Novel Insights on Human NK Cells’ Immunological Modalities Revealed by Gene Expression Profiling

Preparation of NK subset samples

PBLs were isolated from different healthy donors using Ficoll gradients. Isolation of NK cells was performed using the NK Isolation kit II (Miltenyi Biotec, Auburn, CA), according to manufacturer’s instructions. A total of 100 × 10⁶ cells enriched for CD56⁺CD3⁻ NK cells from each donor were subsequently sorted as described below to purify CD56^dimCD16⁺CD3⁻ and CD56^brightCD16⁻CD3⁻ subsets (see Fig. 1⇓A). Purified cells were pooled after isolation and mRNA was extracted immediately and stored at −70°C. Bulk polyclonal in vitro-activated NK cultures were prepared from CD56^dimCD16⁺CD3⁻ sorted cells from nine donors and were grown on feeder cells in medium supplemented with human IL-2 and PHA as previously described (2). Activated NK cells were only used after repurification with the NK Isolation kit II to ensure the elimination of any remnants of irradiated feeder cells (see Fig. 1⇓B).

• Download figure
• Open in new tab
• Download powerpoint

FIGURE 1.

Global gene expression analysis measured in purified NK subsets. A, Sorting gates applied for purifying NK subsets from pbNK-enriched fractions. After exclusion of CD3⁺ cells, CD56^brightCD16⁻ and CD56^dimCD16⁺ subsets were gated. Representative gating conditions on one donor’s cells (of nine performed) is shown. B, Purity of analyzed activated CD16⁺ (CD56⁺CD16⁺) NK cells was evaluated by flow cytometry staining. High quality of total RNA (C) and cRNA (D) prepared from pooled NK subset samples, together with fragmentation of biotinylated cRNA to uniform size (E), were verified by using an Agilent Bioanalyzer. F–H, Reproducibility of normalized gene expression data was assessed by plotting expression levels from two replicate experiments (A and B) performed on each NK subset. Points falling outside of the lines had expression levels >2-fold different in one replicate compared with the other. Linear correlation coefficient values (R²) are indicated for each analysis.

Flow cytometry and cell sorting

The following mouse anti-human mAbs, purified or directly conjugated with FITC, PE, or CyChrome C, were used in flow cytometric analysis: anti-CD56, anti-CD9 (DAKO, Carpinteria, CA); anti-CD16, anti-CD3, anti-CD53, anti-CD81, anti-HLA DR, DP, DQ, anti-CD44, anti-CD62L, anti-galectin 3, anti-CD86, anti-CD70 (all obtained from BD Biosciences, Mountain View, CA); anti-IL-7R α-chain (CD127), anti-NKG2D, anti-NKp30, anti-galectin 1 (R&D Systems, Minneapolis, MN); anti-NKp46 (12); anti-CD3ζ (cross-reactive to human (13)); anti-OX40L (MBL, Nagoya, Japan). For staining and cell sorting, cells were washed in PBS supplemented with 2% FCS and incubated with mAb on ice for 30 min, followed by washing twice. Cell sorting and fluorescence measurements were performed on a MoFlo High Performance Cell Sorter (DakoCytomation, Glostrup, Denmark). Data from single cell events were collected using a standard FACSCalibur flow cytometer (BD Immunocytometry Systems, Mountain View, CA). Data were analyzed using CellQuest (BD Biosciences). The permeabilization and intracellular staining for galectin1, galectin3, and CD3ζ were performed using the Cytofix/Cytoperm Plus kit (BD Pharmingen, San Diego, CA) according to the manufacturer’s instruction.

Preparation of labeled RNA and microarray hybridization

Total cellular RNA from the NK subsets pooled from different donors (CD56^brightCD16⁻, CD56^dimCD16⁺, and in vitro-activated CD16⁺ NK subsets) was isolated and subjected to a cleanup protocol with RNeasy Mini kits (Qiagen, Valencia, CA), according to the manufacturer’s specifications. The quality of total RNA was assessed using an Agilent Bioanalyzer (see Fig. 1⇑C). First and second strand cDNA was prepared from a 0.5-μg RNA template (see Fig. 1⇑D), and the cDNA was subjected to in vitro transcription in the presence of biotinylated nucleoside triphosphates. The biotinylated cRNA was fragmented to uniform sizes (∼100 nt as verified in Fig. 1⇑E). CodeLink Uniset Human 20K I Bioarrays (Amersham Biosciences, Piscataway, NJ) were hybridized with each prepared cRNA target in duplicates and stained with Cy5-streptavidin and subsequently washed according to manufacturer’s instructions, and then scanned on an Axon GenePix 4000B scanner (Axon Instruments, Foster City, CA). Raw data files were obtained after analysis of scanned images with CodeLink expression software (Amersham Biosciences). The complete microarray data is deposited at the National Center for Biotechnology Information’s (NCBI) Gene Expression Omnibus (GEO) under entry names: GSM26200-26205.

Microarray data analysis

The global normalization (total intensity normalization) method was used for the data obtained from the Amersham array. The mean intensities were calculated for each array (Mi, i = 1, 2, … . , n) and used to calculate the mean intensity across all arrays (Ma). The scale factors (Fi, i = 1,2, … . , n) were calculated for each array as Ma/Mi. The normalized intensities for each array were calculated by multiplying scale factors by their measured intensities. This classical normalization method scales the individual intensities so that the mean intensities are the same across all arrays, and were applied to the data by an in-house developed program. Transcripts that demonstrated differences >2-fold in duplicate analysis of the same NK sample examined were excluded from the analysis. The raw intensity data were normalized after exclusion of such genes. Additionally, 20 was set to be the minimal normalized hybridization intensity value (in other words, all intensities below 20 were brought up to this value) after observing that a set of more than 10 transcripts (e.g., CD20, CD4, CTLA4, etc.), known to be absent in NK cells, showed hybridization intensity values below this threshold.

