Hnf1b controls pancreas morphogenesis and the generation of Ngn3⁺ endocrine progenitors

The pancreas is an abdominal gland that is essential for nutrient homeostasis formed by acinar, ductal and endocrine cells. Acinar cells, organized in acini, secrete digestive enzymes into the duodenum via the ductal network. Endocrine cells are organized in the islets of Langerhans, notably composed of α, β and δ cells, which secrete the hormones glucagon, insulin and somatostatin, respectively. In mice, the pancreas is specified by embryonic day (E) 8.5. Early pancreas morphogenesis, known as the primary transition, is characterized by active proliferation of epithelial multipotent progenitor cells (MPCs), followed by a period of growth and differentiation from E12.5, called the secondary transition, to form acinar, ductal and endocrine cells. Regionalization of the early epithelium results in a pre-acinar domain in the tips of the branching organ and a central bipotential ductal/endocrine trunk domain, from which endocrine cells differentiate (Seymour and Sander, 2011). Notch signaling plays an important role in this process (Afelik and Jensen, 2013). Numerous transcription factors have been implicated in the regulation of pancreas development (Pan and Wright, 2011). Foxa1/a2 are dominant regulators of Pdx1 expression (Gao et al., 2008). Pdx1 and Ptf1a determine fate specification of pancreatic MPCs (Pan and Wright, 2011), whereas Sox9 maintains MPCs, by preventing apoptosis and promoting proliferation (Seymour et al., 2007). Ngn3 is required for endocrine cell differentiation (Gradwohl et al., 2000). During the secondary transition, Sox9 and Hnf6 are both expressed in the bipotent duct/endocrine domain and required for maintaining Ngn3 expression (Dubois et al., 2011; Jacquemin et al., 2000; Seymour et al., 2008). After the secondary transition, endocrine and exocrine cell populations expand and differentiate to generate the mature hormone- and enzyme-producing cell types of islets and acini, respectively.

Among transcription factors, hepatocyte nuclear factor 1b (Hnf1b) is expressed in the pre-pancreatic foregut endoderm and in pancreatic MPCs. A sequential transcriptional cascade of Hnf1b→Hnf6→Pdx1 was found to direct differentiation of endodermal cells into pancreatic progenitors (Poll et al., 2006). From ∼E14.5 to adulthood, Hnf1b expression is restricted to the embryonic ductal cords that later form the adult ductal cells (Haumaitre et al., 2005; Kopp et al., 2011; Maestro et al., 2003; Nammo et al., 2008). We have previously shown that Hnf1b-deficient mouse embryos rescued by tetraploid aggregation exhibit pancreas agenesis, with a transient dorsal bud expressing Pdx1 but not Ptf1a, showing that Hnf1b is required for pancreas specification (Haumaitre et al., 2005). Regionalization of the gut was also affected, as revealed by ectopic expression of Shh, which could contribute to reduce the Pdx1⁺ pre-pancreatic domain. Hnf1b-deficient embryos also exhibit impaired specification of other organs derived from the ventral endoderm, including liver (Lokmane et al., 2008). Thus, it was difficult to distinguish the role of Hnf1b in the pancreas from its role in regionalizing the primitive intestine. Moreover, the severity of the phenotype precluded the analysis of crucial later roles of Hnf1b during pancreas differentiation. Indeed, lineage-tracing analyses revealed that embryonic Hnf1b⁺ cells of the branching pancreas are precursors of acinar, duct and endocrine lineages (Solar et al., 2009). In humans, HNF1B heterozygous mutations are associated with ‘maturity onset diabetes of the young type 5’ (MODY5) syndrome, which is characterized by early onset of diabetes, pancreas hypoplasia and multicystic kidney dysplasia (Bellanné-Chantelot et al., 2004; Chen et al., 2010; Edghill et al., 2006; Haldorsen et al., 2008). The identification of two fetuses carrying distinct HNF1B mutations, associated with polycystic kidneys and severe pancreatic hypoplasia (Haumaitre et al., 2006), further suggested an early developmental role of HNF1B in human pancreas, which might be an important cause of MODY5.

In order to elucidate the specific role of Hnf1b in pancreas development, we conditionally inactivated Hnf1b in pancreatic MPCs and at later stages. Combined early and late deletion analyses demonstrate the crucial function of Hnf1b in the regulatory networks controlling pancreatic MPC expansion, duct morphogenesis, acinar cell identity and generation of endocrine precursors.

