Suppression of Hepatocyte Proliferation by Hepatocyte Nuclear Factor 4α in Adult Mice*

Background: HNF4α is a key factor regulating hepatocyte differentiation and liver-specific functions. Results: Acute disruption of HNF4α in the adult liver causes rapid hepatocyte proliferation through direct and indirect control of multiple pathways. Conclusion: HNF4α is necessary to maintain the differentiated state of hepatocytes in adult liver. Significance: HNF4α expression maintains normal liver function, and its loss stimulates hepatocytes proliferation, possibly leading to cancer. Hepatocyte nuclear factor 4α (HNF4α) regulates genes involved in lipid and bile acid synthesis, gluconeogenesis, amino acid metabolism, and blood coagulation. In addition to its metabolic role, HNF4α is critical for hepatocyte differentiation, and loss of HNF4α is associated with hepatocellular carcinoma. The hepatocyte-specific Hnf4a knock-out mouse develops severe hepatomegaly and steatosis resulting in premature death, thereby limiting studies of the role of this transcription factor in the adult animal. In addition, gene compensation may complicate analysis of the phenotype of these mice. To overcome these issues, an acute Hnf4a knock-out mouse model was generated through use of the tamoxifen-inducible ErT2cre coupled to the serum albumin gene promoter. Microarray expression analysis revealed up-regulation of genes associated with proliferation and cell cycle control only in the acute liver-specific Hnf4α-null mouse. BrdU and ki67 staining confirmed extensive hepatocyte proliferation in this model. Proliferation was associated with induction of the hepatomitogen Bmp7 as well as reduced basal apoptotic activity. The p53/p63 apoptosis effector gene Perp was further identified as a direct HNF4α target gene. These data suggest that HNF4α maintains hepatocyte differentiation in the adult healthy liver, and its loss may directly contribute to hepatocellular carcinoma development, thus indicating this factor as a possible liver tumor suppressor gene.

Cancers of the liver and hepatobiliary tract are the seventh most common type of cancer worldwide (1). In the United States, liver cancers have one of the worst 5-year survival rates at 14.4% (2). Along with hepatitis B or C viral infection, the other risk factor for development of hepatocellular carcinoma (HCC) 4 is cirrhosis caused by chronic alcohol consumption (3). In addition, it is now recognized that metabolic syndrome, diabetes, and non-alcoholic fatty liver disease may contribute to the development of HCC due to the deleterious effects of these diseases on liver function (4). As a key regulator of liver homeostasis, disruption of hepatocyte nuclear factor 4␣ (HNF4␣, NR2A1) function results in many of these disorders that contribute to HCC.
HNF4␣ is a member of the nuclear receptor family of transcription factors. Originally designated as an orphan member due to the lack of a known ligand, subsequent studies have identified fatty acids bound in the ligand binding pocket with linoleic acid as the potential endogenous HNF4␣ ligand (5)(6)(7)(8). HNF4␣ is critical for proper hepatic lipid homeostasis, as evidenced by the development of steatosis in Hnf4␣-deficient mouse livers (9). Mutations in the human HNF4A gene result in maturity onset diabetes of the young type 1, a monogenic form of type 2 diabetes mellitus (10). In addition to regulation of multiple metabolic pathways in the liver, HNF4␣ expression maintains the differentiated state of hepatocytes. Conditional mouse knock-out models demonstrated the requirement of HNF4␣ expression to maintain proper liver architecture and many epithelial markers in the developing mouse liver (11,12). Disruption of HNF4␣ in mature hepatocytes results in epithelial to mesenchymal transition (EMT) (13). The EMT process is associated with tumor progression and metastatic potential (14). Thus, alterations in HNF4␣ functionality may increase the risk for development of HCC. Indeed, several mouse and human studies have shown that HNF4␣ expression is diminished in HCC (15)(16)(17)(18). Additionally, it was demonstrated that overexpression of HNF4␣ in rodent HCC models blocks carcinogenesis and metastasis, thus indicating a potential role for HNF4␣ as a tumor suppressor (16,19).
Due to its critical role in liver development, it has been difficult to study the role of Hnf4␣ in the adult mouse. The full-body Hnf4a knock-out mouse is embryonic lethal, and the liver-spe-cific albumin promoter-driven cre recombinase knock-out mice display severe hepatic metabolic disruption and death at 6 -8 weeks of age (9,20). In addition, the possibility exists that gene compensation may occur when Hnf4␣ is lost during early development, thus obscuring the potential effects of HNF4␣ in the adult. To overcome these obstacles, a temporal liver-specific knock-out of Hnf4a was generated using the ERT2cre system, which is activated upon tamoxifen exposure (21). Similar to the Alb-Hnf4a knock-out mouse, acute disruption of Hnf4a resulted in hepatomegaly and hepatosteatosis. Interestingly, microarray analysis identified the up-regulation of proliferative genes only in the acute knock-out. Further investigation revealed the impact of acute Hnf4a disruption on the cell cycle, apoptosis, and growth factors, thus suggesting a potential role for HNF4␣ as a hepatic tumor suppressor.

