Proteome Analysis of Distinct Developmental Stages of Human Natural Killer (NK) Cells

The recent Natural Killer (NK) cell maturation model postulates that CD34+ hematopoietic stem cells (HSC) first develop into CD56bright NK cells, then into CD56dimCD57− and finally into terminally maturated CD56dimCD57+. The molecular mechanisms of human NK cell differentiation and maturation however are incompletely characterized. Here we present a proteome analysis of distinct developmental stages of human primary NK cells, isolated from healthy human blood donors. Peptide sequencing was used to comparatively analyze CD56bright NK cells versus CD56dim NK cells and CD56dimCD57− NK cells versus CD56dimCD57+ NK cells and revealed distinct protein signatures for all of these subsets. Quantitative data for about 3400 proteins were obtained and support the current differentiation model. Furthermore, 11 donor-independently, but developmental stage specifically regulated proteins so far undescribed in NK cells were revealed, which may contribute to NK cell development and may elucidate a molecular source for NK cell effector functions. Among those proteins, S100A4 (Calvasculin) and S100A6 (Calcyclin) were selected to study their dynamic subcellular localization. Upon activation of human primary NK cells, both proteins are recruited into the immune synapse (NKIS), where they colocalize with myosin IIa.

Natural killer (NK) 1 cells are large granular lymphocytes that provide a first innate immune defense. They are able to kill virus-infected and transformed cells and furthermore release cytokines and chemokines to activate adaptive immune cells (1,2).
The balance of signals from activating and inhibitory NK cell surface receptors tightly regulates NK cell activity. Activated NK cells release lytic granules through a process called degranulation. Therefore, NK cell cytotoxicity requires the formation of the F-actin-rich NK immune synapse (NKIS) and the transport of Perforin-containing lytic granules to the NKIS. Furthermore, this process requires granule-associated MYH9 protein (non-muscle Myosin IIa) mediating the interaction of granules with F-actin at the NKIS (3)(4)(5), leading to lytic granule exocytosis. Whereas related phenotypes and functional properties are well characterized, the underlying regulatory protein network mediating differentiation, cytokine release, and cytotoxicity, is still incomplete. NK cells are defined by the expression of the surface molecule CD56 (NCAM1) and the absence of the T cell receptor (TCR) associated protein CD3 and can be further subdivided into subsets (6,7). CD56 expressing cells originate from CD34 ϩ HSCs. Notably, the commitment to the NK lineage includes discrete steps from HSC to cells, expressing high CD56 levels (CD56 bright ) (8,9), which act immune regulatory by the release of various cytokines. NK cells with low CD56expression (CD56 dim ) predominantly constitute cytotoxic responses (10,11). Contact of CD56 (NCAM1) with fibroblasts (12) and neutrophils (13) supports the differentiation process from CD56 bright to CD56 dim NK cells. The progression of early differentiation steps is proven by telomere length investigation (14 ) and early presence in blood after HSC transplantation (HSCT) (14,15). Indeed, CD56 dim NK cells are able to change their phenotypic properties, which can be correlated with continued differentiation throughout their whole lifespan (15)(16)(17)(18). CD57 was determined to be a senescence marker in T cells (19). Recent studies determined CD57 ϩ NK cells as a fully mature NK cell subset among the CD56 dim NK cell pop-ulation (CD56 dim CD57 ϩ and CD56 dim CD57 Ϫ ). Furthermore, the NK cell differentiation process is characterized by a reversible loss of NKG2A in parallel with an irreversible acquisition of KIRs and CD57 (15).
Furthermore, CD57 ϩ NK cells are characterized by a specialized phenotype that includes increased CD16-and Perforin-expression, reduced responsiveness to cytokines and decreased proliferation capacity. CD57 is mostly studied in the context of NK cell education that runs in parallel but uncoupled from NK cell differentiation (15,17,18). The NK cell education process encompasses the acquisition of activating and inhibitory surface receptors, like KIRs, which in turn interact with HLA class I ligands, e.g., during viral infections (19,20). CD57 ϩ NK cells can recognize cytomegalovirus (CMV) and developed memory effects toward this virus (21,22). Likewise, an expansion of the CD57 ϩ NK cell population is observed during Hantavirus (24), Chikungunya virus (25), and HIV-1 (16), but not during HSV-2 (26) infections. Recently, Lanier and colleagues showed that IL-12 is indispensable for NK cell expansion and the generation of memory NK cells during MCMV infection (27). Up to now the molecular network underlying CD57 ϩ phenotypes are mostly characterized by FACS techniques and microarray analyses on the systemic level (15,17,18).
Mass spectrometry can identify and quantify proteins in a global and unbiased manner, and thereby certainly contribute to the elucidation of developmental processes and the acquisition of specialized functions. Proteomic studies on human NK cells are accomplished mainly at the level of NK cell lines, because of the scarcity of primary NK cell subsets. Cytotoxicity, but not development, was studied in NK-92 and YTS NK cells by 2D-PAGE and peptide sequencing approaches pending on activating signals (28 -32). At least one proteome study investigates ex vivo expanded primary human NK cells, and focuses on the characterization of kinases, involved in NK cell activation (33). Attempts to unravel the basics of NK cell development in mice were successful (34) but not completely transferable to the human NK cell system because of a different set of surface receptors. Hence, several studies have contributed to our understanding of the role of surface receptors in different developmental stages, but studies targeting the regulation of intracellular proteins are still missing.
