MicroRNA-23b cluster microRNAs regulate transforming growth factor-beta/bone morphogenetic protein signaling and liver stem cell differentiation by targeting Smads


  • Charles E. Rogler,

    Corresponding author
    1. Department of Medicine, Marion Bessin Liver Research Center, Division of Hepatology, Bronx, NY
    • Albert Einstein College of Medicine, 1300 Morris Park Avenue, Bronx, NY 10461
    Search for more papers by this author
    • These authors contributed equally to this study.

    • fax: 718-430-8975

  • Lauretta LeVoci,

    1. Department of Medicine, Marion Bessin Liver Research Center, Division of Hepatology, Bronx, NY
    Search for more papers by this author
  • Tammy Ader,

    1. Department of Medicine, Marion Bessin Liver Research Center, Division of Hepatology, Bronx, NY
    Search for more papers by this author
  • Aldo Massimi,

    1. Department of Molecular Genetics, Albert Einstein College of Medicine, Bronx, NY
    Search for more papers by this author
  • Tatyana Tchaikovskaya,

    1. Department of Medicine, Marion Bessin Liver Research Center, Division of Hepatology, Bronx, NY
    Search for more papers by this author
  • Raquel Norel,

    1. Department of Medicine, Marion Bessin Liver Research Center, Division of Hepatology, Bronx, NY
    Current affiliation:
    1. Center for Computational Biology and Bioinformatics, Columbia University, New York, NY 10032
    Search for more papers by this author
  • Leslie E. Rogler

    Corresponding author
    1. Department of Medicine, Marion Bessin Liver Research Center, Division of Hepatology, Bronx, NY
    • Albert Einstein College of Medicine, 1300 Morris Park Avenue, Bronx, NY 10461
    Search for more papers by this author
    • These authors contributed equally to this study.

    • fax: 718-430-8975

  • Potential conflict of interest: Nothing to report.


Transforming growth factor-beta / bone morphogenetic protein (TGFβ/BMP) signaling has a gradient of effects on cell fate choice in the fetal mouse liver. The molecular mechanism to understand why adjacent cells develop into bile ducts or grow actively as hepatocytes in the ubiquitous presence of both TGFβ ligands and receptors has been unknown. We hypothesized that microRNAs (miRNAs) might play a role in cell fate decisions in the liver. miRNA profiling during late fetal development in the mouse identified miR-23b cluster miRNAs comprising miR-23b, miR-27b, and miR-24-1 and miR-10a, miR-26a, and miR-30a as up-regulated. In situ hybridization of fetal liver at embryonic day 17.5 of gestation revealed miR-23b cluster expression only in fetal hepatocytes. A complementary (c)DNA microarray approach was used to identify genes with a reciprocal expression pattern to that of miR-23b cluster miRNAs. This approach identified Smads (mothers against decapentaplegic homolog), the key TGFβ signaling molecules, as putative miR-23b cluster targets. Bioinformatic analysis identified multiple candidate target sites in the 3′ UTRs (untranslated regions) of Smads 3, 4, and 5. Dual luciferase reporter assays confirmed down-regulation of constructs containing Smad 3, 4, or 5, 3′ UTRs by a mixture of miR-23b cluster mimics. Knockdown of miR-23b miRNAs during hepatocytic differentiation of a fetal liver stem cell line, HBC-3, promoted expression of bile duct genes, in addition to Smads, in these cells. In contrast, ectopic expression of miR-23b mimics during bile duct differentiation of HBC-3 cells blocked the process. Conclusion: Our data provide a model in which miR-23b miRNAs repress bile duct gene expression in fetal hepatocytes while promoting their growth by down-regulating Smads and consequently TGFβ signaling. Concomitantly, low levels of the miR-23b miRNAs are needed in cholangiocytes to allow TGFβ signaling and bile duct formation. (HEPATOLOGY 2009.)

The embryonic liver of the mouse is an organ of great contrast at late stages in its development (embryonic day [E]16.6 to birth).1, 2 On one hand, newly differentiated hepatocytes are in a very active growth process that is needed for the formation of the liver parenchyma. On the other hand, hepatoblasts near the portal mesenchyme undergo apoptosis or differentiation into cholangiocytes to form the ductal plates. The ductal plates undergo further morphogenesis to form mature bile duct structures in a process that continues in the early postnatal liver (Fig. 1).3, 4 At the same time, the fetal liver is primarily a hematopoietic organ.

Figure 1.

