As a first approach to phenotype NK cells in adipose tissue, we performed LEGENDScreen™ surface proteomic screening to evaluate 315 surface receptors and compare their expression on NK cells in blood (peripheral blood mononuclear cells; PBMCs) and SAT from donors ( Figure 1A ). Substantial overlap in protein expression was observed between total NK cells in blood and adipose tissue, displaying high expression levels of CD11a, CD47, CD48, and CD50 ( Figure 1B ). Several proteins were identified as either blood- or adipose tissue-enriched, such as higher expression of CD181, CD274, and CD137L on SAT NK cells and higher CD99, NKp80, CD81, and GPR56 on blood NK cells ( Figure 1C ). The CD56^bright NK cells are normally considered the tissue-resident subset, and to identify potential adipose tissue-resident NK cell markers, we investigated proteins that were more highly expressed by these cells relative to the CD56^dim NK cells. Interestingly, levels of several proteins, such as CD63, CD54, and CD183, were expressed by higher proportions of the CD56^bright cells in SAT ( Figure 1C ). Taken together, we identified surface proteins enriched on NK cells in blood and adipose tissue, respectively. Moreover, we defined surface proteins characteristic of CD56^bright NK cells in adipose tissue that may potentially define adipose tissue-resident NK cell subsets. Of note, the tissue residency marker, CD69, did not emerge as tissue-specific from the proteomic analysis (data not shown), emphasizing certain limitations with this approach and the need to validate our findings.

To validate the LEGENDScreen™ findings and to further characterize the CD56^bright and CD56^dim NK cells in blood and adipose tissue, we generated a 27-parameters flow cytometry panel that combined established NK cell-associated markers (e.g., granzyme B, perforin, T-bet, Eomes, KIRs, NKp46, NKG2A, and NKG2D) with the adipose tissue-enriched CD56^bright NK cell proteins identified from the screening (CD26, CD54, CD63, CCR5, and Sialyl Lewis^x). CD69 was also included in the panel as an indicator of tissue residency. The protein expression was measured on CD56^bright and CD56^dim NK cells from matched blood, SAT, and VAT samples of 43 individuals with obesity, undergoing bariatric surgery ( Table 1 ) and the gating strategy is presented in Figure 2A .

The proportion of total NK cells was significantly higher in adipose tissue relative to blood and also higher in SAT than in VAT ( Figure 2B ). Similarly, depot-specific differences were found for the relative abundance of CD56^bright and CD56^dim subsets ( Figure 2C ). As expected, the CD56^bright NK cells constituted on average around 10% of the NK cell population in blood, a distribution pattern that was also seen in SAT. In VAT, however, the CD56^bright NK cells made up on average almost 50% of the NK cell compartment.

The expression levels of several proteins differed significantly between the CD56^bright and CD56^dim subsets ( Figures 2D, E , Supplementary Figures 1, 2 ). As expected, the CD56^dim NK cells displayed a cytotoxic- and maturation profile, indicated by consistently higher levels of perforin, granzyme B, T-bet, and KIRs. Interestingly, a proportion of the CD56^bright cells was also positive for these markers, suggesting some level of cytotoxic potential for this subset. Additionally, CD56^bright cells in the adipose tissue expressed higher levels of NKG2A and CD69, which supports the assumption that CD56^bright NK cells represent the major tissue-resident population in adipose tissue. Interestingly, not all CD56^bright NK cells were positive for CD69, indicating that there are subsets that do not display classical tissue residency phenotypes ( Figure 2D ). The proteins identified from LEGENDScreen™ also displayed differential expression between the two subtypes, with significantly higher levels of CD26, CD54, CD63, CCR5, and Sialyl Lewis^x detected on the CD56^bright NK cells in adipose tissue compared with CD56^dim NK cells ( Figure 2D ).

