SMAD3/Stat3 Signaling Mediates β-Cell Epithelial-Mesenchymal Transition in Chronic Pancreatitis–Related Diabetes

The prevalence of chronic pancreatitis is roughly 50 per 100,000 people worldwide (1). Chronic pancreatitis in the United States results in more than 122,000 outpatient visits and more than 56,000 hospitalizations each year (2). Many patients with chronic pancreatitis develop insulinopenia, glucose intolerance, insulin resistance, and eventually diabetes (2), largely as a result of the intimate proximity of the endocrine pancreas to the exocrine pancreas (3). Moreover, patients with chronic pancreatitis often develop a fibrotic pancreas with a reduced β-cell mass and have a 15- to 16-fold increased risk for pancreatic cancer (4). To date, the understanding of the development and pathogenesis of chronic pancreatitis–related diabetes (CPRD) is very limited. In addition, the mechanisms of β-cell loss in CPRD may in part be similar to those in type 2 diabetes (T2D) (5,6) and in cystic fibrosis (7). Thus, elucidation of the underlying mechanisms common to chronic pancreatitis, CPRD, and pancreatic cancer is critical.

Among animal models for acute and chronic pancreatitis (8), partial pancreatic duct ligation (PDL) has been used to generate chronic pancreatitis in mammals (9,10). Ligation of the pancreatic duct immediately at the beginning of the splenic or tail part of the pancreas blocks the drainage of ductal fluid from the distal pancreas, resulting in autodigestion of acinar cells and severe inflammation specifically in the ligated tail of the pancreas, although initially the islets are largely unaffected. Acinar cell death in the tail of the pancreas leads to the complete loss of acinar cells, without significant acinar cell regeneration (11). On the contrary, the nonligated part, or head, of the pancreas appears to be normal, thus offering an excellent internal control (12,13). We recently reported an inflammatory molecular signature in PDL, which induced β-cell proliferation in a transforming growth factor β (TGFβ) receptor signaling–dependent manner (12–15). As a strong stimulant of epithelial-mesenchymal transition (EMT) in epithelial cells (16–18), TGFβ1 is highly upregulated in the ligated tail of the pancreas after PDL (PDL-tail) (14). Thus we were prompted to evaluate the effects of PDL-induced TGFβ1 on the EMT of β-cells.

Forkhead box protein O1 (FoxO1) is a pivotal factor in orchestrating the response of β-cell mass and function to overnutrition and obesity (19) and to oxidative stress (20–22). FoxO1 is predominantly expressed by β-cells in the adult pancreas. We and others have shown that FoxO1 nuclear translocation increases NeuroD1, MafA, and Nkx6.1 expression in β-cells, contributing to the maintenance of a functional differentiated phenotype to resist stress-induced dedifferentiation, dysfunction, and failure (23–25). Nevertheless, a role for FoxO1 during the pathogenesis of CPRD is unknown.

Here we studied long-term PDL (12–24 weeks) as a model of chronic pancreatitis in humans. We further analyzed the molecular mechanisms underlying the gradual β-cell loss in this model, which mimics CPRD in humans.

All mouse experiments were approved by the Animal Research and Care Committee at the Children’s Hospital of Pittsburgh and the University of Pittsburgh Institutional Animal Care and Use Committee. BAC transgenic rat insulin promoter (RIP) Cre reporter (RIP-Cre) mice, MIP-GFP mice (green fluorescent protein reporter under the control of a mouse insulin promoter), and Rosa26CAG-mTmG (mTmG) mice have been described before (13). These mice and C57BL/6 mice were all purchased from The Jackson Laboratory (Bar Harbor, ME). TGFβ receptor II (TBR2) fx/fx mice were generous gifts from Professor Stefan Karlsson (University of Lund, Sweden) and have been described previously (12). All mice were 10-week-old males and had a C57BL/6 background. PDL was performed and validated as described elsewhere (12–15). Intraductal viral infusion was performed as described previously (26). Adeno-associated virus (AAV) 8 viruses (titration of 10¹² genome copy particles/mL in a 150-μL volume) were delivered via a catheter at a rate of 6 μL/min. BrdU-supplemented drinking water was given to mice 1 week before the analysis, as previously described (12).