In this study we characterized distinct developmental stages of primary human NK cells by proteomics. To get better insight into the molecular mechanisms of the NK cell differentiation process we comparatively analyzed freshly isolated primary CD56 bright , CD56 dim , CD56 dim CD57 Ϫ and CD56 dim CD57 ϩ NK cells by isobaric tags for relative and absolute quantification (iTRAQ)-based LC-MS/MS. We obtained relative quantitative data for more than 3400 proteins and observed a specific CD56 ϩ NK cell core proteome. The obtained proteomic data strongly supports the current differentiation model of NK cells by highlighting strong distinctions between CD56 bright and CD56 dim NK cells and high similarity among CD57 Ϫ and CD57 ϩ NK cells. In addition to a significant set of anticipated and well-known proteins involved in NK cell development and effector functions, we detected also 11 novel protein candidates so far undescribed in NK cells. The expression patterns of S100A4 (UniProt accession name S10A4) and S100A6 (UniProt accession name S10A6), both belong to the family of S100 calcium binding proteins and contain two EF-hand domains, correlated with the developmental stages of cytotoxic NK cell subsets. Both proteins were recruited into the NKIS, following NK cell activation.
Human NK Cells-This study was conducted in accordance with the rules of the Regional Ethics Committee of Lower Saxony, Germany and the declaration of Helsinki. Buffy coats from blood donations of healthy human volunteers who provided informed consent were obtained from the Institute for Clinical Transfusion Medicine, Klinikum Braunschweig, Germany. Blood donors' health is rigorously checked before being admitted for blood donation. This process included a national standardized questionnaire with health questions, an interview with a medical doctor and standardized laboratory tests for a) infections HIV1/2, HBV, HCV, Syphilis (serology and/or nucleic acid testing) and b) hematological cell counts.
Buffy Coats were produced from whole blood donations on day 1 by using the Top & Bottom Extraction Bag System (Polymed Medical Devices TM , Triple Blood Bag System, No. 7300; containing CPDA-1. Peripheral blood mononuclear cells (PBMCs) were isolated from these buffy coat products by Biocoll density gradient centrifugation (Biochrome AG) on day 2. PBMCs were cultured overnight in RPMI 1640 medium (GIBCO) supplemented with 10% fetal bovine serum (FBS) gold (PAA Laboratories, Etobicoke, ON, Canada), 2 mM L-glutamine, 50 units/ml penicillin and 50 g/ml streptomycin (all Invitrogen) at 37°C in a humide 7.5% CO 2 atmosphere.
FACS Sorting-PBMCs were incubated with fluorochrome-conjugated anti-CD3, anti-CD56, and anti-CD57 monoclonal antibodies (mAbs) for 15  Cell Lysis, Protein Digestion, and iTRAQ Modification of Peptides-CD3 Ϫ CD56 dim /CD3 Ϫ CD56 bright (CD56 dim versus CD56 bright ) and CD56 dim CD57 ϩ /CD56 dim CD57 Ϫ (CD57 ϩ versus CD57 Ϫ ) NK cells were each isolated from five individual human donors and mass spectrometry data were obtained and processed for each of the 10 donors individually. At a minimum, 0.9 ϫ 10 6 NK cells were used for each MS experiment. NK cells were lysed in ice-cold lysis buffer (50 mM HEPES pH 7.5, 150 mM NaCl, protease inhibitor mixture supplemented with EDTA, 1% Triton-X100) on day 3. Protein concentrations were determined using a NanoDrop spectrophotometer (ND-1000, Peqlab, Biotechnology GmbH, Erlangen). Proteins were extracted from lysates by chloroform/methanol precipitation as described previously (35), and redissolved in dissolution buffer (50 mM TEAB). Equal amounts of protein from each NK cell subset was digested with sequencing grade modified trypsin from Promega, as recommended in a ratio of 1:50 at 37°C on day 3 overnight. Subsequently, digestion was completed by adding a further 1 g of trypsin for 2 h to each sample. Labeling of tryptic peptides with isobaric iTRAQ reagents was performed on day 4, according to the manufacturer's guidelines (Applied Biosystems, Foster City, CA). Peptides derived from CD56 dim NK cells were labeled with iTRAQ reagent 117, those derived from CD56 bright NK cells were labeled with iTRAQ reagent 115 Da. For quantitative MS analyses of peptides from CD3 Ϫ CD56 dim CD57 ϩ (CD57 ϩ ) NK cells, iTRAQ label 114 was used, and iTRAQ 116 for CD3 Ϫ CD56 dim CD57 Ϫ (CD57 Ϫ ) NK cell peptides (supplemental Table  S1; Supplement). Peptides of two different NK cell subsets (CD56 bright and CD56 dim or CD57 Ϫ and CD57 ϩ NK cells) were combined (1:1 ratio), vacuum dried, dissolved in 0.2% trifluoroacetic acid and desalted on self-packed LiChroprep RP-18 (Merck) SPE columns.
Strong Cation Exchange Chromatography (SCX)-The combined iTRAQ-labeled peptide samples were further subfractionated by strong cation exchange chromatography (SCX) on day 4 to support representative and comprehensive protein identification by LC-MS/ MS. Peptides were dissolved in SCX buffer (0.065% formic acid, 25% ACN) and fractionated on a Mono SPC1.6/5 column connected to an Ettan micro-LC system (both GE Healthcare), and separated at a flow rate of 150 l/min for 30 min with a linear gradient from 0% to 35% SCX buffer supplemented with 0.5 M potassium chloride. Fractions were collected by a microfraction collector, every minute (SunCollect). Peptide elution was monitored by an UV detector at 214 nm. Peptide-containing fractions were vacuum-dried, desalted by RP-C 18 chromatography ZipTip pipette tips (Millipore) and analyzed separately by LC-MS/MS.

