Reversible changes in pancreatic islet structure and function produced by elevated blood glucose

Diabetes is characterized by hyperglycaemia due to impaired insulin secretion and aberrant glucagon secretion resulting from changes in pancreatic islet cell function and/or mass. The extent to which hyperglycaemia per se underlies these alterations remains poorly understood. Here we show that β-cell-specific expression of a human activating KATP channel mutation in adult mice leads to rapid diabetes and marked alterations in islet morphology, ultrastructure and gene expression. Chronic hyperglycaemia is associated with a dramatic reduction in insulin-positive cells and an increase in glucagon-positive cells in islets, without alterations in cell turnover. Furthermore, some β-cells begin expressing glucagon, whilst retaining many β-cell characteristics. Hyperglycaemia, rather than KATP channel activation, underlies these changes, as they are prevented by insulin therapy and fully reversed by sulphonylureas. Our data suggest that many changes in islet structure and function associated with diabetes are attributable to hyperglycaemia alone and are reversed when blood glucose is normalized.

D iabetes mellitus currently affects over 330 million people worldwide. It is a heterogeneous disorder with multiple aetiologies, but all result in a persistently elevated blood glucose concentration as a consequence of insufficient insulin release by pancreatic b-cells 1 . In type-1 diabetes, the insulinproducing b-cells are destroyed and patients require lifelong treatment with exogenous insulin. In type-2 diabetes mellitus (T2DM), the b-cells largely remain, but they fail to release sufficient insulin to maintain normoglycaemia. Both reduced b-cell mass and impaired b-cell function have been proposed to underlie the defective insulin secretion, but their relative contribution is debated [2][3][4] . There are also a number of rare monogenic forms of diabetes that present either at birth (neonatal diabetes) or in young adult life (maturity onset diabetes of the young) 1 . The primary problem in almost all of these genetic disorders is insufficient insulin secretion. In addition, glucagon secretion from pancreatic a-cells is usually disturbed.
The extent to which hyperglycaemia per se underlies, or exacerbates, alterations in b-cell and a-cell function and/or mass in diabetes remains poorly understood. Numerous studies have examined the effect of hyperglycaemia on isolated islets, b-cells or b-cell lines 5 . These have concluded that culture in high glucose (420 mM) for several days leads to a reduction in insulin content, impaired insulin secretion and multiple changes in gene expression. However, such studies have the limitation that changes in gene expression may occur as a consequence of in vitro culture. The effects of hyperglycaemia have also been explored using various mouse models of diabetes 6 . Most of these, however, suffer from the disadvantage that the insulin secretory defect cannot be reversed, either because the diabetes is genetic or is artificially induced by b-cell ablation. Furthermore, the diabetogenic gene(s) may be unknown, is usually expressed in all tissues and may be associated with insulin resistance and/or obesity. This makes dissecting the effect of hyperglycaemia per se on pancreatic islet cells in vivo difficult. In this respect, a means of selectively and reversibly switching off insulin secretion would be advantageous. One way to do so is by manipulating ATP-sensitive K þ (K ATP ) channel activity.
The prominent role of the K ATP channel in regulating glucosestimulated insulin secretion is well established 7,8 . These channels are regulated by metabolically generated ATP and thereby link changes in blood glucose concentration to insulin release. When K ATP channels are open, as at low plasma glucose levels, Ca 2 þ influx and insulin secretion are prevented, switching off insulin release. A rise in blood glucose elevates intracellular ATP, closing K ATP channels and leading to membrane depolarization, b-cell electrical activity, Ca 2 þ influx and insulin release. The crucial role of K ATP channel function in controlling blood glucose is illustrated by the fact that mutations in K ATP channel genes can cause aberrant insulin secretion. For example, gain-of-function mutations in either the Kir6.2 or SUR1 subunits of the K ATP channel are a common cause of neonatal diabetes, a rare genetic form of the disease that presents within the first 6 months of life [9][10][11] . All these mutations impair the ability of metabolically generated ATP to close the channel, and thereby prevent insulin secretion. Mice that express activating K ATP channel mutations selectively in their pancreatic b-cells recapitulate many of the characteristics of neonatal diabetes [12][13][14] : in particular, they display hyperglycaemia and hypoinsulinaemia in the absence of obesity and insulin resistance.