When examining the reproducibility of normalized duplicates for each of the three NK samples, there was a significant correlation (R² > 0.98) (Fig. 1⇑, F–H). Additionally, examination of genes up- or down-regulated by at least 2-fold between any two normalized data sets from different NK samples always showed a statistically significant correlation in 12 possible combinations examined (R² > 0.992, data not shown). This confirmed the reproducibility of the 2-fold change criteria between different NK subsets. This reproducibility allowed normalized values for each replicate to be averaged. Genes were classified as up- or down-regulated if they demonstrated at least a 2-fold alteration in expression level.

For Venn diagrams, three paired comparisons were made between NK subsets: 1) CD56^brightCD16⁻ vs CD56^dimCD16⁺; 2) activated CD16⁺ NK vs CD56^dimCD16⁺; 3) activated CD16⁺ NK vs CD56^brightCD16⁻. Venn diagrams were constructed by intersecting the set of genes up- or down-regulated by 2-fold or greater in each of the three comparisons. Gene functional classification was based on information available in the Gene Ontology Consortium functional annotations (www.geneontology.org), Locuslink (www.ncbi.nlm.nih.gov/LocusLink), OMIM (www.ncbi.nlm.nih.gov/entrez/query.fcgi?db-OMIM), and PubMed (www.ncbi.nlm.nih.gov./entrez/query.fcgi) databases.

Cytotoxicity assay

Purified unactivated NK subsets were sorted as described above. Activated CD16⁺ and CD16⁻ NK cells were isolated by the use of anti-CD16 microbeads (Miltenyi Biotec) on bulk polyclonal-activated CD56⁺CD3⁻ NK cells. The cytotoxic activity of various NK subsets against the 721.221 EBV-transformed B cell line was measured after labeling these cells overnight with [³⁵S]methionine. After labeling, target cells were washed, and 5 × 10³ labeled target cells were incubated at various E:T ratios. The killing rate was calculated as percent [³⁵S]methionine release = (cpm sample − cpm spontaneous release)/(cpm total − cpm spontaneous release) × 100. Total [³⁵S]methionine release was measured after incubation of the cells with 0.1 M NaOH. In these cytotoxic assays, the spontaneous release was <25% of maximal release.

T cell costimulation assay

T cell proliferation assays were performed in 96-well flat-bottom plates in triplicate in a final volume of 250 μl of RPMI 1640 + 10% human serum. Plates were coated with a suboptimal concentration of anti-human CD3 mAb (clone T3D, obtained from the American Type Culture Collection (Manassas, VA)) (0.1 μg/ml in PBS incubated overnight at 4°C). Plates were washed twice with RPMI 1640 before incubating with 25 × 10³ irradiated (5000 rad) activated CD16⁺ NK or unactivated polyclonal CD56⁺CD3⁻ NK cells, combined with 50 × 10³ autologous CD3⁺ T cells (Pan CD3⁺ T Cell Isolation kit; Miltenyi Biotec) for 72 h. [³H]Thymidine (1 μCi/well) was added for the last 24 h of the assays and cells were harvested to quantitate [³H] incorporation into DNA. Supernatants from proliferation experiments were used for measuring production of human IL-2 by ELISA using the Quantikine kit (R&D Systems).

Jurkat cell apoptosis assay

A total of 1 × 10⁶ unactivated CD56⁺CD3⁻ NK cells were grown for 7 days at a concentration of 0.5 × 10⁶/ml in RPMI 1640 medium + 10% heat inactivated FCS with or without the addition of 100 U/ml IL-2 and 1 ng/ml PHA. Supernatants of these cultures (together with control media containing the same concentrations of IL-2 and PHA) were added to Jurkat cells (1 × 10⁶/ml) for 8 h at 37°C. Induced cell death was measured using the Annexin V Apoptosis kit (Molecular Probes, Eugene, OR) according to manufacturer’s instructions.

Microarray gene expression analysis of human CD56^brightCD16⁻, CD56^dimCD16⁺, and activated CD16⁺ NK cells

The initial aim of this study was to investigate the gene expression profile of ∼20,000 genes in human CD56^dimCD16⁺ NK, CD56^brightCD16⁻, and in vitro-activated CD16⁺ NK cells. These subsets were purified from nine different healthy donors (Fig. 1⇑, A and B). All samples were re-examined by flow cytometry following sorting, to ensure purity above 99% (data not shown). Subsequently, each purified subset from all donors was pooled and used to obtain a total RNA extract. The quality of total RNA used, and the cRNA that was prepared afterward, was verified by using the Agilent Bioanalyzer (Fig. 1⇑, C and D, respectively). Additionally, we confirmed that the cRNA samples were fragmented to uniform size (Fig. 1⇑E) before continuing with sample labeling and hybridization procedures. Gene expression profiling was performed on these subsets by using CodeLink Uniset Human 20K I Bioarray. The labeled cRNA samples for the three NK subsets were hybridized in duplicate to enhance the significance of the hybridization intensities and to eliminate random noise especially from probe sets at the borderline of expression. Reproducibility of the normalized gene expression data was assessed by plotting gene expression levels from each of the two replicate experiments (Fig. 1⇑, F–H). Linear correlation coefficient for all plots was above 0.98, and <0.01% of the genes examined displayed >2-fold variation in all three comparisons. This observation provided a justification to average normalized hybridization intensity values for each gene in replicate experiments, and to use these values for estimating gene expression patterns. The online version of this article contains supplemental Table I that lists the averaged normalized hybridization values for the three NK subsets studied.⁸

A list of differentially expressed genes among NK cell subsets was compiled using an algorithm that identifies significant differences with minimal false positives (as estimated from an analysis of marker genes that were included or excluded using different thresholds as described in Materials and Methods). The NK subsets were organized based on the overall similarity in gene expression patterns by an unsupervised hierarchical clustering algorithm of variable genes that showed more than a relative 4-fold change in expression in at least one of the examined samples. A dendogram, in which the pattern of length of the branches reflects the comparative difference in gene expression profiles between each of the NK samples, is shown in Fig. 2⇓A. Interestingly, activated CD16⁺ NK cells were strikingly distinct from both subsets of unactivated pbNK, as illustrated by the length of the two terminal branches. Within the group of unactivated pbNK cells, a secondary branching point separated CD56^brightCD16⁻ and CD56^dimCD16⁺ duplicates from each other, thus confirming the fact that these two subsets are distinct. In other words, the gene expression profiles of the two subsets of unactivated NK cells were far more closely related to each other than either was to activated CD16⁺ NK cells (Fig. 2⇓A).