LC-MS/MS Measurement and Protein Identification-LC-MS/MS
analyses were performed with an UltiMate 3000 RSLCnano LC system (Thermo Scientific) connected to an LTQ Orbitrap Velos Fourier transform mass spectrometer (Thermo Scientific). Peptides were applied to a C 18 precolumn (3-m, Acclaim, 75 m ϫ 20 mm, Dionex) and washed with 0.1% TFA for 3 min at a flow rate of 6 l/min. Subsequently, peptides were separated on a C 18 analytical column (3-m, Acclaim PepMap RSLC, 75 m ϫ 25 cm, Dionex) at 350 l/min via a linear 120-min 3.7-31.3% B gradient with UPLC solvent A (0.1% formic acid in water) and UPLC solvent B (0.1% formic acid in 80% ACN). The LC system was operated with Chromeleon Software (version 6.8, Dionex). The effluent from the column was electrosprayed (Pico Tip Emitter Needles, New Objectives) into the mass spectrometer. The mass spectrometer was controlled by Xcalibur software (version 2.1, Thermo Scientific) and operated in the datadependent mode allowing the automatic selection of doubly and triply charged peptides and their subsequent fragmentation. Peptide fragmentation was carried out using High Collision Dissociation (HCD) settings, with Collision Energy (CE) 44 optimized for iTRAQ-labeled peptides. MS/MS raw data from all SCX fractions, corresponding to one experiment, were visualized by Xcalibur software (.raw-file) and merged for Mascot Daemon-aided searches (Mascot version V2.3.02, Matrix Science) against UniProtKB/Swiss-Prot protein database (release 2011_03; with 525,997 entries; taxonomy: Homo sapiens with 20,226 entries). The following search parameters were used: enzyme, trypsin; maximum missed cleavages, 1; fixed modifications, iTRAQ TM 4-plex (K), iTRAQ TM (N terminus), Methylthio (C); variable modifications, oxidation (M); peptide ion mass tolerance, 10 ppm; MS/MS tolerance, 0,2 Da. Mascot result files (.dat files) were uploaded into the software Scaffold (Version Scaffold_3_00_01, Proteome Software Inc.), which was used for inspection of MS/MS data-based identification of peptides/proteins, to support protein quantifications and to prepare statistical validation and comparison of donor specific data sets, respectively. Only proteins with a minimal protein identification probability of 99% and identified by at least two unique peptides showing a minimal peptide identification probability of 95% and a Mascot (peptide) Score of at least 30 were accepted in this survey. Mascot-aided decoy searches were performed against the randomized Uniprot/Swiss-Prot protein database Mascot version V2.3.02, Matrix Science) against UniProtKB/Swiss-Prot protein database (release 2011_03; with 525,997 entries; taxonomy: Homo sapiens with 20,226 entries) and protein false discovery rates (FDRs) were calculated by Scaffold on the basis of FDR ϭ FP (false positive)/(FP ϩ TP (true positives)). On average we determined FDRs always lower than 0.9% (with significance level 0.05 in Scaffold). All MS-data associated with this manuscript are published in the PROteomics IDEntification Database (PRIDE) (www.ebi.ac.uk/pride/).
Protein Quantification-The Scaffold software (Version Scaf-fold_3_00_01, Proteome Software Inc.) was used for relative quantification of protein expression in NK cell subsets. Mascot result files (.dat files) were uploaded into Scaffold and all identified proteins as defined by the criteria described above were considered for quantitative data analyses. For quantification, only unique peptides were used. Scaffold provided normalized log 2 -regulation factors (RFs), basing on the iTRAQ-label intensity of individual peptides, respectively. Log 2 regulation factors represent the relative abundance of a protein in CD56 dim compared with CD56 bright or CD57 ϩ compared with CD57 Ϫ NK cells (RF ϭ CD56 dim /CD56 bright ; CD57 ϩ /CD57 Ϫ ). Regulation factors calculated by Scaffold were exported to Excel (Microsoft Office 2007) and statistically analyzed.
Statistical Data Analysis-The aim of the statistical evaluation was to distinguish random fluctuations of protein regulation factors (RF) or potential donor variations from statistically significant and conserved regulations within the NK cell data sets. To identify significant protein regulations the variation of the regulation factors was statistically estimated. In our statistical model we assume that the log 2 -regulation factors (RF) of each protein follow a normal distribution (5 ϫ log 2 RFs per protein because of five analyzed donors per NK cell subset comparison) with different expected values, but with the same standard deviation 0 . It is necessary to estimate 0 to distinguish significant (donor-dependent) deviations from no regulation. The limited number of replicates (n ϭ 5 donors in each of both comparative NK subset analyses) required an estimation of the standard deviation by taking RFs of all proteins into account. In our model we used the mean of the MADs (Median Absolute Deviation from the median) of all proteins as an estimator for the standard deviation of a normal distribution (with a correction for small sample sizes). The MAD was corrected by the factor of 1.4826 (determined by Monte Carlo simulation) according to the number of k ϭ 5 data sets (donors) to obtain an unbiased and robust estimator for the standard deviation 0 . In this way we obtain many estimates for the MAD with a large variance. But averaging this larger number of not very reliable estimates leads to a reliable estimate of the MAD.
Based on the previous estimation we are able to construct a hypothesis test for the identification of significantly regulated proteins. Thereby, the hypothesis of the test encompasses that a protein is considered to be significantly regulated if its mean regulation deviates significantly from 0. The function ⌽ represents the cumulative distribution function of the standard normal distribution. As previously shown, we propose a strict test (36), where ␣ is the significance level of the test (here ␣ ϭ 5%; before FDR correction) and where the absolute value of the log 2 -regulation exceeds a given threshold c ␣ in at least m (m ϭ 3, 4, 5) out of the k ϭ 5 replicates. n m is the number of proteins that have been measured in at least m replicates. We used this value for the FDR correction, because we cannot apply the test to proteins that have been measured in less than m replicates. Taken all the parameters together we obtained the value for the threshold as follows: These thresholds (for m ϭ 5, m ϭ 4, and m ϭ 3 donors) were used to define a set of significantly and donor-independently regulated proteins (36 proteins).
