Insulin Regulates Lipolysis and Fat Mass by UpregulatingGrowth/Differentiation Factor 3 in Adipose TissueMacrophagesYun Bu,1 Katsuhide Okunishi,1 Satomi Yogosawa,1 Kouichi Mizuno,1 Maria Johnson Irudayam,2

Chester W. Brown,2 and Tetsuro Izumi1,3

https://doi.org/10.2337/db17-1201

Previous genetic studies in mice have shown that func-tional loss of activin receptor–like kinase 7 (ALK7), a type Itransforming growth factor-b receptor, increases lipol-ysis to resist fat accumulation in adipocytes. Althoughgrowth/differentiation factor 3 (GDF3) has been sug-gested to function as a ligand of ALK7 under nutrient-excess conditions, it is unknown howGDF3 production isregulated. Here, we show that a physiologically low levelof insulin converts CD11c2 adipose tissue macrophages(ATMs) intoGDF3-producingCD11c+macrophages ex vivoand directs ALK7-dependent accumulation of fat in vivo.Depletion of ATMs by clodronate upregulates adiposelipases and reduces fat mass in ALK7-intact obese mice,but not in their ALK7-deficient counterparts. Furthermore,depletion of ATMs or transplantation of GDF3-deficientbone marrow negates the in vivo effects of insulin on bothlipolysis and fat accumulation in ALK7-intact mice. TheGDF3-ALK7 axis between ATMs and adipocytes repre-sents a previously unrecognized mechanism by whichinsulin regulates both fat metabolism and mass.

The worldwide prevalence of obesity increases morbidityand mortality and imposes a growing public health burden.Most excess food intake is converted into fat, and specif-ically into triglycerides (TGs), which are stored in adipocytesof white adipose tissue (WAT). As adipocytes accumulate fatand increase in size, they start to secrete proinflammatoryadipocytokines, recruit or polarize macrophages and other

hematopoietic cells inside WAT, and cause chronic inflam-mation and obesity-related disorders (1). The TG content inadipocytes is determined by the balance between the syn-thesis and breakdown of TG. Although TG synthesis dependson the uptake of nutrients, the rate of lipid removal throughlipolysis is proportional to the total fat mass as well as theactivities of lipases, and is regulated by external factors, suchas catecholamine and insulin. It is important to understandthe mechanisms of fat accumulation to dissect the patho-physiology of obesity. Our previous genetic analyses usingF2 progeny between the Tsumura, Suzuki, obese diabetes(TSOD) and control BALB/c mice revealed a naturally oc-curring mutation in Acvr1c encoding the type I transforminggrowth factor-b (TGF-b) receptor activin receptor–like ki-nase 7 (ALK7) in BALB/c mice (2–5). Themutation gives riseto a stop codon in the kinase domain of ALK7. The congenicstrain T.B-Nidd5/3 is isogenic with TSODmice except for theBALB/c-derived ALK7 mutation and exhibits decreased ad-iposity because of enhanced lipolysis. The activation of ALK7downregulates themaster regulators of adipogenesis, C/EBPaand peroxisome proliferator–activated receptor g (PPARg), indifferentiated adipocytes, which leads to the suppression oflipolysis and to increases in adipocyte size and TG content.

To understand the regulatory mechanisms associatedwith ALK7, it is essential to determine its physiologicalligand. TGF-b family members such as Nodal, inhibin-bB(activin B or activin AB), growth/differentiation factor(GDF) 3, and GDF11 bind ALK7 and mediate its signals

1Laboratory of Molecular Endocrinology and Metabolism, Department of MolecularMedicine, Institute for Molecular and Cellular Regulation, Gunma University,Maebashi, Japan2Division of Genetics, Department of Pediatrics, University of Tennessee HealthScience Center, Memphis, TN3Research Program for Signal Transduction, Division of Endocrinology, Metabolismand Signal Research, Gunma University Initiative for Advanced Research, GunmaUniversity, Maebashi, Japan

Corresponding author: Tetsuro Izumi, tizumi@gunma-u.ac.jp, and KatsuhideOkunishi, okunishik@gunma-u.ac.jp.

Received 4 October 2017 and accepted 30 May 2018.

This article contains Supplementary Data online at http://diabetes.diabetesjournals.org/lookup/suppl/doi:10.2337/db17-1201/-/DC1.

© 2018 by the American Diabetes Association. Readers may use this article aslong as the work is properly cited, the use is educational and not for profit, and thework is not altered. More information is available at http://www.diabetesjournals.org/content/license.

Diabetes 1

METABOLISM

Diabetes Publish Ahead of Print, published online June 26, 2018

under specific conditions (6–9). Among these ligands,GDF3 seems to function under nutrient-excess conditions,because both GDF3 and ALK7 knockout mice attenuate fataccumulation in the face of high-fat diet (HFD)–inducedobesity (9,10). However, it has not been shown that GDF3directly activates ALK7 in adipocytes. Besides, neitherthe producer nor the upstream regulator of GDF3 undernutrient-excess conditions is known. In the current study,we establish GDF3 as the physiological ligand that acti-vates ALK7 in adipocytes, and CD11c+ adipose tissuemacrophages (ATMs) as the main cell source of GDF3.We further demonstrate that insulin upregulates GDF3 inATMs ex vivo and stimulates fat accumulation in vivothrough the GDF3-ALK7 signaling pathway. Our findingsreveal a novel mechanism by which insulin regulatesadiposity through ATMs in addition to its classically de-fined direct effect on adipocytes.

