Opposing Roles of the Forkhead Box Factors FoxM1 and FoxA2 in Liver Cancer

Hepatocellular carcinoma (HCC) is the second most fatal malignancy in men worldwide (1, 2). Development of HCC has been linked to viral hepatitis, alcohol abuse, as well as nonalcoholic steatohepatitis (3–5). Irrespective of its etiology, 1-year survival with intervention is very low. That is partly due to high rate of recurrence of the cancer resulting from intrahepatic and extrahepatic metastasis (6, 7). Recent studies have linked aggressive progression of HCC to overexpression of the forkhead box transcription factor FoxM1. For example, overexpression of FoxM1 has been shown to strongly correlate with poor prognosis and high-grade progression of HCC (8–10). Studies with mouse models provided strong causal link between FoxM1 and aggressive progression of HCC. It was shown that FoxM1 is essential for development of HCC in a chemical carcinogenesis model (11). Deletion of FoxM1 in the adult liver blocked diethylnitrosamine (DEN)-induced HCC development. Moreover, in the same model of chemical carcinogenesis, deregulated FoxM1 drives highly aggressive, metastatic progression of HCC (12). In the absence of p19Arf, FoxM1 stimulates all steps of metastatic progression (12). Consequently, inhibition of FoxM1 impedes metastatic progression of HCC (12).

Similar observations were made with a transgenic mouse model expressing activated HRas in the liver, driven by the albumin promoter (10). In that model, HRas-induced HCC coincides with an increased expression of FoxM1. Conditional deletion of FoxM1 after HCC development causes inhibition of cancer progression. Hepatic progenitor cells for HCC have been characterized in the chemical carcinogenesis model. They express cell-surface markers CD44 and EpCAM (13). Those cells were also detected in an HRas-transgenic mouse model for HCC. They account for about 30% to 40% of the HCC cells in tumor sections (10). Interestingly, deletion of FoxM1 causes a preferential loss of those cells in tumor nodules, indicating that FoxM1 is critical for the CD44⁺ and EpCAM⁺ HCC cells (10). CD44 and EpCAM are also expressed by human hepatic cancer stem cells. Moreover, hepatic cancer stem cells in human HCC lines are dependent upon FoxM1, as deletion of FoxM1 causes a preferential loss of the cancer stem cells (10). In that regard, it is noteworthy that FoxM1 is a critical downstream factor of a variety of cancer signaling pathways, including Wnt/b-catenin signaling, that promote cancer stem cells (14). Moreover, FoxM1 stimulates expression of the pluripotency genes c-Myc, Oct4, Sox2, and Nanog (15–17).

Our recent studies on mammary luminal differentiation identified a transcriptional repression function of FoxM1 that is involved in regulation of the differentiation gene GATA3 (18). FoxM1 is expressed at high levels in mammary stem and luminal progenitor cells. Deletion of FoxM1 decreases the population of the stem/progenitor cells and increases differentiated luminal cells (10). Expression of FoxM1 has opposite effects, in that it inhibits luminal differentiation (18). FoxM1 inhibits luminal differentiation by repressing the luminal differentiation gene GATA3. We showed that FoxM1 represses GATA3 expression by recruiting the retinoblastoma protein (Rb) and DNMT3b onto the GATA3 promoter (18). Moreover, in breast cancer, there is an inverse correlation between the expression of FoxM1 and GATA3 (18).

We sought to investigate whether the differentiation gene-repression function of FoxM1 is active in other systems where FoxM1 is overexpressed. FoxM1 is overexpressed in high-grade HCC (10). FoxA genes are important hepatic differentiation genes. For example, FoxA1 and FoxA2 were shown to be essential for liver development and hepatocyte differentiation (19). Interestingly, FoxA2 was shown to inhibit metastasis of HCC cells (20). FoxM1, on the other hand, drives metastasis of HCC (21). Therefore, we investigated whether FoxM1 represses expression of FoxA2 in HCC. Here we show that, as in the case of GATA3 in breast cancer cells, FoxM1 inhibits FoxA2 in HCC cells by recruiting Rb and DNMT3b. Moreover, we show that FoxA2 inhibits FoxM1 expression, providing evidence for opposing roles of FoxM1 and FoxA2 in HCC.