Here we use a transgenic mouse that expresses a human neonatal diabetes mutation (Kir6.2-V59M) specifically in b-cells 12 to investigate the effects of chronic hyperglycaemia on islet cell structure and function. In this mouse model, insulin secretion is rapidly switched off following K ATP channel activation but can be restored by treatment with sulphonylurea (SU) drugs, which are specific K ATP channel blockers. This allowed us to assess the effects of reversible hyperglycaemia on pancreatic islet cells. Our results suggest that many of the changes in islet structure and function associated with diabetes are attributable to hyperglycaemia, and can be reversed by normalization of blood glucose. These data have implications for understanding the aetiology and treatment of diabetes.

Results
Expression of Kir6.2-V59M in b-cells rapidly induces diabetes. We generated an inducible mouse model selectively expressing a gain-of-function K ATP channel mutation (Kir6.2-V59M) in pancreatic b-cells (bV59M mice). Expression was induced at 12 weeks of age by tamoxifen injection. This resulted in a rapid rise in the blood glucose concentration of free-fed mice that exceeded 20 mM within 2 days and was sustained throughout the next 4 weeks (Fig. 1a). Free-fed ( Supplementary Fig. 1a) and fasted ( Supplementary Fig. 1b) plasma insulin levels were significantly reduced (B60%) by Kir6.2-V59M gene expression. Plasma glucagon levels did not change ( Supplementary Fig. 1c,d).
We examined the effects of insulin or glibenclamide therapy using slow-release pellets implanted subcutaneously once blood glucose had risen above 20 mM. Free-fed blood glucose levels were not significantly different in bV59M mice implanted with a placebo pellet (containing no drug) from those that did not receive a pellet (Fig. 1a).   Implantation of an insulin pellet (releasing 0.2-0.3U per day) caused a rapid fall in blood glucose from 20.3 ± 1.4 mM (n ¼ 6) on the day of implantation to 8.2±3.2 mM 2 days later (Fig. 1b). However, insulin therapy failed to maintain normoglycaemia consistently throughout the 4-week period, and sporadic episodes of hypoglycaemia (o2 mM) and hyperglycaemia (420 mM) were common.
Glycaemic control was greatly improved in mice treated with glibenclamide pellets (Fig. 1c). However, the dose needed to control the blood glucose level varied with the duration of hyperglycaemia. When glibenclamide was administered immediately following a rise in blood glucose to 420 mM, a dose of 17 mg kg À 1 per day was sufficient to produce normoglycaemia (Fig. 1c), whereas a dose of 34-95 mg kg À 1 per day was necessary to normalize glycaemia in mice that had been diabetic for 4 weeks (Fig. 1d).
Glibenclamide therapy caused a rapid fall in free-fed blood glucose: 2 days after drug implantation, glucose concentrations were 4.6 ± 0.2 mM (n ¼ 19; Fig. 1c) in mice previously exposed to 24 h of hyperglycaemia and 4.6±0.4 mM (n ¼ 6; Fig. 1d) in those exposed to 4 weeks of hyperglycaemia. These values are close to those measured in the same animals before gene induction (5.6 ± 0.1 mM; n ¼ 25) or in control mice (5.9 ± 0.2 mM; n ¼ 41). Blood glucose levels were stable throughout the 4 weeks of glibenclamide treatment.
Effect of long-term hyperglycaemia on islet morphology. Hyperglycaemia for 4 weeks led to marked changes in islet morphology. There was a dramatic decrease in insulin-positive (ins þ ) cells and a concomitant increase in glucagon-positive (glu þ ) cells, which were no longer confined to the mantle but populated the core of the islet (compare Fig. 2c,d with Fig. 2a,b). Quantitative analysis revealed a marked reduction (B70%) in the area of the islet staining for insulin and an equivalent increase in the area staining for glucagon ( Fig. 3a,b). A reduction in the percentage area of the pancreas staining for insulin (a surrogate for b-cell mass) and an increase in the area staining for glucagon was also observed (Fig. 3c,d). Chronic hyperglycaemia did not produce a statistically significant change in the percentage area of the pancreas occupied by islets ( Supplementary  Fig. 2a) or in islet density (islet number per cm 2 pancreas) ( Supplementary Fig. 2b,c).
Insulin therapy largely prevented the reduction in insulin staining and increase in glucagon staining when data were expressed relative to islet area (Fig. 2e,f and Fig. 3a,b) or per cm 2 pancreas (Fig. 3c,d). Thus, the decline in ins þ cells (and corresponding increase in glu þ cells) is principally the result of hyperglycaemia per se and not K ATP channel hyperactivity.