(A) Schematic of biliary morphogenesis from E16.5 days in the mouse to P1. (B) Clustergram depicting the miRNA expression at E16.5, 17.5, and P1 normalized to adult liver. Black: adult liver level. Blue: lower than adult liver. Yellow: higher than adult liver. Data are calculated as the log2RAT of expression day X/expression in adult liver. Boxes are outlined in red. Box 1: hematopoietic cell miRNAs-transiently induced. Box 2: epithelial cell differentiation miRNAs expanded box to right of main clustergram. Red stars mark miRNAs that are also up-regulated during hepatocytic differentiation of HBC-3 cells. The relative abundance is equal to the mean of spot intensity / the average spot intensity for the entire set of expressed miRNAs. (C) Quantitative display of miRGEM data for miRs-23b, 27b, and 24 from E16.5 (black), E17.5 (light gray), P1 (gray), or adult 8-week-old (white). (D) Relative expression changes in the top 31 expressed miRNAs as HBC-3 cells undergo hepatocytic differentiation. Vertical axis: Log2 ratio of hepatocytic/undifferentiated expression levels. MiRNAs are sorted (left to right) from the most highly induced to the most inhibited.

The murine fetal hepatoblast cell line, HBC-3, can be induced to differentiate toward the hepatocytic pathway with treatment of dimethylsulfoxide (DMSO), or to form bile ducts by plating in Matrigel.5 During hepatocytic differentiation, transforming growth factor-beta (TGFβ) receptors, and Smads (mothers against decapentaplegic homolog) 3, 4, and 5, which are the primary signal transduction molecules for TGFβ, are down-regulated, along with many downstream targets of TGFβ signaling.7 In contrast, during bile duct formation in Matrigel the TGFβ receptors and the same Smads are up-regulated.6

Transgenic knockout mice for TGFβ pathway genes have shown a requirement for TGFβ signaling for balanced development of hepatocytic and bile ductular compartments of the liver.8, 9 Specifically, TGFβ is necessary for the formation of bile ducts.9, 10 Heterozygous TGFβ knockout mice revealed a haploinsufficiency phenotype and the haploinsufficiency extended to Smads 2 and 3. Mouse embryos that were heterozygous for both Smad 2 and Smad 3 were hypomorphic and were deficient in bile ducts.8

A prominent broad-based function of TGFβ is strong inhibition of the cell cycle.11 This occurs through direct repression of c-Myc by activated Smad complexes, and activation of at least two cell cycle-inhibiting tumor suppressors.11 These effects occur in virtually all normal cell types that express TGFβ receptors and Smads.11, 12 Both TGFβ ligands and TGFβ receptors are present throughout the hepatocytic and bile ductular compartments of the embryonic liver.13

Therefore, a fundamental problem that has existed in the field of embryonic liver development is the question of how hepatocytes can avoid growth inhibition by TGFβ, whereas cells immediately adjacent to them (next to the portal mesenchyme) are positively affected to differentiate into cholangiocytes and then remodel into bile ducts.10 Various explanations have been suggested, involving Notch and hepatocyte growth factor (HGF) signaling.9, 10 However, a molecular mechanism for the establishment of a TGFβ response gradient across a single layer of cells in the embryonic liver has remained elusive.9, 10

MicroRNAs (miRNAs) are small noncoding RNAs that are products of highly conserved genetic loci.14 They generally function by inhibiting gene expression through posttranscriptional mechanisms.15 They are also highly developmentally regulated and function in embryonic organogenesis.16 They are also differentially expressed in different cell types in developing organs.16 Furthermore, recent evidence has shown that expression of miRNAs can regulate the divergent differentiation pathways of stem cells.17, 18 As such, they represent cell autonomous genetic elements that can establish different differentiation pathways in adjacent cells.

Therefore, we reasoned that miRNAs could be responsible for regulating cell fate decisions in the embryonic liver by regulating the response of cells to ubiquitous growth factors, such as TGFβ. In this report, we identify a cluster of miRNAs (miRs 23b, 27b, and 24-1, from the miR-23b cluster) that are highly expressed in hepatocytes compared to developing bile ducts in fetal mouse liver. We present evidence that the miR-23b cluster miRNAs serve as a molecular switch regulating TGFβ response of liver stem cells, and their differentiation fate.


BMP, bone morphogenetic protein; miRGEM, microRNA gene microarray; miRNA, microRNA; TGFβ, transforming growth factor-beta; UTR, untranslated region.

Materials and Methods


See supporting data for details of liver dissection protocols.

