To identify TS-miRNAs, we first performed an integrative approach using function- and expression-based screenings in six HCC cell lines (Hep G2, Hep 3B, HLE, Huh7, JHH-4, and sK-Hep-1). We focused on miRNAs, which showed remarkable inhibition of cell proliferation in vitro together with significant downregulation in HCC cell lines compared with normal liver tissues (Fig. 1A), as candidate TS-miRNAs for HCC.

In the function-based screening in six HCC cell lines using a synthetic miRNA mimic library containing 470 pre-miRNAs, 113 miRNAs demonstrated remarkable inhibitory effects on cell growth in more than 3 of 6 cell lines (relative growth ratio <0.8 compared with control non-specific miRNA; Fig. 1B, and Dataset S1).

In the expression-based screening in the same six cell lines and normal liver tissues (C20, C40) using a miRNA microarray containing 866 human miRNAs, 265 miRNAs were able to be quantified at least in one cell line. Among them, 194, 194, 77, 190, 199, and 168 miRNAs were downregulated (>2-fold decrease) in the Hep G2, Hep 3B, HLE, Huh7, JHH-4, and sK-Hep-1 cell lines, respectively, compared with C20 and C40, and 45 miRNAs were commonly downregulated in all these six cell lines (>2-fold decrease; Fig. 1C, and Dataset S2).

By combining results of the two screenings, we identified seven miRNAs, i.e. miR-101, -126, -15a, -192, -195, -378 and -497, as candidate TS-miRNAs for HCC (Fig. 1A and 1D). Among them, miR-101 and miR-195 had already been reported as possible TS-miRNAs for HCC [13], [14], suggesting that the approach employed in this study could successfully identify TS-miRNAs in HCC cells.

In order to narrow down those seven candidates by frequency of their downregulation in HCC cells, we next performed expression analyses in a panel of 19 HCC cell lines (Fig. S1A) and a panel of paired tumorous and non-tumorous tissues from 18 primary HCC cases (Fig. S1B). miR-15a was excluded due to a low frequency of downregulation in a panel of HCC cell lines (42.1%) and primary cases (16.7%, Fig. 1D). miR-126 and miR-192 were also excluded due to a low frequency of tumor-specific down-regulation in a panel of primary HCC cases (27.8% and 44.4%, respectively; Fig. 1D). Finally, four miRNAs, miR-101, miR-195, miR-378, and miR-497, were selected as the most promising candidates for TS-miRNA for HCC showing frequent (>50%) tumor-specific downregulation both in HCC cell lines and primary HCC cases (Fig. 1D). Among them, miR-101 was excluded from further analysis because of its well-known function as a TS-miRNA in HCC [13]. Notably, miR-378 had been reported as an oncomiR in various types of cancers [15], [16], while showing TS function in our screening performed in HCC. In addition, miR-378 was also reported to be downregulated during liver regeneration after partial hepatectomy in mice with a delay in cell cycle progression involving the G1 to S phase transition [17], suggesting miR-378 to have cell growth suppressive activity in liver and supporting our findings in HCCs. These inconsistent findings indicate that miR-378 may have multiple functions in a tissue- or lineage-specific manner. Since we focused on miRNAs showing direct inhibition of HCC cell growth in this screening, we excluded miR-378 from further analysis. Among the 4 candidates, miR-497 and miR-195 belong to the same miRNA family and are also located very close together: pre-miR-195 is 209 bp downstream from the 3′ end of pre-miR-497 at 17p13.1, and is frequently deleted in human cancers [18], [19]. We focused on these two miRNAs for further analysis, because (a) reported evidence and evolutional conservation suggest that the clustered miRNAs may have related physiological and pathophysiological functions including carcinogenesis [20]–[22], and (b) miR-497 is poorly characterized in human cancer, especially in HCC, indeed, few target molecules for those two miRNAs are known, although several reports have suggested miR-195 to be a possible TS-miRNA for various human cancers including HCC [14], [23]–[25].

Computational analysis using a data set of miRNA expression profiles in 89 primary HCV-related HCCs obtained from the database of the Broad Institute (GSE20596) revealed a significantly positive correlation between expression levels of mature forms of miR-195 and miR-497 (Fig. S2A, left panel). In addition, the expression status of miR-195 and miR-497 seemed to be correlated regardless of their viral infection status not only in tumor samples but also in non-tumorous liver tissues, when we recalculated the data used in Fig. S1B (Fig. S2A right panel), although the number of samples are too small for statistical analysis. These results suggesting those miRNAs to be transcribed concomitantly as a single primary transcript and expressed as a cluster not only in HCCs but also in non-tumorous liver tissues. Indeed, this hypothesis is supported by the existence of the transcript NR_038310.1, mir-497-195 cluster host gene (non-protein coding) (MIR497HG), and non-coding RNA in the NCBI reference sequence (http://www.ncbi.nlm.nih.gov/nuccore/NR_038310?report=GenBank).

