Cytokine-mediated modulation of the hepatic miRNome: miR-146b-5p is an IL-6-inducible miRNA with multiple targets

Interleukin-6 (IL-6)-type cytokines playimportant rolesin liver (patho-)biology. For instance, they regulate the acute phase response to inflammatory signals and are involved in hepatocarcino-genesis. Much is known about the regulation of protein-coding genes by cytokines whereas their effects on the miRNome is less well understood. We performed a microarray screen to identify microRNAs (miRNAs) in human hepatocytes which are modulated by IL-6-type cytokines. Using samples of 2 donors, 27 and 68 miRNAs (out of 1,733) were found to be differentially expressed upon stimulation with hyper-IL-6 (HIL-6) for up to 72 h, with an overlap of 15 commonly regulated miRNAs. qPCR validation revealed that miR-146b-5p was also consistently up-regulated in hepatocytes derived from 2 other donors. Interestingly, miR-146b-5p (but not miR-146a-5p) was induced by IL-6-type cytokines (HIL-6 and OSM) in non-transformed liver-derived PH5CH8 and THLE2 cells and in Huh-7 hepatoma cells, but not in HepG2 or Hep3B hepatoma cells. We did not find evidence for a differential regulation of miR-146b-5p expression by promoter methylation, also when analyzing the TCGA data set on liver cancer samples. Inducible overexpression of miR-146b-5pinPH5CH8cellsfollowedbyRNA-SeqanalysisrevealedeffectsonmultiplemRNAs, including those encoding IRAK1 and TRAF6 crucial for Toll-like receptor signaling. Indeed, LPS-mediatedsignalingwasattenuateduponoverexpressionofmiR-146b-5p,suggestingaregulatory loop to modulate inflammatory signaling in hepatocytes. Further validation experiments suggest DNAJC6, MAGEE1, MPHOSPH6, PPP2R1B, SLC10A3, SNRNP27, and TIMM17B to be novel targets for miR-146b-5p (and miR-146a-5p).

Its action is mainly mediated by the Janus kinase (JAK)/signal transducer and activator of transcription 3 (STAT3) signaling pathway. Upon phosphorylation, STAT3 homodimers are formed and translocate to the nucleus where they modulate the expression of many genes including several cytokines (e.g., EGF, HGF, and VEGF), apoptosis inhibitors such as Bcl-2, Bcl-xL, survivin, Mcl-1, and XIAP, 11,12 cell-cycle-related genes such as cyclin D1, 13 metastasis-related genes matrix metalloproteinase 1 (MMP1), MMP2, 14 and MUC1, 15 and acute phase proteins (APP) specifically in the liver. 16 Both IL-6 and STAT3 have been shown to promote HCC development in mouse models. While 100% of male mice develop liver tumors upon N,N-diethylnitrosamine (DEN) treatment, only 20% of IL-6 −/− male mice treated with DEN develop HCC. 17 Similarly, a liverspecific knock-out of STAT3 in mice results in a significant decrease in HCC prevalence and smaller tumor size. 18 In addition, deletion of STAT3 in cultured DEN-induced mouse hepatocytes leads to cell death whereas cells with only partial reduction are viable but with a senescent phenotype and without capacities to form subcutaneous tumors, showing that activated STAT3 is required for survival of HCC cells. 18 Finally, aberrant activation of this pathway, e.g. due to gain-offunction mutations affecting STAT3 or gp130 (as observed in inflammatory hepatocellular adenomas), 19,20 or the constitutive release of cytokines which activate STAT3 (mainly IL-6), 21 has also been shown to occur in cancer cell lines and solid tumors (reviewed in Guo et al. 5 Subramaniam et al. 22 ).
MicroRNAs (miRNAs) are small non-coding RNAs characterized by their length, between 20 and 30 nucleotides, and their association with Argonaute (AGO) proteins. They function as guides during the RNA interference process by binding to complementary sequences, mainly in the 3 ′ UTR of mRNAs, leading to the cleavage of the target (in case of perfect complementarity), its translational repression or its deadenylation (in case of imperfect complementarity) (reviewed in Gu and Kay 23 ). In humans, most of miRNAs are intergenic or located in introns of both non-coding and coding transcripts. However, a small portion can also be found inside exon sequences (reviewed in Olena and Patton 24 ). When primary-miRNA (pri-miRNA) sequences are located inside coding mRNAs, they usually share the host promoter and are thereby co-transcribed. In some cases, the miRNA promoter can be distinct from the one of the mRNA host gene, allowing for an independent expression. 25,26 While the regulation of the mRNA transcriptome mediated by  has been well studied, the modulation of the miRNome remains more elusive. So far, few miRNAs have been identified as being regulated by IL-6 and/or STAT3 (reviewed in Cao et al. 27 ): let-7 family members, 28 miR-17∼92 cluster, 29 miR-181b-1, 30 miR-21, 30,31 and also miR-146b-5p. 32 Therefore, we decided to further investigate the impact of IL-6 on the miRNome of both primary hepatocytes and liver-derived cell lines.
In the current study, we show that in both primary hepatocytes and non-neoplastic liver-derived cell lines miR-146b-5p is induced upon IL-6 stimulation whereas this is only the case in 1 out of 3 HCC cell lines. In contrast to the situation in breast cancer, 32 the methylation of specific CpG islands in the promoter of pri-mir-146b does not correlate with its inducibility in cell lines. In addition, miR-146b-5p does not seem to be differentially expressed in tumor samples in comparison to the matched healthy adjacent tissue. Finally, we identified new target candidates of this miRNA, which may be related to its suggested tumor suppressor role in cancer, and found evidence for a regulatory feedback loop involving IL-6, miR-146b-5p, and the NF B signaling pathway.

