HNF1α Inactivation Promotes Lipogenesis in Human Hepatocellular Adenoma Independently of SREBP-1 and Carbohydrate-response Element-binding Protein (ChREBP) Activation*

Biallelic inactivating mutations of the transcription factor 1 gene (TCF1), encoding hepatocyte nuclear factor 1α (HNF1α) were identified in 50% of hepatocellular adenomas (HCA) phenotypically characterized by a striking steatosis. To understand the molecular basis of this aberrant lipid storage, we performed a microarray transcriptome analysis validated by quantitative reverse transcription-PCR, Western blotting, and lipid profiling. In mutated HCA, we showed a repression of gluconeogenesis coordinated with an activation of glycolysis, citrate shuttle, and fatty acid synthesis predicting elevated rates of lipogenesis. Moreover, the strong down-regulation of liver fatty acid-binding protein suggests that impaired fatty acid trafficking may also contribute to the fatty phenotype. In addition, transcriptional profile analysis of the observed deregulated genes in non-HNF1α-mutated HCA as well as in non-tumor livers allowed us to define a specific signature of the HNF1α-mutated HCA. In these tumors, lipid composition was dramatically modified according to the transcriptional deregulations identified in the fatty acid synthetic pathway. Surprisingly, lipogenesis activation did not operate through sterol regulatory element-binding protein-1 (SREBP-1) and carbohydrate-response element-binding protein (ChREBP) that were repressed. We conclude that steatosis in HNF1α-mutated HCA results mainly from an aberrant promotion of lipogenesis that is linked to HNF1α inactivation and that is independent of both SREBP-1 and ChREBP activation. Finally, our findings have potential clinical implications since lipogenesis can be efficiently inhibited by targeted therapies.

Hepatocyte nuclear factor 1-␣ (HNF1␣) 3 is a transcription factor that controls the expression of liver-specific genes, such as ␤-fibrinogen, ␣1-antitrypsin, and albumin (1). Heterozygous germline mutations in the gene encoding HNF1␣ (TCF1 (transcription factor 1)) are responsible for an autosomal dominant form of non-insulin-dependent diabetes mellitus called maturity onset diabetes of the young type 3 (MODY3), in which subjects usually develop hyperglycemia before 25 years of age (2). More recently, we identified HNF1␣ as a tumor suppressor gene involved in human liver tumorigenesis since we found biallelic inactivating mutations of this gene in 50% of hepatocellular adenomas (HCA) and in rare cases of well differentiated hepatocellular carcinomas developed in the absence of cirrhosis (3). HCA are rare benign primary liver tumors closely related to oral contraceptive intake (4).
Recently, in a comprehensive analysis of genotype-phenotype correlations in a large series of 96 HCA, we showed that HNF1␣ mutations define a homogeneous group of tumors phenotypically characterized by the recurrent presence of a marked steatosis (5). To get insight into the underlying molecular mechanisms that drive the fatty phenotype in human HNF1␣mutated HCA, we performed a transcriptome analysis using cDNA and Affymetrix microarrays. Gene expression profiles were compared between non-tumor livers and HNF1␣-mutated HCA. Among the differentially expressed genes, we focused our analysis on genes involved in lipid homeostasis, and we searched for a possible alteration in fat transport, degradation, and synthesis processes. Transcriptomic deregulations were further validated in an additional series of tumors, but also * This work was supported by a grant from the INSERM (Ré seau de Recherche at the protein level, and we analyzed the precise composition of accumulated lipids.

