HepaRG cells were induced to differentiate into morphologically and functionally mature hepatocyte-like cells. Differentiated HepaRG cells showed features of normal hepatocytes, such as refractile cellular borders, clearly delineated nuclei and tridimensional polarization with the appearance of refringent circular canaliculi vertically (Figure 1). In order to identify the genomic events associated with HepaRG cell differentiation, total RNA and polysome-bound RNA were isolated at the proliferative stage and at the end of the differentiation protocol and analyzed on Affymetrix Human Genome U133A arrays (Figure 1). We separated polysomes from free messenger ribonucleoproteins (mRNPs) using sucrose gradient centrifugation with the assumption that translationally inactive mRNAs are present as free cytoplasmic mRNPs, whereas actively translated mRNAs are contained within polysomes. Total RNA was processed in parallel for each sample.

Out of the 22,283 probe sets spotted on the array, 3,071 (13.8%) were modulated by at least 2-fold upon differentiation and in 3 independent experiments, either in the total RNA or the polysome-bound RNA compartments. Total RNA fold changes were plotted against polysome-bound RNA fold changes for these 3,071 probe sets (Figure 2a). The correlation coefficient for the regression curve calculated from all values was 0.38, demonstrating a poor correlation and, therefore, an uncoupling phenomenon between changes in the polysome-bound fractions and changes in total RNA upon differentiation of HepaRG cells. We then determined the distribution of up- and down-regulated transcripts in each RNA population upon differentiation. In the total RNA compartment, 547 and 1,636 probe sets (a total of 2,183) were up-regulated and down-regulated, respectively. In contrast, in the polysome-bound RNA compartment, 1,325 and 124 probe sets (a total of 1,449) were up-regulated and down-regulated, respectively (Figure 2b). Transcription is, therefore, largely down-regulated during HepaRG differentiation while translation of specific genes is up-regulated. Probe sets that are similarly up-regulated or down-regulated in both RNA populations correspond to genes modulated as a result of transcriptional regulation without any subsequent translational control. These probe sets represented only a small number of genes with 359 up-regulated and 88 down-regulated probe sets. They represented 14.6% of the initially selected 3,071 regulated probe sets (Figure 2b, dark portions of the graph bars). On the other hand, 2,624 probe sets (85.4% of the total number of regulated probe sets) were modulated due to translational control (Figure 2b, gray portions of the bar graphs).

A subset of genes was selected for validation. Validation was performed using real-time PCR on the total RNA and the polysome-bound RNA populations, for ten genes: those encoding apolipoprotein H, solute carrier (SLC)27A3, cytochrome P450 isoforms 3A4 and 7B1, vascular endothelial growth factor (VEGF), E-cadherin, insulin receptor, leptin receptor, transforming growth factor (TGF) beta receptor 2 and membrane metallo-endopeptidase (MME). The PCR results obtained on the three independent experiments confirmed the microarray data for all ten genes (Figure 3a). Validation was also performed using real time PCR on each fraction of the sucrose gradient separating free mRNPs and polysomes, for three genes: those encoding latent transforming growth factor beta binding protein 1 (LTBP1), spectrin repeat-containing nuclear envelope 1 (SYNE-1) and matrix metalloproteinase 3 (MMP3). A shift was observed upon HepaRG differentiation for all three transcripts from the free mRNP fractions to the heavier polysome fractions on the sucrose gradient as shown in Figure 3b for LTBP1. These results demonstrate an increased translation of these transcripts and validate the array data indicating no change or a slight decrease in LTBP1, SYNE-1 and MMP3 transcript levels in the total RNA compartment and a strong up-regulation of all three transcripts in the polysome-bound RNA compartment.

All together, these results suggest that translational control plays a prominent role in the hepatocytic differentiation of liver progenitor cells and that the total RNA content may not be representative of the mature phenotype of hepatocyte-like cells. In addition, transcriptional changes did not overlap with translational changes. The large majority of polysome-bound (that is, translated) genes modified were up-regulated whereas the majority of genes modified at the total RNA level were down-regulated, suggesting that the mature hepatocyte phenotype is acquired by increased translation of pre-existing transcripts. The total RNA population can be considered as a stock of translated and untranslated transcripts that can be utilized by the cell rapidly. The more diverse the total RNA population is, the greater the options the cell has in selecting protein expression patterns. Therefore, the extensive down-regulation of genes in the total RNA compartment can be interpreted as a decrease in cellular RNA diversity, consistent with the commitment of a dedifferentiated epithelial progenitor into a defined, in this case hepatocytic, lineage.

