Adipose tissue macrophages induce PPARγ-high FOXP3+ regulatory T cells

Numerous regulatory T cells (Tregs) are present in adipose tissues compared with other lymphoid or non-lymphoid tissues. Adipose Tregs regulate inflammatory state and insulin sensitivity. However, the mechanism that maintains Tregs in adipose tissue remains unclear. Here, we revealed the contribution of adipose tissue macrophages (ATMs) to the induction and proliferation of adipose Tregs. ATMs isolated from mice under steady state conditions induced Tregs with high expression of PPARγ compared with splenic dendritic cells in vitro. Furthermore, ATMs from obese mice prompted the differentiation of PPARγ low Tregs. Adoptive transfer of ATMs induced differentiation and proliferation of Tregs, whereas depletion of ATMs by clodronate-liposome resulted in reduction of adipose Tregs, in vivo. Deficiency of anti-inflammatory adipocytokine, Adipoq, resulted in small proportions of ATMs and adipose Tregs without alteration of other immune cells in vivo. Therefore, these data suggest that the abundance of Tregs in adipose tissue could be partly attributed to the ability of ATMs to induce PPARγ-expressing Tregs.

However, little is known about the mechanisms responsible for the presence of high proportion of Tregs in WAT.
Adiponectin is an adipocytokine secreted by adipocytes and has anti-diabetic and anti-inflammatory properties 11 . The adipose tissue stromal vascular fraction (SVF) is a rich source of preadipocytes, mesenchymal stem cells, endothelial progenitor cell, T cells, B cells, mast cells as well as adipose tissue macrophages (ATMs). The parabiosis models of wild type (WT) and Adipoq (− /− ) demonstrated that adiponectin goes back to adipose tissue especially into adipose SVF 12 , suggesting that adiponectin exerts some effect on adipose SVF that contains immune cells such as macrophages and T cells. In this regard, however, the direct effect of adiponectin on adipose immune cells remains to be investigated.
The aim of the present study was to determine the mechanism that maintains Tregs in adipose tissue. Towards this goal, we examined the ability of ATMs to induce Tregs. The results showed that ATMs induce PPARγ -expressing Tregs and thus explain, at least in part, the abundance of Tregs in adipose tissue.

Results
Local proliferation of adipose tissue Tregs and phenotype of adipose tissue antigen presenting cells. Large numbers of Tregs are present in WAT. One possible explanation of this phenomenon is a proliferation of Treg in WAT. First, the intensity of Ki67, a cell proliferation marker, was measured in immune cells in adipose tissues. The expression of Ki67 was the highest in CD3 + CD4 + FOXP3 + Tregs among immune cells, including CD3 + CD4 + FOXP3 − T cells, CD3 + CD8 + T cells, CD19 + B220 + B cells, and F4/80 + CD11b + macrophages (Fig. 1A,B). This result suggested that antigen presenting cell (APC) seem to drive the proliferation of Tregs in adipose tissues. Next, we analyzed the characteristics of APCs in low-density leukocytes from the SVF based on CD11c and CD11b expression over a wide expression range 13 . Among these cells, at least three subsets of cells were recognized based on the CD11c/CD11b expression patterns: CD11c int CD11b low (R1), CD11c high CD11b low (R2), and CD11b high (R3) (Fig. 1C). R3 was the most abundant fraction, and flow cytometric analysis of the co-stimulatory molecules and surface antigens showed the expression of several DC-related molecules (CD40) and several important molecules for antigen presentation (MHC class II, CD80 and CD86) (Fig. 1D). In addition, R3 also expressed plasmacytoid DC-related molecule mPDCA1 as well as macrophage-related molecule (CD11b and F4/80), and CD206, a marker of M2 macrophages (Fig. 1D). With regard to the other two fractions, R1 expressed MHC class II, CD103, CD205, B220, and mPDCA1 (Fig. 1D), which is in agreement with the characteristics of plasmacytoid dendritic cells (pDCs). On the other hand, R2 overexpressed CD11c, MHC class II, CD103, and CD205. The expression of these co-stimulatory molecules was consistent with the conventional features of DC (Fig. 1D). Based on these results, we used the F4/80 + R3 subset as ATMs.
To further characterize the ATMs, we analyzed the morphology of R3. Microscopic analysis using Pappenheim staining showed that R3 contained droplets or cytoplasmic vacuoles characteristic of macrophages (Fig. 1E). Furthermore, Nile red staining showed more neutral lipid content in ATMs from ob/ ob mice (ob/ob ATMs) than those from C57BL6/J mice (NC ATMs) (Fig. 1F).