Ethical approval
Human primary hepatocyte collection was approved by the respective ethics committees in Germany (Ethik-Kommission der Aerztekammer des Saarlands, 79/12) and Luxembourg (Comité National d'Éthique de Recherche, 201309/07), and signed statements of informed consent were obtained from all patients.
Briefly, cells were transduced with lentiviral particles at a multiplicity of infection of 0.3 and 4 g/ml polybrene (Sigma-Aldrich, Diegem, Belgium) following the manufacturer's instructions and grown under normal conditions. Three days after transduction, the antibiotic selection (0.3 g/ml puromycin, InvivoGen, Toulouse, France) started and was maintained for 1 week.

Flow cytometry
Huh-7 and Hep3B cells were resuspended in cold PBS supplemented with 5% FBS and 0.1% sodium azide and incubated with a mouse antibody specific for OSM-R (sc-9992, Santa Cruz, Heidelberg, Germany) or the corresponding IgG control antibody (21275534, Immunotools, Friesoythe, Germany) for 1 h at 4 • C. Cells were washed with cold PBS/azide and incubated with a secondary antibody against mouse IgG and coupled with R-phycoerythrin (1:100, 12-4015-82, eBioscience, Vienna, Austria) for 1 h at 4 • C. Cells were then washed twice with cold PBS/azide and analyzed on a FACSCanto II flow cytometer using FACSDiva (BD Biosciences, Erembodegem, Belgium) software. Overlays were created using FlowJo software (Ashland, Oregon, USA).

Demethylation studies
Briefly, cells were seeded at day 0 and 8 h later treated with the vehi-

Gene expression analysis
Five hundred nanograms of total RNA was reverse-transcribed with the miScript II RT kit (Qiagen, Venlo, The Netherlands) in a volume of 10 l according to the manufacturer's instructions. Quantitative real time PCR (qPCR) was carried out on a CFX384 Detection System (BioRad, Temse, Belgium) using 5 (miRNA detection) or 50 ng (mRNA detection) RNA input in a 10 l reaction volume, 2× iTaq SYBR Green Supermix (BioRad, Temse, Belgium) and 1 l of 10× miRNA-specific primers (Qiagen, Venlo, The Netherlands) or either 2.5 pmol gene-specific primer pairs (sequences available in Table 1, Eurogentec, Liège, Belgium) or 1 l of 10× QuantiTech primers (Qiagen, Venlo, The Netherlands). miR-NAs and mRNAs of interest as well as small RNA used as miRNA normalizers (3 out of RNU1A, RNU5A, SCARNA17, and SNORD95) and reference genes used for RNA normalization (HRPT, PPIA, and TBP),

Gene
Primer Sequence Forward Primer Sequence Reverse Size (bp) were assessed in parallel for each sample and run in triplicates. Calculations were carried out by using the CFX Manager software (BioRad, Hercules, California, USA) provided with the machine.

RNA-Sequencing
Library preparation for sequencing was done with 1 g of total RNA

Pyrosequencing of selected CpG islands
Genomic DNA of non-neoplastic (PH5CH8 and THLE2) and cancerous (Hep3B, HepG2, and Huh-7) liver cells was extracted with the kit NucleoSpin Blood following the manufacturer's instructions (Macherey-Nagel, Düren, Germany). Pyrosequencing experiments on extracted gDNA as well as the analysis were performed by the company Varionostic (Ulm, Germany).

