Hepatocyte nuclear factor 1α suppresses steatosis-associated liver cancer by inhibiting PPARγ transcription

Hepatocellular carcinoma is the third leading cancer-related cause of death worldwide. This is partly due to late diagnosis and the fact that no efficient treatment is available. Recently, in the framework of the International Genome Consortium and The Cancer Genome Atlas, the largest genome profiling of liver cancers was conducted (1–4). Large-scale analyses, including exome sequencing, transcriptome, copy-number, and methylome analyses, uncovered a broad landscape of genetic alterations and highlighted the extraordinary diversity of benign and malignant liver lesions. Multiple molecular pathways were found dysregulated in hepatic lesions, including p53 and cell cycle regulators, WNT/β-catenin pathway, chromatin modifiers, and oxidative stress and growth factor signaling pathways. The latter group was found activated in the majority of malignant liver lesions due to mutations in RAS gene members; PIK3CA, PTEN, RPS6KB1, and RPS6KA3 genes; and growth factor tyrosine kinase receptors and ligands (MET, FGF19, VEGF, and IGF); and its activation was positively correlated with the severity of the disease (5, 6). In particular, the insulin signaling pathway has a major role in the metabolic adaptations to nutrition, including increased hepatic lipid accumulation (steatosis). Interestingly, overnutrition and obesity are also risk factors for liver tumorigenesis, in part by promoting an inflammatory environment favorable for cancer growth (7, 8). However, it is unclear whether and how genetic insults leading to steatosis provide hepatocytes with a cell-autonomous growth advantage in malignant transformation.

The transcription factor and lipid sensor peroxisome proliferator–activated receptor-γ (PPARγ, encoded by Pparg in mice) has received a lot of attention in the liver steatosis response, though its role in liver tumorigenesis remains to be clarified. PPARγ is known as a master regulator of adipocyte differentiation, consistent with its highest levels of expression and activity in adipose tissue, where it orchestrates lipid uptake, synthesis, and storage (9). However, in ob/ob and db/db mouse models of obesity, liver Pparg mRNA levels are substantially increased (10, 11). In addition, PPARγ expression is also induced by genetic insults, e.g., by the deletion of the PIP3-lipid phosphatase and tumor suppressor phosphatase and tensin homolog (Pten) gene in liver (12, 13). The involvement of PPARγ in the steatosis response in liver is implied as mice with hepatic deletion of Pparg are protected from high-fat diet–induced steatosis and show improvements in glucose tolerance (14). Yet, the accumulated data on the implication of PPARγ in tumorigenesis are not conclusive and are in some instances contradictory.

Depending on the cancer type, both tumor-suppressive and tumor-promoting functions for PPARγ were reported. While a tumor-suppressive role is described in colon, breast, and prostate cancers, PPARγ activation promotes polyp formation in colon cells carrying mutations in the APC gene (15–17). In liver, loss-of-function mutations of negative regulators of PPARγ, such as histone deacetylase 3 (HDAC3) and nuclear hormone corepressor (N-CoR), promote steatosis and pathological liver growth culminating in cancer in mice and humans (18, 19). Conversely, loss of 1 allele of Pparg sensitized mice to chemically induced liver tumorigenesis (20). Similarly, in the STAM mouse model of liver cancer, combining diabetes and high-fat diet, pharmacological activation of PPARγ significantly ameliorated liver damage and reduced tumor numbers without affecting tumor size or hepatocyte proliferation in nontumoral liver tissue (21). One possible explanation for these contradictory findings is the distinction between steatosis induced by genetic insults and that induced by environmental factors. Interestingly, a subclass of hepatocellular adenoma in humans is associated with loss-of-function mutations in the transcription factor hepatocyte nuclear factor 1α (HNF1α) and is signified by important steatosis of unconfirmed origin independent of nutritional status (22). The requirement for PPARγ depending on the liver cancer genotype and the relevance for human malignancies remain an open question.

In this work, we screened a large annotated collection of human liver cancers for PPARγ expression. We find that the expression and activity of PPARγ are significantly increased in benign lesions characterized by loss of function of HNF1A and a subset of malignant hepatic lesions characterized by activated Akt signaling. In functional studies in mice, we provide a link between genetic loss of Hnf1a and Pparg transcription, revealing HNF1α as a novel transcriptional repressor of PPARG under control of Akt2. Finally, preclinical studies in a Pten-deficient mouse liver cancer model demonstrate that PPARγ inhibition by genetic or pharmacological tools has potent antitumoral action. In sum, our findings show a novel functional interaction between 2 transcriptional master regulators of hepatocyte and adipocyte cell fate, HNF1α and PPARγ, in the integration of metabolism and hepatocyte transformation.

High PPARγ expression in a subset of human hepatocellular carcinomas. In our recent work, we demonstrated that aberrant upregulation of the transcription factor PPARγ downstream of activated PI3K/Akt2 signaling in liver resulted in coordinated activation of aerobic glycolysis and lipogenesis (13). We hypothesized that PPARγ might be relevant to human liver pathophysiology and could be abnormally expressed in a subset of hepatocellular carcinomas (HCCs) and adenomas (HCAs), thereby conferring on them an adaptive growth advantage. Hence, we interrogated a comprehensive collection of human benign and malignant liver lesions, which was previously characterized by transcriptomic, metabolomic, and genomic analyses with full medical annotations comprising etiology, sex, and ethnic origin (1–3). In total, transcript levels of PPARG were evaluated in 315 HCC and 117 HCA samples as compared with 52 nontumoral liver samples of tumor-bearing patients and 5 tissue samples of non–tumor-bearing patients. Quantitative real-time PCR analysis revealed that PPARG transcript levels were relatively increased in HCA and HCC as compared with the nontumoral liver tissue biopsies (Figure 1A). These analyses also revealed variations in PPARG levels of expression in the HCA and HCC lesions. We made similar observations by using publicly available microarray data sets GSE14520 and GSE36376, containing 246 human hepatocellular carcinoma (hHCC) and 231 nontumoral liver samples, and 240 hHCC and 193 nontumoral liver samples, respectively (Supplemental Figure 1A; supplemental material available online with this article; https://doi.org/10.1172/JCI90327DS1). From this analysis, 33% of hHCC samples in both sets showed high PPARG mRNA expression levels that are at least 4 SD above the mean expression in nontumoral liver tissue samples.

Figure 1

PPARγ is induced in a subset of human liver cancers. (A) Expression profiles of PPARG transcript in normal liver (N, n = 5), nontumoral liver (NT, n = 52), HCA (n = 117), and HCC (n = 315) biopsies by real-time quantitative PCR. Data are presented as a ribosomal 18S–normalized mean fold (log₂) compared with the mean value in nontumoral samples ± SEM, Mann-Whitney test. *P < 0.05; ***P < 0.001. (B) Kaplan-Meier analysis and the log-rank test depending on the expression profiles of PPARG in HCC biopsies. Groups are defined according to the PPARG transcript levels using a fold change of 3 as threshold. HCC (n = 237), Mann-Whitney test, P = 0.0051. (C) Immunoblot analysis of PPARγ levels in HCC biopsies of G3 (HIGH) versus other subgroups (LOW). Immunoblot with anti-GAPDH antibody served as a loading control. Total PPARγ signal was quantified. Data are means ± SEM, n = 6–7. *P < 0.05, 2-tailed, unpaired Student’s t test. (D) Immunoblot analysis of PPARγ levels and Akt signaling activation in HCC and in nontumoral (NT) liver biopsies. Immunoblot with anti-GAPDH antibody served as a loading control. Densitometric analysis of GAPDH-normalized total PPARγ signal is presented. Data are means ± SEM, n = 4–6. *P < 0.05 vs. NT; ^#P < 0.05 vs. LOW; 1-way ANOVA with Tukey’s multiple-comparisons test. The Pearson correlation between total PPARγ expression and Akt phosphorylation on Ser473 in individual HCC samples is 0.65. See all complete unedited blots in the supplemental material.

