Bile system morphogenesis defects and liver dysfunction upon targeted deletion of HNF1β

Liver organogenesis starts around day 10 of mouse embryonic development and is completed after birth. This process requires the simultaneous differentiation of several cell types, including hepatocytes and biliary epithelial cells (BEC), and the formation of precise three-dimensional structures. Several groups have attempted to unravel the genetic program that controls hepatocytes and BEC differentiation by characterizing transcription factors enriched in these cells (Cereghini, 1996). Among these factors, the homeoprotein HNF1β/vHNF1 (hepatocyte nuclear factor 1β or variant hepatocyte nuclear factor 1) presents an interesting expression pattern. It is detectable in the hepatic bud and is abundantly expressed in the gallbladder primordium (Coffinier et al., 1999a). It is detected later in the intrahepatic bile ducts (IHBD) as they develop within the liver.

At the adult stage, HNF1β is strongly expressed throughout the biliary system. It is also expressed in periportal hepatocytes. This expression pattern distinguishes HNF1β from the closely related HNF1α protein that is uniformly distributed in all hepatocytes, and is expressed at lower levels in BEC (Pontoglio et al., 1996). The difference in expression of the two genes is interesting in light of the liver zonation phenomenon (Jungermann and Katz, 1982). Hepatocyte activities are specialized along the porto-central axis; catabolic metabolism occurs in the periportal hepatocytes, whereas anabolic functions are performed by hepatocytes closer to the central vein. Thus, the balance between the two HNF1 activities along the lobules may determine which hepatic genes are activated at a given position.

In addition to its hepatic expression, Hnf1β (Tcf2 – Mouse Genome Informatics) is strongly expressed in several epithelia organized in tubules, such as the pancreatic exocrine ducts and the kidney tubules (De Simone et al., 1991; Lazzaro et al., 1992). As HNF1β is expressed from the very onset of formation of such structures, it might play a role during differentiation and organogenesis (Coffinier et al., 1999a; Ott et al., 1991). Furthermore, gene inactivation demonstrated an early requirement for Hnf1β in the differentiation of another epithelium, the visceral endoderm (Barbacci et al., 1999; Coffinier et al., 1999b). HNF1β mutation blocks the differentiation of visceral endoderm and prevents its specialization into the two cell types that are specific for the embryonic and extra-embryonic territories. As visceral endoderm fails to differentiate, HNF1β-null mice die around the time of gastrulation, preventing further study of the role of HNF1β during later differentiation events.

To address the role of HNF1β in liver organogenesis, we generated a conditional allele of the gene and performed its tissue-specific inactivation in the liver using the Cre/loxP system (Gu et al., 1994). The loss of HNF1β in both hepatocytes and bile ducts resulted in a severe phenotype, including growth retardation and jaundice. Histological analysis of the gallbladder and the larger IHBD revealed epithelial abnormalities. We also observed a strong decrease in the number of the smaller IHBD and a persistence of the ductal plate from which IHBD are formed. These data suggest a requirement for HNF1β in biliary epithelium formation from the onset of the biliary system development. Biochemical and molecular studies of hepatic markers led to the identification of two HNF1β specific target genes, Oatp1 (Sla21a1 – Mouse Genome Informatics) and Vlcad (Acadvl – Mouse Genome Informatics), both of which are involved in lipid metabolism in hepatocytes.