EXPERIMENTAL PROCEDURES
Patients and Samples-A series of 40 HCA, 25 non-steatotic non-tumor livers, and 11 steatotic non-tumor livers was collected in nine French surgery departments from 1992 to 2004. Liver tissues were immediately frozen in liquid nitrogen and stored at Ϫ80°C until used for molecular studies. Among the 38 patients with an HCA, the sex ratio (male: female) was 1:9, and the mean age was 37 (median ϭ 37 years ranging from 14 to 56). All the patients were recruited in accordance with French law and institutional ethical guidelines. The study was approved by the ethical committee of the Hôpital Saint-Louis, Paris, France. All HCA were screened for HNF1␣ and ␤-catenin mutation, pathological slides were reviewed, and all HCA were classified as described previously (5). Twenty-five HCA from 23 patients were HNF1␣-mutated, 11 demonstrated steatosis in more than two-thirds of hepatocytes, eight demonstrated steatosis in one-third to two-thirds of hepatocytes, three demonstrated steatosis in less than one-third of hepatocytes, and three were not steatotic. Fifteen HCA had no mutations in HNF1␣, four of them were ␤-catenin-mutated, five were inflammatory, and nine demonstrated steatosis. Non-tumor liver tissues were taken from patients resected with primary liver tumors developed in the absence of cirrhosis. Steatotic non-tumor livers demonstrated lipid storage either in more than two-thirds of hepatocytes (four cases) or in one-third to two-thirds of hepatocytes (seven cases), and were of various etiologies: hepatitis C virus infection (one case) dysmetabolic syndrome (six cases), alcohol (three cases), and unknown etiology (one case).
Microarray Analysis-Total RNA was extracted from frozen tissues using Qiagen RNeasy kits (Qiagen) according to the manufacturer's instructions. RNA integrity was assessed using RNA 6000 nano chips and the Agilent 2100 bioanalyzer. RNA quality control was performed as described previously (6). Transcriptional profiling of HNF1␣-mutated HCA and nontumor liver tissues was performed using two different microarray approaches. Minimum Information about a Microarray Experiment (MIAME)-compliant data (7) have been deposited in Gene Expression Omnibus (GEO) at NCBI and are accessible through GEO Series accession number GSE7473. The detailed procedure of analysis is provided in the supplemental experimental procedures.
Western Blotting-Total protein extracts were obtained after homogenization in radioimmunoprecipitation assay buffer (Santa Cruz Biotechnology). Nuclear extracts were prepared using the NE-PER nuclear and cytoplasmic extraction reagent kit (Pierce), and protein concentration was determined using a Pierce BCA protein assay kit. Primary antibodies were used at the following dilutions: rabbit polyclonal anti-L-FABP 1:2000 (a gift of Dr. J Gordon), mouse monoclonals anti-fatty acid synthase (anti-FAS) and anti-SREBP-1 (detecting both the SREBP-1a and the SREBP-1c isoforms) 1:500 (Pharmingen), polyclonals rabbit anti-glucokinase (anti-GK) (Santa Cruz Biotechnology) and anti-ACL (Cell Signaling Technology) 1:200, and polyclonal rabbit anti-ChREBP 1:1000 (Novus Biologicals). We used polyclonal rabbit anti-actin (1:3000, Sigma) and polyclonal rabbit anti-lamin A/C (1:500, Cell Signaling Technology) as loading controls to normalize the signal obtained for total and nuclear protein extracts, respectively. Detection of signals was performed using the ECL SuperSignal West Pico chemiluminescent substrate (Pierce) with either anti-mouse (1:4000, Amersham Biosciences) or anti-rabbit (1:2000, Santa Cruz Biotechnology) horseradish peroxidase-conjugated IgG as second antibodies.
Glucose-6-phosphate Measurement-Glucose-6-phosphate (G6P) concentrations were determined enzymatically in normal and tumor samples by exploiting the selective and quantitative conversion of G6P to 6-phosphogluconic acid in the presence of NADP ϩ and glucose-6-phosphate dehydrogenase, as described previously (9).
Histochemical Analysis of Glycogen Storage-For this analysis, we used liver tissues frozen through immersion in isopentane cooled in liquid nitrogen. Periodic acid Schiff histochemical staining was performed on frozen section from five HNF1␣-mutated HCA and five non-mutated HCA as well as their matched non-tumor livers.
Lipid Profiling-Detailed protocols used for lipid profiling are provided in the supplemental experimental procedures.
Statistical Analysis-All the values reported are mean Ϯ S.D. Statistical analysis was performed using GraphPad Prism version 4 software, and significance was determined using the nonparametric Mann-Whitney test for unpaired data. Difference was considered significant at p Ͻ 0.05.