• Download figure
• Open in new tab
• Download powerpoint

FIGURE 2.

CD56^brightCD16⁻, CD56^dimCD16⁺, and activated CD16 cells represent three different NK cell subsets. A, Unsupervised hierarchical clustering of NK samples based on expression profile of genes with variable expression levels across all subset samples. Normalized data for genes that were up- or down-regulated by at least 4-fold in at least two of the samples were filtered and log transformed before cluster analysis. Mean levels for each gene across all samples were calculated and the magnitude of relative expression of a particular gene relative to the calculated mean expression was reflected by use of color representation. Brighter red means higher expression, brighter green means lower expression, and black means average intensity across samples. The organization and length of the branches in the resulting dendogram reflect the similarity in gene expression profiles between each of the samples. B, Expression levels for selected genes whose expression has been extensively characterized on NK subsets. C and D, Venn diagrams generated by the intersection of the list of genes up-regulated (C) or down-regulated (D) by at least 2-fold (based on selection criteria described in Materials and Methods) in at least one of the three comparisons performed: 1) CD56^brightCD16⁻ vs CD56^dimCD16⁺; 2) activated CD16⁺ NK vs CD56^dimCD16⁺; 3) activated CD16⁺ NK vs CD56^brightCD16⁻. For example, a total of 380 genes were overexpressed in CD56^brightCD16⁻ when compared with CD56^dimCD16⁺ (C), 216 of which were overexpressed in this paired comparison only, and not when activated CD16⁺ NK cells were compared with any of the two peripheral blood-derived unactivated NK subsets. Thirty-two genes were commonly up-regulated in all three comparisons performed. One-hundred thirty-two transcripts were up-regulated in CD56^brightCD16⁻ and activated CD16⁺ NK vs CD56^dimCD16⁺ NK, while no genes were overexpressed in both CD56^dimCD16⁺ and activated CD16⁺ NK vs CD56^brightCD16⁻ NK.

To confirm the validity of our data and correlate with gene expression data in the literature, we compiled a list of hybridization intensity values of selected marker genes expected to be absent or present in the NK subset in question (Fig. 2⇑B). All samples did not contain transcripts specific for monocytes, B, or T cells (CD14, CD22, CD28, and CD4), emphasizing the purity of the cells used in the study. In correlation with our sorting criteria, CD16 mRNA expression was not detected in the CD56^brightCD16⁻ subset, while CD56 mRNA expression was significantly higher in this subset compared with CD56^dimCD16⁺. In agreement with previous observations, inhibitory receptor KIR3DL2 mRNA was preferentially expressed in unactivated CD56^dimCD16⁺ NK cells (9), while activating receptor KIR2DL4 mRNA was specifically transcribed in unactivated CD56^brightCD16⁻ and in vitro-activated CD16⁺ NK, but not unactivated CD56^dimCD16⁺ NK (14). Our results also confirm L-selectin mRNA-specific expression on CD56^brightCD16⁻ NK cells (6), preferential expression of granzyme B in CD56^dimCD16⁺ NK cells regardless of their activation state (9), and specific expression of the NK lysis receptor NKp44 transcript following activation (15). Finally, it was important to assess whether protein expression of the corresponding gene products of interest correlated with the transcriptional differences in these genes. As shown in Fig. 3⇓, detection of several membrane (Fig. 3⇓, A–I) and secreted proteins (Fig. 3⇓, J–K) by flow cytometry analysis, correlated with normalized hybridization intensities obtained for the same genes from our gene array analysis. Overall, the concordance between the hybridization intensity values of the various marker genes and published gene expression patterns in NK subsets validates our normalization methods and the quality of NK samples used in this study.

• Download figure
• Open in new tab
• Download powerpoint

FIGURE 3.

Confirmation of expression patterns of selected differentially expressed genes at the protein level. CD56^dimCD16⁺, CD56^brightCD16⁻, and activated CD16⁺ NK were sorted and purified as described in Materials and Methods. Each data set (A–K) shows flow cytometry staining for different molecules (light gray) compared with background staining (empty histograms). Dashed lines were added to assist in demonstrating the observed differences in protein expression. Mean fluorescence intensities (MFI) for each sample are indicated in the right upper corner of the panels. MFIs for background stainings were between 2 and 4. Bar graph above each histogram data set represents the normalized hybridization intensity of a probe set specific for the same gene. Data shown are representative stainings obtained from three to five different donors.

Comparison of gene expression between unactivated CD56^brightCD16⁻ and CD56^dimCD16⁺ NK cells

