BMP4 and BMP Antagonists Regulate Human White and Beige Adipogenesis

Obesity and its associated negative metabolic and health consequences are globally increasing at an epidemic rate. The subcutaneous adipose tissue (SAT) is the largest adipose depot of the body, and it is also the major sink for excess fat storage. However, SAT has a limited expandability; when the limit is exceeded, fat accumulates in different ectopic sites, including the liver and visceral adipose tissue, and this is the major driver of the metabolic consequences of obesity (1). Studies have also shown that individuals with type 2 diabetes (T2D), compared with nondiabetic subjects, have increased amount of ectopic fat for the same amount of total body fat, supporting a reduced ability to expand the subcutaneous depot (2). Consequently, ability to store excess fat in the subcutaneous depot has a protective effect against the obesity-associated complications, a concept supported by the ability of peroxisome proliferator–activated receptor (PPAR) γ ligands to reduce ectopic fat while subcutaneous body fat is increased (3). However, these ligands can only enhance the differentiation of already committed preadipocytes and cannot promote the commitment and recruitment of new adipose cells.

SAT contains a pool of preadipocytes and other precursor/stem cells that can differentiate into mature adipocytes (4). Regulation of adipogenesis is an important question, since adipose cell expansion (hypertrophic adipocytes) is associated with a dysfunctional adipose tissue with local and systemic insulin resistance and both in vivo and ex vivo studies have shown that hypertrophic obesity is characterized by a reduced recruitment of new cells (4–9). Importantly, we have recently shown that adipose cell size in the abdominal SAT is considerably larger in individuals with a genetic predisposition (defined as being a first-degree relative) for T2D than in matched individuals lacking a known heredity for diabetes (5,8). This novel finding links hypertrophic obesity, ectopic fat accumulation, and associated insulin resistance with genetic risk for T2D, a concept that has received strong recent support from large clinical studies (10,11).

The adipose tissue mesenchymal stem cells serve as a reservoir and allow a continued renewal of precursor cells that can differentiate into adipocytes (12–14). The bone morphogenetic proteins (BMPs) are of particular interest, since some members have been shown to recruit adipose precursor cells into the adipose lineage (6). BMP7 is reported to be a regulator of brown adipogenesis (15), while BMP2 and -4 are related to white adipogenesis (6,12,16). We recently demonstrated that human SAT preadipocytes induce BMP4 during differentiation and that BMP4 increased both commitment and differentiation of human precursor cells (6). In addition, recent studies with human precursor cells found BMP4 also to promote the induction of a beige phenotype (17).

Adipogenic commitment of early precursor cells by BMP4 is mediated by the dissociation of an intracellular complex consisting of the PPARγ transcriptional activator zinc finger protein-423 (ZNF423) (18) and the mesenchymal cell canonical WNT1-inducible signaling pathway protein-2 (WISP2), thereby allowing nuclear entry of ZNF423 and PPARγ induction (19). Thus, BMP4 signaling and its cross-talk with canonical WNT/WISP2 is an essential component of the early induction of adipogenesis. Consequently, inability to adequately increase BMP4 in precursor cells would decrease adipogenesis and, instead, promote enlargement of available cells, i.e., hypertrophic obesity similar to what was found when the commitment factor, early B cell factor (Ebf)1, was genetically deleted in mice (20).

The amount of BMP available for signaling is tightly regulated by the complex BMP receptor signaling pathways including a number of structurally distinct BMP antagonists that alter the ability of BMPs to bind to their receptors and regulate development of many different cell types (21,22). Very little is currently known about the endogenous BMP antagonists and their effects on BMP4 action in human adipogenesis and hypertrophic obesity.

Activin A, a secreted homodimer of inhibin-βA (INHBA) subunits (23,24), follistatin, (25) and the pseudoreceptor BMP and activin membrane-bound inhibitor (BAMBI) (26) are all expressed in the adipose tissue. Follistatin is secreted by adipose tissue explants (27), and noggin is a well-established and secreted inhibitor that binds to the BMP receptors, but its biological functions are mostly undetermined (6,28). Chordin and chordin-like-1 (CHRDL1) are secreted proteins that bind BMP2, -4, and -7 (29,30), and CHRDL1 and has been shown to enhance the proliferation of human mesenchymal stem cells (31). Gremlin-1 (GREM1) is a potent extra- and intracellular inhibitor of BMP4 (32) and is involved in fibrosis and arthritis development (33).