Glibenclamide therapy also prevented the diabetes-induced changes in insulin (Fig. 2g) and glucagon ( Fig. 2h) staining, and in the area of individual islets, or whole pancreas, composed of ins þ and glu þ cells (Fig. 3a-d). Drug treatment was slightly more effective than insulin, perhaps because it produced more stable control of blood glucose (compare Fig. 1b and c).
Remarkably, 4 weeks of glibenclamide therapy almost fully reversed the histological changes produced by 4 weeks of hyperglycaemia. Insulin staining was observed throughout the islet and glucagon staining was once more confined to the mantle (Fig. 2i,j). The islet area (Fig. 3a,b) and pancreas area (Fig. 3c,d) occupied by ins þ or glu þ cells were restored to levels similar to those found in control mice.
Consistent with the decrease in the percentage of the islet area showing insulin immunoreactivity, and the concomitant increase in glucagon immunoreactivity, insulin messenger RNA and insulin content were reduced (Fig. 3e,g), and preproglucagon mRNA and glucagon content were enhanced (Fig. 3f,h), in islets isolated from mice that had been hyperglycaemic for 4 weeks, when compared with control animals.
Effects of hyperglycaemia on islet cell ultrastructure. Exposure to chronic diabetes resulted in striking changes in b-cell ultrastructure. Compared with control littermates (Fig. 4a), the number of insulin granules was considerably reduced in mice exposed to diabetes for 4 weeks (Fig. 4b). Surprisingly, in many b-cells, large areas of cytoplasm were filled with a homogeneous unstructured substance that displaced the intracellular organelles. The shape of the nucleus was occasionally distorted, but there was no indication of any apoptotic changes. No morphological signs of cellular stress were detected: the mitochondria and endoplasmic reticulum were not swollen and appeared normal. Furthermore, transcript levels of the endoplasmic reticulum (ER) stress markers Chop and spliced Xbp1 were unaltered in islets isolated from 4-week diabetic mice compared with controls ( Supplementary Fig. 3a). The ultrastructural changes were specific to b-cells, and not observed in adjacent a-cells or d-cells ( Supplementary Fig. 3b).
No effect on b-cell ultrastructure was observed following Kir6.2-V59M gene induction and only 24 h of hyperglycaemia and b-cells remained well granulated ( Supplementary Fig. 3c). This supports the view that K ATP channel activation acutely inhibits insulin release by preventing b-cell electrical activity 12,14 rather by affecting insulin content.
Immediate insulin therapy prevented the ultrastuctural changes associated with diabetes ( Fig. 4c). Even more remarkably, the ultrastructural changes associated with 4 weeks of diabetes were completely reversed after 4 weeks of glibenclamide therapy (Fig. 4d). In glibenclamide-treated islets, the majority of b-cells were densely packed with insulin granules, which had a size and morphology similar to those observed in control animals. Taken together, the data indicate that b-cell ultrastructural changes are due to hyperglycaemia and/or hypoinsulinaemia andimportantly-are reversible on restoration of euglycaemia. Representative electron micrographs of pancreatic sections from control mice (a), bV59M mice exposed to hyperglycaemia for 4 weeks (b), bV59M mice treated with insulin for 4 weeks (c) and bV59M mice that were hyperglycaemic for 4 weeks and then treated with glibenclamide for 4 weeks (d). Scale bar, 2 mm (refers to all panels). Images are representative of: a, 9 mice/9 islets/150 b-cells; b, 3 mice/9 islets/134 b-cells; c, 2 mice/3 islets/100 b-cells; d, 3 mice/3 islets/105 b-cells.
Effect of chronic hyperglycaemia on islet cell identity. We next investigated the mechanism(s) underlying the alterations in islet cell composition in chronically diabetic bV59M mice. We first explored whether they were caused by changes in cell turnover. Because of marked b-cell degranulation in bV59M mice, we used electron microscopy to identify b-cells unequivocally and nuclear morphology (condensed chromatin) as a measure of apoptosis. We found 3.7% of b-cells (that is, cells containing typical insulin granules) from 4-week diabetic bV59M mice had apoptotic nuclei ( Supplementary Fig. 3d); similar numbers were found in control islets (4.0%). Cell proliferation, determined by Ki67 positivity, was also unchanged in glu þ cells ( Supplementary Fig. 4a,b). Thus, the changes we observe are unlikely to be due to b-cell death or a-cell proliferation.