TCGA data analyses
Datasets from the liver hepatocellular carcinoma cohort of The Can-

Statistical analyses
For the statistical analysis of the normalized and prefiltered expression data of the microarrays, we used the limma package of R/Bioconductor, as previously described. 34 False discovery rates (FDR) were assigned based on F-test P-values in order to identify miRNAs with non-random profiles. Features with FDR < 0.05 were considered significant.
Non-parametric one-way Anova (Kruskal-Wallis test) followed by Dunn's multiple comparisons test were performed to assess the statistical significance between the untreated/unstimulated cells (Ctrl) and the different stimulations/treatments (Figs. 2B-F, 3C, 5, and 6).
Two-tailed paired t-test was used to assess statistical significance of the effects of miR-146b-5p overexpression (with and without Dox) on the expression levels of specific mRNA genes (Fig. 4A, B, and D).
All comparisons were carried out with the statistical program Prism 7 (GraphPad Software, La Jolla, California, USA).

miR-146b-5p is an IL-6-regulated miRNA in primary human hepatocytes
We stimulated human primary hepatocytes isolated from 2 donors

Expression of miR-146b-5p, but not miR-146a-5p, is induced by HIL-6 and Oncostatin M in non-transformed liver cell lines
miR-146b-5p and miR-146a-5p, the other member of the miR-146 family, are located on different chromosomes (10 and 5, respectively) and differ by only 2 nucleotides at the 3 ′ end (letters in gray, Fig. 2A), i.e. the seed region is shared by both miRNAs (in bold, Fig. 2A). We investigated the cytokine-mediated regulation of these miRNAs in 2 non-neoplastic immortalized cells lines (PH5CH8 and THLE2) and 3 HCC cell lines (Huh-7, HepG2, and Hep3B). As additional stimuli, another STAT3-activating cytokine, the IL-6-type cytokine OSM, as well as the STAT1-activating cytokine IFN and an activator of the NF B pathway, TNF , were included. The activation was monitored by Western blot analysis (Fig. 2G). HIL-6 and OSM led to increased levels of pSTAT3 (except OSM-treated Hep3B cells which lack the OSM receptor, Supplementary Fig. 1A), and IFN led to increased pSTAT1 levels. TNF did not affect phosphorylation of STATs. In PH5CH8 cells,

Promoter methylation does not account for lower induction of miR-146b-5p
In primary hepatocytes, the basal expression of this miRNA was already higher than in liver-derived cell lines (Supplementary Table 2) and we also observed a stronger up-regulation upon cytokine stimulation. To further address this difference in expression, we investigated the methylation state of the pri-mir-146b promoter. Specifically, we measured the methylation percentage of 4 CpG islands (at positions -63, -56, -26, and +70 relative to the 1st nucleotide of pre-mir-146b, Fig. 3A), which were shown to be crucial for the reduced miR-146b-5p expression in breast cancer cells 32 (Fig. 3B). Interestingly, CpG island methylation for all 4 investigated sites was lower in the non-responsive HepG2 cells (Fig. 3B), while in the responsive Huh-7 and PH5CH8 cells, the percentage of methylation for each CpG island was higher (Fig. 3B). In addition, we performed 5-AZA demethylation studies in Hep3B, Huh-7, and PH5CH8 cells (showing a high degree of CpG island methylation, Fig. 3B). However, neither basal nor inducible miR-146b-5p expression was affected by the demethylation agent ( Fig. 3C). To ensure the effectiveness of the 5-AZA treatment, we checked the expression level of SOCS3 in Hep3B and Huh-7 cells, for which it has been described that the promoter is subject to CpG island modification. 46 Indeed, upon 5-AZA treatment, we observed an increased expression of this gene (Supplementary Fig. 1B). To study if other CpG islands located in the presumed promoter region of pri-mir-146b (Fig. 3A) could be implicated, we analyzed the methylome data from the Liver Cancer cohort of The Cancer Genome Atlas (TCGA LIHC) available for 50 HCC tumor tissues and the matched solid "normal" tissues (black and gray boxes, respectively, Fig. 3D). The which both information on the miR-146b-5p expression levels and the methylation status was available, were compared to the corresponding adjacent "normal" tissues (Fig. 3E). A lower CpG islands methylation in the tumors than in the surrounding tissues was observed throughout, irrespective of an increased, unchanged, or decreased miR-146b-5p expression in the corresponding samples (Fig. 3F, G, and H). Taken together, these results indicate that the methylation status of the tumor tissues was different from the one in the surrounding tissues but this was not reflected in the miR-146b-5p expression levels.

