A Whole-Genome RNA Interference Screen Reveals a Role for Spry2 in Insulin Transcription and the Unfolded Protein Response

Worldwide, 354 million people have diabetes, and this is projected to increase to 592 million by the year 2035 (1). The majority of this rise will be in type 2 diabetes, although the incidence of type 1 diabetes also continues to rise (2,3). The pathophysiological common ground of diabetes is β-cell dysfunction and loss. Genome-wide association studies of patients with type 2 diabetes have identified approximately eighty risk loci, many of which are located near genes that would be predicted to affect β-cell function (reviewed by Morris [4]). However, only a few of the identified single nucleotide polymorphisms (SNPs) have been experimentally verified to work through the genes to which they have been assigned. We hypothesized that augmenting current genome-wide association studies results with unbiased screens for the genes that control normal β-cell function would reveal novel targets for future diabetes therapeutics and generate a more complete picture of the genes involved in the pathogenesis of human diabetes.

We performed a whole-genome RNA interference (RNAi) screen for regulators of the human insulin promoter. We identified novel regulators of insulin production including a type 2 diabetes locus, Spry2 (5–7). Spry2 is known to negatively regulate growth factor signaling, but its link to diabetes is not clear. We discovered a novel role for Spry2 in the unfolded protein response (UPR) and established it as a UPR target. Our study provides a blueprint for functionally annotating the β-cell insulin production pathway and reveals a novel mechanism linking Spry2 to human diabetes.

MIN6 cells were a gift from Dr. J. Miyazaki (Osaka University). The screening MIN6 cell line, containing a lentivirally delivered copy of the human insulin promoter driving EGFP and a distinct integrated copy of the rous sarcoma virus (RSV) promoter driving mCherry, was previously described (8). A total of 1.25 pmol of each small interfering RNA (siRNA) from the Qiagen Whole Genome Mouse siRNA library version 1.0 was placed individually into wells of 384-well plates. Totals of 0.05 μL of RNAiMax (Life Technologies) and 9.95 μL of Opti-MEM (Life Technologies) followed by 10,000 reporter MIN6 cells in 40 μL of growth media (DMEM high glucose, 10% FBS, penicillin, streptomycin) were added to each well. Five days after transfection, fluorescence was read using an Analyst HT (Molecular Devices). GFP fluorescence was divided by mCherry fluorescence. Wells with mCherry fluorescence 1.5 SD below the mean of that plate were removed because of presumed cellular toxicity. Scores were median two-way polished to remove edge effects. A subset of the whole-genome library was repeated using 0.5 μL of HiPerFect (Qiagen) and 9.5 μL of Opti-MEM (Life Technologies) per well instead of RNAiMax. Transfection efficiency for both reagents was typically >90%.

Total RNA was prepared with TRIzol Reagent (Life Technologies) followed by DNA digestion with TURBO DNase (Ambion). Oligo-dT– and random hexamer–primed cDNA was synthesized with SuperScript III (Life Technologies). For TaqMan primers, the delta-delta CT method was used using the average CT of β-actin, β-glucuronidase, cyclophilin, and TBP as reference genes. For SYBR Green primers, data were normalized to β-actin. Primer sequences are listed in Supplementary Table 4.

A single RNA isolation from the screening cell line was used to generate a polyA-primed mRNA-Seq library as previously described, except Cufflinks 2.0.2 was used for the fragments per kilobase of transcript per million mapped reads (FPKM) determination (9,10). Reads have been deposited in the National Center for Biotechnology Information Sequence Read Archive (SAMN06250888).

Database for Annotation, Visualization and Integrated Discovery v6.7 was used to find enriched gene ontology (GO) terms (11).

Spry2 floxed mice were a gift from Dr. O. Klein (University of California, San Francisco). Mouse insulin promoter–CreERT (MIP1-CreERT) mice were a gift from Dr. N. Tamarina (University of Chicago). Mice received three subcutaneous 2-mg tamoxifen injections (Sigma-Aldrich) at 6 weeks of age; each injection was separated by a day. Intraperitoneal glucose tolerance testing was performed at 12 weeks: 2 g/kg glucose was injected into the peritoneal space after a 6-h fast, and blood glucose was monitored with a FreeStyle Lite Glucometer (Abbott). Glucose tolerance P values between the three genotypes were calculated using a one-way ANOVA with repeated measures followed by pairwise comparisons corrected with the Benjamini-Hochberg method for multiple testing. The glucose area under the curve above fasting was measured using the trapezoidal method. Insulin tolerance tests were performed at 14 weeks: after a 6-h fast, 0.1 mU/gram body weight of regular insulin was injected into the peritoneal space. Plasma insulin levels were measured at 15 weeks: after a 16-h fast and after a 2 g/kg intraperitoneal glucose injection. Islets were isolated as previously described (12). RNA was harvested with TRIzol 1 day after isolation.

