Potential conflict of interest: Nothing to report.
Molecular classifications defining new tumor subtypes have been recently refined with genetic and transcriptomic analyses of benign and malignant hepatocellular tumors. Here, we performed microRNA (miRNA) profiling in two series of fully annotated liver tumors to uncover associations between oncogene/tumor suppressor mutations and clinical and pathological features. Expression levels of 250 miRNAs in 46 benign and malignant hepatocellular tumors were compared to those of 4 normal liver samples with quantitative reverse-transcriptase polymerase chain reaction. miRNAs associated with genetic and clinical characteristics were validated in a second series of 43 liver tumor samples and 16 nontumor samples. miRNA profiling unsupervised analysis classified samples in unique clusters characterized by histological features (tumor/nontumor, P < 0.001; benign/malignant tumors, P < 0.01; inflammatory adenoma and focal nodular hyperplasia, P < 0.01), clinical characteristics [hepatitis B virus (HBV) infection, P < 0.001; alcohol consumption, P < 0.05], and oncogene/tumor suppressor gene mutations [β-catenin, P < 0.01; hepatocyte nuclear factor 1α (HNF1α), P < 0.01]. Our study identified and validated miR-224 overexpression in all tumors and miR-200c, miR-200, miR-21, miR-224, miR-10b, and miR-222 specific deregulation in benign or malignant tumors. Moreover, miR-96 was overexpressed in HBV tumors, and miR-126* was down-regulated in alcohol-related hepatocellular carcinoma. Down-regulations of miR-107 and miR-375 were specifically associated with HNF1α and β-catenin gene mutations, respectively. miR-375 expression was highly correlated to that of β-catenin–targeted genes as miR-107 expression was correlated to that of HNF1α in a small interfering RNA cell line model. Thus, this strongly suggests that β-catenin and HNF1α could regulate miR-375 and miR-107 expression levels, respectively. Conclusion: Hepatocellular tumors may have a distinct miRNA expression fingerprint according to malignancy, risk factors, and oncogene/tumor suppressor gene alterations. Dissecting these relationships provides a new hypothesis to understand the functional impact of miRNA deregulation in liver tumorigenesis and the promising use of miRNAs as diagnostic markers. (HEPATOLOGY 2008.)
Hepatocellular adenomas (HCAs) and hepatocellular carcinomas (HCCs) are liver tumors that are heterogeneous in nature and have various risk factors, genetic alterations, and clinical features.1-3 HCAs are rare benign liver tumors that usually develop in women after oral contraceptive use.4 They are sometime difficult to discriminate from well-differentiated HCC or from focal nodular hyperplasia (FNH), a benign regenerative lesion. Recently, we defined a new molecular classification for HCA according to genetic [hepatocyte nuclear factor 1α (HNF1α) or β-catenin mutations] and phenotypic (inflammation) features.5, 6 We showed that β-catenin–activated HCAs were more at risk of malignant transformation than the other HCA subtypes.
HCC is the most common malignant tumor in the liver and the third cause of death from cancer. Viral [hepatitis B virus (HBV) and hepatitis C virus (HCV) infection] and nonviral (alcohol and aflatoxin B1 ingestion) agents are known to be associated with HCC occurrence. In each tumor, several oncogenic pathways are deregulated,2 and different subtypes of tumors are defined according to genetically and transcriptomically based classification.7, 8 Briefly, two main tumor groups related to chromosome stability/instability have been identified, and six subgroups (called G1 to G6) have been found to be closely related to genetic factors (TP53 and β-catenin mutations), activation of oncogenic pathways (AKT and insulin-like growth factor activation), clinical features (local invasion), and risk factors (HBV exposure).
Apart from genetic and epigenetic abnormalities modifying oncogene and tumor suppressor genes, deregulation of microRNAs (miRNAs) may also contribute to carcinogenesis. miRNAs are small, endogenous, noncoding RNAs (∼22 nucleotides) responsible for a posttranscriptional expression regulation of targeted genes. They promote messenger RNA (mRNA) degradation and repress mRNA translation by sequence-specific interaction with the 3′-untranslated region of targeted mRNAs.9 Physiologically, miRNAs have been shown to be involved in several processes such as development, apoptosis, proliferation, and differentiation.10, 11 Moreover, their levels of expression are deregulated in a large number of tumors, some of them were found to be directly implicated in carcinogenetic mechanisms, and several altered expressions of miRNAs have previously been described in rat and human HCC.12-14
We hypothesized that miRNA deregulation could be related to specific subgroups of hepatocellular tumors defined by either oncogene/tumor suppressor mutations or histological or clinical features. We first analyzed the expression of 250 miRNAs in a series of 46 malignant and benign hepatocellular tumors extensively characterized with respect to 4 normal liver tissues. According to tumor classification, the most significant deregulated miRNAs were selected, and their differential expression was validated in a second series of 43 tumors and 16 nontumor liver tissues including cirrhosis and chronic hepatitis of various etiologies.