Pancreas digestion, islet isolation, and β-cell purification from MIP-GFP mice have been described previously, taking advantage of the specific expression of green fluorescent protein on β-cells (12–14,27–29). Dissociated pancreatic cells were sorted based on α-smooth muscle actin (SMA) positivity after the cells were fixed in 4% formalin and then sequentially incubated with anti-α-SMA (Abcam, Cambridge, MA) and phycoerythrin-conjugated antirabbit antibody (BD Biosciences, San Jose, CA).

Mouse islets were cultured in Ham’s F10 medium (Life Technologies, St. Louis, MO) supplemented with 0.5% BSA (Sigma-Aldrich, St. Louis, MO), 2 mmol/L glutamine, 2 mmol/L calcium, and 5 mmol/L glucose at 37°C in 95% air and 5% CO₂. After overnight culture, islets were either treated or not treated with 20 ng/mL TGFβ1 (with or without SMAD3 and Stat3 inhibitors) and harvested after 0, 0.5, 1, 3, 12, and 48 h. The SMAD3 inhibitor SIS3 (2 µmol/L) or Stat3 inhibitor cryptotanshinone (CTSN; 5 µmol/L) or control DMSO solution was added with the TGFβ1.

Western blotting was performed as previously described (14,30). Primary antibodies for Western blotting are rabbit polyclonal anti-MafA (Abcam), anti-GAPDH, anti-Snail, anti-Slug, anti-Twist, anti-ZEB1, anti-ZEB2, anti-SMAD3, anti-pSMAD3, anti-Stat3, anti-pStat3, anti-NeuroD1, anti-Nkx6.1, anti-Pdx1, and anti–epithelial cadherin (E-cadherin), all from Cell Signaling Technologies, San Jose, CA). The secondary antibody was horseradish peroxidase–conjugated antirabbit (Dako, Carpinteria, CA).

AAV serotype 8 vectors were used to generate AAV-RIP-null and AAV-RIP-FoxO1 viruses, as described before (25).

RNA extraction and quantitative RT-PCR have been described previously (12,13,27). The following primers all were purchased from Qiagen (Valencia, CA): CycloA (QT00247709), FoxO1 (QT00116186), Pdx1 (QT00102235), NeuroD1 (QT00251265), MafA (QT01037638), Nkx6.1 (QT00143318), TGFβ1 (QT00145250), vimentin (QT00159670), α-SMA (QT00140119), F4/80 (QT00099617), collagen I (QT00162204), E-cadherin (QT00121163), Snail (QT00240940), Slug (QT00098273), ZEB1 (QT00105385), ZEB2 (QT00148995), and fibronection (QT00135758). Quantitative RT-PCR values were normalized against CycloA, which proved to be stable across the samples. Fold changes from the control are shown in the figures.

All pancreas samples were fixed and cryoprotected in 30% sucrose overnight before freezing, as described before (12). Masson trichrome staining was performed using a Trichrome Stain (Masson) Kit (Sigma-Aldrich). The fluorescent membrane-targeted Tomato (mT) and membrane-targeted EGFP (mG) were detected by direct fluorescence. Primary antibodies for immunostaining were guinea pig polyclonal anti-insulin (Dako), rabbit polyclonal anti-F4/80 (Invitrogen, CA, Carlsbad), goat polyclonal antiamylase (Santa Cruz Biotechnology, Dallas, TX), rabbit polyclonal anti-FoxO1 (made by H.D.), goat polyclonal anti-E-cadherin (R&D Systems, Los Angeles, CA), rabbit polyclonal anti-α-SMA (Abcam), and rat polyclonal anti-CD45 (BD Biosciences). BrdU staining has been described before (12). Secondary antibodies for indirect fluorescent staining were Cy2, Cy3, or Cy5 conjugated rat-, rabbit- goat-, and guinea pig–specific antibodies (Jackson ImmunoResearch Labs, West Grove, PA). Nuclear staining was performed with Hoechst solution (BD Biosciences). β-Cell mass was quantified as described previously (12).

All values are depicted as means ± standard errors. Five repeats were analyzed in each condition. All data were statistically analyzed using one-way ANOVA with a Bonferroni correction, followed by the Fisher exact test. Data were considered significant when P < 0.05.

