In Vivo Misfolding of Proinsulin Below the Threshold of Frank Diabetes

Rabbit antisera against GFP was from Immunology Consultants (Newberg, OR); anti–α-tubulin was from Santa Cruz Biotechnology (Santa Cruz, CA); anti-immunoglobulin heavy chain-binding protein (BiP) was from Cell Signaling (Danvers, MA); AlexaFluor-488–conjugated anti-GFP was from Invitrogen (Carlsbad, CA); peroxidase-conjugated anti-rabbit and peroxidase-conjugated anti–guinea pig were from Jackson ImmunoResearch Laboratories (West Grove, PA); AlexaFluor-555–conjugated anti-rabbit was from Invitrogen; rat insulin radioimmunoassay (RIA) was from Millipore (Billerica, MA); mouse proinsulin ELISA was from ALPCO (Salem NH); and collagenase-P and proteinase inhibitor mixture were from Roche Applied Science (Indianapolis, IN).

The emerald GFP cDNA was inserted into the C-peptide–coding sequence within the human insulin cDNA to create hProCpepGFP (17). The hProC(A7)Y-CpepGFP then used PCR mutagenesis to replace Cys(A7) with Tyr in the coding sequence. XhoI-flanking restriction sites were used to insert these constructs downstream of the 8.3-kb mouse Ins1 promoter and upstream of the nontranslated human growth hormone gene (18).

The linearized hProC(A7)Y-CpepGFP transgene was injected into pronuclei of fertilized mouse eggs at the University of Michigan Transgenic Animal Model Core. PCR genotyping was performed by PCR with GFP-specific primers (forward primer, 5′-AGG TCT ATA TCA CCG CCG ACA-3′; reverse primer, 5′-TGC AGT AGT TCT CCA GCT GGT AG-3′), yielding a 400-bp product. One line (of 10) was propagated, housed in a pathogen-free facility, and fed standard rodent chow in accordance with the University of Michigan Animal Care Committee. Unless otherwise stated, all experiments were done with 3–6-month-old nondiabetic male mice (e.g., Figs. 1A, 4A, 5, and 8). Intraperitoneal glucose tolerance tests were performed on fasted animals with 1.0 mg dextrose per gram body weight. For some experiments, hProCpepGFP and hProC(A7)Y-CpepGFP mice were mated to Ins2^−/− mice (from C. Polychronakos, McGill University, Montreal, Quebec, Canada) to increase the ratio between the products of transgenic and endogenous Ins1 alleles. In other experiments, compound heterozygotes bearing both wild-type hProCpepGFP and Akita mutant proinsulin were used as a control for the hProC(A7)Y-CpepGFP line that carries both elements in a single allele.

Fixed pancreata (formaldehyde plus 10% sucrose) were snap-frozen; 7-μm sections were permeabilized in acetone, blocked, and immunolabeled with guinea pig anti-insulin or antiglucagon and secondary antibodies in PBS containing 0.5% BSA. For BiP and GFP, de-paraffinized sections were heated in citrate (pH 5.0), cooled, blocked, incubated overnight with anti-BiP and appropriate secondary, blocked again, incubated overnight with AlexaFluor-488–conjugated anti-GFP, and mounted for confocal microscopy.

Pancreata were digested in 2 mg/mL collagenase-P for 30 min at 37°C, washed, and handpicked and incubated in RPMI 1640 medium containing either 4 or 16.7 mmol/L glucose (plus 10% fetal bovine serum and 1% penicillin–streptomycin) for 48 h.

Islet proteins were resolved by 4–12% gradient SDS-PAGE under reducing conditions; electrotransferred to nitrocellulose; probed with anti-insulin, anti-GFP, and anti–α-tubulin (loading control); and probed with appropriate peroxidase-conjugated secondary antibodies for enhanced chemiluminescence.

