OstemiR: A Novel Panel of MicroRNA Biomarkers in Osteoblastic and Osteocytic Differentiation from Mesencymal Stem Cells

MicroRNAs (miRNAs) are small RNA molecules of 21–25 nucleotides that regulate cell behavior through inhibition of translation from mRNA to protein, promotion of mRNA degradation and control of gene transcription. In this study, we investigated the miRNA expression signatures of cell cultures undergoing osteoblastic and osteocytic differentiation from mesenchymal stem cells (MSC) using mouse MSC line KUSA-A1 and human MSCs. Ninety types of miRNA were quantified during osteoblastic/osteocytic differentiation in KUSA-A1 cells utilizing miRNA PCR arrays. Coincidently with mRNA induction of the osteoblastic and osteocytic markers, the expression levels of several dozen miRNAs including miR-30 family, let-7 family, miR-21, miR-16, miR-155, miR-322 and Snord85 were changed during the differentiation process. These miRNAs were predicted to recognize osteogenic differentiation-, stemness-, epinegetics-, and cell cycle-related mRNAs, and were thus designated OstemiR. Among those OstemiR, the miR-30 family was classified into miR-30b/c and miR-30a/d/e groups on the basis of expression patterns during osteogenesis as well as mature miRNA structures. In silico prediction and subsequent qRT-PCR in stable miR-30d transfectants clarified that context-dependent targeting of miR-30d on known regulators of bone formation including osteopontin/spp1, lifr, ccn2/ctgf, ccn1/cyr61, runx2, sox9 as well as novel key factors including lin28a, hnrnpa3, hspa5/grp78, eed and pcgf5. In addition, knockdown of human OstemiR miR-541 increased Osteopontin/SPP1 expression and calcification in hMSC osteoblastic differentiation, indicating that miR-541 is a negative regulator of osteoblastic differentiation. These observations indicate stage-specific roles of OstemiR especially miR-541 and the miR-30 family on novel targets in osteogenesis.


