Glucose-Induced O2 Consumption Activates Hypoxia Inducible Factors 1 and 2 in Rat Insulin-Secreting Pancreatic Beta-Cells

Background Glucose increases the expression of glycolytic enzymes and other hypoxia-response genes in pancreatic beta-cells. Here, we tested whether this effect results from the activation of Hypoxia-Inducible-factors (HIF) 1 and 2 in a hypoxia-dependent manner. Methodology/Principal Findings Isolated rat islets and insulin-secreting INS-1E cells were stimulated with nutrients at various pO2 values or treated with the HIF activator CoCl2. HIF-target gene mRNA levels and HIF subunit protein levels were measured by real-time RT-PCR, Western Blot and immunohistochemistry. The formation of pimonidazole-protein adducts was used as an indicator of hypoxia. In INS-1E and islet beta-cells, glucose concentration-dependently stimulated formation of pimonidazole-protein adducts, HIF1 and HIF2 nuclear expression and HIF-target gene mRNA levels to a lesser extent than CoCl2 or a four-fold reduction in pO2. Islets also showed signs of HIF activation in diabetic Leprdb/db but not non-diabetic Leprdb/+ mice. In vitro, these glucose effects were reproduced by nutrient secretagogues that bypass glycolysis, and were inhibited by a three-fold increase in pO2 or by inhibitors of Ca2+ influx and insulin secretion. In INS-1E cells, small interfering RNA-mediated knockdown of Hif1α and Hif2α, alone or in combination, indicated that the stimulation of glycolytic enzyme mRNA levels depended on both HIF isoforms while the vasodilating peptide adrenomedullin was a HIF2-specific target gene. Conclusions/Significance Glucose-induced O2 consumption creates an intracellular hypoxia that activates HIF1 and HIF2 in rat beta-cells, and this glucose effect contributes, together with the activation of other transcription factors, to the glucose stimulation of expression of some glycolytic enzymes and other hypoxia response genes.


Introduction
Hypoxia-Inducible-Factors (HIFs) are basic helix-loop-helix-PAS domain transcription factors composed of a regulated a subunit (HIF1a or HIF2a) and a constitutively expressed HIF1b subunit (Aryl-hydrocarbon-Receptor Nuclear Translocator (ARNT)) [1,2]. Under normoxic conditions, Prolyl-Hydroxylase-Domain proteins (PHD1-3) hydroxylate HIFa subunits on proline residues in an O 2 -, Fe 2+ -and a-ketoglutarate-dependent manner. This hydroxylation promotes HIFa binding to von Hippel-Lindau protein, followed by their polyubiquitylation and proteasomal degradation. Under hypoxic conditions (O 2 partial pressure (pO 2 ),2.3-38 mmHg) or after inhibition of PHDs with CoCl 2 , HIFa subunits are no longer degraded and translocate with ARNT to the nucleus where they activate the transcription of HIF-target genes including glucose transporter 1 (Glut1), glycolytic enzymes, monocarboxylate transporter 4 (Mct4), the vasodilating peptide adrenomedullin (Adm), vascular endothelial growth factors (Vegfs), and erythropoietin (Epo). This response favours cell survival by triggering a switch from aerobic mitochondrial to anaerobic glycolytic ATP production at the cellular level, an increase in blood flow and capillary growth at the organ level, and an increase in O 2 transport capacity at the organism level [1][2][3].
The glucose stimulation of insulin secretion (GSIS) by pancreatic beta-cells critically depends on the acceleration of glycolysis and mitochondrial Krebs cycle, with consequent increases in NAD(P)H and ATP production as well as export of Krebs cycle intermediates, including a-ketoglutarate, to the cytosol [4]. Subsequent plasma membrane depolarization and Ca 2+ influx through voltage-dependent-Ca 2+ -channels trigger insulin granule exocytosis [5]. In addition, glucose stimulates various ATP-consuming processes such as gene transcription, protein synthesis, and Ca 2+ pumping [6]. In beta-cells, glucoseinduced acceleration of ATP production is coupled to an increase in mitochondrial O 2 consumption [7][8][9]. In vivo, the concomitant increase in islet blood flow prevents the fall in intra-islet pO 2 [10,11]. In isolated islets maintained in vitro or transplanted in vivo, however, the glucose stimulation of betacell O 2 consumption leads to a reduction in intra-islet pO 2 [12][13][14], of which approximately one third depends on the stimulation of Ca 2+ influx [15]. However, as a drop in pO 2 and an increase in a-ketoglutarate exert opposite effects on PHD-mediated HIFa hydroxylation [2], it remains unclear whether glucose eventually activates HIFs in beta-cells and, if so, to what extent such activation contributes to the glucose regulation of islet gene expression. In this context, it has recently been shown that glucose activates HIF1 in MIN6 cells and mouse islets only if cultured under slightly hypoxic conditions [16].
Others and we have previously shown that islet expression of hexokinase (Hk) 1, lactate dehydrogenase A (Ldha), Mct1 and 4 and Hypoxia up-regulated 1 (Hyou1) is increased in hyperglycemic rats [17][18][19][20]. We more recently reported that the glucose stimulation of cultured rat islets increases their mRNA levels of most glycolytic enzymes (except glucokinase (GK)), of other HIFtarget genes like Adm, and of genes that are induced by hypoxia independently from HIF activation, like Hyou1 [21]. We now demonstrate that glucose activates HIF1 and HIF2 in rat beta cells and that both HIF isoforms play distinct roles in the glucose stimulation of expression of glycolytic enzymes and Adm. We also provide some evidence that HIFs are activated in islets from diabetic mice, suggesting that hyperglycaemia could induce betacell hypoxia in vivo.