miR-146b-5p affects expression of multiple genes, including IRAK1 and TRAF6 involved in TLR signaling
We showed that miR-146b-5p was reproducibly responding to HIL-6 stimulation, hence we investigated potential target genes of this miRNA. To do so, we engineered PH5CH8 liver cells, which inducibly overexpressed this miRNA, named PH5CH8-146b cells. Upon Dox treatment, a strong induction could be observed when miR-146b-5p levels were monitored by qPCR (Fig. 4A). Its induction affected the expression of several mRNAs, as revealed by RNA-Seq analysis.  Table 2. These included known targets of miR-146b-5p such as IRAK1, CARD10, and ZNRF3. TRAF6, another well-described target, was also down-regulated, however, with a lower log 2 |FC| (0.38) than the above-mentioned ones (complete lists of all significantly up-and down-regulated mRNAs can be found in Supplementary Table 3 Reduced mRNA levels of IRAK1 and TRAF6, part of a complex involved in NF B activation upon TLR4 triggering, were confirmed by qPCR following miR-146b-5p overexpression ( Fig. 4B and D) as well as a reduced protein level of TRAF6 (Fig. 4E). We also monitored the expression of downstream targets of NF B signaling, the genes encoding the chemokine MCP1 (also named CCL2) and IL-6, and found both to be induced by TNF and LPS (Fig. 4D). Importantly, only the LPS-mediated up-regulation of MCP1 and IL-6 mRNA expression was attenuated when miR-146b-5p was overexpressed (Fig. 4D).

miR-146a-5p and miR-146b-5p regulate similar mRNA target genes
miR-146a-5p and miR-146b-5p differ only by 2 nucleotides but have the same seed region ( Fig. 2A). Therefore, we next investigated whereas they indeed regulate the same genes by transiently transfecting liver-derived cells (PH5CH8 and Huh-7) with a negative mimic control (NCM1), miR-146a-5p, or miR-146b-5p mimic and quantified the expression levels of the previously validated mRNA candidates ( Fig. 5A and B, respectively). In contrast to the induction in stable PH5CH8-146b cells, transient transfection of 20 nM mimic led to a huge overexpression (more than 5,000 times in both cell lines, Supplementary Fig. 1C) of the corresponding miRNA. Not surprisingly, crossreactivity of both primer pairs was observed under these conditions of strong miRNA overexpression. All 17 mRNA candidates identified by RNA-Seq of miR-146b-5p overexpressing PH5CH8 cells were confirmed following transient transfection of PH5CH8 and Huh-7 cells.
Some candidate genes like FOSL1 and SHISAL1 could not be confirmed in Huh-7 cells. Importantly, both miR-146a-5p and miR-146b-5p showed very similar effects on the various potential targets, confirming the expected redundancy due to their identical seed regions.
To investigate if the effects observed after mimic transfection were due to a direct interaction between the miRNAs and the 3 ′ UTR of potential targets (CPA4, DNAJC6, MAGEE1 MPHOSPH6, PPP2R1B, PTX3, SLC10A3, SNRNP27, and TIMM17B), we performed dual luminescence reporter assays. These mRNAs were selected as they were so far unknown targets of miR-146b-5p or miR-146a-5p. Hek293T cells were co-transfected with the vector containing the gene of the secreted Gaussia luciferase followed by the 3 ′ UTR of the candidate genes and with NCM1, miR-146a-5p, or miR-146b-5p mimics. We observed similar results for both miRNAs with a reduction of 30-40% in luciferase activity upon mimic transfection for the 3 ′ UTRs of TIMM17B and DNAJC6, 40-50% for the 3 ′ UTRs of MAGEE1, MPHOSPH6, and SNRNP27, and more than 60% for the 3 ′ UTRs of PPP2R1B and SLC10A3 ( Fig. 6 and Table 2). These data confirmed a direct binding of the miRNAs to the 3 ′ UTRs of the selected mRNAs.
While we observed a reduction of CPA4 expression at the mRNA level in PH5CH8 cells (Fig. 5A), only a moderate diminution of luciferase activity was measured. The data suggested an indirect effect of the miRNAs or that the binding sites were not only localized in the 3 ′ UTR of the CPA4 mRNA, but rather elsewhere (5 ′ UTR or coding sequence).
Surprisingly, PPP2R1B showed the strongest effect on luciferase activity even though at the mRNA level the down-regulation was moderate in both cell lines (Fig. 5). Interestingly, most of the 3 ′ UTRs further studied were not predicted as potential miR-146b-5p targets, highlighting the fact that experimental screenings and validation are necessary to identify new miRNA targets.