Cell Culture.

HBC-3 cells were cultured as described.5 Hepatocytic differentiation was induced by treatment with 3.5% DMSO in Hepes-buffered medium (HBM).5 Bile duct morphogenesis was induced by plating on Matrigel as described.6 STO cells are a murine embryonic fibroblast cell line from an SIM mouse that is thioguanine-resistant, ouabain-resistant, and neomycin-resistant. Mitomycin treatment was carried out as reported.5

MiRNA Isolation.

Small RNA fractions were isolated from liver tissue using the miRVana miRNA isolation kit (Ambion). Total RNA was isolated from HBC-3 cells as described19 for northern blot and quantitative reverse-transcribed polymerase chain reaction (qRT-PCR) analysis.

MicroRNA Gene Microarray (miRGEM).

A set of 372 DNA oligonucleotides of 20-23 nucleotides homologous to the mature (sense) orientation of murine miRNAs listed on the Sanger database (2006). Specifics of slide printing, probe preparation, hybridization, and data processing are in the Supporting Methods.

In Situ Hybridization.

Eight-micron cryosections were fixed in 4% paraformaldehyde in phosphate-buffered saline (PBS). Slides were pretreated as described.20 Details of hybridization protocol and photography are found in the Supporting Methods.

cDNA Microarray.

The cDNA microarrays containing 27,400 sequence verified mouse clones used were obtained from the Albert Einstein College of Medicine Functional Genomic Facility. Details of probe preparation, hybridization, and data analysis and bioinformatics were carried out as described6 and in the Supporting Methods. The time course described in the text was performed in triplicate using a total of 24 microarrays.

Bioinformatics of Target Identification.

Putative targets were identified by comparison of lists of putative targets from Miranda and TargetScan with the HBC-3 cDNA array expression database. 3′-UTR (untranslated region) sequences were recovered from GenBank and the sequences were fed to RNA hybrid (RNAhybrid.org) to identify sites of potential seed matches. Additional criteria used to judge potential target sites included: a match of greater than six nucleotides in the seed sequence21; low ΔG; additional homology to the remaining sequences within the microRNA.

MiRNA Reporter Assays.

Fragments of the 3′ UTRs of Smads 3, 4, and 5 outlined in Fig. 3 were amplified with the primers listed in Supporting Methods, which carried XhoI/NotI sites and cloned into those sites downstream of Renilla luciferase in the psiCheck2 vector (Promega). Details of the methods for the Dual Luciferase reporter assays are in the Supporting Methods.

Figure 3.

(A) Effects of miRNA mimics on reporter gene expression constructs containing 3′ UTR fragments from Smad 3, Smad 5, Smad 4, and PsiCheck2miR-16. Light gray bars: no-mimic. Dark gray bars: miR-16 mimic. Black bars: mixture of mimics for miRs-23b, 27b, and 24. Normalized R/F: data are the ratio of Renilla/Firefly luciferase normalized to the no-mimic control for each plasmid. Mimics transfected into Hek293 cells and assayed by the Dual Luciferase Assay 20 hours after transfection in 96-well plates. Quadruplicate samples measured. Statistical analysis by Student's t test, *P = 0.05, **P = 0.006. (B) Effects of miRNA inhibitors on marker gene expression in HBC-3 treated with DMSO (hepatocytic differentiation). Fold change: increase in gene expression versus undifferentiatied, due to treatment. Gray bars: DMSO treatment only. Black bars: DMSO treatment plus mixture of inhibitors against miRs-23b, 27b, and 24. (C) Effects of miRidian mimics on bile duct differentiation of HBC-3 cells plated on Matrigel. Control: no transfection. Mock: transfection reagent alone. Mimic mix: transfection of a mixture of miRidian mimics of miRs 23b, 27b, and 24. Negative control: transfection with miRidian negative control mimic. Scrambled SiRNA: scrambled SiRNA from IDT. Smad4 siRNA-: siRNA against Smad 4 from IDT. Images taken 24 hours after plating on Matrigel. Magnification 25×. (D) Quantitation of the effects of miRidian mimics on bile duct formation 24 hours after plating transfected HBC-3 cells on Matrigel. miRidian mimic transfections as in (C). Percent of total area (pixels2): a minimum of 27 nonoverlapping images were made of each condition and the area of tubules and clusters was calculated using ImageJ (NIH). Treatment with the miRidian mimics effectively blocked tubule formation by HBC-3 cells.