We also examined transcript levels of green fluorescent protein (GFP), which is located downstream of Kir6.2-V59M in the transgene cassette and thus serves as a marker of gene induction 12 . A 430-fold increase in islet GFP mRNA was observed 24 h after establishment of a free-fed blood glucose of 420 mM by gene induction (Fig. 5a). Islet GFP mRNA levels were unchanged after 4 weeks of chronic hyperglycaemia (Fig. 5a), despite the reduction in islet insulin immunostaining (Fig. 2), insulin mRNA (B50%, Fig. 3e) and insulin content (B70%, Fig. 3g). This provides further support for the idea that the reduction in ins þ cells we observe is not due to b-cell death.
Chronic hyperglycaemia was associated with a B20-fold increase in cells positive for both insulin and glucagon (ins þ /glu þ cells) in islets from 4-week-diabetic bV59M mice (Fig. 5b,c). However, in mice treated with insulin for 4 weeks immediately after establishment of diabetes, or treated with glibenclamide following 4 weeks of hyperglycaemia, the number of ins þ /glu þ cells was not significantly different from control animals (Fig. 5b). This argues the increase in ins þ /glu þ cells is caused by hyperglycaemia/hypoinsulinaemia and is reversed when blood glucose is normalized.
We used lineage tracing to test whether b-cells start to express glucagon when exposed to chronic hyperglycaemia. Although GFP was located downstream of Kir6.2-V59M in the transgene cassette, expression was too weak to be visualized by immunofluorescence. Thus, a Rosa26 RFP reporter was crossed into RIPII-CreER mice with (bV59M-RFP mice) or without (b-RFP control mice) the floxed Kir6.2-V59M transgene, to label all b-cells and their progeny with red fluorescent protein (RFP) after tamoxifen injection ( Fig. 6a and Supplementary Fig. 5a,b).
Although the a-cell marker MafB appeared elevated, Arx and Pax6 mRNA levels were unchanged (Fig. 7e).   In contrast to what was observed in islets, mRNA levels of Nkx6.1, MafA and Glut2 were unaltered in FAC-sorted RFP þ b-cells (Fig. 7f). The difference between the effect of diabetes on mRNA expression in islets and FAC-purified RFP þ b-cells is probably due to loss of a specific b-cell subpopulation during FAC sorting: B30% fewer RFP þ b-cells were FAC sorted from 4-week diabetic mice than from b-RFP mice. It is possible that the absent cells represent b-cells with large areas of unstructured cytoplasm (identified in electron microscopy), which may be more fragile and thus do not survive FAC sorting.
Expression of the a-cell transcription factors MafA and Arx was increased in RFP þ FAC-sorted b-cells (Fig. 7f), which explains the elevated glucagon content. Taken together, the data confirm that cells with a b-cell lineage express glucagon in response to chronic hyperglycaemia by increasing expression of a-cell transcription factors. In addition, the islet progenitor cell marker Ngn3 was elevated in both islets (Supplementary Fig. 6b) and FAC-sorted RFP þ b-cells from 4-week-diabetic bV59M mice (Supplementary Fig. 6c).
We also assessed the electrophysiological fingerprint of islet cells from control and bV59M mice following 4 weeks of diabetes. All cells were identified by immunolabelling for insulin and glucagon after patch clamping. In bV59M mice, the Cre-lox approach restricts expression of the mutant Kir6.2 subunit to cells expressing insulin at the time of induction. As the Kir6.2-V59M mutation markedly decreases K ATP channel inhibition by ATP, and thereby increases current amplitude 12,15 , large amplitude K ATP currents also serve as a lineage marker for b-cells. Cellattached K ATP currents from control b-cells were small (5±2 pA; n ¼ 27), due to inhibition by intracellular ATP. In contrast, ins þ cells from bV59M mice had B10-fold larger currents (49±6 pA; n ¼ 48). Large on-cell K ATP currents were also recorded from ins þ /glu þ cells, and even from a number of cells that expressed glucagon alone (Fig. 8a,b).