A 5.8 kb genomic fragment containing the promoter and the first exon of Hnf1β was previously isolated from a mouse 129SVJ genomic library (Coffinier et al., 1999b). A conditional allele was generated by flanking the first exon with loxP sites for the Cre recombinase. A loxP site was inserted into a HincII site located 400 bp upstream of the transcription start site (Power and Cereghini, 1996) and a pgk-neo selection gene flanked by two additional loxP sites was introduced into a KpnI site located 150 base pair downstream of the first exon. All three loxP sites were introduced in the same orientation as verified by sequencing. In addition, a pgk-tk gene isolated from pPNT was added downstream to the right recombination arm to allow double (Neo+ tk−) selection (Tybulewicz et al., 1991). CK35 ES cells (Kress et al., 1998) were electroporated with the linearized vector. After selection, G418-resistant clones genotype was probed by Southern blot analysis after XbaI digest using the two external probes previously described (Coffinier et al., 1999b), and five correctly recombined clones were identified. Cre-mediated recombination was obtained after electroporation of 10⁷ recombinant ES cells using 15 μg pIC-Cre vector (Gu et al., 1993). After low-density plating, individual colonies were cultured in duplicate, in the presence or in absence of 300 μg/ml G418. After a 7-day culture, G418-sensitive clones were amplified and genotyped by PCR using primers flanking the loxP sites (VHP9, GGG GTG GCC TGC TCT AGG TGG C; VHP8, GCT CTC CAG GTC CTC CTC AGC TCC G; VIR3, CTT TTC GCT GCA CCC ACC AGC C; EX11, GTC CAA GCT CAC GTC GCT CC; VIR4, CCA GGT CTT TGC AGA GAA CTG C) (Fig. 1A). Out of 240 clones isolated after Cre expression, 50 had lost the selection gene, including 48 Hnf1β^del and two Hnf1β^flox clones, as identified by PCR. Both clones containing a floxed first exon and a deleted loci were isolated. Mice strains carrying either a Hnf1β^flox or a Hnf1β^del allele were established on a mixed C57BL/6 × 129SVJ background, and genotyped by PCR using the primers described above.

The AlfpCre transgenic line (established on a FVB/N background) was used to produce HNF1β liver-specific inactivation (Kellendonk et al., 2000). β-galactosidase staining was performed according to standard procedures on animals resulting from the crossing of AlfpCre strain with the R26R strain (Soriano, 1999). Mice were genotyped using the primers described above and additional primers specific for the Hnf1β^lacZ allele (Coffinier et al., 1999b) and the Cre transgene (Lakso et al., 1992). Cre recombination efficiency was assessed on liver DNA by Southern blot analysis after HindIII digest using a KpnI-HindIII fragment located in the first intron as a probe. Control animals used in this study were of either Hnf1β^flox/+, Hnf1β^flox/+AlfpCre or Hnf1β^flox/lacZ. All animals were maintained under standard housing conditions.

Northern blot analysis were performed on 30 μg of liver total RNA or 10 μg of kidney total RNA using Hnf1β and Vlcad cDNA probes, and a Hprt probe to normalize the samples. Expression of the differentiation markers were studied using semiquantitative RT-PCR as described previously (Coffinier et al., 1999b). The PCR cycle number were estimated for each primer pair to assure linear range amplification. The primer sequences are available on request.

Representational difference analysis of cDNA was carried out following a protocol supplied by M. Hubank (Hubank and Schatz, 1994). PolyA mRNA were purified from 2-week-old mutant and control liver total RNA (Dynabeads, Dynal). Double-stranded cDNA was prepared using oligo dT primer (Universal RiboClone cDNA Synthesis System, Promega).

Blood total and conjugated bilirubin concentrations were measured by colorimetry (Sigma – 552/553). After overnight fasting, blood triglycerides and cholesterol concentrations were measured enzymatically (Sigma – 336 and 352). Blood albumin and biliary acid concentrations were determined by Vebiotel laboratory (94110 Arcueil, France). Organs were fixed in formaldehyde 4% or Bouin’s fixative, and embedded in paraffin using routine procedures. Sections (5 μm) were stained with Hematoxylin and Eosin.

Livers were frozen in isopentane cooled in a liquid nitrogen bath, and sectioned at 7 μm using a cryostat. Sections were air-dried overnight, fixed in acetone at 4°C for 10 minutes and washed in two changes of Tris-buffered saline and treated with a blocking buffer (TBS/normal goat serum 10%/bovine serum albumin 1%/Triton X-100 0.3%) for 30 minutes. Thereafter, sections were washed in TBS, incubated with primary antibody (rabbit anti-human keratin A0575, Dako or monoclonal anti-α smooth muscle actin 1A4, Dako; 1/250 dilution in TBS/normal goat serum 1%/BSA 0.1%/Triton X-100 0.3%) overnight at 4°C, washed again and incubated with secondary antibody (Envision+Peroxidase Rabbit, Dako or anti-mouse IgG, HRP-linked NA931, Amersham; 1/400 dilution) for 30-60 minutes at room temperature. Sections were stained with 3-amino-9-ethylcarbazole (ICN Biochemicals) and counterstained with Hematoxylin.