EXPERIMENTAL PROCEDURES
Animals and Treatments-The Hnf4a F/F and Alb-Hnf4a Ϫ/Ϫ mouse lines were described previously (9). To generate the conditional temporal Hnf4a F/F;AlbERT2cre mice, Hnf4a F/F mice were crossed with the tamoxifen-inducible hepatocyte-specific Cre recombinase expressing mouse SA ϩ/Cre-ERT2 (21). The mice were on a mixed SvJ129 and FVB background and used at 6 -8 weeks of age. For Hnf4a F/F;AlbERT2cre mice, floxed cre-negative or -positive littermates without tamoxifen treatment were used as controls. Mice were fed a diet containing tamoxifen (1 g/kg diet) for up to 5 days (days 1-5) and then returned to regular chow and euthanized at the indicated times (Fig. 1A). The Vhl F/F;AlbERT2cre mice were described previously and treated with tamoxifen for 5 days and then returned to a chow diet for 11 days prior to euthanasia (22). Hepatic deletion of VHL results in constitutive activation of HIF2␣ in the liver, leading to rapid inflammation, lipid accumulation, and fibrosis (22). For BrdU incorporation experiments, miniosmotic pumps containing sterile BrdU were implanted subcutaneously, and mice were euthanized 6 days later. Serum chemistry analysis was performed using the VetScan VS2 comprehensive and liver profiles (Abaxis, Union City, CA). For determination of liver bile acid levels, 20 mg of frozen liver was homogenized in 400 l of 75% ethanol, incubated at 50°C for 2 h, and then centrifuged. The supernatant (aqueous fraction) was retained, evaporated, and resuspended in 200 l of 0.9% saline. Twenty microliters was used for bile acid quantification using the VetSpec Bile Acids kit (Catachem, Oxford, CT). Mice were housed in a temperatureand light-controlled facility and given food and water ad libitum. All animal studies were performed in accordance with the guidelines and approval of the NCI, National Institutes of Health, Animal Care and Use Committee.
Immunohistochemistry-Liver tissue was fixed in 10% phosphate-buffered formalin for 24 h and then processed in paraffin blocks. Four-micrometer sections were used for H&E staining and immunohistochemistry. A rat anti-BrdU antibody (AbD Serotec, Oxford, UK) and a rabbit anti-human ki67 antibody (ab16667, Abcam (Cambridge, MA)) were used for immunohistochemical detection of proliferation.
RNA Analysis-Total RNA was isolated from frozen liver using the RNeasy minikit (Qiagen, Valencia, CA). One microgram of RNA was reverse transcribed and used for qRT-PCR analysis. Primers were designed for gene specificity and to cross exon-exon junctions using Primer-BLAST (NCBI). Results are normalized to actin gene expression. Values given are -fold over control or relative expression value, where appropriate, calculated using the 2 ⌬Ct QPCR calculation method (23).
Microarray Analysis-Dye-coupled cDNA was hybridized to Agilent 44K mouse 60-mer oligonucleotide microarrays (Agilent Technologies, Santa Clara, CA). Three mouse liver samples were independently processed for each genotype analyzed. Microarray data were processed and analyzed using Genespring GX 11.5.1 software (Agilent Technologies). Microarray data were deposited in the NCBI Gene Expression Omnibus under GEO accession number GSE34581. Only those genes changed Ն2.0-fold with a p value of Յ0.05 were considered significant and used for further analysis with Ingenuity Pathway Analysis (IPA) software (Ingenuity Systems, Redwood City, CA). For gene network analysis using IPA, networks were considered highly relevant with a score of Ն30. The network score is the Ϫlog of the p value determined by Fisher's exact test result, which takes into account the number of genes in the network, the size of the network, expression values of genes in the network, the number of genes in the data set, and the total number of genes in the IPA database.
Luciferase Assays-The mouse Perp promoter luciferase plasmids were constructed by cloning the upstream region from ϩ49 (relative to the transcription start site) to Ϫ578 and successive truncations into the pGL4-basic luciferase vector using primers listed in supplemental Table S1 (Promega, Madison, WI). The luciferase reporters and a Renilla luciferase control vector were cotransfected into HepG2 or COS-1 cells using Fugene 6 transfection reagent (Roche Applied Science). COS-1 cells, which do not endogenously express HNF4␣, were cotransfected with a control empty vector or a previously described rat HNF4␣ expression vector (24). Luciferase assays were performed using the Promega Dual-Luciferase assay kit.
ChIP Assay-Livers of Alb-Hnf4a F/F and Alb-Hn4a Ϫ/Ϫ mice (n ϭ 3) were perfused with Hanks' balanced salt solution (without Ca 2ϩ and Mg 2ϩ ) and EDTA followed by Hanks' balanced salt solution with Ca 2ϩ and collagenase. Hepatocyte slurries were washed in DMEM and filtered through a 70-m strainer and immediately fixed in 1% formaldehyde. Chromatin was prepared using the SimpleChIP Enzymatic Chromatin IP kit (Cell Signaling Technologies). Digested chromatin was incubated overnight with 4 g of anti-HNF4␣ (C-19, sc-6556x, Santa Cruz Biotechnology, Inc. (Santa Cruz, CA)). One microliter of recovered DNA was used for QPCR analysis. Primers were designed to span an ϳ100-bp region of the putative HNF4␣ binding site (supplemental Table S1). Values were normalized using 2% input control. Values given are -fold over the negative control IgG pull-down using the 2 ⌬Ct QPCR calculation method (23).