To further characterize the differentiated phenotype of HepaRG cells, we selected all polysome-bound up-regulated probe sets (n= 1641) and all polysome-bound down-regulated probe sets (n= 204), regardless of their fold-change status at the total RNA level. The content of these two lists of genes were separately analyzed using the Ingenuity Systems Pathways Knowledge Base 22. This database enables one to search for gene products' interactions and annotations coming from curated data from publications and peer-reviewed resources. Networks displaying significant overlap between the selected regulated genes found in our study and the software-preselected members were selected. The Ingenuity pathway analysis identified nine networks (networks A-I) and one network (network J) generated from the up-regulated and down-regulated transcripts, respectively (Table 1 and Additional data file 1). These ten networks can be divided into six groups based on their associated biological top functions: cell cycle, cell death, innate immunity, lipid and drug metabolism, cell morphology, and cell environment and movement.

Most of the genes identified in this study and contributing to the differentiation phenotype were modulated by translational control. Translationally regulated transcripts are underlined in Table 1 and indicated in blue in Additional data file 1. To investigate whether translational control specifically affects transcripts involved in defined cellular functions, we calculated the percentage of translationally controlled probe sets in each of the ten networks A-J described above. Paired t-tests were performed between groups of networks sharing the same cellular functions (Figure 4). A significantly greater involvement of translational control was observed in networks related to cell cycle and cell death functions than in networks related to lipid metabolism and drug metabolism (p = 0.005). Likewise, a significantly stronger involvement of translational control was found in innate immunity-related networks compared to cell environment and cell movement-related networks (p = 0.027). The high percentage of translationally controlled probe sets in cell cycle and cell death-related networks is in agreement with the ability of the hepatocyte to massively and rapidly proliferate under acute liver injury, as well as with the hypersensitivity of the hepatocyte to cell death in response, for example, to drug-associated toxicity. Translationally regulated transcripts associated with cell cycle included the nuclear proteins SMARCA2 and SMARCB1, the transcription factors MEF2C, MEF2D and EGR1 and the NF-KB inhibitor NFKBIA. Translationally regulated transcripts associated with cell death included oncostatin M receptor/IL6ST and the initiator caspases 8 and 9. Translationally regulated transcripts associated with innate immunity included several interferon-associated genes, such as those encoding OAS1, IRF3 and IFI16. Finally, numerous transcription factors associated with inflammation were translationally upregulated and included the three liver-enriched transcription factors RARA, RXRA and RXRB and STAT6 (Table 2).

Numerous transcription factors were translationally upregulated while left unchanged or even decreased at the total RNA level. Translational control of these transcription factors provides the cell with a means to modify its phenotype in a timely manner, rapidly expressing genes downstream of these transcription factors. The hepatocyte has to be a highly versatile cell because of at least two of its functions: the ability to generate the acute phase reaction and to maintain blood homeostasy after meals as the first line organ downstream of the portal vein that carries nutrients from the digestive tract.

The importance of translational control during liver progenitor cell differentiation raises the question of the identity of the actors involved. We recently reported a functional down-regulation of the mTOR/4E-BP1/p70S6 kinase pathway during differentiation of HepaRG cells 58. Moreover, forced expression of an activated mutant of mTOR impairs hepatocytic differentiation in this model 58. This pathway may therefore contribute at least partially to some of the translational events described here.

Some genes were similarly modified upon differentiation of HepaRG cells, in both the total and the polysome-bound RNA populations, indicative of a transcriptional regulation. These include 435 up-regulated and 142 down-regulated probe sets (Figure 2b), indicated in yellow in Additional data file 1 and not underlined in Table 1. These genes corresponded in the majority to liver-enriched transcripts and to genes involved in lipid and drug metabolism. They included those encoding PPARA, PPARD, CEBPA, the hepatic leukemia factor (HLF) and the alcohol dehydrogenases 1B, 1C and 6. Other transcriptionally regulated genes included those encoding plasma proteins synthesized in the liver: the SERPINs A1, A4, F2, several complement system subunits (C1S, C3, C4A, C5 and C6) and three forms of fibrinogen (A, B and G). Finally, several cytokines, chemokines or hormones and their receptors were transcriptionally regulated as well: TNFSF10/TRAIL, IL6R, BMP2 and PDGFB (Table 2).