Human HCC Huh7, HepG2, and SNU449 cells were obtained from American Type Culture Collection. All experiments were performed within 8 to 10 passages. No further authentication of cells was performed. Cells were maintained in Dulbecco's modified Eagle's medium (DMEM) supplemented with 10% fetal bovine serum (HyClone Laboratories Inc.) or 10% Tet System Approved FBS (Clontech; cat no. 631105; for inducible cell lines), with 100 units of penicillin/streptomycin used to culture cells at 37°C with 5% CO₂. Cells were transfected with plasmid DNA or siRNA using Lipofectamine 2000 (Invitrogen) in serum-free tissue culture medium following the manufacturer's protocol. Six hours after transfection, cells were fed with complete DMEM containing 10% fetal bovine serum.

IHC staining was performed following the standard procedure. Antigen retrieval was done using sodium citrate buffer, and sections were then treated with antibodies overnight. An additional blocking step was performed using an avidin/biotin Vectastain kit following the manufacturer's protocol. Visualization was done using DAB and counterstained using hematoxylin (Polyscientific). For antibodies of mouse origin, mouse on mouse kit was used. All reagents are from Vector Labs. Information about the antibodies is included in Supplementary Table S1.

All animal experiments were preapproved by the UIC Institutional Animal Care and Use Committee. Previously described Alb-H-ras12V mice were crossed with FoxM1fl/fl MxCre C57/BL6 mice to obtain FoxM1fl/fl MxCre Alb-H-ras12V and FoxM1^+/+ MxCre Alb-H-ras12V mice. For deletion studies, 8-month-old male mice (FoxM1^+/+ MxCre Alb-H-ras12V and FoxM1fl/fl MxCre Alb-H-ras12V) were subjected to 5 or 10 intraperitoneal (i.p.) injections (every other day) with 250 μg of synthetic polyinosinic-polycytidylic acid (polyIpolyC; Sigma-Aldrich) to induce expression of the Mx-Cre transgene. The mice were sacrificed 3 weeks following the last injection, and liver tissues and HCC nodules were harvested. For FoxA2 expression in mouse liver, tail-vein injections (30-gauge needle) with 0.2 mL of adenovirus expressing FoxA2 or LacZ (1.7 × 10¹⁰ pfu/mL) in 10-month-old Alb-H-ras12V male mice were carried out under anesthesia.

RNA was TRIzol extracted (Invitrogen) and cDNA was synthesized using Bio-Rad reverse transcriptase (Bio-Rad). cDNA was amplified using SYBR Green (Bio-Rad) and analyzed via iCycler software. Western blots and chromatin-IPs were performed following previously described procedures (18). For chromatin-IPs, signals obtained with IgG and specific antibodies were first normalized with signals obtained with those antibodies on a nonspecific site in the GATA3 promoter (18). Normalized values were used to plot the fold enrichment with FoxM1-ab over IgG. For Rb-ChIP and DNMT3b-ChIP, we compared enrichments with the same antibody, after normalization, in the presence and absence of FoxM1-siRNA, and the fold enrichments in control siRNA over FoxM1-siRNA were plotted. All primer sequences and antibodies are included in Supplementary Table S1.

Genomic DNA (gDNA) was obtained from Huh7 cells and mouse tissue using the Wizard Genomic DNA Purification kit as instructed by the manufacturer's manual.

Genomic DNA samples were treated with EZ DNA methylation kit (Zymo Research) according to the manufacturer's recommendation. The extent of methylation of desired gene was then measured by qPCR amplification with pairs of specific primers as mentioned in Supplementary Table S1, which were designed using MethPrimer MSP/BSP prediction and primer designing tool. Quantitative MSP was performed with using SYBR Green (Bio-Rad) and analyzed via iCycler software. Each reaction contained 20 ng of bisulfite-treated DNA as a template, 6.25 μL SYBR Green PCR (Bio-Rad) and 4 mmol/L each forward and reverse primers in a total volume of 12.5 μL. The quantification cycle (C_q) was determined for each reaction with methylation-specific primers (MSP), and the ratio of unmethylated to total amplifiable bisulfite-treated DNA was calculated.

Statistical significance was calculated by the Student t test (22) and Pearson correlation coefficient. Statistically significant changes were indicated with asterisks (***, P < 0.001; **, P < 0.01; *, P < 0.05).