We first examined copy-number aberrations around miR-195 and miR-497 in 18 HCC cell lines by q-gPCR, and found that only 3 of them (HLE, JHH-6, PLC/PRF/5) have a copy number loss at this locus in spite of frequent downregulation of those miRNAs in a number of HCC cell lines (Fig. S2B). Since the biogenesis of miRNAs has been reported to be altered in cancer cells [26], we tried to test the possibility that the down-regulation of those miRNAs were caused by abnormal miRNA biogenesis. Since downregulation of miR-195 and miR-497 was also observed in pri-miRNA level, those miRNAs seem to be downregulated at a transcriptional level rather than a post-transcriptional level (Fig. S2C). To assess other factors contributing to the downregulation of miR-195 and miR-497, we next searched for a regulatory element for the putative polycistronic primary transcript, pri-miR-497-195. The putative transcriptional start site (TSS) was detected 1.6 kb upstream of pre-miR-497 by SwitchGear Genomics Transcription Start Sites (CHR17_M0118_R1) and ENCODE Transcription Factor ChIP-seq (TAF1) binding site (UCSC Genome Bioinfomatics, Feb. 2009 [GRCh37/hg19] assembly; http://genome.ucsc.edu/cgi-bin/hgGateway) (Fig. S3A). Promoter activities around the TSS were confirmed in three HCC cell lines, Hep G2, sK-Hep-1 and HuH-6, regardless of the expression status of miR-195 and miR-497 (Fig. S3B). Although Suzuki et al. [27] reported that miR-497 and miR-195 are possibly regulated by non-CpG island methylation in colorectal cancer based on data from a genome-wide combined analysis of mRNA expression, chromatin signature, and DNA methylation, only minimal restoration was observed in the expression of those miRNAs after treatment with 5-aza-dCyd in HCC cell lines (Fig. S3C) and almost no difference in DNA methylation pattern around the TSS of pri-miR-497-195 was observed among HCC lines regardless of their expression status (Fig. S3D). Moreover, the inhibition of the class I and II histone deacetylase (HDAC) families with Tricostatin A (TSA) did not induce restoration of miR-195 nor miR-497 in most of cell lines, suggesting that these HDACs do not strongly contribute to the decreased expression of those miRNAs in HCC cells (Fig. S3C). In addition, ChIP analyses around TSS for H3K4me3 and H3K27me3, which are histone modification patterns activating and inactivating gene expression, revealed that the promoter region of pri-miR-497-195 showed no significant repressive pattern in the H3K27me3 modification, but a clear inactive pattern in the H3K4me3 modification in non-expressing HCC cells compared with miR-195-expressing HuH-6 cells (Fig. S3E). Taken together, frequent downregulation of miR-497 and miR-195 expression occurred not mainly by genomic loss, DNA hypermethylation or altered miRNA biogenesis, but at least partly through repressive histone modifications in the HCC cell lines examined, although HDAC inhibitor had little or no effect on the expression and the epigenetic factors contributing to the downregulation remain to be clarified.

Growth suppressive activities of double stranded RNAs (dsRNAs) mimicking miR-195 and miR-497 in HCC cell lines (Fig. 1B and Dataset S1) were validated by transfection of synthetic dsRNAs mimicking miR-195 and miR-497 and control non-specific miRNA purchased from a different company (Fig. 2A), excluding the off-target effects of synthetic dsRNAs. A similar growth suppression pattern for miR-195 and miR-497 was observed in 3 of 6 cell lines (Hep 3B, Hep G2 and JHH-4) by ectopic overexpression of those miRNAs, whereas comparatively weak growth suppression was observed after restoration of miR-497 compared with miR-195 in the remaining 3 cell lines (HLE, Huh7 and sK-Hep-1).