As the contribution of transcription appeared restricted to selective genes during HepaRG cell differentiation, we sought to investigate the expression levels and phosphorylation status of the canonic hepatocytic transcription factors HNF1α and HNF4α throughout differentiation. HNF1α is a major player in the acquisition of central hepatocytic functions, including gluconeogenesis, carbohydrate synthesis and storage, lipid metabolism (synthesis of cholesterol and apolipoproteins), detoxification (synthesis of cytochrome P450 monooxygenases), and synthesis of serum proteins (albumin, complements, and coagulation factors) 59. Interestingly, neither total nor polysome-bound RNA levels of HNF1α were modulated (-1.38 and +1.48-fold, respectively). This observation was confirmed by real time PCR (+1.38 ± 0.08 fold (mean ± standard error of the mean (SEM)) in total RNA and +1.02 ± 0.19 fold (mean ± SEM) in polysome-bound RNA; Figure 5a). In addition, no changes were observed at the protein expression level nor in phosphorylation status for HNF1α (55% of HNF1α is phosphorylated at the proliferative stage versus 38% at the differentiated stage; Figure 5b). HNF4α was slightly increased in both total and polysome-bound RNA (+1.89-fold and +1.35-fold, respectively). These slight increases were confirmed by real time PCR (+2.71 ± 0.13 fold (mean ± SEM) in total RNA and +1.74 ± 0.06 fold (mean ± SEM) in polysome-bound RNA; Figure 5c). However, HNF4α phosphorylation was strongly induced upon differentiation (Figure 5d), suggesting that, in contrast to HNF1α, HNF4α may contribute to HepaRG cell differentiation. Mutations of HNF1α associated with metabolic diseases have been described 6061 and, therefore, we cannot exclude that the lack of regulation of HNF1α found in this study results from mutation(s) disrupting its biochemical characteristics. However, the patient that gave rise to HepaRG cells was not known to be affected by any of these diseases.

In conclusion, transcriptional control appears to play a highly selective role in the phenotype of liver progenitor cell maturation and specifically targets liver-enriched transcripts characteristic of the mature hepatocytic phenotype. Novel findings suggest that the complement system is induced during maturation following transcriptional regulation.

Several transcripts were strongly transcriptionally induced upon HepaRG cell differentiation while unchanged or induced to a much weaker level in the polysome-bound RNA population, suggesting a translational repression control. Examples include E-cadherin, involved in hepatocytic polarization, cytochrome P450 3A4, a steroid-inducible cytochrome P450 isoform, cytochrome P450 7B1, a cytochrome P450 isoform involved in cholesterol metabolism, cytochrome P450 2A6 and 2C19, cytochrome P450 isoforms involved in drug metabolism, TGF-β receptor 2 and VEGF, an important regulator of angiogenesis and metastasis (Table 2).

Interestingly, four isoforms of cytochrome P450 were strongly up-regulated at the total RNA level but not at the polysome-bound RNA level. Given that cytochromes are inducible proteins involved in drug and lipid metabolism, high levels of untranslated RNA could serve as a stock that may be rapidly translated and used for the detoxification and acute phase-associated functions of the hepatocyte.

The HepaRG cell line was cultured in William's E medium (Invitrogen, Carlsbad, CA, USA) supplemented with 10% fetal calf serum (Mediatech, Manassas, VA, USA), 100 units/ml penicillin, 100 μg/ml streptomycin (Invitrogen), 5 μg/ml insulin (Sigma-Aldrich, St. Louis, MO, USA), and 5 × 10^-5 M hydrocortisone hemisuccinate (Sigma-Aldrich). To induce differentiation, a two-step procedure was used as previously described 1415. Cells were seeded at a density of 4 × 10⁴ cells/cm² and maintained for 2 weeks in the growth medium. Then, the culture medium was supplemented with 1% DMSO (Sigma-Aldrich) and 20 ng/ml EGF (PeproTech, Rocky Hill, NJ, USA) for 2 additional weeks. Cells were harvested either at 2 days (proliferative stage) or at 28 days (differentiation stage) after seeding. Cell culture pictures were taken using a phase contrast microscope (Nikon). Differentiation was evaluated morphologically by counting bile canaliculi (refringent area) at the intersection of two or three hepatocyte-like cells.

Total RNA was extracted, precipitated and resuspended in RNAse-free water using Trizol reagent (Invitrogen) according to the manufacturer's instructions. For polysome fractionation, cycloheximide (100 μg/ml) was added to the medium for 3 minutes prior to harvest. The medium was then removed and the cells were washed with ice-cold phosphate-buffered saline containing 100 μg/ml cycloheximide. The cells were then scraped, centrifuged at 800g for 5 minutes at 4°C and cytoplasmic RNA was obtained by lysis of the cell pellet in 1 ml of polysome buffer containing 10 mM Tris-HCl (pH 8.0), 140 mM NaCl, 1.5 mM MgCl2, 0.5% Nonidet P-40, and a ribonuclease inhibitor, RNasin (500 units/ml; Promega, Madison, WI, USA). After the removal of nuclei, the cytosolic supernatant was supplemented with 100 μg/ml cycloheximide, 665 μg/ml heparin, 20 mM dithiothreitol, and 1 mM phenylmethanesulfonyl fluoride. Mitochondria and membrane debris were removed by centrifugation, and the post-mitochondrial supernatant was overlaid onto a 15-40% sucrose gradient and spun at 38,000 rpm for 2 h at 4°C in a SW41Ti rotor (Beckman Coulter, Fullerton, CA, USA). Fractions (750 μl) were collected from the top of each gradient and deproteinated with 100 μg of proteinase K in the presence of 1% SDS and 10 mM EDTA. After acid phenol extraction, RNA integrity was controlled by electrophoresis analysis on 1.2% agarose gel. Densitometry (GelDoc, Bio-Rad Laboratories, Hercules, CA, USA) was used to identify the fractions in which the 28S/18S ratio equals 2 (that is, fractions corresponding to polysome-bound RNA). These fractions were pooled from each sucrose gradient.