Brown adipose tissue (BAT) is specialized for energy expenditure and maintaining body temperature. Until now, the physiological significance of BAT for whole-body metabolism in adult man has been unclear, but cold exposure in men increased resting energy expenditure, glucose oxidation, and insulin sensitivity (34). An intermediate kind of brown adipocytes, the beige cells, has been demonstrated in SAT (35,36). Increasing the activation of beige cells in mice was associated with reduced weight gain and improved glucose tolerance (37).

In the current study, we examined the effects of BMP4 and several BMP inhibitors during adipogenic differentiation of human subcutaneous preadipocytes. Our results provide evidence for the concept that hypertrophic obesity is a condition of preadipocyte resistance to BMP4 as a consequence of increased secretion of GREM1. In addition, GREM1 regulates the ability of BMP4 to induce beige/brown adipogenesis, making it an interesting target in obesity.

Genes/proteins were studied in isolated mature adipose cells and adipose tissue needle biopsies of the SAT from 33 individuals. The subjects were between 26 and 52 years old, with mean BMI 24.4 ± 2.3 kg/m² (range 19.5–27.5) and adipose cell size 92.8 ± 9.7 µm (range 71.5–118.4). Additional SAT was obtained from 24 individuals by needle biopsy (n = 23) or bariatric surgery (n = 1) for the adipogenic studies. The subjects were between 27 and 66 years of age and had a mean BMI of 27.5 ± 7.1 kg/m² (range 19.3–54.8) and adipose cell size 95.9 ± 17.0 µm (range 52.8–122.5). All subjects had normal glucose levels and had no known chronic diseases. The ethics committee of the University of Gothenburg approved the study design, and written informed consent was received from participants prior to inclusion in the study.

SAT was digested with collagenase and cell size measured as previously described (6,7). The remaining stromal-vascular progenitor cells were isolated and differentiated as previously described (6). When indicated, 100 ng/mL (3 nmol/L) BMP4 was added.

Details of real-time PCR assays were described previously (6). Gene-specific primers and probes were designed using Primer Express software or purchased as Assay-on-Demand (Life Technologies, Stockholm, Sweden).

For overexpression of CHRDL1, cells were transfected with an myc-DDK–tagged ORF clone of CHRDL1 obtained from Origene (TrueORF Gold RC202635; BioNordika Sweden) using Lipofectamine 2000 (Life Technologies) according to the manufacturers’ protocol. An empty vector was used as negative control and green fluorescent protein expression to monitor transfection efficiency. Adipogenic differentiation was induced 24 h after transfection.

Human isolated preadipocytes were transfected with CHRDL1, noggin, or GREM1 small interfering RNA (siRNA) using RNAiMAX (Life Technologies) according to the manufacturer’s instructions. After 48 h, medium was changed to adipocyte differentiation media with and without BMP4 (10–100 ng/mL) as stated. RNA was extracted with an EZNA total RNA kit (VWR, Stockholm Sweden) after 6 days of culture.

Human adipose tissue stromal cells were grown on glass slides, fixed with 4% formaldehyde for 15 min, and permeabilized in 0.1% Triton X-100 for 5 min. Cells were then blocked with 20% FBS for 30 min followed by incubation with anti-BMP4 antibody (MAB1049; Merck Millipore, Solna, Sweden) and UCP1 (ab10983; Abcam, Cambridge, U.K.) for 3h. After washing in PBS and incubation with secondary antibody conjugated with Alexa-594 for 1 h, confocal images were collected by the Leica SP5 confocal system. Cells were also stained with MitoTracker Red according to the manufacturer’s instructions (Life Technologies).

Protein lysates and immunoblotting were performed as previously described (38) using the following antibodies: SMAD1/5/8 (sc-6031 Santa Cruz; AH Diagnostics, Solna, Sweden), pSMAD1/5/8 (Cell Signaling 9511; BioNordika, Stockholm, Sweden), pSMAD3 (Cell Signaling 8769), AKT (Cell Signaling 9272), SMAD3 (sc-8332 Santa Cruz), BMP4 (ab93939), and BMP7 (sc-53917).

AlphaLISA was used to measure secreted human GREM1 (MedImmune, Gaithersburg, MD), and secreted BMP4 was detected with an ELISA from Abcam (ab99982) according to the manufacturer’s instructions.