Methods for quantitative real-time PCR and transfection of HBC-3 cells, northern blotting, and RT/PCR of miR-23b cluster miRNAs are in the Supporting Methods.


miRGEM Profiling of the Fetal Liver and Hepatoblast Differentiation In Vitro.

We conducted miRNA profiling of fetal murine liver from E16.5 to birth using a custom-made oligonucleotide miRNA Gene Expression Microarray (miRGEM) comprising murine miRNAs (see Materials and Methods). We compared the relative expression level of miRNAs at E16.5, E17.5, and perinatal day (P)1 with adult mouse liver. Hybridization data from the significantly expressed miRNAs were clustered (Fig. 1B). Many of the highly expressed miRNAs were members of families of miRNAs that were expressed in a coordinate pattern (Supporting Table 1).19 Cluster analysis revealed several temporally regulated sets of miRNAs including one set that was highly up-regulated in only the E17.5 and P1 samples, compared to adult liver (Fig. 1B, box 1). This cluster contained miR-181a, miR-181b, miR-136, and miR-154, all of which have been causally linked to hematopoiesis,17, 18 which occurs in the fetal liver at this time and is absent from the adult liver.

Another major cluster of miRNAs that were expressed at very low levels at E16.5 and were up-regulated from E17.5 through adult liver are highlighted in the expanded box in Fig. 1B, box 2. Included in these miRNAs was the miR-23b cluster, including miR-23b, miR-27b, and miR-24-1. These three miRNAs are derived from a single pri-miRNA encoded on chromosome 13 of the mouse19 and are referred to as the miR-23b cluster miRNAs throughout this report. Quantitative analysis of the miRGEM data showed that the three miRNAs were coordinately up-regulated from E17.5 to adult liver (Fig. 1C). Additional up-regulated miRNAs included three members of the let-7 family (let-7a, e, d), and miRs-126, 26a, 10a, and 30a (Fig. 1B). Interestingly, miR-122, which is the major miRNA in liver,19 did not vary in its expression from E16.5 to adult. This is consistent with previous work showing that miR-122 is induced soon after liver specification before E16.5.22

We used HBC-3 cells to address the role(s) of miRNAs in hepatocytic and bile ductular differentiation. HBC-3 cells are a euploid hepatoblast cell line that can be maintained in an undifferentiated state on STO feeder layers and can be induced toward the hepatocytic lineage by treatment with DMSO and toward the bile ductular lineage by plating on Matrigel.5 HBC-3 cells have been extensively characterized with regard to their transplantation properties6, 19 gene expression profiles during both hepatocytic7 and bile ductular6 differentiation in vitro.

miRNA profiling of HBC-3 cells during hepatocytic differentiation revealed 16 up-regulated miRNAs and nine down-regulated miRNAs (Fig. 1D). Several of the highest up-regulated miRNAs in the HBC-3 cells were similarly up-regulated in the fetal liver (marked by red stars in Fig. 1B), suggesting that the HBC-3 model closely mimics changes in the fetal liver. The two highest up-regulated miRNAs in HBC-3 cells were miR-23b and miR-23a, and the other two miR-23b cluster miRNAs, miR-27b and miR-24, were within the top seven most highly up-regulated miRNAs. Northern blot and qRT-PCR data confirmed the coordinate up-regulation of miRs-23b, 27b, and 24 during hepatocytic differentiation (Supporting Fig. 1). Interestingly, expression of these microRNAs was low and did not appreciably change during bile ductular differentiation induced by Matrigel (Supporting Fig. 1).

The miR-23b cluster is located in an intron of a cDNA for which a protein has not been identified (Riken gene 201011I01RiK). Data mining of the expression of the RIKEN 201011I01 locus, from our cDNA microarray studies,6, 7 confirmed that this locus is up-regulated during hepatocytic differentiation (DMSO treatment) and is down-regulated during bile ductular differentiation plated on Matrigel (Supporting Fig. 2). Additionally, we identified two previously uncharacterized CAAT and TATA boxes and a set of transcription factor binding sites upstream of the miR-23b cluster, suggesting that it may be independently transcribed (Supporting Figs. 2, 3). We used RT-PCR to independently verify expression of the miRNA locus, using one primer between the two CAAT/TATA sequences and one primer immediately downstream from miR-24-1. This procedure detected the expected 1,700 basepair transcript in both undifferentiated and hepatocyte differentiated HBC-3 cell RNAs (Supporting Fig. 2).

MiR-23b Is Specifically Expressed in Developing Hepatocytes.