Microscopy of fixed NK Cells-NK cells were freshly isolated from PBMCs of healthy donors by negative selection using the NK cell Isolation Kit (Miltenyi Biotec). Target cells (K562-leukemia cells) were allowed to settle overnight on poly-L-lysine coated coverslips (Sigma Aldrich). Pure NK cells (Ն 96%), were added at a ratio of 1:1 and incubated for 5, 10, 15, 20, and 30 min, at 37°C and 7.5% CO 2 . Cells were fixed for 20 min with 2% paraformaldehyde (Sigma) in sterile PBS, permeabilized for 3 min in 0.5% Saponin (Sigma) in PBS and 1h blocked with 1% BSA in PBS, supplemented with 0.05% Tween-20 (Roth). Cells were stained (3) with anti-Perforin, anti-CD107a, anti-Myosin IIa, anti-S100A4, and anti-S100A6 mabs. Secondary antibody staining was performed with goat anti-mouse IgG, conjugated with Alexa 594 and goat anti-rabbit IgG, conjugated with Alexa 488) in 1:400 dilution supplemented with DAPI (1:1000). Imaging was performed on an Axiovert 135 microscope with HBO lamp and CCD camera. Images were analyzed by MetaMorph software, version 7.5.3.0 (Analytical Technologies). Five independent experiments were performed using pure NK cells and 20 -30 images were acquired per coverslip in one experiment. Image analysis was performed with ImageJ (version 1.44p).

RESULTS
The Proteome of Primary Human CD56 ϩ NK Cells-Comparative proteome studies of two distinct developmental steps, from CD56 bright to CD56 dim and from CD57 Ϫ to CD57 ϩ , each from five healthy human blood donors were performed.
Peripheral blood mononuclear cells (PBMCs) were freshly isolated from buffy coats of clinically approved healthy blood donors. These PBMCs were separated to generate the different developmental stages of primary human CD56 ϩ NK cells. The used gating strategy for distinct CD56 ϩ NK cell subsets is depicted in Fig. 1A. PBMCs were gated on forward scatter area (FSC) versus side scatter area (SSC), which revealed the lymphocyte population. These were further gated on FSC area versus FSC height and SSC area versus SSC height to exclude doublets and aggregates (not shown). CD3 Ϫ CD56 ϩ NK cells were gated on CD56 dim and CD56 bright subsets. The challenge here was to overcome the scarcity of CD56 bright NK cells, which account for only 5-10% of all NK cells. Therefore, 6 ϫ 10 8 PBMCs were sorted for CD56 ϩ NK cell subsets to obtain about 1 ϫ 10 6 CD56 bright NK cells from each donor, sufficient material to study the first developmental step from CD56 bright to CD56 dim . Similarly, CD56 dim NK cells from further five healthy human blood donors (D6-D10; supplemental Table S1; supplemental Table S2) were isolated separately and sub-divided into CD57 Ϫ and CD57 ϩ NK cell subsets to study the second developmental step.
CD56 bright NK cells were used to normalize (1) the cell numbers of subsets analyzed by proteomics and (2) the obtained protein regulatory data. Proteins were extracted directly from the freshly sorted primary NK cell subsets and digested with trypsin (Fig. 1B). Peptides originating from the distinct subsets were then labeled differentially by iTRAQ, combined and further subfractionated by strong cation exchange chromatography (SCX) to reduce ion suppression effects and to improve protein sequence coverage. Sorting procedure for NK cell subset isolation. Dot plot A1 depicts the PBMC population with gate on lymphocytes (blue). Gating on CD3 Ϫ CD56 ϩ NK cells, with separate gates on CD56 dim and CD56 bright NK cells within the lymphocyte population is depicted in dot plot 2. Further discrimination of the CD56 dim NK cell population into CD57 ϩ and CD57 Ϫ NK cell subsets is shown in dot plot 3. B, Experimental design of MS-based proteomic analysis of NK cell subsets. NK cell subsets were sorted from healthy human blood donor PBMCs to produce specific CD56 pos NK cell subsets. Isolated NK cells were prepared for MS analysis and differentially labeled with iTRAQ for further quantification. Labeled peptides were sequenced by nanoLC-MS/MS. Database searches supported protein identification and comparing iTRAQ reporter intensities revealed protein regulation factors. Statistical evaluation was performed on the generated datasets for quality control and to filter out conserved donor-independent protein regulations. teins in CD56 ϩ primary NK cells. 3345 of those proteins were identified commonly in both comparative proteome approaches ( Fig. 2A). Only 13 proteins were exclusively detected in the CD56 dim/bright subsets and 54 in CD57 ϩ/Ϫ NK cell subsets (supplemental Table S3). Among these uniquely identified proteins, several signaling components, like UB2Q2 (Ubiquitin-conjugating enzyme E2 Q2; CD56 dim/bright ), PTN9 (Tyrosine-protein phosphatase non-receptor type 9; CD57 ϩ/Ϫ ), P52K (52 kDa repressor of the inhibitor of the protein kinase; CD57 ϩ/Ϫ ), SLAP2 (Src-like-adapter 2; CD57 ϩ/Ϫ ), RN167 (E3 ubiquitin-protein ligase RNF167; CD57 ϩ/Ϫ ), CPPED (Calcineurin-like phosphoesterase domain-containing protein 1; CD57 ϩ/Ϫ ), and the C-X-C chemokine receptor type 2 CXCR2 (CD57 ϩ/Ϫ ) were detected. These proteins may possibly represent novel markers for the corresponding NK cell subsets, but a multiplicity was identified only in one donor (8 of 13 proteins in CD56 dim/bright and 37 of 54 proteins in CD57 ϩ/Ϫ NK cell subsets, whereas four proteins showed RFs above the statistical thresholds, namely GTPB6 (Putative GTPbinding protein 6; CD56 dim/bright ; Log 2 RF ϭ Ϫ0.8); ZN574 (Zinc finger protein 574; CD57 ϩ/Ϫ ; Log 2 RF ϭ 0.7); CLN5 (Ceroid-lipofuscinosis neuronal protein 5; CD57 ϩ/Ϫ ; Log 2 RF ϭ 0.8), and AMPN (Aminopeptidase N; CD57 ϩ/Ϫ ; Log 2 RF ϭ 0.8). All uniquely identified proteins are highlighted with an * in supplemental Tables S4 and S5). This minor difference regarding the number of proteins identified by our two-dimensional proteome strategy indicates that NK cell development does not depend on the presence but on the regulation of certain proteins. By evaluating individual donors, the comparison of CD56 dim versus CD56 bright NK cell subsets revealed 2941 common proteins, whereof 40% (1162) could be detected in all investigated donors. Corresponding numbers for the MS analyses of CD57 ϩ versus CD57 Ϫ NK cells were 3224 and 1427 (44%) (supplemental Fig. S1). The 2941 identified proteins of the CD56 dim versus CD56 bright MS analysis are listed in supplemental Table S4 (supplemental excel file: CD56dimversusCD56bright.xls) and the detected proteins of the CD57 ϩ versus CD57 Ϫ NK cell subset analysis (3224) are summarized in supplemental Table S5 (Supplemental excel file: CD57ϩversusCD57-.xls), including their Mascot Protein Scores, number of unique peptides, log 2 regulation factors and protein coverage for each individual donor investigation and as median value, including the five investigated donors.
