Increased Inflammatory Properties of Adipose Tissue Macrophages Recruited During Diet-Induced Obesity

PKH26 and β-Actin monoclonal antibodies were obtained through Sigma-Aldrich (St. Louis, MO). Anti-CD68 antibody was from R&D Systems (Minneapolis, MN). Unconjugated anti-F4/80 antibodies were from Abcam (Cambridge, MA). PE-Cy5–conjugated anti-F4/80 and isotype antibodies were from eBiosciences (San Diego, CA). Alexa 488 CD11b and isotype controls were from Serotec (Raleigh, NC). CD14 antibodies were from BD Biosciences (San Jose, CA). Anti-Fgr, anti–apolioprotein (apo) E, and anti–G-protein–coupled receptor kinase 3 (GRK3) antibodies were from Santa Cruz Biotechnologies (Santa Cruz, CA).

Male C57BL/6 mice were rendered obese by feeding a high-fat diet consisting of 45% of calories from fat (Research Diets, New Brunswick, NJ) starting at 8 weeks of age for 20 weeks or the indicated times. Normal diet controls were fed a diet consisting of 4.5% of calories from fat (Labdiet, St. Louis, MO). CCR2 knockout mice on a C57BL/6 background were kindly provided by Steven Kunkel (University of Michigan). Animals were housed in a specific pathogen–free facility with a 12-h light/dark cycle and given free access to food and water. Food intake was monitored by assessing chow weight every 5 days in individually housed mice over 20 days. All animal use was in compliance with the Institute of Laboratory Animal Research Guide for Care and Use of Laboratory Animals and approved by the University Committee on Use and Care of Animals at the University of Michigan.

Epididymal fat pads from male C57BL/6 mice fed a normal diet or high-fat diet were excised and minced in PBS with 0.5% BSA. Tissue suspensions were centrifuged at 500g for 5 min to remove free erythrocytes and leukocytes. Collagenase (Sigma-Aldrich) was added to 1 mg/ml before incubation at 37°C for 20 min with shaking. The cell suspension was filtered through a 100-μm filter and spun at 300g for 5 min to separate floating adipocytes from the stromal vascular fraction (SVF) pellet.

For immunofluorescence and lipid staining experiments, the pellet containing SVF cells was resuspended in growth media (Dulbecco’s modified Eagle’s medium plus 10% heat-inactivated, endotoxin-free fetal bovine serum) and was plated on glass coverslips. For flow cytometry, the SVF pellet was incubated in 0.5 ml red blood cell lysis buffer (eBioscience) for 5 min before resuspension in sorting buffer (PBS with 0.5% fetal bovine serum, 2 mmol/l EDTA, and 25 mmol/l HEPES) at a concentration of 10⁷ cells/ml. Cells were incubated with Fc Block (BD Biosciences) before staining with conjugated antibodies. After washing, cells were resuspended in sorting buffer supplemented with 4,6-diamidino-2-phenylindole (DAPI) and then analyzed using a fluorescence-activated cell sorter (FACS; FACSaria; BD Biosciences). Viable cells (DAPI⁻) were sorted directly into RNA lysis buffer for RNA extraction. ATMs were defined as F4/80⁺CD11b⁺ cells.

Isolation of CD11b⁺ macrophages from SVF isolates was performed by magnetic immunoaffinity isolation with anti-CD11b antibodies conjugated to magnetic beads (MACS; Miltenyi Biotec, Auburn, CA). Cells were isolated using positive selection columns before preparation of whole-cell lysates.

For tissue sections, fat pads were isolated and fixed in 10% formalin overnight before cryosectioning. For cultured cells, SVF cells were plated onto glass coverslips for 2 h, and nonadherent cells were removed by washing in PBS. After 24 h, the cells were fixed in 10% formalin for 20 min. For CD68 antibody staining, cells were permeabilized with 0.5% Triton X-100 in PBS. Cells were blocked in 2% BSA in PBS and incubated with primary antibodies diluted in blocking buffer. Alexa 488–conjugated secondary antibodies were then added with or without rhodamine-conjugated phalloidin (Invitrogen, Carlsbad, CA) to label actin structures. Lipid staining was performed by incubation of fixed cells in 1 μg/ml Nile Red (Sigma-Aldrich) in PBS for 5 min. Coverslips were mounted on slides with Vectashield (Vector Laboratories, Burlingame, CA) and imaged by confocal microscopy (Olympus IX SLA; Olympus, Center Valley, PA).