We performed a conditional deletion of Hnf1b in pancreatic MPCs using a Hnf1b-floxed mouse line (Heliot et al., 2013) crossed with the Pdx1-Cre (Wells et al., 2007) or the tamoxifen (TM)-inducible Sox9-CreER^T2 line (Kopp et al., 2011), as Pdx1 and Sox9 share a common expression domain with Hnf1b in the early pancreas (Dubois et al., 2011; Maestro et al., 2003; Seymour et al., 2007). Pdx1-Cre;Hnf1b^+/LacZ and Hnf1b^Flox/Flox mice were crossed to generate Pdx1-Cre;Hnf1b^Flox/LacZ embryos, referred to as mutants. Hnf1b^Flox/LacZ and Hnf1b^+/Flox embryos without the Pdx1-Cre transgene were referred to as controls, as haploinsufficient embryos with the LacZ-null Hnf1b allele did not show any phenotype (Barbacci et al., 1999; Kornfeld et al., 2013). Heterozygous Pdx1-Cre;Hnf1b^+/Flox embryos also showed no phenotype (Fig. 1E; supplementary material Fig. S1). By contrast, Pdx1-Cre;Hnf1b^Flox/LacZ mutant embryos displayed severe pancreatic hypoplasia at E18.5 (Fig. 1A-D), corresponding to a 45% and 90% decrease in pancreatic weight at E16.5 and E18.5, respectively (Fig. 1E). We also generated Pdx1-Cre;Hnf1b^Flox/LacZ;R26R^+/YFP mutants and observed uniform YFP labeling in the remnant pancreatic epithelium, revealing the high efficiency of the Pdx1-driven Cre recombination (Fig. 1C′,D′). Hnf1b/GFP co-immunostainings at E10.5 confirmed the extensive deletion of Hnf1b in mutants, showing only 16% of remaining Hnf1b⁺/GFP⁻ cells due to a slight mosaic expression of the Pdx1-Cre line (Fig. 1F-G′). In accordance, we found a 70% decrease in wild-type (WT) Hnf1b transcripts at E12.5 (Fig. 1H). Histological analysis by Hematoxylin and Eosin staining revealed a severe decrease in acinar cells with dispersed clusters of acini, cystic ducts and an apparent absence of endocrine islets in mutant pancreata at E16.5 and E18.5 (Fig. 1I-L). This phenotype was associated with high lethality of mutant pups, as 70% died during the first week of life (Fig. 1M). Interestingly, we found that mutant newborns were hypoglycemic, with a 30% decrease of blood glucose (Fig. 1O). This phenotype correlates with a 93% decrease in glucagon-expressing cells (see Fig. 7O). Hypoglycemia is likely the main cause of mutant lethality at P0/P1 (40%), because immediately after birth, glycogenolysis stimulated by glucagon allows mobilisation of hepatic glycogen, which is the only energetic source available at this stage. Hnf3a^−/− mice also die perinatally of hypoglycemia, associated with a 50% decrease in glucagon (Kaestner et al., 1999; Shih et al., 1999). By contrast, at P2, mutants became hyperglycemic, with a 44% increase of blood glucose (Fig. 1O), in correlation with the 93% decrease in insulin-expressing cells (see Fig. 7O). Furthermore, we found a massive increase of blood amylase in mutants (250%) (Fig. 1P), as is the case in pancreatitis. This was associated with acinar cell defects (see Fig. 4) and a 32% decrease in mutant body weight at P2 (Fig. 1N). This suggests that mutant lethality after P2 may be due to hyperglycemia and to chronic malabsorption.

We also conditionally inactivated Hnf1b with the Sox9-CreER^T2 line. Pregnant females from crosses between Hnf1b^Flox/Flox and Sox9-CreER^T2;Hnf1b^+/LacZ mice were injected with TM at E9.5 (Kopp et al., 2011). Sox9-CreER^T2;Hnf1b^Flox/LacZ mutants exhibited strong pancreatic hypoplasia at E18.5 (supplementary material Fig. S2A,B), corresponding to a 40% decrease in pancreas weight at E16.5 (supplementary material Fig. S2G). Hematoxylin and Eosin staining of mutant pancreata showed cystic ducts with a severe decrease in acinar cells and absence of islet structures (supplementary material Fig. S2C-F). Efficient Sox9-driven Cre recombination was observed with a 70% decrease in Hnf1b transcripts at E14.5 (supplementary material Fig. S2H), and further confirmed by Hnf1b and GFP co-immunostainings at E11.5 in Sox9-CreER^T2;Hnf1b^Flox/LacZ;R26R^+/YFP mutants (supplementary material Fig. S2I-J′). These data strongly corroborate the findings obtained with the Pdx1-Cre line. Therefore, Hnf1b inactivation in MPCs leads to severe pancreatic hypoplasia, associated with cystic ducts, a decrease in acinar cells and absence of endocrine cells.