HNF1β null mutation resulted in an early embryonic lethality, owing to a defect in visceral endoderm differentiation (Barbacci et al., 1999; Coffinier et al., 1999b). We generated a conditional allele of the Hnf1β gene using the Cre/loxP system (Gu et al., 1994) to analyze potential functions at later stages of development. loxP sites for the Cre recombinase were introduced on both sides of the first exon followed by a PGK-neo selection marker also flanked by a loxP site (Fig. 1A). Removal of the neo selection cassette, by transient expression of the Cre should result in a conditional or floxed allele, as the remaining loxP should not interfere with gene expression (Fig. 1A). Recombination between the first and third loxP sites resulted in deletion of the first exon and creation of a null allele, Hnf1β^del. Two ES clones, one Hnf1β^del and one Hnf1β^flox, were injected into blastocysts to generate chimeric mice and heterozygous offspring. As previously reported, the Hnf1β^del/del mutation resulted in an early embryonic lethality (Coffinier et al., 1999b). By contrast, Hnf1β^flox/flox mice were born at a normal Mendelian ratio and were healthy and fertile. Thus, the Hnf1β^flox allele is equivalent to a wild-type allele and deletion by the Cre recombinase should result into a null mutation, Hnf1β^del.

In the liver, Hnf1β displays a complex expression pattern. It is strongly expressed in the biliary epithelial cells, and more weakly in the hepatocytes localized at the periphery of the hepatic lobules (Coffinier et al., 1999a). To inactivate the Hnf1β gene in both cell types, we required a Cre recombinase transgene expressed in these cells or in a common precursor. The AlfpCre transgene contains both albumin and α-fetoprotein regulatory elements, drives Cre recombinase expression in the hepatic bud as early as embryonic day 10 (E10) and targets both cell lineages (Kellendonk et al., 2000). To extend these initial data, AlfpCre activity was analyzed after crossing the AlfpCre transgenic mice with the ROSA26 Cre reporter line (R26R) carrying a lacZ gene that is activated after Cre-driven recombination (Soriano, 1999). Liver sections prepared at birth showed specific β-galactosidase staining in all hepatocytes and in all cells of the IHBD. By contrast no staining was detected in endothelial cells or portal vein mesenchyme (Fig. 1B). The homogenous staining suggested that efficient recombination occurred in both hepatocytes and biliary epithelial cells or more probably in the hepatoblast, the postulated common precursor of both cell types. Furthermore, these data confirm that Cre expression occurred before bile duct morphogenesis, which starts around E15, as differentiated biliary epithelial cells do not express the albumin gene (Shiojiri, 1997). In contrast to the uniform staining observed in the liver, the gallbladder sections showed only patches of β-galactosidase staining in the inner epithelium (Fig. 1C). As the gallbladder separates from the liver bud and stops expressing the albumin gene at day E10.5, the patchy staining suggests a shorter exposure to Cre activity, resulting in recombination events occurring only in a few precursor cells.

Liver-specific inactivation of the Hnf1β gene was achieved by crossing Hnf1β^flox/flox mice with animals carrying one Hnf1β null allele and an AlfpCre transgene (Hnf1β^lacZ/+AlfpCre) (Coffinier et al., 1999b). One recombination event per cell was sufficient to generate a homozygous null mutation in Hnf1β^lacZ/floxAlfpCre offspring. Pups presenting the four possible genotypes were born in normal Mendelian ratios (data not shown). Recombination efficiency at the Hnf1β locus was confirmed by Southern blot analysis of liver DNA using a probe distinguishing between all four Hnf1β alleles: wild type, Hnf1β^lacZ, Hnf1β^flox and Hnf1β^del (Fig. 1D). To confirm the loss of Hnf1β expression in Hnf1β^lacZ/floxAlfpCre mice, a northern blot analysis was performed on total liver RNA using a probe specific for Hnf1β transcript. No Hnf1β RNA was detected in the mutant (Hnf1β^lacZ/+AlfpCre) livers compared with controls, either Hnf1β^flox/+, Hnf1β^flox/+AlfpCre or Hnf1β^flox/lacZ (Fig. 1E). The loss of the transcript was tissue-specific as equal amounts of Hnf1β RNA were detected in total kidney RNA from both control and mutant animals (data not shown). Based on AlfpCre activity described above and on both DNA and RNA analysis, we concluded that Cre expression in the mutant mice led to the specific inactivation of the Hnf1β gene in hepatocytes and IHBD epithelial cells. In addition, our data suggested that partial deletion may have occurred in the gallbladder.