RESEARCH DESIGN AND METHODS

Animal ProceduresAnimal experiments were performed in accordance withthe rules and regulations of the Animal Care and Exper-imentation Committee, Gunma University. The TSODmouse was originally established from an outbred ddYstrain as an inbred strain with obesity and urinary glucose(11). The congenic mouse strain T.B-Nidd5/3 was devel-oped and characterized previously (3,4). The GDF3 knock-out mouse with a genetic background of C57BL/6J wasdescribed previously (10). C57BL/6N and BALB/cA micewere purchased from CLEA Japan. Only male mice werephenotypically characterized in the current study. Micehad ad libitum access to water and standard laboratorychow (CE-2; CLEA Japan) in an air-conditioned room with12-h light/dark cycles. An HFD (55% fat, 28% carbohy-drate, and 17% protein in calorie percentage; OrientalYeast Co., Ltd.) was given to mice from 4 weeks of agefor the indicated duration. For macrophage depletion,liposomes containing 110 mg/kg body weight clodronate(ClodronateLiposomes.org) were injected intraperitoneallytwice per week. For the in vivo insulin administration,saline or 0.75 units/kg body weight insulin (Humulin R;Lilly), the amount generally used for insulin tolerancetests, was injected intraperitoneally twice daily. For bonemarrow (BM) transplantation, recipient C57BL/6Nmice at8–10 weeks of age were irradiated twice with an individualdose of 5.4 Gy with a 3-h interval, and subsequentlyreceived an intravenous injection of 2 3 106 BM cellsfrom donor wild-type or GDF3 knockout mice. Mice weresacrificed after anesthetization by isoflurane inhalation.Blood was collected from the inferior vena cava using23-gauge needles and syringes. Serum nonesterified fattyacid (NEFA) levels were measured as a marker of lipolysisby NEFA C-test (Wako).

Cell Fractionation of Epididymal WATEpididymal WAT (epiWAT) was minced and digested with1 mg/mL collagenase type I (Invitrogen) for 1 h at 37°C

during shaking. The digested cells were filtered througha 250-mm nylon mesh (Kyoshin Rikoh) and centrifuged at50g for 10 min. The floating adipocytes were washed withPBS twice. After dispersing the pellet containing thestromal-vascular fraction (SVF), the medium was filteredthrough a 40-mm nylon mesh and centrifuged at 300g for10 min. The pellet was then incubated with erythrocyte-lysing buffer consisting of 155 mmol/L NH4Cl, 5.7 mmol/LK2HPO4, and 0.1 mmol/L EDTA at room temperature for1 min and washed twice with PBS.

The cells in the SVF were resuspended in PBS, 2 mmol/LEDTA, and 2% FBS, and were incubated with excess Fcblock (anti-CD16/CD32 antibodies; BD Biosciences) to blockFc receptor–mediated, nonspecific antibody binding. Cellsurface markers were stained on ice in the dark for 20 minusing CD11b-phycoerythrin-Cy7, F4/80-allophycocyanin(Tonbo Biosciences), and CD11c-phycoerythrin (BD Bio-sciences) monoclonal antibodies. Some cells were stainedas negative controls with fluorochrome-matched isotypecontrol antibodies. After excluding dead cells by stainingwith 7-aminoactinomycin D, live cells were subjected tocharacterization of cell populations or to sorting of spe-cific cell populations by a FACSVerse or a FACSAriaII FlowCytometer (BD Biosciences).

RNA Preparation and Gene Expression AnalysesRNA was extracted using Sepasol-RNA I Super (NacalaiTesque). Total RNA (1 mg) was reverse transcribed usingoligo-(dT)12–18 primer and Superscript III (Invitrogen).Quantitative PCR was performed with SYBR Premix ExTaq (TaKaRa Bio) using a LightCyler 480 System (Roche).The results were normalized against 36B4 mRNA expres-sion. The primer sequences are listed in SupplementaryTable 1.

Antibodies, Immunoblotting, and ImmunostainingRabbit polyclonal anti-ALK7 antibody was described pre-viously (4). Rabbit monoclonal antibodies toward Smad3,phospho-Smad3 (Ser 423/425), Akt, and phospho-Akt (Ser473) were purchased from Cell Signaling Technology. Ratmonoclonal anti-Cripto and goat polyclonal anti-GDF3antibodies were purchased from R&D Systems. Mousemonoclonal antibodies toward b-actin and a-tubulinwere purchased from Sigma-Aldrich. For immunoblotting,isolated adipocytes and the SVF were lysed with buffer(20 mmol/L HEPES, pH 7.4, 150 mmol/L NaCl, 1% TritonX-100, 0.2 mmol/L EDTA, and 1 mmol/L dithiothreitol)containing protease and phosphatase inhibitors. The pro-tein extracts (8–10 mg for macrophages and 20 mg forother cells) were loaded onto polyacrylamide gels forelectrophoresis. For imaging of whole-mount epiWAT,euthanized mice were perfused with 40 mL of fresh 1%formaldehyde in PBS via intracardiac injection over a fewminutes. EpiWAT was subdivided into small pieces(;0.1 cm3) by scissors, and was then fixed in 1% form-aldehyde in PBS and blocked in 5% BSA in PBS at roomtemperature for 30 min. For immunostaining of CD11c+

ATMs, cells attached on slide glasses by Cytospin (Thermo

2 Roles of the Insulin-GDF3-ALK7 Axis in Obesity Diabetes

Fisher Scientific) were fixed with 3.7% formaldehyde inPBS for 30 min at room temperature. With permeabiliza-tion by 0.1% Triton X-100, the tissues or the cells wereincubated with 10 mg/mL anti-GDF3 antibody or controlIgG overnight at 4°C followed by the Alexa Fluor 488–conjugated secondary antibody (diluted at 1:500; Invitro-gen) for 1 h at room temperature, and were observed undera laser-scanning confocal microscope. The concentration ofGDF3 in a medium was measured by mouse GDF3 ELISAkit (Elabscience).