All identified proteins were analyzed using GeneGo TM , a tool which gives an overview about preferentially covered pathways. This bioinformatics analysis gave directly insight into the functionality of the investigated CD56 ϩ NK cell proteomes and revealed predominant expression of two main biological processes: cytoskeleton re-arrangements and immune response pathways. Thereby, we observed striking similarities in pathway coverage between CD56 dim/bright compared with CD57 ϩ/Ϫ NK cells, as indicated before ( Fig. 2A). Also numerous NK cell signaling pathways were found highly represented by the identified proteins, i.e. CD16 signaling (identified proteins covered 50% of the entire pathway); formation of the immunological synapse (56%) and Granzyme A signaling (86%) (Fig. 2B). Hence, the GeneGo TM analysis confirmed a functionally consistent protein set in CD56 ϩ NK cells subsets and underlines the presence of a putative core proteome. 13 proteins were unique in CD56 dim/bright NK cells and 54 in CD57 ϩ/Ϫ NK cells. B, Data sets (all proteins, n ϭ 5, median) were loaded into GeneGo TM . This program assigns proteins to certain pathways by computing a statistical relevance value (negative log p value). The negative log p value depicts the ratio of proteins identified in this study and proteins known to be present within this pathway. Hierarchical order was defined by CD57 Ϫ NK cell subsets. Numbers depict the order within the CD56 bright/dim GeneGo TM analysis.
Evaluating NK Cell Specific Protein Expression-Donor and subset specific peptide samples were labeled differentially with iTRAQ reagents for relative quantification in addition to peptide sequencing. Thus, each individual MS fragment ion spectrum provides quantitative information in parallel to the identified peptide sequence. The ratios of iTRAQ reporter intensities were used to determine the relative protein abundances in the different CD56 ϩ NK cell subsets-1D-LC-MS/MS experiments were performed with 1% of each sample to confirm the completeness of peptide labeling. Next, manual inspection of the quantitative data using Scaffold uncovered a set of well-described NK cell-specific proteins. For example, expression of CD56 (NCAM1) and Perforin (PERF) is fully in accordance with the literature: CD56 expression declines from CD56 bright to CD56 dim NK cells, whereas Perforin abundance increases simultaneously (Figs. 3A and 3B) (37). Furthermore, a stable CD56 expression in CD57 ϩ and CD57 Ϫ NK cells was recently described (15). MS data of this study indeed confirm non-regulated CD56 expression profiles and in parallel revealed a slight increase in Perforin abundance in CD57 ϩ NK cells. Thus, abundances of well-described NK cell proteins are in perfect accordance with the literature (supplemental Table S6) (15)(16)(17)37). Regarding to NK cell differentiation the level of cytotoxicity related proteins, like Perforin, multiple Granzymes and LAMP1/3 increases from CD56 bright to CD56 dim NK cells and induction of Perforin, as well as KIR expression coincides with CD57 appearance (15,37).
Characterizing Donor-specific Variations-Although the blood donor volunteers were clinically approved to be healthy, we paid particular attention to determine potential individual variations.
A global inspection of regulatory data from each donor was performed by box plot analysis. Protein regulation factors in assessed donors were symmetric and normally distributed within the box plots, which indicates data robustness and only minor donor variations in both separate comparative analyses of CD56 dim/bright and CD57 ϩ/Ϫ NK cell data. A wide regulation range from ϩ3.8 to Ϫ4.0 (log 2 RF ϭ log 2 regulation factor) was detected in CD56 dim/bright NK cells and only minor regulation differences in CD57 ϩ/Ϫ NK cells (ϩ2.3 to Ϫ2.3; Fig. 4A, right hand panel).
Next, regulatory information were sorted at the protein level and inspected by heat maps (Fig. 4B). Generally, the expression profiles indicate a remarkable ratio of proteins similarly regulated in each set of five donor samples. As already indicated by the box plot analyses, a higher total number and more pronounced protein regulations were observed in CD56 dim/bright NK cells compared with CD57 ϩ/Ϫ NK cells. Only slight donor variations could be detected for donor 3 of the CD56 dim and CD56 bright NK cell subsets and donor 8 of the CD57 ϩ/Ϫ NK cell subsets. Hence, the heat map analyses generally confirmed reproducible quantification and highly conserved protein regulation signatures within both sets of donors. Individual variations are limited and thus likely not compromise the definition of conserved regulated proteins. However, their identification might be of interest as clinical markers. Thus, we characterized strongly deviating proteins, which occur donor-dependently, through MAD analysis. Proteins with a MAD higher than 0.3, were considered to show donor-dependent protein regulations. The vast majority of proteins exhibited remarkable consistent protein. However, 13 proteins were found with MAD values higher than 0.3 and therewith donor-dependent protein deviations. Ten of those cases were detected in CD56 bright/dim and three in CD57 ϩ/Ϫ NK cells. The 13 candidates include GRAH (Granzyme H) and GNLY (Granulysin), which were already previously described to be donor-dependently regulated in human primary lymphocytes (38, 39) (supplemental Fig. S2). Another strongly deviating candidate is CD44, whose regulation occurs infectiondependently in mice and steers human postnatal thymocyte development (40,41). CD63 (Granulophysin) is well characterized in NK cells and was shown to be differentially regulated in decidual human primary NK cell subsets (42). CATW (Cathepsin W) is exclusively expressed in CD8 ϩ T cells and NK cells. This protein may mediate cytolytic activity, but is not essentially involved in cytotoxicity (43). Unfortunately, donor-dependent expression patterns of CD44, CD63, CATW and the other proteins with high MAD values have not been accessed to this date in primary human NK cells.