Liver-specific inactivation of the Hnf1β gene resulted in a clear phenotype. Hnf1β^lacZ/floxAlfpCre mice showed severe growth retardation, hypertrophy of the liver and chronic jaundice. Although some variations in the growth retardation were observed, the jaundice and the liver enlargement were fully penetrant among the mutant animals.

Body weight differences could be measured as early as day 2 after birth and increased with time (data not shown and Fig. 2A,B). The weight differences between mutant and control littermates were more dramatic by the time of weaning, reaching a growth plateau. In addition to their smaller size, mutants were cachectic with reduced adipose tissue and muscular atrophy. Despite the severity of their phenotype, most mutants survived for several months.

Jaundice, or high plasma levels of bilirubin, is a hallmark of hepatic or bile system damage. Bilirubin, an end metabolite of heme degradation, is formed in the liver and further processed into a conjugated form before secretion into the biliary system. Bilirubin conjugation is catalyzed in hepatocytes by uridine diphosphate glucuronosyltransferase (UGT). Accumulation of unconjugated bilirubin is an indicator of hepatic failure, whereas accumulation of conjugated bilirubin implies a post-hepatic defect, such as bile duct obstruction. Increased bilirubin levels were detectable 1 week after birth and reached 30-fold excess at 2 month stage (Fig. 2C). Despite this dramatic excess, conjugated bilirubin represented similar percentages of the total bilirubin in mutant and in control serum (mutants: 64.0±8.1%, n=5; controls: 71.5±10.2%, n=3). These results indicated a defect downstream of bilirubin processing. In addition, accumulation of conjugated bilirubin suggested that the conjugation machinery functions correctly in HNF1β mutant hepatocytes as a large excess of bilirubin was processed. Bile acids, other bile components generated by the catabolism of cholesterol, also showed a dramatic increase in adult serum (Fig. 2D).

Despite a probably incomplete deletion of the Hnf1β gene in the gallbladder, as suggested by the data reported in Fig. 1E, its morphology was strongly affected in mutant animals and was characterized by an irregular shape interrupted by several constrictions (Fig. 3A-C). In some extreme cases, the cystic duct was completely dilated and no duct was really identifiable; at the same time, the common bile duct was affected and the boundary between the two ducts could not be determined (Fig. 3C). Histological staining of tissue sections revealed that the inner layer of the gallbladder presented areas with disorganized epithelia in place of a typical cuboidal epithelium (Fig. 3D,E). Some of the abnormal cells presented a morphology reminiscent of mucus-secreting cells (Fig. 3E). Despite these abnormalities, the gallbladders were filled with bile and communicated normally with the liver and the duodenum. We conclude that extrahepatic biliary obstruction was not the cause of cholestasis.

Analysis of the IHBD revealed a dysplasia of the larger bile ducts and a reduced number, or paucity, of small IHBD. Immunostaining of liver sections with anti-cytokeratin (CK) antibodies specific for biliary cells (van Eyken et al., 1987) showed that, in two-month old mutant mice, most portal tracts did not contain any recognizable bile ducts (Fig. 4A,B). As this defect is already seen in 1-week-old mutant mice, it indicated the lack of formation of interlobular bile ducts. The larger bile ducts still present in mutant livers were abnormal. In contrast to normal bile ducts that are formed of a simple cuboidal epithelium, they were lined by a dysplasic multi-layered epithelium with delocalization of the nuclei from the basal membrane, both indicative of epithelial disorganization (Fig. 4C,D). These observations provide the first evidence that HNF1β participates in the genetic program that directs IHBD morphogenesis.