cDNA from template mRNA (RNeasy; QIAGEN, Valencia, CA) was generated by reverse transcriptase (SuperScript III; Invitrogen) and amplified with Taq polymerase in a real-time thermal cycler (denaturation 30 s, annealing 30 s, and extension 3 min) using appropriate primers, and SYBR green fluorescence followed each cycle. The following primers were used: GFP, 5′-AGG TCT ATA TCA CCG CCG ACA-3′ and 5′-TGC AGT AGT TCT CCA GCT GGT AG-3′; preproinsulin-II, 5′-CCC TGC TGG CCC TGC TCT T-3′ and 5′-AGG TCT GAA GGT CAC CTG CT-3′; BiP, 5′-GGC CAA ATT TGA AGA GCT GA-3′ and 5′-GCT CCT TGC CAT TGA AGA AC-3′; CHOP, 5′- CAT ACA CCA CCA CAC CTG AAA G-3′ and 5′-CCG TTT CCT AGT TCT TCC TTG C-3′; XBP-1, 5′-TGG CCG GGT CTG CTG AGT CCG-3′ and 5′-GTC CAT GGG AAG ATG TTC TGG-3′; spliced XBP-1, 5′-CTG AGT CCG AAT CAG GTG CAG-3′ and 5′-GTC CAT GGG AAG ATG TTC TGG-3′; and β-actin, 5′-TAT TGG CAA CGA GCG GTT CC-3′ and 5′-GGC ATA GAG GTC TTT ACG GAT GTC-3′. mRNA levels (normalized to β-actin) were calculated using the comparative threshold cycle (CT) method (2^−ΔΔCT).

A 175-bp, digoxigenin–labeled oligonucleotide probe complementary to GFP mRNA was used for in situ hybridization with anti–digoxigenin-phosphatase detection as described elsewhere (19). As a negative control, pancreatic islets from C57BL/6 mice were negative for detectable mRNA.

For each genotype, epifluorescence images from twenty 0.5-cm diameter pancreatic (tail) cryosections from three mice, with identical exposures, were quantified in grayscale using Adobe Photoshop CS3. To identify the predominant (pro)insulin (GFP-labeled or endogenous) in each β-cell, a sensitivity threshold was established at midrange of the pixel intensity histogram for each section. Cellular areas with signals above threshold for each type of (pro)insulin were demarcated, and DAPI staining identified each nucleus. Of the total cellular area of fluorescence, cells containing >80% GFP fluorescence area were classified as cells expressing primarily GFP-labeled proinsulin; cells expressing >80% insulin immunofluorescence area were classified as cells expressing primarily endogenous insulin; and β-cells expressing both in which neither label predominated were classified as expressing both. Cell boundaries (not shown) were corroborated by immunolabeling with anti–β-catenin.

Quantification was performed using NIH ImageJ 1.37c software. Results presented are mean ± SEM. Statistical analyses were performed using one-way ANOVA with Tukey multiple comparisons; P ≤ 0.05 was considered significant.

Pancreatic tissue from nondiabetic hProC(A7)Y-CpepGFP mice was fixed in buffered glutaraldehyde, postfixed in OsO₄, dehydrated with graded alcohols and propylene oxide, embedded in Spurr’s resin, and sectioned at 70 nm, grids-stained with uranyl acetate and lead citrate, and examined on a Philips CM-100 electron microscope.

Two transgenic lines were used. The hProCpepGFP mouse carries a GFP-labeled proinsulin (17), whereas the hProC(A7)Y-CpepGFP mouse carries a similar GFP-labeled proinsulin bearing the misfolding-inducing Akita mutation. In islets isolated from the two mouse lines, we examined an abundance of mRNAs encoding these GFP-labeled preproinsulins and the endogenous Ins2 mRNA levels (Ins2 represents the majority of total insulin transcripts in mouse islets [20]). Using qPCR (normalized to β-actin mRNA) with concentration standards of hProCpepGFP plasmid DNA, we confirmed that the levels of hProCpepGFP and hProC(A7)Y-CpepGFP mRNA represented only 5.5 and 4.2%, respectively, of that measured for endogenous Ins2 mRNA (Fig. 1A).

We examined pancreatic cryosections from neonatal male hProCpepGFP and hProC(A7)Y-CpepGFP mice (sex determined by Y-chromosome PCR) using intrinsic GFP fluorescence and insulin immunofluorescence. At postnatal day 1, hProCpepGFP and hProC(A7)Y-CpepGFP male mice demonstrated GFP fluorescence (either properly folded or misfolded mutant proinsulin) in essentially all of the cells that were immunolabeled for endogenous insulin + proinsulin, i.e., in the vast majority of β-cells (Fig. 1B; quantified in Fig. 7A). Thus, the data in Fig. 1 indicate that the two animal models demonstrate comparable levels and patterns of transgene mRNA and protein expression across the β-cell population in vivo.