RESULTS
Gene Expression Profiles in HNF1␣-mutated HCA-Our cDNA and Affymetrix microarray experiments analyzed the expression of a total of 15,000 different genes comparing, respectively, eight HNF1␣-mutated HCA with their corresponding non-tumor liver and five HNF1␣-mutated HCA with four non-tumor livers. Computational analysis of both experiments identified 375 and 222 genes, respectively, significantly down-and up-regulated in HNF1␣-mutated HCA. Among these genes, a large fraction was related to the normal hepatocyte function including carbohydrate and lipid metabolism, detoxification, and synthesis of secreted proteins such as complement and coagulation factors (supplemental Table 1). Among differentially expressed genes, promoter regions of 33 were previously identified to bound HNF1␣ in primary human hepatocytes using chromatin immunoprecipitation-on-chip (ChIP-on-chip) technology (10). All but four of these genes were down-regulated in HNF1␣-mutated HCA (supplemental Table 1), which is in accordance with the well known transactivating function of HNF1␣. In addition, we identified 82 genes showing a common pattern of expression with liver from hnf1␣-null mice (11) (supplemental Table 1). Interestingly, 19 of these common deregulated genes were involved in glucidolipidic metabolism and particularly in bile acid metabolism, cholesterol, and fatty acid synthesis (supplemental Table 1).
Fatty Acid Transport and Oxidation-In HNF1␣-mutated HCA, we identified an important modification in the expression profiles of apolipoprotein genes. Particularly, components of the high density lipoprotein particles encoded by APOM, LPAL2, APOF, LPA, and APOA4 were strongly down-regulated, whereas APOL3 was up-regulated (supplemental Table 1). In contrast, the expression of critical genes for hepatocyte very low density lipoprotein assembly and secretion such as APOB and MTP was not affected. We also found a dramatic decrease in the mRNA of FABP1 encoding the liver fatty acid-binding protein (L-FABP). This result was confirmed by quantitative RT-PCR (Ϫ68-fold change, Fig. 1A) and by Western blotting analysis that showed an absence of L-FABP in mutated HCA (Fig. 1B). In contrast, we did not find any change in the expression of genes encoding mitochondrial and peroxisomal ␤-oxidation enzymes. The expression level of genes encoding plasma membrane transporters such as CD36 was also normal in mutated HCA as well as the LDLR gene. However, we identified a number of deregulated genes related to the lipogenesis pathway ( Figs. 2A and 3A).
Gluconeogenesis Repression and Glycolysis Activation-The expression of four genes (FBP1, encoding the fructose-1,6bisphosphatase; PCK1 and PCK2, encoding the cytosolic and the mitochondrial forms, respectively, of the phosphoenolpyruvate carboxykinase (PEPCK) enzyme; and G6PT1 encoding the glucose-6-phosphate transporter 1) critical to promote hepatic glucose production was significantly decreased in HNF1␣-mutated HCA (Fig. 2B), indicating a repression of the gluconeogenesis pathway in these tumors ( Fig. 2A). In contrast, glycolysis was activated in HNF1␣-mutated HCA since we found a strong up-regulation of glucokinase at the mRNA (GCK, 11-fold) and protein level (GK, Fig. 2, B and C). In addi-tion, we found a 3-fold decrease in the expression of the GCKR transcript whose product acts as a negative regulator of glucokinase (Fig. 2B). According to a predicted increased glucose phosphorylation rate, we showed a significantly higher level of glucose-6-phosphate in HNF1␣-mutated HCA (Fig. 2D). Moreover, periodic acid Schiff staining revealed a glycogen overload in these tumors consistent with G6P increase (Fig. 2E). We also found an up-regulation of GPI mRNA (4-fold, Fig. 2B), which encodes the glucose phosphate isomerase, a second glycolytic enzyme. In contrast, PKLR mRNA, which encodes the liver-specific pyruvate kinase that catalyzes the last irreversible step of glycolysis, was 3-fold decreased in HNF1␣-mutated HCA (Fig. 2F). However, PKLR down-regulation may be balanced by a 2-fold increase of PKM2 gene that encodes the muscle-specific pyruvate kinase (Fig. 2F).
Activation of the Citrate Shuttle-Besides the observed gluconeogenesis inhibition and glycolysis activation that predict an overproduction of pyruvate, a substrate for the mitochondrial synthesis of acetyl-CoA, the citrate shuttle system that exports acetyl-CoA into the cytosol was also activated in HNF1␣-mutated HCA ( Fig. 2A). Indeed, we found a 12-fold overexpression of ME1 mRNA, which encodes the malic enzyme, and a moderate 2-fold increase of MDH1 transcript, and ACLY mRNA encoding the ATP citrate lyase (ACL) was 3.7-fold overexpressed in HNF1␣-mutated HCA (Fig. 2B). As ATP citrate lyase is crucial to provide the unique cytoplasmic source of the acetyl-CoA lipogenic precursor, we confirmed by Western blotting the ACL protein overexpression (Fig. 2C).