Introduction
RNA transcripts from the genome spontaneously form stem, loop and/or bulge structures, and many of these have been reported to act as primary miRNA (pri-miRNA). Drosha, an RNase III in mammalian, then cuts the pri-miRNA to create pre-miRNA that has a simpler stem-bulge structure. Exportin-5 subsequently exports the pre-miRNA to the cytoplasm [1,2]. Dicer, another RNase III, next cuts the pre-miRNA and creates double strand RNA (dsRNA). Each strand of the dsRNA is a 21-25mer nucleotide, and acts as miRNA by forming the RNAinduced silencing complex (RISC) along with the Argonaute protein [3]. The miRNAs can then hybridize partially with mRNAs, inhibit mRNA translation [4] and promote mRNA degradation [5,6]. Small RNAs have been also reported to regulate transcription in fission yeast [7], in Drosophila [8] and in human [9]. Moreover, miRNA have also been detected in exosomes suggesting a potential role in regulating gene expression in a paracrine or autocrine manner [10]. Recent studies also imply that miRNA may be involved in a feedback system of gene expression where the miR-371-373 cluster was controlled by Wnt target beta-catenin/TCF and this miRNA cluster could then inhibit DKK1, a Wnt/LRP inhibitor [11].
Osteoblasts are crucial for bone formation during development and metabolism in adult animals by coupling with osteoclasts and thus playing key roles in osteocytic differentiation. Runx2/Cbfa1, Sp7/Osterix and beta-catenin, which are the master transcription factors for osteocytic differentiation, regulate the differentiation of osteoblasts [12,13]. In addition, the CCN family proteins are known as crucial growth factors for bone formation [14]. The ccn2/ctgf gene is inducible by TGF-beta/smad and through TRENDIC [15,16], and is expressed in mesenchymal/fibroblast and vascular endothelial lineages to induce chondro-, and osteogenesis in target cells [17,18]. Recent studies have demonstrated that beta-catenin/T-cell factor (Tcf)/lymphoid enhancer factor (Lef) shares a binding site with sox9, a master transcription factor for chondrogenesis, in the ctgf/ccn2 promoter region and provides stage specific control of ctgf/ccn2 expression [19].
Osteocytes, the terminally differentiated cells derived from osteoblasts, are crucial for the mechanotransduction/mechanostress response, leading to inhibition of osteoclastic bone resorption, mineralization, and mechanical strengthening of bone while preventing osteoporosis and fracture [20,21,22]. Osteoblasts are differentiated cells derived from mesenchymal stem cells or bone marrow stromal cells [23], and can further maturate to form osteocytes, which reside individually in bone cavities called lacunae [24]. Dmp1, Fgf-23 and sclerostin are known osteocyte markers [20,25]. Dmp1, which is a secretory protein, was the first osteocyte marker to be isolated [26]. In addition, FGF-23 plays an important physiological function in mineral homeostasis and bone formation in the body by modulating phosphate excretion and vitamin D activation in the kidney [27,28]. Finally, sclerostin, the Wnt antagonist and the product of the sost gene that inhibits bone formation and plays a causal role in sclerosteosis, is transcriptionally activated by mef2c and is another osteocyte marker [29,30]. Induction of bone formation by PTH/PHT1R-induced downregulation of SOST expression is mediated through MEF2 [31].
In this study, we have carried out a series of experiments on KUSA-A1, a murine bone-marrow-derived mesenchymal stem cell line with the potential to differentiate into several different cell types, although being highly oriented towards osteocytic differentiation [32]. KUSA-A1 cells produce rich amounts of alkaline phosphatase (ALP), synthesize bglap/osteocalcin that encodes the bone gamma-carboxyglutamic acid-containing (gla) protein and mineralize very efficiently [32]. The KUSA-A1 cells present Sca-1, CD44, Ly-6C and CD140 markers on the surface [33]. More significantly, subcutaneous injection of KUSA-A1 cells in mice generates ectopic bone formation.
Previous studies have shown that miRNAs regulate osteoblastic differentiation. miRNAs control the readout of the mRNAs encoding Runx2 [33], FAK [35] and Connexin43 [36] and thus regulate osteoblastic differentiation. In our study, we have analyzed alterations in the miRNA expression signature associated with osteoblast maturation and/or osteocytic differentiation utilizing a qRT-PCR-based miRNA array. We then investigated the relationship between osteocytic differentiation and the deduced patterns of miRNA expression.

Cell culture, osteo-induction
Mouse mesenchymal stem and bone marrow stromal cell (mMSC) line KUSA-A1 and calvaria-derived osteoblastic cell line MC3T3-E1 were provided from the RIKEN cell bank. These cell lines were cultured in DMEM supplemented with 10% FBS in humidified air containing 5% CO2 at 37uC. For osteogenic induction, cells were cultured until confluency, and then Lascorbic acid phosphate, dexamethasone and beta-glycerophosphate (DS pharma) were added every other day for osteogenic differentiation.

Alizarin red staining
For alizarin red staining, cells were fixed with 4% PFA for 15 min, washed with distilled water, stained with 1% Alizarin red S staining solution for 20 min, and then washed again to remove excess dye as previously described [37].

miRNA qPCR array
Total RNA including small RNA was prepared by using Qiazol and miRNeasy mini kit (Qiagen) with DNase treatment. cDNA was synthesized with a stem-loop adaptor by using RT 2 First Strand kit (SABiosciences). miRNA qPCR array was carried out by using miFinder miRNA PCR array (SABiosciences). The data were analyzed by using miScript array data analysis (SABiosciences).