Marked differences in the voltage-dependence of Na þ -current inactivation were observed between control mouse band a-cells in intact freshly isolated islets, with half-maximal inactivation (V 0.5 ) occurring at À 103 ± 1 mV (n ¼ 5) and À 75 ± 1 mV  (Fig. 8c). In ins þ /glu þ cells (Fig. 8c,d), V 0.5 was À 117±2 mV (n ¼ 6; also measured in acutely isolated intact islets): this is similar to control b-cells and provides further evidence that ins þ /glu þ cells derive from b-cells and retain many b-cell properties.

Discussion
Our data demonstrate that expression of a human activating K ATP channel mutation in adult mouse b-cells leads to rapid diabetes and marked alterations in islet morphology and ultrastructure. Following 4 weeks of diabetes, islet insulin mRNA and protein were dramatically reduced and glucagon mRNA and protein increased. Importantly, the reduced insulin staining seen with light microscopy did not equate with loss of b-cells, as evident from electron microscopy. Some b-cells now expressed glucagon and several a-cell transcription factors. However, they retained b-cell characteristic proteins such as Glut2 and b-cell voltagegated Na þ channels. We show that hyperglycaemia, rather than K ATP channel activation per se, accounts for these changes, as they can be prevented by insulin therapy. Strikingly, the effects of chronic hyperglycaemia were also reversed when blood glucose was normalized with the SU glibenclamide.
K ATP channel activation led to a rapid and sustained rise in blood glucose on gene induction. Treatment of bV59M mice with either insulin or glibenclamide normalized blood glucose within 24 h of implementing therapy. Glibenclamide produced far more stable control of plasma glucose than insulin (as it does in patients with neonatal diabetes 16 ). This may be because SUs exert their effect indirectly by stimulating insulin release from pancreatic b-cells 17 . K ATP channel closure by SUs enables additional mechanisms that enhance meal-induced insulin secretion, such as the stimulatory effects of incretins (triggered by the presence of food in the gut 18 ) and the non-K ATPdependent actions of glucose 19 . This should help reduce plasma glucose fluctuations by directly linking insulin release to plasma glucose concentrations and food ingestion. In contrast, insulin therapy results in large spikes in blood glucose (as also seen in bV59M mice), because an insulin dose sufficient to fully control meal-stimulated spikes in plasma glucose cannot be used, as it would produce hypoglycaemia at fasting plasma glucose levels.
Surprisingly, glibenclamide therapy rapidly normalized plasma glucose levels, even after 4 weeks of diabetes. This agrees with clinical data: many patients with neonatal diabetes have been able to transfer to SU therapy, albeit with a very high dose, even after years of insulin treatment 18,20 . Similar to patients with many years of poor glycaemic control, bV59M mice required higher drug doses following 4 weeks, rather than 2 days, of hyperglycaemia. This may explain previous reports that glibenclamide was unable to reverse established diabetes in mice with a different activating K ATP channel mutation, which used a lower drug dose 14,21 .
Dramatic changes in insulin content, islet morphology and b-cell ultrastructure were evident after 4 weeks of diabetes. There was a marked reduction in islet cells staining for insulin, consistent with the fall in islet insulin mRNA and protein levels. Nevertheless, our data indicate that the loss of insulin immunolabelling did not equate to loss of b-cell mass. This is clear from the fact that many b-cells, albeit with very few insulin granules, are observed at the ultrastructural level. The nuclear morphology of these b-cells also confirms they are not undergoing apoptosis. Furthermore, islet GFP mRNA levels do not decrease with diabetes duration, as would be expected if b-cells die (GFP serves as a b-cell lineage marker in bV59M mice). We also found no change in apoptosis. Thus, our data argue that the marked decrease in insulin staining we observe by immunohistochemistry is not primarily due to b-cell loss but rather to a dramatic reduction in insulin gene expression and insulin granule density. The idea that reduced insulin content can give rise to the fallacious impression of b-cell loss has also been suggested for islets from patients with type T2D (ref. 22) and rodent models of diabetes 23 .
In our mouse model, chronic hyperglycaemia not only led to a remarkable reduction in insulin granule content but also to the appearance of large areas of unstructured cytoplasm within b-cells. We saw no morphological evidence of increased numbers of lysosomes or autophagic bodies, and these cells did not show any morphological signs of apoptotic death or ER stress. Furthermore, the structural changes were fully reversible on administration of glibenclamide; hence, they are not a sign of permanent b-cell damage. Such dramatic changes in b-cell ultrastructure have not been reported previously, either in mouse models of diabetes or in islets from patients with T2DM. Thus, it is possible they arise from a combination of hyperglycaemia/ hypoinsulinaemia and K ATP channel activation.
