Epigenetic Modulation of miR-122 Facilitates Human Embryonic Stem Cell Self-Renewal and Hepatocellular Carcinoma Proliferation

The self-renewal capacity ascribed to hESCs is paralleled in cancer cell proliferation, suggesting that a common network of genes may facilitate the promotion of these traits. However, the molecular mechanisms that are involved in regulating the silencing of these genes as stem cells differentiate into quiescent cellular lineages remain poorly understood. Here, we show that a differentiated cell specific miR-122 exemplifies this regulatory attribute by suppressing the translation of a gene, Pkm2, which is commonly enriched in hESCs and liver cancer cells (HCCs), and facilitates self-renewal and proliferation. Through a series of gene expression analysis, we show that miR-122 expression is highly elevated in quiescent human primary hepatocytes (hPHs) but lost or attenuated in hESCs and HCCs, while an opposing expression pattern is observed for Pkm2. Depleting hESCs and HCCs of Pkm2, or overexpressing miR-122, leads to a common deficiency in self-renewal and proliferation. Likewise, during the differentiation process of hESCs into hepatocytes, a reciprocal expression pattern is observed between miR-122 and Pkm2. An examination of the genomic region upstream of miR-122 uncovered hyper-methylation in hESCs and HCCs, while the same region is de-methylated and occupied by a transcription initiating protein, RNA polymerase II (RNAPII), in hPHs. These findings indicate that one possible mechanism by which hESC self-renewal is modulated in quiescent hepatic derivatives of hESCs is through the regulatory activity of a differentiated cell-specific miR-122, and that a failure to properly turn “on” this miRNA is observed in uncontrollably proliferating HCCs.


Introduction
One of the hallmark traits ascribed to human embryonic stem cells (hESCs) is a capacity for extensive self-renewal [1]. While the indefinite cell division capacity of hESCs is an asset as an unlimited source of pluripotent cells for use in tissue replacement, the apparent similarity between stem cell self-renewal and cancer cell proliferation has raised concerns over the tumorigenic risk of hESCs and their derivatives in the transplanted tissue [2,3]. The phenotypic parallels between these two classes of cells have raised speculation that a common network of genes may be involved in facilitating the promotion of these traits, and that a failure to properly control the silencing of these genes in quiescent derivatives of hESCs may give rise to uncontrolled proliferation that is typical of many cancer cells. However, to date, the molecular mechanisms that involved in governing the regulation of this shared network of genes in hESCs and cancer cells remain poorly understood.
Although both hESCs and cancer cells possess an extensive cell division capacity that gives rise to identical progenies, unlike cancer cell proliferation, stem cell self-renewal is tightly regulated by extrinsic signals and intrinsic factors in the developing embryo [4][5][6][7]. During development, the transition from pluripotent to lineage specified cells is carefully orchestrated by turning on genes that are required for the specification of lineage determining cells and restricting the expression of genes that promote self-renewal [8,9]. As stem cells move further along the differentiation pathway toward various cell fate lineages, many differentiated cell types enter a quiescent or terminally differentiated state by exiting the cell cycle. This is especially evident during liver morphogenesis when stem cells differentiate into quiescent hepatocytes. While much attention has been focused on elucidating the mechanisms that are involved in promoting ESC self-renewal, less attention has been given to understanding how this trait is down-regulated as hESCs differentiate into quiescent cell types, and whether a failure to properly turn ''on'' these regulatory mechanisms can be observed in proliferating cancer cells.
In this study, we have examined the role of a differentiated cellspecific miR-122 as a regulator of a common network of genes that promote the facilitation of stem cell self-renewal as well as hepatocellular carcinoma cell (HCC) proliferation. microRNAs (miRNAs) are endogenous regulatory RNAs that refine and limit the expression of mRNAs post transcription by suppressing translation, and influencing the ultimate proteome [10]. Even though the number of miRNAs in the human genome is small relative to the number of protein coding genes, this family of RNAs has been described as one of the master regulators of the cell, because of their involvement in a wide range of cellular or organ programs including, for example, maintaining stemness, guiding differentiation, and functioning as tumor suppressors or oncomirs depending on the cell type in which they are expressed [11][12][13][14][15]. In the liver, miR-122 has been described as a critical facilitator of various homeostatic functions, notably, fatty acid and cholesterol metabolism, as well as hepatitis C viral replication [16][17][18][19][20]. Of relevance to this study are recent reports demonstrating the role of miR-122 as a potential tumor suppressor, because of its ability to regulate cell cycle progression and metastasis in liver cancer cells [21][22][23][24][25][26][27]. The ability of miR-122 to inhibit cell cycle progression in liver cancer cells raises the possibility that this miRNAs may also regulate hESC cell self-renewal during hepatocyte differentiation, by suppressing a common network of genes that promote hESC self-renewal as well as HCC proliferation.