For short-term labeling experiments, lean mice were injected via tail vein with 100 μl of saline or PKH26 (0.1 mmol/l in Diluent B) per the manufacturer’s instructions. SVFs were isolated 24 h after injection and analyzed by FACS. For RT-PCR analysis, SVFs were isolated 8 weeks after injection. For microarray experiments, mice were injected with PKH26 after 8 weeks on the high-fat diet. Mice were killed 8 weeks postinjection and SVFs isolated. F4/80⁺CD11b⁺PKH26⁺ and F4/80⁺CD11b⁺PKH26⁻ cells were isolated by FACS and sorted into lysis buffer. Samples were isolated from three independent groups of 3–4 mice for use in microarray analysis.

RNA extraction was performed with an RNeasy kit (Qiagen, Valencia, CA). For microarray experiments, RNA quality was assessed on Agilent Bioanalyzer Picochip. RNA was amplified and cRNA prepared using Ovation Biotin labeling system (Nugen, San Carlos, CA) before hybridization to Mouse 430 2.0 arrays (Affymetrix, Santa Clara, CA). Expression values for each gene were calculated using a robust multiarray average (21). Data were filtered to remove genes without differential expression between recruited PKH26⁻ and resident PKH26⁺ ATM RNA samples and then fit to a linear model by the method of Smythe (22). The false discovery rate was set at an adjusted P value of 0.05 (23). Gene ontology was assessed using open-source databases (http://apps1.niaid.nih.gov/david/). RT reactions were performed using standard methods (Stratagene, LA Jolla, CA). cDNA were analyzed by semiquantitative PCR (30 cycles) or real-time PCR analysis using SYBR Green normalized to glyceraldehyde-3-phosphate dehydrogenase (GAPDH 7200; ABI, Foster City, CA). Semiquantitative PCR primers were: Gapdh forward 5′-ACCCAGAAGACTGTGGATGG-3′, Gapdh reverse 5′-GGAGACAACCTGGTCCTCAG-3′ (amplicon size 300 bp); Tnfa forward 5′-CCACATCTCCCTCCAGAAAA-3′, Tnfa reverse 5′-AGGGTCTGGGCCATAGAACT-3′ (amplicon size 258 bp); Nos2 forward 5′ CAGAGGACCCAGAGACAAGC-3′, Nos2 reverse 5′-TGCTGAAACATTTCCTGTGC-3′ (amplicon size 283 bp); Il6 forward 5′-GTTCTCTGGGAAATCGTGGA-3′, Il6 reverse 5′-GGAAATTCGGGGTAGGAAGGA-3′ (amplicon size 338 bp); Ccr2 forward 5′-GGGTCATGATCCCTATGTGG-3′, Ccr2 reverse 5′-TCCATGAGCAGTGGTTTGAA-3′ (amplicon size 215 bp); and Cd14 forward 5′-CAAGTGGGGAACCTGTCACT-3′, Cd14 reverse 5′-TGGCTTTTACCCACTGAACC-3′ (amplicon size 289 bp). Real-time PCR primers were: Gapdh forward 5′-TGAAGCAGGCATCTGAGGG-3′, Gapdh reverse 5′-CGAAGGTGGAAGAGTGGGAG-3′; Adfp forward 5′-ATTCTGAACCAGCCAACGTC-3′, Adfp reverse 5′-CTTATCCACCACCCCTGAGA-3′; Apoe forward 5′-CTGACAGGATGCCTAGCCG3′, Apoe reverse 5′-CGCAGGTAATCCCAGAAGC-3′; Emb forward 5′-TGAGGGCGATCCCACAGAT-3′, Emb reverse 5′-CCGTCACTGAGATATTACAGCTC-3′; Sirpb1 forward 5′-GCTCTTGGTGAACATATCTGCC-3′, Sirpb1 reverse 5′-GGGGTGAAGACCTTACCAGAC-3′; Pparg forward 5′-GGAAGACCACTCGCATTCCTT-3′, Pparg reverse 5′-TCGCACTTTGGTATTCTTGGAG-3′; Fgr forward 5′-CGGCTGAAGAACGCTATTACC-3′, Fgr reverse 5′-GGGCGACGAATATGGTCACTC-3′; Fn1 forward 5′-TATAGCAGGCTACCGACTGAC-3′, Fn1 reverse 5′-GGTCGCGGTGTACTCAGAC-3′; Grk3 forward 5′-AGGAGGGTTTGGGGAAGTTTA-3′, Grk3 reverse 5′-CATGATCCTCTCGTTCAAAGCC-3′; Timp2 forward 5′-CTGGACGTTGGAGGAAAGAAG-3′, Timp2 reverse 5′-GGTGATGCTAAGCGTGTCCC-3′; Ccr2 forward 5′-ATCCACGGCATACTATCAACATC-3′, Ccr2 reverse 5′-CAAGGCTCACCATCATCGTAG-3′; Vsig forward 5′-CCTGGGCCACCTAATAGTGC-3′, Vsig reverse 5′-GCCTCTCAGGGGATCATAGAT-3′; Tnfa forward 5′-ACGGCATGGATCTCAAAGAC-3′, Tnfa reverse 5′-AGATAGCAAATCGGCTGACG-3′; Il6 forward 5′-GTTCTCTGGGAAATCGTGGA-3′, Il6 reverse 5′-GGAAATTGGGGTAGGAAGGA-3′; Nos2 forward 5′-CAGAGGACCCAGAGACAAGC-3′, Nos2 reverse 5′-TGCTGAAACATTTCCTGTGC-3′; and Cd14 forward 5′-CAAGTGGGGAACCTGTCACT-3′, Cd14 reverse 5′-TGGCTTTTACCCACTGAACC-3′. Relative expression was assessed by comparative C_T method, correcting for amplification efficiency of the primers, and performed in triplicate.