We next determined modes of action on the cell cycle by FACS analysis in several HCC cell lines (Fig. 2B). The accumulation in G0/G1 phase was significant in all cell lines transfected with miR-195, while the effect was slightly weak in HLE and sK-Hep-1 cells transfected with miR-497 compared with miR-195. These findings were consistent with results of the cell growth assays in vitro (Fig. 2A). Although a small increase in cells in the sub-G1 fraction was observed in Hep 3B, no remarkable change in the number of cells was observed 48 hours after transfection (Fig. 2A), indicating G1 arrest to be predominant for the cell growth-suppressive effect of miR-195 and miR-497 in HCC cells. Taken together, miR-195 and miR-497 are likely to inhibit cell growth mainly by suppressing cell cycle progression, especially in the G1 to S phase.

For exploration of signaling pathways downstream of miR-195 and miR-497, we performed a genome-wide mRNA expression analysis of Hep G2 and sK-Hep-1 cells 48 hours after transfection with miR-195, miR-497, or control dsRNA. Among significantly altered GO terms (correlated p-value<0.05) in biological processes and functions selected from genes showing 2-fold changes between miR-195 or miR-497 transfectants and their counterparts with a false discovery value (q) <0.001 (Table S1–1 to S–4), GO terms involved in the cell cycle most frequently appeared in all conditions. This result is consistent with the inhibitory effect of miR-195 and miR-497 on cell cycle (Fig. 2B), suggesting molecules contributing to cell cycle regulation to be major targets for miR-195 and miR-497.

To determine direct targets for miR-195 and/or miR-497 responsible for hepatocaricinogenesis, we performed RNA coimmunoprecipitation with anti-Ago2 antibody (Ago2-IP) for trapping overexpressed miRNA incorporated into the RNA-induced silencing complex (RISC) together with their target mRNAs in Hep G2 cells transfected with miR-195 or miR-497. Profiling of miRNAs obtained from the Ago2-IP fraction by miRNA microarray analysis showed a significantly increased rate of each transfected miRNA compared with most other endogenous miRNAs in Hep G2 cells (Fig. S4), indicating transfected miRNAs to be incorporated successfully and specifically into RISC. However, miRNAs other than miR-195 or -497 also showed increased rates, suggesting that mRNAs trapped by RISC include targets for non-specific miRNAs as false-positives. In this condition, we performed RNA sequencing using Ago2-IP RNAs (Ago2-IP-seq) along with total RNAs (mRNA-seq) from the same samples in miR-195-, miR-497-, or non-transfected Hep G2 cells, and calculated the abundance of each mRNA in the Ago2 binding fractions compared with total RNA fractions (Ago2-IP-seq/mRNA-seq of each RPKM value) in each sample (abundant-Ago2 fraction). Finally, we obtained the fold enrichment score by calculating the ratio of abundant-Ago2 fractions between miR-195- or miR-497-transfected samples and non-transfected samples.

In GSEA, genes selected as top 10% candidates by Ago2-IP in miR-195 transfected cells were significantly enriched in genes downregulated by miR-195 transfection, whereas those in miR-497-transfected cells were not significant (p<0.001 and = 0.0709607, FDR<0.001 and = 0.041126948, and normalized enrichment score = -1.7712895 and 1.1393404 for miR-195 and miR-497, respectively; Fig. S4B). Through excluding genes without a target sequence of miR-195 and miR-497 in their 3′UTR, enrichment scores in GSEA were dramatically improved (p<0.001 and <0.001, FDR<0.001 and <0.001, and normalized enrichment score = -2.5568738 and -1.7213858 for miR-195 and miR-497, respectively; Fig. S4C).

Since the total numbers of true targets for miR-195 and miR-497 in Hep G2 cells are unclear, genes enriched in Ago2IP-deep sequencing with the target sequence were cut off by the most efficient rates with maximum enrichment scores in GSEA. The highest enrichment scores were detected when genes were cut off by top 8% (577 genes) or 36% (2,010 genes) for candidates of miR-195 or miR-497 targets, respectively (p<0.001 and <0.001, FDR<0.001 and <0.001, and normalized enrichment score = -2.5925288 and -1.9934999 for miR-195 and miR-497, respectively; Fig. 3A and Fig. S4D). The top 20 candidate targets enriched in the Ago2-IP fraction of miR-195 or miR-497 combined with target predictions are listed in Table S2.