Expression of FoxM1 is low in nontumor human liver sections, whereas FoxA2 is expressed at high levels in those sections (Supplementary Fig. S1A). To investigate expression of FoxM1 and FoxA2 in HCC, we analyzed expression of the FoxM1 and FoxA2 proteins using tissue microarrays. We carried out IHC staining of tissue microarrays derived from consecutive sections of HCC specimens using specific antibodies against FoxM1 and FoxA2. There was an obvious difference in the expression pattern of FoxM1 and FoxA2. In grade 1 samples, FoxA2 is vividly detectable, whereas the nuclear expression of FoxM1 is low (Fig. 1A and B). In grade 3 specimens, on the other hand, expression of FoxA2 is low, but expression of FoxM1 is abundant. Pearson correlation analyses of FoxM1 and FoxA2 expressions in the consecutive TMAs indicated a strong negative correlation (Fig. 1C).

Recently, we studied the roles of FoxM1 in HCC progression using a transgenic mouse model that expresses oncogenic HRas in the liver (10). In that study, the floxed alleles of FoxM1 were deleted after HCC development using the MxCre deletion system, which is induced by injecting mice with double-stranded RNA (polyIpolyC). This system efficiently deletes floxed alleles in liver as well as in blood cells (23), and that somewhat mimics what would be expected from a drug that inhibits FoxM1. In those experiments, FoxM1 deletion in the HCC nodules was detected mainly in HCC cells (10). Moreover, we showed that deletion of FoxM1 after HCC development inhibits HCC progression (10). Sections from those tumor nodules were analyzed for FoxA2 expression by IHC. The HRas-derived tumor nodules without FoxM1 deletion exhibited very little expression of FoxA2 (Fig. 2A and B). But, in the FoxM1-deleted samples, there was a significant increase in the FoxA2 (Fig. 2A and B). The observation was confirmed by Western blot assays using extracts from tumor nodules with and without FoxM1-deletion (Fig. 2C). The increase in the expression of FoxA2 in the FoxM1-deleted samples provides in vivo genetic evidence that FoxM1 plays a role in the inhibition of FoxA2 in HCC. In normal mouse liver, FoxM1 is expressed at a low level, whereas the expression of FoxA2 is abundant (Supplementary Fig. S1B).

Next, we determined the effects of FoxM1b overexpression and FoxM1-knockdown on the levels of FoxA2 using the widely used HCC cell lines HepG2, Huh7, and SNU449. HepG2 cells express FoxM1 at low levels (Supplementary Fig. S2A); therefore, we used that line for overexpression experiments. SNU449 cells express FoxM1 at high levels, and we used that line for knockdown studies. Expression of FoxM1 in the Huh7 cells is in a range in between that observed in HepG2 and SNU449. Therefore, Huh7 was used for both knockdown and overexpression experiments. Expression of T7-tagged FoxM1b in Huh7 and HepG2 cells inhibited the levels of FoxA2 at both mRNA (Fig. 3A and B) and protein levels (Supplementary Fig. S2C). GAPDH-mRNA values were used to calculate relative fold expression of FoxA2-mRNA. Knockdown of FoxM1 in Huh7 and SNU449 cells caused increase in the levels of FoxA2 in both mRNA (Fig. 3B and C) and protein levels (Supplementary Fig. S2D-E). Also, we developed Huh7 stable cell lines in which FoxM1-shRNA can be expressed in an inducible manner by adding doxycycline in the culture medium. Expression of FoxM1-shRNA in 3 independent clones increased expression of FoxA2 proteins (Supplementary Fig. S2F).

To determine whether a direct mechanism is in play, we sought to determine whether FoxM1 directly targets the FoxA2 promoter. The human FoxA2 gene contains at least 4 putative FoxM1-binding elements (Supplementary Fig. S3A). Chromatin-IP experiments using FoxM1-ab detected enrichment of DNA fragments encompassing the sites at −1,294 and −4,156 in the FoxA2 gene (Fig. 3D). The other sites in the FoxA2 upstream regions did not show any significant enrichment over that with the IgG (Fig. 3D). Previously, we showed that FoxM1 binds to both DNMT3b and Rb, forming a repressor complex in breast cancer cells (18). Immunoprecipitation of Huh7 cell extracts with a monoclonal antibody against FoxM1 (Supplementary Fig. S3B) coimmunoprecipitated DNMT3b and Rb. Therefore, we carried out chromatin-IP experiments with Rb and DNMT3b antibodies using the Huh7 cells expressing control siRNA or FoxM1-siRNA. Both Rb and DNMT3b bound to the same promoter fragments that were enriched in chromatin-IP with FoxM1-ab (Fig. 3E and F). Moreover, knockdown of FoxM1 caused significant reduction in the bindings of Rb (Fig. 3E) and DNMT3b (Fig. 3F) onto the specific sites in the FoxA2 promoter. Chromatin-IP experiments with SNU449 cells further confirmed the binding of FoxM1, Rb, and DNMT3b onto the FoxA2 promoter (Supplementary Fig. S3C–S3E).