Vector Construction and Luciferase Reporter AssayThe binding site of Smad3 and Smad4 (CAGA)14 (12) wasinserted into the pGL4.10[luc2] vector (Promega). Humanembryonic kidney 293 T (HEK293T) cells cultured inDMEM containing 10% FBS and 1 mmol/L L-glutaminewere transfected with 20 ng of the reporter plasmid, 10 ngof the control plasmid pGLA474[hRluc/TK] (Promega),12.5 ng of plasmid containing ALK7 cDNA (4), and 6.25 ngof that containing Cripto cDNA derived from mouse em-bryo, using Lipofectamine 2000 reagent (Invitrogen). After48 h, the recombinant proteins of human GDF3, bonemorphogenetic protein 3 (BMP3), activin B, and TGF-b1(R&D Systems) were added to the medium. After a further24 h, the luciferase activities were measured by the Dual-Luciferase Reporter Assay System (Promega). The lightunits were normalized to Renilla luciferase activity.

Lipolysis AssayIsolated mouse adipocytes (600 mL) were incubated at37°C for 3 h in Krebs-Ringer HEPES buffer (20 mmol/LHEPES, pH 7.4, 120 mmol/L NaCl, 5 mmol/L KCl,2 mmol/L CaCl2, 1 mmol/L MgCl2, and 1 mmol/LKH2PO4) containing 2 mmol/L glucose and 1% fattyacid–free BSA. Lipolysis was assessed by measuring theconcentration of glycerol in the buffer using a Free Glyc-erol Determination Kit (Sigma-Aldrich).

Statistical AnalysisAll quantitative data were expressed as the mean 6 SD.Data analysis used GraphPad Prism software. The P valueswere calculated using Student t test, one-way ANOVA withTukey multiple-comparison test, or repeated-measuresANOVA with Bonferroni multiple-comparison test, asappropriate, to determine significant differences betweengroup means.

RESULTS

GDF3 Produced from CD11c+ ATMs Functions asa Ligand of ALK7 in AdipocytesBecause ALK7 knockout mice show reduced fat accumu-lation when fed an HFD, but exhibit normal weight whenfed regular chow (9), the ALK7 signal could be activatedunder nutrient-excess conditions. We thus screened TGF-bsuperfamily members that exhibit differential expressionsdepending on nutritional states and also between theabsence or presence of functional ALK7. For this purpose,we isolated tissues potentially involved in nutritional

metabolism from ALK7-intact C57BL/6 and ALK7-deficient BALB/c lean mouse strains fed either regularchow or an HFD. We also isolated these tissues from ALK7-intact TSOD and ALK7-deficient T.B-Nidd5/3 obese mousestrains, both of which have the same genetic background(3,4). Among the 33 members of the mammalian TGF-bsuperfamily (13), GDF3, BMP3, inhibin-bB, and TGF-b1showed differential expression in WAT (Fig. 1A and Sup-plementary Fig. 1). Their expression in WAT was stronglyupregulated in C57BL/6 mice fed an HFD compared withthose fed regular chow. Some of them were also upregulatedin obese TSOD mice fed regular chow and even in ALK7-deficient BALB/c mice fed an HFD. In contrast to the otherthree ligands, GDF3 showed a remarkably high and specificexpression in epiWAT of TSOD and HFD-fed C57BL/6 mice,which is consistent with previous findings (4,10). We thenexamined the ligand activity through ALK7 inHEK293T cellsexpressing a luciferase reporter containing a Smad3/4 re-sponsive element (12), which acts downstream of ALK7 inadipocytes (4). Consistent with a previous finding (9), GDF3activated the reporter in a dose-dependent fashion only inthe presence of exogenously expressed ALK7 and Cripto,a coreceptor that enhances signaling via the type I and type IIreceptor kinase complex (14) (Fig. 1B). In contrast, BMP3 didnot show such enhancement. Activin B, a dimer of inhibin-bB, and TGF-b1 activated the reporter even in the absenceof ALK7 and Cripto, although both induced slight activa-tion with the receptor expression. These findings makeGDF3 the most likely candidate ligand for ALK7.

ALK7-deficient T.B-Nidd5/3 mice at 7 weeks of ageshowed a significant reduction in epiWAT weight com-pared with control TSOD mice (Supplementary Fig. 2A).The mice exhibited increased levels of mRNA encodingthe transcription factors PPARg and C/EBPa, and theirdownstream genes encoding adipose TG lipase (ATGL)and hormone-sensitive lipase (HSL), as previously re-ported in older mice (4). Serum levels of NEFA reflectingenhanced lipolysis were also elevated relative to controlTSOD mice. Therefore, the ALK7-deficient phenotypesbecome overt at 7 weeks of age. GDF3 inhibited lipolysisin adipocytes from TSODmice at this age, whereas BMP3,activin B, or TGF-b1 did not (Fig. 2A), which is consistentwith the findings from the luciferase assays (Fig. 1B).Importantly, GDF3 inhibited lipolysis and activated thedownstream Smad3 by phosphorylation only in ALK7-intact adipocytes from TSOD mice, but not in ALK7-deficient adipocytes from T.B-Nidd5/3 mice (Fig. 2B andC). These findings establish that GDF3 can signal throughALK7 in adipocytes.