We compared body weights and random glucose levels of hProC(A7)Y-CpepGFP transgenic mice with those of hProCpepGFP mice that are phenotypically identical to C57BL/6 littermates (17). Mean body weight of hProC(A7)Y-CpepGFP males was similar to that of normal control subjects, and mean random glucose of hProC(A7)Y-CpepGFP males (between 8:00 a.m. and 12:00 p.m.) was only modestly elevated and significantly lower than hProCpepGFP Akita males (Fig. 2A). A histogram of individual males of each genotype showed that random blood glucose levels in the vast majority (>85%) of hProC(A7)Y-CpepGFP males had a distribution largely overlapping with that of normal controls (Fig. 2B). Although a small subset of males (∼10% of total) exhibited random hyperglycemia, the results indicate that the majority of animals with transgenic expression of hProC(A7)Y-CpepGFP (driven by the Ins1 promoter) do not have frank diabetes (Figs. 2A and B), i.e., a considerably milder phenotype than Akita mice who bear the same mutation in one of the endogenous Ins2 genes.

To identify subtler defects in islet performance in hProC(A7)Y-CpepGFP males, we subjected the animals to intraperitoneal glucose tolerance testing after an overnight fast. Although fasting glucose values were similar to those of hProCpepGFP and C57BL/6 control males, the hProC(A7)Y-CpepGFP males had a significantly higher mean blood glucose at 30 min poststimulation than control subjects (309 ± 32.6 vs. 253.1 ± 16.1 mg/dL; P = 0.03). Although a significant difference was not sustained at 120 min postchallenge, the results indicate that hProC(A7)Y-CpepGFP mice have impaired glucose tolerance (Fig. 2C).

Homozygous knockout of the endogenous Ins2 gene (with homozygous expression of wild-type Ins1 still remaining) itself does not cause diabetes (20). Therefore, we were interested to use homozygous Ins2 knockout to examine the effect of increasing the relative expression of mutant hProC(A7)Y-CpepGFP (transgene) to wild-type insulin. In this background, mice with transgenic expression of wild-type hProCpepGFP were normoglycemic, whereas mice with transgenic expression of hProC(A7)Y-CpepGFP developed hyperglycemia that was particularly severe when both Ins2 alleles were missing, including the cohort of female animals (Fig. 3, bottom). Thus, although the majority of mice exhibit sufficient compensation to accommodate low-level expression of misfolded proinsulin, these animals are nevertheless predisposed to diabetes, and the relative expression levels of misfolded and wild-type proinsulin dictate the phenotype (Fig. 3).

To determine whether pancreatic insulin content was comparable in nondiabetic hProC(A7)Y-CpepGFP transgenic mice and nonmutant control subjects, we measured total insulin + proinsulin (endogenous and transgenic) in isolated pancreatic islets. By RIA, islet insulin + proinsulin content per cell was significantly less than that measured for hProCpepGFP transgenic controls (Fig. 4A), despite that all islets were derived from nondiabetic animals. In the small subset of diabetic hProC(A7)Y-CpepGFP males, the insulin + proinsulin content was lower still (<5% of control).

We also used an independent assay to follow proinsulin (and insulin) content in isolated islets incubated for 48 h in either 4 or 16.7 mmol/L glucose. Islet lysates were analyzed by reducing SDS-PAGE, electrotransferred to nitrocellulose, and immunoblotted with either anti-GFP (to follow the transgene product) or anti-insulin (to follow the endogenous product). In islets from both hProCpepGFP and hProC(A7)Y-CpepGFP mice, protein encoded by the transgene was up-regulated in response to incubation at high glucose. As expected, hProC(A7)Y-CpepGFP did not exhibit maturation, as measured by its inability to be endoproteolytically converted to CpepGFP, whereas hProCpepGFP was processed in secretory granules to CpepGFP (Fig. 4B). Endogenous proinsulin protein expression also was increased by high glucose; however, the amount of proinsulin was markedly less in islets of hProC(A7)Y-CpepGFP transgenic mice (Fig. 4B). The immunoblotted insulin band in islets of hProC(A7)Y-CpepGFP transgenic mice did not appear to be greatly affected by the high-glucose incubation; however, we found that when loading amounts that linearly report changes in proinsulin content (Fig. 4B, inset), insulin band-intensity changes were rather insensitive and may underestimate the insulin recovery defect in hProC(A7)Y-CpepGFP islets. Indeed by rodent proinsulin–specific RIA, we found that proinsulin content of the islets was 1.26% of total hormone content for both hProCpepGFP and C57BL/6 mice and 1.4% for nondiabetic hProC(A7)Y-CpepGFP mice. Altogether, the data of Fig. 4 suggest that the presence of misfolded mutant proinsulin decreases the levels of endogenous bystander proinsulin (Fig. 4B), concomitant with a decrease in islet insulin that occurs even in nondiabetic animals (Fig. 4A).

