AMPK Activation by Metformin Suppresses Abnormal Extracellular Matrix Remodeling in Adipose Tissue and Ameliorates Insulin Resistance in Obesity

In obesity, visceral adipose tissue dysfunction contributes to the development of insulin resistance and type 2 diabetes mellitus. Physiologically, extracellular matrix (ECM) components such as collagens and fibronectin provide mechanical support for tissue. As obesity-induced adipose tissue expansion progresses, chronic, continuous production and deposition of ECM lead to the destruction of normal adipose tissue architecture, which is referred to as adipose tissue ECM remodeling (1). Genetic deletion of collagen VI (Col6a3), which is highly expressed in adipose tissue, improves adipose fibrosis and systemic metabolic dysfunction in mice fed a high-fat diet, highlighting the critical role of adipose tissue ECM in regulating metabolic homeostasis (2). Adipose tissue fibrosis is increasingly appreciated as a major contributor of metabolic dysregulation in obese humans and those with type 2 diabetes (3). However, how persistent aberrant ECM remodeling is initiated by obesity remains poorly understood. The major goal of this study is to identify the signaling molecules that modulate fat tissue ECM remodeling in obesity.

Metformin is the first-line drug that treats obesity-related type 2 diabetes (4,5). Our initial studies and others indicate that the lipid-lowering effect of metformin is largely attributable to the activation of AMPK, the energy sensor, in hepatocytes (6–9). Furthermore, metformin can repress de novo lipogenesis in hepatocytes by activating AMPK (10). Although the adipose tissue acts as a target of AMPK and its activator metformin (11–14), whether AMPK regulates adipose tissue ECM homeostasis remains undefined.

We report here that the interplay of AMPK and transforming growth factor (TGF)-β1 signaling is implicated in pericellular fibrosis in the omental fat of obese humans. Activation of AMPK by metformin suppresses TGF-β1/Smad3 signaling, resolves the interstitial accumulation of ECM in white adipose tissue (WAT), and improves local and systemic insulin resistance in animals with genetic and diet-induced obesity. Pharmacological and molecular activation of AMPK inhibits collagen induction, lysyl oxidase (LOX)–related collagen cross-linking, myofibroblast-like cell differentiation, and adipocyte apoptosis and insulin resistance upon TGF-β1 treatment in vitro. Targeting adipose tissue ECM may benefit adipose and whole-body insulin resistance in obesity.

Male 8- to 10-week-old ob/ob C57BL/6 mice, as well as their lean littermates and wild-type mice, were purchased from The Jackson Laboratory (Bar Harbor, ME). All animal experiments were approved by the Institutional Animal Care and Use Committee at Boston University School of Medicine.

Stromal vascular fraction (SVF) cells in epididymal fat depots from male C57BL/6 mice were isolated and cultured as described previously (15,16).

Omental adipose tissue samples were collected from obese patients (BMI 28.7 ± 1.4 kg/m²) and age-matched normal-weight individuals (BMI 19.8 ± 1.5 kg/m²). The ethics committee of the First Affiliated Hospital of Chongqing Medical University in China approved the study (IRB#15–39). Subjects gave written informed consent after individual explanation.

Data are presented as mean ± SEM. Using GraphPad Prism 5.0 software, results were analyzed by one-way ANOVA (between multiple groups), when appropriate, and by a two-tailed Student t test (between two groups), as described previously (6,8,10,17–20). P < 0.05 was considered statistically significant.

As shown in Fig. 1, body weight gain was moderately but significantly reduced in metformin-treated ob/ob mice, as was seen in early studies (21). While lean mass was unaffected by metformin, the weight loss was mainly attributed to a 14% decrease in the fat mass of ob/ob mice, consistent with a moderate weight loss and fat mass reduction in patients with type 2 diabetes after metformin treatment (14,22,23). Histologically, hematoxylin-eosin (H-E) staining showed that metformin lessened features of hypertrophic adipocytes in epididymal WAT (eWAT) of ob/ob mice, with a switch toward smaller adipocytes. Masson trichrome staining showed an approximately sevenfold increase in collagen deposition and distinct distribution patterns of interstitial collagen in eWAT of ob/ob mice, in which fibrous regions appeared denser and thicker and gave rise to a larger lobe or a lobule-like structure. Sirius Red staining further confirmed that collagen fibers were mainly organized in fibrotic bundles surrounding adipocytes in ob/ob mice. Strikingly, abnormal collagen deposition was reduced approximately 50% in metformin-treated ob/ob mice. Quantification of hydroxyproline concentrations is considered an accurate indicator of tissue collagen content (24–26). An increase in hydroxyproline content in the eWAT of ob/ob mice was also reduced by metformin, reaching a concentration comparable with that of lean mice. Thus metformin protects against adiposity and pronounced ECM remodeling caused by obesity.