Statistical Analysis-All results are expressed as the mean Ϯ S.D. Significance was determined by t test or one-way ANOVA with Bonferroni post-test using Prism 5.0 software (GraphPad Software, La Jolla, CA). A p value of Ͻ0.05 was considered significant and is indicated in graphs (*, p Ͻ 0.05; **, p Ͻ 0.005; ***, p Ͻ 0.0005; ****, p Ͻ 0.00005).

Generation of Temporal, Hepatocyte-specific Hnf4a
Knock-out Mouse-Disruption of the Hnf4a gene in the hepatocyte-specific Alb-Hnf4a Ϫ/Ϫ mouse results in severe metabolic disturbances in lipid and bile acid handling, which contributes to the early mortality of knock-out mice at 6 -8 weeks after birth (9). In order to study the functions of HNF4␣ in the adult mouse liver without the confounding factors of early mortality and accumulated, severe hepatic malfunction, the Hnf4a F/F mice were crossed with the tamoxifeninducible hepatocyte-specific Cre recombinase expressing mouse SA ϩ/Cre-ERT2 to generate Hnf4a F/F;AlbERT2cre mice (9,21). To confirm the inducible recombination of the floxed Hnf4a allele, Hnf4a F/F;AlbERT2cre mice were fed a tamoxifencontaining diet for up to 5 days (day 0 start diet, day 5 remove), at which point the diet was replaced with a normal chow diet. Mice were euthanized on days 2, 5, 9, 12, and 19 after starting the tamoxifen diet. Analysis of Hnf4a mRNA and protein confirmed the complete loss of HNF4␣ in livers of Hnf4a F/F;AlbERT2cre mice within 48 h of tamoxifen treatment (Fig. 1, A and C). HNF4␣ expression in normal chow Hnf4a F/F or Hnf4a F/F;AlbERT2cre mice or tamoxifen-treated Hnf4a F/F mice was unchanged, confirming the specificity of the CRE gene activation. A trace amount of HNF4␣ protein is detectable 19 days after first tamoxifen exposure. The source of this protein may be proliferation and differentiation of hepatic stem/oval cells or loss of Cre recombinase expression in adult hepatocytes after several passages through the cell cycle. Loss of HNF4␣ transcriptional function was confirmed by the dramatic reduction in mRNA expression of classic HNF4␣ targets L-Fabp, ApoC3, and Ppara ( Fig. 1D) (9,25,26). Hnf4g mRNA levels remained low and even decreased slightly after Hnf4a disruption, eliminating the possibility of HNF4␥ compensation after HNF4␣ loss (Fig. 1B).
Short term knock-out of Hnf4a recapitulates the physiological changes observed in the developmental Alb-Hnf4a Ϫ/Ϫ mouse. Serum chemistry analysis of 19-day Hnf4a F/F;AlbERT2cre knock-out mice confirmed liver dysfunction, as revealed by increased levels of alkaline phosphatase, alanine aminotransferase, bile acids, and total bilirubin (Table 1). Similar to the Alb-Hnf4a Ϫ/Ϫ mouse, the Hnf4a F/F;AlbERT2cre mice also have decreased levels of serum cholesterol, a reflection of the known regulation by HNF4␣ of genes involved in hepatic cholesterol synthesis and transport (9,(25)(26)(27). Hepatomegaly is evident in Hnf4a F/F;AlbERT2cre knock-out mice as early as 5 days post-tamoxifen treatment, peaking at 9 days ( Fig. 2A). Although still significantly increased compared with control mice, the liver/ body mass ratio decreases slightly by 19 days. This decrease is not due to increased body mass but rather a reduction in liver mass (data not shown). Histological examination of 9-day Hnf4a F/F;AlbERT2cre ϩ tamoxifen mice reveals significant vacuolization of hepatocytes, hypertrophy, and scattered eosinophilic hepatocytes (Fig. 2, B and D). The full Alb-Hnf4a Ϫ/Ϫ mouse has more extensive eosinophilia and hypertrophy with slightly smaller vacuoles (Fig. 2, C and E). Previously published reports confirmed lipid deposition in the livers of both Hnf4a F/F;AlbERT2cre ϩ tamoxifen and Alb-Hnf4a Ϫ/Ϫ mice by Oil Red O staining (9, 28). Thus, the temporal hepatocyte-specific Hnf4a F/F;AlbERT2cre mouse model mirrors changes observed in FIGURE 1. Characterization of the temporal and tissue-specific Hnf4a F/F knock-out mouse. A, Hnf4a mRNA expression as determined by qRT-PCR. Livers (n ϭ 3) from Hnf4a F/F;AlbERT2cre mice (Cre ϩ) were collected at the indicated times after tamoxifen treatment. Nineteen-day Hnf4a F/F mice (Flox) were used as a control for tamoxifen exposure. Values were normalized to actin and represented as -fold change over expression in control Hnf4a F/F mice. Statistically significant changes from Hnf4a F/F control mice as determined by one-way ANOVA are indicated (***, p Ͻ 0.0005). B, qRT-PCR expression analysis for Hnf4a and Hnf4g was performed on RNA from 9-day tamoxifen-treated Hnf4a F/F;AlbERT2cre mice. Values shown are relative expression levels. Statistically significant changes from Hnf4a F/F;AlbERT2cre control mice as determined by one-way ANOVA are indicated (****, p Ͻ 0.00005). C, Western blot analysis of nuclear protein extracts from two mice per group. Blots were probed with anti-HNF4␣ antibody or the nuclear loading control TATA-binding protein (TBP). Nineteen-day tamoxifen-treated Hnf4a F/F mice were used as a control for tamoxifen exposure. D, qRT-PCR analysis of known HNF4␣ target genes in 19-day tamoxifen-treated livers (n ϭ 3). Values were normalized to actin expression and represented as -fold change over control Hnf4a F/F;AlbERT2cre . Statistically significant changes from Hnf4a F/F;AlbERT2cre control mice as determined by one-way ANOVA are indicated (***, p Ͻ 0.0005). Error bars, S.D.