In conclusion, donor dependent variations were observed to a minor degree (13 proteins among 3412) in this study. Regulations of well-described NK-specific proteins serve without exception as proof-of-concept data. A broader dynamic range of regulations coinciding with the step from CD56 bright to CD56 dim , compared with the subsequent step from CD57 Ϫ to CD57 ϩ , confirmed the recently suggested differentiation model. On the other hand, regulations observed for proteins so far not described in this context likely contribute to the biology of distinct NK cell subsets, including NK cell effector functions, status of differentiation and responsiveness.
Subset-specific Protein Regulation in CD56 ϩ NK Cell Development-After investigating donor variation, particular attention was paid to the detection of the most significant and subset-specific protein regulations. Therefore, a statistical model was developed to detect the most pronounced and conserved protein regulations (36). Standard deviations were calculated based on the assumption that log 2 -regulation factors of each protein follow a normal distribution, taking the limited number of 5 donors into account. The 3412 identified proteins were used to estimate the standard deviation and this served as a base to determine statistical significance thresholds depending on the number of donors. Notably, the thresholds considering only 3 donors (0.83 for CD56 dim/bright and 0.51 for CD57 ϩ/Ϫ NK cells) did not vary significantly from thresholds calculated for four or five donors (log 2 RF CD56dim/bright : 0.809 and 0.745 and log 2 RF CD57Ϯ : 0.503 and 0.467; see supplemental Table S8). Hence, the thresholds 0.83 for comparing CD56 dim versus CD56 bright NK cells and 0.51 for CD57 ϩ versus CD57 Ϫ NK cells were applied to extract subset-specific protein regulations. Although the threshold for CD56 dim/bright NK cell subsets was significantly higher than in CD57 ϩ/Ϫ NK cells, a list of 31 proteins was generated (Table I), instead of 5 proteins in CD57 ϩ/Ϫ NK cell subsets (Table II).
None of these 36 proteins belongs to the group of donordependently regulated candidates with MAD values over 0.3. Thus, the regulation of these 36 proteins can be directly related to the distinct developmental NK cell stages. They can be grouped into the following categories: (1) NK signaling; (2) cytoskeletal dynamics; (3) differentiation and (4) cytotoxicity. Top-regulated novel CD56 ϩ NK cell protein candidates (11 proteins; depicted in bold; with Ϫ1.0Ͻlog 2 RFϾ1.0, Fig. 5) are comparatively and subset-specifically summarized together with six key components of NK cell biology in FIGURE 5. Among those, the expression of the AK1C3 (Aldo-keto reductase family 1 member C3) protein was highest in CD56 dim NK cells together with Perforin and declined within the last step of differentiation into CD57 ϩ NK cells. Expression of NHRF1 (Na ϩ /H ϩ exchange regulatory cofactor NHE-RF1), SH2K1 (SH3 domain-containing kinase-binding protein 1), EFHD2 (Swiprosin), ANXA2 (Annexin A2) and PTCA (Protein tyrosin phosphatase receptor type C-associated protein) increased during the first differentiation step from CD56 bright into CD56 dim NK cells and was stagnating during the last step from CD57 Ϫ to CD57 ϩ NK cells.
However, we also detected proteins highly abundant in CD56 bright NK cells, with a decline in expression during all following differentiation steps (from CD56 dim to CD57 Ϫ and finally to CD57 ϩ NK cells), e.g. FLNB (FilaminB), LMNA (PrelaminA/C), and COTL1 (Coactosin-like protein), which indicates a pivotal role especially in this early NK cell subset. In total three proteins showed continuous increase in expression during both differentiation steps, namely S100A4 (Calvasculin), S100A6 (Calcyclin), and LEG1 (Galectin-1). LEG1 was previously described in NK cells, but without NK cell subsetspecific regulation. Of note, all three proteins were shown previously to be involved in the thymocyte differentiation process (52). S100A4 and S100A6 are Recruited to the NKIS and Colocalize with Myosin IIa-NK-cell cytotoxicity depends on the formation of the F-actin-rich NKIS and the Myosin IIa (MYH9) mediated transport of Perforin-containing lytic granules to the NKIS (3)(4)(5). In this study, we detected MYH9 expression highest in mature CD56 dim (log 2 RF ϭ 0.7), but reduced in terminally maturated CD57 ϩ NK cells (log 2 RF ϭ 0.1), confirming its relevance for cytotoxic NK effector functions. Like MYH9, other subset-specific proteins exhibit a similar expression profile, e.g. S100A4 and S100A6. Notably, MYH9 was previously shown to interact with S100A4 in murine fibroblasts (NIH3T3) and human endothelial cells (CSML-0 and Ϫ100) (44,45). Therefore, both proteins (S100A4 and S100A6) were selected for subcellular localization studies that were performed using cytotoxic CD56 dim (with high Perforin expression levels) NK cells to analyze S100A4 and S100A6 contribution to the immunological synapse of NK cells. NK activation was performed by coincubation with their natural targets -K562 cells that we conducted in a time-resolved manner for 5, 10, 15, 20, and 30 min. The induced formation of the NKIS was documented by fluorescence microscopy. Staining for Perforin together with CD107a (46) and MYH9 (3) allowed following the dynamics of formation and recruitment of lytic granules to the interphase between NK and target cell. All three proteins localized time-dependently into the NKIS, thus favoring this generally Fig. 6 and Fig. 7, left panel). Intensity of colocalization was determined with the help of 3D density surface plots, generated by ImageJ.