To gain further mechanistic insights into the observed increase in PPARG expression, we analyzed its relative transcript levels according to the G1–G6 transcriptional signatures that were previously determined in HCCs (23). We revealed that PPARG expression was maximal in the G3 group of hHCCs, followed by the G1 and G2 groups, while the G4 group did not differ from control (Supplemental Figure 1B). These results were encouraging since the G1–G3 subgroups of hHCCs, unlike G4–G6, are characterized by activated PI3K/Akt signaling (23). In addition, high PPARG mRNA expression in the liver cancer atlas collection was associated with worse survival in 237 HCC patients treated by curative surgical resection (Figure 1B). Next, we performed immunoblot analysis using anti-PPARγ antibody, to address whether the increased transcript level of PPARG is reflected in increased protein levels. Biopsy samples were selected based on the results of PPARG transcript analyses as the representative average values of the respective group. We observed that PPARγ protein was expressed in tumoral lesions and its levels were significantly higher in the hHCC biopsies of the G3 (HIGH) group as compared with the samples from other subgroups of hHCCs (Figure 1C). In murine and human cells, expression of multiple PPARγ isoforms was reported, with PPARγ1 and PPARγ2 being the best characterized (24). Interestingly, both PPARγ1 and PPARγ2 isoforms were upregulated, although PPARγ1 to a larger extent, in the G3 (HIGH) subgroup of hHCCs compared with other hHCCs.

Next, we asked whether increased PPARγ protein expression correlated with changes in Akt signaling in hHCCs, using a distinct biopsy collection. We found that, as expected, Akt is activated in the majority of lesions, as witnessed by the increased phosphorylation of Akt-Ser473 and its downstream target the proline-rich Akt substrate of 40 kDa (Pras40) (Figure 1D). Importantly, we revealed a positive correlation of total PPARγ protein levels and an activation of Akt signaling in hHCCs (Figure 1D).

In sum, these observations in hHCCs show that PPARγ is upregulated in a portion of liver tumors at both the transcript and the protein level and its induction positively correlates with increased Akt signaling.

Hepatic deletion of Pparg rescues Akt2-driven liver tumorigenesis. To address in vivo the functional importance of PPARγ expression in liver tumorigenesis, we used a mouse model of liver cancer induced by hepatocyte-specific inactivation of the Pten gene, which we combined with Akt2 or Pparg deletions. To this end, we crossed Pten, Akt2, or Pparg floxed mouse lines with a transgenic line overexpressing Cre under the albumin enhancer/promoter (Alb-Cre), which achieves efficient deletion of targeted genes at the early postnatal stage in both hepatocytes and biliary cells (25). Characterization of double mutants revealed that deletion of Akt2 in Pten-null hepatocytes (Pten/Akt2 double mutants) was sufficient to rescue liver hypertrophy, liver damage as assessed by the activity of hepatic enzymes aspartate transaminase and alanine transaminase in plasma, and tumor burden in 1-year-old mice (Figure 2, A and B, and Supplemental Figure 2, A and B). This analysis confirmed previous studies using the whole-body deletion of Akt2 (13, 26) and also demonstrated, for the first time to our knowledge, that the activation of Akt2 specifically in hepatocytes is required for liver tumorigenesis. Strikingly, hepatocyte-specific deletion of Pparg phenocopied to a large extent the deletion of Akt2, providing a remarkable resistance to liver tumorigenesis (Figure 2A). Notably, Pparg deletion decreased liver hypertrophy by 50% (Figure 2A and Supplemental Figure 2A) and fully rescued liver damage (Figure 2B and Supplemental Figure 2B) in 1-year-old tumoral Pten mutants. Importantly, tumor incidence and size distribution of lesions in Pten mutants were significantly reduced by codeletion of either Akt2 or Pparg. Interestingly, Pten/Akt2 double mutants showed more protection as compared with Pten/Pparg double-mutant mice (Figure 2A and Supplemental Figure 2C). These observations suggest that PPARγ is not a unique player in liver tumorigenesis downstream of activated Akt2 signaling. Nevertheless, the predominant role of PPARγ was indicated by the significantly lower incidence of lesions upon deletion of Pparg in Pten mutants (Supplemental Figure 2C). We have also confirmed that lesions found in Pten/Pparg double mutants originated from Pparg-null hepatocytes as evidenced by the recombination in the Pparg locus in dissected tumors (Supplemental Figure 2D). Importantly, the protection from tumorigenesis in double mutants was also long-lasting. While 100% of Pten mutants died before 14 months, the double mutants at the age of 15 months did not show any significant progression of the disease as compared with 12-month-old animals, as assessed by the liver macroscopic appearance and organ hypertrophy (Figure 2A and Supplemental Figure 2A).

Figure 2

Liver tumorigenesis driven by activated insulin signaling is rescued by hepatocyte-specific deletion of Akt2 and Pparg. (A) Representative images of livers of 12- and 15-month-old male mice. Scale bar: 1 cm. (B) Plasmatic aspartate transaminase levels in random-fed 12-month-old male mice (n = 4). (C and D) Relative transcript levels (C) and immunoblot analysis of total protein extracts (D) of PPARγ in livers of 5-month-old male mice of indicated genotypes, n = 6–8. (E) Relative liver weight (n = 8–12). (F) Representative images of immunohistochemical analyses with anti-BrdU/anti–β-catenin antibodies, H&E, and Oil Red O (ORO). Scale bar: 50 μm. (G and H) Hepatocyte proliferation presented as the ratio of BrdU⁺ nuclei to total number of hepatocyte nuclei (n = 3–9) (G) and hepatic triglyceride levels (n = 7) (H) in liver tissue of 5-month-old male mice of indicated genotypes. Data are means ± SEM. *P < 0.05 vs. WT; ^#P < 0.05 vs. Pten LKO mice; 1-way ANOVA with Tukey’s multiple-comparisons test. Pten LKO, liver-specific Pten knockout; Pten/Akt2 LDKO, liver-specific Pten/Akt2 double knockout; Pten/Pparg LDKO, liver-specific Pten/Pparg double knockout.

To determine whether PPARγ was required in the early stages of liver disease development in the Pten-null model, the phenotype of mutants was characterized at a pretumoral age of 5 months. First, the efficient deletion of targeted genes was confirmed by the immunoblotting analyses in liver extracts of respective mutants (Supplemental Figure 3, A and B). Importantly, significant PPARγ induction was observed in the liver tissue of Pten mutants at both the transcript and the protein level (Figure 2, C and D, and Supplemental Figure 3B). Strikingly, it was fully rescued by codeletion of Akt2 (Figure 2, C and D). To get further insight into the upregulation of PPARγ in Pten-null liver, we analyzed the expression of PPARγ isoforms using specific primer pairs in real-time quantitative PCR analysis. As a result, we revealed that transcript of PPARγ2 isoform was potently induced in Pten-null liver already at an early pretumoral age (Supplemental Figure 3C). PPARγ1 was significantly upregulated in aged Pten mutants (Supplemental Figure 3C). Importantly, these analyses also revealed that deletion of Pparg in Pten-null hepatocytes did not affect Akt activation, as witnessed by phosphorylation of Ser473 in Akt and its downstream target Pras40, further suggesting that PPARγ activity is downstream of PI3K/Akt2 signaling (Supplemental Figure 3B).