A whole series of 109 liver samples (55 HCC, 29 HCA, 5 FNH, and 30 nontumor liver samples) was collected from 93 patients surgically treated in Bordeaux from 1992 to 2004. Liver tissues were immediately frozen in liquid nitrogen and stored at −80°C until they were used for molecular studies. All the patients were recruited in accordance with French law and institutional ethical guidelines. Liver samples were fully clinically, histologically, and genetically characterized and divided into two sets (Table 1). HCAs were classified according to the pathomolecular classification, which took into account the HNF1α or β-catenin mutations and the presence of inflammatory infiltrates.5, 6 HCCs were classified according to the clinical, pathological, and genetic features as previously described.7
Table 1. Clinical, Genetic, and Transcriptomic Features of the Studied Tumors
First Set of Samples (n = 50)
Validation Set of Samples (n = 59)
FNH indicates focal nodular hyperplasia; HBV, hepatitis B virus; HCA, hepatocellular adenoma; HCC, hepatocellular carcinoma; HCV, hepatitis C virus; and HNF1α, hepatocyte nuclear factor 1α.
Three patients were HBV-infected and HCV-infected.
Thirteen nontumor livers were paired with analyzed tumors.
Total RNA extraction was performed with Trizol reagent (Invitrogen). A 2-mm3 frozen tissue sample was homogenized in 800 μL of Trizol reagent with an MM 300 tissue lyser mixer mill (Quiagen). The extraction was performed according to the Invitrogen protocol. RNAs were quantified with NanoDrop ND-1000, and the quality was tested by the migration of 200 ng of RNA on 0.8% agarose gel.
miRNA quantitative RT-PCR was performed with TaqMan miRNA human assays and the ABI-Prism 7900HT system (Applied Biosystems). The first set of samples was analyzed with a set of 250 miRNAs (Applied Biosystems; Supplementary Table 1). The expression of 22 miRNAs was further analyzed in the validation set of tumors. For each sample and each miRNA tested, an independent retrotranscription using a specific reverse-transcription primer was performed on 5 ng of total RNAs according to the manufacturer's protocol. Ribosomal 18s (R18S) was used for the normalization of expression data. The relative amount of miRNAs and all mRNAs (target) studied in the samples was determined with the 2−ΔΔCT method, where ΔΔCT = (CTtarget − CTR18S)sample − (CTtarget − CTR18S)calibrator. Briefly, for the first set of samples, expression of an miRNA was normalized with R18S and with the mean expression of the corresponding miRNAs in the 4 normal liver samples (calibrator). For the validation set of samples, expression was normalized with R18S and with one normal liver sample (calibrator). As the same normal liver RNA was previously included in the first set of experiments, we verified the reproducibility of miRNA expression values obtained in the two experiments. Results of the validated miRNA are provided in Supplementary Table 2.
HNF1α/transcription factor 1, glutamine synthetase ligase (GLUL), leucine-rich repeat-containing G protein-coupled receptor 5 (LGR5), T-box 3 (TBX3), and pantothenate kinase 1 (PANK1) mRNA expression was measured as previously described15 with predesigned primers and probe sets from Applied Biosystems (probe set numbers: Hs00167041_m1, Hs00374213_m1, Hs00374213_m1, Hs00255591_m1, and Hs00379740_m1).
In the first set of 50 samples, quantitative RT-PCR expression results for the 250 tested miRNAs were filtered, and 130 miRNAs were selected on the basis of a Ct value below 31 in at least 10% of the tested samples (Supplementary Table 1). These 130 selected miRNAs were used for an unsupervised clustering analysis. Dendrograms classifying samples and genes with average linkage methods using the Pearson correlation as a distance metric were obtained on the basis of the expression profile of quantitative RT-PCR with dChip 2006 software (http://biosun1.harvard.edu/complab/dchip/). Clusters were considered significantly associated with a feature when the P value was less than 0.05 with Pearson correlation. miRNAs that were significantly differentially expressed according to clinical or genetic features identified in the unsupervised clustering analysis were selected for supervised analyses (Supplementary Table 3).