Microscopic investigation revealed time-resolved colocalization events of the here characterized NK cell proteins S100A4 and S100A6 together with MYH9. First colocalization between S100A4 and MYH9 was observed already after 5 min, formation of complete complexes after 10 min that peaked after 15 min, when specific recruitment and intense accumulation of this complex to the NKIS was observed (Fig.  6). Instead, peak accumulation of the S100A6 and MYH9 complex occurred already after 10 min (Fig. 7). The disassembly of this complex then starts immediately after 15 min whereas the S100A4 and MYH9 interactions lasted longer and started to disintegrate after 20 min and were apparent for  further 10 min until full disintegration and detachment of the NK cell from the target cell.
To summarize, complex formation of S100A6 and MYH9 occurred very quickly, but between S100A4 and MYH9 much more intense and over a longer period of time. Hence, the microscopic investigation on the NK cell subset specifically regulated proteins S100A4 and S100A6, revealed distinct localization of both proteins at the NKIS after NK cell activation, followed by intense colocalization together with MYH9.

DISCUSSION
Research on NK cell development has rapidly emerged during the past few years, starting with the introduction of the four-step NK cell differentiation model, leading to fully active CD56 bright cells (9), their conversion into CD56 dim cells (12,13) and their final complementation into terminally differentiated CD56 dim CD57 ϩ NK cells (15,17,18).
NK cell subset analyses mostly relied on FACS-based approaches and microarray analyses to determine subset specific protein expression from ex vivo expanded primary NK cells (11,15,37). As a complementation of these approaches, mass spectrometry can provide an unbiased and systematic overview about protein expression and regulation in the distinct developmental stages. Here, we provide the first insight into the proteome of primary CD56 ϩ NK cells of distinct developmental stages isolated from healthy human blood donors. We investigated the proteomes of CD56 bright versus CD56 dim and CD57 Ϫ versus CD57 ϩ NK cell subsets. Quantitative data were determined for 3400 proteins and describe already known as well as novel functional aspects of NK cell subsets. Notably, regulatory data of well-characterized NK cell proteins were in accordance with gene expression data from previous studies (16,37). CD56 bright cells characterize an immune regulatory NK cell subset able to release various pro-inflammatory cytokines for activating, modulating and recruitment of adaptive immune cells (11). CD56 dim NK cells also release cytokines, but their main biological function encompasses the detection of transformed and virus infected target cells, followed by target cell lysis (47). This cytotoxic capacity requires expression of cytolytic components, like Granzymes, Perforin, and Granulysin, as well as vesicle forming and transport related proteins (CD107a, Rab proteins, WASP), also involved in endocytosis. Our proteome approach covered a significant portion of known cytolytic and vesicle-forming components, which were consistently, detected less abundantly in CD56 bright NK cells but significantly higher amounts in CD56 dim cells that present the mature and prototypic cytotoxic NK subset.
Recently, it was reported that CD57 expression occurs independently of, but simultaneously with KIR acquisition and thus NK cell differentiation proceeds un-coupled, but in parallel to NK cell education. Our proteome results confirmed this observation, because KIR2DS2, KIR3DL2, and KIR3DL1 were solely detected regulated in the CD57 ϩ NK cell subsets (supplemental Table S6). This study also provides evidence for the FIG. 5. Changes in protein expression during distinct stages of NK cell differentiation. Selected proteins are shown together with their corresponding regulation intensity in CD56 bright/dim and in CD57 Ϫ/ϩ NK cells. Proteins with previously undescribed regulation in NK cells are shown in bold. Regulation intensity is depicted with the following symbols: ϩϩϩ log 2 RF over 1.5; ϩϩ log 2 RF over 1.0; ϩ log 2 RF over 0.5;/unregulated (in between Ϫ0.5 and 0.5); -less than Ϫ0.5; -less than Ϫ1.0; -less than Ϫ1.5. specialization of terminally mature CD57 ϩ NK cells, because expression of KIRs and broad coverage of CD16 signalingassociated components (50% of the whole pathway) were determined only for this NK cell subset. It has been shown previously, that the intensity of NK cell cytotoxicity occurs donor-dependently (48,49). Interestingly, donors of this study were clinically approved as healthy but statistical proteome analyses revealed slight but significant variations that might serve as markers to better understand these donor-variations. Although the majority of identified regulations (Table I and  Table II) were found conserved and widely donor-independent, a limited number of protein regulation showed a MAD value higher than 0.3 (in summary 13 proteins). Among them we detected nine proteins with significant regulations, crossing the statistical threshold (supplemental Table S7). Four of them, GRAH (Granzyme H), CATW (Cathepsin W), and CD63 (LAMP-3), are well-characterized NK cell specific proteins involved in cytotoxicity and antigen processing, whereas CD44 is also known as marker for early T cell development in the thymus. The other three, protein kinase C (PACN1), SH2 domain containing protein 1A (SH21A) and MAP kinase 3 (MK03), are generally involved in NK cell signaling but were characterized as regulated in distinct developmental stages here for the first time. Unfortunately, the function of FETUA (Alpha-2HS glycoprotein) and CLIC3 (Chloride intracellular channel protein 3), showing most pronounced regulation, remain unknown in NK cells.