Results are the means ± SD. Statistical analyses were conducted with unpaired Student’s t tests, with significance set at a P value of <0.05.

ATMs have been identified primarily by expression of F4/80, a marker that is found on many, but not all, macrophage types (e.g., low expression in alveolar macrophages) (5,6,24). Given the heterogeneity in macrophage surface antigen expression, we sought to identify ATMs using a complementary method based on their function as phagocytic cells. To label ATMs, we injected mice intravenously with PKH26, an inert fluorescent dye that is avidly taken up by phagocytic cells in vivo and that efficiently labels tissue macrophages (18,19).

After injection with PKH26, PKH26⁺F4/80⁺ cells in the SVF were detected by FACS (Fig. 1A). PKH26 labeled >90% of the F4/80⁺ cells by flow cytometry. This was confirmed in adipose tissue sections, where PKH26⁺ cells colocalized with CD11b⁺ cells between adipocytes and associated with vascular structures (Fig. 1B). Specific and efficient labeling of ATMs was further confirmed in SVF cultures. PKH26⁺ cells expressed CD11b, F4/80, and CD68 but showed minimal expression of CD14. ATMs extended long cellular processes and had an elongated phenotype in culture.

By immunofluorescence, variability in F4/80 expression in PKH26-labeled cells was seen with high- and low-F4/80–expressing PKH26⁺ cells (data not shown). In contrast, CD11b labeled PKH26⁺ cells uniformly. Therefore, subsequent studies used both CD11b and F4/80 expression to identify ATMs. PKH26 injection did not label cells lacking expression of macrophage markers F4/80 or CD11b.

Because PKH26 is retained in macrophages for prolonged periods of time (18,19,25), the efficient PKH26 labeling of ATMs in vivo allowed us to differentiate resident tissue macrophages present at the time of injection from those ATMs recruited to adipose tissue on high-fat diet exposure. Because monocytes do not take up the dye, macrophages that are recruited to adipose tissue in the days after the injection will be PKH26⁻ (Fig. 2A).

To demonstrate this, mice were injected with PKH26 at 2 months of age and put on a normal diet or high-fat diet (45% kcal fat). In the high-fat diet mice, the percentage of F4/80⁺PKH26⁻ cells found in the SVF increased with time after injection (Fig. 2B). PKH26⁺ ATMs were detectable as long as 20 weeks after injection in high-fat diet mice. In contrast, mice injected and maintained on a normal diet did not demonstrate persistence of PKH26⁺ ATMs because they made up <5% of the ATM population at 8 weeks after injection (Fig. 2C). When normalized for total cell number and fat weight 8 weeks after injection, mice placed on a high-fat diet had fourfold more PKH26⁺ ATMs compared with normal diet mice (0.25 × 10⁶ vs. 0.063 × 10⁶ cells/g fat). This indicates that a population of ATMs labeled by PKH26 is retained longer in high-fat diet compared with normal diet mice and suggests that ATMs in normal diet mice have a faster turnover rate than high-fat diet mice.