To begin the comparison between the various NK subsets, a list was generated of differentially expressed genes by at least 2-fold between: 1) CD56^brightCD16⁻ vs CD56^dimCD16⁺; 2) activated CD16⁺ NK vs CD56^dimCD16⁺; 3) activated CD16⁺ NK vs CD56^brightCD16⁻. The online version of this article contains supplemental Table II that lists all genes and their changes in expression calculated from the three comparisons mentioned above. Pairwise comparison of CD56^brightCD16⁻ and CD56^dimCD16⁺ revealed that, quantitatively, the difference in gene expression between these subsets (Fig. 2⇑, C and D) was largely due to down-regulation of genes in CD56^brightCD16⁻, rather than up-regulation, as illustrated by a Venn diagram (Fig. 2⇑, C and D). There were 888 genes transcribed at significantly lower levels in the CD56^brightCD16⁻ subset, while 380 genes were specifically up-regulated. Interestingly, CD56^brightCD16⁻ cells seemed to have gene expression similarities with the activated CD16⁺ NK subset. One-hundred-thirty two genes of the total 380 up-regulated genes in CD56^brightCD16⁻ NK vs CD56^dimCD16⁺ were also up-regulated, mostly at higher fold changes, in activated CD16⁺ NK when compared with CD56^dimCD16⁺ (Fig. 2⇑C). In comparison, there were no genes that were overexpressed in both CD56^dimCD16⁺ and activated CD16⁺ NK compared with CD56^brightCD16⁻ NK. A similar pattern was observed for down-regulated genes (Fig. 2⇑D). Taken together, this analysis suggests that the unactivated low cytotoxic CD56^brightCD16⁻ NK subset has increased expressed levels of a number of activation-induced markers that can also be found on activated CD16⁺ NK. After classification of selected genes into functional categories (Fig. 4⇓⇓), differentially expressed genes were evaluated to determine whether these genes could potentially contribute to the functional differences between these subsets and/or provide new insights on NK cell behavior.

• Download figure
• Open in new tab
• Download powerpoint

FIGURE 4.

Gene expression patterns in human CD56^dimCD16⁺ and CD56^brightCD16⁻ NK subsets. Gene expression profiles of selected genes differentially expressed by at least 2-fold between CD56^dimCD16⁺ and CD56^brightCD16⁻ subset. Bars represent fold change of the mRNA level of a particular gene when comparing these subpopulations. Positive values indicate that the transcript was more abundant in the CD56^brightCD16⁻ subpopulation and negative values indicate the opposite. Genes were grouped according to their presumed function (A–G) based on information available in public databases or in the literature.(Figure continues)

• Download figure
• Open in new tab
• Download powerpoint

FIGURE 4A.

(Continued)

CD56^brightCD16⁻ vs CD56^dimCD16⁺ NK cells: molecules involved in regulation of cytotoxicity

Previous studies of unactivated CD56^dimCD16⁺ NK revealed that these cells are naturally more cytotoxic than unactivated CD56^brightCD16⁻ NK, due to higher levels of granzyme and perforin expression (4, 9). Similarly, our analysis shows that effector cytotoxic molecules granzyme B and CTLA-1 were expressed at lower levels in the CD56^brightCD16⁻ subset (by ∼2.2- and ∼2-fold, respectively) (Fig. 4⇑⇑C), while differences in granzymes A and M were not statistically significant. Surprisingly however, granzyme K, that also has the ability to lyse target cells, was robustly up-regulated in CD56^brightCD16⁻ NK (by ∼20-fold) (Fig. 4⇑⇑C). This observation, together with the fact that CD56^brightCD16⁻ cells express adequate levels of other granzyme subtypes, led us to investigate additional possibilities for their low cytotoxic ability. We examined whether expression of adaptor proteins associated with NCR and NKG2D activating receptors (such as DAP10, KARAP/DAP12, FCεR1γ, and CD3ζ chain) could provide a complementing explanation for this phenomenon. These molecules contain ITAM or YxxM motifs in their cytoplasmic domains and are known to facilitate NK activation signals (16). The CD3ζ chain was down-regulated by >3-fold in CD56^brightCD16⁻ NK (Fig. 4⇑⇑B). However, no significant differences in expression of other adaptor molecules were detected. The detection of CD3ζ protein by intracellular flow cytometric staining consistently showed ∼4-fold down-regulation in CD56^brightCD16⁻ NK in all donors tested (Fig. 5⇓A). Recent evidence shows that the generally impaired T cell function in chronic inflammation results from down-regulation of the CD3ζ chain in vivo (while other TCR complex subunits remain preserved), thus reflecting the impact of this adaptor molecule on T cell functionality (13). As CD3ζ is the main signaling adaptor molecule for NKp30 and NKp46 cytotoxicity receptors (17), the relative deficiency in this molecule in CD56^brightCD16⁻ NK suggests an additional explanation for the lower cytotoxic capabilities of this subset.

• Download figure
• Open in new tab
• Download powerpoint

FIGURE 5.

Correlation between CD3ζ chain and NKp30 expression levels and CD16⁻ NK cell cytotoxicity. A, Staining of isolated unactivated polyclonal pbNK cells for the CD3ζ chain (intracellular staining) and surface expression of NKp30, NKp46, and NKG2D receptor. Staining for CD56 was used as a second marker to discriminate between CD56^bright and CD56^dim subsets. Mean fluorescence intensities for each subset are indicated. B, Staining of the activated polyclonal CD56⁺CD3⁻ NK line for the CD3ζ chain (intracellular staining) and the NKp30 receptor. Staining for CD16 was used to differentiate between CD16⁺ and CD16⁻ activated NK subsets. C, Killing of 721.221 cells by various purified NK subsets at various E:T ratios. Data are representative of three independent experiments.

When examining differences in surface abundance of activating receptors on pbNK subsets, NKp30 was the only receptor found at lower levels on the CD56^brightCD16⁻ NK subset (probes not represented on the gene microarray used), while NKp46 and NKG2D surface expression on this subset was preserved (Fig. 5⇑A). The observation that the impaired cytotoxic ability of CD56^brightCD16⁻ NK cells was abolished following IL-2 activation and correlates with the reconstitution of both CD3ζ and NKp30 on activated CD16⁻ NK cells (Fig. 5⇑, B and C) further strengthens our hypothesis regarding the involvement of these molecules in controlling CD56^brightCD16⁻ NK cytotoxicity. A pattern of enrichment was also observed for cytotoxicity costimulatory (enhancing) receptors on CD56^dimCD16⁺ NK. NK, T, and B cell Ag and CD58 (both belong to CD2 family), LLT1- lectin-like NK receptor (a member of NK gene complex), and transmembrane protein 2 (TMEM2), all characterized as enhancers of NK cytotoxicity upon ligation (18, 19, 20), were up-regulated on CD56^dimCD16⁺ NK (Fig. 4⇑⇑A). CD2 was the only costimulatory receptor expressed at higher levels on the CD56^brightCD16⁻ NK subpopulation. To conclude this aspect of the analysis, our data indicate that suppression and regulation of CD56^brightCD16⁻ NK killing might be a result of complex mechanisms in different hierarchies of the cytotoxicity activation cascade that complement each other. These mechanisms involve down-regulation of certain activating (NKp30) and costimulatory receptors, key adaptor molecules (CD3ζ), and cytotoxicity effector-secreted proteins (perforin, granzyme B, and CTLA1 molecules).