Here, we show that miR-122 is highly enriched in differentiated human primary hepatocytes (hPHs), and functions as a modulator of hESCs self-renewal and HCC proliferation by suppressing the translation of a metabolic protein, PKM2. Through a series of global gene expression analysis, we show that miR-122 expression is either lost or severely attenuated in hESCs and HCCs, while an opposing expression pattern is observed for Pkm2. Both depleting hESCs and HCCs of Pkm2, or gain of miR-122 function, leads to a notable deficiency in self-renewal and proliferation. Likewise, during the differentiation process of hESCs into hepatocytes, a reciprocal expression pattern is observed between miR-122 and Pkm2, suggesting a possible role for this miRNA as a modulator of self-renewal during hepatic lineage specification. An examination of the genomic region up-steam of miR-122 uncovered hypermethylation in hESCs and HCCs, while the same region is demethylated and occupied by a transcription initiating protein, RNA polymerase II (RNAPII), in hPHs. Our findings suggest that one possible mechanism by which hESC self-renewal and HCC proliferation are modulated is through the regulatory activity of a differentiated cell-specific miR-122, which directly suppresses the translation of a gene, Pkm2, that is commonly enriched in hESCs and HCCs, and plays a role as a facilitator of these traits.

Results
mRNAs that are commonly enriched in hESCs and HCCs possess Gene Ontology (GO) functional terms describing actively dividing cells As a starting point of our study, we sought to determine whether the similarity between hESC self-renewal and HCC proliferation is reflected in the functional properties of mRNAs that are commonly enriched in these classes of cells. To do this, we performed a series of global mRNA expression profiles in hESCs, hPHs and HCCs (HepG2 and Hep3B), and compared their expression patterns. We show using hierchical clustering and Pearson correlation analyses that global gene expression patterns within each cell class are highly similar, while notable differences are observed among the three classes of cells ( Figure 1A-C). These observations indicate that hESCs, hPHs and HCCs possess gene expression signatures that are specific and unique for each class of cells. Pearson correlation further revealed that while a large dissimilarity exists between the correlation coefficients undifferentiated hESCs and differentiated hPHs (mean r 2 = 0.5441), HCCs appear to possess a signature that resembles ESCs (mean r 2 = 0.6258) as well as hPHs (mean r 2 = 0.6570), relative to the correlation between hPHs and hESCs ( Figures 1B and 1C). These observations led us to speculate whether a large fraction of genes that are commonly enriched in HCCs and hPHs may reflect a common tissue origin, while genes that are commonly enriched in HCCs and hESCs may possess functional relevance to self-renewal and proliferation. In order to evaluate this, the normalized data generated from the mRNA arrays were log 2 -transformed, and categorized as ''enriched'' based on a parameter set at log 2 value of 7.0 or higher. For HCCs, genes were considered ''enriched'' if both HepG2 and Hep3B cells met the enrichment criteria. We found that among the 17,172 mRNAs that were evaluated, 367 (2.14%) were commonly enriched in hESCs and HCCs, but not in hPHs, and 249 (1.45%) were commonly enriched in hPHs and HCCs, but not in hESCs ( Figure 1D). To evaluate the functional properties of these sets of genes, the program DAVID [28] was utilized to assess the functional classification of each gene based on Gene Ontology (GO) descriptions. GO classifications were then evaluated further using a feature in DAVID which permits the clustering these genes according to the similarity of their functional terms. Not surprisingly, we found that a large fraction of the gene set that is commonly enriched in hESCs and HCCs generated GO clusters describing functional activities that are highly relevant for actively dividing cells (e.g., M phase, DNA replication, DNA replication initiation, regulation of cell cycle, and chromosome segregation) ( Figure 1D, highlighted in red). In contrast, mRNAs commonly enriched in HCCs and hPHs, but not in hESCs, generated generated clusters describing functions that are generally associated with the homeostasis of differentiated cells (e.g., response to external stimulus, response to wounding, and immune and inflammatory responses), and various functions relating to the liver (e.g., reverse cholesterol transport and lipid metabolic process) ( Figure 1D). Taken together, these results show that a large fraction of mRNAs that are commonly up-regulated in both hESCs and HCCs, but not in hPHs, embody GO functional descriptions that reflect the self-renewal and proliferation traits ascribed to stem cells and cancer cells.