We hypothesized that macrophages recruited to fat tissue during high-fat diet have unique inflammatory gene expression compared with the resident macrophages. C57BL/6 male mice were injected with PKH26 and placed on high-fat diet for 8 weeks. SVF cells were then separated into an F4/80⁺CD11b⁺PKH26⁺ resident ATM pool and an F4/80⁺CD11b⁺PKH26⁻ recruited ATM pool (Fig. 2D). Gene expression in these pools was evaluated by semiquantitative RT-PCR for proinflammatory genes (Fig. 2E). F4/80⁺CD11b⁺ ATMs were also isolated by FACS from age-matched lean male mice maintained on a normal diet for comparison. Although CD14 and TNF-α were expressed at equal levels in resident and recruited ATM pools, IL-6, iNOS, and CCR2 were expressed at higher levels in the F4/80⁺PKH26⁻ recruited ATMs. Expression of these inflammatory genes in both recruited and resident high-fat diet ATM populations were higher than ATMs from normal diet mice. This suggests that ATMs newly recruited to adipose tissue have an increased inflammatory capacity compared with resident ATMs.

These results demonstrated that for the candidate genes examined, qualitative differences exist between resident and recruited ATMs. To broadly assess gene expression differences between the two ATM subtypes, we analyzed resident and recruited ATMs using cDNA microarrays. ATMs were isolated from epididymal fat pads of three independent pools of mice injected with PKH26 and placed on high-fat diet. Because the most significant influx of macrophages into adipose tissue occurs after 8 weeks of high-fat diet (6), we designed the experiment to identify and capture the cells that accumulate in fat during the window of 8–16 weeks on a high-fat diet. Therefore, male mice were placed on a high-fat diet 8 weeks before injection and then continued on high-fat diet for 8 more weeks after injection.

Comparative analysis of gene expression between F4/80⁺PKH26⁻ recruited and F4/80⁺PKH26⁺ resident ATMs identified a total of 46 unique genes differentially expressed between the two ATM populations. Of these genes, 35 were overexpressed in recruited F4/80⁺PKH26⁻ ATMs (Table 1), and 11 were overexpressed in resident F4/80⁺PKH26⁺ ATMs (Table 2). Gene ontology analysis showed that 26% of the genes overexpressed in recruited ATMs are involved in response to stimulus or cell-cell communication (Fig. 3A). Genes preferentially expressed in recruited F4/80⁺PKH26⁻ ATMs included Ccr2, which agrees with studies showing that CCR2 is required for recruitment of inflammatory ATMs. Additionally, genes important in macrophage activation (signal regulatory protein β1 [Sirpb1] and major histocompatibility complex class II receptor [H2-DMb1]) and cellular adhesion and migration (embigin [Emb], Src kinase Fgr [Fgr], and fibronectin [Fn1]) were overexpressed in recruited ATMs. The expression differences of these genes were confirmed by real-time RT-PCR in independent unamplified RNA samples (n = 3) (Fig. 3B). Expression of all of these genes were increased in the SVF from high-fat diet mice compared with lean mice (Fig. 3C). Analysis of Fgr, apoE, and GRK3 by immunoblot demonstrated that these proteins were primarily expressed in ATMs compared with nonmacrophage cells in the SVF (Fig. 3D).

Based on the current models of macrophage diversity and our microarray results (14), we hypothesized that CCR2 knockout (CCR2KO) mice would be deficient in the recruited macrophage population with high CCR2 expression, but it would retain the resident pool of ATMs (PKH26⁺) with low CCR2 expression. Gene expression in ATMs from high-fat diet–fed CCR2KO mice was analyzed. Male CCR2KO mice did not differ in weight from C57BL/6 control mice fed a normal or high-fat diet for 16 weeks (Table 3). We did not detect significant differences in food intake between the knockout and control mice (5.95 ± 1.2 vs. 5.58 ± 0.83 g/day intake for C57/Bl6 vs. CCR2KO, P value = 0.41). A significant increase in the mass of epididymal fat pads in CCR2KO compared with control mice was observed that correlated with increased adipocyte size in the high-fat diet CCR2KO mice (data not shown) (12).