Cells were lysed in buffer containing the following: 50 mmol/L TrisHCl pH 6.7, 150 mmol/L NaCl, 1% Triton X-100, cOmplete Protease Inhibitor Cocktail (Roche), Phosphatase Inhibitor Cocktail 2 (Sigma-Aldrich), and Phosphatase Inhibitor Cocktail 3 (Sigma-Aldrich). The following antibodies were used: anti-Wfs1 (11558–1-AP; Proteintech), phospho-PERK (3179; Cell Signaling), phospho-Eif2α (3398; Cell Signaling), total Eif2α (9722S; Cell Signaling), total PERK (3192S; Cell Signaling), GAPDH (G8795; Sigma-Aldrich), Spry2 (ab50317; Abcam), and Serca2 (sc-8095; Santa Cruz Biotechnology). Islet protein samples were isolated from the TRIzol Reagent organic phase after RNA isolation as described in the protocol, except the pellet was resuspended in 5 mol/L urea in 0.5% SDS.

Unmodified MIN6 cells were infected with a lentivirus expressing endoplasmic reticulum (ER)-targeted fluorescent cameleon D1ER. If indicated, the cells were transfected with the indicated siRNAs and rested for 5 days prior to analysis. For measurement of ER cells low in calcium, the cells were trypsinized and resuspended in complete media. Flow cytometry was performed using an Attune Cytometer (Life Technologies) using VL2 and VL1 as the fluorescence resonance energy transfer (FRET) emission channel and the CFP (donor) channel, respectively. VL2 was divided by VL1 to calculate the FRET signal. The FRET low population was defined as those having >33% reduction in the FRET signal compared with the geometric mean of the population. FRET low was typically 5–8% of the cell population for the control siRNA transfection.

Anti-luciferase siRNA, AllStars Negative Control siRNA, and Negative Control siRNA were obtained from Qiagen.

Parental MIN6 cells were stained with Annexin V Alexa Fluor 488 & Propidium Iodide (Life Technologies) and analyzed by flow cytometry (Attune; Life Technologies). The percentage of total events that were annexin V positive and propidium iodide negative is shown. For percent cell death, MIN6 cells were trypsinzed and the percentage of events in the live gate determined by forward and side scatter was subtracted from 100%.

Unless otherwise specified, P values were calculated by two-tailed Student t test and corrected for false discovery using the Benjamini-Hochberg method. For quantitation of Western blot bands (phospho-PERK, phospho-Eif2α, Spry2 induction by UPR inducers, Spry2 or Wfs1 knockdown), the log-transformed ratio of the experimental condition to the control condition was used for a one-sample Student t test.

To elucidate genes that alter insulin transcription, we previously generated a MIN6 mouse insulinoma clonal cell line that harbors an EGFP reporter driven by the proximal 362 bases of the human insulin promoter. To avoid selecting hits that affect general transcription, this line was engineered to constitutively express a control mCherry reporter driven by the RSV promoter (8). This reporter cell line was used in a high-throughput RNAi screen using a commercial library containing 75,366 unique siRNAs targeting ∼22,000 mouse genes with 3–4 unique siRNAs per gene. Each siRNA was individually reverse transfected into the reporter cell line in 384-well plates. Five days after transfection, the ratio of GFP to mCherry fluorescence was measured (Fig. 1A). The plate-based assay performance metric Z′ typically ranged from 0.6–0.8 (0.5 is considered acceptable; Supplementary Fig. 1A) (13). Gene-level P values for positive regulators of insulin expression (reduced GFP/mCherry) were determined using the redundant siRNA analysis (RSA), where P values are assigned based on the number and strength of siRNAs to that gene (14).