Differentiated cell-specific miR-122 is highly expressed in hPHs, and predicted to target a hepatic progenitor stem cell associated gene, Pkm2, that is commonly enriched in hESCs and HCCs Because mRNAs commonly up-regulated in hESCs and HCCs embody GO functional descriptions associated with self-renewal and proliferation, we sought to determine whether a differentiated cell-specific miR-122 facilitates the suppression of these traits in hESCs and HCCs by targeting mRNAs commonly up-regulated in these two classes of cells. To do this, we performed global miRNA expression profiles on hESCs, HCCs and hPHs to determine whether miR-122 is highly expressed in hPHs, but attenuated in hESCs and HCCs. By using this approach, we identified seven miRNAs, including miR-122, that are exclusively enriched in hPHs ( Figure 2A). We then utilized miRNA target prediction tools provided by miRNAMap 2.0 [29] and Target Scan [30] to examine whether these seven miRNAs contain seed sequences that are predicted to target the 39UTR of the 367 mRNAs that are commonly enriched in hESCs and HCCs ( Figure 1D and Figure 2B). Our analysis uncovered 18 out of 367 mRNAs that are predicted to be targeted by at least one of the seven hPH-specific miRNAs ( Figure 2C and 2D). In this list, we observed that miR-122 is predicted to target three genes (Ndrg3, Npepps, Pkm2). Among this list of genes that are predicted to be targeted by miR-122, we selected miR-122:Pkm2 pair for an indepth analysis, because the embryonic isoform of pyruvate kinase (Pkm2) is reported in the literature to be highly expressed in actively proliferating hepatic progenitor stem cells (hepatoblasts), as well as many cancer cells [31][32][33][34]. An examination of the 39UTR of Pkm2 using Target Scan and miRNAMap 2.0 revealed two potential binding sites for miR-122 ( Figure 2E). We, therefore, speculated that Pkm2 may facilitate the promotion of hESC selfrenewal and HCC proliferation, and that miR-122 may play a role as a modulator of these traits in differentiated hepatocytes by suppressing the translation of Pkm2.

hPH-specific miR-122 suppresses the translation of Pkm2
To determine whether miR-122 is capable of attenuates the expression of Pkm2 in vitro, we first validated the opposing expression patterns of Pkm2 and miR-122 in hESCs, HCCs and hPHs, using quantitative real time PCR (RT-qPCR). In addition to these cell types, telomerase-immortalized human fetal hepatocytes (hFHs) were evaluated as a representative sample of hepatic progenitor stem cells -during liver morphogenesis, actively proliferating fetal hepatocytes give rise to quiescent mature hepatocytes. The RT-qPCR results were consistent with the mRNA expression array data, and showed an elevated Pkm2 expression in hESC and HCCs, while in hPHs, Pkm2 was severely attenuated ( Figure 3A). An opposite expression pattern was observed for miR-122 in these cell types ( Figure 3B). As expected, hFHs mirrored the Pkm2 and miR-122 expression patterns that were observed in hESCs and HCCs.
Because established cell lines cannot fully recapitulate clinical malignancy, we also examined the enrichment pattern of miR-122 in primary liver tissues from three patients. Consistent with the observations in HCCs, we found that miR-122 expression was  Figure 1B. (D) The data generated from the mRNA expression arrays were normalized, log 2 -transformed, categorized as ''enriched'' based on an enrichment parameter set at log 2 value of 7.0 or higher. The enriched genes are distributed according to the Venn diagram. The three circles represent hESCs (green), hPHs (blue), and HCCs (orange). GO functional annotation clusters were derived from mRNAs that were commonly enriched in HCCs and hESCs (367), and HCCs and hPHs (249). The clusters were ranked according to the EASE scores of each term, and the top 20 clusters are listed in the tables. For each cluster, the GO term with the lowest P-value is shown as a representative functional term. The columns list the cluster ID, the enrichment score, GO ID of each term, P-value, fold enrichment, and the false discovery rate (FDR) of each term. Boxed in red are GO terms describing cell cycle regulation. doi:10.1371/journal.pone.0027740.g001 significantly attenuated in human liver tumor tissues ( Figure 3C). These results indicate that miR-122 expression is pervasively attenuated in self-renewing hESCs and proliferating liver cancer cells, and espouses an opposing expression pattern to Pkm2.
We then performed a series of tests using a luciferase reporter vector to evaluate whether miR-122 hybridizes to the in silico predicted sites in the 39UTR of Pkm2 and inhibit translation ( Figure 2E). To do this, a portion of the 39UTR of Pkm2 containing either of the two predicted miR-122 target sequences, and two derivative sequences with three-mismatch mutations, were cloned into separate luciferase reporter vectors ( Figure 3D). Each of these vectors was then co-transfected into HepG2 cells with precursor miR-122 molecules or mock precursor miR molecules. The cells were assayed for luciferase activity 36 hours post-transfection. We found that HepG2 cells that were transfected with either of the two wild-type luciferase constructs resulted in a significant reduction of the luciferase activity in the presence of precursor miR-122 relative to the control ( Figures 3E and 3F). Suppression of the luciferase activity was not observed when precursor miR-122 was co-transfected into HepG2 cells with luciferase constructs that contained the three-base mismatch mutation. These data indicate that miR-122 directly binds to the two predicted target sites in the 39UTR of Pkm2 and suppresses translation. Protein immunoblot assay validated the direct translational suppression of Pkm2 by miR-122 ( Figure 3G).
To ascertain the endogenous translational suppression of Pkm2 by miR-122, we performed a series of loss-of-function analyses by ectopically transfecting HepG2 and Hep3B cells with siPkm2, precursor miR-122 molecules, or mock precursor miR molecules. The transfected cell cultures were assayed 36 hours later for the level of Pkm2 expression using RT-qPCR. Consistent with the luciferase reporter assays, over-expression of miR-122 in HepG2 and Hep3B cells significantly reduced the level of Pkm2 transcripts relative to the mock miR transfections ( Figure 3H), indicating that miR-122 hybridizes to the 39UTR of Pkm2 and induces transcript destabilization. These results suggest that an absence of miR-122 expression in hESCs and HCCs may permit an increased expression of Pkm2. In contrast, the high level of miR-122 expression in hPHs likely attenuates the endogenous Pkm2 expression.