Induction of Tregs by ATMs.
Tregs are abundant in adipose tissues accounting for almost 40% of T cells 10 , and proliferate in adipose tissues (Fig. 1A,B). Considering that ATMs are the most abundant APCs (Fig. 1C), we hypothesized that ATMs might be potent inducers of Tregs. To test this hypothesis, ATMs, SPDCs, and CD4 + CD25 − CD62L + CD44 − (non Treg) T cells were purified using FACS. Subsequently, non Treg T cells were stimulated with ATMs or SPDCs plus anti-CD3 mAb, with or without TGF-β . After 6-day culture, SPDCs induced FOXP3 + Tregs (without TGF-β : 0.58%, with TGF-β : 4.5%) ( Fig. 2A). Surprisingly, the proportion of FOXP3 + Tregs induced by ATMs was higher than by SPDCs in the presence of TGF-β (NC ATMs: 26.8%, HFD ATMs: 29.4%, ob/ob ATMs: 31.1%) ( Fig. 2A). SPDCs and ATMs induced Tregs with quite low efficiency in the absence of TGF-β , indicating that this effect was dependent on TGF-β . The proportion of induced Tregs was higher in three ATMs than SPDCs ( Fig. 2A,B), however, the number of Tregs was much higher in ob/ob ATMs (Fig. 2C). SPDC increased the number of Tregs more than NC or HFD ATMs (Fig. 2C). These results indicated that SPDCs and ob/ob ATMs increased both Tregs and other T lymphocytes, whereas, NC and HFD ATMs efficiently increased Tregs compared to other T lymphocyte.
To further characterize Treg induction in vitro, we analyzed CD25, folate receptor (FR)-4, glucocorticoid-induced tumor necrosis factor receptor (GITR), and cytotoxic T-lymphocyte-associated protein (CTLA) 4, which are hallmarks of Treg 5,14-16 . In Tregs induced by SPDCs and NC ATMs, all Treg markers, such as CD25, FR4 and GITR, were highly upregulated to levels similar to those induced by SPDCs (Fig. 2D). CTLA4 induction was relatively low in Tregs induced by ATMs (Fig. 2D). Next, number of CFSE (carboxyfluorescein diacetate succinimidyl ester) divisions were measured during Treg induction in vitro. In Tregs induced by ATMs, number of CFSE division was lower than that in Tregs induced by SPDCs (Fig. 2E,F). These results indicated that ATMs converted non-Treg T cells into de novo Tregs more efficiently than SPDCs.
To further confirm the induction of Tregs by ATMs, CD3 + CD4 + FOXP3 − (non Treg) T cells were collected from FOXP3-EGFP mice, which co-express EGFP and FOXP3 under the control of endogenous FOXP3 promoter. These non-Treg T cells were cocultured with ATMs from control mice, and examined by time-lapse confocal microscopy. In this system, conversion of Tregs was reflected by GFP fluorescence. The frequency GFP-positive cells increased with time (Fig. 2G), indicating that coculture with ATMs converted non-Treg T cells into Tregs. During microscopic observation, firm adhesion was observed between Tregs and ATMs (Fig. 2H, green cells: Tregs, red cells: ATMs). Previously, Cahalan et al. reported that contacts between Tregs and DCs last longer than those between conventional T cells and DCs, and this interaction is mediated by CTLA4 17 . Similarly to this report, distance analysis indicated that the distance between ATMs and Tregs was significantly shorter than that between ATMs and non-Treg T cells (Fig. 2I, supplementary Figure 1A,B). Moreover, adhesion between ATMs and Tregs was more frequent than that between ATMs and non-Treg T cells (Fig. 2J). Considering that ATMs induced CTLA4 positive Tregs in vitro (Fig. 2D), interaction between Tregs and ATMs should last longer than non-Tregs. Collectively, these results indicate that ATMs had potency to convert non-Treg T cells into Tregs in vitro. In vivo and in vitro characterization of Tregs. Since Tregs from adipose tissues of NC mice specifically express PPARγ 10 , we further characterized Tregs induction in vitro by measuring PPARγ expression. PPARγ expression level was higher in Tregs induced by NC ATMs than those induced by SPDCs of NC (Fig. 3A). Moreover, PPARγ expression level was significantly higher in Tregs induced by NC ATMs than those by HFD ATMs (Fig. 3B). Furthermore, PPARγ expression was lower in Tregs induced by ob/ ob ATMs than those by HFD ATMs (Fig. 3B), indicating that lean ATMs induced PPARγ -high Tregs, whereas obese ATMs induced PPARγ -low Tregs in vitro.