To investigate the underlying cause of pancreatic hypoplasia in Pdx1-Cre;Hnf1b^Flox/LacZ mutants, we analyzed the pool of MPCs at E12.5 using Pdx1 immunostaining and observed a 35% decrease in Pdx1⁺ progenitor cells (Fig. 2A). We further analyzed proliferation and apoptosis. We quantified the percentage of mitotic and apoptotic Pdx1⁺ cells using phospho-histone H3 (PHH3) and TUNEL assay, respectively. A 20% decrease in Pdx1⁺ cell proliferation (Fig. 2B,D,E), as well as an 11-fold increase in Pdx1⁺ cell apoptosis (Fig. 2C,F,G) were observed in mutants compared with controls. Thus, both decreased proliferation and increased cell death contribute to the reduction of MPCs in mutants.

Activation of the fibroblast growth factor (FGF) pathway via binding of mesenchymal FGFs to epithelial FGF receptors (FGFRs) is fundamental to promote proliferation of early pancreatic MPCs, especially through the Fgf10/Fgfr2b pathway (Bhushan et al., 2001; Hart et al., 2003; Pulkkinen et al., 2003). Surprisingly, we observed no difference in Fgfr2b expression in Pdx1-Cre;Hnf1b^Flox/LacZ mutants at E12.5, but we found a 70% decrease in Fgfr4 expression by qRT-PCR (Fig. 2H). Moreover, by in vivo chromatin immunoprecipitation (ChIP) on E12.5 pancreata (Fig. 2I), we found Hnf1b bound to a region containing two previously described Hnf1 DNA-binding sites (Shah et al., 2002), at +280 and +355 bp downstream of the Fgfr4 transcription start site (TSS) in its first intron. These data suggest that modulation of FGF signaling through direct regulation of Fgfr4 by Hnf1b may sustain MPC expansion.

The Notch signaling pathway also plays an important role in the maintenance, proliferation and differentiation of pancreatic MPCs (Apelqvist et al., 1999; Hald et al., 2003; Jensen et al., 2000; Murtaugh et al., 2003). In Pdx1-Cre;Hnf1b^Flox/LacZ pancreata, we observed a downregulation of the Notch ligand Dll1 (40%), and an upregulation of the effectors Hey1, Hey2 and Heyl at E12.5 (Fig. 3A). Interestingly, we found that expression of Hey and other Notch members, such as Notch2 and Jag1, remained abnormally high in mutant pancreata at E14.5 contrary to controls, in which these genes are downregulated at this stage (Fig. 3B). Thus, lack of Hnf1b is associated with deregulation of some Notch pathway components. These findings show that Hnf1b is essential for proliferation and maintenance of MPCs, at least in part through modulation of FGF and Notch pathways.

The severe loss of acinar cells and the impaired architecture of remaining acini with enlarged acinar lumens observed by Hematoxylin and Eosin staining in mutants (Fig 1I-L) led us to investigate acinar defects. Amylase immunostaining (Fig. 4A-H) and amylase⁺ cross-sectional area quantification at E16.5 showed a 20% decrease in acinar area in Pdx1-Cre;Hnf1b^Flox/LacZ mutants (Fig. 4J). A 67% decrease in Amylase expression was observed at E16.5 by qRT-PCR (Fig. 4I), suggesting that Amylase expression might be reduced in the remaining mutant acinar cells. We also found an eightfold increase of apoptotic mutant acinar cells (Fig. 4L), which cannot be fully compensated for by a twofold increase of mitotic acinar cells (Fig. 4K). In addition, we analyzed expression of the early acinar markers Ptf1a, Mist1 (Bhlha15 – Mouse Genome Informatics) and Nr5a2 (Hale et al., 2014; Pin et al., 2001). Expression of Mist1 was severely reduced before and during the onset of acinar differentiation, with a 50% and an 80% decrease at E12.5 and E14.5, respectively (Fig. 4I). We also observed a 64% decrease of Ptf1a at E14.5 and a 35% decrease of Nr5a2 at E16.5 (Fig. 4I). Notably, an equivalent severe decrease in acinar cells was found in Sox9-CreER^T2;Hnf1b^Flox/LacZ (TM E9.5), with a 74% downregulation of amylase gene expression at E16.5 (supplementary material Fig. S3E).

Interestingly, the ductal marker Hnf6 was almost undetectable in Pdx1-Cre;Hnf1b^Flox/LacZ pancreatic ducts, but was found to be ectopically expressed in some differentiated acinar cells at E16.5 (Fig. 4A-D). Ectopic expression of another ductal marker, Sox9, was also observed in mutant acini (Fig. 5C,D). This was accompanied by expanded expression of the ductal marker cytokeratin (pan-CK) (Fig. 4E-H) and by a twofold upregulation of Ck19 observed by qRT-PCR (Fig. 4I). In agreement with these observations, the Sox9⁺ cross-sectional area increased threefold in mutant pancreata (Fig. 4J), even though the proliferation rate in this compartment was not significantly changed (Fig. 4K).

Thus, lack of Hnf1b leads to increased acinar cell death and reduced expression of acinar cell markers, accompanied by ectopic expression of ductal markers. These results uncover an early role of Hnf1b in acinar differentiation and maintenance of acinar cell identity.