Pkm2 facilitates the promotion of hESC self-renewal and HCC proliferation To determine whether Pkm2 facilitates the promotion of hESC self-renewal and HCC proliferation, we performed a series of in vitro loss-of-function analysis by silencing the expression of Pkm2 or over-expressing miR-122 in these cells. To do this, we transfected HepG2 and Hep3B cells with siPkm2, precursor miR-122 molecules, or mock control miR molecules, and evaluated the culture for cellular proliferation 36 hours following the transfection procedure using an MTT based assay. Compared to the control, cells that were treated with siPkm2 or precursor miR-122 molecules yielded approximately 10-15% lower absorbance at 570 nm ( Figure 4A). Hoechst stained images of HepG2 and Hep3B cells at 24 and 36 hours post-transfection showed a notable difference in the amount of nuclei between the control, and both the siPkm2 and precursor miR-122 transfected conditions ( Figures 4B and 4C). Quantification of the Hoechst stained nuclei from randomly chosen regions of the cultures validated the proliferation deficiency in the siPkm2 and miR-122 transfected  Figure 2E were inserted into separate miR-REPORT vectors, immediately downstream of the Firefly luciferase gene, which is under the control of a constitutive CMV promoter. Downstream of MRE insert is polyadenylation signal (pA). The overlapping target sequences are colored in gray. In the mutant reporter constructs of WT#1 and WT#2 (MUT#1 and MUT#2), a three-basepair mismatch mutation (red, underlined) was introduced into the seed region. Reporter constructs as these are a standard method used to provide experimental evidence that miRNAs can directly repress translation initiation. (E, F) Each Firefly luciferase construct was cotransfected with precursor miR-122 molecules or mock miRNA molecules into HepG2 cells, and assayed for Firefly luciferase activity after 36 hours. Renilla luciferase construct was used as a normalization tool. The bar graphs show a decrease in luciferase activity when the cells were cotransfected with the wild-type (WT#1 or WT#2) MREs and precursor miR-122 molecules compared to the mock control. The difference in luciferase activity was not observed in mutant MREs (MUT#1 or MUT#2). (G) Detection of PKM2 by Western blot assay in HepG2 and Hep3B cells 3 days after transduction with lentivirus over-expressing miR-122 (miR-122) or scambled oligo sequence as control HCCs, relative to the control ( Figures S3A and S3B). When hESCs were transfected with siPkm2, precursor miR-122 molecules, or mock control miR molecules, we observed, in general, a more frequent occurrence of smaller sized colonies in the siPkm2 and miR-122 transfected conditions compared to the control ( Figure 4D). In addition, the morphology of a larger fraction of hESC colonies in the miR-122 transfected condition appeared differentiated/unhealthy ( Figure 4D and 4E). To examine the effect of knocking-down Pkm2 and over-expressing miR-122 on hESC pluripotency, we assayed the expression patterns for hESC markers (Pou5f1, Nanog, Sox2) ( Figure 5F), mature hepatocyte markers (Alb, a1At, Tf) ( Figure 4G), and early differentiation markers following the transfection procedure (Sox17, Brach, Gsc, Cxcr4, Foxa2, Mixl1) ( Figure 4H). We observed that hESCs that were transfected with siPkm2 maintained similar expression levels of these genes relative to the control. In contrast, hESCs that were transfected with precursor miR-122 molecules revealed an increased expression of one of the stem cell markers (Sox2), as well as one of the early differentiation markers (Gsc). The markers for mature hepatocytes (Alb, a1At, Tf) did not show a statistically significant difference compared to the control when hESCs were transfected with siPkm2 or miR-122 precursor molecules. These findings suggest that over-expressing miR-122 or knocking-down Pkm2 in hESCs and HCCs modulates self-renewal and proliferation. However, while knocking-down Pkm2 did not affect the ESC pluripotency associated genes that were observed, keeping the differentiated cell-specific miR-122 attenuated in hESCs appears essential in order to prevent the destabilization of a ESC marker genes, Sox2, and also, to inhibit the induction of an early differentiation gene, Gsc. An inverse expression pattern is observed between Pkm2 and miR-122 during hepatocyte differentiation of hESCs During the lineage specification process of hESCs into hepatocytes, the narrowing of the pluripotent potential is accompanied by a gradual loss of self-renewal. By the time stem cells have differentiated into mature hepatocytes, the cells are no longer self-renewing, and have settled into a quiescent state. Given the high enrichment of Pkm2 in rapidly self-renewing hESCs and proliferating HCCs, and the role of miR-122 as a direct suppressor of this gene, we asked whether the loss of self-renewal during the hepatocyte differentiation process may be accompanied by a decreased Pkm2 expression, and a corresponding increased expression of miR-122. To evaluate this, we utilized a three-step in vitro differentiation method for directing the specification of a large fraction of hESCs into hepatocyte-like cells [35][36][37], and examined the expression patterns of Pkm2 and miR-122 at 5 day intervals during the differentiation timeline. To verify the