Substantial variations in NK cell protein expression were observed between the various tissue compartments ( Figure 2D ). This was particularly evident for the CD56^bright subset, of which CD54, CD63, CD69, CCR5, Eomes, and NKG2D, were not only enriched in adipose tissue compared with blood but also significantly higher in VAT than in SAT. Conversely, CD26 and Sialyl Lewis^x emerged as SAT-enriched proteins, although CD26 was also expressed by nearly 50% of the CD56^bright NK cells in VAT. Perforin, granzyme B, KIRs, T-bet, and CD49e displayed the opposite tissue expression pattern on the CD56^bright NK cells, with significantly reduced levels in adipose tissue compared to blood, and lower in VAT than in SAT. The CD56^dim NK cells also showed tissue-specific phenotypes, although less striking. Similar to the CD56^bright subset, NKG2D was increased on the CD56^dim NK cells in adipose tissue compared with blood, also displaying a higher VAT/SAT ratio ( Figure 2D ). Furthermore, a small fraction of the CD56^dim NK cells in adipose tissue, particularly in VAT, were positive for CD69, indicating that subsets of CD56^dim NK cells also exhibit signs of tissue residency. In summary, our data suggests the existence of tissue-resident CD56^bright NK cells enriched in VAT, characterized by the expression of additional surface markers that may potentially reflect adaptations to the microenvironment in this tissue depot.

Since adipose tissue NK cells are suggested contributors for the development of insulin resistance and metabolic disease, we performed a correlation analysis between the individual proteins expressed by the CD56^bright and CD56^dim NK cell subsets and glucose and lipid parameters related to cardiometabolic disease. Interestingly, no significant associations were found between proteins on NK cells in blood, SAT, or VAT and HOMA-IR, as a measure of systemic insulin resistance ( Figure 3A ). However, we did observe several, mostly negative, statistically significant associations between glucose parameters and protein expression in blood, such as between the long-term glucose indicator HbA1c and CD38 on both NK cell subsets, and between glucose and NKG2D on the CD56^bright NK cells ( Figure 3A ). In adipose tissue, the few identified correlations with nominal significance were found with lipid parameters, such as positive correlations between both LDL and total cholesterol and CCR5 expression on the CD56^bright NK cells in SAT ( Figure 3A ). Given the observed associations of HbA1c and fasting glucose with NK cell proteins, we further investigated potential differences in NK cells in patients with and without T2D. However, in our material we did not find significant differences between the two groups, neither the NK cell abundance nor NK cell protein expression, which may possibly be due to the relatively low number of people with T2D in our cohort (n=9) ( Figure 3B , Supplementary Figure 3 ). In summary, we observed some significant associations between NK cell proteins and circulating glucose- and lipid parameters, however, the weak associations with systemic insulin resistance and T2D in our cohort suggest that expression of individual proteins may not provide sufficient power to identify such link.

For a higher resolution depiction of NK cell diversity, we employed the high dimensional reduction technique, Uniform Manifold Approximation and Projection (UMAP), on CD56^bright and CD56^dim NK cells from blood, SAT, and VAT to identify protein combinations on a single cell level. UMAP identified multiple clusters of CD56^dim and CD56^bright NK cells deriving from the different tissue compartments ( Figure 4A ). The CD56^dim NK cells from both blood, SAT, and VAT primarily co-localized in two clusters, whereas the CD56^bright NK cells widely dispersed into multiple SAT and VAT-derived clusters, suggesting higher degree of phenotypic diversity among this subset ( Figure 4A ). The UMAP plots display variation in expression levels of each of the proteins in the flow cytometry panel across the different clusters within both the CD56^dim and CD56^bright NK cell populations ( Figure 4B ). CD38 and Sialyl Lewis^X in CD56^bright cells from VAT and SAT, respectively, and variation of KIR expression in CD56^dim cells from all compartments were examples of surface proteins showing high variation ( Figure 4B ). Thus, the analysis identified multiple clusters of CD56^bright and CD56^dim NK cells with different levels of protein expression, pointing to the existence of several subpopulations of NK cells.

Indeed, PhenoGraph clustering analysis identified 18 defined clusters of NK cells within the UMAP plot with various size proportions ( Figures 4C-E , Supplementary Figure 4 ). Most clusters were enriched with either CD56^dim or CD56^bright NK cells, with the exception of cluster 6, containing equal proportion of both subsets ( Figure 4F , left). The majority of clusters contained cells deriving from all tissue compartments ( Figure 4F , right), but as expected, the VAT-enriched (#9, #16, #18) and the SAT-enriched (#8, #11) clusters were dominated by CD56^bright NK cells, and the blood-enriched cluster (#2) were dominated by CD56^dim NK cells ( Figure 4E, F ). A hierarchical clustering, summarizing protein expression patterns across the 18 clusters, confirmed the similarities between the VAT-enriched CD56^bright NK cell clusters (#9, #16, #18), all displaying considerably higher levels of CD63, CD69, and NKG2D compared with the other clusters ( Figures 4G, H ). However, differences were also observed across these clusters, such as higher levels of Eomes, NKG2A, and CD38 in #9 and CCR5 and KIRs in #16, revealing proteins that may define the different subclusters of CD56^bright NK cells in VAT. As expected, the clusters enriched with CD56^dim NK cells (e.g., #1, #2, #10) shared phenotypic signatures, such as higher levels of perforin, granzyme B and T-bet ( Figure 4G ).