Effects of glucose on HIF-target gene mRNA levels in cultured rat islets and INS-1E cells
To characterize the role of HIF in the glucose stimulation of islet gene expression, we first tested the effect of a 18 h culture in the presence of 2, 5, 10 or 30 mmol/l glucose (G2, G5, G10, or G30) on the mRNA levels of known HIF-target genes and compared it with the effect of HIF activation by CoCl 2 , hypoxia or knockout of vhlh, the gene coding the von Hippel-Lindau protein [22].
Due to limited O 2 diffusion, large islets frequently suffer from central necrosis under normoxic culture conditions [14,23]. Therefore, islets with central necrosis (usually with a diameter .150 mm) were systematically discarded during preculture, and the islet density per cm 2 and the medium depth were kept constant between groups. Under these conditions, glucose significantly increased the mRNA levels of many (but not all) HIF-target genes  that were up-regulated by more than 2-fold in vhlh-knockout islets [22], including Glut1, most glycolytic enzymes, Mct4, pyruvate dehydrogenase kinase 1 (Pdk1), Adm and carbonic anhydrase 12 (Car12) ( Table 1 and Table S1). This effect was larger for genes expressed at low levels under control conditions (Ldha, Adm and Car12) than for the highly expressed genes Gapdh and aldolase A (Aldoa). Glucose also increased the mRNA levels of Gapdh, triose phosphate isomerase 1 (Tpi1) and Adm (but not Ldha that remained below detection limit) in INS-1E cells cultured for 18 h at 70% confluence, indicating that the glucose stimulation of HIF-target gene expression was not restricted to devascularized islets ( Table 2). In contrast, glucose failed to affect the islet expression of HIFtarget genes that were not or only slightly increased in vhlh-KO islets, like hexokinase 1 and 2 (Hk1 and Hk2), and vascular endothelial growth factors (Vegf). (Table S1). As expected, the mRNA levels of Gapdh, Tpi, Ldha, Mct4, Pdk1 and Adm were significantly increased in CoCl 2 -treated rat islets or in islets exposed overnight to hypoxia (pO 2 ,38 mmHg) ( Table 3), a condition under which all islets developed central necrosis. In contrast, CoCl 2 and hypoxia did not affect the mRNA levels of genes that are markedly induced by glucose in rat islets but are not HIF-target genes, like thioredoxin interacting protein (Txnip) and aldolase B (Aldob) (data not shown).