At the organ level, similar to what is observed in 1-year-old tumoral mice, the liver hypertrophy at pretumoral age in Pten single mutants was corrected by 80% and 74% by codeletion of Akt2 or Pparg, respectively (Figure 2E). The increase in liver size of Pten mutants at this age was due to a 2-fold increase in hepatocyte proliferation and cell size (Figure 2, F and G, and Supplemental Figure 3D). Both of these parameters were rescued to a similar extent by ablation of Akt2 or Pparg expression in Pten-null hepatocytes (Figure 2, F and G, and Supplemental Figure 3D).

PPARγ specifically controls the transcription of hexokinase 2 (HK2) and M2 pyruvate kinase (PKM2), 2 enzymes that greatly enhance aerobic glycolysis and promote lipogenesis, well-known metabolic rearrangements in tumoral cells (13). As shown in Figure 2H, the steatosis of Pten mutants, as reflected by a 5-fold increase in triglyceride levels, was inhibited by 58% and 78%, respectively, upon deletion of Pparg or Akt2. These biochemical measurements were also reinforced by the histological assessment of liver tissue sections using H&E and Oil Red O staining (Figure 2F). These observations were further corroborated by changes in transcript and protein levels of metabolic enzymes in Pten-null livers in a PPARγ-dependent manner (Supplemental Figure 3, E and F). In sum, the characterization of liver-specific Pten/Akt2 and Pten/Pparg double mutants validates Pparg as an essential gene product for the metabolic rearrangements and liver tumorigenesis downstream of activated Akt2. Altogether, these genetic epistasis experiments strongly suggest that Akt2 and PPARγ act in a linear pathway in a cell-autonomous manner and corroborate our analyses in human patient HCC samples.

PPARγ is induced in liver lesions characterized by HNF1α inactivation. To gain further mechanistic insights into the regulation of PPARγ expression in liver cancer, we considered the following points: (a) PPARG transcript levels are induced in premalignant HCA lesions in patients, (b) Pten/Pparg double-knockout mice are resistant to liver tumorigenesis, and (c) HCC commonly develops in an environment of HCA both in Pten mutant mice and in humans. Altogether, these data suggest that activation of PPARG transcription might be an early event in hepatocyte transformation. Recent genomic studies in HCAs revealed a progressive accumulation of genetic alterations in the course of malignant transformation to HCC. Unlike HCCs, for which each lesion contains on average about 60 damaging mutations, HCAs are more homogeneous, starting with fewer than 10 alterations with progressive accumulation during the course of transformation (1–3, 27). We therefore reasoned that PPARG expression analysis in HCA subgroups is more likely to identify direct molecular regulators. Using our annotated HCA biopsy collection, we showed that PPARG transcript levels were significantly induced in a subgroup of lesions harboring loss-of-function mutations in the transcription factor HNF1A (Figure 3A). These lesions constitute a homogeneous group of HNF1A-mutated adenomas (H-HCA) and represent about 30% of HCA. Next, we asked whether the expression of HNF1A and its downstream targets was modified in HCCs in which we found PPARG transcript upregulated. We discovered that, although transcript levels of HNF1A were not modified between nontumoral and HCC samples characterized by increased PPARG transcript levels, the expression of known HNF1α target genes (ALB, FABP1, and UGT2B7) was significantly decreased (Supplemental Figure 4A). These observations in HCA and HCC advocate a link between HNF1α and PPARG expression. Furthermore, a presence of diffuse steatosis devoid of inflammation is a striking feature uniformly observed in all H-HCA lesions (22, 28, 29). Given that PPARγ is a prosteatogenic transcription factor, increased PPARG transcript levels in H-HCA suggest that HNF1α and PPARγ might functionally interact in hepatocytes. To test this possibility, we used a whole-body mouse mutant of Hnf1a. Consistent with the initial characterization of Hnf1a mutants (30), 2-month-old Hnf1a knockout mice presented a striking 280% liver hypertrophy (Figure 3B). This liver hypertrophy was due to increased hepatocyte proliferation as revealed by nuclear PCNA and Ki67 localization (Figure 3C and Supplemental Figure 4B). Furthermore, consistent with the steatotic nature of H-HCA lesions, Hnf1a mutants showed a 4-fold increase in lipid accumulation in liver, as assessed by biochemical triglyceride (TG) measurements and histological analysis of liver tissue (Figure 3D and Supplemental Figure 4C). Importantly, we observed a 10-fold induction of hepatic PPARγ transcript and protein expression in Hnf1a-null liver (Figure 3E and Supplemental Figure 4D). Notably, both isoforms of PPARγ, PPARγ1 and PPARγ2, were upregulated in the liver of Hnf1a mutants, albeit PPARγ1 protein to a larger extent, unlike in Pten mutants (Supplemental Figure 4, D and E). This was accompanied by PPARγ activation, as indicated by nuclear accumulation of PPARγ protein (Figure 3F) and increased expression of its target genes (Figure 3E). In summary, PPARγ expression and activity are upregulated both in H-HCA and in liver of Hnf1a mouse mutants, suggesting a functional interaction between these 2 transcription factors.

Figure 3

PPARγ expression and activity are induced in liver lesions characterized by inactivation of the transcription factor HNF1α. (A) Expression profile of PPARG transcript in normal liver (N, n = 5), nontumoral liver (NT, n = 52), biopsies of HNF1A-mutated HCAs (n = 29), and non–HNF1A-mutated HCAs (n = 87) by real-time quantitative PCR. Data are presented as a ribosomal 18S–normalized mean fold (log₂) compared with the mean value in nontumoral samples ± SEM, Mann-Whitney test. *P < 0.05. (B–F) Relative liver weight (n = 7) (B), hepatocyte proliferation revealed by anti-PCNA immunohistochemistry (the inset shows the magnified view of the PCNA⁺ hepatocytes) and analyzed as a ratio of PCNA⁺ nuclei to total number of hepatocyte nuclei (n = 3) (C), hepatic triglycerides (n = 5) (D), relative transcript levels of Pparg and PPARγ target genes (n = 3) (E), and immunohistochemistry analysis using anti-PPARγ antibody (F) in livers of 3-month-old random-fed WT and Hnf1a KO male mice. Data are means ± SEM. *P < 0.05 vs. WT; 2-tailed, unpaired Student’s t test. Scale bar: 25 μm. (G and H) Hepatic triglycerides (G) and immunoblot analysis of total protein extracts (H) from liver tissue of 10-week-old random-fed WT and Hnf1a KO female mice sacrificed 5 days after transduction with adenoviral vectors expressing PPARγ shRNA or GFP. Densitometric analysis of actin-normalized signals is presented. Data are means ± SEM, n = 4–5. *P < 0.05 vs. WT/AdGFP; ^#P < 0.05 vs. Hnf1a KO/AdGFP; 1-way ANOVA with Tukey’s multiple-comparisons test.

Steatosis in Hnf1a mutants is dependent on PPARγ. Previous studies focused on the molecular characterization of H-HCA provided evidence that lipid accumulation in H-HCA lesions is positively correlated with the transcriptional induction of enzymes implicated in glycolysis and fatty acid synthesis (22), without revealing the nature of the transcription factors involved in this response. To address the contribution of PPARγ to the fatty liver phenotype of Hnf1a mouse mutants, we acutely depleted hepatic PPARγ levels in Hnf1a-null mice using shRNA knockdown (Supplemental Figure 5, A and B). As a result, a 5-fold decrease of PPARγ expression in the livers of Hnf1a knockout mice normalized hepatic TG content (Figure 3G and Supplemental Figure 5C) and reversed lipogenic enzyme (ACC, FAS, ATPCL) and fatty acid–mobilizing protein (aP2) expression (Figure 3H). Notably, hepatic downregulation of PPARγ did not affect Akt2 activation, as assessed by phosphorylation of Ser473 (Figure 3H). Furthermore, to rule out any possible non–cell-autonomous effects associated with the responses in vivo, we also downregulated PPARγ expression in primary hepatocytes isolated from Hnf1a mutant mice (Supplemental Figure 5D). Consistent with our observations in vivo, PPARγ depletion in primary hepatocytes efficiently rescued lipid accumulation (Supplemental Figure 5E) and had a profound effect on prolipogenic gene expression (Supplemental Figure 5D). Taken together, these observations indicate that PPARγ expression is essential for lipogenesis and lipid accumulation in Hnf1a mutant hepatocytes.