We performed Sox9 immunostaining at different stages of pancreas development to further analyze duct morphogenesis (Fig. 5A-F). Cystic ducts at E16.5 in Pdx1-Cre;Hnf1b^Flox/LacZ mutants showed some multilayered epithelium (Fig. 5B,D). Moreover, lack of Sox9 (Fig. 5E,F) or AQP1 expression (Fig. 5K-L′) revealed loss of ductal cell identity in large cystic ducts. We further analyzed the localization of polarity markers in mutants. Similar to PanCK (Fig. 4E-H), strong and expanded β-catenin expression to basal and apical membranes was observed in terminal enlarged ductal cells (Fig. 5G-H′). The basal expression of dystroglycan (Fig. 5G-H′) and laminin (Fig. 5I-J′) was also disrupted in most mutant acinar cells, further illustrating the defects in acinar cells and acquisition of ductal features. Importantly, although control ducts showed strong apical localization of ezrin, PKCz and MUC1 in epithelial cells around the duct lumen, expression of these apical markers in the cells lining cysts was reduced and discontinuous (Fig. 5M-R′). Cyst formation is often associated with an absence or dysfunction of primary cilia (Ware et al., 2011). Immunostaining of acetylated tubulin, a specific component of the cilium axoneme, revealed that cystic cells were devoid of primary cilia (Fig. 5S,T), whereas cilia were still present in non-cystic ducts, suggesting that Hnf1b is not required for primary cilium formation. Similar cystic ducts were observed in Sox9-CreER^T2;Hnf1b^Flox/LacZ (TM E9.5) mutants (supplementary material Fig. S4A,B), with altered ductal cell polarity and abnormal localization of β-catenin (supplementary material Fig. S4C,D). Notably, Sox9-CreER^T2;Hnf1b^Flox/LacZ (TM E12.5) mutants also displayed cystic ducts (supplementary material Fig. S5D,E). These data show that Hnf1b is required for morphogenesis and for epithelial polarization of ductal cells.

To gain insight into how Hnf1b controls ductal cell differentiation, we further investigated the expression of cystic disease genes in Pdx1-Cre;Hnf1b^Flox/LacZ mutants. Although Hnf1b and Sox9 are both expressed in pancreatic ducts, and despite the cystic phenotype of pancreatic Sox9 mutants (Shih et al., 2012), we found no change in Sox9 expression in our mutants (Fig. 5U). By contrast, Hnf6 expression was strongly decreased in mutant ductal cells both by immunostaining (Fig. 4A,C) and by qRT-PCR at E14.5 (Fig. 5U), in line with the cystic phenotype of Hnf6^−/− pancreata (Pierreux et al., 2006; Zhang et al., 2009). We also observed a decrease in Spp1 expression, which is directly regulated by Hnf1b in renal cells (Senkel et al., 2005). Importantly, we found a 90% downregulation of the autosomal recessive PKD gene Pkhd1 (Ward et al., 2002), and a dramatic decrease in the expression of the key cystic disease genes Kif12 (Mrug et al., 2005), Cys1 (Hou et al., 2002), Bicc1 (Cogswell et al., 2003; Lemaire et al., 2015) and Glis3 (Kang et al., 2009b) (Fig. 5U). Cys1 and Glis3 are of particular interest. Cys1 is responsible for congenital polycystic kidney (CPK) disease (Tao et al., 2009) and is involved in ciliogenesis and polarization of cholangiocytes (Raynaud et al., 2011), whereas Glis3 is implicated in polycystic disease in both kidney (Kang et al., 2009a) and pancreas (Kang et al., 2009b). Among these genes, Kif12, Pkhd1, Pkd2 and Bicc1 were identified as direct Hnf1b targets in the kidney (Gong et al., 2009; Gresh et al., 2004; Verdeguer et al., 2010). Thus, we analyzed whether Hnf1b could be a major regulator of these genes in the pancreas by ChIP experiments on E12.5 pancreata (Fig. 5V). Our results showed that Hnf1b is recruited to Hnf1-binding sites in the first intron of Hnf6, within a region known to drive Hnf6 expression in E8.75 pancreatic endoderm (Poll et al., 2006). Moreover, Hnf1b bound to a region carrying a site in the Pkhd1 promoter previously identified in the kidney (Gresh et al., 2004). Remarkably, we identified two novel Hnf1b target genes: Cys1 and Glis3 (Fig. 5V). These results suggest that Hnf1b is a key regulator of duct morphogenesis, exerting direct control of crucial genes involved in duct morphogenesis and cystogenesis.