A portion of the screen was repeated with an alternative lipid transfection reagent to ensure results were not specific to the transfection reagent used. Normalized scores for each siRNA were well correlated (Supplementary Fig. 1B), and gene-level P values for positive regulators of insulin transcription were also well correlated (Supplementary Fig. 1C). To reduce off-target effects, we did not further consider siRNAs targeting genes that were not expressed using mRNA-Seq of the screening cell line (Supplementary Table 1). As would be expected, the fraction of hits that were not expressed in the screening cell line increased as more inferior hits were considered (Supplementary Fig. 1D).

When ranked based on low GFP/mCherry (i.e., specific suppression of insulin expression), the key β-cell transcription factors, Glis3, Pdx1, and Neurod1, were ranked 3rd, 17th, and 29th, respectively (Fig. 1B), providing strong validation for the screen. Wfs1, a known regulator of both insulin production and insulin secretion, was also identified. GO terms involving mitochondria, DNA transcription, vesicular transport, and cytoskeleton were significantly enriched among the top 300 genes that, when knocked down, reduced GFP/mCherry (Fig. 1C).

To validate the primary screen, we retested the top 40 siRNAs targeting putative positive regulators, selecting the top 2–3 performing siRNAs for each gene by GFP/mCherrry reduction. These siRNAs were transfected into the screening cell line in biological triplicate, and GFP and mCherry were measured. Of the top 40 siRNAs targeting putative positive regulators, 36 (90%) reduced GFP/mCherry compared with a control siRNA during retesting. Of 50 random siRNAs selected from the library, only one showed a statistically significant reduction in GFP/mCherry (Fig. 1D). We then selected additional siRNAs for retesting based on β-cell expression, interconnectivity (based on GO term enrichment), and published data for a specific role in β-cell function (Fig. 1E). In total, 141 siRNAs targeting positive regulators were tested in the screening assay with RSA ranks up to 6,313. A total of 112/141 (79%) showed a statistically significant reduction in GFP/mCherry compared with a control siRNA (Fig. 1E and Supplementary Table 2).

As GFP/mCherry fluorescence is a surrogate for insulin promoter activity, siRNAs were tested for their ability to reduce endogenous, unspliced insulin mRNA, as this has a shorter half-life than mature insulin message (15,16). We used a quantitative RT-PCR (qRT-PCR) assay for preIns1 mRNA, as we found that this was more sensitive to Pdx1, Glis3, and NeuroD1 knockdown than preIns2 mRNA (data not shown). A total of 78/108 (72%) siRNAs reduced preIns1 mRNA. We considered genes with two siRNAs showing reduced preIns1 mRNA as high-confidence hits. We found 29 such genes, 26 of which did not have a known role in insulin promoter activity. An additional 15 genes had one of two siRNAs reduce insulin mRNA levels (Fig. 2A). With the exception of three genes without a known human ortholog, all of our validated hits are expressed in human islets at reads per kilobase of transcript per million mapped reads (RPKM) >0.1, suggesting that they could play a similar role in insulin production in human β-cells (Supplementary Table 3) (17). Although the screen was not optimized to detect increases in GFP/mCherry, when the siRNAs were ranked according to increased GFP/mCherry (targeting putative negative regulators), the top 40 siRNAs did tend to subtly increase GFP/mCherry during retesting (Fig. 1D). However, a low rate of these siRNAs increased preIns1 mRNA (data not shown), so we focused on the positive regulators.

To begin to validate our screen in vivo, we examined global knockout (ko) mice of Gpr75, one of the G-protein–coupled receptor hits in the screen. We found that Gpr75 ko mice were modestly hyperglycemic compared with their wild-type (wt) or heterozygous littermates (Supplementary Fig. 2A and B), despite being more insulin sensitive and having lower body weight (Supplementary Fig. 2C and D). Plasma insulin levels were lower before and after a glucose challenge in Gpr75 ko mice as compared with their wt littermates (Supplementary Fig. 2E). Finally, insulin secretion in vitro was also blunted in islets from Gpr75 ko mice (Supplementary Fig. 2F). These data show that our in vitro whole-genome RNAi screen was capable of detecting in vivo regulators of β-cell function.

Among the other novel hits, we were particularly interested in Spry2 because it has been identified in multiple studies as a type 2 diabetes risk locus (5,7). We first returned to the original confirmation qRT-PCR for preIns1. Of the two Spry2 siRNAs initially retested, only Spry2_4 statistically significantly reduced preIns1 mRNA; Spry2_6 had a weaker effect and a false discovery rate (FDR)-adjusted P value of 0.13 after three biological replicates (Fig. 2A and Supplementary Table 2). Therefore, we tested all three siRNAs targeting Spry2 that were identified as hits in the primary screen. After nine biological replicates, all three independent siRNAs to Spry2 statistically significantly reduced preIns1 mRNA levels in MIN6 cells after FDR correction (Fig. 2B). All three siRNAs reduced Spry2 mRNA levels (Supplementary Fig. 3).