the developmental Alb-Hnf4a Ϫ/Ϫ mouse with the added advantage of studying early responses to Hnf4␣ deficiency in healthy adult hepatocytes in vivo.
Hepatocyte Proliferation in Response to Acute HNF4␣ Deficiency-To determine the pathways influenced by HNF4␣, gene expression analysis was performed on livers from Hnf4a F/F;AlbERT2cre ϩ tamoxifen (19 day) and Alb-Hnf4a Ϫ/Ϫ mice and their respective wild-type controls (Fig. 3). The overall number of genes significantly altered in the two different knockouts was similar: 2388 genes in Hnf4a F/F;AlbERT2cre ϩ tamoxifen mice and 2535 in Alb-Hnf4a Ϫ/Ϫ mice. Those genes commonly regulated in both the Hnf4a F/F;AlbERT2cre ϩ tamoxifen (19 day) and Alb-Hnf4a Ϫ/Ϫ mice may be direct HNF4␣ transcriptional targets. The 1384 commonly regulated genes identified were further subjected to IPA gene network analysis with a score of 30 or greater considered highly relevant within the data set. As expected, the highest scoring network identified was lipid metabolism (Table 2). Interestingly, gene networks involved in cell cycle, cellular growth, proliferation, and the inflammatory response were also significantly affected by HNF4␣ deficiency. When microarray data were analyzed separately for the genotypes, cell growth and proliferation pathways were highly significant only in the Hnf4a F/F;AlbERT2cre ϩ tamoxifen-treated mice. Although gene network analysis indicated significant changes in cell growth and proliferation for both knock-out genotypes, the highest scoring proliferation network in the Alb-Hnf4a Ϫ/Ϫ mouse data set was 14, indicating that the Hnf4a F/F;AlbERT2cre ϩ tamoxifen data set was driving the identification of proliferation gene networks in the combined analysis (data not shown). The microarray data suggest that the increase in liver/body mass ratio in acutely HNF4␣-deficient livers may be due in part to hepatocyte proliferation and not solely to lipid accumulation. Hepatocyte proliferation was assessed by BrdU incorporation in the livers of both Hnf4a F/F;AlbERT2cre and Alb-Hnf4a Ϫ/Ϫ mice. Tamoxifen treatment resulted in a higher degree of BrdU labeling in Hnf4a F/F ϩ tamoxifen (9 day) livers compared with normal diet Hnf4a F/F mice (Fig. 4, A and E). The small, oblong nuclei labeled after tamoxifen treatment are characteristic of non-parenchymal cells, such as Kupffer or stellate cells. Acute disruption of Hnf4␣ expression in the Hnf4a F/F;AlbERT2cre ϩ tamoxifen mice resulted in widespread hepatic nuclei labeling (Fig. 4G), whereas limited foci of proliferation were evident in Alb-Hnf4a Ϫ/Ϫ mice (Fig. 4C). Proliferation in the livers of Hnf4a F/F;AlbERT2cre ϩ tamoxifen mice was further confirmed using the endogenous proliferation marker ki67 (Fig. 4H). Similar results were obtained with proliferating cell nuclear antigen (PCNA) staining (data not shown). Because the proliferation signal initiated after HNF4␣ deficiency is more robust in the acute knock-out model, the expression of cell cycle regulators in 9-day tamoxifen-treated Hnf4a F/F;AlbERT2cre mice was examined.