Statistical analysis was performed with dChip and GraphPad Prism software. Quantitative RT-PCR data were first analyzed for differences of expression between sample subgroups with the dChip t test and the nonparametric Mann-Whitney test. miRNAs with a fold change superior to 3 and statistically significant in the two tests were selected for validation (Supplementary Table 4). All miRNA expression data were presented as the mean ± standard error of the mean in subgroups of tumors. The nonparametric Spearman test was used to compare quantitative values of expression. All reported P values were two-tailed, and differences were considered significant when the P value was under 0.05.
Cell Culture and Transfections.
HepG2 cells were cultured in Dulbecco's modified Eagle medium high glucose (Invitrogen) supplemented with 10% fetal bovine serum and penicillin/streptomycin. Cells were transfected with Lipofectamine RNAiMax (Invitrogen) in six-well plates according to the manufacturer's protocol. HNF1α extinction was performed with a small interfering RNA (siRNA) targeting exons 8 and 9 of HNF1α (siRNA 3544; sense: GGUCUUCACCUCAGACACUtt, antisense: AGUGUCUGAGGUGAAGACCtg; Ambion). Several siRNA concentrations were tested (0, 0.01, 0.05, 0.1, 0.2, 0.4, 0.6, 0.8, 1, 5, 10, and 50 nM) in order to modulate the HNF1α expression. Maximal inhibition of HNF1α protein was observed 72 hours post-transfection (see Supplementary Fig. 1). BLOCK-iT Alexa Fluor Red Fluorescent Oligo (Invitrogen) was used as a transfected double-stranded RNA control. Western blotting was performed with HNF1α antibody sc-6548 (Santa Cruz Biotechnology). At this time, cells were lysed in Trizol reagent (Invitrogen) for RNA extraction.
Global Profiling of miRNAs Is Related to Well-Defined Subgroups of Hepatocellular Tumors.
We analyzed the expression level of 250 miRNAs in the first set of samples, which included 28 HCCs, 13 HCAs, and 5 FNHs, compared to 4 normal liver tissues. An unsupervised analysis was performed after selection of the 130 miRNAs expressed in at least 10% of the samples (Fig. 1A). It showed a significant partition between nontumor and tumor samples (P < 0.01); benign tumors were divided into three clusters significantly parted from HCC (P < 0.01). Adenomas were grouped into two distinct clusters corresponding to HNF1α-mutated (P < 0.01) and inflammatory HCA (P < 0.01). FNHs were located on a third cluster (P < 0.01) distinct from HCA. In contrast, β-catenin–mutated HCA was spread between two HCA clusters. HCC samples were partitioned into three different clusters significantly characterized either by a specific clinical risk factor (HBV, P < 0.001; alcohol intake, P < 0.05) or by a gene mutation (β-catenin, P < 0.01). Two supervised analyses were performed to classify separately benign and malignant tumors with the list of genes differentially expressed in each clinical or genetic feature significantly associated with clusters in the unsupervised analysis (Fig. 1B,C and Supplementary Table 3).
Specific miRNA Deregulations Related to Benign or Malignant Features.
As shown previously, miRNA profiling could classify hepatocellular tumors according to pathological, clinical, and genetic features. Thus, we aimed to identify and validate, in a second series of tumors, miRNA with a deregulated expression most closely associated with such specific features. First, we identified miRNAs specifically deregulated in tumors when compared with nontumor liver samples, and several were validated (Fig. 2). Three miRNAs were found to be significantly up-regulated (miR-224) or down-regulated (miR-422b and miR-122a) in both benign and malignant tumors (Fig. 2A). Second, we validated miRNAs with a level of expression significantly different between malignant and benign tumors. Corresponding to these criteria, miR-200c and miR-203 were underexpressed in benign tumors (Fig. 2B), whereas miR-21, miR-222, and miR-10b were significantly overexpressed in HCC (Fig. 2C). Finally, miR-224, which was differentially expressed between tumor and nontumor liver tissues, was also significantly overexpressed in HCC when compared to benign tumors (Fig. 2A).
miRNA Deregulation Related to Tumor Suppressor/Oncogene Mutations in HCA.