We wondered whether the genes overexpressed in the recruited ATMs might be decreased in ATMs from obese CCR2KO mice. The expression levels of these genes in ATMs from high-fat diet–fed CCR2KO mice were decreased compared with the recruited PKH26⁻ ATMs and were expressed at levels comparable to the resident PKH26⁺ ATM population (Fig. 3B). Similar results were seen for the genes overexpressed in resident ATMs, such as tissue inhibitor of metalloproteinase-2 (Timp2) and apoE (Apoe) (Fig. 3E). Expression levels of these genes in high-fat diet–fed CCR2KO ATMs were comparable to the resident PKH26⁺ ATM population and were increased compared with the recruited PKH26⁻ population. Combined with the morphometric data, this indicates that the qualitative changes in ATMs in the CCR2KO mice are not attributable to decreased adiposity.

ApoE (Apoe) gene expression was decreased in the recruited ATMs compared with resident macrophages (Fig. 3E). This suggested that cholesterol efflux might be impaired in recruited macrophages, similar to macrophage foam cells in atherosclerosis (26). Microarray analysis also suggested that lipid and cholesterol metabolism might be increased in recruited ATMs because expression of sterol regulatory element binding factor 1 (Srepf1) and apoB48 receptor (Apob48r) were both increased in recruited ATMs, although this did not reach full significance (P = 0.073 and 0.077, respectively).

Therefore, we examined the expression of several genes important in lipid metabolism in ATMs from normal diet and high-fat diet mice. Apoe was downregulated in ATMs with high-fat diet. Expression of both adipocyte differentiation–related protein (Adfp) and peroxisome proliferator activation receptor-γ (Pparg) were increased in high-fat diet ATMs (Fig. 4A). To determine the functional significance of these findings, we examined the lipid content of ATMs isolated from normal diet–and high-fat diet–fed animals by Nile Red staining. Both macrophage (CD11b⁺) and nonmacrophage (CD11b⁻) cells in SVF cultures had Nile Red–positive vesicles; however, ATMs had more pronounced lipid accumulation, especially in obese mice (Fig. 4B). Lipid-rich vesicles were qualitatively larger and more abundant in ATMs from high-fat diet compared with normal diet mice (Fig. 4C).

We have examined the properties of ATMs in obese animals with a focus on the macrophages that are recruited to adipose tissue during high-fat diet exposure, based on the hypothesis that these cells have unique inflammatory properties. Our initial experiments examined the utility of PKH26 to label ATMs in vivo. This dye helped confirm that ATMs in mice express F4/80, CD11b, and CD68, but not CD14. Low CD14 expression appears to be a characteristic of ATMs from mice, but not from humans (27). These experiments also suggest that ATMs are the predominant phagocytic cells in the SVF.

PKH26 labeling also permitted a gross assessment of ATM turnover (rate of loss of ATMs from fat tissue). Interestingly, at 8 weeks after injection, mice on a normal diet had few PKH26⁺ ATMs in the fat tissue, whereas those on a high-fat diet still had a significant quantity of PKH26⁺ ATMs along with significantly more total macrophages in fat tissue compared with normal diet mice. This shows that some macrophages are retained in fat in obese mice longer than in lean animals and suggests that ATMs are lost at a greater rate in normal diet compared with high-fat diet mice. This is consistent with studies showing that CCR2⁺ monocytes are recruited to peripheral tissues in normal states at a constant low level (28). Retention signals, such as MCP-1 and macrophage migration inhibitory factor, that are increased in obesity may serve to maintain macrophages in adipose tissue (8,29). Further studies are required to confirm this hypothesis, using other methods such as bone marrow transplantation.

Stratification of ATMs based on PKH26 labeling allowed us to purify ATMs that are recruited to white fat during high-fat diet exposure (PKH26⁻). This pulse labeling technique has been used in other inflammatory settings to examine the properties of recruited tissue macrophages. Recruited ATMs had increased expression of Ccr2, Il6, and Nos2, genes that have been previously shown to contribute to inflammation in adipose tissue (5,6,12). These observations support the importance of the CCR2/MCP-1 axis in recruiting ATMs during high-fat diet.

To better understand these cell types, we used microarray analysis to examine differential gene expression between resident and recruited ATMs in obese animals. The validity of this approach was highlighted by the detection of increased CCR2 expression in PKH26⁻ recruited ATMs. We identified a number of other genes involved in macrophage inflammatory responses, migration, and phagocytosis that are upregulated in recruited PKH26⁻ ATMs.

