplification and clean-up protocol was adapted from Baugh et al. (28). Biotinylated cRNA was hybridized to human genome HGU95Av2 chips displaying probes for ∼10,000 full-length genes for expression analysis. Chips were processed according to manufacturer's instructions (for further experimental details see Online Supplemental Material). Gene transcript levels were determined using algorithms in the Microarray Suite 5.0 software (Affymetrix, Inc.). Based on Affymetrix, Inc. decision matrices, each probe was assigned a call of present (expressed) or absent (not expressed). 19 samples out of 22 samples were of sufficient quality to be further analyzed (nine dNK, five CD56^bright pNK, and five CD56^dim pNK cells; Table SI).

Sample clustering analysis was performed with D-chip software (http://www.dchip.org). Intensity values were normalized by the model-based expression analysis method (29). Genes with variable expression levels across the 19 samples were selected by applying two criteria as follows: genes should be expressed (have presence calls) in at least 20% of the 19 samples, and σ_i/μ_i ratio should be >0.5 and <25, where σ_i and μ_i are the standard deviation and mean of the hybridization intensity values of each particular gene across all samples, respectively. An unsupervised hierarchical clustering algorithm (30) was applied to group the 19 samples based on the similarity of the expression profiles of the selected genes.

The 19 samples were normalized by setting a trimmed mean (using middle 95% of the distribution) to be the same; there were no noticeable problems with saturation or other nonlinear effects. To reduce the number of genes for subsequent analysis, a conservative requirement was applied that a gene must be present (according to the Affymetrix, Inc. algorithm) in at least 5 of the 19 samples, given the smallest sample size of 5 in both pNK cell groups.

Samples were grouped according to their biological origin in dNK (nine samples), CD56^dim pNK (five samples), and CD56^bright pNK cells (five samples). Three pairwise comparisons (dNK vs. CD56^bright pNK, dNK vs. CD56^dim pNK, and CD56^bright pNK vs. CD56^dim pNK) were performed to identify differentially expressed genes using the Student's t test with unequal variances. At P ≤ 0.01 (unadjusted for multiple testing), >1,100 genes were significantly differentially expressed in the pairwise comparisons involving dNK cells, whereas ∼300 were differentially expressed in the CD56^bright pNK versus CD56^dim pNK cell comparison. To rule out the possibility that this result was generated by the larger sample size in the case of dNK (nine samples vs. five of each pNK subset), the analysis was repeated with all possible sets of five samples from the nine original decidual samples. Although the total number of differentially expressed genes between all three subsets was slightly less, the number of differentially expressed genes between dNK and pNK^bright or pNK^dim cells was still nearly three times the number of such genes between pNK^bright and pNK^dim cells (Table SII).

278 genes showing at least a threefold change in transcript levels at Student's t test P ≤ 0.001 in at least one of the three pairwise comparisons were considered for further analysis. For genes represented by multiple probesets, the result for only one representative probeset is shown. Genes were classified, based on information available on the worldwide web (see Online Supplemental Material).

For Venn diagrams, genes were considered to be up-regulated in a particular NK subset if they showed a significant fold increase (P ≤ 0.001) in at least one of the Student's t test comparisons involving that particular subset. Venn diagrams were constructed by intersecting the set of genes considered up-regulated in each of the three NK cell groups.