Recent studies of the effects of PDL on β-cell mass have primarily focused on the short-term (0–8 weeks). PDL can induce robust pancreatitis, as shown here in immunostaining for the pan-leukocyte marker CD45 (Fig. 1A) and the macrophage-specific marker F4/80 (Fig. 1B). Long-term PDL is a model of chronic pancreatitis. However, the effect of long-term PDL on β-cell mass has not been studied. Fasting blood glucose remained normal (Supplementary Fig. 1), presumably as a result of the head part of the pancreas remaining intact after PDL (PDL-head; Supplementary Fig. 2). Here, with gross imaging, we detected increasing exocrine tissue atrophy and fibrosis in the PDL-tail over the 12 to 24 weeks after PDL (Fig. 1C–F), which was confirmed by fibrosis quantified with Masson trichrome staining (Fig. 1G) and by quantification of transcript levels of fibrotic markers—collagen I, fibronectin, and vimentin—in the PDL-tail compared with the PDL-head (Fig. 1H). These findings are consistent with what has been seen in patients with chronic pancreatitis (2) and suggest that long-term PDL is a reasonable model of human chronic pancreatitis.

When we examined the morphology of the pancreas over time after PDL, we found that islets were easily visualized with the naked eye early after PDL as a result of elimination of acini, as shown grossly in Fig. 1C and D and in previous publications (12,13,31). Surprisingly, however, although acinar cells do not regenerate after PDL, islets were barely visible by 12 weeks after PDL (Fig. 1E) and were not grossly visible at all by 24 weeks after PDL (Fig. 1F). Further, insulin staining and β-cell mass quantification revealed a gradual β-cell loss, with a time-dependent decrease in β-cell mass, all specifically in the PDL-tail (Fig. 2A–K). Thus, β-cell loss occurs over the long-term after PDL.

Recent studies strongly suggest that protection of the differentiated phenotype of existing β-cells is critical to maintain a functional β-cell mass and thus to prevent T2D (23,24). In line with this concept, the β-cell–specific transcription factors Pdx1, NeuroD1, Nkx6.1, and MafA seem to be required for β-cells to be fully functional, whereas their loss correlates with β-cell dysfunction and β-cell loss (23–25). As such, we performed PDL on MIP-GFP mice, and sorted β-cells via FACS at serial time points after PDL to asses Pdx1, NeuroD1, Nkx6.1, and MafA mRNA levels in β-cells by quantitative RT-PCR and Western blotting. We detected a gradual decrease in both mRNA and protein levels for these genes in β-cells (Fig. 2L and Supplementary Fig. 3), suggesting that β-cell dedifferentiation may occur in the tail of the pancreas over the long-term after PDL.

We then analyzed whether dedifferentiation of β-cells followed by β-cell apoptosis may be responsible for the progressive loss of β-cell mass. We performed both cleaved caspase 3 (Fig. 3A) and terminal deoxynucleotidyl TUNEL (Fig. 3B) staining to evaluate β-cell apoptosis. To our surprise, no significant increase in β-cell apoptosis was detected by either method at different time points after PDL (Fig. 3A and B), suggesting that β-cell loss over the long-term after PDL is not primarily attributable to β-cell apoptosis.

We previously showed that large amounts of TGFβ1 were secreted by M2 macrophages in the PDL-tail, which led to SMAD7 expression and TGFβ receptor signaling in β-cells to coordinate a modest increase in β-cell proliferation. These actions occurred in conjunction with simultaneously activated epidermal growth factor receptor signaling (14). TGFβ1 is also a potent and well-defined trigger for EMT, a process by which epithelial cells lose their cell polarity and cell-cell adhesion and gain migratory and invasive properties to become mesenchyme-like cells (16–18). Because we detected high TGFβ1 levels (14) and progressive fibrosis in the PDL-tail (Fig. 1A–G), we hypothesized that TGFβ1 may also trigger EMT of β-cells. Loss of E-cadherin is well established as a fundamental event in EMT, during which the transcription factors ZEB1, ZEB2, Snail, and Slug bind directly to E-box consensus sequences on the promoter region of E-cadherin and repress its expression (16–18). Thus we analyzed expression levels of these EMT-related genes in purified β-cells from the tails of pancreases of MIP-GFP mice after PDL. We found that ZEB1 and Snail protein levels in β-cells increased with time, whereas E-cadherin levels decreased with time. In addition, ZEB2 and Slug mRNA (but not protein) increased with time (Fig. 3C and Supplementary Fig. 3). These results suggest that an EMT process may indeed occur in β-cells over time after PDL.