Because GDF3 is expressed in thymus, spleen, and BMas well as in WAT (Fig. 1A), as originally reported (15), itmight be expressed in hematopoietic cells rather than adi-pocytes in WAT. To identify the cell source of GDF3, wefirst dissociated the epiWAT into the SVF and mature adi-pocytes, then further fractionated SVF cells by fluorescence-activated cell sorting using fluorochrome-conjugatedantibodies targeting macrophage surface markers (16).

diabetes.diabetesjournals.org Bu and Associates 3

GDF3 transcripts were enriched in the SVF, particularly inCD11b+ F4/80+ macrophages (defined as ATMs), with thegreatest elevation seen in those expressing CD11c (Fig. 2Dand Supplementary Fig. 2B). Immunostaining with anti-

GDF3 antibody revealed that most of the CD11c+ ATMsexpress GDF3 (94.8 6 2.2%; n = 3: ;100 cells wereexamined in total). GDF3+ cells were located around in-dividual adipocytes in WAT, consistent with localization

Figure 1—Screening TGF-b superfamily members to identify ALK7 ligands. A: TSOD mice and their ALK7-deficient counterparts, T.B-Nidd5/3mice, fed regular chow (RC) were sacrificed at 10 weeks of age. C57BL/6 (B6) and BALB/c (BALB) mice fed either RC or an HFD from4 weeks of age were sacrificed at 14 weeks of age. Total RNA was isolated from the indicated tissues, including epiWAT, inguinal WAT(ingWAT), and brown adipose tissue (BAT), and mRNA levels of GDF3, BMP3, inhibin-bB, and TGF-b1 were quantified and normalized to theaverage values in epiWAT of C57BL/6 mice fed RC (n = 3). B: HEK239T cells were transfected with plasmids encoding ALK7 and/or Cripto.The protein levels of ALK7 and Cripto were examined at 48 h post-transfection by immunoblotting (left panel). HEK293T cells weretransfected with plasmids encoding ALK7 and Cripto, and simultaneously with a luciferase reporter fused with the Smad-binding promoterelement. At 48 h post-transfection, different concentrations (0, 50, 150, and 400 ng/mL) of the indicated recombinant protein were added tothe cells. The luciferase activities were measured after a further 24 h (middle panel: GDF3, n = 4; BMP3, n = 3; right panel, n = 3). *P, 0.05,**P , 0.01, ***P , 0.001; Student t test. +P , 0.05, ++P , 0.01, +++P , 0.001; repeated-measures ANOVA.

4 Roles of the Insulin-GDF3-ALK7 Axis in Obesity Diabetes

in ATMs. In contrast, BMP3 and inhibin-bB were ex-pressed mainly in mature adipocytes, whereas TGF-b1was ubiquitously expressed in every cell fraction (Supple-mentary Fig. 2C). Concomitant increases in the CD11c andGDF3 transcripts were also found in the SVF of HFD-fedC57BL/6 mice (Supplementary Fig. 1D). Although inflam-masome activation has recently been shown to induceGDF3 in ATMs from aged mice (17), the GDF3 inductionin TSOD or HFD-treated C57BL/6 mice was not accom-panied by the upregulation of inflammasome activation–related genes, such as tumor necrosis factor-a (TNF-a),

monocyte chemotactic protein-1 (MCP-1), NLR family pyrindomain containing 3 (NLRP3), and Caspase-1 (Supple-mentary Fig. 1B and D).

MacrophageDepletion Reverses the Effects of ALK7 onAdiposityTo evaluate the role of GDF3-producing ATMs in vivo, weintraperitoneally injected clodronate to deplete macro-phages (18). Clodronate treatment partially but signifi-cantly decreased the percentage of ATMs, including that ofCD11c+ ATMs, as well as the expression of F4/80, CD11c,

Figure 2—GDF3 acts on ALK7 within WAT. Primary adipocytes derived from epiWAT of 7-week-old TSOD or T.B-Nidd5/3 mice wereincubated with the indicated recombinant protein (400 ng/mL) for 3 h (A and B) or 30 min (C). Glycerol release was measured and normalizedto the average values of control TSOD adipocytes (A andB, n = 3). Phosphorylation of Smad3 in adipocyteswas examined by immunoblottingwith the indicated antibodies (C). The band with a black arrowhead in the p-Smad3 panel is a nonspecific protein. D: EpiWAT of 7-week-oldTSODmice were biochemically separated into adipocytes and the SVF. The SVF was then fractionated by FACS as shown in SupplementaryFig. 2B. GDF3mRNA levels were quantified in each of the cell fractions (left panel: epiWAT, n = 4; adipocytes, n = 3; SVF, n = 4; CD11b2 cellsin the SVF, n = 5; CD11b+ cells in the SVF, n = 4; CD11b+ F4/802 nonmacrophage cells, n = 3; CD11b+ F4/80+ macrophages, n = 3; CD11c2

macrophages, n = 4; and CD11c+ macrophages, n = 6). The extracts of CD11c+ and CD11c2 ATMs were immunoblotted with the indicatedantibodies (middle top panel). CD11c+ ATMs (middle bottom panel: white bars, 10 mm) and whole-mount epiWAT (right panel: yellowbars, 50 mm) were immunostained with control IgG or anti-GDF3 antibody. ##P , 0.01, ###P , 0.001; one-way ANOVA. +P , 0.05, ++P ,0.01, +++P , 0.001; repeated-measures ANOVA.

diabetes.diabetesjournals.org Bu and Associates 5

and GDF3, in both TSOD and T.B-Nidd5/3mice (Fig. 3A andSupplementary Fig. 3A and B). However, clodronate de-creased total body weight, particularly epiWAT weight, onlyin TSODmice, indicating that the effects of the drug dependon intact ALK7. Furthermore, clodronate increased thePPARg, C/EBPa, ATGL, and HSL transcripts, and the serumNEFA concentration normalized to the epiWAT weight, inTSOD mice (Fig. 3B). Therefore, the effects of macrophagedepletion from ALK7-intact TSOD mice are remarkablysimilar to the phenotypic changes in ALK7-deficient T.B-Nidd5/3mice when comparedwith control TSODmice (3,4),indicating that the GDF3-ALK7 axis represents a major linkbetween macrophages and adipocytes in the regulation ofwhole-body lipid metabolism and fat accumulation.