efficacy of the differentiation method, we confirmed the gradual downregulation of hESC associated genes (Pou5f1, Nanog, and Sox2) ( Figures 5A-C), and the up-regulation of mature hepatocyte markers (Alb, a1At, and Tf) (Figures 5D-F) during the same 5 day intervals using RT-qPCR. In addition, we also confirmed the presence of definitive endoderm (DE) at day 5 of our differentiation timeline, which recapitulates the gastrulation process in the developing embryo, and gives rise to hepatic and pancreatic progenitor cells [37] (Figure 5G). Our results show that after day 5, the level of Pkm2 expression began to decrease ( Figure 5H). As stem cells moved further along the differentiation pathway, Pkm2 expression continued to decrease with a corresponding increase in miR-122 expression. At day 30 of the differentiation timeline when a large fraction of hESCs have differentiated into mature haptocyte-like cells, a low level of Pkm2 expression was observed. and Tf-at 5 day intervals during the differentiation timeline, relative to day 0. n = 3, error bars represent s.e.m. Student's ttest results are indicated by (*) P,0.05 and (**) P,0.01 relative to hESCs (day 0). (G) RT-qRCR of genes reported to be up-regulated (Cxcr4, Gsc, Sox17, Foxa2, Mixl1) and down-regulated (Brach) in definitive endoderm. Solid bars represent the relative mRNA expression of hESCs. Dotted bars represent the relative mRNA expression of embryoid bodies (EBs) that were not treated with Activin A recombinant protein (Activin A induces definitive endoderm). Checkered bars represent the relative mRNA expression of EBs that were treated with Activin A recombinant protein to induce definitive endoderm cells. n = 3; error bars represent s.e.m. Student's t-test results are indicated by (*) P,0.05 and (**) P,0.01, relative to hESCs. (H) Relative levels of Pkm2 expression compared to miR-122 expression via RT-qPCR at 5 day intervals during the differentiation timeline of hESCs into mature hepatocytes, relative to day 0 (undifferentiated hESCs). The diagram in between the two graphs describes the differentiation method corresponding to the differentiation timeline. doi:10.1371/journal.pone.0027740.g005 However, this level was still slightly higher than the level that was observed in hESCs. We suspect that at this may reflect a heterogeneous population of cells in the culture condition, composed of mature hepatocyte-like cells, as well as some remaining hepatic and other progenitor stem cells. Overall, these observations show that a reciprocal expression exists between Pkm2 and miR-122 during hepatic lineage specification of hESCs. These findings suggest a role for miR-122 as a potential modulator of hESC self-renewal during the hepatocyte differentiation process by suppressing the translation of Pkm2.
The up-stream genomic region of miR-122 is occupied by RNAPII in hPHs but hyper-methylated in hESCs and HCCs In the preceding sections, we have shown that attenuating the expression of miR-122 in hESCs and HCCs allows for a constitutive expression of Pkm2, which in turn, facilitates the promotion of self-renewal and proliferation in these cells. We, therefore, sought to understand the molecular mechanism that may be involved in regulating the expression of this miRNAs in hESCs, HCCs and hPHs. To do this, we examined the 20 kb genomic region up-stream of miR-122, where the putative promoter sequence of this miRNA is likely to be found, for methylation status and binding activity of RNA Polymerase II (RNAPII) in hESCs, HCCs and hPHs. We chose to focus our attention on these epigenetic factors to because: 1) RNAPII has been described in the literature as the most prominent miRNA transcription initiating polymerase in metazoans [38]; and 2) hyper-methylation of the promoter region of genes has been reported to be frequently observed in aberrant gene silencing in a multitude of cancer cells [39]. We utilized a chromatin immunoprecipitation (ChIP) assay to examine the RNAPII binding activity and DNA methylation status of the 20 kb region upstream of the mature miR-122 sequences in the genomes of hESCs, hPHs and Hep3B. The ChIP data revealed that the upstream region of miR-122 in hPHs was covered with RNAPII, while in hESCs and Hep3B cells, the same region was hypermethylated ( Figure 6A). To verify these findings, we designed two PCR primer sets using templates from two separate regions (Chr18:54,264,000-54,265,000 and Chr18:54,268,000-54,269,000) that appeared to be hyper-methylated in hESCs and HCCs, and assayed for the presence of RNAPII and 5-MeC in chromatin immunoprecipitates of hESC, hPH and HCCs ( Figures 6B and 6C). These experiments validated hypo-methylation and the presence of RNAPII in hPHs and hyper-methylation in HCCs and hESCs. To determine whether these regions contain a putative promoter sequence for miR-122, two programs, Promoter 2.0 [40] and BDGP [41], designed to identify possible promoter sequences in the genome were utilized. Both Promoter 2.0 and BDGP revealed a high likelihood that theregion between54,264,000-54,265,000 may contain a possible promoter sequence ( Figure 6A; Figures S4A-C). We, therefore, performed additional experiments using bisulfite treatment of hESCs, HCCs and hPHs to validate whether this genomic region is hyper-methylated in hESCs and HCCs, while hypo-methylation in