To investigate the impact of obesity on ATMs, we performed gene expression microarray analysis of ATMs derived from NC ATMs and ob/ob ATMs (supplementary Figure 2). Ob/ob ATMs showed significantly higher expressions of inflammatory cytokines, such asIL-1β and TNF-α , and lower expression of Th2 cytokine: IL-4 than NC ATMs. In addition, Sema4a expression was significantly lower in ob/ob ATMs than NC ATMs. There were no significant differences in expressions of IL-33 or TGFβ 1, which were cytokines associating Treg induction 18,19 . IL-1β is reported to convert human Treg into Th17 cells 20 . TNF-α induces depohsphorylation of Foxp3, and inhibits Treg function 21 . Sema4a binds to nrp1 in Tregs, and potentiate Treg function and survival 22 . These results indicated that differences between obese and NC ATMs might be derived from such changes in cytokine and surface molecule expressions.
Next, we characterized adipose Tregs in vivo. The proportion of Tregs was higher in adipose tissues than spleen, and higher in adipose tissues of NC than HFD mice ( Fig. 3C; upper), as reported previously 10 . Under these conditions, PPARγ expression was markedly higher in adipose NC Treg than HFD (NC: 55.7%, HFD: 3.6%) ( Fig. 3C; lower). PPARγ expression level was lower in Tregs from the spleen than those from adipose tissue of NC, and the expression level in the spleen was not different between NC and HFD (NC: 8.9%, HFD: 5.3%) ( Fig. 3C; lower). Number of PPARγ -high Tregs were significantly higher in adipose tissues of NC than ob/ob mice (Fig. 3D). Collectively, based on the ability of NC ATMs to induce PPARγ -high Tregs in vitro, PPARγ -high Tregs were abundant in adipose tissues of NC mice in vivo. Feng et al. 23,24 reported that intraperitoneal injection of clodronate-liposomes reduced the number of ATMs in diet-induced obese mice. Using the same experimental protocol, NC mice were treated with PBS-liposome or clodronate-liposome for 8 weeks. No significant differences were observed in the time course of body weight and organ weights (epididymal fat, mesenteric fat, subcutaneous fat, spleen and liver) between PBS-and clodronate-liposome injected groups (Fig. 4A,B). Clodronate-liposome did not alter the proportion of dendritic cells, or CD4 + T cells in adipose tissue, and splenic Tregs (Fig. 4C), but significantly reduced ATMs (Fig. 4D,F). Under this condition, adipose Tregs were significantly decreased (Fig. 4E,F).

Induction of Tregs by ATMs in vivo.
Next, to investigate the differentiation of Tregs in adipose tissues, isolated ATMs or PBS was injected into each side of epididymal WAT in living mice, followed by analysis of WAT after 4 days (Fig. 5A). FOXP3 expression and the proportion of Treg were significantly higher in WAT injected with ATMs than with PBS (Fig. 5B,C). Furthermore, the expression of Ki67 was higher in Tregs than other CD4 + T cells in each WAT (Fig. 5D). Moreover, Ki67 expression was higher in Tregs from ATM-injected WAT than PBS-injected WAT (Fig. 5D). The number of Ki67 + Tregs was significantly higher in ATM-injected WAT than PBS-injected WAT (Fig. 5E). Considered together, ATMs induce differentiation and proliferation of Tregs in vivo.
To rescue the Treg number in WAT of obese mice, isolated NC ATMs or PBS was injected into each side of epididymal WAT in ob/ob mice. The proportion of Tregs was not different between PBS-injected and ATM injected WAT, and the number of Tregs showed tendency to decrease in ATM-injected WAT (supplementary Figure 3A, B). These results indicated that NC ATMs could not rescue Treg cell number in ob/ob mice.
To further investigate the mechanism, Tregs were induced by ATMs from WT mice and Adipoq (− /− ) mice in vitro. Coculture of the same number of ATMs with non-Treg T cells resulted in the induction of similar proportions of Tregs in WT and Adipoq (− /− ) (Fig. 6F), and PPARγ expression in Tregs was not different between those induced by ATMs from WT or Adipoq (− /− ) (Fig. 6G).
To confirm the effect of ATMs, NC ATMs were introduced into epididymal WAT of Adipoq (− /− ) mice. The rate and number of Tregs were significantly increased in ATM-injected WAT than PBS (Fig. 6H,I). These results suggest that the small proportion of ATMs in Adipoq (− /− ) mice should be related to the small proportion of adipose Tregs.