HNF1α is a novel negative regulator of PPARG expression. Since PPARG transcript levels are induced in H-HCA and in the liver tissue of Hnf1a mouse mutants, we hypothesized that HNF1α might be a novel negative regulator of PPARG expression. Therefore, we performed a bioinformatics analysis of PPARG gene promoter for the presence of putative HNF1α response elements (HREs). In silico analysis of the human PPARG locus identified multiple putative HREs, spanning upstream of the PPARG1 and PPARG2 promoters and into the intronic sequences of the PPARG gene (Figure 4A). Namely, they were localized at positions HRE#1, –45,519 bp; HRE#2–4, –3,000 to –200 bp; HRE#5, +1,791 bp; HRE#6, +34,022 bp; HRE#7, +56,554 bp; and HRE#8–9, +75,457 to +76,000 bp from the transcription start site. Importantly, these putative HREs are conserved among mammals (Figure 4A). To evaluate whether HNF1α could directly bind to those putative HREs, we performed chromatin immunoprecipitation (ChIP) and analyzed enrichment in HNF1α binding using primers nested around the identified sequences. As shown in Figure 4B, in human hepatocellular carcinoma HUH7 cells, endogenous HNF1α was found to bind to HRE#1, #2, and #7 with enrichment of more than 10-fold over IgG control immunoprecipitations. The observed binding to HREs in the PPARG promoter was similar to HRE binding in the promoter of ALB, an established HNF1α target gene (Figure 4B). We also detected a consistent enrichment of HNF1α in the HRE#3 and HRE#9 regions, however, with lower affinity as compared with the other HREs (Figure 4B). Taken together, these findings indicate that PPARG is a novel transcriptional target of HNF1α.

Figure 4

HNF1α is a novel negative regulator of PPARG gene transcription. (A) Localization of putative HREs in the human PPARG gene. Gray and white rectangles represent noncoding and coding exons, respectively. Black rectangles represent putative HREs. Arrows mark localization of the PPARG isoform–specific promoters. (B) Relative enrichment of endogenous HNF1α at putative HREs on the PPARG and ALB genes in unsynchronized HUH7 cells. Data are means ± SEM, n = 3 independent experiments. *P < 0.05 vs. IgG; 2-tailed, unpaired Student’s t test. (C) The luciferase reporter activity normalized to β-gal activity in primary hepatocytes overexpressing HNF1α or GFP using HNF4A or PPARG promoter constructs. Data are presented as fold difference over empty vector–transfected AdGFP-infected condition. Data are means ± SEM, n = 3 independent hepatocyte cultures. *P < 0.05 vs. empty vector; 1-way ANOVA with Tukey’s multiple-comparisons test. (D) Immunoblot analysis of PPARγ protein after HNF1α overexpression in primary hepatocytes 24 hours after transduction with increasing doses of adenoviruses (0.5, 1, and 5 MOI). Immunoblot with anti-GAPDH antibody served as a loading control. (E) Luciferase activity of FGB-LUC, normalized to Renilla, measured in livers of 2-month-old male mice sacrificed 24 hours after hydrodynamic shock delivery. Data are means ± SEM, n = 3–7. *P < 0.05 vs. WT; ^#P < 0.05 vs. Pten LKO; 2-tailed, unpaired Student’s t test. (F and G) Immunoblot analysis of total protein extracts (F) and relative transcript levels (G) of HNF1α and HNF1α target genes in liver tissue of 4-month-old random-fed male mice. Densitometric analysis of actin-normalized signals is presented as a graph. Data are means ± SEM, n = 4–5. *P < 0.05 vs. WT; ^#P < 0.05 vs. Pten LKO mice; 2-tailed, unpaired Student’s t test.

Next, we investigated whether HNF1α was sufficient to inhibit PPARG promoter activity. Transfection of primary hepatocytes with a 3-kb human PPARG promoter luciferase reporter construct revealed a 50% inhibition of luciferase expression upon HNF1α overexpression (Figure 4C). At the same time, overexpression of HNF1α induced luciferase expression from a hepatocyte nuclear factor 4α (HNF4α) reporter construct, a known positive HNF1α target gene (Figure 4C). The role of HNF1α in the control of PPARG transcription was further supported by the downregulation of PPARγ transcript and protein levels in primary mouse hepatocytes upon HNF1α overexpression (Figure 4D and Supplemental Figure 6A). Previous reports suggested that in addition to HNF4A being a target gene of HNF1α, HNF4α protein binds and acts as a cofactor of HNF1α, potently stimulating its transcriptional activity (31). We tested the synergism of these factors on PPARG expression. By using both pharmacological inhibition of HNF4α by the selective antagonist BI6015 and HNF4α ectopic expression, we did not observe the functional cooperation between HNF1α and HNF4α in the repression of PPARG expression (Supplemental Figure 6, B and C). Together, these results reveal PPARG as a novel target under negative transcriptional control by HNF1α.

Activated Akt2 inhibits HNF1α. Our observations that PPARγ expression in Pten-null liver is under positive control of Akt2 and that in hepatocytes PPARγ expression is negatively regulated by HNF1α made us hypothesize that there might be a functional link between Akt2 and HNF1α. We therefore tested HNF1α transcriptional activity in the livers of pretumoral Pten knockout and Pten/Akt2 double-knockout mice. In vivo luciferase assays using a reporter construct of a validated HNF1α transcriptional target (Fgb) demonstrated an 80% inhibition of HNF1α transcriptional activity in liver tissue of Pten mutants (Figure 4E). Importantly, HNF1α activity in the livers of Pten mutants was 1.7-fold upregulated by codeletion of Akt2 (Figure 4E). However, it was 60% of activity in WT livers, suggesting that other mechanisms may be implicated.

Next, we performed expression analyses in liver tissue of Pten and Pten/Akt2 double mutants. Consistent with the results of HNF1α transcriptional activity measurements in vivo, transcript and protein levels of HNF1α and its known targets were significantly downregulated in Pten-null livers in an Akt2-dependent manner (Figure 4, F and G). Notably, a similar level of inhibition of HNF1α transcriptional responses was observed in primary hepatocytes isolated from Pten knockout mice, confirming the cell-autonomous nature of suppression (Supplemental Figure 6, D and E). Also, consistent with the observations in liver tissue extracts, the protein levels of HNF1α and its direct targets, Hnf4a and Alb, were rescued by codeletion of Akt2 in Pten-null hepatocytes (Supplemental Figure 6F). The repressive effect of HNF1α on Pparg expression was counteracted by overexpression of activated Akt2 (Myr-Akt2) in primary hepatocytes (Supplemental Figure 6G). Furthermore, inhibition of HNF1α by Myr-Akt2 was also observed on its transcriptional target, HNF4α, in a luciferase assay (Supplemental Figure 6H). Collectively, these analyses in vitro and in vivo reveal HNF1α as a transcriptional repressor of Pparg under negative control of Akt2.