RT-PCR was performed using equal amounts of RNA available from three dNK cell samples and CD56^dim and CD56^bright pNK cells. First strand cDNA was synthesized from 300 ng total RNA using Superscript II Reverse transcriptase for RT-PCR, according to the manufacturer's protocol (Life Technologies). PCR was done with 2 μl cDNA template in a total volume of 50 μl containing 1.5 mM MgCl₂, 10 mM dNTP, 2U Taq DNA polymerase, and 10 μM of each primer in 10× PCR buffer (Taq DNA polymerase PCR kit; QIAGEN). The PCR reaction profile consisted of 1 min at 95°C, followed by 30 cycles of 30 s at 94°C, 30 s at 60°C, and extension for 30 s at 72°C, with a final 5-min extension at 72°C. 10 μl of each product was visualized on 1–2% agarose gels. Primers used were as follows: NKG2E-Fw (5′-AGAATTAAACCTTCAAAATGCTTCT-3′) and NKG2E-Rev (5′-GATTCTTGAAGATCCACACTGGT-3′) for specific amplification of a 553-bp NKG2E product; NKG2C-Fw (5′-GAAGAGAGTTTGCTGGCCTGTA-3′) and NKG2C-Rev (5′-CACTGGGCTGATTTAAGTCGAT-3′) for a 242-bp NKG2C product, β actin–Fw (5′-AACTGGGACGACATGGAGAAAA-3′) and β actin–Rev (5′-TCGGTGAGGATCTTCATGAGGT-3′) for a 353-bp β-actin product; progestagen-associated protein 14 (PP14)/GdA-Fw (5′-AGAGCTCAGAGCCACCCAC-3′) and PP14/GdA-Rev (5′-GTGGGAGTCTGGTCTTCC-3′) for a 544-bp PP14/GdA; and GAL1-Fw (5′-CCCACGGCGACGCCAAC-3′) and GAL1-Rev (5′-AGAGGCAGCTGCCTTTATTG-3′) for a 295-bp galectin-1 product.

The supplemental material, data sets, Tables S1 and S2, and Figs. S1 and S2 are available. The supplemental material provides further details about cRNA preparation and hybridization, and gene classification, as well as explanatory comments for supplemental microarray Data Sets 1–3. Table S1 lists samples used for hybridization to genechips. Table S2 summarizes the number of differentially expressed genes in pairwise comparisons. Fig. S1 supplements Fig. 3. Fig. S2 shows FACS^® and RT-PCR results for relevant markers after applying the procedure used for isolating dNK cells to pNK cells.

CD56^bright dNK cells and the two subsets of pNK cells (CD56^bright and CD56^dim) have been distinguished by only a few differentially expressed surface markers (3). The functional differences among them are still unclear. To further characterize dNK cells in relation to pNK cells and to discover genes with relative overexpression that could provide clues to their function, the gene expression profiles of freshly isolated NK cells from decidual and peripheral blood were compared. Cells obtained as described in Materials and Methods were flow sorted as illustrated in Fig. 1 A. The hybridization results, listing the number of present calls, the average hybridization intensities, and standard deviations for each NK cell subset (nine dNK, five CD56^bright pNK, and five CD56^dim pNK cell samples) and all 12,558 probe sets are available in Data Set S1 .

The 19 samples were organized on the basis of overall similarity in their gene expression patterns by an unsupervised hierarchical clustering algorithm of variable genes (30). A dendrogram, in which the pattern and length of the branches reflect the comparative difference in gene expression profiles between each of the dNK and pNK cell samples is shown in Fig. 1 B. dNK cells are clearly distinct from both subsets of pNK cells, as illustrated by the length of the two terminal branches. Within the group of pNK cells, a secondary branching point separates the CD56^bright and CD56^dim samples. At the transcriptional level, dNK cells differ notably from both subsets of pNK cells and are more distinct than the two peripheral CD56^bright and CD56^dim subsets are from each other. In other words, the two subsets of pNK cells are far more closely related to each other than either is to dNK cells.

Pairwise Student's t test comparisons of the hybridization data with the three NK cell subsets using P ≤ 0.01 subsequently revealed differential expression of 1,785 genes, most of which were found in the comparisons involving dNK cells. Even after adjusting for the uneven sample sizes, the number of genes differentially expressed between the pNK subsets was about three times smaller (Table SII).