Discussion
In the present study, we characterized ATMs from NC and obese mice. These cells differentially promoted the induction of Tregs. ATMs of NC induced PPARγ -high Tregs while ATMs of obese mice induced PPARγ -low Tregs. Moreover, we demonstrated that PPARγ expression was markedly reduced in adipose Treg of HFD-fed obese mice compared with control mice. In agreement with these results, ATMs of HFD-fed mice induced Tregs with lower PPARγ expression than control mice in vitro. Furthermore, ATMs of severe obese ob/ob mice induced Tregs with further lower PPARγ expression than HFD-fed mice in vitro. Consequently, the severity of obesity correlated with the low proportion of adipose Tregs and their PPARγ expression in vivo, and was linked with the diminished ability of ATMs to induce Tregs with PPARγ expression in vitro.
To further assess the ability of ATMs to induce Tregs in vivo, we performed both ablation and adoptive transfer of macrophages in epididymal WAT. To deplete ATMs in vivo, we used a chemical reagent that produced local ablation of macrophages in epididymal WAT. In this regard, intraperitoneal injection of clodronate-liposome is reported to reduce the number of macrophages in fat depots located within the peritoneal cavity, particularly in epididymal WAT 23,24 . In the present study, clodronate-liposome significantly reduced ATMs without reducing adipose dendritic cells, as well as significantly reduced the number of adipose Treg in steady state conditions. The results suggest a major role for ATMs in the maintenance of adipose Tregs. Inversely, injection of ATMs resulted in significant increase in the proportion and the number of Ki67-expressing adipose Tregs. The result suggests that ATMs play a role in inducing differentiation and proliferation of Tregs in vivo.
Injection of NC ATMs increased the proportion of Tregs in WAT of NC mice, but not in WAT of ob/ob mice. These results indicated that Treg induction was influenced by environmental state. The expression of IL-6 is increased in adipose tissues of obese mice, and its level in adipose tissue correlates with insulin resistance 25 . Previous studies indicated that in the absence of proinflammatory cytokines, such as IL-6 or IL-21, TGF-β induced differentiation of naive T cells to Tregs, whereas in the presence of IL-6, naive T cells differentiated into proinflammatory Th17 cells 18 . Actually, we confirmed that IL-6 abrogated the efficient induction of Tregs by ATMs (supplementary Figure 3C, D). Collectively, inflammatory state should be partly account for the low efficiency of Treg induction in WAT of obese mice.
Adiponectin is a protein secreted abundantly by adipocytes 11 . Serum adiponectin levels correlate inversely with anthropometric indices and insulin resistance in obese subjects 26 . Furthermore, adiponectin promotes polarization of anti-inflammatory macrophages 27 . Previous studies also showed accumulation of adiponectin protein in immune cells-containing SVF, in adipose tissues 12 . However, the effects of adiponectin on adipose tissue inflammation and ATMs remain unclear. The present study showed that the effect of adiponectin on adipose tissue immune cells includes the maintenance of ATM number without affecting FOXP3 and PPARγ induction. The proportions of ATMs and Tregs were lower in adipose tissues of Adipoq (− /− ) compared with WT mice. The ability of ATMs to induce PPARγ -expressing Tregs in both WT and Adipoq (− /− ) mice was similar in vitro. Consequently, the low number of Tregs is related to the small number of ATMs in Adipoq (− /− ) mice. Adiponectin deficiency leads to diet-induced insulin resistance 11 , and Treg depletion worsens insulin resistance in obese mice 28 . Considered together, the small numbers of ATMs and Tregs may partly contribute to the development of insulin resistance in Adipoq (− /− ) mice.  Recently, Mathis et al. 29 reported that the adipose Treg compartment is seeded from thymocytes, and that the accumulation of Treg depends on APCs, including DCs and ATMs 29 . Our results demonstrate that ATMs, but not DCs, have a large impact on the proportion of adipose Tregs. In addition, our experiments clearly demonstrated the potency of ATMs in converting non-Treg T cells into Tregs in vitro. Moreover, in Tregs induced by ATMs, the number of cell division was smaller than those by SPDCs, indicating that ATMs generated de novo Tregs more efficiently than SPDCs. These results suggest that ATMs could induce de novo generation of Tregs in adipose tissues.
In summary, the present study demonstrated that ATMs of control mice promoted the differentiation of PPARγ -high Tregs, whereas ATMs of obese mice triggered PPARγ -low Tregs. In obese adipose tissues, the diminished capacity of ATMs to induce PPARγ -high Tregs and low adiponectin is likely linked to the low number of Tregs.