Effects of glucose on the expression of HIF subunits in cultured rat islets and INS-1E cells
To characterize the role of HIF in the glucose stimulation of islet gene expression, we next compared the effects of glucose, CoCl 2 and hypoxia on the expression of components of the HIF signalling pathway. As shown in Table 1 and Table S1, the mRNA coding the main HIF subunits, HIF-regulating and HIFinteracting proteins were detected in rat islets, some of them being significantly affected by glucose. Most noticeably, glucose (between G2 and G30) decreased Hif1a mRNA levels by 60% and Arnt mRNA levels by 40% while increasing Hif2a mRNA levels 2fold. Glucose similarly affected Hif subunits mRNA levels in INS-1E cells, except that Hif2a mRNA levels increased at lower glucose concentrations than in rat islets and tended to decrease between G10 and G30 (Table 2). Interestingly, CoCl 2 but not hypoxia also decreased Hif1a mRNA levels, while both treatments increased Hif2a mRNA levels in rat islets (Table 3).
Because HIF activation mainly results from the stabilization of its alpha subunits and their nuclear translocation with ARNT, we next tested the effect of glucose, hypoxia and CoCl 2 on HIF1a and ARNT protein levels in cultured rat islets by immunohistochemistry. After culture in G5, HIF1a was only detected in the nuclei of a few insulin-negative cells while ARNT was detected in the cytosol and nuclei of most islet cells (Figure 1 and Figure S1). After culture in G30, HIF1a staining increased in insulin-positive but not insulin-negative islet cells, while ARNT staining was unaffected. The increase in HIF1a staining was heterogeneous between beta-cells ( Figure 1). In comparison, hypoxia and CoCl 2 markedly increased HIF1a staining in most islet cells outside the central necrotic area. Exposure to CoCl 2 also tended to increase the intensity of ARNT staining in islet cell nuclei ( Figure S1).
Glucose also increased HIF nuclear levels in INS-1E cells ( Figure 2). Thus, compared with G2, culture in G30 induced a 4fold increase in HIF1a and HIF2a nuclear levels and a 2-fold increase in ARNT. Glucose also decreased cytosolic ARNT levels by ,50%. These glucose effects were, however, of smaller amplitude than those of CoCl 2 ( Figure 2B). These results indicate that, upon glucose stimulation, HIF1a and HIF2a translocate with their dimerization partner ARNT to beta-cell nuclei.

Effects of Hif1a and Hif2a knockdown on the expression of glycolytic enzymes and Adm in INS-1E cells
The relationship between HIF1/HIF2 expression and the upregulation of their target genes was tested in INS-1E cells using small interfering RNAs (siRNAs) against Hif1a and Hif2a. We first checked the effects of selected siRNAs on the mRNA levels of Hif subunits during culture in the presence of G2 and CoCl 2 ( Figure   markedly reduced both Hif1a and Hif2a mRNA levels. Interestingly, Arnt mRNA levels were not affected by either siRNA. We next tested the effects of Hifa knockdown on CoCl 2mediated induction of Adm, Tpi and Gapdh mRNA expression ( Figure 3D-F). As expected, CoCl 2 treatment significantly increased the mRNA levels of the three HIF-target genes in INS-1E cells treated with siLuc. Under these conditions, Adm mRNA levels were unaffected by siHif1a but were inhibited by ,90% by siHif2a. In comparison, the CoCl 2 -mediated stimulation of Tpi1 and Gapdh mRNA expression was unaffected by siHif2a alone, partly reduced by siHif1a alone, and markedly reduced by their combination.
We finally tested the effects of Hif knockdown on the glucose stimulation of HIF-target gene expression ( Figure 4D-F). Of note, siHif1a significantly reduced GSIS by INS-1E cells while siHif2a tended to increase basal insulin release in G2 and G5 ( Figure 4A). Again, siHif2a, but not siHif1a, significantly reduced the glucose induction of Adm mRNA expression in INS-1E cells. In comparison, both siHif1a and siHif2a, alone or in combination, only partly reduced the glucose stimulation of Tpi1 and Gapdh mRNA expression. These results indicate that HIF1a and HIF2a are both involved but play distinct roles in CoCl 2 -and glucosemediated HIF-target gene expression in INS-1E cells.