CD56^brightCD16⁻ vs CD56^dimCD16⁺ NK cells: adhesion molecules

The complex process of lymphocyte trafficking is dictated via interactions between cell surface adhesion molecules and their ligands. Specific expression of L-selectin on CD56^brightCD16⁻ NK has provided a partial explanation for the ability of this subset to traffic to secondary lymphoid organs (6). Our analysis confirms this observation (Figs. 3⇑A and 4⇑⇑A) and shows additional differences in adhesion molecule expression. CD58 (LFA-3), ICAM2, and integrin α_E were up-regulated in the CD56^dimCD16⁺ subset, while integrin α₅, integrin α_M, integrin α_X, ICAM3, and CD44 were up-regulated in the CD56^brightCD16⁻ NK subset (Fig. 4⇑⇑A).

The CD44 protein has been implicated in lymphocyte trafficking to lymph nodes and enhancing IFN-γ secretion (21, 22), two modalities that CD56^brightCD16⁻ NK efficiently perform. The CD56^brightCD16⁻ NK subset expresses CD44 in higher levels both at the transcript level and at the cell surface (by ∼3.75- and ∼8-fold, respectively) (Figs. 4⇑⇑A and 3⇑B, respectively). Integrin α_E is considered an important factor for localizing certain lymphocyte subsets to mucosal-epithelial tissues such as the gut and pancreatic islets (23, 24). Recent observations in mice showed increased NK-specific transcripts in pancreatic islets of type I diabetic mice and that depletion of these cells inhibited diabetes development (25). Others demonstrated in rat models that gut-derived NK cells were equally cytotoxic to pbNK cells (26). In light of these observations, human cytotoxic CD56^dimCD16⁺ NK cells may preferentially localize to gut-associated lymphoid tissue and play a role in local immune surveillance.

CD56^brightCD16⁻ vs CD56^dimCD16⁺ NK cells: implications for NK development and homeostasis

NK cell differentiation occurs when precursors interact with cytokines and stromal cells in the bone marrow (4). The up-regulation of IL-7R α-chain mRNA (by ∼11-fold, Fig. 4⇑⇑A) and specific surface expression (Fig. 6⇓A) on CD56^brightCD16⁻ NK cells, but not CD56^dimCD16⁺ NK, might also relate to certain developmental stages of this subset. Despite the low levels of IL-7R expression, this receptor was functional and induced specific proliferation in CD56^brightCD16⁻ NK cells (Fig. 6⇓B). This receptor is considered as one of the key players in maintaining homeostasis and survival of naive and memory T cells (27), which raises the question of whether it has a similar function in CD56^brightCD16⁻ NK homeostasis. Also, it has been shown in mice that the IL-7R is present on NK cell precursors; however, they lose its expression upon terminal maturation and acquisition of cytotoxic capabilities (11). Because it is not known yet whether mice have NK cell subsets analogous to CD56^bright and CD56^dim, data regarding human NK subset development has been limited. Nevertheless, based on the inverse correlation between maturation/cytotoxic ability and IL-7R expression in the mouse NK cell lineage, our results can be interpreted in a way that supports a current hypothesis regarding human NK subset development, claiming that CD56^dim cells are derived from the CD56^bright NK subset following further maturation (4).

• Download figure
• Open in new tab
• Download powerpoint

FIGURE 6.

Unique expression of IL-7 receptor α-chain on CD56^brightCD16- NK subset. A, Flow cytometry analysis of sorted CD56^brightCD16⁻ and CD56^dimCD16⁺ NK subsets for IL-7R α-chain. B, Fold increase in proliferation of purified CD56^brightCD16⁻ and CD56^dimCD16⁺ NK subsets in response to addition of 20 ng/ml recombinant human IL-7 for 72 h in 37°C. Data presented are representative of two independent experiments.

The interaction between Notch and its ligands, delta or jagged family members, constitutes an evolutionary conserved pathway for determining cellular development pathways (28). Notch signaling is intimately involved in the process of T vs B lymphocyte differentiation from common progenitor precursors (28). Moreover, Notch1 signaling has been shown to attenuate peripheral T cell activation by inhibiting activation and cytokine production (29). Our analysis yielded significant levels of Notch1 and Notch2 transcripts in CD56^dimCD16⁺ and, more abundantly, in CD56^brightCD16⁻ NK cells (Fig. 4⇑⇑A) thus suggesting possible involvement of this pathway in regulating NK subset development, function, and homeostasis.

CD56^brightCD16⁻ vs CD56^.dimCD16⁺ NK cells: secreted immune effector molecules

Differentially expressed genes coding for secreted proteins, such as chemokines, were studied. It was found that CD56^dimCD16⁺ NK expressed significantly higher levels of IL-8, MIP-1α, MIP-1β, and RANTES transcripts (Fig. 4⇑⇑C). This group of chemokines has been demonstrated to be expressed in mouse NK cells in vivo, and play an essential role in recruiting inflammatory effector cells to peripheral inflammation sites during murine CMV infection (30). Therefore, the specific abundance of these chemokines in CD56^dimCD16⁺ NK correlates with the abundance of this subset in inflamed tissues rather than secondary lymphoid organs. Lymphotactin, which was also transcribed significantly in CD56^dimCD16⁺ NK and was also known to be important in lymphoid recruitment to inflammatory tissues, was enriched in CD56^brightCD16⁻ (by ∼4-fold, Fig. 4⇑⇑C). This chemokine has been uniquely shown to have the capacity to modulate costimulatory levels of human CD4⁺ and CD8⁺ T cells following TCR priming in lymph nodes (31), implying that lymphotactin secretion by CD56^brightCD16⁻ NK cells present in T cell areas of secondary lymphoid organs (6) might constitute a cross-talk pathway with neighboring T cells.