To further confirm this possibility, we performed a lineage tracing study using our previously published RIP-Cre;mTmG mouse model (13). These mice should express membrane-bound red fluorescence in all cells except insulin-positive β-cells and their progeny, whose floxed mT cassette is deleted, leading to constitutive expression of the mG cassette located just downstream. We noticed that fibroblasts tend to not express the mT well, and therefore many of the cells in a fibrotic pancreas do not fluoresce. In these mice, if any β-cells undergo dedifferentiation to become insulin-negative, they still keep mG expression and appear green. We detected quite a few mG-positive/vimentin-positive/insulin-negative cells 12 weeks after PDL; these cells likely represent β-cells that have undergone EMT (Fig. 3D). These data strongly suggest the occurrence of a β-cell EMT, which may be a critical contributor to β-cell loss over the long-term after PDL.

We previously reported that PDL entails recruitment of M2 macrophages that secrete large amounts of TGFβ1 (14). However, macrophage infiltration in the PDL-tail peaked 1 week after PDL and then subsequently decreased, as shown previously (12–14), and as shown here by F4/80 transcript enrichment in the PDL-tail (Fig. 4A). Because tissue fibrosis and β-cell EMT seem to occur over the long-term after PDL, we hypothesized that either TGFβ1 levels remain high over the long-term after PDL because the TGFβ1 comes from other sources, or non-TGFβ1 EMT triggers may be involved. We analyzed the transcript levels of TGFβ1 in the PDL-tail over time and found that they remained high, peaking ∼4 weeks after PDL, when F4/80 transcript levels were already significantly decreased (Fig. 4A). These data suggest that TGFβ1 may be the trigger of tissue fibrosis and β-cell EMT but that macrophages may not be the major source of TGFβ1 at these later time points after PDL (14). Indeed, previous studies have highlighted a specific cell population called myofibroblasts, also called pancreatic stellate cells (32), which, when activated, produce high levels of TGFβ1 to promote EMT and fibrosis (16–18). Because α-SMA is a specific marker for myofibroblasts, we examined the transcript levels of α-SMA in the PDL-tail and found a significant and sustained increase in α-SMA by 4 weeks after PDL (Fig. 4A); this was further confirmed by immunostaining (Fig. 4B). Moreover, when we sorted α-SMA–negative and α-SMA–positive cells from the PDL-tail, we detected significantly higher levels of TGFβ1 in the α-SMA–negative population than in the α-SMA–positive population 1 week after PDL, but significantly higher levels of TGFβ1 in the α-SMA–positive population than in the α-SMA–negative population 4 weeks after PDL (Fig. 4C). This indicates that activated myofibroblasts may be the major source of TGFβ1 over the long-term after PDL and may play a more important role than macrophages in pancreatic tissue remodeling, fibrosis, and β-cell EMT.

To confirm TGFβ1 as a necessary trigger for β-cell EMT in this model, we performed PDL on β-cell–specific TBR2-null mutant mice (RIP-Cre; TBR2fx/fx) (12,13) (Fig. 5A). While β-cell–specific suppression of TGFβ1-activated TGFβ receptor signaling did not alter pancreas fibrosis in PDL-tails (Fig. 5B), the gradual loss of β-cell mass over the long-term after PDL was significantly reduced (Fig. 5C and D). These data strongly suggest that TGFβ1 is a necessary trigger for β-cell EMT and β-cell loss over the long-term after PDL.

To understand the molecular basis underlying TGFβ1-activated β-cell EMT, we treated cultured primary mouse islets with 20 ng/mL TGFβ1 and harvested the cells at 0, 0.5, 1, 3, 12, 24, and 48 h after TGFβ1 administration. Western blotting was performed to examine the EMT regulatory proteins Twist, Snail, Slug, ZEB1, and ZEB2; among these, Snail was the only one found to be strongly activated (Fig. 6A). Moreover, Snail was activated as early as 1 h after TGFβ1 treatment, and Snail expression seemed to be sustained (Fig. 6A). In addition, Snail mRNA was activated in a pattern similar to that of the protein (Fig. 6B), suggesting that Snail may be activated primarily at the transcription level. Thus we focused on the regulation of Snail by TGFβ signaling. SMAD3 is a key factor in mediating signal transduction of activated TGFβ receptors. Activated TGFβ/SMAD3 has been shown to induce phosphorylation and activation of Stat3 to promote the transcription of Snail (33). TGFβ1-treated mouse islets were cotreated with either a specific SMAD3 phosphorylation inhibitor (SIS3) or a specific Stat3 phosphorylation (Tyr705) inhibitor (CTSN) in order to suppress phosphorylation-associated activation of SMAD3 and Stat3, respectively. Islets were analyzed 12 h after treatment. We found that TGFβ1 induced phosphorylation of SMAD3 and Stat3 and activation of Snail (Fig. 6C–F). Suppression of SMAD3 phosphorylation by SIS3 prevented both TGFβ1-induced phosphorylation of Stat3 and TGFβ1-induced activation of Snail, but suppression of Stat3 phosphorylation by CTSN prevented only TGFβ1-induced activation of Snail, without affecting TGFβ1-induced phosphorylation of SMAD3 (Fig. 6C–F). Thus TGFβ1 may trigger EMT of β-cells through a SMAD3/Stat3 signaling cascade, and SMAD3 phosphorylation occurs upstream of Stat3 phosphorylation, consistent with a previous report (33).