To examine the specific requirement of Hnf1b in the ductal compartment, we inactivated Hnf1b at late embryogenesis (∼E15), using the inducible Sox9-CreER^T2 line with TM injection at E14.5. Overall pancreas morphology appeared to be normal in mutants at E18.5, and no change in expression of the ductal marker Sox9 and the acinar marker amylase were observed (Fig. 6A). However, we observed a dramatic decrease in expression of the cystic disease genes Pkhd1, Kif12, Cys1 and Glis3 (Fig. 6A). We further analyzed Sox9-CreER^T2;Hnf1b^Flox/LacZ (TM E14.5) mutant pancreata at postnatal day 8 (P8) and observed a significant reduction in pancreas weight (Fig. 6B). Histological analysis at P8 revealed many cystic ducts associated with a severe loss of acinar cells with enlarged acinar lumen often connected with enlarged terminal ducts (Fig. 6C,D). CPA1 immunostaining confirmed the decrease in the number of acinar cells in mutants (Fig. 6E,F). β-Catenin and MUC1 immunostaining revealed altered polarity of mutant ductal cells (Fig. 6G,H), which were also devoid of primary cilia (Fig. 6I,J). These data reinforce the specific role of Hnf1b in the control of duct morphogenesis, and suggest that its function in ducts contributes indirectly to the maintenance of acinar cells.

As Hnf1b⁺ cells were defined as immediate precursors of Ngn3⁺ cells by immunohistochemistry (Maestro et al., 2003; Nammo et al., 2008; Rukstalis and Habener, 2007), and lineage-tracing analyses showed that embryonic Hnf1b⁺ cells give rise to precursors of endocrine cells (Solar et al., 2009), we analyzed whether Hnf1b is required for the generation of these cells. Interestingly, we observed an almost complete loss of Ngn3⁺ cells (Fig. 7A-F), associated with a 70%, 85% and 93% decrease in Ngn3 expression observed by qRT-PCR in Pdx1-Cre;Hnf1b^Flox/LacZ pancreata at E12.5, E14.5 and E16.5, respectively (Fig. 7M). Endocrine cell differentiation was almost completely abrogated (Fig. 7G-L), as evidenced by a 93% decrease in insulin⁺ and glucagon⁺ areas at E16.5 (Fig. 7O), and decreased transcripts for insulin, glucagon and somatostatin (Fig. 7N). Sox9-CreER^T2;Hnf1b^Flox/LacZ (TM E9.5) mutants showed a similar dramatic loss of Ngn3⁺ cells (supplementary material Fig. S3A,B) and an 86% decrease in Ngn3 expression at E16.5 (supplementary material Fig. S3E). Insulin⁺ and glucagon⁺ areas were also severely reduced in size (supplementary material Fig. S3C,D), and insulin, glucagon and somatostatin mRNA levels decreased (supplementary material Fig. S3E).

To determine whether Hnf1b specifically controls the generation of endocrine precursors, excluding possible indirect effects due to the early Hnf1b deficiency in MPCs, we conditionally inactivated Hnf1b in the pancreatic epithelium during the secondary transition at ∼E13, when the major wave of endocrine cell neogenesis occurs. The Hnf1b-floxed locus was efficiently recombined in Sox9-CreER^T2;Hnf1b^Flox/LacZ (TM E12.5) pancreata at E16.5, as confirmed by a 75% decrease in Hnf1b expression (Fig. 7T). Mutant pancreas morphology and organ size appeared normal, and expression of the acinar markers Ptf1a, Mist1, Nr5a2 and amylase were only partially affected at E16.5 (supplementary material Fig. S5A). Strikingly, we found a dramatic loss of Ngn3⁺ endocrine precursors (Fig. 7P,Q) and a 90% decrease in Ngn3 expression at E16.5 (Fig. 7T). This coincided with a 81% decrease in insulin⁺ and a 69% decrease in glucagon⁺ sectional areas (Fig. 7R,S,U), and downregulation of insulin, glucagon and somatostatin transcripts (Fig. 7T). These results support the specific requirement for Hnf1b in the generation of Ngn3⁺ endocrine progenitors.

These data suggested that Hnf1b directly regulates Ngn3 expression. We performed ChIP experiments on E12.5 pancreata and found no Hnf1b enrichment −3318 base pairs (bp) upstream Ngn3 TSS, a region homologous to the human NGN3 cluster 1 enhancer containing a putative Hnf1-binding site (Lee et al., 2001). Importantly, we found that Hnf1b was recruited to a proximal and a distal region containing Hnf1-binding sites at −697 bp and −4890 bp, respectively (Fig. 7V).These results demonstrate that Hnf1b is specifically required to generate endocrine precursors, very likely by directly regulating Ngn3.

By Hnf1b conditional inactivation in the pancreas, our data demonstrate the essential functions exerted by Hnf1b in pancreatic MPC expansion and differentiation of exocrine and endocrine lineages, placing this transcription factor in a prominent position in the regulatory networks involved in these processes.