HNF1α is a novel substrate of Akt2. To get further mechanistic insights into the HNF1α regulation by Akt2, we asked whether HNF1α could be a novel protein substrate of Akt2. By using phosphospecific antibodies recognizing an Akt-phosphorylation motif, we revealed that endogenous HNF1α was phosphorylated in Pten-null liver in an Akt2-dependent manner (Figure 5A). Next, by using a bioinformatics approach, we identified Ser247, a highly evolutionary conserved residue in the DNA-binding domain of HNF1α, as a putative Akt-phosphorylation motif (see below). We confirmed that Ser247 was a major phosphorylation site for Akt in HNF1α, as a substitution of Ala or Asp for Ser247 abolished recognition of HNF1α by phosphospecific antibodies (Figure 5B). Next, we generated antibodies that specifically recognized phosphorylated Ser247 (Figure 5C). Phosphorylation of ectopically expressed WT HNF1α was induced by serum stimulation and was sensitive to inhibition of Akt signaling by Torin treatment (Figure 5, D and E). Analyses of the endogenous HNF1α protein immunoprecipitated from the liver extracts demonstrated that phosphorylation of Ser247 was induced in Pten-null liver in an Akt2-dependent manner (Figure 5F). To investigate the role of the newly identified HNF1α phosphorylation, we studied subcellular localization of HNF1α. Our analysis of nuclear extracts from livers of Pten-null mice revealed a nuclear exclusion of endogenous HNF1α, which was dependent on Akt2 expression (Figure 5G). Furthermore, the analyses in HUH7 cells revealed that nuclear localization of transiently overexpressed WT HNF1α protein was promoted under serum starvation conditions, while HNF1α protein was readily observed in the cytoplasm upon serum stimulation conditions that activate Akt (Supplemental Figure 7B). Importantly, the pharmacological inhibition of Akt signaling using the mTOR inhibitor Torin counteracted the effect of serum stimulation on the localization of WT HNF1α (Supplemental Figure 7B). Notably, the phospho-mimicking substitution of Asp for Ser247 in HNF1α promoted cytoplasmic localization of HNF1α (Supplemental Figure 7B). In agreement with these subcellular localization studies, the transcription activity of the phospho-mimicking mutant of HNF1α was significantly reduced, as measured by levels of Hnf4a transcript upon overexpression of HNF1α S247D mutant (Supplemental Figure 7C). Altogether, these analyses demonstrate that HNF1α is a novel Akt2 substrate whose nuclear localization and transcriptional activity are inhibited by phosphorylation of Ser247.

Figure 5

HNF1α is a novel substrate of Akt2 phosphorylated on Ser247. (A) Immunoblot analysis of endogenous HNF1α immunoprecipitated from liver tissue extracts of 5-month-old random-fed male mice of indicated genotypes with antibody recognizing an Akt-phosphorylation motif. The membrane was reprobed with anti-HNF1α antibody. Densitometric analysis of phosphorylated HNF1α normalized to immunoprecipitated HNF1α signals is presented. Data are means ± SEM, n = 4–5. *P < 0.05 vs. WT; ^#P < 0.05 vs. Pten LKO; 1-way ANOVA with Tukey’s multiple-comparisons test. (B) Immunoblot analysis of immunoprecipitated HNF1α-WT-MYC, HNF1α-S247A-MYC, and HNF1α-S247D-MYC proteins transiently overexpressed in HEK293T cells, using antibody recognizing an Akt-phosphorylation motif. The immunoblot with anti-HNF1α of the same membrane served as a loading control. (C) Immunoblot analysis of immunoprecipitated HNF1α-WT-MYC and HNF1α-S247A-MYC proteins transiently overexpressed in HEK293T cells, using antibody raised against phosphorylated S247HNF1α. (D and E) Immunoblot analysis of total protein extracts of immortalized WT mouse embryonic fibroblast cells transiently overexpressing Myc-tagged HNF1α WT protein. Twenty-four hours after transfection, cells were starved in Earle’s Balanced Salt Solution for 2 hours followed by 30 minutes of treatment with 100 nM Torin before stimulation with 10% FBS for 1 hour (D) or stimulation with 10% FBS for indicated times (E). (F) Immunoblot analysis of endogenous HNF1α immunoprecipitated from liver tissue extracts of 5-month-old random-fed male mice of indicated genotypes with antibody raised against phosphorylated S247HNF1α. Densitometric analysis of phosphorylated HNF1α normalized to total HNF1α is presented. Data are means ± SEM, n = 5. *P < 0.05 vs. WT; ^#P < 0.05 vs. Pten LKO; 1-way ANOVA with Tukey’s multiple-comparisons test. (G) Immunoblot analysis of nuclear protein extracts from livers of 3-month-old male mice with indicated antibodies. Densitometric analysis of lamin A/C–normalized signals is presented as a graph. Data are means ± SEM, n = 4. *P < 0.05 vs. WT; ^#P < 0.05 vs. Pten LKO mice; 2-tailed, unpaired Student’s t test.

Pharmacological targeting of PPARγ activity modulates liver tumorigenesis. PPARγ activation was proposed as an anticancer treatment in certain malignancies, promoting cell differentiation and cell cycle withdrawal (32–34). However, our data suggest that PPARγ transcriptional activity could be protumorigenic in the context of activated Akt2 signaling in the liver. While the nature of endogenous PPARγ ligands is still obscure, a plethora of synthetic PPARγ agonists, including the thiazolidinediones (TZDs), is available. For example, the typified TZD pioglitazone was widely used in the clinic as an insulin sensitizer. In contrast, only a few selective PPARγ antagonists with potent in vivo activity are available. The most recent compound of this class is SR2595, which displays satisfactory pharmacokinetics to support in vivo studies (35). To clarify the outcomes of hepatic PPARγ activity modulation in vivo, we performed chronic treatments with pioglitazone and SR2595. Treatment with pioglitazone starting from 5 months of age for 3 months resulted in a striking 85% increase in liver size of Pten-null mice (Figure 6, A and B). Importantly, this effect was limited to Pten mutants, as control mice were completely resistant to the growth-promoting effect of the PPARγ agonist, which is consistent with low expression of PPARγ in the livers of WT mice (Figure 6, A and B). To rule out any non–PPARγ-related effects of pioglitazone, we also treated Pten/Pparg double-knockout mice. Pten/Pparg mutants were remarkably resistant to the effect of the agonist, confirming the selectivity of the drug (Figure 6, A and B). Furthermore, this chronic pioglitazone administration significantly aggravated liver damage in Pten mutants as reflected by a 70% increase in aspartate transaminase enzymatic activity in plasma while having no effect in the 2 control groups (Figure 6C). Consistent with PPARγ activation by pioglitazone, liver steatosis was significantly upregulated in Pten-null mice treated with the agonist, as assessed by hepatocyte morphological appearance, by Oil Red O staining, and by biochemical measurements of triglycerides in liver extracts (Figure 6D and Supplemental Figure 8, A and B). This was paralleled by increased expression of PPARγ targets in livers of Pten mutants treated with pioglitazone (Supplemental Figure 8, C and D). Importantly, these metabolic rearrangements in livers of Pten mutants upon pioglitazone administration were accompanied by significant expansion of hepatic lesions (Figure 6, E and F). Notably, in Pten mutants at this age, the majority of lesions were classified as adenomas, while in pioglitazone-treated mutants these lesions were predominantly the high-grade proliferating bile duct neoplasia and HCCs (Figure 6, E and F). In sum, pharmacological activation of PPARγ in Pten-null mice significantly aggravates liver damage and accelerates liver tumorigenesis.