The experimental data are presented as means ± SEM. Data analyzes were carried out with the PASWstatistics (SPSS, Inc.) for Macintosh. For cells differentiated in vitro, Student paired t test was used for comparison of gene expression with basal samples within groups, and one-way ANOVA (post hoc Tukey test) was used for direct comparisons between groups, correcting for multiple variables where applicable. Differences were considered statistically significant at P < 0.05 level.

To address the potential role of BMP4 in hypertrophic obesity, we first asked whether BMP4 is expressed in mature adipose cells and, if so, whether it is reduced in enlarged cells. However, BMP4 transcript levels were high in isolated mature adipocytes and correlated positively with the cell size (Fig. 1A), albeit not with BMI over this limited range (19.5–27.5 kg/m²) (Fig. 1B).

We have previously shown that BMP4 mRNA levels are increased during human preadipocyte differentiation (6). To ensure that BMP4 protein also is present in adipose cells, we examined differentiated and undifferentiated preadipocytes with both immunofluorescence and Western blot. BMP4 protein was clearly induced after differentiation (Fig. 1C). Furthermore, intermediate– and low–molecular weight secreted BMP4 protein in adipose tissue biopsies was also positively correlated with cell size (Fig. 1E). To validate that BMP4 also was secreted from differentiated human adipocytes, we collected cell culture medium from day 0 and days 0–3 and 9–12 from preadipocytes undergoing differentiation and measured BMP4 protein secretion with ELISA. BMP4 protein was also secreted by differentiated (pre)adipocytes (Fig. 1F).

A possible explanation for the unexpected finding of increased BMP4 in hypertrophic obesity could be that the endogenous BMP inhibitors also are increased, thereby inducing a cellular resistance to BMP4 and its proadipogenic effect. All measured BMP4 inhibitors were robustly expressed in isolated mature adipose cells as well as in undifferentiated preadipocytes and adipocytes differentiated in vitro (Supplementary Table 1). Furthermore, transcript levels of GREM1 and CHRDL1 correlated positively with those of BMP4 in both mature adipocytes and intact adipose tissue biopsies (Supplementary Table 2), supporting the concept of a BMP4 resistance. In addition, GREM1 and CHRDL1 mRNA levels in the human adipose tissue biopsies correlated positively and significantly (P = 0.02 and P = 0.003, respectively) with adipose cell size of the donors, while there was no significant correlation for noggin or follistatin (data not shown). These results show that both BMP4 and certain BMP4 inhibitors are increased in adipose tissue characterized by expanded adipose cells, supporting the possibility that hypertrophic obesity is a condition of cellular BMP4 resistance.

Of note, CHRDL1 is the most highly expressed BMP inhibitor in differentiated (pre)adipocytes and, surprisingly, the highest expression is seen in mature adipose cells (Supplementary Table 1). Human tissue expression pattern also shows CHRDL1 to be most highly expressed in the SAT (Supplementary Fig. 1A) and to be higher in obese than in lean individuals in this depot (Supplementary Fig. 1B).

To characterize the potential role of the BMP4 inhibitors, we examined their expression after differentiation of subcutaneous adipogenic precursor cells. BMP4 increased as expected, and all but two of the inhibitors (CHRDL1 and noggin) were reduced in differentiated cells (Fig. 2A). We also analyzed BMP2 and BMP7 after differentiation. However, BMP2 transcript levels decreased (Fig. 2A), while BMP7 was generally not expressed in human preadipocytes or differentiated adipose cells and we could not detect BMP7 protein in Western blots (data not shown). Thus, these were not further studied.

CHRDL1 increased gradually during differentiation, while noggin showed a transient early increase followed by inhibition (Fig. 2B and C). The expression of these inhibitors was accentuated by the presence of BMP4, probably due to the enhanced differentiation induced by BMP4 (Fig. 2B and C). We characterized the time course for these inhibitors in relation to PPARγ transcript levels, and both the increase in CHRDL1 and the decrease in GREM1 followed a time course similar to that of PPARγ induction (Fig. 2D). The other BMP inhibitors (CHRD, FST, and BAMBI) only showed minor differences in mRNA levels after preadipocyte differentiation, suggesting that they played a less important role for the apparent BMP4 resistance (Supplementary Table 1).