To reduce the number of differentially expressed genes to a manageable number, genes with a fold change of three or greater in at least one of the three pairwise comparisons at P ≤ 0.001 were considered. These stringent criteria were met by 278 genes. An image of these genes clustered by similar expression patterns is represented in Fig. 1 C. Overall, ∼70% of the 278 differentially expressed genes were overexpressed in dNK cells (197 genes) as compared with either CD56^bright or CD56^dim pNK cells, whereas roughly 30% of genes were overexpressed in pNK cells versus dNK cells, as illustrated by a Venn diagram (Fig. 2 A).

As an internal control, CD56, a prototypical NK cell marker, was first considered. CD56 (NCAM-1) appeared in the list of 278 genes, and revealed higher intensities in dNK as compared with both pNK subsets. This was also observed by FACS^® analysis (Fig. 1 A and see Fig. 4 A). CD16 identifies CD56^dim NK cells and was absent in CD56^bright pNK and dNK cells (see Fig. 4 B). The latter result was foreseen because cell separation included CD16 expression as one parameter (Fig. 1 A). After classification of the genes into functional categories, the largest number of transcripts enriched in dNK cells were those encoding membrane proteins (Fig. 2 B). Additionally, the 278 differentially expressed genes are shown with fold changes for each comparison in Fig. 3 (also see Fig. S1). Expression of several of these genes was examined by either flow cytometry (Fig. 4) or RT-PCR (Fig. 5).

Among the 278 differentially expressed genes, a large number of genes were surface molecules and receptors (Fig. 2 B and Fig. 3 A). Among the genes in this category, three members of the transmembrane 4 superfamily (CD151, CD9, and tetraspan-5; also known as tetraspanins) were exclusively expressed in dNK cells. Two integrin subunits, the αD (ITGAD;CD11d) and the β5 (ITGB5) subunits were also highly overexpressed by dNK. Other tetraspanins (CD53 and CD63) and integrin subunits (β1 [ITGB1] and αx [ITGAX;CD11c]) were relatively overexpressed in dNK cells in relation to at least one of the peripheral subsets. Integrins are involved in cell–cell and cell–extracellular matrix interactions and influence diverse processes such as cell adhesion, migration, activation, and interaction with target cells. Many of the tetraspanins may facilitate these processes by assembling molecules into membrane clusters, distinct from lipid rafts (e.g., by participating in functional signaling complexes with other integrins, lineage-markers, or other tetraspanins; references 31–36). The overexpression of several of these proteins in dNK cells may be related to the mechanisms used by dNK cells for migration to, retention within, and/or their effector function within the pregnant uterine mucosa. FACS^® analysis confirmed surface expression of both CD9 and CD151. CD9 is exclusively expressed on dNK cells and emerges as a specific marker for these cells ex vivo (Fig. 4, C and D).

Interestingly, a ligand for murine CD9, murine pregnancy-specific glycoprotein (PSG) 17, was recently identified (37). This prompts the interesting hypothesis that dNK cells, through their CD9 receptor, may interact with placental PSGs and consequently produce cytokines that support successful pregnancy. PSGs are a family of highly similar, placentally secreted proteins, originally isolated from the circulation of pregnant women. In humans, PSG concentration in the bloodstream increases exponentially until term (38) and low PSG levels are associated with recurrent spontaneous abortions (39). Administration of anti-PSG Abs induced spontaneous abortion in primates, indicating that PSGs are essential for successful pregnancy (40). In addition, PSGs may induce production of antiinflammatory cytokines by monocytes (41).

Of particular interest, NKG2C and NKG2E genes both were highly significantly overexpressed in dNK cells. Both showed increased hybridization in the three- to fivefold range (Fig. 3 A and Fig. 5). The specificity of this result was confirmed by RT-PCR. NKG2C and NKG2E are CD94-associated members of the C-type lectin like receptors with presumed activating functions. The CD94/NKG2C heterodimer associates with DAP12 and binds to HLA-E (42, 43). Although the resemblance of NKG2E to NKG2C in terms of an ITIM-lacking cytoplasmic domain (both containing a Lys residue in the transmembrane region that may lead to association with DAP12) suggests an activating function, the ligand and functions of NKG2E have not been described previously (44). Moreover, detection of NKG2E protein has not been reported.