mRNA qRT-PCR (realtime PCR)
Primer pairs for mRNA were designed by utilizing the NCBI gene bank, primer 3 software and perfect PCR system (TaKaRa), in which each 59-and 39-primer are designed to recognize different exons for avoiding genomic DNA amplification if available. cDNA synthesis was carried out using mixture of random primer and oligo dT. The cDNA was diluted 4-fold in TE0.1 (10 mM Tris and 0.1 mM EDTA). A mixture of samples and step dilutions (10-fold and 100-fold) were used as a standard to clarify amplification efficiency. Realtime PCR was carried out as described previously [15]. Expression levels were normalized to that of the internal control gapdh or S29. Relative mRNA levels among different gene products were obtained utilizing D D Ct calculation. DDCt method when amplification efficiencies were similar among those. miRNA qRT-PCR Forward primers specific for each mature miRNA, which detect both human and mouse miRNAs, were designed based on sequence of mature miRNA obtained from the miRBase [38]. cDNA was synthesized by A-tailing and reverse transcription using universal primer by using qScript microRNA cDNA synthesis kit (QuantaBiosciences), the All-in-One miRNA qRT-PCR kit (Gene-Copoeia) or Mir-X TM miRNA first-strand synthesis. cDNA was diluted 4-20 fold in TE0.1 and step dilution was prepared as describe above. Realtime PCR for miRNA was carried out: 95uC for 5 min initially, then 30-50 cycles of 95uC for 5 s and 60uC for 40-60 s, and final dissociation analysis. miRNA levels were normalized to that of the internal control snord48 or snord66. miRNA target prediction TargetScan 5.1 was utilized either for in silico search for miRNAs which recognize 39-untranslated region (39-UTR) in a specific mRNA of interest or for prediction of target mRNA of specific miRNA of interest [39]. Target of miR-30 family, miR-34 family, let-7 family, miR-15/16 family (including miR-322/424), miR-21 family, miR-541/654 was predicted and selected using cut off score 20.2. Development, RNA regulation, epigenetics, transcription, protein modification-related factors were preferentially selected.

Statistics
Data was expressed as means6standard deviations, and the statistical significance of differences in mean values was assessed by Student's unpaired t tests for KUSA-A1 cells or by one-way ANOVA with Tukey post-hoc correction tests for hBMSC/MSC. Differences among the mean values were considered significant at a P of ,0.05. Realtime PCR was repeated 4 times for each samples in experiments using KUSA-A1 cells. Total RNA was isolated from independent triplicate culture of hBMSC/MSC and realtime PCR was repeated in triplicate.
Stable miRNA expression miRNA sequences of step loop part of pre-miR-21, pre-miR-30d, and pre-miR-322 were obtained from miRBase. The genomic counterparts of the miRNA stem loop were searched using Blast. Hundred base native flank sequences to both upstream and downstream of the miRNA stem loop were added for PCR amplification using following primers: miR21-F, cagctttctttcctagaattggcattaag; miR21-R, ccaaggtataagggctccaagtctcac; miR30d-F, ttattgtttgtcttttccccccaagatg; miR30d-R, ttagaagctgccagcagaagcaagcag; miR322-F, tcctccccactatccaccacaccctg; miR322-R, caggcccttggactgtgtagagtgac. The amplified fragments were cloned into pSMPUW-miR-GFP/Puro (Cell Biolabs, San Diego, CA) and sequenced to verify the contents. Recombinant lentivirus was generated by Lenti-X Lentiviral expression system (Takara Bio, Shiga, Japan) and infected to KUSA/A1 cells for 36 hours, and then the medium was replaced to normal medium. The infected cells were selected by puromycin (2 ug/mL) in 10 days for cloning of miR-21 and miR-30d stable transfectant and in 2 weeks for miR-322 stable cells.

Anti-miRNA-mediated knock down
Cells were plated in 6 well plates at a density of 26105 cells with culture medium without antibiotics. On the following day, cells were transfected with anti-miR541miR-541 (20 nM, miRvana TM miRNA inhibitor, Invitrogen) or negative control (20 nM, mirVana TM miRNA Inhibitor negative control #1, Invitrogen) using RNAiMax (Invitrogen). Cells were cultured for 2 days before total RNA extraction, or culture until confluency before osteogenic induction.