In order to narrow down promising targets of miR-195 and miR-497, we focused on 190 of 577 and 290 of 2,010 genes remarkably downregulated (>2-fold change) by miR-195 and miR-497 transfection in Hep G2 cells (Fig. 3B), respectively, and carried out a pathway analysis. All of those genes were mapped to genetic networks describing functional relationships among gene products based on known interactions as defined by the IPA tool. Consistent with our findings showing that miR-195 and miR-497 induced G1 arrest in HCC cells, the IPA tool identified the canonical pathway “Cell Cycle: G1/S Checkpoint Regulation” as a significantly enriched pathway for possible target genes (p<0.001 and p = 0.00287, respectively; Fig. 3C). Therefore, we focused on candidate genes involved in this pathway, such as BTRC, CCND3, CCNE1, CDC25A, CDK4 and CDK6 as a target for miR-195 and miR-497, as functionally important direct targets in Hep G2 cells.

Since several cell cycle regulators, such as CCND1, CDK6 and E2F3, have been reported as direct targets of miR-195 in HCC [14], we included them as positive controls in our experiments to validate predicted targets. We first assessed the protein expression of predicted targets 48 hours after transfection with dsRNA mimicking miR-195 or miR-497 into Hep G2 and sK-Hep-1 cells (Fig. 3D). A reduction in CDK6 and E2F3 proteins was observed on transfection of each of those miRNAs in both cell lines, whereas no reduction in the CCND1 protein level was observed in Hep G2 cells. Protein levels of CCNE1, BTRC, CDC25A, CCND3 and CDK4 were reduced in both miR-195 and miR-497 transfectants compared with their control counterparts. To determine whether miR-195 and/or miR-497 directly inhibit the expression of those targets, we performed 3′UTR reporter assays using reporter constructs for each target mRNA containing putative binding sites for these miRNAs (Fig. 3E). Significant reductions in luciferase activity were observed in cells cotransfected with each reporter construct for all 8 genes in miR-195 or miR-497 transfectants compared with mock transfectants (Fig. 3E). Taken together, CCNE1, CDC25A, CCND3, CDK4 and BTRC seem to be novel direct targets in addition to three known targets for miR-195 and miR-497 in HCC cells (Fig. 4A).

We confirm the effect of novel target genes on cell growth by knocking down gene expression using specific small interfering RNAs (siRNA). The knock down efficiency of each gene was determined by Western blotting (Fig. S5). Decreased cell growth with G1 arrest was observed by downregulation of CCNE1, CDC25A, CCND3, CDK4 or BTRC in Hep G2 cells (Fig. 4B), suggesting these genes to be functional targets (Fig. 4A). To assess the consequence of miR-195/miR-497 and their target genes in HCC cases, we first analyzed the expression of those 8 targets genes in paired primary tissues of HCC used in Fig. 1D by qRT-PCR (Fig. 4B). Among them, only CCNE1, CDC25A and CDK4 showed >2-fold up regulation in over 50% of cases in tumorous compared with non-tumorous tissues. In addition, we also analyzed the correlation between each target gene and miR-195 or miR-497 expression using the data set of miRNA expression profiles in 89 primary HCV-related HCCs obtained from the database of the Broad Institute (GSE20596). Surprisingly, only the top-ranked 4 genes, CDK6, CCNE1, CDC25A and CDK4, frequently overexpressed in HCCs showed a slight significant inverse correlation with miR-195 and miR-497 expression (p<0.05, correlation coefficient = -0.257, -0.359, -0.315, and -0.355 for miR-195 and -0.309, -0.299, -0.245, and -0.292 for miR-497, respectively; Table S3). In conclusion, CCNE1, CDC25A, and CDK4 together with CDK6 seem to be the most important factors among 8 target genes because of their frequent tumor-specific up-regulation in HCC together with significant inverse correlation with miR-195 and miR-497 expression in primary HCCs relevant to cell cycle progression regulated by miR-195 and miR-497 in HCCs.

A total of 19 HCC cell lines were maintained in appropriate media as described elsewhere [32]. To examine the restoration of mRNA expression in genes of interest, the cell lines were cultured with or without 5 µM 5-aza-2′-deoxycytidine (5-aza-dCyd) for 5 days and/or 300 nM of TSA for the last 24 hours.

A total of 18 frozen primary tumor samples and corresponding non-tumorous tissue samples were obtained from HCC patients, and two frozen normal liver tissues samples (C20, C40) were obtained from patients with hepatectomy due to metastatic liver tumor treated at Tokyo Medical and Dental University with written consent from each patient and approval by the local ethics committees of Medical Research Institute and Faculty of Medicine, TMDU. The TNM classification of the Union for International Cancer Control (UICC) was used.