In many cell types, Spry2 negatively regulates Ras-MAPK signaling (18). Since MAPK signaling is thought to positively regulate insulin transcription, we sought another mechanism for Spry2’s positive influence on insulin production (19). We hypothesized that Spry2 might play a role in the UPR, as dominant negative Spry2 expression in the mouse lens has been reported to increase ER stress (20). Furthermore, SPRY2 was induced in human islets treated with palmitate, which induces ER stress (21).

We first asked whether Spry2 is induced by established chemical UPR inducers. Exposure of MIN6 cells to the sarcoplasmic/endoplasmic reticulum calcium ATPase (Serca) inhibitor thapsigargan increased Spry2 mRNA and protein expression (Fig. 3A and B). The degree of upregulation was similar to that of Wfs1 mRNA, a gene known to be upregulated by ER stress (22,23). The N-linked glycosylation inhibitor and ER stress inducer tunicamycin also increased Spry2 mRNA and protein (Supplementary Fig. 4A and B). Mouse embryonic fibroblasts (MEFs) also upregulated Spry2 mRNA in response to thapsigargan (Fig. 3C) or tunicamycin (Fig. 3D). However, MEFs deficient in PERK did not upregulate Spry2 mRNA in response to thapsigargan or tunicamycin. Finally, since knockdown of Wfs1 is known to increase ER stress in β-cells, we asked whether knockdown of Wfs1 might also increase Spry2 expression in MIN6 cells (24,25). Multiple siRNAs targeting Wfs1 increased Spry2 mRNA and Spry2 protein expression in the absence of any other ER stress inducer (Fig. 3E and F). Notably, these Wfs1 siRNAs were effective in knockdown of Wfs1 mRNA (Supplementary Fig. 4C) and Wfs1 protein (Supplementary Fig. 4D) and in reducing preIns1 mRNA (Supplementary Fig. 4E).

Next, we hypothesized that Spry2 could play a role in regulating the UPR. Knockdown of Spry2 by multiple independent siRNAs resulted in increased levels of the UPR markers CHOP and spliced Xbp1 (Fig. 4A and B). BiP expression did not increase after either Spry2 or Wfs1 knockdown (data not shown). To further confirm activation of the UPR, we examined the activation of the ER stress sensor PERK and its target Eif2α after siRNA knockdown of Spry2 or Wfs1. Both phospho-PERK and phospho-Eif2α levels increased after either Spry2 or Wfs1 knockdown (Fig. 4C–E). ATF4 and ATF6 levels did not change after Spry2 knockdown (Supplementary Fig. 5). Since ER stress in β-cells can trigger apoptosis, we measured the frequency of annexin V⁺ cells after Spry2 knockdown. Indeed, knockdown of Spry2 increased the fraction of annexin V⁺ cells (Fig. 4F).

Since PERK activation can be triggered by low levels of ER calcium, we measured ER calcium levels using the D1ER cameleon reporter by flow cytometry (26,27). Treatment of D1ER-expressing MIN6 cells with EDTA and ionomycin resulted in an ∼60% increase in the number of cells with low levels of FRET, confirming our ability to monitor ER calcium concentrations with D1ER (Supplementary Fig. 6A and B). When D1ER MIN6 cells were transfected with anti-Wfs1 siRNAs, there was a 40% increase in ER calcium–low cells, in line with what has been previously reported (26). Spry2 knockdown by two independent siRNAs also increased the fraction of cells with low levels of ER calcium by 25% (Fig. 5A).

Decreased ER calcium could be caused by increased activity of IP3 receptors. However, neither IP1 levels (a breakdown product of IP3) nor expression of PLCγ1, a producer of IP3 that can be regulated by Spry2, changed after Spry2 knockdown (Supplementary Fig. 6C and D). Phosphorylated PLCγ1 was not detectable with or without Spry2 knockdown (data not shown). Therefore, we examined levels of Serca2 after Spry2 knockdown. Serca2 protein levels were reduced by 50% after Spry2 knockdown (Fig. 5B and C), while mRNA levels were not changed (Fig. 5D), suggesting posttranscriptional control of Serca2 protein by Spry2.