To determine the effect of metformin on systemic glucose homeostasis, glucose tolerance tests showed that glucose clearance was significantly enhanced by metformin compared with ob/ob mice. Insulin tolerance tests also showed that systemic insulin sensitivity was increased by metformin. Integrated glucose concentrations, as calculated by the area under the curve were consistently significantly decreased. Likewise, fasting blood glucose concentrations were reduced ∼30% by metformin. A positive correlation between fat collagen content and fasting blood glucose was observed in ob/ob mice treated or untreated with metformin (r² = 0.6323; P < 0.01) (Fig. 1M–R).

To elucidate the mechanisms by which metformin attenuates pericellular fibrosis, we measured mRNA expression of collagen type I (Col1a1), type III (Col3a1), and type VI (Col6a3), all of which were shown to be highly expressed in obese adipose tissue (2). As shown in Fig. 2A, these collagens were increased five- to sevenfold in ob/ob mice, despite the basal level of Col6a3 being much higher than that of Col1a1 and Col3a1 in lean mice, as described previously (2). By contrast, overexpression of these collagens was attenuated ∼50% by metformin. Among major fibrogenic genes, expression of fibronectin, a functional integrin ligand, was also stimulated by obesity and was downregulated by metformin. LOX, a copper-dependent amine oxidase that catalyzes the covalent cross-linking of fibers through posttranslational modification of collagens, is essential for the cross-linking of collagen fibrils and fibers and the modification of biochemical and mechanical properties of the ECM (27). The antifibrotic effect of metformin was also evidenced by downregulation of fat LOX in ob/ob mice.

The development of tissue fibrosis occurs as a consequence of the dysregulation of matrix metalloproteinases (MMPs) and tissue inhibitors of MMPs (TIMPs) (28–31). To test the hypothesis that fat collagen deposition might be suppressed by metformin through altered matrix degradation, mRNA levels of MMP-2, MMP-9, MMP-13, and MT1-MMP (MMP-14), which are predominantly upregulated in obese adipose tissue (32), were assessed by quantitative RT-PCR. As shown in Fig. 2B–I, expression of different MMPs was induced 5- to 10-fold in eWAT of ob/ob mice, suggesting that the perturbation of the normal fat ECM structure and/or its replacement with newly synthesized ECM may act as a driver to promote a metabolically unfavorable fat microenvironment. Interestingly, the induction of MMP-9 and TIMP-1 by obesity was reduced ∼60% by metformin. The protein expression and distribution of MMP-9 and TIMP-1, as assessed by immunohistochemical and immunoblotting analyses, were also reduced. No significant changes in TIMP-2 and TIMP-3 were evident among the three groups. Collectively, metformin may inhibit abnormal ECM synthesis and degradation in fat tissue, reduce LOX-related collagen cross-linking, and consequently provoke a favorable ECM flexibility.

To better understand the potential effect of metformin on fat ECM remodeling process in obesity, we expanded our studies to a diet-induced obesity animal model. As shown in Fig. 3, hypertrophic adipocytes in epididymal and inguinal fat of mice with high fat/high sucrose (HFHS)–induced obesity were significantly reduced by metformin. Masson trichrome staining revealed that the collagen fibers surrounding adipocytes were increased seven- and threefold in the eWAT and inguinal WAT (iWAT) of HFHS-fed mice, respectively. To our surprise, the abundance of collagen deposition had a distinct feature between fat depots of obese mice: higher amounts were found in eWAT and much lower amounts in iWAT. Since the basal expression levels of Col1a1, Col6a3, and fibronectin in epididymal fat, a visceral fat associated with metabolic disease, were approximately fivefold higher than those in subcutaneous fat, it may reflect a depot-specific effect of fat fibrosis. The excessive collagen deposition of eWAT, as assessed histologically and biochemically by hydroxyproline content, was significantly reduced by metformin. The beneficial effect of metformin on interstitial fibrosis was also evidenced by reduced fibrogenic gene expression and concurrently inhibited ECM regulators such as MMP-2, MMP-9, MMP-14, and TIMP-1. The modulation of ECM components also occurred in the iWAT of metformin-treated mice. Decreased hydroxyproline concentrations in eWAT were positively correlated with lowered fasting glucose concentrations in metformin-treated HFHS-fed mice (r² = 0.5911; P < 0.01). Thus, metformin benefits both moderate obesity in HFHS-fed mice and severe obesity in leptin-deficient ob/ob mice.