The CD94-associated lectinlike inhibitory receptor NKG2A, which bears two ITIMs in the cytoplasmic domain and for which the ligand is HLA-E, was not represented on the chip because the polymorphism and isoforms of this gene makes it difficult to select specific probes. Its expression on the three NK cell subsets in association with CD94 was detected by FACS^® analysis using the NKG2A-specific mAb Z199 and anti-CD94. CD94 was not differentially expressed in the chip analyses (unpublished data). The manner in which signals mediated through activating and inhibitory receptors on the same cell are integrated remains to be explored.

Ly-49L (Fig. 3 A, KLRA1) was present and overexpressed in nearly all of the dNK cells but absent from both sets of pNK cells (Fig. 3 A, P-calls). This unusual transcript is derived from a gene that encodes a member of the lectin superfamily closely related to the large family of murine Ly-49 genes. The single human gene has been found expressed only as transcripts that would lead to a truncated protein, due to a single nucleotide substitution preceding a splice donor site in Exon 5 (45, 46). However, the full-length transcript was found in both baboons and the orangutan (47, 48). It remains to be seen whether the full-length transcript and the corresponding protein can be produced in human materials, particularly in dNK cells, and if so, what its function might be.

All of the KIR genes for which probes were present on the genechip (KIR3DL1, KIR3DL2, KIR2DL3, and KIR2DL4) were overexpressed by dNK cells (Fig. 3 A). KIRs have been known to be expressed on both dNK and CD56^dim pNK cells, but not on CD56 ^bright pNK cells. Staining for KIRs using the available mAbs CD158a (unpublished data), CD158b, and NKB1 (Fig. 4, E and F) consistently showed a much higher proportion of positive cells in dNK cells than in CD56^dim pNK cells, in keeping with the increased transcript levels found on the average in dNK cells. These data are in accordance with previously published data on KIR expression in peripheral and decidual blood from the same individual (22).

KIR2DL4 expression has been reported on both dNK and pNK cells (49, 50). Although it has been suggested that it is the HLA-G receptor, other recent data do not support this conclusion (51, 52). The present findings indicate that KIR2DL4 is transcribed in dNK cells to a higher degree than in CD56^dim or CD56^bright pNK cells (Fig. 3 A). Using anti-KIR2DL4–specific antibodies (27), the surface expression of this receptor protein was found in 5 out of 11 tested dNK cell samples, but was not detectable on pNK cells in 5 tested samples (Fig. 4 G). In contrast with many of the other KIRs that are variably present in individual genomes, KIR2DL4 is found in all haplotypes. Yet, a single base deletion in the KIR2DL4 gene, which predicts generation of a molecule with a shortened cytoplasmic tail (53) and possibly a protein which is ultimately not surface expressed, occurs. A considerable number of individuals are homozygous for this deletion (54), which may partly explain the lack of reactivity seen with the anti-KIR2DL4 mAb.

Although galectin-1 is expressed by all NK cell types, its transcription level is greatly elevated in dNK cells relative to the other two subsets (Fig. 3 B), a result that was confirmed by RT-PCR (Fig. 5). Galectin-1 is a member of a small family of carbohydrate-binding proteins (galectins 1–14; reference 55). Although galectins recognize N-acetyl-lactosamine (galactosyl-N-acetylglucosamine), they have a higher affinity for polylactosamine moieties present in N-glycosylated and core 2 O-glycosylated glycoproteins. Two enzymes are essential for the presence of these moieties on N- or O-glycan structures. For N-glycans, Mgat 5 initiates a fourth branch to form a tetraantenary glycan, which is elongated with polylactosamine groups. For O-glycans, C2GNT initiates a second branch (core 2) on O-glycans that is also further elongated with polylactosamine residues (56, 57). Divalent galectin-1 homodimers binding to cell surface O- and N-glycosylated glycoproteins mediate cell–cell interactions, and generate lattices that can segregate, group, or affect clustering of cell surface glycoproteins (57). Microarray expression profiles show that Mgat5 is not expressed on any of the three subsets of NK cells studied here. Strikingly, however, C2GNT required for core 2 O-glycan synthesis is expressed on dNK cells but on neither of the two subsets of pNK cells (Fig. 3 B).