Chronic hyperglycaemia was also associated with a marked increase in preproglucagon mRNA and glucagon content, as well as glucagon immunostaining, in b-cells. However, the increase in glucagon immunostaining was significantly less than the reduction in insulin immunostaining, when normalized to pancreatic area. This is consistent with the presence of many sparsely granulated b-cells that do not contain sufficient insulin for detection by conventional immunostaining. The low levels of glucagon in b-cells (average B0.02 pg/RFP þ cell) compared with pre-existing a-cells (B1.5 pg/RFP À cell) ( Fig. 6d and Supplementary Fig. 5d) may explain why there was no obvious change in plasma glucagon despite the increase in glu þ cells and islet glucagon content. Furthermore, glucagon expressed by b-cells is unlikely to be released due to the hyperpolarizing effect of the K ATP channel mutation.
Lineage tracing and electrophysiological recordings confirmed that ins þ /glu þ cells and a number of cells expressing glucagon alone were of b-cell origin, as they expressed RFP and displayed very large K ATP currents (indicating they carried the Kir6.2-V59M mutation). These cells also retained several b-cell functional characteristics; they expressed the b-cell transcription factor Pdx-1, the b-cell glucose transporter Glut2 and the voltagedependence of Na þ current inactivation was characteristic of b-cell, not a-cell, Na þ channels 24,25 . However, the a-cell transcription factors MafB and Arx were upregulated in RFP þ b-cells, which probably explains their increased glucagon expression 26,27 .
In previous studies, insulin and glucagon content, and/or islet cell transcription factors have commonly been used to define b-cell or a-cell identity. What constitutes a b-cell may be a matter of semantics; nevertheless, our results raise the question of whether transcription factor expression and insulin/glucagon content are sufficient to define b-cell identity. In our mouse model, b-cells (identified by lineage tracing) showed lower insulin and increased glucagon content following 4 weeks of diabetes. However, they maintained a partial b-cell identity; they expressed Pdx1 and Glut2, and had Na þ channels characteristic of b-cells (not a-cells).
Several studies have highlighted the remarkable plasticity of islet cells. b-Cells can transdifferentiate into a-cells 23,[27][28][29][30] . Similarly, conversion of a-cells into b-cells following near-total b-cell ablation has been observed 31 and alterations in the expression and activity of key b-cell transcription factors, including Nkx6. 1 (refs 32-35), FoxO1 (ref. 23) and Pdx-1 (ref. 35), are commonly observed following dedifferentiation or transdifferentiation of islet cells. Chronic hyperglycaemia in both diabetic rodent models and human type 2 diabetic islets can also lead to changes in transcription factor expression and the appearance of ins þ /glu þ cells 23,31,[33][34][35][36] . Our data consolidate these findings by showing that hyperglycaemia per se can drive alterations in islet cell identity.
A recent study of mice with an activating K ATP channel mutation (DN30-K185E) also found that hyperglycaemia led to a dramatic loss of insulin content without marked changes in islet cell death or proliferation 37 . They attributed this to b-cell dedifferentiation due to upregulation of the islet progenitor marker Ngn3. Although we found a similar increase in Ngn3 mRNA, we were unable to observe significant protein expression. Wang et al. 37 also reported a marked increase in glucagon immunostaining but found few glu þ cells that were of b-cell lineage. One possible reason for these differences may be that bcells from DN30-K185E mice might have dedifferentiated further than those of bV59M mice, perhaps because of the higher blood glucose level of DN30-K185E mice (433 versus 25 mM). How the severity and duration of hyperglycaemia affects b-cell plasticity is currently unknown. Thus, we cannot exclude the possibility that the changes we observe reflect an early intermediary state in b-cell dedifferentiation or transdifferentiation to a-cells. However, at this stage, their glucose transporters and Na þ channels are more characteristic of b-cells than a-cells or islet progenitor cells.
Our results suggest many of the changes we observed in islet morphology and ultrastructure can be attributed to hyperglycaemia rather than K ATP channel activation. Insulin therapy, which normalizes blood glucose without affecting the open K ATP channel, prevented the changes in insulin-and glucagon-staining, b-cell ultrastructure and number of ins þ /glu þ cells produced by 4 weeks of diabetes. Furthermore, normalization of blood glucose by glibenclamide following 4 weeks of diabetes reversed these changes, suggesting that b-cell functional mass can be immediately restored when blood glucose is controlled. This may explain why glucoseinduced insulin secretion is reinstated in patients with T2DM given a low-calorie diet 38 . Although we cannot unequivocally distinguish between the effects of hyperglycaemia and hypoinsulinaemia, the fact that deletion of the insulin receptor in b-cells 39 does not alter their ultrastructure, or the relative proportions of ins þ and glu þ cells, favours the former possibility.