To determine whether expression of misfolded mutant hProC(A7)Y-CpepGFP below the diabetogenic threshold triggers ER stress response, mRNA levels for BiP and spliced XBP-1 were measured in freshly isolated islets from nondiabetic hProCpepGFP and hProC(A7)Y-CpepGFP transgenic mice. Indeed, islets of hProC(A7)Y-CpepGFP mice had significantly increased BiP mRNA, as well as a tendency to increased splicing of XBP-1 (Fig. 5). Furthermore, hProC(A7)Y-CpepGFP islets also exhibited a trend suggesting increased mRNA for CHOP (Fig. 5), a downstream component of the unfolded protein response pathway that predisposes to cell death (21–23). The results suggest that proinsulin misfolding by hProC(A7)Y-CpepGFP does induce ER stress, even in the absence of or before frank diabetes.

A recent report of PERK knockout mice has shown that a subpopulation of islet β-cells exhibits an “impacted-ER” phenotype characterized morphologically by an expanded ER with accumulation of proinsulin (4), consistent with ER crowding (1,2). Because of the evidence of ER stress in islets of nondiabetic hProC(A7)Y-CpepGFP mice (Fig. 5), we looked for morphological correlates in the β-cells of these animals. In cryosections of adult male transgenic mice, hProCpepGFP or hProC(A7)Y-CpepGFP protein expression was detected by intrinsic GFP fluorescence, whereas anti-insulin immunostaining was used to reflect endogenous insulin protein.

As in neonatal males (Fig. 1B), islets of hProCpepGFP adult males (17) expressed the GFP-positive transgene product in virtually all β-cells, which were simultaneously positive for endogenous insulin (Fig. 6A; quantified in Fig. 7B). Remarkably, in contrast to neonatal animals, islets of nondiabetic hProC(A7)Y-CpepGFP males reproducibly exhibited heterogeneous β-cell subpopulations (Fig. 6B). Approximately 40% of the β-cells accumulated the fluorescent, misfolded mutant proinsulin, which, at higher magnification, exhibited an ER distribution pattern (Fig. 6C). Within each islet, β-cells rich in hProC(A7)Y-CpepGFP fluorescence tended to be largely devoid of insulin immunostaining; another major subpopulation of β-cells exhibited little GFP fluorescence and immunostained clearly for endogenous insulin; and only a small, third subpopulation accumulated both types of molecules (Fig. 7B).

To begin to understand the origin of the heterogeneity of hProC(A7)Y-CpepGFP accumulation within the β-cell population, we also examined hProC(A7)Y-CpepGFP mRNA distribution within the islets by in situ hybridization with a specific complementary oligonucleotide RNA probe. As exemplified in Fig. 6D, as much as 60% of transgene mRNA–positive β-cells within the population exhibited little or no hProC(A7)Y-CpepGFP fluorescence. Evidently, the other ∼40% subpopulation of β-cells with misfolded proinsulin accumulation and little or no insulin immunostaining (Figs. 6B and 7B) develops during postnatal life. Even if the larger subpopulation of β-cells that failed to accumulate or had not yet accumulated significant quantities of misfolded proinsulin within the ER (Fig. 6B) contained all of the islet insulin that is recovered (Fig. 4A), the insulin content, even in this subpopulation of β-cells, must still be decreased compared with that of hProCpepGFP control islet β-cells.