Recruitment of DNMT3b onto the FoxA2 promoter suggests that the repression by FoxM1 would involve methylation of CpG islands. To investigate that, Huh7 cells were transfected with FoxM1b-expresion vector or FoxM1-siRNA. Genomic DNAs from the transfected cells were treated with bisulfite followed by PCR using primers for the CpG islands in the FoxA2 promoter. Expression of FoxM1b caused an increase in CpG methylation near the FoxM1-binding sites in the FoxA2 gene (Fig. 4A; Supplementary Fig. S3F). Moreover, knockdown of FoxM1 caused a decrease in CpG methylation at those sites in the FoxA2 promoter (Fig. 4B; Supplementary Fig. S3G).

Next, we used an Rb-shRNA construct that allows inducible depletion of Rb in the presence of doxycycline (Fig. 4C). As shown in Fig. 4D, expression of FoxM1 increased methylation of the FoxA2 promoter in the presence of Rb (no doxycycline), but upon depletion of Rb (doxycycline) there was no increase in the CpG methylation at the indicated sites. Moreover, depletion of Rb caused increases in the expression of FoxA2 (Fig. 4E). Also, knockdown of DNMT3b increased expression of FoxA2 and reduced methylation of the FoxA2 promoter (Supplementary Fig. S4A–S4C). These observations demonstrate that the FoxM1/DNMT3b complex methylates and represses FoxA2 promoters requiring Rb. The extent of FoxA2 repression by FoxM1b varied between 40% and 75%, which is likely due to variations in the levels of active Rb in the transfected cells. Changes in CpG methylation were also detected in mouse HCC samples following deletion of mouse FoxM1 (Supplementary Fig. S4D–S4E) that binds to Rb and DNMT3b (Supplementary Fig. S4F).

Given that Rb is required for FoxM1-mediated repression of FoxA2, we predicted that repression occurs mainly in the G₁ phase in which Rb is in the underphosphorylated form and is most active. Also, it is the underphosphorylated form of Rb that was shown to associate with FoxM1 (24). The doxycycline-inducible FoxM1-shRNA cells were treated with vehicle or doxycycline (300 ng/mL) for 96 hours, and then treated with Hoechst that allows fluorescent staining of DNA in live cells. The cells were then fractionated using a cell sorter to obtain cell populations enriched for G₁, S, or G₂–M cells. As expected, expression of the FoxM1-shRNA (Fig. 5A) caused an increase in the G₁ population and reductions in the S and G₂–M cells (Fig. 5B and C). Clearly, depletion of FoxM1 (open bars in Fig. 5D) caused increase in the expression of FoxA2 in G₁ phase. Expressions of pluripotency genes, on the other hand, were inhibited in the FoxM1-depleted cells, and the inhibition was observed mainly in the S and G₂–M cells (Fig. 5E and F). There was no significant difference in FoxA2 expression in the G₂–M phases (Fig. 5F) because FoxM1 is phosphorylated by Plk1, which blocks binding to Rb (25). We repeated this experiment with 3 independent clones of FoxM1-shRNA cells and observed similar results. The changes in gene expression were not indirect effects of G₁ inhibition because similar changes were not observed for Oct-4 and Nanog in G₁ cells. Also, an unrelated gene, Fyn, did not show any significant changes in the G₁ and S-phase cells. Doxycycline treatment alone, in the absence of FoxM1-shRNA, did not exhibit any significant effect on FoxM1 or FoxA2 expression in parental Huh7 cells (Supplementary Fig. S5).