CD56^brightCD16⁻ NK expressed higher levels of lysozyme mRNA (by ∼5.6-fold, Fig. 4⇑⇑C), whose natural substrate is the bacterial cell wall peptidoglycan (cleaving the β[1–4] glycosidic linkages between N-acetylmuramic acid and N-acetylglucosamine), and granulysin which is a cytolytic granule member known to reduce the viability of a broad spectrum of intra- and extracellular pathogens (32). CD56^dimCD16⁺ NK preferentially expressed defensin-6 (by ∼5.8-fold) which is known to disrupt bacterial membranes (33). A secreted protein specifically up-regulated in CD56^brightCD16⁻ NK cells (by ∼6.9-fold) was amphiregulin, a member of the epidermal growth factor (EGF) family that interacts with the EGF/TGF-α receptor to promote the growth of normal epithelial cells and inhibits the growth of certain aggressive carcinoma cell lines (34). These observations suggest that highly purified NK cells might posses additional novel and direct cytotoxic pathways against invading microorganisms or cancerous cells by secreting certain natural antimicrobial or growth-suppressing agents.

The proinflammatory cytokine, osteopontin, has been shown to critically aggravate development and progression of experimental autoimmune encephalomyelitis (EAE) in mice through enhancing Th1 activity (35). Osteopontin was specifically transcribed in CD56^dimCD16⁺ NK cells (Fig. 4⇑⇑C). Interestingly, these cells have been shown to infiltrate inflammatory lesions in CNS, especially at peak stages in the development of the disease (36). This observation calls for further evaluation of osteopontin production by this NK subpopulation in EAE models and exploring direct and indirect links between these cells and autoreactive T and B cells.

CD56^brightCD16⁻ vs CD56^dimCD16⁺ NK cells: other differences

As mentioned earlier, one aim of this study was to analyze certain members of the group of proteins that were up-regulated both in CD56^brightCD16⁻ and activated CD16⁺ NK cells compared with the CD56^dimCD16⁺ subset (Fig. 2⇑C). Interestingly, several of the genes in this group belong to the MHC class II family (HLA-DRA, HLA-DRB, HLA-DMA, HLA-DPA, HLA-DQA, HLA-DPB, and CD74) (Figs. 4⇑⇑A and 3⇑C). This expression pattern suggests the functional possibility that superantigens may bind to MHC class II-positive CD56^brightCD16⁻ or activated NK cells in lymph nodes or peripheral inflammatory tissues and potentiate the pathologic and nonspecific T cell proliferation and activation.

Lymphotoxin β is a type II membrane protein of the TNF family. It was uniquely transcribed in CD56^brightCD16⁻ NK cells (up-regulated by ∼8-fold, Fig. 4⇑⇑A), and robustly overexpressed after activation on CD16⁺ NK cells (up-regulated by ∼50-fold, Fig. 7⇓⇓A). This protein anchors lymphotoxin α to the cell surface through a heterotrimer complex that has the ability to bind the lymphotoxin β receptor (37). Surface expression of lymphotoxin β on B cells in secondary lymphoid organs has a crucial role in supporting correct lymphoid architecture that is critical for an effective immune response (37). The fact that CD56^brightCD16⁻ NK cells are preferentially found in lymph nodes might supply the biological rationale for specific expression of lymphotoxin β on this subset. In contrast, this molecule is an inducer of proinflammatory cytokines and chemokines in effector cells localizing to inflamed tissues (38). It is possible that activated CD16⁺ NK use this robust induction in lymphotoxin β to contribute to the progression and potentiation of the ongoing immune response.

• Download figure
• Open in new tab
• Download powerpoint

FIGURE 7.

Gene expression patterns in human activated CD16⁺ and CD56^dimCD16⁺ NK subsets. Gene expression profiles of selected genes differentially expressed by at least 2-fold between the unactivated CD56^dimCD16⁺ and activated CD16⁺ NK subset. Bars represent the fold change of the mRNA level of a particular gene when comparing these two subpopulations. Positive values indicate that the transcript was more abundant in activated CD16⁺ NK and negative values indicate the opposite. The presented genes were grouped according to their presumed function (A–G) based on information available in public databases or in the literature. (Figure continues)

• Download figure
• Open in new tab
• Download powerpoint

FIGURE 7A.

(Continued)

CD56^brightCD16⁻ NK cells have been shown to be generally better responders to various cytokine stimulations (response can be observed at relatively lower concentrations of cytokine stimulation), as compared with the CD56^dimCD16⁺ NK subset, due to preferential expression of a number of membrane-associated receptors (e.g., c-kit receptor, high affinity IL-2R, etc.) (39). The results indicated that several proteins expressed at higher levels on the CD56^brightCD16⁻ NK subset could also contribute to the enhanced cytokine responsiveness of this subset: 1) ITAM containing signal transducing adaptor molecule STAM (by ∼2.5-fold, Fig. 4⇑⇑B) that is known to associate with JAK2 and JAK3 kinases and participates in downstream signaling of the cytokine receptor upon phosphorylation (40); 2) SRY-box 4 (SOX4) transcription factor (by ∼10-fold, Fig. 4⇑⇑D) that is involved in controlling cytokine induced cellular transcriptional regulation (41); 3) ribosomal protein family members L36, L35, L27A, L3, L22, L10A, and S26 (Fig. 4⇑⇑G) that catalyze protein synthesis were also preferentially expressed in CD56^brightCD16⁻ NK. This observation might suggest that this subset is more “metabolically ready” to engage in the protein synthesis process that is required for functional responsiveness to cytokines.