Embigin, fibronectin, and the Src kinase Fgr are important in cell migration and were all increased in recruited ATMs. The importance of cell adhesion in ATM function was supported by the long processes seen in cultured ATMs from obese mice. This supports observations that macrophage-adipocyte cell contact activates macrophages and influences adipocyte insulin sensitivity (30). Additionally, recruited ATMs had increased expression of major histocompatibility complex class II receptors, suggesting that these macrophages have increased antigen-presenting properties.

Based on a panel of genes identified in microarrays, ATMs from obese CCR2KO mice have characteristics similar to the resident population of ATMs and differ from the recruited population of ATMs. This is consistent with the model that resident ATMs arrive in fat via CCR2-independent signals and have decreased inflammatory properties. How such cells get targeted to adipose tissue is unknown. Overall, these observations demonstrate a diversity in ATMs from obese mice that has not been previously appreciated, and they imply that some macrophages in fat tissue may have decreased inflammatory properties.

Our analysis of diet-induced obesity in CCR2KO mice differs from other reports on this model (12,31). Similar to Chen et al. (31), we observed equivalent weight gain on a high-fat diet in CCR2KO mice and controls. This differs from the data of Weisberg et al. (12), who observed less weight gain in CCR2KO mice based on decreased food intake. These discrepancies may be due to different diet compositions between studies (60% fat for Weisberg et al. vs. 45% fat in this study), differences in animal care, or differences in food intake analysis. Similar to Weisberg et al. (12), we observed increased adipocyte size and protection from hepatic steatosis in high-fat diet–fed CCR2KO mice. However, we observed increased weight of epididymal fat pads in normal diet–and high-fat diet–fed CCR2KO mice compared with controls, whereas Weisberg et al. did not note differences in total fat mass by dual-energy X-ray absorptiometry analysis. This difference may be attributable either to the different feeding protocols used (Weisberg et al. fed 60% fat for 24 weeks) or to the different methodologies used to assess fat composition.

We did not note differences in TNF-α expression between the resident and recruited ATMs from high-fat diet mice in both RT-PCR and microarray analysis, and we observed that both types express more TNF-α than ATMs from lean animals. This suggests that obesity generates signals that may activate all macrophages similarly in fat, despite the qualitative differences between subtypes of ATMs. Although the role of TNF-α in adipose tissue function is well known, these findings suggest that other macrophage characteristics should be examined further, including cell adhesion and migratory properties.

Our microarray analysis failed to identify significant differences in cytokine gene expression between the recruited and resident ATM populations, which differed from our initial analysis of these cytokines. This may be attributable to loss of sensitivity with the RNA amplification step required for the microarray hybridization. Alternatively, the experimental design used for the microarray analysis may examine slightly different ATM populations. Nevertheless, this analysis provides evidence of qualitative differences in ATM subpopulations that is a starting point for future investigations.

High-fat diet appears to increase ATM expression of genes important in lipid metabolism, such as Adfp, Pparg, Srebf1, and Apob48r. Increased Adfp expression supports studies that show increased adipocyte differentiation–related protein in ATMs in crown-like structures (9). Increases in Apob48r and Srebf1 are consistent with the increased lipid uptake seen in high-fat diet ATMs. These observations are similar to what has been seen with foam cells in atherosclerosis and suggest that similar mechanisms may relate the two processes.

It is important to point out that these observations were made in murine models of diet-induced obesity and that the changes identified here may not be recapitulated in humans. Mouse and human macrophages differ with respect to activation profiles (15). In particular, iNOS activity in macrophage populations differs between murine and human models of inflammation (32,33). However for many inflammatory markers, ATMs from obese humans and mice have been shown to be similar (5–7). Therefore, mice remain a useful model for examining ATM biology to generate new hypotheses to be confirmed in humans. Future studies will focus on defining the importance of the identified genes on ATM function and the development of type 2 diabetes. They will enable us to better understand the range of function of these unique cells and to investigate how their diversity may be used to control inflammation in adipose tissue.

Research was supported by a training grant from the National Institutes of Health to the Department of Pediatrics and Communicable Diseases (T32HD007513), a grant to the Michigan Diabetes Research Training Center (P60DK20572), and a grant from the National Institutes of Health (R01DK60591) to A.R.S.

We thank the University of Michigan Flow Cytometry Core for their assistance with FACS analysis and the University of Michigan Microarray Core for microarray hybridization and data analysis. We thank the Michigan Diabetes Research Training Center at the University of Michigan for subsidizing microarray analysis costs. We thank Jiandie Lin for his assistance with real-time PCR.