Galectins have been implicated in immune maturation and modulation through various mechanisms (55). Mgat5 expression in T cells negatively regulates T cell activation and autoimmunity. Galectin binding restricts TCR recruitment to the synapse by formation of a lattice resulting from its binding to tetraantennary N-glycosylated glycoproteins (58). Most recently, galectin-1 has been implicated in pre–B cell and bone marrow stromal cell synapse formation as a stromal cell ligand of the pre–B cell receptor (59). Core 2 O-glycans are expressed by CD4 + CD8 + T cells in a regulated manner during intrathymic maturation (56). Galectin-1 mediates thymocyte binding to stromal cells. It also induces apoptosis of immature CD8^low CD4^low thymocytes and peripheral activated T cells through galectin-1–mediated segregation of O-glycosylated CD45 from CD7 and CD43. This process is dependent on C2GNT expression (57, 60). Other effects of galectin-1 on T cells have been reported previously (55). Two hypotheses are suggested by these data as follows: (a) galectin-1 and C2GNT, the enzyme required to initiate formation of its ligand, are expressed by NK cells in the decidua because they are needed to down-modulate cytotoxic activity; and (b) galectin-1 present on trophoblasts in the placental bed (61) may originate from dNK cells and may function on the trophoblast to down-modulate cytotoxicity.

PP14 was uniquely expressed in dNK cells (Fig. 5). Although its expression level was variable (Fig. 5) and, therefore, did not reach the statistical significance level required for inclusion in Fig. 3, it was present and highly expressed in seven of the nine dNK samples but in neither of the pNK cell subsets. Its differential expression was confirmed by RT-PCR. PP14 is a 28-kD glycoprotein with a unique carbohydrate configuration that is very unusual in mammals. Although it was first isolated from human placenta, it was later shown to be synthesized by endometrial glands and gestational deciduas, with the highest production at 10–14 wk of gestation (62). PP14 has been reported to inhibit T lymphocyte proliferative responses and to negatively regulate T cells by localizing to the APC–T cell contact site and elevating activation thresholds (63). It was also reported to inhibit K562 lysis by pNK cells (64).

Granularity is a characteristic of both dNK and CD56^dim pNK cells, both of which are also called large granular lymphocytes (65), whereas CD56^bright pNK cells are known to be less granular (15). Granzyme B and perforin (components of cytotoxic granules) were expressed at similar levels in dNK and CD56^dim pNK cells, whereas granzyme A was 3- and 10-fold overexpressed in dNK cells versus the CD56^dim and CD56^bright pNK cells, respectively (Fig. 3 C). Thus, the potential cytolytic function of dNK cells, probably regulated, cannot be ignored. This nascent activity could be manifest in the decidua under special circumstances (e.g., virus infection of the trophoblast or decidual remodeling).

Many other genes overexpressed in dNK cells that are presented in Fig. 3 (e.g., CD31, CD38, MIP-1 α/β, Syk, LEF1, AARS, arp2/3 subunit, and tubulins) may be of considerable interest but are beyond the scope of the present paper. Likewise, genes that are absent or have lower expression in dNK cells in relation to pNK cells may be relevant to dNK or pNK cell function and interesting candidates for further study. For instance, toll-like receptor 1 (Fig. 3 A) and SH2D1A (Fig. S1 A) are expressed in both pNK subsets but not in dNK cells. CD6 is expressed only in CD56^dim pNK cells, whereas the receptor activator of NF-κB, a new member of the TNF superfamily of receptors, transcripts are found only in CD56^bright pNK cells (Fig. 3 A). Both may be relevant to interactions between these and other cells (66, 67). Finally, functions other than a role in fetal-maternal tolerance have been suggested for dNK cells. In particular, roles for them in placental development and in vascular remodeling of uterine spiral arteries have been suggested in the mouse (11, 66, 67). dNK cells may have different functions in response to novel ligands in distinct uterine microenvironments. The data presented in Fig. S1, as well as that included in Fig. 3 (A–C), should be particularly useful to those interested in these and other phenomena associated with pregnancy.