Lastly, many other genes overexpressed in one of the unactivated pbNK subsets that are presented in Fig. 4⇑⇑ or in on-line supplementary Table II (e.g., EBV-induced gene (EBI2), VEGFβ, lymphopain, TCF7, annexin A2, CD55, etc.) are of considerable interest but are beyond the scope of the present paper. The data presented in these figures should be useful to those interested in analyzing differentially expressed genes that were not discussed in this manuscript.

Characterization of CD16⁺ NK cell transcriptome following activation

To gain more insight into potential functions that NK cells might exert upon activation in inflamed tissues, we analyzed transcriptional changes in CD16⁺ NK cells following activation. Activated CD16⁺ NK cells analyzed in this work were activated with a combination of stimuli including IL-2, PHA, cytokines produced by irradiated nonautologous PBLs (used as feeders), and addition of classical NK target cells (irradiated RPMI 8866 cells). Therefore, it was not surprising to observe the dramatic alteration in gene expression profile in the activated CD16⁺ NK cells (Fig. 2⇑, A, C, and D).

Initially, genes up-regulated in the activated CD16⁺ NK line compared with both pbNK subsets (Fig. 2⇑C) were evaluated. Expectedly, a group of over 30 genes known as inducers of cell cycle progression and proliferation (e.g., CDC2, CDK4, CCNA2, CCNB2, CCNA2, MCM2, CKS2 etc) (Fig. 7⇑⇑F and data not shown) were robustly, and in most cases, specifically expressed in the highly proliferating activated CD16⁺ NK cell subset, while antiproliferative genes BTG1 and cyclin dependent kinase inhibitor 1C were down-regulated. We then focused on selected differentially expressed genes between CD56^dimCD16⁺ and activated CD16⁺ NK subsets, and classified them into functional categories (Fig. 7⇑⇑).

Activated vs unactivated CD16⁺ NK cells: molecules involved in the regulation of non-NK immune functions

The naive T cell responses to Ag are directed by immunogenic stimuli that can be divided into two main categories. First, TCR ligation mediated by peptide-MHC complexes expressed on professional APCs. Second, costimulatory signal conferred by a counter-receptor expressed on APC or other accessory cells (42), which is mandatory for efficient priming of the immune response and for avoiding anergy (43). OX40L, CD70, and CD86 which are ligands for the classical TCR costimulatory molecules OX40, CD27, and CD28, respectively (43), were exclusively transcribed in activated NK cells (by ∼2.5-, 4-, and 6-fold, respectively) (Fig. 7⇑⇑A). The observed transcriptional pattern for these molecules was verified by detecting cell surface protein expression (Fig. 3⇑, G–I). Class IV semaphorin Sema4A, usually expressed on DCs and known to costimulate T cell responses (44), was also transcribed in all three NK subsets examined at similar levels (data not shown). We extend the relevance of these observations by showing that irradiated activated CD16⁺ NK cells were extremely potent in costimulating proliferation and IL-2 secretion of peripheral blood CD3⁺ T cells that were activated by suboptimal levels of plate-bound anti-CD3 mAb (Fig. 8⇓, A and B). Although it is possible that the irradiated activated NK cells used can produce part of the IL-2 detected in this experiment, direct intracellular staining of T cells confirmed the up-regulation of IL-2 expression upon incubation with anti-CD3 and activated CD16⁺ NK cells (data not shown). The presence of anti-CD3 mAb, in addition to activated NK cells, was mandatory for obtaining robust T cell proliferation further strengthening the conclusion that TCR costimulation enhanced CD3⁺ T cell activation in this experiment (Fig. 8⇓, A and B). It is important to note that OX40-OX40L interaction has been described as a major pathway in maintaining long-term survival of neutrophils and primed T cells in vivo (45). Taken together, these results provide direct evidence for novel and unexpected cross-talk pathways between NK and T cells as TCR costimulators, linking the innate and adaptive immune responses.

• Download figure
• Open in new tab
• Download powerpoint

FIGURE 8.

Characterization of functional interactions between activated CD16⁺ NK and T cells. A and B, Activated human CD16⁺ NK cells costimulate TCR responses of human CD3⁺ T cells. A, Proliferation of purified peripheral blood CD3⁺ T cells (50 × 10³ cells per well) in the presence of irradiated unactivated polyclonal NK cells or activated CD16⁺ NK cells (25 × 10³ cells per well) with or without suboptimal levels of anti-CD3 mAb (0.1 μg/ml). As a positive control, we used soluble anti-CD28 Ab (20 ng/ml) in addition to anti-CD3. Proliferation was (Figure legend continues) determined by [³H]thymidine incorporation after 72 h of total incubation time. B, Before cell harvesting, 100 μl of supernatant from each well was removed for ELISA measurement of IL-2. Values are mean ± SD for triplicate samples. C, A total of 1 × 10⁶ unactivated polyclonal CD56⁺CD3⁻ NK cells were grown for 7 days at a concentration of 0.5 × 10⁶/ml in RPMI 1640 medium + 10% heat inactivated FCS with or without the addition of 100 U/ml IL-2 and 1 ng/ml PHA. Subsequently, we removed supernatants of these cultures and introduced them to Jurkat cells at a concentration of 1 × 10⁶/ml for 8 h. Incubation of Jurkat cells with control medium that contained IL-2 and PHA was performed as a control to rule out that the apoptotic effect observed was a result of activation-induced cell death. Jurkat cells were then analyzed for apoptosis levels by flow cytometry. Propidium iodide-positive (necrotic) cells were excluded from the analysis, while the remaining cells were characterized for Annexin V staining percentages. All experiments shown are representative samples of three total repeats performed.