Pancreas hypoplasia observed in Pdx1-Cre;Hnf1b^Flox/LacZ embryos is correlated with a reduced pool of MPCs, as it was previously shown that the progenitor pool defines the final pancreas size (Stanger et al., 2007). Hnf1b mutants display similar defects in MPC proliferation to Gata4/Gata6 compound pancreatic mutants (Xuan et al., 2012), and similar MPC apoptosis to Sox9 mutants (Seymour et al., 2007). Surprisingly, expression of the key transcription factors Pdx1, Sox9 and Ptf1a, which are involved in MPC expansion (Rieck et al., 2012), was not changed at E12.5 (data not shown), excluding their contribution to the MPC phenotype observed.

We observed that Notch signaling pathway is deregulated in Hnf1b mutant pancreata, showing a decrease in Dll1 expression and upregulation of Hey repressors. A direct regulation of Dll1 by Hnf1b, as recently shown in the kidney (Heliot et al., 2013; Massa et al., 2013), does not seem to occur in the pancreas, as ChIP on E12.5 pancreata showed no Hnf1b recruitment to a conserved region of Dll1 (data not shown). Recent data suggested a role for Dll1, which is activated by Ptf1a, in MPC proliferation (Ahnfelt-Ronne et al., 2012). Hes and Hey factors were found to inhibit the activity of the Ptf1 transcriptional complex by direct interaction with Ptf1a, without changing Ptf1a mRNA levels (Esni et al., 2004; Ghosh and Leach, 2006). Therefore, in Hnf1b mutants, the increased levels of Hey factors observed could inhibit Ptf1 transcriptional complex activity resulting in reduced Dll1 expression and MPC proliferation.

The reduced proliferation in Hnf1b-deficient pancreata could also be attributed to a decreased FGF signaling via downregulation of Fgfr4. Although the pancreatic phenotype of Fgfr4^−/− embryos was not analyzed (Weinstein et al., 1998), several recent studies showed that Fgfr4 positively regulates proliferation and has anti-apoptotic effects in models of liver, prostate and gastric cancer (Drafahl et al., 2010; Ho et al., 2009; Miura et al., 2012; Ye et al., 2011). Interestingly, both Fgfr2b and Fgfr4 were downregulated in Sox9-deficient pancreas (Seymour et al., 2012), raising the possibility that they might serve partially redundant functions in MPC proliferation.

The severe reduction in the number of acinar cells in Pdx1-Cre;Hnf1b^Flox/LacZ mutants was associated with a differentiation defect and apoptosis. The acinar compartment exhibited ectopic expression of the ductal markers Hnf6 and Sox9, which were shown to be required for acinar metaplasia and repression of acinar genes (Prevot et al., 2012). Moreover, the ductal compartment was increased in size without changing its proliferation rate, suggesting that acini were replaced by the expanded ductal compartment. This acinar defect could be associated with the dramatic decrease in Mist1 expression, as inhibition of Mist1 in acinar cells leads to severe defects, including acinar-to-ductal metaplasia (Zhu et al., 2004). The loss of acinar cell identity may also be explained by the previously described link between reduced Mist1 and Ptf1a expression, conversion of acinar cells into ductal cells and upregulation of Notch signaling (Rooman et al., 2006; Rovira et al., 2010; Shi et al., 2009). Hnf1b deletion at the onset of acinar cell differentiation (TM at E12.5) also resulted in increased Hey2 expression, even if less pronounced than in Pdx1-Cre;Hnf1b^Flox/LacZ mutants, which correlated with downregulation of Mist1 and Ptf1a (supplementary material Fig. S5F), and fewer amylase-expressing cells (supplementary material Fig. S5B,C). These results suggest an early role for Hnf1b in the acquisition of pancreatic acinar cell identity from MPCs, possibly through Mist1, Ptf1a and Hey factors. Later Hnf1b deletion (TM at E14.5) was associated with a defect in acinar cell maintenance. As Hnf1b is not expressed in acinar cells, these late acinar defects might be an indirect consequence of abnormal duct morphogenesis, as described in pancreata deficient for Jag1 (Golson et al., 2009) or Kif3a (Cano et al., 2006).