Osteoblastic and osteocytic differentiation of KUSA-A1 cells
We first investigated KUSA-A1, MC3T3 cells and hBMSC/ hMSC in tissue culture as osteoblastic/osteocytic differentiation models (Fig. 1). After adding osteogenic differentiation supplements to cells, calcium was gradually deposited in KUSA-A1 and hMSC culture until 2 weeks after the addition (Fig. 1). The levels of calcium deposition were markedly increased in KUSA-A1 cells after 2 weeks induction, while only mild levels were observed in MC3T3 cells. KUSA-A1 cells were therefore used in the subsequent studies on miRNA signature involved in osteocytic differentiation from MSCs. In order to confirm the differentiation stages, relative expression of markers for mesenchymal, osteoblastic and osteocytic cells were quantified after osteogenic induction. Mesenchymal cell markers ctgf/ccn2, cyr61/ccn1 and nov/ccn3 were induced at initial stimulation (4 hrs) but declined as cells differentiated along the osteocyte lineage (4w+: Fig. 2A). The Osteoblast marker spp1/osteopontin, increased on 4 hours (0.16 day), 2 days, 7 days and 14 days after starting osteogenic induction (Fig 2B), and the spp1 level remained at a decreased level in confluent cells without osteoinduction. Another osteoblast marker Bglap/osteocalcin was induced at 2 weeks after osteoinduction compared with non-induced control ( Fig 2B). Dmp1, an osteocyte marker, was powerfully induced at 4 hours, 7 days and 14 days after osteo-induction (Fig. 2C). Late stage osteocyte marker sost/ sclerostin was increased on day 2 and especially on day 14, while under detection limit on other conditions and time points (Fig 2C), These results suggest that this MSC differentiation model of KUSA-A1 is useful for analysis of either osteoblastic and/or osteocytic differentiation from MSCs.

Osteoblastic differentiation of hMSC/hBMSC
In order to firstly establish osteoblastic differentiation from hMSC, osteo-induction, calcium staining and quantification of mRNA levels of marker genes were carried out. Calcium was accumulated on hMSC upon osteo-induction around 1 and 2 weeks after the induction gradually as shown by alizaline red S staining (Fig 1), while non-osteo-induced cells were not stained. ALP and SPP1/Osteopontin (OPN) mRNA were quantified in realtime qRT-PCR during osteo-induction. Both ALP and OPN/ SPP1 were upregulated in two independent experiments with different peak timings. In the first sample, both ALP and OPN/ SPP1 peaked on day14, while in the 2nd sample, both ALP and OPN/SPP1 peaked on day 7 and decreased thereafter (Fig 2D), indicating that differentiation is slower in sample #1, while faster in sample #2 relatively each other. These data suggest that hMSC/hBMSC differentiated into osteoblasts after the induction.

Overview of the miRNA array approach
To investigate novel miRNAs that could potentially regulate osteoblastic/osteocytic differentiation and stemness (OstemiR), a screening with a qRT-PCR-based miRNA array was utilized. We first obtained an overview of changes in overall miRNA levels. Total miRNA expression levels after osteo-induction for 2 weeks were significantly lower than observed in other conditions (Fig. 3A). All the tested miRNAs were then analyzed by clustergram and the findings indicated that most miRNAs were expressed at a reduced level in mature osteocytes (Fig. 3B). Based on the results of the clustergram, miRNAs were then categorized into eight groups ( Table 1). The most abundant group was the miRNAs that increased in level after prolonged culture without osteo-induction (2w2). miRNA downregulated by two weeks osteo-induction included members of the let-7 and miR-30 families (miR-30a/d/e) ( Table 1). Thus, the total amount of miRNA is reduced in osteocytes compared to osteoblast precursors, and release of translational repression may be involved in induction of osteocytespecific factors and the osteocytic phenotype.
For a better understanding of OstemiR, the miRNAs involved in osteocyte differentiation, heat maps were created to compare miRNA expression levels between experimental conditions of induction (4h and 2w+/2 differentiation stimulus). The expression level of miR-30d was higher in the 4h+, 2w-and 2w+ condition than in the 4h2 condition ( Fig. 4A-F). Indeed, by a single stimulation for osteocytic differentiation, not only the miR-30d but also miR-155 was induced (Fig. 4A, D). The expression level of miR-16 was decreased by the single stimulation (Fig. 4A, D). After repeated osteo-induction (2w+), miR-30d and miR-30c were induced, and the expression levels of miR-503, miR-322 and miR-125b-3p were the most powerfully repressed (Fig. 4B, E). miR-30d and miR150 as well as other miRNAs were induced by long-term culture for 2 weeks in the absence of differentiation stimulus, while miR-503 and miR-744 were reduced by the long-term culture (Fig. 4C, F).