The above observations encouraged us to search for other molecules on NK cells that might mediate direct interaction with immune cells. Tim3 (T cell Ig- and mucin-domain-containing molecule) was significantly induced on activated CD16⁺ NK cells (by ∼3-fold, Fig. 7⇑⇑A). This molecule is known to be expressed on Th1, but not Th2, CD4⁺ T cells (46), where it is thought to down-regulate macrophage homeostasis and activation levels and autoregulate the severity of Th1-mediated immune responses (46, 47). Additionally, as Th1 and Th2 cells negatively regulate each other’s functions, Th1 cells might be implicated in promoting asthma and atopy (typical Th2 immune responses) via Tim3-Tim3 ligand interactions (48). We therefore speculate that NK cells might also influence macrophage behavior and participate in balancing Th1-Th2 cross-regulation.

Activated vs unactivated CD16⁺ NK cells: alterations in trafficking capabilities

Chemokines and chemokine receptors are pivotal players in trafficking, homing, and retention of immune and nonimmune cells. The activity of several chemokines is regulated by CD26 (dipeptidyl-peptidase IV)-mediated cleavage (49). The results indicate that mRNA encoding CD26 was uniquely transcribed in activated CD16⁺ NK and up-regulated by >12-fold (Fig. 7⇑⇑A). The inactivation of chemokines by CD26, together with down-regulation of CXCR4 (Fig. 7⇑⇑A), may contribute to the fine control of chemotactic migration of these cells by providing a “stop” signal that keeps these cells at the site of inflammation.

The cell surface molecules: CD9, CD53, CD63, CD81, and CD151, all belonging to the tetraspanin family (also known as transmembrane 4 superfamily), were up-regulated on CD16⁺ NK cells after activation (Fig. 7⇑⇑A). Flow cytometry analysis confirmed CD9, CD53, and CD81 up-regulation on this NK subset (Fig. 3⇑, D–F). Many of the tetraspanins assemble with various integrin subunits into functional signaling complexes and facilitate alteration in cell-cell and cell-matrix interactions (50, 51, 52, 53). Thus, overexpression of certain tetraspanin family members on activated NK cells might underlie mechanisms used by these cells to enhance or alter their migration and retention in inflamed tissues. Additionally, CD81 is known to confer inhibition of NK cell proliferation (54). Interestingly, the IFN-induced transmembrane proteins 1 and 3 (IFITM1 and IFITM3) that are known to be involved in preventing overproliferation of T cells (55) were also up-regulated on CD16⁺ NK cells following activation (Fig. 7⇑⇑A). This suggests that NK cells, similarly to T cells, might use the up-regulation of these receptors together with CD81 as part of an autoregulatory “brake” mechanism that controls their activation-induced proliferation.

Activated vs unactivated CD16⁺ NK cells: secreted immune effector molecules

In agreement with previous observations, granzyme A, granzyme B, CTLA1, and granzyme K were up-regulated on activated CD16⁺ NK, while granzyme M displayed a unique transcription pattern compared with other members of the granzyme family (down-regulated by ∼2.5-fold upon activation; Fig. 7⇑⇑C). Granzyme M is known to be exceptional in that its expression levels have been previously shown not to correlate with the cytotoxic ability of NK cells. Granzyme M also induces cell death by a unique mechanism not featuring DNA fragmentation and occurring independently of caspases (56, 57).

Galectins, also referred to as S-type lectins, are a conserved family of proteins defined by the presence of at least one characteristic carbohydrate recognition domain (58). Activated CD16⁺ NK cells up-regulated expression of galectin 1 and galectin 12 (Fig. 7⇑⇑C), both characterized as inducers of apoptosis (59). Galectin 1 expression at the protein level on activated CD16⁺ NK cells was verified by intracellular flow cytometry staining (Fig. 3⇑J). This protein has been shown to mediate apoptosis of peripheral activated T cells by segregating O-glycosylated CD45 from CD7 and CD43 (58, 59). Interestingly, supernatants of activated CD16⁺ NK cultures were able to induce low, but significant, levels of apoptosis of the CD7⁺ Jurkat T cell line (Fig. 8⇑C). This effect was not a result of IL-2 and PHA induced cell death, because addition of control medium containing these materials did not induce such apoptosis. This experiment strengthens the possibility of proapoptotic factor production by CD16⁺ NK cells following activation. One hypothesis may be that induction of activated T cell apoptosis via NK-derived galectin-1 and 12 might act in concert with the described ability of syngeneic NK cells to recognize and kill activated, but not resting, CD4⁺ and CD8⁺ T cells. This would lead to attenuation of T cell activation and resolution of the immune response (60). Galectin 3 was another family member up-regulated in CD16⁺ NK cells following activation (Figs. 7⇑⇑C and 3⇑K). It was shown previously that intracellular galectin-3 induction following activation of B cells is important for long-term survival of these cells during the immune response (61). Studies are currently evaluating whether galectin 3 has a similar effect on activated NK cell survival and homeostasis.

Activated vs unactivated CD16⁺ NK cells: other differences

In correlation with the enhanced cytotoxic activity of activated CD16⁺ NK cells, the lysis inhibitory receptor NKG2A and the orphan NKRp1 lectin-like receptor were down-regulated upon NK activation (Fig. 7⇑⇑A). In contrast, NKG2C and NKG2E (KLRC2 and KLRC3) were up-regulated on activated NK cells (Fig. 7⇑⇑A). Both receptors are CD94-associated members of the C-type lectin like receptor family and lack an ITIM in their cytoplasmic domain. NKG2C associates with DAP12 and acquires activating functions. It is presumed that, due to the structural similarities between these two receptors, NKG2E might also facilitate activating functions. However, the ligands and function of NKG2E have not been characterized so far. Up-regulation of the NK costimulatory molecules CD2 and CD59, together with the FcεRIγ chain on activated NK cells might also contribute to the lytic potency of these cells (Fig. 7⇑⇑A).

Many transcripts induced or down-regulated in CD16⁺ NK cells upon activation are presented in Fig. 7⇑⇑ or in on-line supplementary Table II (e.g., TRAIL, CLECSF2, SIGLEC7, CD55, dopamine D4 receptor, several G protein coupled receptors, MMP25, survivin, etc.) and are of considerable interest, but are beyond the scope of the analysis presented in this paper.