FoxO1 is a transcription factor that plays a key protective role against β-cell failure during stress (23–25). Previous reports have shown that FoxO1 and Stat3 antagonize each other in the regulation of T-cell function (34,35) and leptin signaling transduction (36,37). We were thus prompted to evaluate the role of FoxO1 during the process of β-cell EMT in this CPRD model, because the inflammatory environment in the PDL-tail presumably induces significant local cell stress. We performed immunostaining for FoxO1 and, compared with the PDL-head and the sham-operated control, found enhanced expression and nuclear localization of FoxO1 in β-cells in the PDL-tail 4 weeks after PDL, suggesting that FoxO1 may be activated during β-cell stress after PDL (Fig. 7A). However, FoxO1 was absent from β-cells 12 weeks after PDL (Fig. 7A). To document a protective role for FoxO1 against EMT of β-cells, we forcibly expressed FoxO1 specifically in β-cells using an intraductal viral infusion technique recently developed in our laboratory (26) (Fig. 7B). Here, forced expression of FoxO1 in β-cells by intraductal infusion of AAV carrying recombinant FoxO1 under RIP in mice undergoing PDL prevented FoxO1 loss in islets after PDL (Fig. 7C) and significantly ameliorated β-cell loss over time (Fig. 7D). Moreover, forced expression of FoxO1 by AAV-mediated gene transfer in purified β-cells from tails of pancreases 12 weeks after PDL inhibited activation of EMT genes and increased E-cadherin (Fig. 7E). Together, these data suggest that FoxO1 may prevent β-cell EMT over the long-term after PDL (Fig. 7F).

Chronic pancreatitis is highly prevalent (1,2) and predisposes patients to a high risk of developing diabetes (CPRD) (5). The current treatment of chronic pancreatitis includes surgical removal of part or all of the pancreas, with the potential to isolate and transplant the islets back into the patient. Many patients already have diabetes, however, and therefore are not candidates for this islet salvage. Also, islet isolation from a pancreas with chronic pancreatitis characterized by severe tissue fibrosis is difficult and inefficient. Moreover, transplanted islets are situated in the liver, where they may gradually lose function (1,2). Therefore a better understanding of the molecular mechanisms underlying the development and pathogenesis of CPRD may lead to novel, more effective therapeutic strategies.

The development of CPRD may largely result from the proximity of the endocrine pancreas to the exocrine pancreas. When the exocrine pancreas is severely inflamed and fibrotic, the inflammatory milieu results in the presence of many cytokines and factors that negatively affect the neighboring islet cells in a paracrine way. In addition, the vasculature is typically severely disrupted. These alterations may significantly affect the biology and homeostasis of islet β-cells.

Complete PDL and partial PDL have been used for decades as models for acute and chronic pancreatitis in humans (9,10). Compared with complete PDL, the partial PDL model has several advantages. The nonligated head of the pancreas (PDL-head) is minimally affected and thus can be used as an internal control for the PDL-tail. Also, the islets and acinar cells in the head of the pancreas are adequate to meet physiological needs for glucose homeostasis and secretion of pancreatic digestive enzymes, respectively. The disadvantage of the partial PDL model is that the intact head of the pancreas prevents changes in blood glucose at the late stage of CPRD. Since 2008, PDL over the short-term—traditionally regarded as a model of acute pancreatitis—has been extensively investigated as a model of β-cell regeneration. Although controversy still exists, most of these short-term studies have shown that little or no β-cell neogenesis occurs, but β-cell proliferation increases modestly, which contributes minimally to β-cell mass (13,31,38–40).