Expression of FoxM1 in HepG2 cells caused a significant increase in the number of spheres when cells were plated in sphere formation media (Fig. 6A and B). Expression of FoxA2, on the other hand, instead of increasing the number of spheres, caused decreases in the number of spheres in both HepG2 (Fig. 6A) and SNU449 cells (Supplementary Fig. S6A). That is also consistent with the observation that expression of FoxA2 caused inhibition of the pluripotency genes along with FoxM1 and FoxM1 target genes (Fig. 6C; Supplementary Fig. S6B and S6C). Moreover, expression of FoxA2 caused increases in the expression of the hepatocyte differentiation markers ALB, AAT, and HNF4α (Fig. 6D). It is noteworthy that inhibition of FoxM1 also increased expression of those differentiation genes (Supplementary Fig. S6D).

The inhibition of FoxM1 by FoxA2 is interesting because that could be the mechanism by which FoxA2 inhibits pluripotency genes. We did not detect an interaction between Rb and FoxA2. Therefore, we considered other possibilities. For example, FoxM1 was shown to autoactivate its own transcription (26). In chromatin-IP assays, we detected interactions of FoxM1 with multiple sites in the FoxM1 promoter (Fig. 6E). Because FoxA2 binds to similar cognate DNA-elements, we considered the possibility that FoxA2 could compete with FoxM1 and inhibit its binding. Consistent with that notion, we observed strong inhibitions of FoxM1 binding to its own promoter when FoxA2 was overexpressed (Fig. 6F). We analyzed the promoter-proximal binding sites because the enrichments on those sites were dependent upon FoxM1.

FoxM1 is a proproliferation transcription factor that also inhibits apoptosis and drives aggressive progression of cancers when overexpressed (21). If repression of FoxA2 is important, the prediction is that expression of FoxA2 would inhibit the FoxM1 pathways in HCC cells. We observed that expression of FoxM1 led to significant increases in clonogenicity of Huh7 cells (Supplementary Fig. S7A–S7B). Expression of FoxA2 alone did not show any significant effect over control, but when expressed in combination with FoxM1, they strongly inhibited the FoxM1-induced increased clonogenicity of Huh7 cells (Supplementary Fig. S7A–S7B). Similarly, coexpression of FoxA2 inhibited FoxM1-induced increase in soft-agar colonies (Supplementary Fig. S7C–S7D). As expected from the observation that FoxA2 inhibits autoactivation of FoxM1, we consistently observed that expression of FoxA2 affected the levels of total FoxM1. FoxA2 did not have any significant effect on the coexpressed Flag tagged FoxM1b levels (Supplementary Fig. S7E, Flag panel). However, expression of Flag-FoxM1b increased the levels of the endogenous FoxM1 (Supplementary Fig. S7E, top), and coexpression of FoxA2 inhibited the increase of the endogenous FoxM1 (Supplementary Fig. S7E, top). The results are consistent with a model in which FoxM1 and FoxA2 have opposite regulatory effects on each other.

We showed that FoxM1 is essential for progression of Ras-induced HCC (10). Based on the observation that FoxA2 inhibits FoxM1 expression, we predicted that FoxA2 would inhibit Ras-induced HCC progression. We tested that by tail-vein injection of FoxA2-expressing adenovirus (Ad-FoxA2) in 10-month-old Alb-Hras12V transgenic mice harboring HCC. Five male mice were injected with Ad-FoxA2 or Ad-LacZ twice. Second injection was done 12 days after the first, and livers were harvested 10 days after the second injection. Expression of FoxA2 caused significant reductions in the number of tumor nodules and tumor burden. Representative livers with HCC nodules are shown in Fig. 7A, and quantifications are shown in Fig. 7B-C. The inhibition of HCC progression was associated with inhibition of FoxM1 expression as well as Ki67⁺ cells (Fig. 7D–F). These observations provide in vivo evidence that FoxA2 inhibits FoxM1 and Ras-induced HCC progression.

Results presented here are significant in several ways. First, our observations provide insight into the mechanisms by which FoxM1 drives accumulation of poorly differentiated cancer cells during high-grade progression of HCC. We show that FoxM1 suppresses expression of the hepatocyte differentiation gene FoxA2 in G₁ phase, and that suppression is important for expression of the pluripotency genes in S–G₂ phases of the cell cycle. Also, we show that FoxA2 regulates FoxM1 expression by blocking its autoactivation mechanism. Our results suggest that it is the levels of FoxM1 versus FoxA2 that determine the differentiation state of HCC cells.