Given the few associations observed between the single proteins and metabolic parameters ( Figure 3A ), we next investigated the relationship between the NK cell clusters and clinical traits. To this end, the patients were stratified into “low” and “high” based on the median value of the clinical parameters from all patients ( Table 1 ). Overall, the clusters shared relatively similar distribution of NK cells deriving from “low” and “high” patients for most of the parameters, with exception of cluster #14 that contained higher proportions of cells from patients with “high” BMI and HOMA-IR (77% and 78%, respectively) and cluster #16 that contained higher proportions of cells from patients with “low” BMI, HOMA-IR, and LDL (74%, 80%, 79%, respectively) ( Supplementary Figure 5 ).

Taken together, by employing high dimensional clustering analysis, we identified protein signatures defining distinct subsets of CD56^bright and CD56^dim NK cells in blood, SAT, and VAT from people with obesity. Importantly, we discovered potential subpopulations of tissue-resident CD56^bright NK cells, mainly residing in VAT, defined by the co-expression of CD63, CD69, and NKG2D.

Several clinical cohorts were included in the current study. This was approved by the Regional Committees for Medical and Health Research Ethics in Bergen (REK: 2010/502 and 2015/2343) and written consent was obtained from all participants. First, subcutaneous adipose tissue from liposuction aspirates of individuals undergoing plastic surgery was used for proteomic screening. Buffy coats from anonymous blood donors were used as blood samples and internal controls for the cell surface screening. For the 27-color flow cytometry analysis, matched fasting blood samples and subcutaneous- and visceral adipose tissue biopsies were obtained from 43 patients with obesity undergoing bariatric surgery at Voss Hospital. The clinical characteristics and biochemical measurements of this cohort are presented in Table 1 .

Subcutaneous adipose tissue and visceral adipose tissue biopsies obtained from bariatric surgery were immediately stored in Krebs-Ringer Phosphate (KRP) buffer until processing. The biopsies were cut into pieces and enzymatically digested with collagenase Type I (Life Technologies) for 1 hour at 37°C with constant shaking. The subcutaneous adipose tissue from liposuction was washed in 0.9% NaCl, diluted in KRP buffer and digested with Liberase (Roche). The dissolved tissues were filtered, and the stromal vascular cells (SVC) were isolated from the mature adipocytes and washed with Phosphate Buffered Saline (PBS). The liposuction SVC was freshly stained, whereas the SVC from biopsies were preserved in freezing media containing FBS and 10% DMSO and stored in liquid nitrogen until further flow cytometry experiments. Peripheral blood mononuclear cells (PBMC) were isolated from blood of bariatric surgery patients and healthy donors using density gradient centrifugation, as previously described (33).

LEGENDScreen™ Human PE kit (Biolegend, Cat#700007) was used to screen the cell surface proteome. Fresh SVC from subcutaneous liposuction aspirates and PBMC from buffy coat blood were used in the screening as previously described (33). In brief, the tissue-derived and blood-derived cells were CD45 barcoded and stained with a backbone antibody panel allowing identification of the cells of interest. The cells were added to the plates provided in the LEGENDScreen™ kit and the staining was performed according to the manufacture protocol. Plates were run on an 18-color flow cytometer (LSR Fortessa, BD Biosciences) with 407, 488, 561 and 640 lasers using the BD FACSDiva™ Software (BD Biosciences). Flow cytometry data was analyzed using FlowJo v10 (Treestar, USA). Two manual gating approaches were performed to identify the NK cell subsets, a “broad” gating followed by a “fine” gating. Several adipose-enriched proteins identified from both gating approaches were validated for expression in additional adipose tissue samples, and a selection of these proteins was included in the 27-parameter flow cytometry experiment.