Since ER stress is known to impact insulin promoter activity, we asked whether amelioration of ER stress could restore insulin production in Spry2 knockdown cells. Treatment with tauroursodeoxycholic acid (TUDCA) was able to rescue the reduced insulin mRNA seen after Spry2 knockdown (Fig. 5E). Furthermore, TUDCA treatment was able to partially rescue increased β-cell death in MIN6 cells after Spry2 knockdown (Fig. 5F).

To demonstrate a role for Spry2 in vivo, we crossed a conditional allele of Spry2 to MIP1-CreERT mice (28,29) (Fig. 6A). Since we were unable to identify a Spry2 antibody suitable for immunohistochemistry, we used droplet digital PCR to measure ko efficiency from islet genomic DNA. Compared with Spry2 wt/wt MIP1-CreERT mice injected with tamoxifen (Spry2 wt), there was a 50% reduction in Spry2 islet genomic DNA in Spry2 flox/flox MIP1-CreERT mice injected with tamoxifen (Spry2 β-ko) (Supplementary Fig. 7A). The heterozygous Spry2 mice (Spry2 β-het) showed an intermediate degree of reduction. As ∼70% of the cells from islets are expected to be β-cells, we estimate that the recombination efficiency was approximately 70%. As Spry2 mRNA is present at approximately twofold higher levels in the mouse α-cell than in the mouse β-cell, we observed a modest but significant reduction in Spry2 from total islet mRNA of these animals (Supplementary Fig. 7B) (30). Protein levels of Spry2 were reduced in Spry2 β-ko islets compared with Spry2 β-het islets (Supplementary Fig. 7C and D).

We asked whether loss of Spry2 could result in changes in glycemia. After glucose challenge, the Spry2 β-ko mice were statistically significantly hyperglycemic relative to either Spry2 wt or Spry2 β-het mice (Fig. 6B and C). Spry2 β-ko mice also had lower plasma insulin levels 15 min after glucose injection (Fig. 6D), while their insulin tolerances and weights were not different (Supplementary Fig. 7E and F). Islets from Spry2 β-ko mice had slightly lower levels of mature Ins1/2 mRNA, confirming our findings in MIN6 cells (Fig. 6E).

Despite lower insulin mRNA levels, total pancreatic insulin content normalized to total protein was not reduced in the Spry2 β-ko as compared with β-het islets (Supplementary Fig. 7G), and we also did not find any difference in β-cell area in the Spry2 β-ko as compared with the Spry2 β-het mice (Supplementary Fig. 7H). Batch in vitro glucose-stimulated insulin secretion of islets from Spry2 β-het or Spry2 β-ko mice was also not different (Supplementary Fig. 7I).

Confirming our data in MIN6 cells, phospho-PERK levels were elevated in the islets of Spry2 β-ko mice as compared with those from Spry2 wt mice (Fig. 6F and G). Phospho-Eif2α levels and CHOP mRNA were also increased in Spry2 β-ko islets, but these increases did not reach our statistical significance cutoff (P = 0.07 and P = 0.055, respectively) (Fig. 6H and Supplementary Fig. 8A). As in MIN6 cells, we did not detect differences in ATF4 in islets from Spry2 β-ko mice (Supplementary Fig. 8B), which suggests that it is predominantly the PERK pathway that is activated by Spry2 loss in the β-cell. Unlike in MIN6 cells, we did not detect a difference in spliced Xbp1 mRNA in Spry2 β-ko islets (Supplementary Fig. 8A). We also did not detect a difference in fasting proinsulin levels between Spry2 β-ko and Spry2 β-het mice (Supplementary Fig. 8C).

We describe the first whole-genome RNAi screen for regulators of the insulin promoter. In terms of absolute insulin mRNA reduction, the strongest hits came from knockdown of the well-known transcription factors Glis3, NeuroD1, and Pdx1—perhaps not surprising since these proteins bind directly to the insulin promoter. Other hits were previously known to positively regulate insulin secretion in the β-cell but not insulin transcription. For example, Gpr142 was known to positively regulate insulin secretion in vivo and in vitro, and we now identify it as a potential regulator of insulin transcription (31,32). Gpr75 positively regulates insulin secretion in cell lines, and our data further these findings by showing that loss of Gpr75 results in impaired insulin secretion in vivo (33). More importantly, these data show that our screen was able to identify regulators of β-cell function that are relevant in vivo. Since most of our hits have no known connection to β-cell function, the study of these novel regulators will greatly expand our understanding of insulin regulation.