Role of the acceleration of mitochondrial metabolism in glucose-induced HIF-target gene expression
The non-metabolised glucose analogue 3-O-methyl-D-glucopyranose did not reproduce the effect of glucose on insulin secretion, Gapdh and Adm mRNA levels ( Figure 5A). These results indicate that the stimulation of HIF-target gene expression by glucose does not result from a putative osmotic stress but rather depends on its metabolism and activation of downstream events. In agreement, succinic acid monomethyl ester and a-ketoisocaproate, two nutrient secretagogues that bypass glycolysis and directly stimulate mitochondrial metabolism in cultured rat islets, significantly augmented GSIS and the glucose stimulation of Gapdh and Adm mRNA expression ( Figure 5B). Similar results were obtained with a combination of 5 mmol/l leucine and 5 mmol/l glutamine (data not shown). These results are compatible with the hypothesis that the acceleration of mitochondrial metabolism and islet O 2 consumption with consequent reduction in intra-islet pO 2 plays a role in the glucose stimulation of HIF-target gene expression.
We therefore used pimonidazole to detect hypoxia in isolated islets cultured in the presence of increasing glucose concentrations. Under hypoxic conditions, reductively-activated pimonidazole forms protein adducts by reacting with cysteine residues independently from the pyridine nucleotide redox state [24]. As shown in figure 6A-C, glucose concentration-dependently increased pimonidazole-protein adducts in cultured islets, but to a much lesser extent than hypoxia. This increase was not restricted to the islet centre and was heterogeneous between islet cells. In comparison, hypoxia triggered central necrosis and strongly increased pimonidazole-protein adducts in surviving cells at the islet periphery.
As the glucose-stimulation of HIF-target gene expression likely results from hypoxia-mediated HIF activation, we next tested the effect of a 3-fold increase in pO 2 on the glucose stimulation of HIF-target gene expression. As shown in figure 6D, glucose stimulated insulin secretion and Gapdh mRNA expression to a similar extent under control and hyperoxic conditions. In contrast, glucose increased Tpi1, Ldha, Adm and Car12 mRNA levels to a significantly lesser extent under hyperoxic conditions, supporting the role of the increase in islet O 2 consumption in these glucose effects. Of note, the glucose stimulation of pimonidazole-adduct formation was also suppressed by culture in the presence of 90% O 2 ( Figure S2). Also in INS-1E cells, culture under hyperoxic conditions markedly reduced the glucose stimulation of pimonidazole-protein adduct formation, HIF1a and HIF2a nuclear accumulation, and Adm and Tpi1 mRNA expression ( Figure 7). It did not, however, significantly affect GSIS and the stimulation of Gapdh mRNA expression. These results indicate that, depending on the HIF-target gene studied, the glucose stimulation of mRNA expression is independent (Gapdh) or partly results from hypoxia (Tpi, Adm), not only in isolated islets, but also in INS-1E cells.

Role of Ca 2+ influx in the glucose stimulation of HIFtarget gene mRNA expression
It has previously been shown that approximately one third of the glucose stimulation of islet O 2 consumption is Ca 2+ -dependent [15]. The L-type Ca 2+ channel blocker nimodipine, which almost fully inhibited insulin secretion during culture in G30, only slightly reduced the mRNA levels of Gapdh but markedly reduced Aldoa and Adm mRNA levels in G30 ( Figure 8). Interestingly, the latter inhibition was not prevented by addition of exogenous insulin to the medium, indicating that Ca 2+ influx contributes to the stimulation of HIF-target gene expression independently from changes in insulin concentration. In contrast, nimodipine exerted opposite effects on Txnip and Aldob mRNA levels (data not shown). Similar results were obtained with diazoxide, a K ATP channel opener that inhibits glucose-induced Ca 2+ influx and insulin secretion [25] (data not shown). Also in INS-1E cells, nimodipine significantly reduced the glucose stimulation of Adm and Aldoa mRNA expression without affecting that of Gapdh ( Figure S3).

Hypoxia-mediated HIF activation in islets from diabetic mice?
To test whether in vivo hyperglycaemia also induces hypoxia and activates HIF in pancreatic islets, we first measured HIF1a protein levels in islets from diabetic Lepr db/db and non-diabetic Lepr db/+ mice. Interestingly, a few HIF1a-positive nuclei were detected in some islets from diabetic mice, whereas none were observed in sections from non-diabetic mice ( Figure 9A-H). That the lack of HIF1a staining did not result from a problem in tissue fixation/ processing was confirmed by the observation, on the same section, of a large number of HIF1a-positive epithelial nuclei in the villi of the intestinal mucosa ( Figure 9U-V). For technical reasons, we could not determine whether the few HIF1a-positive islet cells detected in Lepr db/db mice are beta-cells or not.
We next measured pimonidazole-protein adducts in the same model of diabetes. Interestingly, islets from both diabetic and nondiabetic mice were more heavily stained than the surrounding exocrine acini ( Figure 9I-T), suggesting that they experience low intensity hypoxia irrespective of the glucose tolerance status. This staining seemed, however, less intense than that observed in the villi of the duodenal mucosa ( Figure 9W-X). Although the intensity of pimonidazole staining looked similar or even slightly lower in islets from diabetic vs. non-diabetic mice, there were clear differences regarding its heterogeneity between cells. Thus, pimonidazole staining was almost uniform throughout the islets of non-diabetic mice ( Figure 9I-L), whereas it was heterogeneous in islets from diabetic mice, with a few cells displaying a higher intensity ( Figure 9M-T). Unfortunately, we could not determine whether these islet cells with higher pimonidazole staining are beta-cells or not.
We finally measured the islet mRNA levels of Hifa subunits and of several HIF-target genes in Lepr db/db and Lepr db/+ mice. As shown in Table 4, the mRNA levels of Hif1a were not different in islets from Lepr db/db and Lepr db/+ mice, but those of Hif2a and of all HIFtarget genes tested (except for Ldha tested in a previous study [26]), were significantly up-regulated in the islets of Lepr db/db mice, further supporting the hypothesis that HIF1 and HIF2 are activated in islets from diabetic mice.

Discussion
This study demonstrates that, even in small islets and cell monolayer, glucose and other nutrient secretagogues that bypass glycolysis activate HIF1 and HIF2 in rat beta-cells following the induction of a state of moderate hypoxia. This effect, which was not simply due to limited O 2 diffusion in culture but also depended on the glucose stimulation of O 2 consumption in beta-cells, contributed to the stimulation of expression of glycolytic enzymes and other hypoxia-response genes. In a recent study very similar to ours but carried out in MIN6 cells and mouse islets, glucose stimulation rapidly (within one hour) triggered beta-cell hypoxia only if the pO 2 was reduced from 20 to 10% [16]. Despite this difference, both studies converge in showing that in vitro glucose stimulation of beta-cell O 2 consumption can induce intracellular hypoxia and activate HIF. Depending on its intensity, this response could either play an important role in beta-cell adaptation to increased insulin demand under physiological conditions (physiological hypoxia) or be involved in the detrimental effect of chronic hyperglycemia.
Earlier studies have convincingly shown that glucose rapidly increases O 2 consumption in islets from rats [8,15,27], mice [12], non-human primates and humans [9]. Although we did not repeat these measurements of islet O 2 consumption or intra-islet pO 2 , we have shown that glucose increases the formation of pimonidazoleprotein adducts in islets and INS-1E cells in parallel with changes in HIF-target gene expression and HIF nuclear accumulation. Moreover, these effects were inhibited under hyperoxic conditions. Pimonidazole-protein adducts were not more abundant in the center than at the islet periphery, indicating that our results do not simply result from central necrosis. In addition, this staining was heterogeneous between individual cells, as would be expected from the metabolic heterogeneity of beta-cells [28].
The increase by glucose of HIF1a nuclear levels in rat beta-cells is in good agreement with the fact that glucose induces moderate hypoxia and with current and previous observations that hypoxia activates HIF1 in INS-1E cells and cultured islets [14,23,29]. It is also compatible with the recent report that Hif1a gene inactivation corrected the stimulation of HIF-target gene expression following vhlh inactivation in mouse beta-cells [30]. In contrast, the increase of HIF2a nuclear levels in INS-1E cells by glucose, CoCl 2 and hypoxia is surprising because HIF2a was detected neither in hypoxic embryonic pancreatic explants [31] nor in vhlh-knockout mouse islets [30]. However, the role of HIF2a in INS-1E cells is strongly supported by the observation that Hif2a expression is  , siHif1a and siHif2a). Then, the transfection medium was replaced with RPMI medium containing 10% foetal calf serum and increasing glucose concentrations. After 18 h culture, the medium was collected for insulin concentration determination (A) and cells were processed for measurement of gene mRNA levels (B-F). Gene to Tbp mRNA ratios were expressed relative to the ratio in INS1-E cells treated with siLuc and cultured in G2. Data are means 6 SEM for 3 experiments. *, p,0.05 for the effect of glucose vs. G2 and # , p,0.05 for the effect of siRNA treatment at the same glucose concentration (two-way ANOVA+test of Bonferroni). For Adm/Tbp mRNA ratio, the reduction by siHif2a in G30 was significant only after removal of siHif1a data that were highly variable. doi:10.1371/journal.pone.0029807.g004 absolutely required for the stimulation of Adm mRNA expression. Of note, glucose, CoCl 2 and hypoxia also affected Hifa mRNA levels in rat beta-cells through unknown mechanisms, but the relative contribution of these changes to the global increase in HIF activity and HIF-target gene expression has not been investigated. In any case, these changes in Hifa mRNA levels should only modulate the increase in HIFa protein levels that mainly result from their stabilization under hypoxic conditions [1].
It is well established that glucose stimulates the expression of various glycolytic enzymes in cultured insulin-secreting cells and rodent islets, including GK, GAPDH and liver pyruvate kinase [32,33]. It has been shown that the stimulation of expression of liver pyruvate kinase results from a reduction in AMPK activity [34] and from activation of the transcription factors ChREBP and c-MYC [35,36]. In contrast, the stimulation of GK expression has been ascribed to SREBP1c activation [37]. However, the transcription factors involved in the glucose-induced expression of other glycolytic enzymes in beta-cells are poorly characterized. In other tissues, HIF1a is preferentially involved in the regulation of glycolytic enzymes while HIF2a stimulates the expression of genes related to angiogenesis [38]. In INS-1E cells, knockdown of Hif1a and Hif2a had a stronger effect on the glucose stimulation of Tpi1 and Gapdh than knockdown of either isoform alone. This suggests that both HIF1 and HIF2 modulate the expression of glycolytic enzymes, at least in rat beta-cells. However, neither hyperoxia, nor Hif1a/Hif2a knockdown or inhibition of Ca 2+ influx with nimodipine were able to fully inhibit the glucose induction of glycolytic enzymes (except for the complete inhibition of Ldha mRNA by hyperoxia in whole rat islets), confirming that other transcription factors, e.g. Myc [39], are also involved. This remark is particularly important in the case of Gapdh mRNA, the glucose induction of which was unaffected by hyperoxia and nimodipine treatment. On the other hand, hyperoxia, Hif2a but not Hif1a) knockdown and nimodipine almost fully inhibited the expression of Adm, thereby demonstrating the specific role of HIF2 in Adm expression by glucose-induced decrease in islet pO 2 . Thus, both HIF isoforms are not redundant and play distinct roles in beta-cell gene expression.

Possible relevance for the physiology and pathophysiology of beta-cells
High expression of GK and downstream glycolytic enzymes is critical for GSIS [40,41]. In agreement, global down-regulation of glycolytic enzymes in Hif1a or Arnt knockout beta-cells markedly reduced GSIS and in vivo glucose tolerance in some [42,43] but not all studies [22,30]. In siRNA-treated INS-1E cells, knockdown of Hif1a markedly inhibited GSIS during culture while the inhibition of Hif2a tended to increase basal insulin release, suggesting that the absence of each isoform differently affects beta cell function. In this context, it is important to note that the moderate activation of HIF by glucose stimulation under physiological conditions, i.e. in vivo where the islet pO 2 may be lower than in in vitro culture systems, could play an important role in the maintenance of the beta-cell phenotype or in their adaptation to changes in insulin demand. Such ''physiological hypoxia'' is compatible with the fact that HIF is activated at glucose concentrations at which mitochondrial ATP production is not reduced [44]. In support of this hypothesis, it was recently shown that, in the rat, the proportion of islets showing pimonidazole-protein adduct staining in vivo is modulated by changes in insulin demand [45]. In contrast with that study, however, we did not observe major differences in pimonidazole staining between islets from the same mouse pancreatic section.
On the other hand, almost complete repression of low-K m HK (I-III), LDH and MCT (isoforms 1, 2 and 4) is also critical to prevent inappropriate stimulation of insulin secretion at low glucose and during exercise [41,46]. The lack of LDH and high expression of malate/aspartate and glycerol-phosphate shuttle enzymes also contribute to optimal GSIS by preventing pyruvate diversion from its mitochondrial metabolism while coupling reoxidation of cytosolic NADH with increased mitochondrial ATP production [47]. Following vhlh inactivation in beta-cells, sustained activation of HIF-target gene expression under normoxic conditions tended to increase insulin secretion at low glucose while reducing the maximal GSIS with consequent development of glucose intolerance [22,30,48]. Moreover, expression of a constitutively active form of HIF1a significantly decreased GSIS, O 2 consumption and pimonidazole staining in MIN6 cells [16]. Hypoxia should therefore be considered as a possible contributing factor when interpreting the in vitro effects of high glucose concentrations on beta-cell gene expression, function and survival, even in beta-cell lines. Our recent attempt to prevent beta-cell glucotoxicity by culturing islets for 1 week in the presence of 60% O 2 led to the opposite effect, i.e. a ,10-fold increase in islet cell apoptosis in G10 and a ,50-fold increase in G30 ( Figure S4). Other strategies will have to be developed to check the role of hypoxia in beta-cell glucotoxicity.

Possible relevance for the pathophysiology of type 2 diabetes
Together with alterations of islet microvasculature and fibrosis that impair vessel integrity and O 2 supply [20], the high metabolic demand imposed by hyperglycaemia may promote beta-cell hypoxia and HIF activation in vivo. So far, the available evidence suggests, but do not prove, that some beta-cells may indeed suffer from hypoxia in type 2 diabetes. Thus, HIF-target gene expression (this study and [18][19][20]) and pimonidazole-protein adducts [16] were increased in islets isolated from various rodent models of diabetes, with a few islet cells (the identity of which could not be clarified) displaying higher levels of pimonidazole-protein adducts when measured by immunohistochemistry. Moreover, HIF1apositive islet nuclei, although rare, were clearly detected on pancreatic sections of diabetic but not normoglycaemic mice. If these cells were unambiguously identified as hypoxic beta-cells, one could suggest that in vivo hypoxia, HIF activation and HIFtarget gene expression in only a fraction of beta-cells could contribute to the slow deterioration of beta-cell function and survival in type 2 diabetes. Such in vivo hypoxia could result from increased beta-cell O 2 consumption at high glucose, to a decrease in islet perfusion following changes in islet microvasculature by chronic hyperglycemia, or to both processes.
In conclusion, glucose-induced O 2 consumption creates an intracellular hypoxia that activates HIF1 and HIF2 in rat betacells, and this glucose effect contributes, together with the activation of other transcription factors, to the glucose stimulation of expression of some glycolytic enzymes and other hypoxia response genes.

Islet isolation and culture
Pancreatic islets were isolated from ,200 g male Wistar rats as described [50]. They were precultured for 1 week in serum-free RPMI 1640 medium (Invitrogen, Carlsbad, CA, USA) containing 10 mmol/l glucose, 5 g/l BSA (fraction V, Roche, Basel, Switzerland), 100 IU/ml penicillin and 100 mg/ml streptomycin (Invitrogen). Islets that developed central necrosis were discarded during preculture. After preculture, islets were cultured 18 h in the G2, G5, G10, and G30 and various test substances at different incubator pO 2 (38, 151, 456 and 680 mmHg). After culture, the medium was collected for insulin concentration determination (RIA using rat insulin as a standard) and the islets were processed for further analysis.

In vivo studies
Lepr db/db and Lepr db/+ mice on a C57BL/KSJ background were from Janvier (Le Genest-Saint-Isle, France) or from the animal facility of the Garvan Institute. They were used from 10 to 26 weeks of age. Blood glucose was measured with a glucometer (Accu-Check Sensor, Roche, Mannheim, Germany). After cervical dislocation, the duodenal loop, the pancreas and the spleen were removed as a block and fixed in less than 3 minutes. All experiments were approved by the local ethics committee for animal experimentation (Université catholique de Louvain, Faculté de Médecine, Comité d'éthique facultaire pour l'expérimentation animale, projet UCL/MD/2009/009: ''Mécanismes moléculaires de la plasticité du phénotype des cellules B pancréatiques en conditions physiopathologiques'' accepted for 4 years). ''Principles of laboratory animal care'' (NIH publication no. 85-23, revised 1985) were followed.