A lack of formation of interlobular arteries was also seen in mutant mice and confirmed by anti-α smooth muscle actin immunostaining (Fig. 5). As Hnf1β is not expressed in endothelial cells, this defect is probably a secondary consequence of the mutation. This observation points to a link between arterial and biliary development within the liver. As suggested by other groups, arteries-bile ducts interaction may play an important role in establishing three-dimensional liver structures during organogenesis (Desmet, 1992).

To further analyze IHBD development in mutant mice, we stained liver sections at different stages with anti-cytokeratin (CK) antibodies. IHBD formation begins with the differentiation of hepatoblasts located at the interface between portal mesenchyme and liver parenchyma (Desmet and Callea, 1990). At E15.5, a continuous layer of CK-positive cells, also called ductal plate, is detected along the periportal mesenchyme. From E16.5 onwards, the bile ducts begin to differentiate by forming a lumen within the ductal plate that is surrounded by CK-positive cells (Fig. 6A,B). Bile duct differentiation progresses by the incorporation of the ducts within the periportal mesenchyme, while the ductal plate located between two adjacent bile duct regresses. By day 10 after birth (P10), bile duct differentiation is almost complete, although some remnant ductal plate can still be seen [see accompanying paper (Clotman et al., 2002) and Fig. 6E]. In contrast to the pattern observed for the control animals, E17.5 mutant embryos showed a very disorganized ductal plate with irregular lumens and the layer of CK-positive cells surrounding lumens displayed interruptions (Fig. 6C,D).

IHBD developmental defects resulted in a very disorganized biliary system. For each mutant animal (n=4), less than 5% of the portal tracts presented a normal duct pattern at day 8. 42±4% of the portal tracts lacked any bile duct-like cells or showed few CK-positive remaining from the ductal plate (Fig. 6F). By contrast, 46±9% of portal tracts presented strong CK staining, either as remnants of the ductal plate or as clusters of CK-positive cells incorporated in the portal mesenchyme but lacking a normal lumen (Fig. 6G). In some cases, dysplasic bile ducts were also observed (Fig. 6H). In conclusion, cell-specific Hnf1β deletion led to important morphogenetic defects of the developing IHBD. The paucity of IHBD observed postnatally in the biliary tract is probably caused by a failure to organize normal ducts. Therefore HNF1β plays an essential role in the differentiation of the IHBD.

To further probe the consequence of HNF1β inactivation in the liver, a panel of biochemical parameters was analyzed in the serum of both mutant and control littermates. General hepatic functions were assessed by measuring the level of serum albumin. No significant difference was detected between the mutants (22.2±2.9 g/l; n=4) and controls (25.7±3.1 g/l; n=4). However, in addition to the accumulation of bilirubin and bile acids described above, the levels of serum cholesterol and triglycerides were found to be dramatically increased, as early as one week after birth (Fig. 7A,B). These levels increased between 1 and 2 weeks of age and then became stabilized. Bile represents a major route for the excretion of organic solutes, such as bilirubin. Furthermore, cholesterol degradation into bile acids and biliary cholesterol secretion are the major pathways to eliminate excess cholesterol (Repa and Mangelsdorf, 2000). Thus, bilirubin, bile acids and cholesterol accumulation observed in mutant mice could result from a reduced bile flow due to IHBD paucity. These accumulations could also be due to the combined effect of a hepatocytic defect and a bile flow defect. Indeed, high levels of triglycerides suggested that hepatocyte metabolism may be affected by Hnf1β liver-specific deletion.