Stimulation of the Fatty Acid Synthetic Pathway-As shown in Fig. 3A, the fatty acid synthetic pathway was also stimulated  MAY 11, 2007 • VOLUME 282 • NUMBER 19 in HNF1␣-mutated HCA. We found an increase of the acetyl-CoA carboxylase transcript (ACACA, 2.6-fold) and a strong overexpression of fatty acid synthase at both the transcriptional (FASN, 7-fold) and the protein level (FAS, Fig. 3, B and C). In the downstream steps of fatty acid elongation and desaturation, we also identified an increase in the mRNA level of three genes encoding elongases including ELOVL1 (3.4-fold), ELOVL2 (2.2-fold), and ELOVL5 (2.3-fold) and of three genes encoding desaturases such as SCD (6-fold), FADS1 (3-fold), and FADS2 (4.7-fold, Fig. 3B).

HNF1␣ Inactivation Promotes Hepatic Lipogenesis
Transcriptional Alteration of Glucido-lipidic Genes Defines an Expression Pattern That Is Specific of HNF1␣-mutated HCA-To search whether the pattern of alteration in the expression of the glucido-lipidic genes was specific of HNF1␣ inactivation, we compared the expression profiles of the 19 aforementioned genes between HNF1␣-mutated HCA and a group of non-mutated HCA containing or not containing steatosis, using quantitative RT-PCR. In this analysis, we also included steatotic non-tumor livers from different etiologies and non-steatotic non-tumor livers. We then performed a non-supervised analysis of the results using a hierarchical clustering algorithm to group the genes as well as the samples on the basis of similarity in their expression pattern. This analysis accurately classified all samples in two major clusters according to their HNF1␣ status (Fig. 4). Steatotic non-tumor livers and non-steatotic non-tumor livers were gathered with the non-HNF1␣-mutated HCA. Gene cluster analysis identified two main groups corresponding to down-regulated (cluster A) and up-regulated (cluster B) genes in HNF1␣-mutated HCA (Fig. 4). As expected, four out of five genes identified in cluster A had been previously characterized as HNF1␣ transactivated targets. Interestingly, all up-regulated genes have never been previously found regulated by HNF1␣. Within cluster B, three subgroups of genes (subgroups 1, 2, and 3) showed a highly correlated overexpression in HNF1␣-mutated HCA (supplemental Fig. 1). For example, ACLY level of expression was correlated to ACACA overexpression in 16 HNF1␣-mutated HCA (Spearman r ϭ 0.72, p ϭ 0.001) (supplemental Fig. 1).
In non-HNF1␣-mutated liver tissues, only few genes demonstrated a significant transcriptional deregulation (supplemental Fig. 2). In steatotic non-tumor livers, expression of only two genes, PCK2 and ME1, followed an expression pattern as seen in HNF1␣-mutated HCA (supplemental Fig. 2). In non-HNF1␣mutated steatotic HCA, nine genes involved in gluconeogenesis, glycolysis, and citrate shuttle were significantly deregulated when compared with the non-steatotic non-tumor livers (supplemental Fig. 2). One of these, GCK, showed an inversed pattern of deregulation when compared with the HNF1␣-mutated  HCA. However, except for a modest elevation in ELOVL1 transcript, expression of genes related to the fatty acid synthetic pathway were unchanged in this group of steatotic adenomas. Finally, in the group of non-HNF1␣-mutated non-steatotic HCA, only four genes showed a significant change in expression pattern (FBP1, GPI, ACLY, ELOVL1).
These results indicated that the patterns of expression of the 19 analyzed genes were specific to HNF1␣ inactivation. Downregulations could be directly attributed to the loss of the classical HNF1␣ transactivation activity. In contrast, overexpressed genes demonstrated a coordinated activation, but none are known to date to be directly controlled by HNF1␣.