DISCUSSION
Hepatocellular carcinoma most often develops in an inflamed liver, and the inflammatory cytokine IL-6 plays a role in hepatocarcinogenesis, as demonstrated in the DEN-induced HCC mouse model. 17 In order to elucidate the role of IL-6 in the modulation of the hepatic miRNome, we performed microarray analyses on primary hepatocytes stimulated with HIL-6 and identified miR-146b-5p as being the most robustly induced miRNA.
This miRNA, belonging to the miR-146 family along with miR-146a, was shown to have a potential tumor-suppressive role in HCC development 50 and to be differentially expressed in inflammatory liver TA B L E 2 Top 20 of significantly (adjusted P-value < 0.05) negatively differentially expressed mRNAs upon DOX treatment in PH5CH8 inducibly overexpressing miR-146b-5p found by RNA-Seq analysis  32 We therefore inspected various liver-derived cell lines regarding their ability to respond to cytokine stimulation with an increased expression of miR-146b-5p. We have shown that miR-146b- TNF is also able to induce the expression of miR-146b-5p in mature human adipocytes, 54 although this cytokine is more generally  This study provides evidence of (i) cytokine-inducible miR-146b-5p expression ( Figs. 1 and 2), (ii) effects of (overexpressed) miR-146b-5p on expression of well-known targets IRAK1 and TRAF6 (Figs. 5 and 4E), 55,58,72 and (iii) effects of (overexpressed) miR-146b-5p on LPS-regulated target genes (Fig. 4D). STAT3 binding to the pri-miR-146b-5p promoter has been shown for mammary cells. 32 subunit Abeta), and SRPRB (SRP Receptor Beta Subunit). The DNAJC6

% of Reduction Compared to Ctrl
gene, encoding the HSP40 Auxilin, is up-regulated in HCC tissues and associated with poor outcome and enhancement of cell proliferation and invasion. 65 SRPRB is part of the small G protein superfamily and contains a small GTP-binding protein domain. 66 Its overexpression in HCC cell lines leads to an up-regulation of p21, a down-regulation of TIMP3 and finally G1 arrest. 67 The third one, PPP2R1B, is a PPP2A family member with serine/threonine phosphatase properties. Deletions in the gene sequence were more frequent in HCC tumors than in the surrounding tissue, 68 and a decreased expression was associated with inhibition of cell proliferation, migration and invasion in HepG2 cells. 69 In a next step, functional consequences of the up-regulation of miR-146b-5p and a resulting down-regulation of DNAJC6, SRPRB, and PPP2R1B could be investigated in the context of healthy and HCC cells.
As the major mature forms of miR-146a/b (-5p) have very similar sequences and identical seed regions, they are predicted to target the same subset of genes. Therefore, we transiently trans-  (mcp1 and il6). In Fig. 7, we schematically show a potential negative regulatory crosstalk in which LPS-(but not TNF -) mediated signaling is inhibited by the IL-6-induced miR-146b-5p, as previously suggested for mammary cells. 32 Our results obtained with cells overexpressing miR-146b-5p (Fig. 4D) are in line with the fact that TNF , in contrast to LPS, does not utilize IRAK1 and TRAF6 for NF B activation. It is, however, well possible that under physiological conditions a single miRNA plays only a role in fine modulation of complex signaling networks. It would therefore be interesting to test whether exogenous administration of miR-146b-5p could have beneficial effects in diseases involving IL-6, as shown in high-fat diet-induced non-alcoholic steatohepatitis in rats. 73 IL-6, as well as other cytokines, has been reported to play a pivotal role in inflammatory events in the context of fatty liver diseases (reviewed in Hassan et al. 74 ) which are known to increase the risk of cirrhosis and HCC development. 75 Furthermore, autocrine IL-6 stimulation of liver cancer progenitor cells has been described to promote malignant progression of HCC. 76 Interestingly, the IL-6-induced miR-146b-5p was found to be up-regulated during progression to steatosis and steatohepatitis in rats 52 as well as in liver diseases (e.g., in obese subjects with non-alcoholic fatty liver diseases (NAFLD) 53 and in NASH patients 51 ). On the contrary, it has been shown that this miRNA (as well as miR-146a-5p 77 ) was reduced in HCC tumor tissues compared to the surrounding tissues, indicating a tumor-suppressive role for this miRNA. 50 Hepatoma most often arises in the context of chronic inflammation and cirrhosis. 1 Thus, the "normal" tissues surrounding hepatic tumor tissues may reflect a wide range of inflammatory states. It can be envisaged that the discrepancy between our results on the TCGA HCC cohort (Fig. 3)

DISCLOSURES
The authors declare no conflicts of interest.