To validate the potential importance of CHRDL1 and noggin in regulating human white adipogenesis, we examined their expression after 9 days in preadipocytes undergoing poor or very good adipogenic differentiation. We also examined factors in the differentiation cocktail, which affected their expression. Table 1 shows that induction of CHRDL1 was positively related to that of PPARγ as well as ability of the cells to undergo differentiation. This is consistent with an overall positive correlation between CHRDL1 and PPARγ expression in fully differentiated (pre)adipose cells (Fig. 2E). Thus, CHRDL1 is a good marker of adipogenic differentiation consistent with the very high expression in mature adipose cells (Supplementary Table 1).

In contrast to CHRDL1, there was no clear difference in noggin expression. This finding, together with the lack of correlation with cell size/BMI, suggests that noggin is an unlikely contributor to the preadipocyte BMP4 resistance in hypertrophic obesity. In contrast, the ability to inhibit GREM1 after differentiation was positively associated with PPARγ transcript activation, suggesting that GREM1 could be an important endogenous regulator of BMP4 resistance and adipogenesis (Fig. 2F).

CHRDL1 was silenced by ∼90% with siRNA (Fig. 3A); however, this reduction did not increase, but markedly reduced, transcriptional activation of PPARγ, adiponectin, and FABP4 during differentiation (Fig. 3B). Furthermore, addition of BMP4 did not rescue the inhibitory effect of small interference CHRDL1 (siCHRDL1) on adipogenic differentiation (Fig. 3B).

Interestingly, cells transfected with CHRDL1 siRNA showed increased ACTA2 (αSMA), a marker of the myofibroblast phenotype, as well as INHBA mRNA levels (Fig. 3C), suggesting that CHRDL1 may cross-talk with other signaling pathways, possibly transforming growth factor (TGF)β, which is known to inhibit adipogenesis and to increase these markers of fibrosis (23,39–41). Thus, CHRDL1 is not an endogenous inhibitor of BMP4 in human preadipocytes, and we also saw no inhibitory or positive effect on pSMAD1/5/8 activation by BMP4 after CHRDL1 overexpression in human preadipocytes (Fig. 3D and E).

Consistent with the concept that CHRDL1 cross-talks with other signaling pathways, we did not find that overexpressing CHRDL1 in preadipocytes directly enhanced differentiation (Fig. 3F); instead, it decreased ACTA2, CTGF, and INHBA, further supporting potential cross-talk with TGFβ (Fig. 3G). We also examined whether silencing CHRDL1 altered initial upstream signaling of TGFβ, measured as pSMAD3 increase by TGFβ after 120 min, but saw no such direct upstream effect (data not shown).

GREM1 mRNA levels in whole SAT biopsies were positively correlated with adipose cell size of the donors (Fig. 4A), and ability to inhibit GREM1 during differentiation of preadipocytes was also markedly reduced in hypertrophic obesity and correlated with the cell size of the donors (Fig. 4B). This is consistent with the negative correlation between GREM1 and PPARγ transcriptional activation seen in differentiated cells (Fig. 2F).

Since GREM1 is a secreted BMP inhibitor, we measured GREM1 in the culture medium of human preadipocytes undergoing differentiation. Secretion of GREM1 correlated positively with GREM1 mRNA levels (Fig. 4C), and it remained essentially stable after day 6 of differentiation (Fig. 4D). This is in accordance with the rapid downregulation of GREM1 during initiation of differentiation and the partial rebound effect seen at later time points (Fig. 4E). Furthermore, GREM1 secretion by differentiated preadipocytes correlated positively with BMI of the donors and also tended to correlate with cell size (Fig. 4F and G). We also verified that GREM1 protein is a bona fide inhibitor of BMP4 (as well as BMP7 [data not shown]) signaling and pSMAD1/5/8 activation (Fig. 4H and I).

We also analyzed the expression of GREM1 in different human tissues. It is highly expressed in the adipose tissue and has a higher expression in omental tissue than in SAT (Supplementary Fig. 2).

We then examined the effect of silencing GREM1 (>90% inhibition [Fig. 3A]) in undifferentiated preadipocytes and found PPARγ induction to be significantly (P < 0.05) increased to an extent similar to that seen in control cells incubated with BMP4, and the effect of pioglitazone was also significantly higher (P < 0.02) (Fig. 5A). Similarly, the PPARγ transcriptional activator ZNF423 (42) was significantly increased (Fig. 5B), and the effect of both BMP4 and pioglitazone was also higher in siGREM1 cells (Fig. 5B). These effects of silencing GREM1 indicate that it is an important regulator of the proadipogenic effect of endogenous BMP4 in human preadipocytes.