As shown in Fig. 4, phosphorylation of AMPK at Thr172, required for AMPK activation (33), declined ∼50% in eWAT of both ob/ob mice and HFHS-fed mice, and the impairment was restored by metformin, returning to nearly normal levels. Stimulation of fat AMPK activity was further confirmed by concomitantly increased Ser79 phosphorylation of acetyl-CoA carboxylase (ACC), a well-known substrate of AMPK, consistent with metformin-induced AMPK in insulin-resistant cells (6,7,34) and in the livers of diabetic mice (8,35).

The fibrogenic regulator TGF-β binds to TGF-β receptors I and II and subsequently phosphorylates and activates downstream effectors (Smad2 and Smad3), which translocate into the nucleus and transcriptionally activate fibrogenic target genes (36–38). Because TGF-β1 is implicated in the progression of liver and heart fibrosis (36–38), we determined whether adipose tissue AMPK could be functionally linked to the modulation of TGF-β1. In agreement with elevated plasma TGF-β1 concentrations in obese humans (39), mRNA amounts of TGF-β1 were increased 5.0- and 3.4-fold in the eWAT of ob/ob mice and HFHS-fed mice, respectively. Positive staining for TGF-β1 was consistently highly elevated and located in hypertrophic adipocytes and other cells in the stromal space of ob/ob mice. Strong positive staining for Ser423/425 phosphorylation of Smad3, an indicator of TGF-β1 activity, was primarily located in nuclei of adipocytes and other cells surrounding adipocytes, which overlapped on areas of TGF-β1-positive cells. The induction of TGF-β1 expression and Smad3 phosphorylation were evident in eWAT and iWAT of HFHS-fed mice. Conversely, TGF-β1 activity, as determined by the phosphorylation and nuclear translocation of Smad3, was inhibited by metformin. Taken together, these data show that metformin activates AMPK, inactivates TGF-β1/Smad3 signaling, and ameliorates fat ECM accumulation in obesity.

MMP-mediated degradation of the ECM not only facilitates tissue remodeling but also generates bioactive cleaved peptides and releases chemokines that promote inflammation and apoptosis (31). The crown-like structure positively stained by MAC-2, a lectin that is expressed by activated macrophages (40), was dramatically increased in ob/ob mice (Supplementary Fig. 3A and B). Accordingly, an apoptotic phenotype, as determined by cleaved caspase-3, a key effector that executes the apoptotic program (41), was predominantly located in adipocytes surrounded by crown-like structures (Fig. 4O and P). Consistent with a decreased fat inflammatory response in ob/ob mice lacking Col6a3 (2), the inflammatory response and adipocyte apoptotic phenotypes in ob/ob mice were attenuated by metformin. Interestingly, mRNA levels of proinflammatory cytokines, including tumor necrosis factor-α, interleukin-1β, and MCP-1, were downregulated ∼60% by metformin (Supplementary Fig. 3C). Taken together, these data show that metformin decreases TGF-β/Smad3 signaling and relieves continuous and progressive ECM remodeling in visceral adipose tissue in obesity, activities that are associated with repressed macrophage recruitment and adipocyte apoptosis.