Research Design and Methods
Animals. The experimental protocol was approved by the Ethics Review Committee for Animal Experimentation of Osaka University, Graduate School of Medicine. All animal experiments were carried out in accordance with the Institutional Animal Care and Use Committee Guidelines of Osaka University. Male C57BL6/J and ob/ob mice were purchased from Charles River Japan (Yokohama, Japan) and used in experiments at 11-16 weeks of age. Control mice were fed normal chow (NC). For diet-induced obesity, 5-week-old C57BL6/J mice were fed high-fat/high-sucrose diet (HFD) for 9-12 weeks and used at 14-17 weeks of age. FOXP3 bicistronic reporter knock-in mice expressing EGFP were kindly provided by Dr. Kiyoshi Takeda (Osaka University, Japan). Adiponectin-deficient mice were generated as described previously 11 and used at 10-14 weeks of age. All mice were maintained under specific pathogen-free conditions and had free access to water and chow.
Proliferation assay. CD3 + CD4 + Foxp3 − splenocytes from control mice were labeled with 1.25 μ M CFSE (Biolegend) in PBS. CFSE labeled T cells were cultured with DCs or ATMs at 1:1 ratio and stimulated with 1 μ g/ml anti-CD3e mAb for 72 hours. Cellular proliferation was calculated by CFSE dilution using FACS Verse.
Confocal time-lapse imaging. PE-CD11b labeled ATMs and CD4 + FOXP3 − T cells from FOXP3-EGFP mice were co-cultured on glass bottom plate (Greiner Bio One, Stonehouse, UK). Time-lapse confocal microscopy was performed using A1 confocal microscope (Nikon), followed by cell position analysis using Imaris software (Bitplane, Zurich, Switzerland). Using position data, contour plots of ATMs distribution were generated by the JMP software (SAS Institute, Cary, NC). After selection of Scientific RepoRts | 5:16801 | DOI: 10.1038/srep16801 eight local maximum points containing abundant ATMs, the minimum distance between the point and each T cell was measured as the distance from macrophages (supplementary Figure 5A, B).
Clodronate depletion from macrophages. Clodronate liposomes were purchased from clodronateliposomes.org (Vrije Universiteit, Netherlands) at concentration of 1 mg/ml, and prepared as described previously 23,24 . Five-week-old mice were fed normal chow and concomitantly treated with 150 μ L liposome-clodronate or an equivalent volume of liposomes containing phosphate buffered saline (PBS) once per week for 8 weeks.
Quantitative real-time PCR. Total RNA was prepared from freshly isolated cells using RNAprotect Cell Reagent (Qiagen) using the protocol supplied by the manufacturer. RNA purification was performed using RNeasy microkit (Qiagen) following the instructions provided by the manufacturer. The cDNA was synthesized using the Transcriptor First Strand cDNA Synthesis Kit (Roche, Indianapolis, IL). Real-time PCR was performed on the LightCycler system (FastStart DNA Master SYBR Green I, Roche) according to the protocol provided by the manufacturer.
Macrophage adoptive transfer. 9 × 10 5 F4/80 + CD11b + macrophages derived from adipose tissues were injected into one side of adipose tissues with PBS as vehicle. Four days after injection, adipose tissues were collected. FOXP3 expression was quantified by quantitative polymerase chain reaction (qPCR) and the proportion of Ki67 + Tregs was analyzed by flow cytometry.
Microarray analysis. The macrophage subsets were sorted from epididymal adipose tissues of male C57BL6/J and ob/ob mice at 11 weeks of age, based on their expression of CD11b, CD11c and F4/80. RNA was extracted with RNeasy micro kit. Gene expression profiling was performed with Mouse Gene Expression 8 × 60 K Microarray (Agilent, Santa Clara, CA) by Takara Bio Inc (Tokyo, Japan). Cyanine-3 (labeled cRNA was prepared from 0.1 μ g Total RNA using the Low Input Quick Amp Labeling Kit (Agilent). Fragmentation, hybridization, and washing steps were also carried out as recommended by the manufacturer (Agilent). Slides were scanned on the Agilent SureScan Microarray Scanner G2600D using one color scan setting for 8 × 60 K array slides. The scanned images were analyzed with Feature Extraction Software 11.5.1.1 (Agilent) using default parameters to obtain processed signal intensities.
Statistical analysis. All data were expressed as mean ± SEM. Differences between two groups were examined for statistical significance by the Student's t-test. A P value < 0.05 denoted the presence of a statistically significant difference.