Reduced expression of FoxA2 coincides with overexpression of FoxM1 (Fig. 1). The need for overexpression of FoxM1 might be related to low abundance of active Rb in HCC cells. High levels of FoxM1 would be needed to seek out active underphosphorylated Rb protein because it is the underphosphorylated Rb that binds to FoxM1 (24). We suspect that the FoxM1-mediated inhibition of FoxA2 might be involved in dedifferentiation of the HCC cells or maintenance of poorly differentiated HCC cells. Alternatively, in the event that the low- and high-grade HCCs develop from different progenitors, we speculate that the FoxM1-mediated repression of FoxA2 is important for the progenitors that give rise to high-grade HCC. It is noteworthy that, although we show regulation at the RNA levels, TCGA data sets did not show opposite expression patterns of FoxM1 and FoxA2, which is likely related to RNA contributions from other cell types in the HCC nodules.

The involvement of Rb in the suppression of FoxA2 also suggests that the mechanism might be more active in G₁ phase where underphosphorylated Rb is abundant. Consistent with that, depletion of FoxM1 caused accumulation of FoxA2 mainly in G₁ phase. Interestingly, studies on embryonic stem (ES) cell differentiation indicated that G₁ phase is the phase in which the chromatin is available for the differentiation mechanisms, and that the ES cells retain pluripotency by suppressing differentiation mechanisms in G₁ and increasing expression of pluripotency genes in the S–G₂ phases (27). In the light of those observations in ES cells, our observations that FoxA2 inhibits expression of the pluripotency genes and that FoxM1 inhibits FoxA2 are interesting because they explain how FoxM1 overexpression maintains poorly differentiated state of cells in high-grade HCC. The involvement of Rb in the suppression of the FoxA2 gene is surprising because it suggests that Rb participates in progression of high-grade HCC. We show that FoxM1 recruits Rb and DNMT3b onto the promoter of FoxA2 and increases methylation of CpG islands in that promoter. Moreover, depletion of Rb blocks FoxM1-mediated increase in promoter methylation and suppression of the FoxA2 gene. Together, the results suggest that, in the context of overexpressed FoxM1, there is a gain of function for Rb, and that function is related to the suppression of differentiation genes, which is likely involved in high-grade progression of HCC.

We provide evidence for a new regulatory loop in which overexpression of FoxA2 inhibits expression of FoxM1 and the FoxM1 target genes in HCC cells. TCGA database analyses revealed that FoxA2 is rarely mutated in HCC. Therefore, an understanding of how the FoxM1 level increases in the presence of FoxA2 during HCC progression will require further analyses. FoxM1 enhancer/promoter are activated by multiple signaling pathways and cancer-relevant transcription factors, including Myc (28), HIF1 (29), and AP1 (30). Also, FoxM1 is regulated at the protein level. For example, CDK4 was shown to stabilize FoxM1 (31). Consistent with that, FoxM1 is expressed at high levels in HCC developed in the H-Ras12V transgenic mice, and that could be related to activation of the JNK1-AP1 as well as the cyclin D/Cdk4 pathways by activated Ras (32, 33). Also, the mechanism by which FoxA2 inhibits FoxM1 is unclear. It remains possible that a repression partner of FoxA2 is downregulated during HCC progression. We speculate that a combination of those pathways overrides inhibition by FoxA2 to increase expression of FoxM1 during high-grade progression of HCC.

No potential conflicts of interest were disclosed.

Conception and design: V. Chand, G. Guzman, P. Raychaudhuri

Development of methodology: V. Chand, G. Guzman

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): V. Chand, A. Pandey, D. Kopanja, G. Guzman

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): V. Chand, A. Pandey, G. Guzman, P. Raychaudhuri

Writing, review, and/or revision of the manuscript: V. Chand, G. Guzman, P. Raychaudhuri

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): V. Chand, G. Guzman

Study supervision: V. Chand, G. Guzman, P. Raychaudhuri

Other (contributed mainly to human liver tissue pathological data portion of the study): G. Guzman

The authors thank Dr. S. Elledge (Harvard Medical School, Boston, MA) for sharing the Rb-shRNA construct. The work was supported by grants from the NIH (CA 177655 and CA 175380) to P. Raychaudhuri. P. Raychaudhuri is also supported by a Merit Review Grant (BX000131) from the Veterans Administration.

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.