It is worthwhile to ask whether the overexpression of a large number of genes in dNK cells may have been due to activation by the procedure used to isolate them (i.e., disaggregation with collagenase and DNase, incubation to allow adhesion of stromal cells and macrophages to plastic and Ficoll density gradient separation, a procedure requiring 4–6 h [Materials and Methods]). The procedure has been applied previously to isolate lymphocytes from other organs such as pancreas, liver, and intestines without apparent activating effects. Nevertheless, pNK cells were put through a mock isolation procedure and the markers that distinguished dNK cells from their blood counterparts were analyzed. The mock isolation was performed with and without the addition of human decidual stroma.

FACS^® and PCR analysis of the most relevant markers (Figs. 3–5; CD69, a marker for activation; CD9, a characteristic marker of dNK cells as shown in this paper; CD16, a specific marker of CD56^dim pNK cells; and PP14, a protein exclusively expressed by dNK cells) was performed. No changes in the expression of these markers on CD56^dim or CD56^bright pNK cells were observed when the isolation procedure was performed in the absence or in the presence of decidual stroma (Fig. S2). Nevertheless, it cannot be excluded that changes in a fraction of the 197 up-regulated genes in dNK cells may have been the result of activation during the isolation procedure.

The present paper has provided a detailed characterization of gene expression in the different NK cell subsets present in peripheral blood and early pregnancy decidual tissues. First, dNK cells are remarkably different from either subset of pNK cells; CD56^bright pNK cells are far more similar to the CD56^dim pNK cell subsets than they are to the CD56^bright dNK cells. Second, the majority of differentially expressed genes are overexpressed by dNK cells. In contrast, the relatively small number of differentially expressed genes in the pNK cell comparison are equally distributed between those that were overexpressed in CD56^dim versus CD56^bright pNK cells, and those overexpressed in CD56^bright versus CD56^dim pNK cells. The observed differences in gene expression may be explained by two alternative hypotheses as follows: (a) dNK cells represent a distinct lineage of NK cells, possibly arising from a distinct hematopoietic precursor; or (b) the differences reflect a maturation of pNK cells (most likely the CD56^bright subset) in the decidual microenvironment. Additionally, the nature of the precursors and sites of differentiation, of dNK cells, and, in particular, the effect of hormones and decidual stroma (whose effect on differentiation have been scarcely studied) are of considerable interest. Special attention must be paid to general effects common to many stroma and possible specific effects that may lead to expression of dNK cell–specific molecules. Third, and most interestingly, genes that provide exciting new venues for research were overexpressed in dNK cells. These include CD9 and other tetraspanins (at least four integrin subunits: NKG2C and NKG2E, Ly-49L, KIRs, galectin-1, and PP14), all of which could have immunomodulatory functions during pregnancy. The data strongly suggest a distinct function for dNK cells in pregnancy (i.e., a role in maternal-fetal tolerance).

We thank J. Couget and colleagues for advice regarding the Genechip experiments. M. Handley, H. Levine, B. Nostrom, and G. Kenty are acknowledged for flow sorting. We appreciate the contributions of W. Beekhuizen. We are grateful for the help of A. Butte and I. Kohane in establishing the data analysis. We thank S. Choe for advice, and R. Vance for reading the manuscript.

This research was supported by the National Institutes of Health (grants CA47554 and AI053330) and by a travel/research grant of the Stichting Dr. C. van Tussenbroek to L.A. Koopman.