On a global level, this unbiased screen highlights several critical pathways required for insulin transcription. Multiple components of the oxidative phosphorylation cascade were identified as positive regulators of insulin mRNA production (Atp5a1, Atp5j2, Cox6b1, Cox6c, Cox7a2). While oxidative phosphorylation is known to be important in glucose sensing for insulin secretion, our data suggests that it is also important for insulin transcription (34).

ER stress and the UPR were another thematic area identified by the screen. The UPR is known to reduce insulin transcription through multiple mechanisms, including ATF6 and spliced XBP1 (35–37). Besides Wfs1, activation of the UPR likely also explains the hits involved in vesicular trafficking between the ER and the Golgi (Copb1, Gbf1, Sec23b). In fact, mice lacking the small GTPase Sec23b lack professional secretory cells (including β-cells) and have UPR activation in the embryonic pancreas (38).

We chose to focus on Spry2 because a SNP near the gene is associated with type 2 diabetes. We show a novel role for Spry2 in insulin transcription and glucose homeostasis in vivo. Furthermore, we demonstrate that Spry2 is both a target and a negative regulator of the PERK arm of the UPR, suggesting that upregulation of Spry2 may be a part of the homeostatic response to increased ER stress. However, we were unable to detect a change in β-cell area or in in vitro insulin secretion in Spry2 β-ko mice. Since the hyperglycemia of the Spry2 β-ko mice was relatively mild, we might only expect a small reduction in these assays that could be below our limit of detection. Therefore, the defect in the Spry2 β-ko islets could be from reduced mass, reduced insulin secretion, or possibly both.

Our data nonetheless demonstrate an explanation of how a SNP near Spry2 might cause human diabetes—through activation of the β-cell UPR. How could loss of Spry2 activate the UPR? Although Spry1 and Spry2 have been previously shown to bind to and negatively regulate PLCγ1 activity in non–β-cells, thereby reducing inositol triphosphate production and calcium efflux from the ER, we did not observe changes in PLCγ1 phosphorylation or increased inositol phosphate production (39). Instead, we found that Spry2 knockdown in β-cells causes reduced Serca2 protein levels, suggesting a possible mechanism by which Spry2 loss might cause reduced ER calcium and the UPR. However, since ER calcium concentrations can also be reduced by high levels of ER stress, it remains possible that Spry2 loss causes ER stress by an alternative mechanism, such as accumulation of ubiquitylated proteins, decreased autophagy, or altered vesicular traffic (20,26,40,41). Future studies will be required to define how Spry2 loss results in activation of the UPR.

In summary, we have performed the first whole-genome RNAi screen in β-cells. We have identified novel and unexpected positive regulators of the insulin promoter and have begun to elucidate how these genes function together to control pancreatic β-cell insulin production. Specifically, we have demonstrated a novel role for Spry2 in the UPR and in insulin production in β-cells, possibly explaining the association between SNPs near Spry2 and type 2 diabetes susceptibility. Further understanding of the role of Spry2 and the UPR may provide novel therapeutic inroads to understand and treat human diabetes.

Acknowledgments. The authors thank Dr. Feroz Papa (University of California, San Francisco) for useful discussion.

Funding. This work was supported by National Institutes of Health National Institute of Diabetes and Digestive and Kidney Diseases grants 5R01-DK-021344 (M.S.G.), 5RC1-DK-086290 (M.T.M.), 5K08-DK-087945 and 5R01-DK-107650 (G.M.K.), and P30-DK-087945; National Institute of Mental Health grant U01MH105028 (M.T.M.); the A.P. Giannini Foundation (G.M.K.); and JDRF grant 5-CDA-2014-199-A-N (G.M.K.).

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

Author Contributions. M.S.G., M.T.M., and G.M.K. conceived of the experiments. Z.P., D.G.C., T.G.H., H.R., J.C., K.Y., L.B., K.A., S.C., M.A., and G.M.K. performed experiments, analyzed data, and reviewed the manuscript. G.M.K. wrote the manuscript. G.M.K. 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 75th Scientific Sessions of the American Diabetes Association, Boston, MA, 5–9 June 2015.