hPHs. By using PCR primers designed to amplify either the methylated or unmethylated regions of DNA after bisulfite treatment ( Figure 6D), we show that the predicted promoter region of miR-122 is methylated in hESCs and HCCs, but unmethylated in hPHs ( Figure 6E). We verified this finding using primers designed by the program MethPrimer program [42], which also showed that the predicted promoter region of miR-122 is unmethylated in hPHs, but methylated in hESCs and HCCs (Figures 7A-C). To confirm whether the promoter of miR-122 is located in this region, 59RACE was performed to determine the transcription start site (TSS). By using this approach, we show that the TSS is located within the predicted promoter region of miR-122. In addition, the CCAAT-box and TATA-box are found at the upstream region of this TSS, which is frequently observed in the promoters many genes regulated by RNAPII (Figures 8A, B). Hence, the combined results of ChIPchip, bisulfite sequencing, and 59RACE analyses suggest that in hPHs, RNAPII binds to the hypo-methylated promoter of miR-122 and initiates transcription. In contrast, hyper-methylation of the same genomic region appears to prevent the binding of this transcription factor in hESCs and HCCs, andinhibits the transcription of this miRNA. In order to verify these findings in vitro, we treated the HepG2 and Hep3B cells with a demethylating reagent, 5-Aza-29Deoxycytidine (5-aza-dC), and performed RT-qPCR 72 hours following the treatment to measure the level of miR-122 expression. We found that in HCCs that were treated with 5-aza-dC, a significant increase in miR-122 expression was observed compared to the control (Figures 8C and 8D). Moreover, we observed a decreased expression of Pkm2 in the same conditions (Figures 8E and 8F). These findings suggest that miR-122 expression in hESCs, HCCs and hPHs may be associated with the methylation status and RNAPII binding activity at the promoter region of this gene (Figure 9).

Discussion
The present work has provided evidence to suggest the involvement of a differentiated cell-specific miR-122 as a modulator of hESC self-renewal and HCC proliferation through a direct translational suppression of a gene, Pkm2, that is commonly enriched in these classes of cells. Both loss of Pkm2 and gain of miR-122 function in hESCs and HCCs lead to a severe deficiency in self-renewal and proliferation, and during the differentiation process of hESCs into hepatocytes, a reciprocal expression pattern is observed between miR-122 and Pkm2. These findings suggest a possible role for miR-122 as a modulator of self-renewal during hepatic lineage specification, and that a failure to properly attenuate the expression of this miRNA is observed in proliferating HCCs. An examination of the genomic region up-steam of miR-122 uncovered hypermethylation in hESCs and HCCs, while the same region is demethylated and occupied by a transcription initiating protein, RNA polymerase II (RNAPII) in hPHs. Our findings suggest that hyper-methylation of the genomic region up-stream of miR-122 may interfere with the binding of RNAPII, which in turn, inhibits a proper initiation of miR-122 transcription (Figure 9). The coordinated interplay between miR-122 and Pkm2 suggests a novel and elegant mechanism for controlling the expression of a gene that may be beneficial for stem cells, but becomes undesirable as stem cells transition into a differentiated quiescent hepatic cell fate. By initiating the expression of miR-122, the differentiating cells of a hepatic lineage have evolved an effective method for reducing the cell of a transcript that they have outgrown, in order to pave a path to ushering in a new phenotype for which self-renewal is undesired. A failure to elevate the expression of miR-122 is observed in hepatocellular carcinoma cells, and its effect on cellular proliferation suggests that miR-122 embodies a function that is reminiscent of a tumor suppressor.
The activation of a specific metabolic pathway involving the enzyme pyruvate kinase (PK) in highly proliferating cancer cells has been described decades ago by Otto Warburg [43] -a metabolic process that is now known as aerobic glycolysis or the Warburg effect [44]. Since the initial discovery, various studies have shown that while Pkm2 expression is primarily elevated in rapidly proliferating embryonic and progenitor stem cells, in terminally differentiated hepatocytes, PK is altered into a different isozyme, PK-L. Recent studies have demonstrated that in several cancer cells, the primary Pkm2 transcript is reexpressed, however, in an altered isoform that is capable of interacting with a phosphor-tyrosine (pTyr) motif, which enables these cells to redirect oxidative phosphorylation towards aerobic glycolysis for growth and proliferation [5,45,46]. The ramifications of these findings are significant in reference to this study, as they suggest that in differentiated normal hepatocytes, miR-122 may function as a critical tumor suppressor by negatively regulating the level of Pkm2 transcripts in the cell, thereby minimizing the likelihood that PKM2 may interact with pTyr. Hence, in addition to the many homeostatic functions that have been ascribed to miR-122 in the liver, this miRNA may also function as an early defense mechanism against a possible neoplastic transformation.