We next determined which cells within the developing liver expressed the miR-23b cluster miRNA. By day 15.5, a continuous single layer of biliary precursors, called the ductal plate forms around the portal mesenchyme.1 The plate becomes bilayered by day 16.5, and focal dilations form by E17.5 (Fig. 1A). The focal dilations will form the bile duct and the remainder of the ductal plate normally regresses. Therefore, we carried out in situ hybridization for miR-23b on E17.5 murine embryos. This analysis identified miR-23b signal in hepatocytes and no detectable signal in ductal plates, focal dilations, portal mesenchyme, or surrounding tissues (Fig. 2 Aa,e). A control hybridization with the major liver miRNA, miR-122, also revealed a high level of expression in hepatocytes and also no detectable expression in ductal plates or surrounding tissues (Fig. 2 Ab-d). The expression pattern within hepatocytes was distinctly punctate, with many small foci in the cytoplasm and almost always one large distinct focus in the nucleus (Fig. 2 Ac). From these data we concluded that miR-23b cluster miRNAs function primarily in the hepatocytes of the fetal liver.

Figure 2.

(A) In situ hybridization of E17.5 fetal liver and adult liver. (a) Detection of miR-23b in E17.5 liver. (b) Detection of miR-122 in E17.5 liver. (c) Detection of miR-122 in adult liver. (d) Negative control using miR-159 sense probe on adult liver. (e) Negative control showing lack of miR-23b signal over nonexpressing tissue in E17.5 embryo. Dark blue dots: punctate positive hybridization signals. Red line demarcates ductal plate. Dashed black line: focal dilation within a ductal plate. Black solid line demarcates portal mesenchyme surrounding a vessel. Ductal plates and mesenchyme are negative for miR-23 and miR-122. (B) Clustergram of genes expressed in HBC-3 cells during hepatocytic (DMSO) or bile ductular (Matrigel) differentiation. Only genes selected for bioinformatically predicted miR-23b targets are shown. Black: expression equal to time 0 (undifferentiated cells). Blue: expression decreased relative to time 0. Yellow: expression increased relative to time 0. Comparisons are made to mRNA levels in undifferentiated HBC-3 cells. Genes down-regulated during DMSO and up-regulated during Matrigel included Smad 3, Mapre1, Mlp, Lsm 3, and Rbpms2 in top cluster. (C) Quantitative plot of expression of Smads 1, 2, 3, 4, 5, and 7 during bile duct differentiation (squares) versus hepatocyte differentiation (triangles) as determined by cDNA microarrays.6 Median RAT(log2) is the expression relative undifferentiated HBC-3 cells. Days: number of days treated with DMSO (triangles) or plated on Matrigel (squares). (D) Mapping of candidate miR-23b, 27b, and 24 target sites in the 3′ UTRs of Smads 3, 4, and 5 3′ UTRs. Red: miR-23b target sites. Blue: miR-27b target sites. Green: miR-24 target sites. Black arrows under maps: segments of 3′ UTRs that were subcloned and tested for target sites using the dual luciferase reporter assay. *Target sites detailed in Fig. 2E. (E) Sequence complementarity between miRNAs and candidate target sites in Smad 3, 4, and 5. Target: sequence of gene 3′UTR mRNA. miRNA: sequence of the mature miRNA. Position: position in the 3′ UTR. −DG: net negative free energy of secondary structure formation.

Identitification of Targets of MiR-23b Cluster miRNAs.

To identify potential mRNA targets of miR-23b cluster miRNAs we conducted cDNA microarray analysis during 7 days of hepatocytic differentiation of HBC-3 cells treated with DMSO. Triplicate microarrays, containing 27K cDNAs, were hybridized and analyzed as published using an 8,900 gene array.7 To focus our data analysis, we selected genes present on the 27K cDNA microarray that contained miR-23b target sites according to the mirBase website (http://microrna.sanger.ac.uk/) and analyzed their expression data. We clustered the cDNA microarray data from that set of candidate genes and looked for a pattern in which the mRNA level went down during hepatocytic differentiation, based on the hypothesis that a miR-23b target site would destabilize the mRNA leading to a reduction in its steady state level. This cluster analysis revealed 18 genes, containing candidate miR-23b target sites, which were down-regulated during DMSO treatment (Fig. 2B).