To gain direct evidence for the relationship among AMPK, TGF-β1, and ECM components of fibrotic fat tissue in obese humans, histological and pathological changes in omental fat depots were characterized and compared between obese subjects (n = 8) and age-matched normal-weight subjects (n = 10). As shown in Fig. 5, H-E staining revealed that the pathologically expanded adipose tissue had much larger adipocyte areas in obese subjects than in control subjects. Masson trichrome and Sirius Red stainings showed increased numbers of collagen fibers around adipocytes of obese adipose tissue, with a higher percentage of fibrotic area, revealing fat fibrosis as a pathological process that likely reduces tissue plasticity and function in human obesity. Notably, pericellular fibrotic fat areas of obese subjects were evident in both macrophage-rich and non-macrophage–rich areas. As previously described (42), immunofluorescent analysis showed increased collagen I positive staining around adipocytes in visceral adipose tissue from obese patients, but minimal staining was seen in lean subjects. In agreement with increased fat LOX in ob/ob mice (Fig. 2A), LOX was upregulated in the omental fat of obese humans. The distribution patterns of positive staining for LOX were similar to those of collagen I positive staining, indicating that the activation of LOX-mediated collagen cross-linking may create an insoluble fibrotic matrix of collagen fibers in omental fat depots. Furthermore, immunohistochemistry of adjacent fat sections was conducted to analyze the expression and distribution of TGF-β1 and α-smooth muscle actin (SMA), a well-recognized marker of myofibroblasts. TGF-β1-positive cells were highly increased and primarily located in adipocytes and other cell types in fibrotic adipose tissue of obese subjects. Some of the inappropriately activated α-SMA-positive myofibroblast-like cells were observed. Strikingly, the distribution of TGF-β1-positive cells was overlapped by that of myofibroblast-like cells, suggesting a possible role of TGF-β1 in the differentiation of myofibroblast-like cells that were presumably derived from resident fibroblasts and other cells. Moreover, the expression and distribution of MMP-9- and TIMP-1-positive cells were increased in obese subjects, suggesting that excessive ECM degradation and remodeling may contribute to the pathological destruction of the normal matrix structure of fat depots. Similarly, the accumulated ECM deposition was associated with increased apoptotic adipocytes, as reflected by increased staining for cleaved caspase-3. Similar to inactivation of AMPK in obesity-induced insulin-resistant mice (Fig. 4), phosphorylation of AMPK was diminished in WAT of obese individuals without affecting AMPKα. Fasting glucose concentrations were elevated in obese subjects compared with controls. Collectively, suppression of AMPK and overproduction of TGF-β1 contribute to excessive fat ECM remodeling in obese patients, characterized by collagen overproduction and deposition, myofibroblast-like cell activation, and adipocyte apoptosis.

Adipose tissue, in addition to adipocytes, contains SVF cells including preadipocytes, fibroblasts, vascular cells, and immune cells. Given the hyperactivation of adipose tissue TGF-β1 in obese mice and humans (Figs. 4 and 5), primary SVF cells isolated from epididymal fat pads of lean C57BL/6 mice were treated with TGF-β1 in an effort to develop an in vitro model that mimics in vivo conditions of obesity-induced fat fibrosis. As shown in Fig. 6A–I, mRNA amounts of Col1a1, Col3a1, and Col6a3 markedly increased six- to eightfold over the basal level upon TGF-β1 treatment. This increase was paralleled by the increase of fibronectin and α-SMA. Like MMP-2, expression of MMP-9 was stimulated over 10-fold by TGF-β1, and the induction closely correlated with an increase of TIMP-1, but not TIMP-2. We further determined whether AMPK regulates ECM homeostasis in a cell-autonomous manner. Except the complete inhibition of Col6a3, the stimulation of fibrogenesis by TGF-β1 was reduced in a dose-dependent manner by metformin, to ∼50% the levels of TGF-β1-treated cells. In accord with the repression of collagen synthesis, metformin had an inhibitory effect on ECM degradation regulators (MMP-2, MMP-9, and TIMP-1), consistent with the in vivo effect of metformin (Fig. 2). AICAR, an analog of AMP that activates AMPK, also reduced fibrogenic response to an extent similar to that of metformin. A-769662, a specific activator of AMPK, repressed the induction of collagen and TIMP-1 by TGF-β1, although to a lesser extent. Collectively, AMPK agonists attenuate TGF-β1-induced fibrogenic response.

To identify a possible source of fibrogenic cells within obese adipose tissue, the transformation of myofibroblast-like cells derived from cell origins of the SVF population was assessed upon TGF-β1 induction. As shown in Fig. 6J–L, immunocytochemical analysis showed that in the absence of TGF-β1, α-SMA-positive cells were negligible; exposure to TGF-β1 led to an increase in α-SMA-positive cells. After a 96-h incubation of TGF-β1, uniformly spindle-shaped α-SMA-positive cells were greatly increased, presuming that these cells are likely differentiated from resident fibroblasts and other cell populations in primary SVF cells. Strikingly, the amounts of α-SMA-positive myofibroblast-like cells were reduced by metformin. The reduction of α-SMA-positive cells and collagen production also coincided with the suppression of LOX. Therefore, metformin attenuates the development of myofibroblast-like cells that constitute a pool of collagen-producing cells in fibrotic adipose tissue.