We searched for miRNA deregulation that could characterize specific subgroups of benign tumors. Although unsupervised clustering enabled us to correctly classify inflammatory HCA and FNH into two specific clusters, no miRNA displayed a sufficient fold change of expression to grant further validation. In contrast, a robust underexpression of miR-375 was validated in β-catenin–mutated HCA when compared to the other HCA (fold change: −14.2; P < 0.001 in overall samples; Fig. 3A). Among 15 miRNAs significantly deregulated in HNF1α-mutated HCA, we validated the underexpression of miR-107 in HNF1α-mutated HCAs when compared to the other HCA (fold change: −3.6; P < 0.001 in overall samples; Fig. 3B). This observation suggested that the HNF1α transcription factor could regulate, directly or not, the expression of miR-107. To test this hypothesis, we performed an HNF1α silencing using several doses of specific siRNA targeting HNF1α in the HepG2 hepatocellular cell line, and we assayed the consequences on the miR-107 expression level. The silencing of HNF1α significantly correlated with the extinction of miR-107 in a dose-effect manner (Spearman r: −0.92; P < 0.0001; Fig. 3C).This result suggested that HNF1α may control the expression of miR-107. miR-107 is located in intron 5 of the PANK1 gene coding for PANK1. We tested if the miR-107 and PANK1 expression could be coregulated. We showed that PANK1 mRNA expression was not affected by HNF1α silencing, suggesting that regulation of miR-107 by HNF1α was independent of PANK1 status (Supplementary Fig. 2). According to the hypothesis of direct regulation of miR-107 by HNF1α, the in silico search (using Genomatix software) revealed two potential HNF1α binding sites located −1852 and −1543 base pairs upstream of the pre-miRNA initiation site. However, the direct role of HNF1α remained to be fully demonstrated by a functional analysis of the putative promoter region.
miRNA Deregulation Related to Genetic Features and Risk Factors in HCC.
As shown in Fig. 1, miRNA deregulations were associated with alcohol consumption and HBV infection in HCC. We validated miR-126* underexpression in HCC related to alcohol abuse when compared to the other HCC (fold change: 2.7; P < 0.001 in overall samples; Fig. 4A). Similarly, we also validated a significant overexpression of miR-96 in HBV-related HCC when compared to noninfected tumors (fold change: 5.7; P < 0.01 in overall samples; Fig. 4B). In contrast, the level of expression of miR-126* and miR-96 did not vary significantly among the various subtypes of nontumor liver tissues (Supplementary Fig. 3). In particular, no significant variation was observed in cirrhosis or HBV-related nontumor samples. As suggested by the unsupervised clustering, no miRNAs were found to be significantly deregulated in HCV-infected HCCs versus noninfected HCCs. Similarly, no significant difference was found according to the level of cell differentiation.
miRNA deregulations were also related to the β-catenin oncogene mutation in the unsupervised clustering analysis (Fig. 1). As in β-catenin–mutated HCAs, we validated the extinction of miR-375 in HCC characterized by a β-catenin–activating mutation (fold change: −44; P < 0.001 in overall samples; Fig. 4C). We also found a significant anticorrelation between the level of expression of miR-375 and three genes regulated by β-catenin, TBX3, LGR5, and GLUL, in HCA and HCC (Fig. 4D). Finally, no significant differences in miRNA expression were observed in nontumor liver tissues according to HCV or HBV infection, alcohol intake, and cirrhosis.
Aberrant patterns of miRNA expression have already been described in several hematological and solid tumors, including HCC (for a review, see Calin and Croce16). In the present work, we have identified for the first time miRNA expression patterns that can unambiguously differentiate benign hepatocellular tumors, malignant hepatocellular tumors, and several subtypes of hepatocellular tumors according to specific risk factors and oncogene and tumor suppressor gene mutations. These new findings and the most significant deregulated miRNAs are summarized in Fig. 5.
In the present work, several miRNAs previously associated with malignancy were identified as significantly deregulated in a hepatocellular subgroup of tumors. Most notably, we showed that miR-224, previously identified as overexpressed in HCC,14 is also up-regulated in benign hepatocellular tumors, including adenomas and FNHs, which are regenerative polyclonal nodules,17 but to a lesser degree. Similarly, miR-122a has been previously found to be down-regulated in human and rodent HCC12; this miRNA, highly expressed in the liver, is essential for HCV replication,18 and it modulates cyclin G1 expression in the HCC cell line.19 Here, we showed that miR-122a is also down-regulated in both human benign and malignant hepatocellular tumors. Thus, concomitant analysis of benign and malignant tumors enabled us to associate deregulation of expression of miR-224, miR-422b, and miR-122a with common early events in hepatocellular tumorigenesis.