Compared with other chronic pancreatitis models (e.g., serial cerulein injections, serial l-arginine injections, alcohol feeding, and genetic modifications), PDL confers several advantages. First, PDL requires only one treatment, which substantially reduces stress to the animals and reduces experimental variability. Second, PDL induces robust pancreatitis through ductal occlusion, which may mimic the pathogenesis of pancreatitis in many humans. Third, acute pancreatitis by PDL progresses to chronic pancreatitis relatively quickly, which not only significantly shortens the experimental period but also allows easier detection of changes in the parameters of interest (9–11,41–43).

In this study we show that long-term PDL becomes a chronic pancreatitis model, and the gradual loss of β-cell mass after PDL mimics the similar phenomenon in patients with CPRD. We found that β-cell dedifferentiation followed by β-EMT occurs in the PDL-tail, resulting in a progressive loss of functional β-cell mass. In the lineage tracing experiment, no vimentin/insulin double-positive cells were ever detected. However, vimentin-positive β-cell progeny cells were broadly detected by 12 weeks after PDL, suggesting widespread β-cell EMT. Of note, EMT β-cells may then undergo cell death, since the fold-increase in EMT-associated genes was not dramatic in purified β-cells at serial time points after PDL.

Using a loss-of-function experiment in which TGFβ receptor signaling was blocked (TBR2-null mutant mice), we confirmed that TGFβ receptor signaling was necessary for β-cell EMT in the PDL-tail. We also showed evidence that TGFβ1 ligand is initially derived predominantly from M2 macrophages soon after PDL, but it comes from activated myofibroblasts in the later stages after PDL. Myofibroblasts can be activated by TGFβ1, but as a result they begin to produce and secrete TGFβ1 themselves, as a positive feedback loop for TGFβ1 production (16–18). Early after PDL, the recruited macrophages may act as activators of myofibroblasts through their production and secretion of TGFβ1 (14). Furthermore, high TGFβ1 has been detected in the pancreas in patients with acute (44) and chronic pancreatitis (45). Thus a critical role for TGFβ1 in β-cell EMT over the long-term after PDL seems a likely mechanism of CPRD pathogenesis. Mechanistically, we showed that TGFβ1 activated TGFβ receptor signaling in β-cells to induce phosphorylation and activation of SMAD3, which subsequently induced activation of Stat3, possibly through competitive binding to protein inhibitor of activated Stat3 (33).

FoxO1 and Stat3 are known to antagonize one another in several cell types, but this has not been studied in β-cells. However, recent studies have demonstrated a critical role for Stat3 in pancreatic cell homeostasis (46,47). Here, early activation of Stat3 overcomes FoxO1 suppression, leading to β-cell dysfunction. Strengthening FoxO1 expression in β-cells suppresses Stat3 signaling and thereby suppresses β-cell EMT. In our mouse model, we further showed that FoxO1 overexpression is sufficient to prevent β-cell EMT, and the late loss of FoxO1 seemed to portend the completion of β-cell EMT.

The relatively delayed onset of β-cell EMT in vivo, compared with in vitro, possibly results from the in vivo environment of the islets that might protect against EMT through cell-cell communication. In summary, we present here a mouse model for CPRD. To the best of our knowledge, this study is the first to show β-cell EMT in an experimental setting of CPRD. Because β-cell EMT seems to underlie the process of β-cell loss, prevention of β-cell EMT may be an effective treatment for CPRD.

Funding. This work was supported by the startup from the Department of Surgery of Children’s Hospital of Pittsburgh (to X.X.), and by the National Institutes of Health/National Institute of Diabetes and Digestive and Kidney Diseases (grant nos. R01DK111460 and R01DK112836 to G.K.G.).

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

Author Contributions. X.X. conceived and designed the study. X.X., S.F., T.Z., C.C., Q.S., R.Z., S.P., J.F., Y.M., P.G., C.S., K.P., and N.G. acquired data. X.X., S.Z.H., H.D., and G.K.G. analyzed and interpreted the data. X.X. wrote the manuscript. All authors revised the article and approved the final version to be published. X.X. and G.K.G. are the guarantors of this work and, as such, had full access to all the data in the study and take responsibility for the integrity of the data and the accuracy of the data analysis.

Prior Presentation. Part of this study was presented orally (no. 2015-A-1495-Diabetes) at the 75th Scientific Sessions of the American Diabetes Association, Boston, MA, 5–9 June 2015.