Particular attention was paid to determine statistically conserved NK subset-specific proteins. Thirty six proteins were detected by this approach and functionally categorized. LEG1 (Galectin-1), S100A4 (Calvasculin), and S100A6 (Calcyclin) occurred with pronounced increasing expression profiles both during the first and the second maturation step, indicating their general relevance in NK cell biology. LEG1 was described previously as significantly expressed and secreted by activated T reg cells from mice and men (50,51). Furthermore, it was shown that LEG1, S100A4 and S100A6 promote thymocyte differentiation (52). S100A4 and S100A6 belong to the calcium-binding family of S100 proteins and are generally involved in cell cycle progression and differentiation. For S100A4 a direct interaction with MYH9 was determined at the leading edge of migrating cells, which naturally contains a high calcium concentration (45). Because MYH9 is also known to play an important role by mediating the interaction FIG. 6. Time-resolved colocalization studies of S100A4 and MYH9 in activated NK cells. NK cells were incubated and activated with K562 target cells ('T' for target cell and "NK" for NK cell) for 5, 10, 15, 20, and 30 min. First two rows show proof of concept colocalization events between Perforin (red) and CD107a (green), and between Perforin (red) and MYH9 (green). Time-resolved activation of NK cells and stain for S100A4 (red) and MYH9 (green) and DNA (DAPI, blue) with appropriate antibodies revealed remarkable time-dependent formation of colocalization events at the NKIS. Anti S100A4 was stained with goat anti-mouse IgG (Alexa 594) and anti-MYH9 (Myosin IIA) with goat anti-rabbit IgG (Alexa 488) supplemented with DAPI (1:1000). Imaging was performed on an inverted microscope (Axiovert 100TV; Carl Zeiss, Jena, Germany) microscope using standard epifluorescence illumination (light source HXP120, Zeiss) and 63ϫ/NA1.4 or 100ϫ/NA1.4 plan-apochromatic objectives. Imaging was performed at room temperature with immersion oil. Images were acquired with a back-illuminated, cooled charge-coupled-device camera (CoolSNAP HQ2, Photometrics, Tucson, AZ, USA) driven by Metamorph software (Version 7.5.3.0; Molecular Devices Corp., Downingtown, PA, USA). Five independent experiments were performed and 20 -30 images were acquired per coverslip in each experiment. Image analysis was performed with ImageJ (version 1.44p).
of cytotoxic granules with F-actin for further transport to NKIS (3-5), we were wondering whether S100A4 and S100A6 are involved in these processes. Time-resolved localization studies actually revealed the recruitment of S100A4 and S100A6 together with MYH9 to the NKIS in cytotoxic CD56 dim NK cells, with peak levels after 10 and 15 min (see Fig. 6 and Fig.  7). These findings confirm their involvement in NKIS-mediated responses, although their direct functional contribution remains at this time uncharacterized. S100A4 can regulate MYH9 activity (45) and might be an indirect mediator of cytotoxic granule transport.
Further candidates were detected regulated mostly at individual developmental stages, and their functional characterization will likely help to unravel the unique capabilities of distinct NK subsets: The AK1C3 (Aldo-ketoreductase family 1 member C3) converts steroid aldehydes and ketones into alcohols and may play a role in controlling cell growth and/or differentiation. AK1C3 was previously shown to be expressed in NK cells and was also suggested as marker for "healthy and normal" NK cells (53). This proteomic study revealed a subset-specific expression profile for AK1C3 that was detected most abundant in cytotoxic CD56 dim NK cells. This suggests AK1C3 as marker for fully functional and cytotoxic NK cells and a role in NK cell effector functions.
On the other hand, FLNB (Filamin B) was exclusively identified in CD56 bright NK cells in high amounts. Kanters and colleagues identified the actin cross-linking molecule Filamin B as a novel binding-partner for intracellular adhesion molecule-1 (ICAM-1) and confirmed the ICAM-driven transendothelial migration of leukocytes (54). Thus, Filamin B may be an important mediator of transendothelial migration of CD56 bright NK cells from secondary lymphoid tissues (SLO) to peripheral blood during maturation.
In conclusion, results of this study fully support the recent NK cell differentiation model and revealed a limited number of novel donor-independent and subset-specific protein regulations. The systematic inspection of distinct primary NK development stages thereby helps to prioritize candidates such as LEG1, AK1C3, and Filamin B for perspective functional studies. S100A4 and S100A6 most likely are of general importance in NKIS-mediated responses. Following the example of this time-resolved localization studies, single molecule high-resolution imaging may guide us likely to the next molecular hallmarks of NK cell biology. FIG. 7. Time-resolved colocalization studies of S100A6 and MYH9 in activated NK cells. NK cells were incubated and activated with K562 target cells ('T' for target cell and "NK" for NK cell) for 5, 10, 15, 20, and 30 min. First two rows show proof of concept colocalization events between Perforin (red) and CD107a (green), and between Perforin (red) and MYH9 (green). Time-resolved activation of NK cells and stain for S100A6 (red) and MYH9 (green) and DNA (DAPI, blue) with appropriate antibodies revealed remarkable time-dependent formation of colocalization events at the NKIS. Anti S100A6 was stained with goat anti-mouse IgG (Alexa 594) and anti-MYH9 (Myosin IIA) with goat anti-rabbit IgG (Alexa 488) supplemented with DAPI (1:1000). Imaging was performed on an inverted microscope (Axiovert 100TV; Carl Zeiss, Jena, Germany) microscope using standard epifluorescence illumination (light source HXP120, Zeiss) and 63ϫ/NA1.4 or 100ϫ/NA1.4 plan-apochromatic objectives. Imaging was performed at room temperature with immersion oil. Images were acquired with a back-illuminated, cooled charge-coupled-device camera (CoolSNAP HQ2, Photometrics, Tucson, AZ, USA) driven by Metamorph software (Version 7.5.3.0; Molecular Devices Corp., Downingtown, PA, USA). Five independent experiments were performed and 20 -30 images were acquired per coverslip in each experiment. Image analysis was performed with ImageJ (version 1.44p).