Insulin Upregulates GDF3 in ATMsWe next explored the external factors that increase GDF3production under nutrient-excess conditions. CD11c2

ATMs isolated from epiWAT of TSOD mice showed ele-vated levels of the GDF3 transcript during culture in FBS-containing medium (Supplementary Fig. 4A), suggestingthat some FBS component converts CD11c2 to CD11c+

ATMs and concomitantly induces GDF3 expression. Be-cause obesity is frequently coincident with hyperinsuline-mia, we suspected that insulin might upregulate GDF3.Plasma insulin concentrations are ;170 pmol/L in leanBALB/c mice and ;1.7 nmol/L in obese TSOD mice (2).Ex vivo administration of 10 mU/mL insulin (61 pmol/L)increased expression of both CD11c and GDF3 after a 24-hculture in CD11c2 macrophages derived from epiWAT ofTSOD mice, and wortmannin, an inhibitor of phosphati-dylinositol 3-kinase, inhibited insulin-induced GDF3 upre-gulation (Fig. 4A). Insulin also increased the expression ofthe typical M2 markers arginase and chitinase-like 3, butnot that of the M1 markers TNF-a and MCP-1. Although61 pmol/L insulin induced GDF3 in ATMs, it increased

Figure 3—Effects ofmacrophage depletion by clodronate. PBS or clodronate encapsulated in liposomes (CLO) was injected intraperitoneallyinto TSOD and T.B-Nidd5/3mice twice a week for 3 weeks from 4weeks of age. Three days after the final injection at 7 weeks of age, the SVFwas isolated from epiWAT. A: The mRNA level of GDF3 in SVF (TSOD, n = 8; T.B-Nidd5/3, n = 4), body weights at 4 weeks of age, and bodyand epiWAT weights and their ratio at 7 weeks of age (n = 5) in mice with or without CLO treatment. B: The mRNA levels of adiposetranscription factors and lipases in epiWAT and serum NEFA concentrations normalized to the epiWAT weight (TSOD, n = 8; T.B-Nidd5/3,n = 4). #P , 0.05, ##P , 0.01, ###P , 0.001; one-way ANOVA.

6 Roles of the Insulin-GDF3-ALK7 Axis in Obesity Diabetes

GDF3 only weakly in macrophages derived from lung,peritoneum, or BM of TSOD mice (Supplementary Fig.4B). This was evident in the low level of expression of theinsulin receptor in these macrophages in contrast to that

in ATMs. These findings indicate the tissue selectivity ofinsulin sensitivity in macrophages.

Although the above findings raise the possibility thatinsulin inhibits lipolysis and accumulates fat in adipocytes

Figure 4—Effects of insulin administered to CD11c2 ATMs or adipocytes. A: CD11c2macrophages from epiWAT of 7-week-old TSODmice(1.53 106 cells/24-well dish) were incubated with or without 61 pmol/L insulin in Krebs-Ringer HEPES buffer for 24 h (left panel; n = 7). Somewere pretreated with the indicated concentration of wortmannin 10min before the 24-h incubation (right panel; n = 3). ThemRNA levels of theindicated genes were quantified and normalized to those without insulin incubation in each experiment (middle panel). Insulin-inducedphosphorylation of Akt in CD11c2 macrophages pretreated with or without 100 nmol/L wortmannin was examined by immunoblotting withthe indicated antibodies (right panel). B: Primary adipocytes isolated from epiWAT of TSOD or T.B-Nidd5/3mice were incubated for 30 minwith 0 mol/L, 61 pmol/L, or 25 nmol/L insulin or with 400 ng/mL GDF3. The cell extracts were immunoblotted with the indicated antibodies.The band with a black arrowhead in the p-Smad3 panel is a nonspecific protein. C: Adipocytes were incubated with or without 10 mmol/Lisoproterenol plus 0 mol/L, 61 pmol/L, or 25 nmol/L insulin for 3 h. Glycerol levels in the medium were measured and normalized to theaverage value of TSOD adipocytes without insulin or isoproterenol incubation (n = 3). D and E: CD11c2 ATMs from TSODmice were culturedwith or without 61 pmol/L insulin for 24 h as in A. The conditioned medium of the macrophages was harvested after centrifugation of theculture plate at 300g for 10min at 4°C, and the concentration of GDF3wasmeasured (D, left panel; n = 4). The conditionedmediumwarmed to37°C was incubated with adipocytes of TSOD mice for 30 min to examine its effect on Smad3 phosphorylation (D, right panel), or withadipocytes of TSOD (n = 5) or T.B-Nidd5/3 (n = 3) mice for 3 h to examine its effect on glycerol release (E ). *P, 0.05, **P, 0.01, ***P, 0.001;Student t test. #P , 0.05, ###P , 0.001; one-way ANOVA. +P , 0.05, ++P , 0.01, +++P , 0.001; repeated measures ANOVA.

diabetes.diabetesjournals.org Bu and Associates 7

through the upregulation of GDF3 in ATMs, insulin isgenerally believed to do so by directly acting on adipocytes.We next investigated the effects of insulin on isolated

adipocytes. We confirmed that insulin phosphorylates thedownstream Akt kinase, but does not activate Smad3 bya noncanonical pathway, in adipocytes (Fig. 4B). However,

Figure 5—Effects of insulin administered to a whole body. A: Saline or insulin (0.75 units/kg body weight) was injected intraperitoneally intoTSOD (top panels; n = 10) and T.B-Nidd5/3 (bottompanels; n = 8)mice twice daily for 2 weeks from 5weeks of age. Shown are theweight ratioof epiWAT to total body, mRNA levels of ATGL and HSL in epiWAT, and serum NEFA concentration normalized to the epiWAT weight. B:Saline or insulin was injected to C57BL/6 (top panels; n = 8) and BALB/c mice (bottom panels; n = 8) that had been fed an HFD for 3 weeksfrom 4 weeks of age, and the effects of insulin were examined as in A. *P , 0.05, **P , 0.01; Student t test.