To characterize the hepatocyte defects and to identify liver-specific HNF1β target genes, we analyzed the expression of a number of hepatic genes by RT-PCR on total liver RNA. The expression of the liver-enriched transcription factors, Hnf1α (Tcf1 – Mouse Genome Informatics), Hnf4α, Hnf3β (Foxa2 – Mouse Genome Informatics) and Hnf6 (Onecut1 – Mouse Genome Informatics) was not significantly affected by the loss of HNF1β (data not shown). The expression of the secreted protein Transthyretin, the epithelial marker E-cadherin and the hepatocyte-specific connexin 32 were also normal (data not shown). As lipid metabolism was strongly perturbed in the mutants, we studied the expression of the apolipoproteins AI, AII, AIV, B, CII, CIII and E, which were all expressed at similar levels between mutants and controls (data not shown). We also examined the expression levels of molecules involved in bile formation. For example, bile acids are produced in the liver from cholesterol through two alternative pathways (Russell and Setchell, 1992). The cholesterol hydroxylase Cyp7α performs the first step of the major pathway, whereas it is carried out by Cyp27 in the alternative pathway. Both Cyp7a and Cyp27 were normally expressed in the mutant livers compared with controls (data not shown). Recent studies have demonstrated that biliary acid synthesis is tightly regulated by transcription factors of the nuclear receptor family. LXRα stimulates Cyp7α expression after binding to oxysterols (Janowski et al., 1996), whereas FXR, SHP-1 and LRH-1 participate in a negative feedback loop activated by bile acids (Goodwin et al., 2000; Lu et al., 2000). Surprisingly, despite the huge excess of bile acids accumulated in the mutant tissues, no variation was observed in the expression of these nuclear receptors between mutants and controls (data not shown).

Bile traffic involves numerous transporters (Kullak-Ublick et al., 2000; Trauner et al., 1998). Ntcp (Slc10a1), a Na⁺-dependent transporter, and the sodium independent transporters, Oatp1, Oatp2 and Oatp4 (Slc21a1, Slc21a5 and Slc21a6), participate in the reabsorption of bile acids from the blood in the sinusoids toward the hepatocytes. They play an important role in allowing the sensing of bile acid levels in the blood by hepatocytes. cMoat (canalicular multiple organic anion transporter; Abcc2 – Mouse Genome Informatics) and Bsep (bile salt export pump; Abcb11 – Mouse Genome Informatics) deliver, respectively, bilirubin and bile acids from the hepatocytes to the bile ductule. We found that Oatp1 was specifically downregulated in the absence of HNF1β, whereas the other six transporters remained almost unaffected (Fig. 7C and data not shown). This decrease in Oatp1 expression was observed in 2-week-old mice, suggesting that it could be a direct consequence of HNF1β deletion in hepatocytes and not a secondary effect of cholestasis. Interestingly, Oatp1 is also a target of HNF1α factor (Shih et al., 2001). Oatp1 downregulation could also contribute to the bile acid accumulation in the serum, as bile acids will then be less efficiently reabsorbed from blood.

Data resulting from the use of representational difference analysis (RDA) of cDNA (Hubank and Schatz, 1994) of mutant versus control livers led to the identification of another HNF1β target gene. The transcription of very long chain acetylCoA dehydrogenase (Vlcad), one of the four key enzymes of fatty acid oxidation (Beinert, 1963), was strongly decreased in the mutant samples (Fig. 7D). The Vlcad promoter (GenBank Accession Number, AJ012054) contains two HNF1 consensus binding sites located at positions –71 and –332, but its expression is not affected in the Hnf1α^–/– mice (data not shown). Therefore, VLCAD constitutes a HNF1β specific target gene. A defect in fatty acid oxidation could account for the accumulation of triglycerides observed in the mutant serums. As fatty acid oxidation is performed by hepatocytes, this further indicates that HNF1β plays specific roles in hepatocyte functions.

The liver fulfills many essential functions in mammalian organisms. The organ is formed from several cell types that are organized into a well defined three-dimensional structure. HNF1α and HNF1β were initially characterized as two closely related homeoproteins that activate the transcription of liver-specific genes. Subsequent studies showed that both proteins have a more extended expression pattern, including epithelial cells of the kidney, pancreas and intestine (Blumenfeld et al., 1991; Coffinier et al., 1999a; De Simone et al., 1991; Lazzaro et al., 1992; Pontoglio et al., 1996). Hnf1α is not essential for normal mouse development, but its absence results in hepatic, pancreatic and renal functional defects (Pontoglio et al., 1996; Pontoglio et al., 1998; Pontoglio et al., 2000; Shih et al., 2001). By contrast, Hnf1β is required for early mouse development. In particular, it is essential for the proper differentiation of the visceral endoderm during gastrulation (Barbacci et al., 1999; Coffinier et al., 1999b). In the present study we investigated the potential role played by HNF1β during development and organogenesis using cell-specific gene inactivation. We demonstrate that HNF1β is essential for the formation of a functional bile duct system and for several hepatic metabolic functions.