To further characterize the role of AMPK in modulating ECM homeostasis, primary adipose tissue SVF cells were transduced with adenoviruses expressing control green fluorescent protein or constitutively active AMPK (CA-AMPK), a truncated mutation of AMPKα1 (α1 [1–312] T172D) that retains significant kinase activity without requiring its interaction with β- and γ-subunits (7). Overexpression of CA-AMPK enhanced AMPK activity, as described previously (6,7,10). Like the metformin action (Fig. 6), overexpression of CA-AMPK effectively suppresses phosphorylation of Smad3 and subsequently mitigates the transcription of fibrogenic genes in response to TGF-β1 (Fig. 7A–H).

The results described above suggest that activation of AMPK, either through pharmacological or genetic means, exerts an antifibrotic effect. To rigorously illustrate the requirement of AMPK for metformin actions, an adenovirus expressing dominant-negative AMPK (DN-AMPK) (6,7,10,34) was transduced into primary SVF cells. As shown in Fig. 7I–N, metformin and AICAR prevented TGF-β1-induced overproduction of collagens; their effects were largely abolished by DN-AMPK. The inhibitory effect of metformin and AICAR on MMP-9 and TIMP-1 was almost completely diminished by DN-AMPK, highlighting AMPK as a negative regulator of fibrogenesis. The data demonstrate that metformin limits excess ECM accumulation in response to TGF-β1 at least partially through AMPK.

Because adipose tissue TGF-β1 is implicated in the pathogenesis of obesity-induced metabolic dysfunction in humans (39), we hypothesized that adipocytes might be involved in this pathological process. To this end, differentiated 3T3L1 adipocytes were treated with TGF-β1 in the absence and presence of AMPK activators. As shown in Fig. 8, metformin largely decreased TGF-β1-stimulated phosphorylation of Smad3. Metformin-treated adipocytes were also resistant to TGF-β1-induced apoptosis, as evidenced by reduced cleaved caspase-3. Like SVF cells (Fig. 7), CA-AMPK also blocked induction of fibrogenic genes caused by TGF-β1 in 3T3-L1 adipocytes, suggesting the effect of AMPK on fibrogenesis occurs in different cells of adipose tissue. Furthermore, TGF-β1 suppressed insulin-induced phosphorylation of Akt in adipocytes, suggesting a causal relationship between the reduction of WAT fibrosis and insulin resistance in response to metformin. Conversely, metformin or AICAR had similar effects to attenuate TGF-β1-induced insulin resistance. As an independent assessment of insulin sensitivity, insulin-mediated suppression of basal- or isoproterenol-stimulated lipolysis was measured by assessing glycerol release from 3T3-L1 adipocytes. TGF-β1 largely abolished insulin responsiveness, representing a state of insulin resistance in 3T3-L1 adipocytes. Cotreatment with AMPK activators restored the ability of insulin to inhibit lipolysis in adipocytes exposed to TGF-β1. Additionally, the ability of metformin to enhance insulin action was recapitulated by CA-AMPK. Impaired phosphorylation of Akt in eWAT of ob/ob mice was consistently substantially counteracted by metformin (Fig. 4). Thus AMPK seems to inhibit TGF-β1 activity and enhance insulin sensitivity in adipocytes.

This study provides the first biochemical evidence that metformin suppresses the TGF-β1-induced fibrogenic response at least partially through AMPK in vitro, which may explain the beneficial effects of metformin on fat fibrosis and systemic insulin resistance in obesity in vivo (Fig. 8K). In visceral WAT of obese humans, suppression of AMPK and activation of TGF-β1/Smad3 contribute to persistent and aberrant ECM remodeling, characterized by excess collagen synthesis and deposition, LOX-related collagen cross-linking, SMA-positive myofibroblast-like cell activation, adipocyte death, and macrophage recruitment. In mice with genetic and diet-induced obesity, metformin stimulates AMPK, reduces TGF-β1 induction and periadipocyte fibrosis, and improves local and systemic insulin resistance. Pharmacological and molecular activation of AMPK effectively inhibit TGF-β1-dependent fibrogenesis and enhance insulin sensitivity in vitro.