Consistent with the microarray data, several genes important in proliferation were up-regulated in Hnf4a F/F;AlbERT2cre knockout mice. Notably, the cyclins A2, B1, D1, and E2 (Ccna2, Ccnb1, Ccnd1, and Ccne2, respectively) as well as cyclin-dependent kinase 1 (Cdk1), myelocytomatosis oncogene (c-Myc), and Pcna were up-regulated (Table 3). Stem cell marker prominin 1 (Prom1/CD133) expression was up-regulated 72-fold after Hnf4a disruption, suggesting a loss of hepatocyte differentiation. The apoptosis-, senescence-, and cell cycle-related gene p21 (Cdkn1a) was induced 18-fold in Hnf4a F/F;AlbERT2cre knock-out mice. Gene expression of  A and B) and Alb-Hnf4a Ϫ/Ϫ mice (C and D) on a normal chow diet. On day 3 of tamoxifen treatment, a BrdU pump was implanted subcutaneously in Hnf4a F/F (E and F) and Hnf4a F/F;AlbERT2cre (G and H) mice. Tamoxifen was removed on day 5, and mice were returned to a normal chow diet until euthanasia on day 9. Slides were stained for BrdU incorporation (A, C, E, and G, dark brown) or ki67 (B, D, F, and H, dark brown) and counterstained with hematoxylin for visualization of nuclei (light purple). Experiments were performed on four animals per group, and representative images are shown. Examples of hepatocyte nuclei staining are indicated by arrows.

TABLE 2 Gene network analysis
Microarray results were processed through Ingenuity Pathway Analysis software to identify networks of genes most significantly affected by the phenotype and assigned a score based on the degree of relevance of the network to the genes in the data set. Networks with a score higher than 30 were considered highly relevant.
another member of the Cip/Kip family of cyclin-dependent kinase inhibitors, p27 (Cdkn1b), was reduced 40% in tamoxifen-treated Hnf4a F/F;AlbERT2cre mice. Expression of the p53/ p63-regulated gene Perp was dramatically suppressed in tamoxifen-treated Hnf4a F/F;AlbERT2cre mice. Up-regulation of cyclin D1 and p21 at the protein level was confirmed in Hnf4a F/F;AlbERT2cre knock-out mice by Western blot analysis (Fig. 5A). Together, these data indicate that HNF4␣ expression suppresses adult hepatocyte proliferation.
Lack of Inflammatory Signal after Acute Deletion of HNF4␣ in Liver-Hepatocyte proliferation in the context of partial hepatectomy or carbon tetrachloride exposure occurs in a two-step process: priming of the hepatocyte by inflammatory cytokines that in turn stimulates growth factor production and release (29,30). It was proposed that increases in bile acids after such liver injury is one mechanism to initiate the regenerative process via a nuclear receptor-dependent mechanism (31). Liver bile acid levels were maintained at control levels until 9 days post-tamoxifen in Hnf4a F/F;AlbERT2cre mice (Fig. 5B). Serum bile acid levels were statistically increased by 5 days of tamoxifen treatment in Hnf4a F/F;AlbERT2cre mice, reaching a maximum at 9 days of treatment (Fig. 5B). However, increased cyclin D1 gene expression precedes serum and, more importantly, liver bile acid elevation, and thus it is unlikely to be the initiating proliferative signal in Hnf4a F/F;AlbERT2cre mice (Fig. 5C).
It was recently shown that inhibition of HNF4␣ leads to a miRNA-mediated pathway resulting in stimulation of the IL-6-STAT3 inflammatory axis in the context of HCC (32). Expression of the two major inflammatory cytokines, Il-6 and Tnfa, was assessed by qRT-PCR. Basal expression levels were near the lower limit of detection and did not significantly increase at any time point after tamoxifen exposure in the Hnf4a F/F;AlbERT2cre mice (Fig. 6A). Consistent with the lack of cytokine induction, none of the major cellular signaling pathways were activated in Hnf4a F/F;AlbERT2cre mice after 2 days of tamoxifen treatment (Fig. 6B). This suggests that loss of HNF4␣ up-regulates a hepatotrophic factor independent of Kupffer cell activation to drive hepatocyte proliferation.
Loss of Hnf4␣ Induces Hepatotrophic Factor Bone Morphogenetic Protein 7 (Bmp7)-There are a wide variety of growth factors capable of stimulating hepatocyte proliferation, including hepatocyte growth factor (HGF), stem cell factor/kit-ligand (SCF/KITL), BMP7, insulin-like growth factor 1 (IGF-1), and the epidermal growth factor (EGF) receptor ligands EGF heparin-binding EGF-like growth factor, transforming growth factor ␣ (TGF␣), and amphiregulin (reviewed in Refs. 29 and 30). The expression of these growth factors was assessed in livers from 9-day tamoxifen-treated Hnf4a F/F;AlbERT2cre mice. Hgf, Igf-1, and Tgfa are the only mitogens studied that are normally produced at low levels in quiescent hepatocytes, as evidenced by the higher basal transcript levels in control Hnf4a F/F;AlbERT2cre mice (Fig. 7A). Tgfa was induced 1.75-fold in tamoxifen-treated Hnf4a F/F;AlbERT2cre mice, whereas Hgf and  Igf-1 were reduced to 0.44 and 0.21, respectively, compared with control Hnf4a F/F;AlbERT2cre mice. The reduction in Igf-1 expression would be expected as it is regulated by the transcription factor hepatocyte nuclear factor 1␣ (HNF1␣), whose expression is dependent on HNF4␣ (33)(34)(35). Egf expression was too variable in tamoxifen-treated Hnf4a F/F;AlbERT2cre mice to conclude a significant up-regulation. The most striking change was the induction of Bmp7, over 40-fold in Hnf4a-deficient mice.