8 Roles of the Insulin-GDF3-ALK7 Axis in Obesity Diabetes

Figure 6—Insulin regulates fat metabolism and mass through the upregulation of GDF3 in ATMs. A: C57BL/6 mice fed an HFD (n = 7 pergroup) were treatedwith PBS or clodronate from 4weeks of age for 3 weeks, as described in Fig. 3, andwere also treatedwith saline or insulinfrom 5 weeks of age for 2 weeks, as described in Fig. 5B. Shown are GDF3 mRNA levels in the SVF, the weight ratio of epiWAT to total body,serum NEFA concentration normalized to the epiWAT weight, and ATGL and HSL mRNA levels in epiWAT. B: The BM of WT or GDF3knockout (KO) C57BL/6micewere transplanted intoWTC57BL/6mice at 8–10weeks of age. The recipientmice were fed anHFD for 3 weeksand treated with insulin for 2 weeks from 6 and 7 weeks after the BM transfer, respectively (n = 6–9/group). #P , 0.05, ##P , 0.01, ###P ,0.001; one-way ANOVA. C: Scheme of the insulin-GDF3-ALK7 axis. See text in DISCUSSION.

diabetes.diabetesjournals.org Bu and Associates 9

the concentration of insulin (61 pmol/L) we adminis-tered ex vivo to ATMs (Fig. 4A) did not inhibit basal orcatecholamine-induced lipolysis in adipocytes, although ahigher concentration of insulin (25 nmol/L) did so (Fig. 4C).These findings indicate that a much higher dose of insulinis required to directly inhibit lipolysis in adipocytes thanis required to upregulate GDF3 in ATMs. Although ALK7deficiency has been reported to enhance catecholamine-induced lipolysis in adipocytes (19), we found that unsti-mulated lipolysis is already elevated, and that the extent ofstimulation by catecholamine is not changed in ALK7-de-ficient adipocytes (Fig. 4C). These findings confirm theprevious finding that ALK7 deficiency elevates basal lipol-ysis by affecting the expression levels of adipose lipases (4).

To reinforce the functional significance of the activityof insulin through GDF3 production from ATMs, we per-formed reconstitution assays by incubating adipocyteswith the supernatant of CD11c2 ATMs that had beentreated with or without 61 pmol/L insulin. Note that thisconcentration of insulin does not directly inhibit lipolysisin isolated adipocytes (Fig. 4C). Insulin induced the secre-tion of GDF3 into their supernatants (1–2 ng/mL), whichdose-dependently increased the phosphorylation of Smad3and inhibited lipolysis in adipocytes of ALK7-intact TSODmice, but not in those of ALK7-deficient T.B-Nidd5/3mice(Fig. 4D and E). We confirmed that 61 pmol/L insulin didnot change the expression levels of inhibin-bB and TGF-b1in the ATMs (Supplementary Fig. 4C), both of which caninduce Smad3 phosphorylation in adipocytes. These find-ings indicate that the insulin-stimulated release of GDF3from ATMs successfully inhibits lipolysis in adipocytesex vivo.

Insulin Inhibits Lipolysis and Accumulates Fat in anALK7-Dependent Manner In VivoTo clarify whether insulin functions through the GDF3-ALK7 signaling pathway in vivo, we intraperitoneallyadministered insulin twice a day for 2 weeks to TSODand T.B-Nidd5/3 mice. This in vivo insulin treatmentelevated the WAT weight, and decreased the levels ofthe ATGL and HSL transcripts and serum NEFA, in anALK7-dependent manner (Fig. 5A), suggesting that insulininhibits lipolysis and accumulates fat through the upre-gulation of GDF3 in ATMs.

In order to exclude the possibility that the effects ofinsulin via the GDF3-ALK7 axis are applicable only to theTSOD strain, for which the molecular pathogenesis ofobesity and diabetes is unknown (2), we administeredinsulin to a commonly used C57BL/6 strain fed an HFDthat indeed expressed ALK7 in WAT (Supplementary Fig.5A). Insulin increased adiposity in parallel with reductionsin the expression of adipose lipases in epiWAT and serumNEFA concentrations in C57BL/6 mice (Fig. 5B). However,no such effects were found in ALK7-deficient BALB/c micefed an HFD. These findings indicate that the effects ofinsulin via ALK7 under nutrient-excess conditions con-tinue irrespective of the mouse strain.

GDF3 Mediates the Activity of Insulin to PromoteAdiposity In VivoTo further substantiate the role of the GDF3-ALK7 axis ininsulin activity in vivo, we injected clodronate to depletemacrophages and then administered insulin to C57BL/6mice fed an HFD. We confirmed that neither clodronatenor insulin treatment alters the food intake of mice(Supplementary Fig. 5B). Clodronate treatment markedlydecreased ATMs, including CD11c+ ATMs, and concomi-tantly reduced GDF3 levels in the SVF (Fig. 6A andSupplementary Fig. 5B). Remarkably, it eliminated thein vivo effects of insulin to increase CD11c and GDF3in the SVF and adiposity in the whole body, and to decreaseadipose lipases and the serum concentration of NEFA.These findings demonstrate that insulin can regulate fatmetabolism and mass through its effects on macrophagesin vivo.

Finally, we performed BM transplantation experimentsto directly prove the involvement of GDF3 in the insulinactivity. We transplanted the BM of GDF3-deficientC57BL/6 mice (10) to wild-type C57BL/6 mice to evadethe cell elimination by the immune system due to MHCmismatch. The recipient mice were then fed an HFD andtreated with insulin. We confirmed that GDF3 deficiency inBM cells does not affect the number of ATMs (Supple-mentary Fig. 5C). In contrast to the mice harboring thewild-type BM, those harboring the GDF3-deficient BM,and thus losing GDF3 in the SVF failed to mediate thein vivo effects of insulin to inhibit lipolysis in the WAT(Fig. 6B). These findings demonstrate that GDF3 produc-tion is necessary for insulin to regulate fat metabolism andmass under nutrient-excess conditions.