It remains to be seen whether silencing miR-122 in terminally differentiated hepatocytes may give rise to tumorigenesis. Such analysis is hampered at present by the lack of experimental strategy for keeping hPHs stable and/or alive for more than a few days after the isolation procedure from the cadaveric liver or hepatic resections, and the lack of techniques to identify, isolate, and culture mature hepatic derivatives from hESCs in vitro.
Although several studies have knocked-down miR-122 expression in the mouse liver using 'antagomirs,' Pkm2 up-regulation was not reported [19,47]. For this reason, examining the relevance of miR-122 in stem and cancer cells of human origin helps elucidate the important tumor suppressor potential of this miRNA during the differentiation process of hESCs into quiescent hepatocytes. Such information may also provide a useful tool as a biomarker to evaluate the safety and efficacy of hepatic derivatives of hESCs in an in vitro differentiation culture system. Furthermore, studies like this would be helpful for exploring additional avenues through which to study liver cancer development, and facilitate the designing of early prognostic tools and treatment strategies for therapeutic intervention. Given the multifaceted role of miR-122 in the liver [16][17][18][19][20][21][22][23][24][25][26][27], it is likely that this miRNA suppresses genes other than Pkm2 to modulate hESC self-renewal and HCC proliferation. For example, studies have shown that Igf1r is a direct target of miR-122, and may function as a mediator of HCC proliferation [48] (Figure S5). Future challenge will be to further elucidate the involvement of miR-122 as a regulator of hESC selfrenewal and HCC proliferation through direct translational suppression of genes in addition to Pkm2, as well as the role of various other differentiated cell-specific miRNAs that have yet to be explored.

Cell culture and differentiation
Human ES cells. Human H9 ES cell line was obtained from the WiCell Research Institute (WA09), and maintained as a monolayer in 6-well (9.6 cm 2 ) plates on gamma irradiated MEF feeder layers in Dulbecco's modified Eagle's medium/Ham's F-12 medium (Invitrogen) supplemented with 20% KnockOut Serum Replacement (Invitrogen), 4 ng/ml basic fibroblast growth factor (Invitrogen), 1 mM nonessential amino acids (Invitrogen), 2 mM L-Glutamine (Invitrogen), 100 u/ml penicillin/streptomycin (Invitrogen), and 0.55 mM 2-mercaptoethanol (Invitrogen). Cultures were passaged after collagenase treatment at a ratio of 1:3 to 1:6 every 4-6 days. Feeder-independent cultures of hESCs were maintained on matrigel (BD Biosciences) coated 6-well plates, in MEF conditioned medium. Induction of hESCs to definitive endoderm was initiated when hESC cultures were 75%-85% confluent, at which time they were washed with phosphate buffered saline (PBS) without calcium and magnesium (Invitrogen), and cultured in RPMI medium (Invitrogen) supplemented with 1XB27 (Invitrogen), 1 mM sodium butyrate (NaB) (Sigma Aldrich), 2 mM L-Glutamine, 100 u/ml penicilline/ streptomycin, 0.5% KOSR, and 100 ng/ml Activin-A (R&D Systems). Five days later, differentiation of definitive endoderm into hepatic progenitor cell types was initiated by changing to MDBK-MM medium (Sigma-Aldrich), supplemented with 2 mM L-Glutamine, 100 u/ml penicilline/streptomycin, 10 ng/ml FGF-4, and 10 ng/ml HGF. Three days later, differentiation and mRNA and miRNA expression microarrays mRNA arrays. RNA samples were isolated using the Qiagen RNeasy Kit (Qiagen). Prior to array experiments, the quality of the RNA samples were assayed using the Agilent Systems Bioanalyzer 2100. The total RNA from each sample was biotinylated and amplified for hybridization to Illumina Sentrix Expression Beadchip HumanRef-8 v3.0. This array platform consists of eight parallel strips, each strip composed of 24,500 probes from the NCBI refseq database (Build 36.2, Release 22). Arrays were processed and scanned per the manufacturer's instruction, and analyzed using the BeadStudio Software v3.0. All signals were normalized, log 2 -transformed, and ranked according to the log 2 values. For each gene, the criterion for enrichment was set at a log 2 value of 6.0 or higher. Hierarchical clustering was performed with average linkage and Pearson correlation. To generate the heatmap, values were centered and normalized to a mean of 0 and a standard deviation of 5. miRNA arrays miRNAs were prepared using the mirVana miRNA Isolation Kit from Ambion (Ambion), and purified and concentrated for small nucleic acids using flashPAGE Fractionator and flashPAGE Reaction Clean-up Kit (Ambion). Amine-modified nucleotides were added to the miRNAs in the samples, and labeled with Cy3 fluorescent dye per the manufacturer's instructions (Ambion). Labeled miRNAs were hybridized to Human miRNA Array v1.0 (Ambion) containing 365 human miRNA probes. Processed arrays were scanned at 10-um resolution using the GenePix 400B scanner (Axon Instruments). Raw data were normalized, log 2transformed, and ranked according to the log 2 values. For each miRNA, the criteria for enrichment was set at log 2 value of 6.0 or higher. All data is MIAME compliant, and the raw data has been deposited at GEO under accession number (GSE29907).