In conclusion, we show that chronic hyperglycaemia results in the partial loss of b-cell identity in a mouse model of b-cell dysfunction. Strikingly, these alterations are prevented and reversed when blood glucose is normalized with anti-diabetic drugs. This highlights the remarkable plasticity of b-cells and their ability to reversibly alter their gene expression, structure and function in response to changes in circulating glucose levels. Our results also highlight the importance of good glucose control in patients with diabetes and indicate that this should help preserve b-cell structure and function, and may even reverse established changes.

Methods
Generation of bV59M mice. All experiments were conducted in accordance with the UK Animals Scientific Procedures Act (1986) and have been approved by the University of Oxford DPAG local ethical committee. Mice hemizygously expressing Kir6.2-V59M in insulin-secreting cells (bV59M mice) were generated using a Cre-lox approach essentially as described 12 but using an inducible rat insulin promoter II (RIPII-Cre-ER mice 40 ). Kir6.2-V59M expression was induced in 12to 14-week-old male and female mice by a single subcutaneous injection of 0.4 ml of 20 mg ml À 1 tamoxifen in corn oil (Sigma). All experiments were performed in 12-to 14-week-old mice with a mixed (C3H, C57BL/6, 129/sv) genetic background. Wild-type mice, RIPII-Cre-ER mice and mice expressing only the floxed Kir6.2-V59M gene (ROSA) were used as controls. Rosa RFP þ / À mice were produced as described 41 .
Free-fed plasma glucose levels were measured daily (at 1400 h; FreeStyle Lite Blood Glucose Monitoring System, Abbott). Insulin and glucagon were measured in plasma or whole islets by radioimmunoassay (Millipore and Euro Diagnostica, respectively).
Electrophysiology. Islets were isolated and dispersed into single cells, as described 12 . Single cells were cultured in 11 mM glucose and patched within 2 days of isolation. Na þ currents were recorded from cells within freshly isolated intact islets using the standard whole-cell configuration as described 44 and following infusion with biocytin, cell identity confirmed by immunocytochemistry 45 . Macroscopic K ATP currents were recorded from cell-attached membrane patches on isolated cells at À 60 mV, filtered at 5 kHz and digitized at 20 kHz 12 . Cell identity was subsequently confirmed by immunocytochemistry.
Electron microscopy. Isolated islets were fixed in 4% paraformaldehyde and 0.5% glutaraldehyde in phosphate buffer for 1 h, post-fixed in 1% osmium tetroxide, block stained in 2% uranyl actetate, dehydrated in graded ethanol and embedded in Spurr's resin (Agar Scientific, Stansted, UK). Ultrathin sections (70 nm) were cut onto Ni 2 þ grids, contrasted with 2% uranyl acetate and lead citrate, and examined in a Jeol 1010 microscope (Welwyn Garden City, UK) with an accelerating voltage of 80 kV.
b-Cell apoptosis was assessed from electron micrographs as the number of b-cells with apoptotic nuclei (that is, displaying typical chromatin condensation): b-cells (135-150 from n ¼ 3-9 mice; n ¼ 3 islets per mouse) were identified by the presence of characteristic insulin granules.
Immunostaining and morphometric analysis. Immunostaining was performed as described on fixed wax-embedded pancreatic sections 12 and dispersed islet cells were fixed within 2 h of isolation. For antibodies see Supplementary Table 2. Insulin and glucagon area was analysed using the manual histology tool in Zeiss LSM-510 software. Five pairs of 5-mm-thick serial sections, 100 mm apart were analysed, for n ¼ 3-6 animals of each genotype (B160 islets/mouse).
Flow cytometry. Pancreatic islet cell suspensions from b-RFP and bV59M-RFP mice were separated by FAC sorting using a MoFlo Beckman Coulter Cytomation sorter (Coulter Corp., Hialeah, FL). Islet populations were sorted immediately following isolation into RNAse-free collection buffer for qPCR analysis as described previously 45 . For Taqman probes, see Supplementary Table 3.