Furthermore, because miR-23b miRNAs were low or reduced during bile duct differentiation in Matrigel, we added a second criterion, that the same mRNAs should be stable or up-regulated during Matrigel-induced bile duct differentiation. We clustered the DMSO and Matrigel datasets in parallel and identified five genes with the appropriate inverse expression pattern. Remarkably, the most prominently regulated gene of the five was Smad 3, which was represented on the microarray by three separate cDNAs, all of which revealed the same pattern of expression (Fig. 2B).

Because Smad 3 is a key molecule in the TGFβ signaling pathway11 and TGFβ is involved in bile duct differentiation in the liver,4 we asked if other Smads, including Smads, 1, 2, 3, 4, 5, and 7, had an expression pattern similar to Smad 3 (Smad 6 was not on the cDNA microarray) (Fig. 2C). Quantitative analysis of the cDNA microarray data (using 27K microarrays)6 revealed the earliest (day 1) and most significant inverse patterns of expression for Smads 3 and 5. Smad 1 up-regulation began later (day 3) and Smad 2 was not differentially regulated. Smad 2 is generally a minor Smad compared to Smad 3 and cDNA array data showed it had a weaker hybridization signal, suggesting a lower level of expression in HBC-3 cells (data not shown). Smad 4 regulation began early (day 1), but was not as pronounced as Smad 3 and 5. Regulation of the inhibitory Smad 7 was weak in the early phase and peaked very late in the differentiation protocol (day 6) long after defined bile ducts were formed.

Investigations of miRNA regulation of all of the receptor and inhibitory Smads, plus TGFβ/BMP (bone morphogenetic protein) receptors, ligands, and their antagonists and agonists will be important to understand the complete impact of miRNA regulation on these interrelated pathways (see Discussion). However, because Smads 3, 4, and 5 were regulated in a similar manner and are major components of the TGFβ/BMP signal transduction pathway, we focused our initial experimental work on these centrally important Smads. Using the microRNA.org and Target Scan 4.1 websites we carried out a detailed analysis of the 3′ UTRs of Smads 3, 4, and 5 for miR-23b cluster miRNA target sites. Our search criteria included a “seed match” plus additional downstream homology with the target and a significant −DG (net negative free energy of secondary structure formation), parameters similar to published target prediction approaches (see Supporting Methods and Ref.21).

We identified multiple candidate target sites and created a map for all three miR-23b cluster miRNAs in Smads 3, 4, and 5 (Fig. 2D). Some of the sites were conserved and had been previously identified; however, most of the sites with strong seed matches and additional homology had not been previously identified. The specific sequence matches between miRNA and 3′ UTR for one of the strongest target sites for each miRNA and each Smad are illustrated in Fig. 2E, and the remainder of the sites are listed in Supporting Table 2. Overall, we identified 47 candidate target sites in the three Smad genes for an average of 5.6 sites per kilobase of 3′ UTR. We screened the same length of 3′ UTR sequences from randomly chosen genes using the same target identification criteria, and we identified only 1.0 site per kilobase of UTR. Statistical analysis using the z test method revealed the number of sites in the Smad 3′ UTRs to be approximately 5-fold higher than predicted for a random distribution. Therefore, we concluded that Smad genes may be preferentially targeted by miR-23b cluster miRNAs.

Smad 3, 4, and 5 Are Functional Targets of Mir-23b Cluster MiRNAs.

To begin to characterize the functionality of the candidate miRNA target sites, we subcloned segments of the 3′ UTRs of Smads 3, 4, and 5 downstream of the Renilla luciferase in the dual luciferase PsiCheck2 reporter vector (Promega) (Fig. 2D, underlined region). We used Hek 293 cells, which have very low or undetectable levels of miR-23b cluster miRNAs,19 for the reporter assays, and tested the effects of miRNA mimics (Dharmacon) on the Renilla/Firefly luciferase ratio. Specific miRNA targeting of the 3′ UTRs was detected by a reduction in the Renilla-to-Firefly luciferase ratio. All three of the Smad 3′ UTRs mediated significant down-regulation of Renilla in response to the miR-23b cluster mimics (mixture of miR-23b, miR-27b, and miR-24 mimics) and not to the miR-16 mimic (Fig. 3A). A control PsiCheck 2 vector, engineered to contain miR-16 sites downstream of Renilla, responded to miR-16 mimics but not to a mixture of miR-23b, 27b, and 24 mimics (Fig. 3A). These data support the conclusion that miR-23b cluster miRNAs target the 3′ UTRs of Smads 3, 4, and 5. The presence of multiple candidate target sites in the 3′ UTRs suggest that combinations of the miRNAs may be functioning together and site-directed mutagenesis of individual miR-23b cluster miRNA target sites has supported this hypothesis (data not shown).