HCC cells were seeded in 24-well plates a day before transfection. Each synthetic miRNA (5 nM) in Pre-miR™ miRNA Precursor Library-Human V3 (Ambion, Austin, TX) or the control non-specific miRNA (Ambion) was transfected into cells using Lipofectamine™ RNAiMAX (Invitrogen, Carlsbad, CA) according to the manufacturer’s instructions. The numbers of viable cells were assessed 4 or 5 days after transfection by the colorimetric water-soluble tetrazolium salt (WST) assay. Results were normalized to the number of cells transfected with non-specific miRNA. Each assay was performed in duplicate.

Extracted total RNA was subjected to analysis using G4471A 15K Human miRNA Microarrays Ver. 3 and G4112F 44K Whole Human Genome Microarray for miRNA and mRNA expression, respectively, with standard image acquisition according to the manufacturer’s instructions (Agilent Technologies, Santa Clara, CA). Raw microarray data were normalized and analyzed using GeneSpring GX software version 11.5.1 (Agilent Technologies). All samples were analyzed twice.

Real-time qRT-PCR was performed using an ABI Prism 7500 Fast Real-time PCR System (Applied Biosystems, Foster City, CA), KAPA PROBE FAST Universal 2×qPCR master mix (Kapa Biosystems, Woburn, MA), TaqMan® Reverse Transcription Kit (Applied Biosystems), and TaqMan® MicroRNA Assays (Applied Biosystems) according to the manufacturer’s instructions [33]. All reactions were run in duplicate.

Each of miCENTURY OX miNatural mimicking miR-195, miR-497, or “microRNA control, Luciferase (GL3)” (5 nM; COSMO BIO, Tokyo, Japan) was efficiently transfected into cells as described elsewhere [33]. Each of gene-specific siRNA (Hs_CCNE1_0049, Hs_CDC25A_7714, Hs_CCND3_0184, Hs_CDK4_2490, and Hs_BTRC_9438; 10 nM; Sigma, St Louis, MO) or negative control siRNA (Mission_Negative control SIC-001, Sigma) was also efficiently transfected into cells. The numbers of viable cells 24–96 hours after transfection were assessed by WST assay. The cell cycle was evaluated 48 hours after transfection by fluorescence-activated cell sorting (FACS) as described [33].

Ago2-IP was performed using 1×10⁷ cells 24 hours after transfection of miCENTURY OX miNatural mimicking miR-195 or miR-497, or non-transfected cells, using Ago2 antibody and a microRNA Isolation Kit, human Ago2 (Wako Pure Chemical, Osaka, Japan) according to the manufacturer’s instructions. Samples for RNA-sequencing were prepared using 10 ng of Ago2-IP RNA and 3 µg of total RNA with a TrueSeq mRNA Sample Preparation Kit (illumina, San Diego, CA) with or without poly-A selection. Each of the libraries were analyzed by Genome Analyzer IIx (illumina) according to the manufacturer’s instructions. The 36-base single-read sequences were mapped to human genomic sequences (hg19) using the sequence alignment program ELANDv2. For the expression analysis, sequence reads were converted to RPKM (Reads Per Kilobase per Million mapped reads) values by using CASAVA 1.7 software (illumina).

Predicted targets of miR-195 and miR-497 and their sites were analyzed by searching for a possible targeting sequence (5′-ACGACGA-3′) in the 3′UTR of mRNA using the Ensembl Genome Browser (http://asia.ensembl.org/Homo_sapiens/) and Ensembl Perl API as well as by searching databases, such as MicroCosm Targets Version 5 (http://www.ebi.ac.uk/enright-srv/microcosm/htdocs/targets/v5/), microRNA.org (http://www.microrna.org/), and TargetScan (http://www.targetscan.org/).

Cut-off values of Ago2-IP-seq were calculated for genes with miRNA target predictions using Gene Set Enrichment Analysis (GSEA; www.broad.mit.edu/gsea). Candidate target genes were selected as an intersect of Ago2-IP-seq final candidates and downregulated genes by miRNA transfection, followed by pathway analyses using Ingenuity Pathway Analysis (IPA; Ingenuity Systems, Inc., Redwood City, CA). Detailed information for each analysis is provided in Methods S1.

Western blotting and 3′UTR reporter assays were performed as described elsewhere [33]. In the 3′UTR reporter assay, “microRNA control, Non-target RNA” (COSMO BIO) was used as a negative control. PCR primers designed to amplify regions of interests, and antibodies used for Western blotting are provided in Table S4 and Methods S1.

The microarray and sequencing data from this publication have been submitted to the GEO database (http://www.ncbi.nlm.nih.gov/geo/) and assigned the identifier “GSE41081”.