Figure 6

Pharmacological activation of PPARγ aggravates pathological liver growth in Pten mutants. (A–D) Representative images of livers (A), relative liver weight (B), plasmatic aspartate transaminase levels (C), and hepatic triglycerides (D) in 8-month-old male mice of the indicated genotypes fed with control or pioglitazone-containing (PIO) chow for 3 months. Data are means ± SEM, n = 5. *P < 0.05 vs. WT; ^#P < 0.05 vs. Pten LKO; ^§P < 0.05 vs. chow food; 1-way ANOVA with Tukey’s multiple-comparisons test. Scale bar: 1 cm. (E and F) Representative images (E) and quantification of lesions (F) in H&E-stained sections of livers of mice treated as in A. Dashed line marks lesion area. Scale bar: 500 μm (top), 100 μm (bottom). The relative area of adenoma, abnormally proliferating bile ducts (BDPs), and tumor lesions is presented. Data are means ± SEM, n = 4. ^§P < 0.05 vs. chow; 1-way ANOVA with Tukey’s multiple-comparisons test.

Next, to test whether pharmacological inhibition of PPARγ could be therapeutically beneficial, we treated Pten mutants at pretumoral and tumoral age with the PPARγ antagonist SR2595. Treatment of 5-month-old pretumoral Pten mutants during 1 month was well tolerated and did not cause any adverse toxicity, as reflected by the absence of weight loss during the course of treatment (Supplemental Figure 9A). Importantly, 1-month treatment with SR2595 resulted in a significant inhibition of PPARγ activity as revealed by decreased PPARγ target expression in liver tissue of Pten mutants (Supplemental Figure 9, B and C). In addition, a 30% increase in free fatty acids in the plasma of SR2595-treated Pten-null mice, as compared with the placebo-treated group, indicated a potent systemic effect of SR2595 (Supplemental Figure 9D). This is consistent with the inhibition of PPARγ in adipose tissue, where PPARγ is required for lipid storage. Furthermore, expression analyses revealed no inhibitory effect of SR2595 on PPARα activity, ruling out potential off-target effects of this antagonist (Supplemental Figure 9E). In sum, SR2595 is a potent specific PPARγ antagonist efficient in inhibiting PPARγ function in the liver of Pten mutants.

The aggravation of the liver phenotype in Pten mutants was accompanied by an increase in steatosis and hypoglycemia as the mice aged. One month of SR2595 treatment had a striking therapeutic effect by stabilizing glycemia and normalizing the steatosis of Pten mutants (Figure 7, A and B). Consistent with decreased steatosis, hepatocyte size was rescued by SR2595 treatment in the livers of Pten-null mice, as witnessed by changes in cell density (Figure 7C and Supplemental Figure 9F). Finally, the increased hepatocyte proliferation in Pten mutants was normalized in SR2595-treated animals (Figure 7C). In conclusion, consistent with the genetic epistasis experiments, pharmacological inhibition of PPARγ in pretumoral Pten-null mice normalizes liver steatosis and inhibits hepatocyte proliferation.

Figure 7

PPARγ inhibition by SR2595 significantly improves the phenotype of tumoral Pten mutants. (A) Blood glucose levels measured during treatment with SR2595 or placebo from 5 to 6 months of age in random-fed Pten LKO male mice. Data are means ± SEM, n = 3. *P < 0.05 vs. D0; ^#P < 0.05 vs. Pten LKO/placebo; 1-way ANOVA with Tukey’s multiple-comparisons test. (B and C) Hepatic triglycerides (B) and representative images of immunohistochemical analyses with anti-BrdU/anti–β-catenin antibodies and quantification of hepatocyte proliferation presented as a ratio of BrdU⁺ nuclei to total number of hepatocyte nuclei (C) in livers of random-fed mice treated as in A. Data are means ± SEM, n = 3. *P < 0.05 vs. WT; ^#P < 0.05 vs. Pten LKO/placebo; 1-way ANOVA with Tukey’s multiple-comparisons test. Arrowheads point to BrdU⁺ proliferating hepatocytes. The inset shows the magnified view of the BrdU⁺ hepatocytes. Scale bar: 100 μm. (D) Blood glucose levels measured during treatment with SR2595 from 11 to 12 months of age in random-fed Pten LKO male mice. Data are means ± SEM, n = 6. *P < 0.05 vs. placebo; 2-tailed, unpaired Student’s t test. (E and F) Representative images of livers (E) and H&E-stained liver sections (F) of mice treated as in D. Dashed line marks tumoral area (T) and nontumoral area (NT) of the section. Graph represents the quantification of relative area of adenoma, abnormally proliferating bile ducts, and tumor lesions. Data are means ± SEM, n = 50 lesions per group. *P < 0.05 vs. placebo; 2-tailed, unpaired Student’s t test. Scale bar: 1 cm (E), 100 μm (F).

Given the positive outcome of SR2595 treatment in pretumoral Pten knockout mice, we asked whether PPARγ inhibition could provide therapeutic benefit in aged tumor-harboring animals. We submitted 11-month-old Pten mutants, severely affected by liver cancer at this stage, to chronic treatment of the antagonist administered daily for 1 month by oral gavage. Similarly to the treatment of Pten-null mice at pretumoral age, SR2595 administration did not provoke any adverse toxicity and was well tolerated in tumoral Pten mutants as reflected by stable body weight (Supplemental Figure 9G). Furthermore, while Pten mutants treated with placebo showed significant decrease in plasmatic glucose levels, consistent with the aggravation of the tumoral phenotype from 11 months to 1 year, SR2595-treated mice had a stable glycemia during the course of treatment and even demonstrated an improvement to levels found in pretumoral mice (Figure 7D). Most importantly, PPARγ inhibition had a positive therapeutic effect on disease progression marked by 50% lower liver hypertrophy in SR2595-treated Pten mutants (Figure 7E and Supplemental Figure 9H). This observation was further corroborated by a lower number and smaller size of lesions observed in the SR2595-treated mice compared with vehicle-treated animals (Figure 7, E and F, and Supplemental Figure 9I). Finally, proliferation in tumoral lesions (Supplemental Figure 9J), as assessed by BrdU labeling, was severely blunted, while apoptosis (Supplemental Figure 9K), evaluated by TUNEL assay, was induced in SR2595-treated mice as compared with the placebo-treated group. Altogether, these multiple lines of evidence lay a strong foundation supporting the pharmacological modulations of hepatic PPARγ activity as a therapeutically relevant intervention in hepatic malignancies associated with activated Akt2 and PPARγ signaling.

Growth factor signaling is found to be upregulated in a majority of malignancies, including liver cancer. Yet the outcomes of its inhibition in the clinic are disappointing, because of the presence of complex feedback mechanisms, urging identification of specific targets for selective treatments (36, 37). By analyzing one of the largest and best-annotated collections of patient hepatic lesions, in combination with analyses of animal models of liver cancer driven by activated insulin signaling, we identified the transcription factor PPARγ as an essential player in liver tumorigenesis. The most significant findings of our work are: (a) PPARγ expression is upregulated in a subset of human HCAs (hHCA) and hHCCs; (b) activation of Akt2 isoform in hepatocytes is sufficient to induce liver steatosis and tumorigenesis, which is dependent on PPARγ; (c) activated Akt2 relieves the repressive function of HNF1α on PPARγ expression, a mechanism conserved in mice and humans; and (d) pharmacological inhibition of PPARγ is therapeutic in liver disease driven by activated Akt2 signaling.