We next reasoned that if Smads 3 and 5 are targets of miRs 23b, 27b, and/or 24 in HBC-3 cells, then knockdown of the miRNAs during hepatocytic differentiation should cause the inappropriate increased expression of these Smads. To test this hypothesis we treated HBC-3 cells with a mix of miRIDIAN microRNA inhibitors against miR-23b cluster miRNAs and then induced hepatocytic differentiation with DMSO. qRT-PCR analysis verified knockdown of the miRNAs (Supporting Fig. 4), and also revealed dramatic 25-fold and 10-fold increases in Smad 3 and Smad 5 mRNA levels, respectively (Fig. 3B). These data further support the conclusion that Smads 3 and 5 are bona fide targets of one or more of the miR-23b cluster miRNAs.

In the same experiment, we assayed the effect of the knockdown of the miR-23b cluster on expression of additional bile duct and hepatocyte lineage gene markers, during hepatocytic differentiation, using qRT-PCR.6, 7 We compared the expression in DMSO-treated HBC-3 cells with or without inhibitors versus expression in nondifferentiated HBC-3 cells and expressed the difference as a fold change, as we had done for the Smad assays (Fig. 3B). Three out of four of the bile duct gene markers (Aquaporin, H19, Gamma Glutamyl Transpeptidase) were found to have greatly increased expression upon knockdown of the miR-23b cluster, suggesting a more “bile duct-like” gene expression signature associated with low levels of miR-23b cluster miRNAs (Fig. 3B). Analysis of the 3′ UTRs of the “bile ductular” genes did not reveal predicted targets sites, indicating that the increased expression was a downstream effect of loss of the miR-23b cluster miRNAs.

We also analyzed the expression of hepatocyte marker genes in the inhibitor experiment. Knockdown of the miR-23b cluster had no effect on the expression of Albumin, ApoC4, Cyp2c40, HNF4α, and Transthyretin. However, the expression of PEPCK (Pkc1) showed a 10-fold reduction (Fig. 3B). Overall, this analysis showed that hepatocyte differentiation proceeded in the absence of miR-23b cluster miRNAs, whereas aberrant overexpression of bile ductular genes occurred.

Mir-23b Cluster Mimics and Smad 4 Small Interfering (si)RNA Block Bile Duct Differentiation of HBC-3 Cells.

If TGFβ/BMP family signaling is required for bile duct differentiation of HBC-3 cells, we hypothesized that overexpression of the miR-23b cluster should block bile duct formation through knockdown of Smads. To test this hypothesis, we transfected miR-23b cluster mimics into HBC-3 cells and 24 hours later plated them on a 1-mm cushion of Matrigel, and observed whether bile duct formation was blocked 24 hours later. This experiment revealed a specific and virtually complete block in tubule formation in our assay (Fig. 3C,D). Test groups with a negative control mimic (C.elegans miR-159) or mock transfection did not block tubule formation (Fig. 3C,D).

The above experiments support a mechanism in which miR-23b cluster miRNAs target Smads leading to a block in TGFβ/BMP family signaling and suppression of bile duct formation in vitro. This model predicts that if we directly knock down a key Smad, we should also be able to duplicate the block in bile duct formation. Therefore, we transfected siRNA targeting Smad 4 and measured bile duct formation in Matrigel. We again observed a specific blockage of bile duct formation by the Smad 4 siRNA (Fig. 3C,D).


This study identifies miR-23b cluster regulation of Smads as a mechanism for differential effects of TGFβ/BMP on hepatocytes versus cholangiocytes in the fetal liver. The mechanism is cell-autonomous and affects the TGFβ/BMP internal signaling pathways. The data in this report support a model illustrated in Fig. 4. According to the model, when miR-23b cluster miRNAs are high (as in hepatocytes in fetal liver) TGFβ signaling is inhibited, bile duct genes are repressed, and hepatocyte proliferation is favored. In contrast, low levels of these same miRNAs, in cholangiocytes of the ductal plates, allow Smad levels to be stable and TGFβ/BMP signaling to proceed. This in turn promotes cholangiocyte differentiation and bile duct formation.

Figure 4.

Model microRNA modulation of TGFβ family signaling in the developing liver. Mir-23b cluster microRNA modulate TGFβ family signaling by targeting Smads. When Mir-23b cluster microRNA are expressed at high levels (as in hepatocytes in the developing liver), Smads are targeted and TGFβ signaling is blocked. Low levels of these same microRNAs in the cholangiocyte progenitors of the ductal plate permit stable Smad levels and allow TGFβ family signaling to proceed. This allows the differentiation of cholangiocytes and bile duct morphogenesis to proceed.