Activation of Lipogenesis in HNF1␣-mutated HCA Is Independent of SREBP-1 and ChREBP Activation-Among the lipogenic genes whose expression was increased in HNF1␣-mutated HCA, 8 out of 13 were previously described as SREBP-1c targets. However, using quantitative RT-PCR, we did not find any significant variation in the mRNA level of total SREBP-1 in HNF1␣-mutated HCA when compared with non-tumor livers nor in SREBP-1a (Fig. 5A). Moreover, surprisingly, Western blotting studies revealed a decrease in the nuclear active form of SREBP-1 in the tumors when compared with their corresponding non-tumor liver (Fig. 5B), and the microarray study also found an increase in INSIG1 (insulin-induced gene 1) mRNA (2-fold, supplemental Table 1) whose product is known to inhibit SREBP-1 cleavage processing. We then looked for a potential alteration of ChREBP expression, another key transcriptional activator of fatty acid synthesis in the liver. RT-PCR results showed a significant decrease in the mRNA of CHREBP in HNF1␣-mutated HCA (Ϫ1.8-fold, Fig. 5A), and Western blotting studies revealed a decrease of the protein level in the nucleus (Fig. 5B). Using the same samples where we demonstrated a repression of both SREBP-1 and ChREBP, we found an obvious overexpression of the FAS and GK proteins, which are known to be transactivated by these two transcription factors (Fig. 5B). Although unexpected, these results suggest that increased lipogenic gene expression found in HNF1␣-mutated HCA is not related to SREBP-1 and ChREBP activation. FAS and GK overexpression were confirmed in three additional cases of HNF1␣-mutated HCA (supplemental Fig. 3). In two of these three cases, ChREBP was decreased, and the nuclear active form of SREBP-1 was highly reduced or unchanged. In the remaining case, we found a mild elevation of ChREBP and nuclear SREBP-1 when compared with the non-tumoral counterpart. On the whole, we observed an increased expression of FAS and GK together with a clear repression of SREBP-1 and ChREBP in four out of six HNF1␣-mutated HCA analyzed. Interestingly, no similar pattern of expression was found in the non-HNF1␣-mutated HCA tested (supplemental Fig. 3). We also checked for the transcriptional level of other transcription factors that have been implicated in the regulation of glucido-lipidic metabolism (Fig. 5A). Except for a 1.9-fold elevation of PPAR␥ transcript, the mRNA level of FOXO1A (forkhead box O1A), PPAR␣, LXR␣, and HNF4␣ was not significantly different in HNF1␣-mutated HCA when compared with non-tumor livers (Fig. 5A). Finally, the RXR isotypes genes were not differentially expressed in the microarray analysis.
Total Lipid Composition Is Modified in HNF1␣-mutated HCA-We compared lipid profiles of four non-steatotic nontumor livers with that of five HNF1␣-mutated HCA. Total fatty acid content was 5.8-fold increased in HNF1␣-mutated HCA (Fig. 6A), and the fatty acid profile was modified. Saturated (SFA) and monounsaturated (MUFA) fatty acids were 7-and 8-fold increased, respectively (Fig. 6A), as a consequence of the accumulation of all SFA and MUFA species detected (Figs. 3A and 6, C and D). In contrast, polyunsaturated fatty acids (PUFA) were raised only 2-fold (Fig. 6A) with a preferential accumulation of fatty acids containing 18 carbon atoms including the two essential fatty acids linoleic acid (18:2n-6) and ␣-linolenic acid (18:3n-3) (Fig. 6E). Consequently, the SFA/PUFA and MUFA/ PUFA ratios in mutated HCA were significantly higher than in non-tumor livers (Fig. 6B). We also showed that C14 -C18 fatty acids were much more increased than C20 -C23 fatty acids, as indicated by a 5-fold higher C14 -C18/C20 -C23 ratio (Fig. 6B). We also identified a mild elevation in free cholesterol content in HNF1␣-mutated HCA (36 versus 26.2 nmol/mg of proteins, Fig. 6F); however, we did not find any change neither in total cholesterol esters and diglyceride contents (Fig. 6F) nor in specific molecular species related to these two neutral lipid classes (data not shown). In contrast, total and all species of triglycerides were dramatically increased in HNF1␣-mutated HCA (Fig.  6, F and G). This overload was inversely proportional to the   MAY 11, 2007 • VOLUME 282 • NUMBER 19 total number of carbon atoms of each species (ranging from a 53-fold increase for C49 to a 4.8-fold increase for C57 families, Fig. 6G), according to the highest increase in C14 -C16 fatty acid species versus C18 -C20.