Replacement of the first exon of HNF1β with β-galactosidase followed by in situ staining has demonstrated that this gene is abundantly transcribed in the bile duct epithelium (Coffinier et al., 1999a). Much lower expression was observed in periportal hepatocytes. To inactivate the gene in both cell types, we crossed HNF1β floxed mice with an AlfpCre mouse. This mouse was shown to express the recombinase in the liver primordium from day 10 of gestation. At this time point, the Cre should be expressed in the hepatoblast that give rise to both hepatocytes and biliary duct cells. Crosses between the AlfpCre mouse and the ROSA26 floxed β-galactosidase reporter strain, showed that inactivation occurred in both cell types. Furthermore we have seen that partial inactivation occurs in the gallbladder epithelium.

Inactivation of HNF1β in precursors of intrahepatic biliary epithelium led to paucity of interlobular bile ducts and dysplasia of larger hepatic bile ducts epithelia. Immunohistochemical analysis showed that bile duct defects were already present at E17.5, and that ductal plate, from which IHBD develop, abnormally persisted after birth in mutant mice. Paucity of IHBD is thus directly linked to a developmental defect, demonstrating that HNF1β is essential for normal IHBD morphogenesis during liver formation. Similar features were observed in Hnf6^–/– mice, demonstrating that both HNF1β and HNF6 act in a transcription factor network involved in IHBD formation [see accompanying paper (Clotman et al., 2002)]. The results from Clotman et al., suggest that a HNF6→HNF1β cascade may be an important component of this network. These two complementary studies are the first to identify specific transcription factors involved in the IHBD differentiation program. In addition to the paucity of small IHBD observed in newborn, we observed a dysplasia of larger IHBD and of the gallbladder. The epithelium presented abnormal characteristics with a delocalization of the nuclei from the base of the cell, the organization of multiple layers of cells or the presence of ectopic epithelium. These observations suggest an effect of Hnf1β inactivation on the maintenance of epithelial differentiation. As previously suggested for the visceral endoderm, HNF1β could regulate fundamental characters of epithelium identity such as the basal position of the nuclei and the growth in monolayer of cells. Thus, HNF1β seems to have a dual function in controlling both the differentiation and the proliferation of certain epithelial cells.

Paucity of IHBD correlates, in mutant mice, with a lack of interlobular arteries. As Hnf1β is not expressed in endothelial cells, this defect is an indirect consequence of the mutation. These findings demonstrate that proper morphogenesis of bile ducts is necessary for arterial vascularization of the liver. Intercellular signals from biliary epithelial cells may act on endothelial or smooth muscle cells differentiation and/or proliferation to establish the correct three-dimensional liver organization.

Human autosomal dominant mutations in the Hnf1β gene result in a particular form of type II diabetes called maturity onset diabetes of the young type 5 (MODY5) (Horikawa et al., 1997). In some of the carriers of these mutations, kidney developmental defects, which are associated with internal genital abnormalities, were also observed (Lindner et al., 1999; Nishigori et al., 1998). Liver function of individuals with MODY5 was poorly investigated, owing to the low occurrence of cases. However, a liver function exploration in Japanese individuals with MODY5 reported high levels of γ-glutamyl transpeptidase, aspartate and alanine amino transferases, indicative of hepatic failure (Iwasaki et al., 1998). A case of hyperbilirubinemia was also described. These values were significantly higher than normal values and were not observed in individuals with MODY1/HNF4α or MODY3/HNF1α, suggesting that autosomal dominant HNF1β mutations may, at least in some cases, be specifically associated with liver dysfunction. Whether this liver dysfunction is linked to bile duct defects similar to those observed in mice, remains unknown.