The most important implication of this study is the characterization of a previously unrecognized role of AMPK on omental adipose fibrogenesis in obese humans. Pathological changes include insoluble fibrotic matrix of collagen fibers with activation of LOX-mediated collagen cross-linking and an increase in SMA-positive cells and apoptotic adipocytes close to the fibrotic bundle. Studies by Xu et al. (13) also indicate that defective AMPK in visceral adipose tissue correlates with insulin resistance in morbidly obese patients. Increased fat mass and collagen deposition, as assessed by trichrome staining, hydroxyproline content, and fibrogenic gene expression, are likely attributable to AMPK inhibition in ob/ob mice, as AMPKα1- and AMPKα2-deficient mice exhibit adiposity and metabolic dysfunction in their adipose tissues (43). Expounding on previous observations that elevated plasma TGF-β1 correlates with adiposity in obese patients (39), this study reveals that TGF-β1 hyperactivation may contribute to the initiation of ECM remodeling in WAT of obese mice and humans. It is worth noting that ob/ob mice develop less fibrosis in response to chronic toxic liver injury than wild-type mice (44). Further research is required to understand the tissue-specific role of leptin in regulating ECM homeostasis. Since TGF-β1-positive staining was evident in both macrophage-infiltrated areas and non-macrophage–rich areas in WAT of ob/ob mice, we cannot exclude a secondary effect of fat inflammation on fat fibrosis, However, our in vitro studies delineate that TGF-β1 acts to induce adipocyte insulin resistance and stimulate myofibroblast-like cells, a major source of excessive ECM production, which is widely shared by the pathogenesis of liver fibrosis (29,30). These studies support the proposed modeling: the dysregulation of AMPK and TGF-β1 plays a causal role in promoting pathologically uncontrolled adipose tissue expansion and insulin resistance. Future studies are required to elucidate the precise role of each SVF cell type in fibrotic fat tissue.

While metformin has been the mainstay of antidiabetic therapy for over 50 years (4,5), its underlying mechanisms are not well defined. One major finding of this study is the identification of a novel function of metformin—a switch of visceral fat to subcutaneous fat properties—that is, decreased fibrogenic and inflammatory responses—which mirrors antifibrotic phenotypes in ob/ob mice lacking Col6a3 (2). Fibrogenic cell phenotypes (e.g., TGF-β1/Smad3 stimulation, LOX induction, α-SMA expression, and collagen induction in obese fat) are reduced by metformin. Since DN-AMPK abolishes the antifibrotic effect of metformin in SVF cells, metformin represses TGF-β1-induced fibrogenesis in vitro via an AMPK-dependent mechanism. Decreased fat fibrosis and lowered fasting blood glucose concentrations are highly correlated in metformin-treated mice. Metabolic improvements in metformin-treated mice may be a secondary consequence of reduced fat mass and inflammation. Fibrosis also negatively affects human adipocyte function via mechanosensitive molecules involving connective tissue growth factor and the mechanosensitive transcriptional complex YAP/TEAD (42). Further studies will investigate whether connective tissue growth factor and YAP are involved in the role of metformin in limiting fat fibrogenesis in human obesity. Therapeutically, activating AMPK by metformin may ameliorate pericellular fibrosis of WAT, systemic metabolic dysregulation, and insulin resistance. Future studies will explore whether AMPK-targeting agents such as metformin can be used to design new clinical trials to improve the health of patients with obesity or other fibrotic diseases.

In conclusion, the convergence of AMPK inhibition and TGF-β1/Smad3 activation may contribute to the development of WAT fibrosis and insulin resistance, leading to the unfavorable metabolic microenvironment in obese humans. Targeting AMPK represents a potential therapeutic approach to combat adipose tissue fibrosis and metabolic dysfunction associated with obesity.

Acknowledgments. The authors are grateful to Dr. Susan K. Fried at Boston Nutrition Obesity Research Center (BNORC P30DK046200) for important advice in adipocyte work. The authors also thank Dr. Richard A. Cohen at Boston University School of Medicine for insightful discussion, and Jinyan Zang at Harvard University for editorial assistance.

Funding. This work was supported by the National Institutes of Health (grant nos. RO1 DK076942, R01 DK100603, and R21AA021181 [to M.Z.]) and the American Diabetes Association Award 1-15-BS-216 (to M.Z.). M.Z. also received a Boston University Clinical and Translational Science Award (1UL1TR001430).

Duality of Interest. No potential conflicts of interest relevant to this article were reported.

Author Contributions. T.L., A.N., and M.Z. designed the experiments and researched data. T.L., J.F., B.J., X.J.X., J.H., Y.Y., and Q.Y. contributed to the discussion and reviewed the manuscript. A.N., J.F., A.S., X.R., B.J., and X.J.X. researched data and provided technique assistance. B.J. and Q.L. provided reagents and materials. M.Z. obtained the funding and wrote the manuscript. M.Z. is the guarantor of this work and, as such, had full access to all the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis.