Experimental Models of Transcription Factor-Associated Maturity-Onset Diabetes of the Young

MODY3-associated mutations in the HNF-1α gene are highly penetrant and constitute the most common form of MODY in many countries (2). MODY3 patients progressively develop severe hyperglycemia (5) and may exhibit the full spectrum of complications of diabetics if not treated with either oral hypoglycemic agents or insulin (2). In fact, 5–10% of antibody-negative diabetics, previously classified as type 1, are due to mutations in HNF-1α (6). Chronic hyperglycemia in MODY3 patients is due to a primary defect in β-cells rather than in insulin action on target tissues (5,7). Although HNF-1α is expressed in both liver and pancreas, liver function is unaffected in MODY3 patients. Insulin secretion in response to glucose and arginine is severely impaired, and this β-cell defect can already be detected in nondiabetic MODY3 carriers (7–9). Mutations linked to MODY3 have been localized throughout the different functional domains of the HNF-1α protein as well as in the promoter region of the HNF-1α gene (2,10). However, the exact molecular mechanism by which these mutations impair HNF-1α function remains unclear. Both dominant-negative actions and haploinsufficiency have been suggested to participate in the development of the MODY3 phenotype (8,11,12).

Mice with targeted disruption of HNF-1α display the diabetic phenotype but also suffer from multiple organ manifestations such as renal dysfunction with massive glucosuria, pathological liver tests, and hepatomegaly (13,14). Pancreatic islets isolated from these animals exhibit attenuation not only in the generation of ATP but also in the rise of cytosolic Ca²⁺—two signals essential for glucose-stimulated insulin secretion (15,16). However, in contrast to the situation in MODY3 patients who harbor a single mutated HNF-1α allele, insulin secretion in mice heterozygous for the disrupted allele of hnf-1α is normal (14,16). This may be related to compensatory overexpression of HNF-1β in this mouse model (14). Although the HNF-1α^{−/−minus;−/−sol;−/−minus;} mice yielded important information (13,14,16,17), the pleotropic effects observed in these mice have complicated analysis of the precise role of this transcription factor in determining normal pancreatic β-cell function. To further characterize the basis of the β-cell dysfunction induced by HNF-1α deficiency, we established both in vivo and in vitro experimental models for MODY3 (11,12,18). The rationale of the in vivo study is to specifically suppress HNF-1α function by targeted overexpression of a DN mutant of HNF-1α (DNHNF-1α) (11) in β-cells of transgenic mice. Two approaches were followed to achieve β-cell–specific expression of DNHNF-1α in the transgenic mice. First, the rat insulin promoter (RIP) II was used to directly drive the expression of the DNHNF-1α transgene (RIP-DNHNF-1α mice). Second, the binary Tet-On system was used to achieve doxycycline-inducible expression of this transgene (Tet-DNHNF-1α and RIP-rtTA mice).

RIP-DNHNF-1α males became markedly diabetic around 6 weeks of age (Fig. 1A) and were already hyperglycemic in the fasted state (18). The RIP-DNHNF-1α females were glucose intolerant (Fig. 1B) but never developed diabetes. Females also exhibited elevated nonfasting blood glucose values at weaning (134% of controls, P < 0.001) similar to those of males (138% of controls, P < 0.001). This was associated with comparable decreases in pancreatic insulin content (∼40% of control), which thereafter increased in females and declined further in males (18). The males showed a blunted insulin response to intraperitoneal glucose injection at 5 weeks of age and failed to increase plasma insulin at 6 weeks. To examine the consequences of β-cell–specific expression of DNHNF-1α on insulin secretion, we also perfused the pancreas of 5-week-old males (before the onset of diabetes). The first phase of glucose-stimulated insulin secretion was diminished by 50%, whereas the small second phase was less affected (69.4% of control). Arginine potentiation of glucose-induced insulin secretion was even more inhibited (26.9% of control). Progressive β-cell dysfunction was also demonstrated by age-related reduction in pancreatic insulin content and severely damaged β-cell structure. There was swelling of the mitochondria and dilatation of the endoplasmic reticulum (18). The latter is a sign of apoptosis, which has been documented in INS-1 cells expressing the same transgene (19). Islet GLUT2 expression was severely reduced in 6-week-old RIP-DNHNF-1α males. Unfortunately, further studies of gene expression were impossible because we were unable to isolate islets from these transgenic animals because of disrupted islet structure. The islet disorganization could be due, at least in part, to downregulation of the cell adhesion molecule E-cadherin, observed in a similar mouse model (20) exhibiting a more severe phenotype because the animals required insulin therapy. In that model, glucose-stimulated insulin secretion in isolated islets from glucose-intolerant female mice was also markedly attenuated.

As an alternative approach, we used an inducible system that allows conditional expression of DNHNF-1α. Two independent transgenic mouse lines were generated. The first carries the reverse tetracycline-dependent transactivator (rtTA) driven by RIP (RIP-rtTA mice). The second harbors the DNHNF-1α transgene under the control of the Tet operator (Tet-DNHNF-1α mice). Two Tet-DNHNF-1α founder mice were identified by PCR and Southern blotting and used to establish two transgenic families, termed Tet-18 and Tet-23. Tet-DNHNF-1α mice from both families were then crossed with RIP-rtTA mice to obtain double-transgenic (Dtr) RIP-rtTA/Tet-DNHNF-1α mice. Dtr mice constitute ∼25% of each litter, whereas the other littermates are either nontransgenic or single-transgenic. The latter were subdivided into two control groups: 1) those positive for the RIP-rtTA transgene (rtTA⁺) and 2) those carrying the Tet-DNHNF-1α transgene or negative for both transgenes (controls). To induce expression of DNHNF-1α, pregnant mice and their offspring were given doxycycline in their drinking water (0.8 g/l); these doxycycline-treated animals were then compared with untreated mice. To assess glucose homeostasis in mice from both Tet-DNHNF-1α families crossed with the RIP-rtTA mice, glucose tolerance tests were performed on ∼9-week-old mice. The results are summarized in Fig. 2A and B (−dox) and Fig. 2C and D (+dox). Unexpectedly, all RIP-rtTA⁺ mice, whether treated with doxycycline or not, exhibited mild glucose intolerance. Because there is high expression of rtTA in the β-cells of the RIP-rtTA family, normal β-cell function may be altered. Given the glucose intolerance of the rtTA⁺ mice, these animals were used as controls in this study. As expected, in the absence of activation by doxycycline, the glucose intolerance of Dtr mice, whether male or female, was not further increased when compared with rtTA⁺ mice (Fig. 2A and B). This result indicates that although rtTA expression by itself affected β-cell function, it remained unable to transactivate the Tet-DNHNF-1α transgene in the absence of doxycycline. In contrast, when mice from the Tet-18 and Tet-23 families were treated with doxycycline, Dtr males became more severely glucose intolerant (Fig. 2C and D), statistically significant at 60, 90, and 120 min; this result was apparently much more pronounced in males of the Tet-18 family. Interestingly, in this family, the fasting blood glucose values were also significantly higher than those in controls, which was due to two Dtr males exhibiting mild hyperglycemia (∼12 mmol/l). Together these results indicate that treating RIP-rtTA/Tet-DNHNF-1α Dtr mice with doxycycline further aggravates the glucose intolerance observed in the rtTA males. The intolerance was more pronounced in males than in females (data not shown) and in Dtr males of the Tet-18 than in the Tet-23 family (Fig. 2). Interestingly, this difference between males and females parallels the sexual dimorphism observed in RIP-DNHNF-1α mice (18,20). It should also be noted that although doxycycline-treated RIP-rtTA/Tet-DNHNF-1α animals never became overtly diabetic, males from both families occasionally exhibited fasting blood glucose >10 mmol/l. This relative hyperglycemia was never observed in mice not subjected to doxycycline treatment.

Using immunostaining with an antibody against the NH₂-terminus of HNF-1α, we showed that anti–HNF-1α staining of islets isolated from untreated animals did not notably differ between control and glucose-intolerant Dtr males of the Tet-23 family (Fig. 3A and B). However, DNHNF-1α overexpression could be demonstrated in the islets of the doxycycline-treated Tet-23 Dtr males that exhibited some degree of hyperglycemia (Fig. 3C and D). In support of this, insulin staining was attenuated in pancreas sections from Tet-23 Dtr males (Fig. 3F), although not to the same extent as in RIP-DNHNF-1α males. Taken together, the phenotype of the Dtr males is reminiscent of the clinical characteristics reported for marker-positive nondiabetic MODY3 patients (7). This inducible HNF-1α animal model will provide a powerful tool to study the pattern of gene expression over time and determine the primary and secondary effects of DNHNF-1α in MODY3 pathology.

To identify HNF-1α target genes and elucidate the molecular mechanisms underlying β-cell dysfunction after its suppression, we also established INS-1 stable clones permitting expression of the artificial rat HNF-1α mutation DNHNF-1α (11) and the most common human mutation HNF-1αP291fsinsC (21), respectively. As summarized in Table 1, HNF-1α targets multiple genes implicated in metabolism-secretion coupling. Induction of either DNHNF-1α or HNF-1αP291fsinsC resulted in severely impaired glucose- and leucine-stimulated, but not KCl-evoked, insulin secretion (11,21). Accordingly, glucose and leucine oxidation was severely reduced in cells expressing DNHNF-1α (11). This defect lies in the mitochondria because [2-¹⁴C]pyruvate oxidation and glucose-induced mitochondrial hyperpolarization (reflecting electron transport chain activity) were inhibited. The carbon in the [2-¹⁴C]pyruvate can only be lost to ¹⁴CO₂ at either the isocitrate dehydrogenase or 2-oxoglutarate dehydrogenase (OGDH) steps in the tricarboxylic acid (TCA) cycle. Interestingly, the mRNA and protein levels of the catalytic OGDH E1 subunit, which is involved in conversion of [2-¹⁴C]pyruvate to ¹⁴CO₂, were reduced in HNF-1αP291fsinsC cells. Furthermore, the expression of uncoupling protein (UCP)-2 was increased. UCP-2 may act as a protonophore and dissipate the mitochondrial membrane potential, thereby uncoupling respiration from ATP synthesis (21). In addition, the generation of ATP by nutrients was abolished in both the doxycycline-treated DNHNF-1α and HNF-1αP291fsinsC cell lines (11,21). Together, these results strongly point to defective mitochondrial metabolism as a primary mechanism underlying the diminished cytosolic Ca²⁺ rise (11) and impaired insulin secretion in cells with induced HNF-1α deficiency. It is tempting to speculate that β-cell mitochondrial dysfunction and morphological changes in β-cells of RIP-DNHNF-1α mice may not be mutually exclusive. In addition to its apparent importance in regulating mitochondrial function, HNF-1α was also shown to regulate genes involved in glucose transport and glycolysis, such as GLUT2, aldolase B, and liver-type pyruvate kinase (L-PK). In DNHNF-1α cells, the glycolytic flux was indeed decreased, despite unaltered expression of glucokinase, the rate-limiting glycolytic enzyme (11). Our findings contradict those of Shih et al. (22), who showed by RT-PCR that glucokinase expression was reduced in islets of hnf-1α^{−/−minus;−/−sol;−/−minus;} mice and hypothesized that diminished glycolysis rather than defective mitochondrial function impairs glucose-stimulated insulin secretion. Interestingly, Northern blot analysis revealed increased rather than decreased glucokinase mRNA in islets isolated from the same transgenic animals (15). The hypothesis proposed by Shih et al. suggests that MODY3 and MODY2 should display similar clinical phenotypes. However, there are distinct clinical and pathophysiological profiles between these two genetic subtypes of diabetes, reflecting the clear differences in β-cell dysfunction (5). Loss-of-function mutations in glucokinase lead to a stable defect of glucose sensing, whereas the HNF-1α mutations cause progressive impairment of glucose-stimulated insulin secretion rather than glucose sensing (5). Therefore, the abolished insulin secretion in response to glucose in MODY3 patients supports our conclusion that HNF-1α deficiency causes both reduced glycolysis and diminished mitochondrial metabolism.

Dominant-negative suppression of HNF-1α function markedly reduced cellular insulin, both at the mRNA and protein levels, by direct inhibition of RIP I activity (11,21). It has been demonstrated that HNF-1α regulates the PDX-1 promoter activity (23). We found that induction of HNF-1αP291fsinsC for 96 h caused 60% reduction of PDX-1 mRNA levels (Fig. 4A). However, the suppressive effect of HNF-1αP291fsinsC on insulin precedes its action on PDX-1 expression (21). Using our DNHNF-1α and HNF-1αP291fsinsC cell models, genes such as cyclin E, p27^KIP1, and Bcl-xL, which regulate cell proliferation and apoptosis, have been identified as HNF-1α targets (19,24). The altered expression of these genes may also be implicated in the progressive deterioration of β-cell function and cell mass in MODY3 patients. HNF-1α has also been shown to be an essential transcriptional regulator of bile acid and HDL cholesterol metabolism (17) and to control renal glucose reabsorption (25).

MODY1, caused by mutations in the HNF-4α gene, is much less prevalent than other subtypes of MODY. Although it is less penetrant than MODY3 and the average age of patients with MODY1 at diagnosis is slightly higher, clinical features of MODY1 are very similar (1,2). Primary β-cell dysfunction in MODY1 patients also results in impaired insulin secretion and leads to the development of a severe form of diabetes (1,2). HNF-4α acts upstream of HNF-1α in a transcriptional cascade that mediates hepatic gene expression and differentiation (26). A mutation in the HNF-4α binding site of the HNF-1α promoter has been reported in a MODY3 family, suggesting that this transcriptional hierarchy could also exist in β-cells (10). Targeted deletion of HNF-4α results in embryonic lethality that prevents further analysis of HNF-4α function in β-cells (27). Our INS-1 cell model, permitting conditional expression of HNF-4α or its dominant-negative mutant DNHNF-4α, represents a unique system for this analysis (28). Similar to DNHNF-1α, induction of DNHNF-4α leads to defective mitochondrial metabolism and, as a consequence, impaired insulin secretion. Decreased TCA cycle activity, as well as partial uncoupling of the mitochondrial electron transport chain, could be demonstrated. In addition, we also showed that the expression of the OGDH E1 subunit and UCP-2, as well as GLUT2, aldolase B, and L-PK, is regulated by HNF-4α. Induction of DNHNF-4α indeed eliminated endogenous HNF-1α expression and severely reduced HNF-1α promoter activity in a reporter gene assay (28), suggesting that HNF-4α is an upstream regulator of HNF-1α.

Two recent reports showed that a distant upstream promoter in the HNF-4α gene (P2) is specifically used in INS-1 and pancreatic β-cells (29,30). The P2 promoter containing an HNF-1α binding site is functional (29) and is occupied by HNF-1α selectively in pancreatic islets in vivo (30). Accordingly, HNF-4α transcription is severely decreased in islets, but not in liver, of hnf-1α^{−/−minus;−/−sol;−/−minus;} mice (30). These studies thus indicate that HNF-1α can also act upstream of HNF-4α in β-cells, revealing a complex transcription factor circuit operating to maintain normal β-cell function (30). To verify this hypothesis, we analyzed the expression pattern of HNF-4α in our INS-1 stable clones after induction of HNF-1αP291fsinsC for 96 h. As shown in Fig. 4A, dominant-negative suppression of HNF-1α resulted in 40% reduction in the HNF-4α mRNA level. The effect is specific because the transcript of the housekeeping gene cyclophilin remains unaltered. This finding supports the notion of a regulatory loop between HNF-4α and HNF-1α transcription (30). A MODY1 family was reported in which the PDX-1 binding site in the HNF-4α P2 promoter was mutated, suggesting a regulatory circuit between PDX-1, HNF-4α, and HNF-1α. We therefore investigated the consequence of upregulation and downregulation of PDX-1 function on HNF-4α expression in INS-1 cells, which permit inducible expression of PDX-1 or a dominant-negative mutant of PDX-1 (DN-PDX-1) (31). Induction of PDX-1 caused a mild increase in HNF-4α mRNA (Fig. 4B), whereas dominant-negative suppression of PDX-1 resulted in a slight reduction in this transcript (Fig. 4C). These results suggest that PDX-1 does not play a major role in the regulation of HNF-4α expression, at least in our in vitro system. The regulatory loop between PDX-1 and HNF-4α/HNF-1α remains to be established.