HNF1␣ Inactivation Promotes Hepatic Lipogenesis
Phospholipid Profile Is Also Changed in HNF1␣-mutated HCA-As shown in Table 1, HNF1␣-mutated HCA showed a 2-fold increase in total phospholipid quantity that was related to a significant elevation of phosphatidylcholine (PC) and phosphatidylethanolamine (PE), whereas the content of the other phospholipid classes including sphingomyelin was unchanged when compared with non-tumor livers (except for PS that was undetectable in the tumors). Our results also showed a significant elevation in the PE fraction, whereas the PI fraction was reduced ( Table 1), indicating that phospholipid profile was qualitatively changed. The fatty acid profile of PE was not significantly modified, whereas it was moderately altered for PC and PS-PI classes ( Table 2). In PS-PI, we found a significant elevation of the MUFA fraction (Table 2). In PC, the SFA and PUFA fractions were decreased and increased, respectively; consequently, the PC-unsaturation index was significantly increased in HNF1␣-mutated HCA ( Table 2). We also showed that C20 -C24 were over-represented in the PC fraction of HNF1␣-mutated HCA, as indicated by the significant decrease of the C16 -C18/C20 -C24 ratio (1.8-fold). When analyzing fatty acid composition of the total phospholipids, we only found a reduction in the C16 -C18/C20 -C24 ratio in HNF1␣-mutated HCA (Table 2).

DISCUSSION
In the present work, careful analysis of the metabolic pathways altered in the HNF1␣-mutated HCA may explain the severe fatty overload observed in these tumors. First, we identified a repression of gluconeogenesis coordinated with an activation of glycolysis and citrate shuttle predicting an aberrant accumulation of the acetyl-CoA and NADPH lipogenic precursors (12). In addition, these tumors also showed an obvious stimulation of the fatty acid synthetic pathway from palmitate synthesis to the downstream steps of fatty acid elongation and desaturation. Taken together, these findings provide evidence that de novo lipogenesis activity is increased in HNF1␣-mutated HCA, and we showed that it is specifically related to the loss of HNF1␣ activity. The fact that the steatotic non-HNF1␣mutated HCA and the steatotic non-tumor livers did not show the same deregulated expression pattern as in mutated HCA suggests that lipid storage in these two groups of samples likely occurs independently of HNF1␣ inactivation. Moreover, since lipogenesis activity is known to be mainly regulated at the transcriptional level, it is unlikely that higher lipogenic rates would be a significant driver of fat accumulation in the steatotic non-HNF1␣-mutated HCA or in the steatotic non-tumor livers. Indeed, in these two sample groups, no major transcriptional deregulation was found in the lipogenic pathway. However, it does not exclude a possible activation of lipogenesis at a posttranscriptional level. The importance of lipogenesis in human steatosis is still under discussion. Here, we report a particular situation in humans in which hepatocyte steatosis developed in the context of HNF1␣ inactivation seems to be primarily due to an aberrant stimulation of lipogenic rate.
In mice, the loss of HNF1␣ activity was also associated with the development of fatty liver (13,14), and a prior work identified a significant elevation in the liver expression of the acetyl- shown that tumor cell survival may depend on de novo synthesis of fatty acids. Although the fatty acid synthase has been proposed as an interesting therapeutic target for cancer treatment, more recent studies showed that the inhibition of other lipogenic enzymes such as acetyl-CoA carboxylase or ATP citrate lyase can also limit cancer cell proliferation and survival (30,31). Recently, we proposed HNF1␣ as a tumor suppressor gene (3), and we demonstrated in the present work that lipogenesis activation in HNF1␣-mutated HCA results from the loss of HNF1␣ activity. This finding may contribute to reinforce the emerging concept that induction of lipogenesis observed in tumor cells would occur downstream from oncogenic events (29) and open new avenues to understand how metabolism deregulations may contribute to benign liver tumorigenesis. Finally, our findings have potential clinical implications since lipogenesis can be efficiently inhibited by targeted therapies.