Our results show that Hnf1b is required for both early and late control of duct morphogenesis; Hnf1b deletion resulted in cystic ducts with altered polarity and a lack of primary cilia. Cilia loss in ductal cells is an important event in pancreatic cyst development. However, despite lack of cilia throughout development in Kif3a mutants, duct dilatations do not occur before E17.5 (Cano et al., 2006). The pancreatic epithelium in Hnf1b mutants is dilated from E14.5, thus indicating that Hnf1b plays additional roles in duct morphogenesis. Conditional Hnf1b deletion at late embryogenesis also resulted in a polycystic pancreas postnatally, showing the direct requirement of Hnf1b in duct morphogenesis and further suggesting that the cystic phenotype is not a consequence of an early blocked endocrine differentiation (Magenheim et al., 2011). As in Hnf6^−/− pancreata (Pierreux et al., 2006), cysts in Hnf1b mutants were not associated with increased epithelial cell proliferation. In Hnf6 and Sox9 mutants, cyst formation also seems to occur by deregulation of cystic-associated genes, but the genes involved are different. Sox9 mutants were characterized by a decrease in Pkd2 expression (Shih et al., 2012), whereas Hnf6 mutants displayed downregulation of Pkhd1 (Pierreux et al., 2006). We observed a marked decrease in Hnf6 expression in mutant pancreatic ducts at E14.5 upon deletion in MPCs, which correlated with the finding that Hnf1b is recruited at E12.5 to Hnf6 regulatory sequences known to drive expression in the pancreatic endoderm (Poll et al., 2006). Similarly, in Hnf6 mutants, Hnf1b expression is reduced during a narrow time window and then re-induced at late embryogenesis (Pierreux et al., 2006), illustrating the existence of a complex Hnf1b-Hnf6 feed-forward loop involved in duct morphogenesis at least between E12.5 and E16.5. However, duct morphogenetic defects of Hnf1b mutants are not simply explained by this transient cross-regulation. Indeed, duct morphogenesis is more largely affected in Hnf1b mutants than in Hnf6 mutants, as cysts affect all types of ducts, and not only intralobular and interlobular ducts, as in Hnf6 mutants (Pierreux et al., 2006). Moreover, whereas Glis3 expression was unaffected in Hnf6-null pancreas (Kang et al., 2009b), we found that Glis3 is downstream Hnf1b, which is particularly interesting as both factors are associated with cystogenesis (Kang et al., 2009a,,b) and endocrine cell development (Kim et al., 2012). Taking advantage of Hnf1b ChIP-seq analysis on embryonic kidneys, we identified Glis3 and Cys1 as novel Hnf1b targets in pancreas. Cys1 was found decreased in Hnf6 and Hnf1b mutant livers, although a direct regulation of Cys1 by Hnf1b was not established in biliary ducts (Raynaud et al., 2011). These studies show that Hnf1b is an essential regulator of key ductal genes related to cyst development in different organs. However, the regulatory circuits operating in ducts can diverge: Pkhd1 expression decreases in both pancreas and kidney, but not in the liver of Hnf1b mutants; Pkd2 expression decreases in the kidney, but not in the pancreas of Hnf1b mutants (Gresh et al., 2004; Hiesberger et al., 2005; Raynaud et al., 2011). Our study uncovers a crucial transcriptional network in pancreatic ductal cells, in which Hnf1b exerts a prominent role.

We show loss of Ngn3 expression upon Hnf1b inactivation during the first and secondary transitions, as well as the recruitment of Hnf1b to putative regulatory regions of Ngn3. This demonstrates an essential role of Hnf1b in the specification of endocrine progenitors. Additional transcription factors, including Hnf6, Glis3, Pdx1, Foxa2 and Sox9, were found to directly regulate Ngn3 expression (Jacquemin et al., 2000; Kim et al., 2012; Oliver-Krasinski et al., 2009; Seymour et al., 2008). Binding sites for some of these factors in the 5′ regulatory region of Ngn3 are listed in supplementary material Fig. S6. Hnf6^−/− embryos exhibit markedly reduced numbers of Ngn3⁺ cells at mid-embryogenesis, associated with decreased Hnf1b expression (Jacquemin et al., 2000). By E17.5, Hnf1b expression is partially restored and reduction of Ngn3⁺ cells is clearly less pronounced when compared with earlier time points, supporting the notion that downregulation of Hnf1b in Hnf6 mutants might be important for Ngn3 downregulation (Maestro et al., 2003). Pdx1 also contributes to Ngn3 regulation and, together with Hnf6 (Oliver-Krasinski et al., 2009) and Glis3 (Kim et al., 2012; Yang et al., 2011), occupies an evolutionary conserved enhancer homologous to human cluster 1 (Lee et al., 2001). Ngn3 cluster 1 also contains putative Sox9-binding sites (at −3.3 kb), which were occupied by Sox9 in ChIP experiments on mPAC ductal cells (Lynn et al., 2007). However, ChIP on embryonic pancreata demonstrated that Sox9 was not bound to cluster 1, but to three other regions: one distal (at −4.0 kb) and two proximal (at −0.4 kb and −161 bp) (Seymour et al., 2008). Although there is a Hnf1-binding site in the homologous mouse cluster 1, Hnf1b, like Sox9, failed to transactivate this enhancer in transfected HePG2 cells (Oliver-Krasinski et al., 2009), which correlates with our findings showing no Hnf1b recruitment to this region by ChIP. Moreover, Hnf1b did not transactivate the Ngn3 promoter corresponding to −4864 bp to +88 bp (Ejarque et al., 2013), whereas this factor was able to activate the Ngn3 full promoter (−5800 to +40 bp) (Yang et al., 2011), suggesting that the Hnf1b-binding site we identified at −4890 bp might be important for Ngn3 activation. These studies support the existence of distinct enhancer modules with differential binding of essential transcription factors that contribute to the activation of Ngn3. Our results show the absolute requirement of Hnf1b for endocrine specification, placing this factor in a prominent position in the regulatory network controlling Ngn3 expression.