BMP4 has also recently been shown to enhance beige adipogenesis in human precursor cells (17). We also found addition of BMP4 to increase the beige adipose marker TMEM26 in the preadipocytes (P < 0.05), and this effect was markedly increased in siGREM1 cells, while adding the PPARγ ligand pioglitazone alone had no effect on TMEM26 under any of these conditions (Fig. 5C). Interestingly, and as also previously noted (17), the beige adipose cell marker TBX1 was expressed at low levels, but it increased significantly after addition of BMP4 (data not shown). Nevertheless, the expression of other markers of beige cells, TMEM26 and CD137 (TNFRSF9), was closely correlated in human preadipocytes (Fig. 5D).

The brown adipose cell marker ZIC1, but not PRDM16 (data not shown), was clearly increased in siGREM1 cells (P < 0.02), and this was further enhanced by BMP4 (Fig. 5E). Unexpectedly, we also found BMP8B, a regulator of thermogenesis and BAT activation in mice (43), to be robustly expressed in human SAT as well as in the preadipocytes but only slightly increased by the addition of BMP4 and/or by silencing GREM1 (data not shown). However, BMP8B expression correlated closely with that of TMEM26, suggesting that it is a marker of beige adipogenesis in human adipose tissue (Fig. 5F). The mRNA levels of UCP1 in control human preadipocytes were low, but the expression was increased and became measurable in most siGREM1 cells whether or not cAMP or BMP4 was present (data not shown). Importantly, silencing GREM1 in human preadipocytes increased the mitochondrial content as measured by the MitoTracker Red, and this was further increased by BMP4 (Fig. 5G and H). Furthermore, induction of UCP1 was also seen under these conditions (Fig. 5G and H). PGC1α was also increased by both silencing GREM1 and adding BMP4 and further markedly increased by adding pioglitazone (Fig. 5I).

Taken together, GREM1 is an attractive target for overcoming the BMP4 resistance in hypertrophic obesity, since silencing GREM1 enhances endogenous BMP4 signaling and action and increases ZNF423 and PPARγ induction as well as expression of markers of an oxidative beige/brown adipose cell phenotype.

We also examined the beige/brown adipose cell markers CD137, TMEM26, and ZIC1 after induction of white adipose cell differentiation of the preadipocytes. All beige markers were dramatically reduced, suggesting the presence of multipotent precursor cells, which could undergo either beige/brown or white differentiation depending on the ambient signals (Fig. 5J). Similarly, BMP8B was also reduced after induction of white_–adipogenesis (Fig. 5J). However, silencing GREM1 before white adipogenic differentiation reactivated TMEM26 (Fig. 5K) and ZIC1 (data not shown), suggesting that GREM1 and/or the enhanced BMP4 signaling exerted a particularly prominent effect in promoting the beige/brown phenotype of human preadipocytes. Ongoing studies are aimed at clarifying detailed molecular mechanisms for this.

Hypertrophic obesity is associated with a dysregulated adipose tissue, inflammation, and local and systemic insulin resistance (4,5,8,9), and the degree of insulin resistance is positively correlated with adipose cell size (16) as well as future risk of developing T2D (44). Thus, understanding adipose precursor cell recruitment and differentiation in the large SAT can open new possibilities for preventing ectopic fat accumulation and the metabolic complications of obesity. Furthermore, recent animal data have shown that cells in SAT have the greatest plasticity in terms of inducing beige/brown adipose cells and that these cells play an important role in total body energy expenditure and weight gain (45). BMP4 regulates adipose precursor cell commitment into the white adipose lineage (6,13,46) and is induced in human preadipocytes undergoing differentiation (16), overexpression in the adipose tissue in mice leads to an increased beige/brown phenotype in SAT (46), and BMP4 can also activate beige adipose cell development in human precursor cells (17).

Since hypertrophic obesity is associated with an impaired subcutaneous adipogenesis (16), we postulated that this could be due to reduced BMP4 in the precursor cells. However, we found that cellular BMP4 transcript and protein levels are increased in hypertrophic obesity, and BMP4 is secreted by the adipose cells, supporting a functional feedback regulation promoting the recruitment of new adipose cells when needed rather than just expanding existing cells. These results suggest that the precursor cells are resistant to secreted BMP4, possibly due to increased activity of the endogenous BMP inhibitors.