As the islets of hProC(A7)Y-CpepGFP mice contained significantly increased BiP mRNA (Fig. 5), we also looked at the distribution of BiP protein compared with islets of hProCpepGFP mice. Although BiP was increased in many β-cells of hProC(A7)Y-CpepGFP mice, a heterogeneous pattern within the islets was observed (Fig. 8A, left). BiP tended to be increased in cells that had accumulated hProC(A7)Y-CpepGFP, but BiP was also increased in a few other cells that had not accumulated green fluorescence (Fig. 8A, right), suggesting that these cells also were synthesizing increased amounts of misfolded secretory protein.

To examine β-cell heterogeneity at the ultrastructural level, we performed transmission electron microscopy of the islets of nondiabetic hProC(A7)Y-CpepGFP islets. Populations of highly granulated β-cells were readily identified (Fig. 8B, β-cell #1). However, side by side with such cells were other β-cells with a highly expanded ER and many fewer (but definite) insulin secretory granules (Fig. 8B, β-cell #2). Upon close inspection, dilation of the ER was also seen in the β-cells that retained abundant insulin secretory granules (Fig. 8C and D, β-cell #1). β-Cells with a dramatically expanded ER compartment tended to have unusually small insulin secretory granule profiles (Fig. 8C, microgranules). Finally, we were also surprised to discover the unique morphological appearance of a third type of β-cell that also had insulin microgranules but lacked an expanded ER compartment; rather, such cells exhibited a highly shrunken cytoplasm (Fig. 8D, β-cells #2 and #3). It is not clear whether this third kind of β-cell has either sufficient hProC(A7)Y-CpepGFP protein or endogenous insulin protein to be detected either by GFP fluorescence or insulin immunofluorescence as was measured in Figs. 6 and 7.

To further clarify whether the decreased islet insulin is a consequence of a decrease of insulin within β-cells or a decrease in islet size and a resultant decrease of overall β-cell numbers, we examined multiple random pancreatic tail cryosections of 3–6-month-old nondiabetic hProCpepGFP and hProC(A7)Y-CpepGFP male mice. Islets contained within 0.2-cm² pancreatic cross-sections from three different mice of each genotype were identified, and, by use of GFP fluorescence plus immunostaining with anti-insulin, the islet boundaries were determined and the islet cell nuclei counted by DAPI staining. (Antiglucagon immunostaining was also performed in both sets of mice [not shown], confirming that α-cells did not exceed 10% of islet cells in either mouse line.)

On average, the random cross-sectional islet area of nonmutant hProCpepGFP mice (n = 21 islets) was 2,545 µm² (±276 SEM), similar to that obtained from C57BL/6 control mice (not shown), whereas random cross-sectional islet area of hProC(A7)Y-CpepGFP mice (n = 25 islets) averaged 4,221 µm² (±793 SEM), a considerable (65%) increase. On average, each hProCpepGFP islet cross-section contained 40.0 (±6.2 SEM) β-cell profiles, whereas each hProC(A7)Y-CpepGFP islet cross-section contained 62.8 (±23.7 SEM) β-cell profiles. Dividing average islet cross-sectional area by average number of β-cell profiles per cross-section, a rough estimate of average β-cell cross-sectional area was essentially unchanged between the two mouse lines. Thus, rather than β-cell hypertrophy, the data indicate an expansion of β-cell number per islet, suggesting β-cell hyperplasia in compensation for expression of misfolded proinsulin to help these animals avoid diabetes. Even when considering that many islet β-cells accumulating fluorescent misfolded mutant proinsulin have little or no endogenous insulin (Fig. 7B), there was not an actual loss of insulin-positive β-cells in nondiabetic hProC(A7)Y-CpepGFP islets, indicating that decreased islet insulin content in these animals (Fig. 4A) must be caused by a decrease of insulin content per β-cell.

Studies have suggested β-cell ER crowding/ER stress in animals in which hyperglycemia was already present (1,3,24). Glucotoxicity is a potentially confounding variable making other pathogenic mechanisms more difficult to analyze. Herein, we have compared nonmutant hProCpepGFP transgenic males to nondiabetic hProC(A7)Y-CpepGFP transgenic males, with each transgene accounting for only a few percent of proinsulin mRNA (Fig. 1). Whereas permanent neonatal diabetes occurs both in humans (9) and Akita males (10), the hProC(A7)Y-CpepGFP transgene is driven by the weak Ins1 promoter coexpressed with four wild-type alleles; thus, diabetes penetrance is more subtle (Fig. 2). A distinct advantage of this model is that effects of frank diabetes are largely excluded, although hProC(A7)Y-CpepGFP males are rendered severely diabetic upon deletion of endogenous Ins2, i.e., when the relative expression of misfolded proinsulin to wild-type proinsulin is increased (Fig. 3). However, even in nondiabetic hProC(A7)Y-CpepGFP males with a full complement of endogenous Ins genes, islet β-cells have decreased insulin production (Fig. 4A), inhibited by coexpressed mutant proinsulin (Fig. 4B) as observed in β-cells of Akita males (17) (females tend to resist diabetes [25,26]).