Cell culture and RNA interference
INS-1E cells (passage 70-94) were cultured in standard RPMI medium supplemented with 10% heat-inactivated fetal calf serum, 10 mmol/l HEPES, 1 mmol/l sodium pyruvate, 2 mmol/l glutamine, 100 U/ml penicillin, 100 mg/ml streptomycin, and 50 mmol/l b-mercaptoethanol. To downregulate gene expression, INS-1E cells were transfected for 24 h using DharmaFECT1 and 30 nmol/l siRNA in antibiotic-free medium, according to the manufacturer's instructions. Cells (,70%confluence) were then cultured for 18 h in fresh medium containing different glucose concentration and various test substances as indicated.

Real time RT-PCR
Islet and INS-1E total RNA extraction, reverse transcription, real-time PCR and melting curve analysis of PCR products were performed as described previously [50]. Relative changes in Gene to Tbp mRNA ratio between test and control conditions were computed using the 2 2DDCt method. Primer sequences and reaction conditions are detailed in Tables S2 and S3.
Nuclear extraction and western blotting for HIF1a, HIF2a and ARNT INS-1E cells were rinsed with ice-cold PBS, scraped in hypotonic buffer (20 mmol/l HEPES, 5.7 mmol/l NaF, 1 mmol/l EDTA, pH 7.5) and incubated on ice for 15 min. After addition of 0.5% Nonidet P-40, the nuclei were pelleted (8006g for 4 min at 4uC), resuspended in 50 ml Complete Lysis Buffer AM1 (Active Motif, Rixensart, Belgium) supplemented with 1 mmol/l dithiothreitol and 1% protease inhibitor cocktail (Roche Diagnostics, Mannheim, Germany), incubated for 15 min on ice with regular shaking, and centrifuged at 17506g for 10 min at 4uC. Cytosolic proteins in the supernatant were precipitated with trichloroacetic acid, triple extracted with ether, and solubilised in Laemmli buffer. Nuclear and cytosolic extracts were then separated by 7.5% SDS-PAGE and transferred to nitrocellulose membrane (Bio-Rad Laboratories, Hercules, CA, USA). The membranes were incubated with a mouse monoclonal primary antibody followed by a horse-radish-peroxidase-conjugated antimouse antibody (Santa-Cruz, CA, USA) and the signal was revealed by enhanced chemiluminescence (SuperSignalH West-Femto or West-Dura kits, Thermo Scientific). Band intensities were quantified by scanning densitometry (Gel-Doc2000, Bio-Rad), analyzed with Quantity One TM (Bio-Rad) and normalized to Red Ponceau staining or ACTIN band intensity.

Immunodetection of pimonidazole-protein adducts
Pimonidazole (Hypoxyprobe TM Inc., Burlington, MA, USA) was added to RPMI medium at a final concentration of 200 mmol/l 2 h before the end of culture or was injected intraperitoneally (60 mg/kg body weight, 34 mmol/l in NaCl 9 g/l sterile solution) 24 h before killing the mice. Tissues were fixed in 4% formaldehyde and embedded in paraffin before detection of pimonidazole-protein adducts on 5 mm-thick sections. Briefly, deparaffinized sections were treated with H 2 O 2 (0.3% vol/ vol) to inactivate endogenous peroxidase and incubated with either Hypoxyprobe 1 monoclonal antibody (clone 4.3.11.3) diluted 1:100 followed by anti-mouse EnVision+ TM peroxidase complex for 1 h (Dako, Carpintera, USA) for rat islets, or Hypoxyprobe 1 polyclonal rabbit antibody diluted 1:200 followed by anti-rabbit EnVision+ TM peroxidase complex for mouse pancreas. In both cases, the signal was revealed by 3,39-diaminobenzidine. Pimonidazole-protein adducts in cultured islets were also measured by Western Blot and the signal intensity was normalized to that of ACTIN (Ab dilution 1:2000).

Supporting Information
(DOC)