To determine if this was a specific up-regulation of Bmp7 or a general activation of Bmps, expression of several members of the Bmp family, their receptors, and regulators was evaluated. Among the Bmps with detectable hepatic expression, only Bmp7 was significantly altered in tamoxifen-treated Hnf4a F/F;AlbERT2cre mice (Fig. 7B). Interestingly, expression of the Bmp antagonist Gremlin2 (Grem2/Prdc) was significantly reduced in tamoxifen-treated Hnf4a F/F;AlbERT2cre mice.
A recent study found elevated BMP7 levels in a hepatitis B mouse model of HCC and further identified elevated BMP7 in human cirrhotic liver and HCC samples (36). To assess whether the induction of Bmp7 in Hnf4a F/F;AlbERT2cre ϩ tamoxifen mice was associated with steatosis and liver damage or cellular proliferation, Bmp7 levels were measured in the Vhl F/F;AlbERT2cre mouse, which displays hepatic steatosis, inflammation, and fibrosis but not cellular proliferation (22). Transglutaminase 2 (Tgm2), a marker of fibrosis, is significantly up-regulated in Vhl F/F;AlbERT2cre ϩ tamoxifen mice, but Bmp7 levels remain nearly undetectable (Fig. 8A). Cyclin D1 mRNA, one marker for cellular proliferation, was slightly decreased from basal levels. This suggests that Bmp7 is not up-regulated by steatosis or fibrosis alone in mouse liver.
Little is known regarding the transcriptional regulation of Bmp7, and no studies have been conducted on its regulation in liver. A mouse Bmp7 promoter luciferase construct was induced by high glucose levels in cultured kidney cells (37). Fed serum glucose levels were measured at various time points after tamoxifen treatment in Hnf4a F/F;AlbERT2cre mice (Fig. 8B). Serum glucose levels do not exceed control levels at any point after Hnf4a disruption. The initial drop in glucose levels may be due to the tamoxifen diet. The Bmp7 induction, similar to cyclin D1, precedes the onset of cholestasis at day 5 (Fig. 8C). HNF4␣ has been shown to interact with the nuclear receptor corepressor 2 (SMRT/NCoR) complex, resulting in histone deacetylase recruitment and inhibition of transcription (38). The mouse Bmp7 promoter was assessed using Genomatix MatInspector software for HNF4␣ binding sites. Three puta-  tive binding sites were found within the first 3 kb of the promoter (Fig. 8E). To determine if HNF4␣ binds to the Bmp7 promoter in vivo, extracts were prepared from liver of Hnf4a F/F and Alb-Hnf4a Ϫ/Ϫ mice for the ChIP assay. The Alb-Hnf4a Ϫ/Ϫ liver extracts serve as a control for antibody specificity. The HNF4␣ binding site in the promoter of the coagulation factor 13B (FXIIIB) was used as a positive control (39). HNF4␣ binding was significantly enriched on Bmp7 site 1 in Hnf4a F/F liver compared with the negative control RPL30 intron 2 and liver extracts from Alb-Hnf4a Ϫ/Ϫ mice (Fig. 8D). However, this enrichment was minimal compared with the enrichment on the known HNF4␣ binding site in the FXIIIB promoter. Thus, it is unlikely that HNF4␣ recruits histone deacetylase complexes to actively repress Bmp7 transcription in normal adult liver.
Regulation of p53/63-dependent Desmosomal Gene Perp by HNF4␣-The potential repression of apoptosis in HNF4␣-deficient livers was assessed by Western blot analysis of caspase 3 or 9 activation. No significant caspase activation was detected, as indicated by caspase or poly(ADP-ribose) polymerase cleavage (Fig. 9A). There is a slight increase in full-length caspase 3 and 9 in the tamoxifen-treated Hnf4a F/F;AlbERT2cre mice that may indicate defective programmed cell death in these livers. Microarray and gene expression analysis revealed a significant down-regulation of the gene Perp (Table 2). PERP was first identified as an effector of p53-dependent apoptosis and is upregulated during apoptosis but not cell cycle arrest (40,41).
Because the decrease in Perp expression in the HNF4␣-deficient liver is of a similar magnitude as verified direct HNF4␣ targets, the first 2 kb of the Perp promoter were analyzed for potential HNF4␣ binding sites. Three putative binding sites (Ϫ220, Ϫ415, and Ϫ566 bp from the transcription start site) are located within the first 1 kb of the Perp proximal promoter (Fig.  9B). To confirm the presence of a functional HNF4␣ binding site, promoter luciferase constructs were tested in HepG2 cells that basally express HNF4␣. Luciferase expression in HepG2 cells was reduced when putative HNF4␣ binding site 2 (Ϫ415 bp) was deleted (Fig. 9B). In a parallel experiment, rat HNF4␣ was overexpressed in COS-1 cells that do not basally express HNF4␣. The full-length Perp promoter luciferase activity was significantly induced in COS-1 cells when HNF4␣ was co-transfected. This induction was completely lost when site 2 was deleted. Deletion of site 3 resulted in incomplete loss of luciferase activity, suggesting that this site cooperates with site 2 to achieve full transactivation of the Perp promoter. The ChIP assay from Hnf4a F/F mouse liver extracts confirmed recruitment of HNF4␣ to site 2 in the Perp promoter (Fig. 9C). The reduced binding signals at HNF4␣ site 1 and site 3 may be due to the resolution of the ChIP assay (DNA average length 200 bp) and close proximity of the sites. The specificity of the immunoprecipitation was confirmed by the absence of enrichment in Alb-Hnf4a Ϫ/Ϫ mouse liver extracts or the nonspecific gene RPL30.