Frozen PBMC and SVC were thawed and stained with an extracellular primary antibody panel containing the following antibodies from BD Biosciences: BUV496 CD16 (Cat# 612944), BUV563 CD56 (Cat# 565704), BUV661 CD38 (Cat# 612969), BUV737 CD69 (Cat# 612817), BUV805 CD45 (Cat# 612891), BB700 NKG2A (Cat# 747926), APC Vio770 CCR5 (Cat# 557755), V500 CD14 (Cat# 561391), BV510 CD19 (Cat# 562947), BV650 CD98 (Cat# 744505), BV711 Sialyl Lewis x (Cat# 563910), PE-Cy5 CD54 (Cat# 555512). From Biolegend: Biotin NKp46 (Cat# 331906), A700 CD63 (Cat# 353024), BV421 Bcl-2 (Cat# 658709), BV510 CD123 (Cat# 306022), BV750 CD3 (Cat# 344845), BV785 HLA-DR (Cat# 307642), PE CD26 (Cat# 302706), PE-Cy7 NKG2D (Cat# 320812), PE-Cy7 CD162 (Cat# 328816), BB515 CD49e (Cat# 130-110-534, Miltenyi), PE-Cy5.5 KIR2DL1/S1 (Cat# A66898, Beckman Coulter), PE-Cy5.5 KIR2DL2/L3/S2 (Cat# A66900, Beckman Coulter). Staining was performed in FACS buffer (PBS with 2mM EDTA (Cat# AM9260G, Ambion), 2% FCS (Cat# F7524, Sigma). After 20 min incubation at room temperature (RT) and in the dark, the cells were washed twice with FACS buffer and further stained with a secondary antibody panel containing streptavidin (Cat# 624294, BD Biosciences) for 15 min at RT in dark. Cells were washed twice with FACS buffer before adding Fix/perm Buffer (diluted ¼ with eBioscience reagents: Fix/perm diluent (Cat#00.5223.56) and Fix/perm concentrate (Cat#00.5123.43)) and incubated for 45 min RT in dark. Cells were then washed twice with perm/MQ buffer (Perm Buffer 10X (Cat#00.8333.56) diluted 1/10 with MQ water). Further, the cells were stained with an intracellular antibody panel containing the following antibodies from BD Biosciences: BUV395 Ki67 (Cat# 564071), BB755-P Perforin (Cat# 624391, BD Horizon custom reagents), BB790-P Granzyme B (Cat# 624296, BD Horizon custom reagents), and eF660 Eomes (Cat# 50-4877-41, eBioscience), PE-Dazzle594 T-bet (Cat# 644828, Biolegend). Live/dead Fixable Aqua Dead Cell stain kit (Invitrogen, 1:100 dilution) was used to distinguish between dead and live cells. After 30 min incubation, the cells were washed twice with perm/MQ buffer and kept in FACS buffer for immediate flow cytometry analysis. Samples were run on a 29-color Symphony (BD Biosciences) with 405, 488, 561 and 639 lasers using the BD FACSDiva™ software (BD Biosciences). Flow cytometry data generated was analyzed using FlowJo V10 (Treestar, USA).

The FlowJo plugins Uniform Manifold Approximation and Projection (UMAP) and PhenoGraph were performed to visualize the single cell flow cytometry data in a high dimensional structure. To this end, CD56^bright and CD56^dim NK cells from PBMC, SAT, and VAT for each patient (n=33) were electronically barcoded for cell subset, tissue of origin and several biochemical parameters (either “low” or “high” based on median values of each parameter). The patient samples containing the same cell subset and tissue type were concatenated (6 samples in total), which was downsampled to 5118 events/file for an equal number of cell input. The samples were concatenated to a final file that was analyzed for UMAP and Phenograph. UMAP was run using the default settings (Euclidean distance function, nearest neighbors: 15 and minimum distance: 0.5). PhenoGraph was run using the default number of nearest neighbors (K = 30). The parameters included in both analyses were the phenotypic markers of interest (n=21) and excluded the proteins used to gate on the NK cell subsets.

Flow cytometry data was analyzed using Prism version 9.2.0 (GraphPad). D’Agostino & Pearson omnibus normality test was used to determine normality of the data. For normally distributed data, one-way ANOVA with Holm-Šídák’s multiple comparisons test was used. When data was not normally distributed, Friedman test with Dunn’s multiple comparisons test was used. A p-value of < 0.05 was considered statistically significant. Correlation analysis was performed in R using the Pearson correlation coefficient and considered significant at adjusted FDR <0.1.