PDX-1 is required for pancreatic development and for maintaining the β-cell phenotype (31,32). Homozygosity for a mutation in the PDX-1/IPF-1 gene results in failure of pancreas formation, whereas heterozygosity for an inactivating mutation and missense mutations in this gene are associated with MODY4 and late-onset type 2 diabetes in humans, respectively (33,34). The primary defect in MODY4 is impaired glucose-stimulated insulin secretion (35). Targeted disruption of PDX-1 in mice results in pancreatic aplasia (32), whereas mice with β-cell–specific deletion of PDX-1 develop diabetes (36). Brissova et al. (37) recently reported impaired glucose tolerance caused by defective glucose-stimulated insulin secretion in the conventional heterozygous PDX-1 mutant mouse. Reduced GLUT2 expression and decreased glucose-evoked NAD(P)H generation in islets were also observed, suggesting impaired glucose metabolism. It is noteworthy that glucokinase expression is unaffected in islets isolated from these mice, supporting our previous conclusion that PDX-1 does not regulate the glucokinase gene (31), in contrast to previous claims by Watada et al. (38). Unaltered glucokinase levels were also demonstrated in the islets of mice with β-cell–specific deletion of PDX-1 (36). Furthermore, PDX-1 does not bind to the β-cell glucokinase promoter in vivo (39).

The in vivo study on mice with β-cell–specific deletion of PDX-1 (36) and our in vitro work on INS-1 cells expressing DN-Pdx-1 (31) indicate that PDX-1 promotes the β-cell phenotype by maintaining the expression of β-cell–specific genes while suppressing glucagon levels. Suppression of glucagon expression by PDX-1 requires synergism with other β-cell–specific transcription factors because ectopic expression of PDX-1 in glucagonoma cells failed to eliminate glucagon (40). We also showed that maximum induction of Brain-4, an α-cell–specific transcription factor, initiated detectable levels of glucagon but did not attenuate the expression of β-cell–specific genes in INS-1 cells (31). This result is in good agreement with a recent in vivo study by Hussain et al. (41), where PDX-1 promoter-driven expression of Brain-4 in pancreatic precursors transactivates glucagon in β-cell lineage without altering β-cell gene expression. Therefore, both Brain-4 expression and loss of PDX-1 function are necessary for differentiation of the α-cell lineage. The transcription factors that switch on and off Brain-4 and PDX-1, respectively, remain to be defined.

We further investigated the mechanism by which PDX-1 deficiency impairs insulin secretion using our INS-1 cell models established for inducible expression or suppression of PDX-1 (31). After a 4-day induction of DN-Pdx-1, glucose-stimulated insulin secretion was markedly inhibited, whereas overexpression of PDX-1 had no effect (42). In contrast to the dominant-negative suppression of HNF-1α and HNF-4α, loss of PDX-1 function also impaired K⁺-induced insulin release. This was also observed in the pancreas perfusion experiment in PDX-1^+/− mice (37). Glucokinase enzyme activity and mRNA level were not altered by dominant-negative suppression of PDX-1 function. Although GLUT2 expression was decreased by 90% after induction of DN-Pdx-1, glycolysis was unaltered, explained by the large excess of the transporter. Similarly, induction of DN-Pdx-1 did not affect mitochondrial metabolism. We found that induction of DN-Pdx-1 caused downregulation of insulin, prohormone convertases (PC1/3 and 2), ATP-sensitive K⁺ (K_ATP) channel subunits (Sur1 and Kir6.2), fibroblast growth factor receptor 1, and glucagon-like peptide 1 receptor. The molecular basis of the global inhibition of insulin secretion remains undefined at present, but decreased insulin processing combined with decreased K_ATP channel activity may contribute to the secretory defect.