DISCUSSION
The present study demonstrated that acute loss of HNF4␣ in the adult mouse liver initiates a robust proliferative response in normal hepatocytes, suggesting that HNF4␣ may be a true tumor suppressor. This response is transitory in nature because microarray analysis indicated that changes in cell cycle control and other genes involved in hepatocyte proliferation were associated with the Hnf4a F/F;AlbERT2cre knock-out and not with the Alb-Hnf4a Ϫ/Ϫ mice in which only small regenerating foci of hepatic periportal oval cells were observed previously (42). This latter finding was confirmed in the present study using BrdU incorporation. However, the widespread BrdU staining observed in the Hnf4a F/F;AlbERT2cre knock-out mice indicates that loss of HNF4␣ stimulates proliferation in both hepatocytes and oval cells. Although loss of HNF4␣ can initiate proliferation, the severe metabolic disruption in the hepatocytes may prevent continuous cell division. In order to progress to cancer, other genetic changes are probably necessary in conjunction with the loss of HNF4␣.
The findings in the Hnf4a F/F;AlbERT2cre knock-out mouse are in agreement with previous in vitro genomic and transcriptomic analysis identifying cell cycle-, apoptosis-, and other cancer-related genes as direct HNF4␣ targets (43). Among the genes identified in that study were the apoptosis-associated genes Cideb and Sec16b that were strongly down-regulated in FIGURE 8. Regulation of the Bmp7 promoter. A, qRT-PCR analysis of livers of Vhl F/F ϩ tamoxifen (TM) and Vhl F/F;AlbERT2cre ϩ tamoxifen mice. Results are expressed as a relative value, and statistically significant difference as determined by Student's t test from Vhl F/F ϩ tamoxifen is indicated (***, p Ͻ 0.0005). B, fed serum glucose concentrations (mg/dl). Statistically significant changes from Hnf4a F/F;AlbERT2cre control mice as determined by one-way ANOVA are indicated (*, p Ͻ 0.05). C, qRT-PCR analysis of Bmp7 mRNA from livers of Hnf4a F/F;AlbERT2cre treated with tamoxifen for 2 and 5 days. Significant differences from control Hnf4a F/F mice are indicated (*, p Ͻ 0.05; **, p Ͻ 0.005). D, ChIP assay with HNF4␣ antibody performed on liver extracts from Hnf4a F/F (WT) and Alb-Hnf4a Ϫ/Ϫ (KO) mice. QPCR values for immunoprecipitated DNA were normalized to input DNA and expressed as a relative value. Statistically significant changes from WT RPL30 as determined by Student's t test are indicated (**, p Ͻ 0.005; ****, p Ͻ 0.00005). E, schematic of the mouse Bmp7 promoter indicating potential HNF4␣ binding sites as well as the previously characterized Y-box protein binding site (YB-1). The position of the central nucleotide of the putative HNF4␣ binding site is given relative to the transcription start site (TSS). Error bars, S.D.
the Hnf4a F/F;AlbERT2cre knock-out mouse model as well (data not shown). In the current study, Perp was revealed as an HNF4␣ target gene in the liver by microarray, ChIP, and luciferase promoter assays. PERP is induced by p53 in response to apoptosis but not cell cycle arrest and is capable of inducing cell death (40,41). As well as being an important effector of apoptosis, PERP is a desmosomal protein essential for epithelial integrity, particularly in the skin (44). The regulation of Perp expression by HNF4␣ is consistent with the broad number of cell adhesion-related genes previously identified as HNF4␣ targets, including E-cadherin, claudins 1 and 3, and desmocollin 2 (11). Disruption of Perp promoted UVB-induced squamous cell carcinoma that was attributed to loss of PERP apoptotic functions as well as its desmosomal role (45). The cumulative effect of the loss of several apoptosis genes in Hnf4a F/F;AlbERT2cre knock-out mice may shift the steady-state equilibrium between apoptosis and proliferation. Indeed, Western blot analysis revealed a small increase in basal, full-length caspase 3 and 9 protein expression, which suggests that HNF4␣-deficient mice may have a defect in response to mitochondrial damage. It is difficult to interpret the significant increase in p21 expression observed in Hnf4a F/F;AlbERT2cre knock-out mice given the context-specific ability of p21 to both promote and inhibit apoptosis and the cell cycle (Fig. 5A) (reviewed in Ref. 46). The lack of activation of apoptosis and the increase in proliferation markers in Hnf4a F/F;AlbERT2cre knock-out mice suggest that p21 may be functioning in its antiapoptotic/proproliferation role by preventing caspase 3 cleavage and assisting with cyclin D-CDK complex assembly (reviewed in Refs. 46 and 47). p21 was previously identified in HepG2 and HCT116 cells as being positively regulated by HNF4␣ (48). The induction of p21 in the context of HNF4␣ depletion suggests that HNF4␣ does not significantly contribute to the induction of p21 in the whole liver.