In the mouse fetus when the TGFβ signaling gradient is perturbed by the loss of HNF6/onecut function, hepatoblasts differentiate into hybrid cells that exhibit characteristics of both hepatocytes and biliary cells.4, 10 In these animals, the ectopic activation of TGFβ signaling in hepatoblasts also produces an aberrant hybrid-like phenotype. When HBC-3 cells are induced to differentiate with DMSO, they acquire hepatocyte characteristics and the levels of Smad 3 and 5 expression are low.6 When miR-23b cluster miRNAs inhibitors were introduced into HBC-3 cells during hepatocytic differentiation there was an activation of biliary markers and a remarkable up-regulation of Smad 3 and Smad 5 expression, producing a hybrid cell gene expression profile similar in phenotype to the cells produced in onecut null livers. This result suggested that the function of the miR-23b cluster is to suppress the expression of the cholangiocyte differentiation program in hepatoblasts.

In this report we have shown that blocking a key TGFβ family signaling molecule, Smad 4, is sufficient to block bile duct morphogenesis of HBC-3 cells. When HBC-3 cells were transfected with a mix of miR-23b cluster miRNA mimics and plated in conditions that induce cholangiocyte differentiation and bile duct morphogenesis, this process was blocked, producing a phenocopy essentially equivalent to the knockdown of Smad 4 in these cells. These results indicate that the miR23b cluster miRNAs act as a molecular switch in the developing liver to regulate hepatoblast responsiveness to TGFβ and as such are important regulators of liver cell fate determination.

One report has shown that TGFβ induces expression of miRs-23a, 27a, and 24 in Huh 7 cells through a Smad-dependent mechanism.23 Whether TGFβ causes a similar induction of miRNAs in HBC-3 cells under differentiation conditions is unknown. However, this report raises the possibility that there may be a feedback loop in which miRNAs induced by TGFβ feedback to regulate its effects by targeting Smads.

Another report has provided evidence that miR-30 family miRNAs are required for complete bile duct formation.24 In this case, the miR-30 family miRNAs are expressed exclusively in the bile duct compartment of the fetal liver. Our miRNA microarray analysis also detected miR-30 as up-regulated during fetal liver development (Fig. 1B). A molecular mechanism to explain complementary actions of these miRNAs would be that miR-30 miRNAs are needed in differentiating bile ducts to repress hepatocyte genes and that miR-23b miRNAs are needed in growing hepatocytes to repress bile duct genes and repress TGFβ signaling.

A key feature of miRNAs is their extraordinary evolutionary conservation and bioinformatics programs to predict conserved and nonconserved miRNA target sites are improving. In order to obtain a glimpse of possible additional levels of miRNA regulation of the complex TGFβ/BMP signaling network, we used TargetScan 4.1 to survey predicted miR-23b cluster target for sites in genes in the TGFβ/BMP network (Supporting Table 3).

Most strikingly, we identified an abundance of candidate target sites for miR-23b cluster miRNAs in the TGFβ, BMP, and Activin Receptors. In contrast, there was a virtual complete absence of conserved miR-23b cluster sites in the inhibitory Smads and antagonists. This suggests a broader hypothesis, that there has been an evolutionary selection for miR-23b cluster target sites in positively acting components of the TGFβ signaling network and a selection against miR-23b cluster sites in inhibitory molecules of the pathways.

Finally, it should be noted that miRs 23b, 27b, and 24 are not related in sequence. Therefore, their coordinate targeting of a common set of mRNAs suggests a further evolutionary advantage. In this case, the cell only has to activate a single locus to produce a set of miRNAs that multiply target a common set of regulatory proteins. Generally, individual miRNAs have a weak “rheostat” role in knocking down gene expression25, 26 and an amplification scheme may be necessary to bring a complex and powerful signaling network such as TGFβ to a rapid halt.27 The simultaneous attack of miR-23b cluster miRNAs on several Smads plus TGFβ receptors may be such an amplification mechanism. The proposed mechanisms may have wide application to a range of TGFβ-induced phenotypes.11, 28, 29


The authors thank Dr. Marcus Landthaler (Rockefeller University) for graciously providing the psiCheck2miR16 control plasmid for dual luciferase assays, and Erin Connolly and Liang Zhu for critical reading of the article.