The most unexpected finding of our studies is the discovery of a novel role for HNF1α as a negative regulator of PPARγ expression, corroborated by HNF1α gain-of-function and loss-of-function experiments and identification of HNF1α binding sites in the promoter region of PPARG. HNF1α is a homeodomain-containing protein that was originally identified as a hepatocyte-specific transcription factor critical for hepatocyte differentiation and metabolic function of liver (30, 38). Its role in defining hepatocyte fate is supported by recent studies in mouse and human fibroblasts that are differentiated to hepatocytes with functional benefits in vivo by transcription factors including HNF1Α (39, 40). Its function as a tumor suppressor gene in liver was proposed after identification of biallelic loss-of-function somatic mutations in a subclass of hepatocellular adenomas, H-HCA (41). Consistently, expression of HNF1α was reported to be downregulated in patient HCC lesions as well as in mice in chemically induced HCCs (42). Furthermore, HNF1α overexpression both in vitro and in vivo potently inhibited proliferation of cancer cells and inhibited tumor progression (42). So far, no direct targets that are negatively regulated by HNF1α are reported. The published reports show that at the genomic level, HNF1 binding sites are enriched in genes whose expression is normally decreased in Hnf1a knockout mice. However, a group of genes including the lipogenic enzymes was already suggested, but not tested experimentally, to be under negative control of HNF1α (22). In addition, HNF1β, a close paralog of HNF1α normally involved in the positive transcriptional activation of target genes, has been described as a repressor of SOCS3 transcription (43). Furthermore, in vitro studies demonstrated that HNF1α could negatively regulate its own and other promoters in an indirect manner. The corepressor activity of HNF1α was observed in a complex with another transcription factor, the orphan nuclear receptor HNF4α (44). However, our bioinformatics analyses did not reveal HNF4α binding sites in proximity to the identified HNF1α response elements in the promoter of PPARG. Furthermore, our analyses did not reveal the synergism between these 2 transcription factors in the regulation of PPARG transcription.

Importantly, our study suggests that HNF1α is under tight control by the PI3K/Akt2 pathway. We discovered that HNF1α is a novel substrate of Akt2 phosphorylated in position Ser247. Our in vitro observations using a phospho-mimicking mutant of HNF1α suggest that phosphorylation of Ser247 in HNF1α promotes its cytoplasmic localization, thus modulating the activity of the transcription factor. HNF1α was found to be phosphorylated at Ser247 in 3 independent mass spectrometry–based studies (45–47). Another report has suggested that multiple sites in HNF1α, including Ser247, are phosphorylated by Mirk kinase (48). However, the conclusions were based on in vitro kinase assays with recombinant proteins, and no functional implications were proposed. Early studies with naturally occurring MODY mutants of HNF1α have suggested that conformational changes in HNF1α protein may affect not only its interaction with coactivators but also their enzymatic activity on the promoters of target genes (e.g., acetyltransferase activity of CBP and P/CAF) (49). Future studies will be required to address the involvement of Akt2-mediated phosphorylation of HNF1α in the recruitment and activity of cofactors.

HNF1α is clearly not the sole transcription factor that negatively regulates PPARG transcription in the liver in a pathophysiologically relevant way. Recently, the antagonizing roles of different AP-1 transcription factors in hepatic PPARG transcription were uncovered (50). Depending on the presence of either c-Fos or FRA1/2 proteins, the complex with Jun transcription factor either stimulates or represses PPARG transcription, respectively. In relevance to liver pathology, expression of FRA genes not only protected mice from high-fat diet–induced fatty liver disease but also reverted already established condition (50). Interestingly, the PI3K/Akt signaling pathway has been shown to regulate FRA-1 expression and activity (51). Another negative regulator of PPARG transcription is FOXO1, a known substrate under negative control of Akt (52). Although we cannot exclude its contribution to the regulation of PPARG transcription in liver, we observed incomplete inhibition of FOXO1 phosphorylation in Pten mutants by Akt2 depletion, suggesting rescue by another Akt isoform. In addition to FOXO1, GATA2 and KLF2 transcription factors were reported to bind the promoter region of PPARG and to inhibit transcription of the PPARG gene (53, 54). Yet the contribution of these to hepatic PPARG transcription was not investigated. Intriguingly, lower expression of GATA2 in human HCC correlates with increased proliferation rate and poor prognosis following resection (55). Curiously, GATA2, KLF2, FRA-1, FRA-2, and c-Fos proteins contain putative phosphorylation motifs for Akt. It is tempting to speculate that, in addition to HNF1α, these might contribute to PPARG transcriptional control in liver via phosphorylation by Akt2.

One of the major findings of our study is a demonstration of therapeutic efficacy of a novel PPARγ antagonist in the preclinical liver cancer model. Future studies are required to address whether SR2595 could be taken to clinic as a sole agent or in a combination with other drugs. Curiously, a recent report demonstrated that systemic inhibition of Akt signaling modeled by hepatic deletion of Akt1 in whole-body Akt2 mutants provokes an early-onset spontaneous hepatocarcinogenesis, suggesting that pharmacological inhibition of Akt might be deleterious in the clinic (56). Interestingly, previous reports using the whole-body mutants uncovered the tumorigenic role of Akt1, unlike Akt2, in prostate and endometrium (57, 58). At the same time, we and others have shown that whole-body deletion of Akt2 strongly protects from liver cancer driven by loss of Pten expression in hepatocytes (13, 26). Here, for the first time to our knowledge, we demonstrated that hepatocyte autonomous Akt2 signaling is essential for liver tumorigenesis driven by oncogenic insult. Importantly, our study also provides a downstream target of Akt2, PPARγ, whose inhibition is therapeutic.

In summary, in this study we uncovered an unappreciated protumorigenic function of PPARγ in liver and established a novel functional link between 2 master regulators of cellular fate, the hepatocyte differentiation factor HNF1α and the lipogenic transcription factor PPARγ, in liver pathophysiology. The detection of increased PPARγ mRNA and protein in a subset of human liver cancers, as well as the efficacy of PPARγ antagonist in preclinical studies of liver tumorigenesis, should open new therapeutic possibilities to test.

Animals. Pparg^fl/fl, Akt2^fl/fl, Pten^fl/fl, and Hnf1a^–/– mouse lines have been previously described (12, 30, 59, 60). SR2595 was synthesized and purified as previously described (35). Animals were maintained in grouped cages in a temperature-controlled pathogen-free facility on a 12-hour/12-hour (8 am–8 pm) light/dark cycle and had free access to water and standard chow (Teklad global protein diet; 20% protein, 75% carbohydrate, 5% fat). Animals were sacrificed between 2 and 4 pm. For the in vivo pharmacological treatments with pioglitazone incorporated in chow food (200 mg/kg), mice had free access to food and were treated for 3 months, from the ages of 5 months to 8 months. For the in vivo pharmacological treatments, SR2595 was administered daily by oral gavage (20 mg/kg). In the pretumoral group, mice were treated daily by oral gavage for 1 month from the ages of 5 to 6 months. For the tumoral group, treatment was initiated at 11 months of age for 1 month. For 5-bromo-2′-deoxyuridine (BrdU) incorporation at pretumoral age, mice were treated with BrdU (3 mg/ml; Sigma-Aldrich) dissolved in drinking water for 5 days before sacrifice. At tumoral age, BrdU was administered i.p. (50 mg/kg) 2 hours before sacrifice. For in vivo transduction, 10⁹ adenoviral infectious particles were diluted in 0.9% NaCl and administered retro-orbitally in a total volume of 100 μl per animal. Animals were sacrificed 5 days after injection.

Cell culture and recombinant adenoviruses. Human HCC cell line HUH7 and HEK293T cells were obtained from ATCC and were maintained in DMEM supplemented with 10% FBS, 2 mM l-glutamine, 50 U/ml penicillin, and 50 μg/ml streptomycin. Cells were regularly tested and were negative for mycoplasma contamination. Primary hepatocytes from 4- to 6-week-old mice were isolated by liver perfusion as described previously (13). Unless indicated, cells were collected for analysis 12 hours after plating. GFP adenoviral vectors were described previously and were used as a control in all experiments (61). Adenoviral particles expressing PPARγ shRNA were provided by Stephan Herzig (Institute for Diabetes and Cancer, Munich, Germany). Adenovirus expressing HNF1α was provided by Benoit Violet (Cochin Institute, Paris, France). Adenoviruses expressing AdMyr-Akt2 (catalog 1023) and AdHNF4α (ADV-261497) were from Vector Biolabs.