The proliferative signal in Hnf4a F/F;AlbERT2cre knock-out mice is probably due to alteration of several pathways due to the pleiotropic nature of HNF4␣ target genes in hepatocytes. Although current models of hepatocyte regeneration suggest a two-hit hypothesis of inflammation and then proliferation, no changes in hepatic cytokine levels or activation of inflammation-associated signaling pathways were detected. However, it is possible that activation of these pathways occurred very rapidly and transiently and was thus missed during the present analysis. A recent report demonstrated increased proliferation and invasiveness and decreased apoptosis in HepG2 cells and mouse models of HCC in which HNF4␣ was knocked down (32). The proliferative phenotype was linked to HNF4␣ regulation of the IL-6/STAT3 signaling pathway through a miRNA feedback loop. However, no activation of STAT3 or Il-6 expression was observed in Hnf4a F/F;AlbERT2cre knock-out mice. This result suggests that the miRNA-mediated IL-6/STAT3 signaling pathway only functions in the context of active HCC and not in the healthy liver. HNF4␣ was previously shown to regulate the expression of miR-122, a highly expressed liver-specific miRNA important in hepatocyte differentiation and proliferation (49). Thus, HNF4␣ may regulate the expression of a large set of miRNA and RNAs in order to maintain the differentiated hepatocyte phenotype.
In adult mammals, BMP7 is a circulating cytokine produced primarily by the kidney and bone, but elevated expression has been detected in several tumor types, including breast, melanoma, and prostate (50 -53). Although the liver does not normally produce BMP7, BMP7 receptors are found on hepatocytes, and neutralization of circulating BMP7 in a mouse partial hepatectomy model inhibited hepatocyte regeneration (54). In the reverse experiment, these authors demonstrated increased liver proliferation after partial hepatectomy when recombinant BMP7 was infused into the mice. BMP7 was also able to antagonize TGF␤ and inhibit fibrosis in a mouse CCl4-induced liver fibrosis model (55). In contrast to the beneficial effects observed in the rodent liver fibrosis model, BMP7 expression is elevated in patients with chronic liver diseases, raising the question of species-specific effects of BMP7 (56,57). A recent study using a hepatitis B model of HCC in mice as well as human HCC samples confirmed up-regulation of BMP7 in cirrhotic and cancerous liver tissues (36). The same study also demonstrated increased cell viability, migration, and invasiveness in the hepatoma cell line Hep3B.
The mechanism by which loss of HNF4␣ induces Bmp7 remains elusive. It is likely that a transcriptional regulator of Bmp7 is up-regulated after HNF4␣ disruption. A recent report identified BMP7 as a target of MYC (myelocytomatosis oncogene) in medulloblastoma-derived cell lines, binding an unspecified location near the transcription start site (58). Indeed, a small 5.85-fold increase in c-myc mRNA was observed in 9-day Hnf4a F/F;AlbERT2cre knock-out mice. A transcription factor binding site search of the mouse Bmp7 promoter identified a putative E-box/c-MYC binding site Ϫ130 bp from the transcription start site. Additionally, several putative binding sites for the c-MYC-regulated protein E2-F1 are located within the proximal promoter of Bmp7. E2-F1 mRNA was previously shown to be up-regulated in the Hnf4a F/F;AlbERT2cre knock-out mice (28). It is interesting to note that Bmp7 was not induced in the Vhl F/F;AlbERT2cre mouse model, which displays significant steatosis, cholestasis, and fibrosis but no proliferation, suggesting that the mechanism of induction is linked to proliferation factors. Elevated Bmp7 expression in the livers of Hnf4a F/F;AlbERT2cre knock-out mice is consistent with reports of increased circulating BMP7 in patients with liver disease. Thus, BMP7 expression may be a marker of liver disease, and the Hnf4a F/F;AlbERT2cre knock-out mice may be a model for studying the role of BMP7 in liver function.
This study establishes HNF4␣ not only as a key regulator of hepatic differentiation, maintenance of the liver phenotype, and metabolism but also as a potential tumor suppressor. In an otherwise healthy liver, acute disruption of HNF4␣ in hepatocytes initiates a robust proliferative response. The combined effects of HNF4␣ regulation on cellular adhesion, EMT, growth factors, apoptosis, and proliferation position it as an important factor in hepatic tumorigenesis. The new temporal and spatial modulating Hnf4a knock-out mouse model generated in this study circumvents the early mortality observed in the developmental Alb-Hnfa Ϫ/Ϫ mouse, thus allowing for future studies on the role of HNF4␣ in fibrosis and tumorigenesis.