The time course for the induction of the BMP inhibitors during preadipocyte differentiation showed that all but noggin and CHRDL1 were markedly decreased. CHRDL1 increased during differentiation and was a positive marker of PPARγ induction and adipogenesis, and in fact, it is not an inhibitor of BMP4 in human preadipocytes. Expressing CHRDL1 in the undifferentiated preadipocytes had no direct effect on PPARγ but reduced several markers of fibrosis and TGFβ activation. Silencing CHRDL1 reduced PPARγ and increased ACTA2 (αSMA), CTGF, and INHBA as markers of a myofibroblast phenotype. These findings indicate that CHRDL1 cross-talks with, and inhibits, the profibrotic TGFβ pathway, which is consistent with the finding that genetic deletion of CHRDL1 in mice enhances renal fibrosis (31). Interestingly, CHRDL1 has been found in the human adipose tissue secretome (47), indicating that it may be a circulating protein, but nothing is known about potential endocrine effects.

Noggin is a well-established inhibitor of BMP4, and as expected, antagonizing noggin enhances endogenous BMP4-stimulated preadipocyte differentiation to both the white (6) and beige (data not shown) phenotype. However, its temporal expression, lack of association with adipose cell size, and precursor cells differentiation capacity suggest that it is not specifically related to the reduced adipogenesis or BMP4 resistance in hypertrophic obesity.

In contrast, GREM1 is a secreted and powerful inhibitor of BMP4 and BMP7 signaling as well as of adipogenic differentiation. The time course for GREM1 is consistent with an important regulatory effect on adipogenesis. It is secreted by human (pre)adipocytes, secreted protein and mRNA levels were positively correlated, and transcript levels were closely correlated with adipose cell size. Silencing GREM1 also increased the effect of BMP4 on PPARγ induction. Thus, GREM1 is an important regulator of white adipogenesis and the ability of BMP4 to induce commitment and differentiation.

Interestingly, GREM1 also regulates BMP4-induced beige/brown adipogenesis as well as the effect of BMP4 on markers of mitochondrial content and PGC1α induction. Unexpectedly, we also found BMP4 and GREM1 to be involved in the regulation of BMP8B. BMP8B activates thermogenesis in BAT in mice, but it is not expressed in white adipose tissue (43). However, BMP8B is robustly expressed and closely associated with TMEM26, suggesting that it is a marker of beige adipose cells in human SAT.

Induction of white adipogenesis markedly reduced TMEM26, CD137, BMP8B, and ZIC1 in the human cells, suggesting the presence of multipotent precursor cells that could enter both white and beige/brown adipogenesis. This concept is similar to the proposal by Lee et al. (48) in murine SAT, and lineage-tracing experiments have shown that beige adipose cells can revert to white cells (49). Importantly, we also found that silencing GREM1 in partly differentiated white adipocytes reactivated TMEM26 and ZIC1, suggesting a transdifferentiation potential and that GREM1/BMP4 strongly promotes a beige/brown phenotype of already committed and differentiating preadipocytes. Our findings are schematically illustrated in Fig. 6.

Taken together, our results show that BMP4 is an important endogenous regulator of human white and beige adipogenesis. However, the powerful BMP4/7 inhibitor GREM1, which is specifically increased in hypertrophic obesity, antagonizes these effects and is an interesting target in human obesity/T2D.

Funding. The authors thank the Swedish Research Council, the Swedish Diabetes Association, the Novo Nordisk Foundationhttp://dx.doi.org/10.13039/501100004191, the Torsten Söderbergs Foundation, and the West Sweden ALF program for financial support.

Duality of Interest. J.G. and C.R are employees of MedImmune, LLC. Financial support was also received from MedImmune (MA-414379) as a research agreement with the University of Gothenburg. No other potential conflicts of interest relevant to this article were reported.

Author Contributions. B.G. performed experiments, analyzed data, and wrote the manuscript. A.H., S.H., and J.M.H. performed experiments. P.-A.S. contributed with data in human tissues. J.G. and C.R. contributed with discussion and reviewed the manuscript. U.S. designed the experiments, did the data analysis and interpretation, and wrote the manuscript. U.S. 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.

Prior Presentation. Parts of this study were presented in abstract form at the 74th Scientific Sessions of the American Diabetes Association, San Francisco, CA, 13–17 June 2014.