A model depicting Hnf1b functions during pancreas development is presented in Fig. 8. As increasing the pool of endocrine progenitors is a key step for the development of cell-based strategies for diabetes, these findings might be of clinical significance to improve in vitro protocols for cell-replacement therapies. Proper regulation of Hnf1b expression appears to be crucial for endocrine cell formation, and Hnf1b can be used as a marker of progenitor cells with the capacity to robustly produce endocrine cells in vitro. In addition, our results suggest that MODY might occur not only as a consequence of β-cell dysfunction, but also as a consequence of defects during development leading to diabetes later in life.

Mice carrying the Hnf1b-null allele (Hnf1b^tmsc1 known as Hnf1b^+/LacZ), with the LacZ gene replacing the first exon of Hnf1b (Barbacci et al., 1999), were maintained as heterozygotes. The Hnf1b conditional knockout (Hnf1b^tm1Ics denoted as Hnf1b^Flox/Flox) carrying LoxP sites flanking exon 4 (Heliot et al., 2013), Pdx1-Cre (Wells et al., 2007) and Sox9-CreER^T2 (Kopp et al., 2011) lines have been previously described. The R26R^YFP line (B6.129X1-Gt(ROSA)26Sortm1(EYFP)Cos/J) was from The Jackson Laboratory. 4-Hydroxytamoxifen (Sigma) was dissolved at 10 mg/ml in corn oil/10% ethanol and administrated intraperitoneally to pregnant females at a dose of 2 mg. Blood glucose levels were measured using the OneTouch Vita blood glucose meter (LifeScan), blood amylase using the Reflovet Plus analyzer and pancreatic amylase by reflotron assay (Roche). Animal experiments were conducted in accordance with French and European ethical legal guidelines and the local ethical committee for animal care.

Tissues were fixed, embedded in paraffin, sectioned and analyzed by histology, immunohistochemistry and TUNEL assay as described previously (Haumaitre et al., 2005). Primary and secondary antibodies are listed in supplementary material Table S2. The percentage of Hnf1b⁺ cells was quantified by counting the number of Pdx1⁺ cells that were also Hnf1b⁺, on at least five sections per pancreas (n=4). Quantification of Pdx1⁺ cells at E12.5 was performed with at least five sections per pancreas (control, n=4; mutant, n=4). More than 10,000 Pdx1⁺ cells were counted for each genotype. Quantification of amylase⁺, Sox9⁺, insulin⁺ and glucagon⁺ cell surface was performed using ImageJ software, on at least three sections per pancreas at E16.5 co-immunostained with DAPI (control, n=3; mutant, n=5). Primary cilia were analyzed by confocal microscope (LEICA TSC SPE).

Total RNA from embryonic pancreata was isolated using RNeasy Micro-kit (Qiagen) and reverse transcribed using the superscript II RT First-Strand Synthesis System (Life Technologies). qRT-PCR was performed using the Fast SYBR Green Master Mix (Life Technologies). Primer sequences are provided in supplementary material Table S2. The 2^−ΔΔCt method was used to calculate expression levels (Livak and Schmittgen, 2001), normalized to cyclophilin A and relative to wild-type cDNA from E15.5 pancreata. Values are shown as mean+s.e.m. Statistical significance was determined using Student's t-test (NS, not significant; *P<0.05; **P<0.005; ***P<0.001).

Chromatin immunoprecipitation (ChIP) experiments were performed as described previously (Heliot and Cereghini, 2012), adapted for E12.5 wild-type pancreata. Immunoprecipitated chromatin were analyzed by qPCR and expressed as fold-enrichment values over IgG control. Primer sequences are provided in supplementary material Table S2. Values are shown as mean+s.e.m. Statistical significance was determined using Student's t-test (NS, not significant; *P<0.05; **P<0.005; ***P<0.001). A detailed protocol is provided in the methods in the supplementary material.

We thank C. Wright (Vanderbilt University, TN, USA), P. Jacquemin and F. Lemaigre (De Duve Institute, Belgium), and S. Louvet (UMR7622, France) for antibodies; the Centre d'Expérimentation Fonctionnelle (Faculté de Médecine Pitié-Salpêtrière, France) for blood amylase measures; mouse facilities of the UMR722-IBPS; and S. Gournet for illustration.