Hepatic metabolite analyses. TG levels in liver tissue or in primary hepatocytes were determined using the Triglycerides FS Kit (Diasys). Fifty to one hundred milligrams of powdered liver tissue or pellet of 1 × 10⁶ cells was used for acetone extraction.

Protein extraction and immunoblotting. To prepare total protein extracts, cells or tissue were homogenized in lysis buffer containing 20 mM Tris-HCl (pH 8.0), 5% glycerol, 138 mM NaCl, 2.7 mM KCl, 1% NP-40, 20 mM NaF, 5 mM EDTA, 1× protease inhibitors (Roche), and 1× PhosphoStop inhibitors (Roche). Homogenates were spun at 12,000 g for 10 minutes at 4°C. For immunoprecipitation, 500 μg of cleared protein extract was incubated with anti-Myc (9E11) or anti-HNF1α (SC) for 3 hours at +4°C. Then, immune complexes were pulled down using Protein G-Sepharose beads (GE Healthcare) during 2 hours followed by 4 washes with extraction buffer. The protein complexes were eluted by boiling of the beads in 1× SDS–sample buffer for 10 minutes. Protein extracts or immunoprecipitate eluates were resolved by SDS-PAGE before transfer onto PVDF membrane followed by incubation with primary antibodies and HRP-linked secondary antibodies. Immobilon Western Chemiluminescent HRP Substrate (Millipore) was used for detection. Densitometric analysis of immunoblots was performed using ImageJ software (NIH). The antibodies used in the study are listed in Supplemental Table 1. See all complete unedited blots in the supplemental material.

Nuclear extracts were prepared from 50–100 mg of snap-frozen liver tissue mechanically disintegrated in hypotonic buffer (10 mM HEPES, pH 7.9, 1× Protease inhibitors, 1× PhosphoStop inhibitors) followed by 5 minutes of centrifugation at 60 g at 4°C. The supernatant was recuperated as a cytoplasmic fraction. The pellet was washed twice with hypotonic buffer followed by lysis in hypertonic buffer (20 mM HEPES, pH 7.9, 25% glycerol, 0.42 M NaCl, 0.2 mM EDTA, 1× Protease inhibitors, 1× PhosphoStop inhibitors). The nuclear extracts were cleared by centrifugation for 10 minutes at 12,000 g at 4°C. The supernatants of soluble nuclear proteins were used for immunoblotting analyses.

Generation of Hnf1a point mutants and phosphospecific antibody. Site-directed mutagenesis was performed using QuikChange II Site-Directed Mutagenesis Kit (Stratagene) according to the manufacturer’s instructions and pcDNA5-Myc-HNF1A (Addgene, 31104) construct as a template. All inserts were sequence-verified. The phospho-HNF1α antibody was developed in collaboration with Cell Signaling Technology.

Histological and morphometric analyses. For immunohistochemical analysis, liver tissue was fixed overnight in phosphate-buffered 10% formalin and embedded in paraffin. Four-micrometer sections were cut and processed either for staining with H&E or for immunohistochemical analyses. Stained liver tissue sections were digitalized with the NanoZoomer S210 (Hamatsu). The classification and quantification of tumor lesions was performed using NDP.view2 software. For Oil Red O staining, frozen liver sections were fixed with 4% formaldehyde for 15 minutes, followed by incubation in 60% isopropanol for 5 minutes and then with Oil Red O (Bio-Rad) for 15 minutes at room temperature. Sections were washed twice with PBS and twice with water. After counterstaining with hematoxylin, slides were mounted with 90% glycerol. Immunohistochemistry of liver tissue sections was performed using anti-PCNA (Cell Signaling Technology) or anti-Ki67 (Thermo Fisher Scientific) or with a mix of anti-BrdU (Roche) and anti–β-catenin (Calbiochem) antibodies. The results are expressed as the ratio of BrdU⁺, PCNA⁺, or Ki67⁺ nuclei to the total number of nuclei in a total area of at least 10 sequential fields of 33,500 μm² tissue analyzed.

Real-time quantitative PCR. Total RNA was isolated from 50–100 mg of snap-frozen liver tissue by RNeasy Lipid Tissue Mini Kit (Qiagen) and RNeasy Mini Kit (Qiagen) from primary hepatocytes. cDNA was synthesized from 1 μg of total RNA using 125 ng of random hexamer primers and SuperScript II (Life Technologies). Real-time quantitative PCR was performed on an MX3005P instrument (Agilent) using iTaq Universal SYBR Green Supermix (Bio-Rad). The relative amounts of transcripts were determined by the 2^–ΔΔCT method, with pinin, cyclophilin, or S18 as reference gene and control treatment or control genotype as the invariant control. The primer sequences are listed in Supplemental Table 2.

For patient HCC samples, predesigned validated primers and probe sets were used (Life Technologies). The relative PPARG (Hs00234592_m1) gene expression was normalized to ribosomal 18S (Ribosomal 18S; 4352930) transcript levels. The expression levels of PPARG in tumor samples were compared with the mean level (log₂) of the corresponding gene expression in normal liver tissues (n = 5), expressed as n-fold ratio. The relative amount of RNA was calculated with the 2^−ΔΔCT method.

Luciferase reporter assay. Luciferase assay was performed using a Dual Luciferase reporter kit (Promega) according to the manufacturer’s instructions. Luciferase reporter constructs were a gift of L. Fajas (pGL3 empty vector and pGL3-PPARG-p3000; University of Lausanne, Lausanne, Switzerland), Dimitris Kardassis (pGL3-HNF4A and pGL3-APOC3; University of Crete Medical School, Crete, Greece), and Maria-Angeles Navas (pGL3-FGB; Universidad Complutense de Madrid, Madrid, Spain). For in vivo studies, a mix of luciferase promoter construct (40 μg/ml) and Renilla control plasmid (4 μg/ml) was administered by hydrodynamic shock through the penis vein in a volume of 70 μl/g of body weight delivered in 10 seconds. Mice were sacrificed 24 hours after treatment, and liver tissue was used for analysis. For in vitro luciferase assays, isolated primary hepatocytes were transduced with adenoviral vectors (AdHNF1α or AdGFP) at a dose of 10 MOI. Twelve hours after transduction, cells were transfected with a mix of luciferase reporter constructs and control plasmid expressing β-gal using Lipofectamine 2000 (Invitrogen). Twenty-four hours after transfection, cells were collected for luciferase reporter activity assay normalized to β-gal activity.

Chromatin immunoprecipitation. ChIP was performed using HUH7 cells as described previously (13, 62). Endogenous HNF1α was immunoprecipitated with anti-HNF1α antibody (63). The relative amounts of the immunoprecipitated DNA were determined by real-time quantitative PCR using the 2^−ΔΔCT method, with input DNA values for each sample as control and enrichment of 10-fold as a cutoff. Detailed protocol is in Supplemental Methods. The primer sequences are listed in Supplemental Table 2.

Statistics. Statistical analysis was performed with GraphPad Prism 5.0 (GraphPad Software). For parametric data with more than 2 comparisons and biological replicate numbers of less than 10, a 1-way ANOVA with Tukey’s multiple-comparisons test was used unless specified. All data are expressed as means ± SEM, and significance was established at the P ≤ 0.05 level.

Study approval. All studies were approved by the Direction Départementale des Services Vétérinaires, Préfecture de Police, Paris, France (authorization no. A75-14-08), Comité d’Ethique en matière d’Expérimentation Animale Paris Descartes (CEEA 34) (16-040), and le Ministère de l’Enseignement Supérieur et de la Recherche (IE-2011-578).