DISCUSSION

We showed that GDF3 produced from CD11c+ ATMs actsas a ligand of ALK7 in adipocytes to inhibit lipolysis andaccumulate fat under nutrient-excess conditions. TheGDF3-ALK7 axis within WAT should represent the majorinteractive mechanism between macrophages and adipo-cytes in the regulation of adiposity, because nonselectivemacrophage depletion by clodronate highlights the ALK7-specific effects, such as decreases in body and epiWATweights, and increases in the expressions of C/EBPa,PPARg, ATGL, and HSL, as well as NEFA production inWAT, in ALK7-intact TSOD mice, but not in their ALK7-deficient counterparts. Although many studies have fo-cused on the effects of macrophages in the formation ofchronic inflammation associated with obesity, the currentstudy demonstrates the role of ATMs in fat accumulationper se. Although CD11c+ macrophages are conventionallyunderstood to be M1 macrophages that are recruited toand/or polarized in obese WAT to induce a chronic in-flammatory state (20), the GDF3-producing cells expressa substantial level of M2 markers. Similar to our findings,it has recently been shown that a prototypical M2 marker,CD301b, as well as arginase, are selectively expressed inCD11c+ mononuclear phagocytes including ATMs, and

10 Roles of the Insulin-GDF3-ALK7 Axis in Obesity Diabetes

that depleting these cells leads to weight loss and increasedinsulin sensitivity in mice (21).

We found that a physiologically low concentration ofinsulin alters the properties of CD11c2 ATMs ex vivo byincreasing the expressions of CD11c and GDF3. Moreover,in vivo insulin administration inhibits lipolysis andexpands WAT in an ALK7-dependent manner, which indi-cates that insulin regulates fat metabolism and mass viathe GDF3-ALK7 axis. Consistently, the in vivo effects ofinsulin on WAT are absent after the depletion of macro-phages or the transplantation of GDF3-deficient BM. It isintriguing that ATMs appear to specifically express a highlevel of insulin receptor compared with macrophages inother tissues. Although insulin is generally thought toinhibit lipolysis directly in adipocytes by regulating thecAMP-mediated signaling pathway (22–25) and/or by sup-pressing the transcription of adipose lipases (26–28), theseactions in adipocytes have been detected only at higherconcentrations of insulin (1–100 nmol/L) compared withthose applied to ATMs in the current study (61 pmol/L).In fact, we observed that 25 nmol/L insulin can directlyinhibit both basal and catecholamine-stimulated lipolysisin adipocytes, whereas 61 pmol/L insulin cannot. There-fore, insulin can differentially regulate fat metabolismand mass depending on its local concentration in WAT.The novel pathway through ATMs might explain theexquisite sensitivity of lipolysis to suppression by insulininfused into human subjects (half-maximal suppression at101 pmol/L) (29).

Given that GDF3 induction in ATMs requires a minimalconcentration of insulin, the GDF3-ALK7 pathway shouldbe active at the beginning of hyperinsulinemia undernutrient-excess conditions. This has a clinical implica-tion for the importance of “early intervention” in adi-posity before the manifestation of insulin resistance.Insulin resistance–related chronic hyperinsulinemia mayaccelerate fat accumulation even under the same energybalance via activation of the GDF3-ALK7 axis, whichmakesit much harder for obese individuals to reduce adiposity.Future research should focus on novel targeting strategiesfor this pathway, such as inhibitors of GDF3 and ALK7,specific depletion of ATMs, and macrophage-specific in-hibition of insulin receptor expression.

In summary, we present a novel mechanism of obesity(Fig. 6C). Under nutrient-excess conditions, insulin effi-ciently activates insulin receptors expressed on CD11c2

ATMs and converts them to CD11c+ ATMs to produceGDF3. GDF3 locally stimulates ALK7 on adipocytes andactivates Smads 2–4 to downregulate PPARg, C/EBPa, andalso adipose lipases to store excess nutrient as fat (4).However, persistent activation of this physiological pathwayenlarges adipocytes and may change adipocytokine reper-toires to cause chronic inflammation and insulin resistance.In fact, ALK7-intact, aged obese mice exhibit elevated levelsof proinflammatory MCP-1 and TNF-a, a reduced level ofinsulin-sensitizing adiponectin, and greater glucose intoler-ance, compared with their ALK7-deficient counterparts

(4,5). As such, the insulin-GDF3-ALK7 axis plays a pivotalrole in both physiological and pathological fat accumulationin WAT.

Acknowledgments. The authors thank the members of the Laboratory ofMolecular Endocrinology and Metabolism, Gunma University, particularly T. Nara,E. Kobayashi, and T. Ushigome for colony maintenance of mice and S. Shigokafor assistance in preparing the manuscript. The authors also thank the staffs atBioresource Center, Gunma University for help in the breeding of the mice.Funding. This work was supported by the Japan Society for the Promotion ofScience KAKENHI grant JP25860739 to S.Y.; and grants JP24659442 andJP25126702 to T.I.; and grants from the Japan Diabetes Foundation and theNovo Nordisk Insulin Study Award (to T.I.).Duality of Interest. No potential conflicts of interest relevant to this articlewere reported.Author Contributions. Y.B., S.Y., and K.M. performed experiments. K.O.performed experiments, designed experiments, and wrote the paper. M.J.I. andC.W.B provided experimental reagents. T.I. designed experiments and wrote thepaper. T.I. is the guarantor of this work and, as such, had full access to all the datain the study and takes responsibility for the integrity of the data and the accuracyof the data analysis.