Only a few miRNA deregulations have been previously described in benign tumors.20, 21 Interestingly, in pituitary adenoma, the miRNA pattern of expression was significantly related to the tumor subtype according to histotypes and secreting characteristics.20 In thyroid nodules, most of the miRNAs deregulated in carcinomas have also demonstrated an altered expression in adenomas.21 Here, we identified miRNA (miR-200c and miR-203) with an expression specifically down-deregulated in benign hepatocellular tumors when compared to normal liver tissues or HCC. A lower expression of miR-203 was also previously described in nonsecreting pituitary adenomas when compared with growth hormone pituitary secreting adenomas.20 In hepatocellular tumors, we showed that the miRNA profile of expression in a benign tumor differed according to the tumor subtype defined by oncogene and tumor suppressor gene mutation or by histological phenotype. These results highlighted the importance of refining classification based on meticulous molecular characterization of the tumor phenotypes, including benign tumors.
Here we validated a significant overexpression of miR-10b, miR-21, and miR-222 in HCC when compared with benign tumors and nontumor liver tissues. Interestingly, overexpression of these three miRNAs was previously identified as an important key factor in promoting cell invasion or proliferation in various epithelial tumor types. Recently, Ma and collaborators22 identified miR-10b as an miRNA highly expressed in metastatic breast tumors that promotes cell migration and invasion. They showed that miR-10b expression is induced by TWIST allowing miR-10b to inhibit translation of the mRNA encoding homeobox D10, resulting in increased expression of a well-characterized prometastatic gene, ras homolog gene family member C (RHOC). miR-222 overexpression promotes cell proliferation by targeting p27Kip1 in prostate and thyroid carcinoma.23, 24 Finally, miR-21 overexpression down-regulates the Pdcd4 tumor suppressor and stimulates invasion, intravasation, and metastasis in colorectal cancer.25 Higher overexpression of miR-21 was also previously associated with poorly differentiated HCC, and it participates in down-regulating the level of expression of phosphatase and tensin homolog (PTEN).13
Our results identified the deregulation of two miRNAs, miR-96 and miR-126*, as significantly associated with different risk factors of HCC, HBV infection, and alcohol consumption. To our knowledge, miR-96 and miR-126* have not been directly implicated in pathways related to carcinogenesis. Because no significant variation in the expression of these two miRNAs was observed in nontumor tissues according to HBV infection or alcohol intake, it suggests that miRNA deregulations are directly linked to carcinogenetic processes specifically induced by these two risk factors. In fact, using transcriptomic and genetic global analyses, we have previously shown that carcinogenesis mechanisms in HBV-related HCC are particularly closely related to chromosome instability, AKT activation, and fetal features in gene expression.7, 8 In contrast, miRNA profiling is the first global genomic approach that enables us to discriminate HCC related to alcohol abuse from the other tumors. Consequences of the miR-126* deregulation remained to be explored in alcohol-related tumors and HCC associated with other risk factors.
For the first time, we identified miRNA deregulation in hepatocellular human tumors closely related to specific oncogene or tumor suppressor gene mutations. However, in cell line models, some miRNAs have been shown to be under control of an oncogene26 or tumor suppressor.27 The low expression of miR-375 in both HCA and HCC mutated for β-catenin together with the highly significant anticorrelation between the miR-375 level of expression and genes targeted by β-catenin argues for a direct link between β-catenin activation and repression of miR-375. This is an important result because β-catenin is a major oncogene activated in 30% to 40% of HCCs. It increases the complexity of regulation of expression by β-catenin, adding yet another layer of control with miRNA to the RNA and translation control level. In HCA, we showed a specific underexpression of miR-107 in HNF1α-inactivated samples. It has been suggested by in silico analysis that miR-107 could regulate the expression level of several predicted target proteins involved in lipid metabolism. Underexpression of miR-107 is predicted to promote acetyl–coenzyme A and lipid levels.28 Interestingly, we previously showed that HNF1α-inactivated HCA exhibited a steatotic phenotype resulting from an activation of lipogenesis.29 Thus, we can hypothesize that miR-107 underexpression specifically observed in HNF1α-inactivated HCA may contribute to the steatotic phenotype and lipid accumulation. Our data suggest that this miRNA could be regulated by HNF1α like the recently identified miR-194 during intestinal epithelial Caco-2 cell line differentiation.30
In conclusion, this work revealed an additional layer of gene alterations to better characterize hepatocellular tumors. It allowed the identification of miRNA deregulation closely correlated to clinical and histological subtypes of tumors and to oncogene and tumor suppressor gene mutations. The gene and protein networks directly targeted and affected by these miRNAs that participate in tumorigenesis still need to be extensively explored.
We thank Philippe Bois for his critical reading of this manuscript. We also thank Aurélien de Reynies for his statistical help.