Total and polysome-bound RNAs were purified using the RNeasy mini-kit clean-up protocol (Qiagen, Valencia, CA, USA), RW1 buffer being used to efficiently remove heparin from the samples. The first-strand cDNA, the double-strand cDNA, and cRNA were synthesized, and cRNA was fragmented using Affymetrix kits and guidelines 62. All cRNA final products were tested in terms of amount and integrity by Bioanalyzer (Agilent Technologies, Santa Clara, CA, USA) prior to microarray hybridization. cRNA samples were processed on Affymetrix HGU133A arrays with strict adherence to the labeling, hybridization and staining protocols provided by Affymetrix. A 'present' (P), 'marginal' (M) or 'absent' (A) call was assigned to each probe set using Affymetrix GeneChip Operation Software (GCOS v1.4). Probe sets with an 'absent' (A) call on all arrays were filtered out. Background correction and normalization steps were carried out using the GC-RMA method and the R Bioconductor software 63. Microarray data have been deposited in the ArrayExpress repository 64 under the accession number E-MEXP-1082. Three independent differentiation processes were performed and the correlation coefficients between each duplicate at the proliferative and the differentiated level were calculated using scatter plots. For total RNA, the correlation coefficients were 0.98 and 0.95 for proliferative and differentiated cells, respectively. For polysome-bound RNA, the values were 0.98 and 0.97 for proliferative and differentiated cells, respectively. The Ingenuity pathway analysis 22 was used to analyze selected probe sets. Each gene identifier was mapped to its corresponding gene object in the Ingenuity Pathways Knowledge Base. The application utilizes a right-sided Fisher's exact test to identify networks that had higher odds ratio of containing significant genes. These genes, called Focus Genes, were then overlaid onto a global molecular network. Networks of these Focus Genes were then algorithmically generated.

Cells were lysed in 50 mM Tris-HCl (pH 8), 150 mM NaCl, 0.1% SDS, 1% NP-40 supplemented with protease inhibitors (Complete, Roche Diagnostics, Indianapolis, IN, USA). Thirty micrograms of proteins were resolved on 10% SDS-polyacrylamide gels and electrotransferred onto nitrocellulose membrane (Amersham Biosciences, Piscataway, NJ, USA). Equal loadings and homogeneous blotting were confirmed using Ponceau red staining. Membranes were blocked with 5% nonfat milk in Tris-buffered saline and incubated with primary antibodies (anti-HNF1α and anti-HNF4α, Santa Cruz Biotechnology, dilution 1/500, Santa Cruz, CA, USA) overnight. Horseradish peroxidase-conjugated immunoglobulins (Dako, dilution 1/1,000, Carpinteria, CA, USA) were used as secondary antibodies and proteins were visualized with enhanced ECL chemiluminescent reagent (Amersham Biosciences). Densitometry was performed using the Total Lab TL100 software.

One microgram of DNAse I-treated (Promega) total RNA or polysome-bound RNA was reverse transcribed using Moloney murine leukemia virus reverse transcriptase and random hexamers (Invitrogen) for 50 minutes at 42°C. cDNA mixtures (1/10) were mixed with an equal volume of 2 × iQ SYBR green supermix (Bio-Rad Laboratories). Amplification was then performed at an annealing temperature of 55°C or 60°C and an elongation time of 30 s or 1 minute on a MyIQ real-time PCR apparatus (Bio-Rad Laboratories). Primer sequences were obtained through the Primer Bank website 6566 and are described in Additional data file 2. Differential expression ratios between proliferative and differentiated stages were calculated using the Δ(ΔCt) formula. Specificity of all amplicons was assessed by post-run melting curve analysis and agarose gel electrophoresis. For the analysis on sucrose gradient fraction distribution, 100 μl of each fraction harvested from the sucrose gradients were purified using the RNAeasy kit (Qiagen). Ten microliters of collected RNA were DNAse-digested and reverse transcribed as described above. Since the RNA amount in each fraction was different, in order to avoid efficiency differences of the reverse transcriptase, RNA amounts were equalized by adding an appropriate amount of in vitro transcribed irrelevant RNA to each fraction, giving a final amount of 1 μg of RNA on each sample. One tenth of each cDNA reaction was processed by real-time PCR as described above.