Congenital diseases of IHBD can affect different levels of the biliary tree and can be characterized by dilatation or involution of the bile duct structures. All these diseases seem to have a developmental origin and to result from ductal plate remodeling problems, also called ‘ductal plate malformations’ (DPM) (Desmet, 1992). The hepatic phenotype of HNF1β-liver specific inactivation mice resembles, in some aspects, two human syndromes: Alagille’s syndrome, which is characterized by a paucity of interlobular bile ducts, and Caroli’s disease, a congenital non-obstructive dilatation of the larger IHBD. However, the phenotype of our mutant mouse does not perfectly match the description of these congenital diseases. Despite the divergence of phenotypes observed, study of mice lacking Hnf1β may provide some insights into the early development of DPM disease in humans.

Hnf1α and Hnf1β were initially characterized as liver-specific genes participating in a network of transcription factors. In vitro studies showed that both proteins bind the same DNA sequences with similar affinities. It has led to a model proposing that the balance between the strong transactivator HNF1α and the weaker HNF1β constitutes a major element in HNF1 target gene regulation (Baumhueter et al., 1988; Rey-Campos et al., 1991; Song et al., 1998). However, studies of knockout mice have revealed that the molecular basis for this regulation might be more complex. For example, in the Hnf1α^–/– mutants, the expression of several renal transporters were strongly affected despite the presence of large amounts of HNF1β in the same renal cells (Pontoglio et al., 2000), suggesting that HNF1α possesses distinct functions. We present evidence that in hepatocytes, despite the presence of large excess of HNF1α protein, HNF1β is necessary for activation specific target genes as shown by the decreased transcription of hepatic transporter Oatp1 and Vlcad enzyme in the mutant livers.

Oatp1 participates in the reabsorption of bile acids from the blood. Oatp1 loss of expression could contribute to the accumulation of bile acids in the blood. As Oatp1 is also a target of HNF1α (Shih et al., 2001), this gene may depend upon transactivation by HNF1α/HNF1β heterodimers. By contrast, VLCAD, a key enzyme of fatty acid oxidation, is affected by Hnf1β mutation but not by Hnf1α mutation, suggesting a requirement for HNF1β homodimers.

In conclusion, our results show that in hepatocytes HNF1β has a specific role not fully redundant with HNF1α. Moreover, metabolic abnormalities observed in the mutants correlate strikingly with the expression of the Hnf1β gene in the liver. Owing to the functional zonation of hepatocytes in the lobules, the periportal hepatocytes are the major site of bile excretion and of oxidative energetic metabolism, including fatty acid oxidation (Jungermann and Katz, 1982). As reported in a previous study, HNF1β is expressed in a subpopulation of hepatocytes located at the lobule periphery, whereas HNF1α has been shown to be expressed in all hepatocytes (Coffinier et al., 1999a; Pontoglio et al., 1996). Thus, only the periportal hepatocytes would be affected by the loss of HNF1β, which correlates with the alterations of biliary acids transport and fatty acid oxidation. We propose that the abnormal HNF1 activity in this subpopulation of cells, which is due to the lack of HNF1β, resulted in a strong decrease of Oatp1 and VLCAD expression. These observations indicate that HNF1β may participate to establish functional zonation within the liver by specification of periportal hepatocytes.

Finally, HNF1β is also expressed in other tubular structures such as pancreatic exocrine ducts and kidney tubules (Coffinier et al., 1999a). Moreover, this expression starts with the onset of differentiation of these structures. The role of HNF1β in bile duct formation and its specific expression pattern suggests that this gene could also be controlling pancreatic ducts and kidney tubule morphogenesis. HNF1β could be part of a general developmental program leading to establishment of epithelial tubular structures during organogenesis of several organs. This question may be addressed by studying crosses of the Hnf1β floxed mouse with other cell-specific Cre-expressing mice.

We are grateful to F. Lemaigre and his team for sharing results before publication. We thank M. Hadchouel and G. Kullak-Ublick for helpful discussion, P. Soriano for the ROSA26R mouse strain, H. Khun, S. Garbay, A. Doyen, P. Ave, J.-C. Benichou and J.-P. Bourgeois for advice, and C. Médaille from Vebiotel (Arcueil, France) for blood albumin and biliary acids measurement. We are grateful to J. Weitzman for critical reading of the manuscript. C. C. was a recipient of a fellowship of the Foundation Roux. L. G. was supported by a fellowship from the French ministry of Education and Research. M. Y. thanks the Kougie & Bert Vallee Foundation for support.