References1. Gregor MF, Hotamisligil GS. Inflammatory mechanisms in obesity. Annu RevImmunol 2011;29:415–4452. Hirayama I, Yi Z, Izumi S, et al. Genetic analysis of obese diabetes in theTSOD mouse. Diabetes 1999;48:1183–11913. Mizutani S, Gomi H, Hirayama I, Izumi T. Chromosome 2 locus Nidd5 hasa potent effect on adiposity in the TSOD mouse. Mamm Genome 2006;17:375–3844. Yogosawa S, Mizutani S, Ogawa Y, Izumi T. Activin receptor-like kinase7 suppresses lipolysis to accumulate fat in obesity through downregulation ofperoxisome proliferator-activated receptor g and C/EBPa. Diabetes 2013;62:115–1235. Yogosawa S, Izumi T. Roles of activin receptor-like kinase 7 signaling and itstarget, peroxisome proliferator-activated receptor g, in lean and obese adipocytes.Adipocyte 2013;2:246–2506. Reissmann E, Jörnvall H, Blokzijl A, et al. The orphan receptor ALK7 and theActivin receptor ALK4 mediate signaling by Nodal proteins during vertebratedevelopment. Genes Dev 2001;15:2010–20227. Tsuchida K, Nakatani M, Yamakawa N, Hashimoto O, Hasegawa Y, Sugino H.Activin isoforms signal through type I receptor serine/threonine kinase ALK7. MolCell Endocrinol 2004;220:59–658. Andersson O, Reissmann E, Ibáñez CF. Growth differentiation factor11 signals through the transforming growth factor-b receptor ALK5 to regionalizethe anterior-posterior axis. EMBO Rep 2006;7:831–8379. Andersson O, Korach-Andre M, Reissmann E, Ibáñez CF, Bertolino P. Growth/differentiation factor 3 signals through ALK7 and regulates accumulation ofadipose tissue and diet-induced obesity. Proc Natl Acad Sci U S A 2008;105:7252–725610. Shen JJ, Huang L, Li L, Jorgez C, Matzuk MM, Brown CW. Deficiency ofgrowth differentiation factor 3 protects against diet-induced obesity by selectivelyacting on white adipose. Mol Endocrinol 2009;23:113–12311. Suzuki W, Iizuka S, Tabuchi M, et al. A new mouse model of spontaneousdiabetes derived from ddY strain. Exp Anim 1999;48:181–18912. Dennler S, Itoh S, Vivien D, ten Dijke P, Huet S, Gauthier JM. Directbinding of Smad3 and Smad4 to critical TGF b-inducible elements in thepromoter of human plasminogen activator inhibitor-type 1 gene. EMBO J 1998;17:3091–310013. Shi M, Zhu J, Wang R, et al. Latent TGF-b structure and activation. Nature2011;474:343–349

diabetes.diabetesjournals.org Bu and Associates 11

14. Chen C, Ware SM, Sato A, et al. The Vg1-related protein Gdf3 acts in a Nodalsignaling pathway in the pre-gastrulation mouse embryo. Development 2006;133:319–32915. McPherron AC, Lee SJ. GDF-3 and GDF-9: two new members of thetransforming growth factor-b superfamily containing a novel pattern of cysteines.J Biol Chem 1993;268:3444–344916. Gordon S, Taylor PR. Monocyte and macrophage heterogeneity. Nat RevImmunol 2005;5:953–96417. Camell CD, Sander J, Spadaro O, et al. Inflammasome-driven catecholaminecatabolism in macrophages blunts lipolysis during ageing. Nature 2017;550:119–12318. Van Rooijen N, Sanders A. Liposome mediated depletion of macrophages:mechanism of action, preparation of liposomes and applications. J ImmunolMethods 1994;174:83–9319. Guo T, Marmol P, Moliner A, et al. Adipocyte ALK7 links nutrient overload tocatecholamine resistance in obesity. eLife 2014;3:e0324520. Lumeng CN, Bodzin JL, Saltiel AR. Obesity induces a phenotypic switchin adipose tissue macrophage polarization. J Clin Invest 2007;117:175–18421. Kumamoto Y, Camporez JPG, Jurczak MJ, et al. CD301b(+) mononuclearphagocytes maintain positive energy balance through secretion of resistin-likemolecule alpha. Immunity 2016;45:583–596

22. Burns TW, Terry BE, Langley PE, Robison GA. Insulin inhibition of lipolysis ofhuman adipocytes: the role of cyclic adenosine monophosphate. Diabetes 1979;28:957–96123. Choi YH, Park S, Hockman S, et al. Alterations in regulation of energyhomeostasis in cyclic nucleotide phosphodiesterase 3B-null mice. J Clin Invest2006;116:3240–325124. Lafontan M, Langin D. Lipolysis and lipid mobilization in human adiposetissue. Prog Lipid Res 2009;48:275–29725. Degerman E, Ahmad F, Chung YW, et al. From PDE3B to the regulation ofenergy homeostasis. Curr Opin Pharmacol 2011;11:676–68226. Kralisch S, Klein J, Lossner U, et al. Isoproterenol, TNFalpha, and insulindownregulate adipose triglyceride lipase in 3T3-L1 adipocytes. Mol Cell Endocrinol2005;240:43–4927. Kershaw EE, Hamm JK, Verhagen LA, Peroni O, Katic M, Flier JS. Adiposetriglyceride lipase: function, regulation by insulin, and comparison with adipo-nutrin. Diabetes 2006;55:148–15728. Kim JY, Tillison K, Lee JH, Rearick DA, Smas CM. The adipose tissue tri-glyceride lipase ATGL/PNPLA2 is downregulated by insulin and TNF-a in 3T3-L1adipocytes and is a target for transactivation by PPARgamma. Am J PhysiolEndocrinol Metab 2006;291:E115–E12729. Campbell PJ, Carlson MG, Nurjhan N. Fat metabolism in human obesity. Am JPhysiol 1994;266:E600–E605

12 Roles of the Insulin-GDF3-ALK7 Axis in Obesity Diabetes