DAVID analysis
Functional annotation clusterings were performed using the DAVID Bioinformatics Resources 2008. Gene sets that were common between HCCs and hESCs, and HCCs and hPHs, were subjected to separate clustering analyses. Each gene set was entered into DAVID's functional annotation clustering tool, which generated clusters of genes based on the similarity of the functional terms assigned to each gene. The clusters were then ranked according to the EASE scores of each term, and the top ten highest ranked clusters were selected for analysis. Within each cluster, the top five Gene Ontology terms with the lowest P values were selected as representative functional terms. If redundant terms were present (e.g., developmental process and cellular developmental process) the term with higher P value was eliminated.

Chromatin immunoprecipitation assays and amplicon preparation
ChIP assays were performed as described at http://www. genomecenter.ucdavis.edu/farnham. Briefly, for 5-Methylcytidine (5-MeC) ChIPs, genomic DNA was extracted from 1610 7 cells of each sample, sonicated to an average size of 800 bp using a Bioruptor Sonicator (Diagenode), denatured, and incubated with the mouse monoclonal 5-MeC antibody (Eurogentec). RNAPolII ChIP was performed by cross-linking 1610 7 cells from each sample with 1.5% formaldehyde, after which time, nuclear extracts were prepared and chromatin was sonicated to an average size of 800 bp and incubated with mouse monoclonal RNA polymerase II antibody (Covance). For both 5-MeC and RNAPolII assays, secondary rabbit anti-mouse IgG (MP Biomedicals) was used, and the nonspecific rabbit IgG (Alpha Diagnostics) was used as a negative control. Immunoprecipitates were purified (QIAquick PCR purification kit, Qiagen), and amplicons were generated using the Sigma GenomePlex WGA2 Kit (Sigma-Aldrich). PCR positive and negative primers were used for ChIP samples and amplicons to confirm the validity of the final product.

Assay and analysis of ChIP-chip data
The miR-122 promoter array was custom designed by tiling 20 kb region upstream from the mature miR-122 DNA sequence with 50-mer probes on 386K array platforms. Amplicons were labeled, hybridized to the promoter arrays, scanned, and analyzed for signal intensity of each probe by Roche-NimbleGen Systems, Inc. To confirm RNAPII binding and methylation, primers were designed from 501 bp regions of chr18:54,264,000-54,264,500 and chr18:54,268,000-54,268,500, and RT-PCR was performed using 1.0 ng of amplicons generated from RNAPII and 5-MeC ChIP samples (hESCs, hPHs, HepG2, and Hep3B). Quantitative RT-PCR For mRNA RT-qPCR, total RNA was isolated using the Qiagen RNeasy Kit (Qiagen), and SuperScript III RT-qPCR Kit (Invitrogen) was used to synthesize cDNAs. For miRNA RT-qPCR, miRNAs were isolated using mirVana miRNA Isolation Kit (Ambion), and reverse transcribed into cDNAs using miRCURY LNA First Strand cDNA Kit (Exiqon). For both mRNAs and miRNAs, RT-qPCR mixture was prepared using ABI TaqMan or Sybr Master Mix (ABI), and RT-qPCR were performed on the ABI PRISM 7700 Sequence Detection System (ABI). The comparative 22(DDCt) method was used to determine the relative quantitative levels of mRNAs using GAPDH for mRNA normalization and Exiqon's Endogenous Control Primers (Exiqon) for miRNA normalization, and expressed in values as relative difference compared to the relevant controls.

Luciferase reporter construct and assay
The luciferase reporter was constructed by cloning into pMiR-REPORT vector (Ambion) the target PKM2 sequences of miR-122 downstream of the firefly luciferase gene and verified by sequencing. The reporter plasmids were co-transfected with 20 pmol of precursor miR-122 molecules or mock precursor miRNA molecules (Ambion) into 50,000 HepG2 cells in 24-well plates using Lipofectamine 2000 (Invitrogen). Cells were lysed 36 hours later and processed for luciferase assay using the Luciferase Reporter Assay System (Promega).

Lentiviral preparation and transduction
The production and titering of lentinvirus were carried out according to protocols from Tronolab (http://tronolab.epfl.ch).
Cell proliferation assay 50,000 HepG2 and Hep3B cells were grown on 24-well plates, and transfected with 20 pmol of precursor miR-122 molecules or mock precursor miRNA molecules (Ambion) using Lipofectamine 2000 (Invitrogen) 24 hours after initial plating of the cells. Cell proliferation was measured 24 hours later using the Cell Titer Cell Proliferation Assay (Promega) at 570 nm absorbance. 59 rapid amplification of cDNA end (RACE) analysis 59RACE was performed using the GeneRacer Kit (Invitrogen).

Bisulfite analysis
hESCs, HCCs and hPHs were treated with CpGenome Bisulfite Modification Kit (Millipore). The genomic DNA was extracted and PCR amplified. The PCR products were cloned into plasmids for sequencing (Invitrogen).