In hProC(A7)Y-CpepGFP males, misfolded proinsulin and its dominant interference with endogenous proinsulin causes ER stress that increases BiP (Fig. 5), predating frank diabetes. Among other ER stress response targets, CHOP is proapoptotic (21–23). Nevertheless, the initial decrease of insulin production is not accompanied by decreased β-cell mass in Akita mice (4) or in prediabetic hProC(A7)Y-CpepGFP mice (this study). Evidently, either because of or in spite of ER stress responses, there is pancreatic compensation for proinsulin misfolding with decreased insulin production.

Surprisingly, we found that, in islets of nondiabetic adult hProC(A7)Y-CpepGFP transgenic males, despite that most or all β-cells express the transgene mRNA (Fig. 6D), misfolded proinsulin accumulation appears concentrated in a subpopulation of islet β-cells (Figs. 6B and 7B), as has recently been described in other mouse models (4). The fact that β-cell subpopulations are not detected on postnatal day 1 (Fig. 1B) suggests that β-cell heterogeneity develops during postnatal life. Preservation of insulin production includes maintenance of a robust pool of insulin secretory granules in a subpopulation (Figs. 6 and 8) that is undoubtedly linked to the islet compensation necessary to avoid diabetes. One hypothesis needing to be tested is that this might be explained by the slowly progressive appearance of new β-cells that have not yet accumulated the misfolded proinsulin product. Indeed, islets of nondiabetic hProC(A7)Y-CpepGFP male mice exhibit morphological evidence of β-cell hyperplasia within islets. Generation of new β-cells may be needed to compensate for others that have lost insulin production (generating only a few microgranules; Fig. 8C and D). Such a hypothesis interdigitates well with the fact that there are already known to be β-cell subpopulations in adult animals that operate at different rates of proinsulin biosynthesis (27–29), and this may include variability in secretory feedback on proinsulin synthesis (30) and secretion (31) as well as potentially different susceptibilities to cell stress (32). Of course, the mutant hProC(A7)Y-CpepGFP represents only a tiny fraction of total proinsulin mRNA (Fig. 1A) and an even smaller fraction of total β-cell protein synthesis. Thus, whether mutant proinsulin protein is variably synthesized among the different β-cells within islets (especially under different glucose conditions) is unknown.

It seems likely that there is a continuum of proinsulin misfolding, ER crowding, and ER stress in a range of diabetes subtypes, from permanent neonatal diabetes to type 2 diabetes. Unfortunately, not all genetically predisposed individuals exhibit sufficient compensatory responses to prevent diabetes. Our results are consistent with the suggestion that β-cell dysfunction exists before the onset of overt diabetes, and patients with such β-cell dysfunction may benefit from early pharmacological intervention to limit proinsulin misfolding or to preserve insulin production through suitable compensatory responses.

This work was supported primarily by National Institutes of Health grants R00-DK-077441A (to I.H.) and R01-DK-48280 (to P.A.), a grant from the Michigan Diabetes Research and Training Center (to M.L.), and Endocrinology T32 grant (to L.L.).

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

I.H. researched the data, contributed to the discussion, and wrote the manuscript. A.A., M.L., and L.H. researched the data and contributed to the discussion. L.L., D.L., A.A.-K., and A.Z. researched the data. P.A. contributed to the discussion, wrote the manuscript, and reviewed and edited the manuscript.

The authors acknowledge assistance from the University of Michigan Transgenic Mouse Core, Morphology and Image Analysis Core, Molecular Biology and DNA Sequencing Core, supported in part by P60-DK-20572, the University of Michigan Gut-Peptide Center, and the University of Michigan Cancer Center for assistance with electron microscopy. The authors thank Bill and Dee Brehm for improving diabetes research infrastructure at the University of Michigan.