The predominant feature of MODY5 patients, carrying mutations in the gene encoding HNF-1β, appears to be renal dysfunction, usually diagnosed before diabetes (43). HNF-1β–deficient mice die shortly after implantation with abnormal visceral endoderm (44). HNF-1β is highly homologous to HNF-1α, and the two transcription factors can form heterodimers for cognate DNA binding (45). Overexpression of DN-HNF-1α and the human mutation HNF-1αP291fsinsC, whether in mice or in cell lines, also suppresses HNF-1β function. To our knowledge, specific studies on the role of HNF-1β in β-cells have not been performed.

Mutations in the gene encoding the transcription factor NeuroD 1/BETA2 have been linked to MODY6 as well as type 1 diabetes (4,46). Mice with targeted disruption of NeuroD 1 died of diabetes shortly after birth (47). The NeuroD 1–deficient pancreas displayed reduced endocrine cell mass, affecting especially the β-cells (47). However, in an alternative genetic background, the NeuroD 1^−/− mice display neonatal diabetes but become normoglycemic through pronounced β-cell neogenesis (48). The role of NeuroD 1 in the regulation of pancreatic β-cell function remains to be established.

Foxa2 (HNF-3β), which is expressed in islets, has been suggested as the upstream transactivator of HNF-4α, HNF-1α, PDX-1, and HNF-1β in the transcriptional hierarchy (49,50). Because these experiments were performed in embryonic stem cell–derived embryoid bodies, this transcriptional cascade may not apply to differentiated β-cells and hepatocytes. Indeed, genetic analysis has failed to link Foxa2 mutations to MODY pedigrees (51). In addition, Sund et al. (52) showed that HNF-1α and HNF-1β mRNA levels are unchanged, whereas HNF-4α mRNA is slightly elevated in mice with liver-specific deletion of Foxa2. Most recently, Tan et al. (53) demonstrated that adenovirus-mediated overexpression of Foxa2 in mouse liver led to a drastic decrease in the expression of HNF-4α and HNF-1α. The phenotype of mice with β-cell–specific deletion of Foxa2 is severe hyperinsulinemic hypoglycemia (54) rather than diabetes. We found that PDX-1 expression in INS-1 cells was not regulated by Foxa2 (55). Overexpression of Foxa2 suppressed rather than activated the expression of HNF-4α and HNF-1α in INS-1 cells. These results confirm that Foxa2 is not a candidate gene for MODY (51).

The principal sites of action of HNF-1α, HNF-4α, and PDX-1 in β-cell function are illustrated in Fig. 5. Insulin expression is regulated by all three genes. HNF-1α and HNF-4α regulate metabolism-secretion coupling and generation of ATP, whereas PDX-1 appears to act beyond mitochondrial metabolism, affecting insulin exocytosis. The stable diabetic phenotype of the RIP-DNHNF-1α males and the glucose intolerance of the females make these mice attractive models of MODY3, amenable to molecular biological and pharmacological investigation. The INS-1 cell lines engineered to manipulate the MODY genes have helped to identify new target genes and to define the regulatory circuit of β-cell transcription factors.

This work was supported by Swiss National Science Foundation Grants 32-49755.96 (to C.B.W.) and 31-53027.97 (to P.L.H.); the Bonizzi-Theler Stiftung, Zürich (to C.B.W.); and the Novartis Foundation, Basel (to C.B.W.).

We are grateful to E.-J. Sarret, D. Cornut-Harry, J. Ritz, and Y. Dupré for expert technical assistance.