Standardized quantification of miRNA expression
In order to establish an internal control for miRNA quantification in the KUSA-A1 cell during the conditions of osteocytic differentiation, Snord85, Snord66, Rnu6, miRTC and PPC expression patterns were examined from the results of the miRNA array (Fig. 5A). Among these candidate control RNAs, the expression patterns of Snord66 and PPC were found to be the most stable in all-experimental conditions, while Rnu6 was not stably expressed, and Snord85 levels became elevated after 2 weeks of osteo-induction. Therefore, Snord66 or PPC were adopted as internal controls for miRNA quantification in the KUSA-A1 osteocytic differentiation system in later experiments. In addition, Snord85 was established as a novel marker of osteocytic differentiation.
We next attempted to confirm the reliability of the results of the miRNA array analysis, designed specific primers for each miRNA, and quantified miRNA levels using qRT-PCR. Expression of both miR-21 and miR-155 was stimulated by the initial differentiation stimulus (4h+) although expression was suppressed at the terminal differentiation stage (2w(+) as indicated by both qRT-PCR and miRNA array (Fig. 5B, C, D). miR-322 was investigated by the same methodology, and shown to be decreased at 2 weeks after osteo-induction in both the qRT-PCR and miRNA array (Fig. 5E, F). miR-16 was also repressed by osteogenic stimulus as quantified by either miRNA array or qRT-PCR (Fig. 5G, H). miRNA array analysis showed that miR-30d was induced by single stimulation (4h+), repeated stimulation (2w+) and prolonged culture in the absence of stimulation (2w2). However, miR-30d was increased only by single stimulation as indicated by qRT-PCR (Fig. 5I, J). Overall, these results indicate that while the qRT-PCR-based miRNA array is useful and reliable for screening, individual qRT-PCR is more reliable and flexible for the assay of single-peaked amplification of cDNA in detailed kinetic analysis. The functions and targets for OstemiRs, whose expression levels were significantly changed during osteocytic differentiation, were predicted by analysis of database (Table 2). These OstemiRs were predicted to recognize mRNAs encoding a number of key transcription factors for osteocytic differentiation. These included: (runx2, sox9, sox5, smad family, msx1, tcf, jun and fos), transcription factors for pluripotency (sox2, klf5/klf, mycn), EMT-related factors (snail, zeb2, claudin-1/2), essential growth factors for osteocytic differentiation and their receptors (wnt/lrp, fgf/fgfr, bmp/gdf/bmpr/ tgfbr/actvr, ihh, lifr, notch, igf/igfr, insr), matrix-related molecules (reck, timp), neuronal markers (gfap), epigenetic factors (mll, jmj, jhdm, hdac/sirt1, tet, jarid, mbd, ncoa1), cell cycle proteins (cyclin/cdk) and  heat shock proteins (hsp, dnaj) ( Table 2). The cohort of miRNA, which was upregulated during osteoblast maturation, including miR-30d, miR-155, miR-21 and miR-16, constitutes a marker of osteocytic differentiation and these miRNA may possibly repress stemness maintenance in osteoblasts. Both miR-34c and miR-16, which increased at 2w+, the stage of osteocytes, are possibly osteocyte markers and repressors of osteoblast-maintaining genes.

Prediction of miRNAs that target osteo-regulators
By the prediction of miRNAs that target osteo-regulators, miRNA recognition sites in the 39-UTR regions in the target mRNAs were predicted and classified into groups that were: (a) conserved among all vertebrates, (b) conserved only among mammalian species or (c) poorly conserved. The conserved miRNA recognition sites among all vertebrates (a) and only among mammalians (b) are described below. A let-7/miR-98 recognition site was predicted in the 39-UTR region of dmp1 mRNA with broad conservation among vertebrates (Fig. S1). Since all members of the let-7 family are downregulated during osteocytic differentiation (Table 1) it was predicted that the downregulation of the let-7 family could be associated with an eventual repression of the dmp1 gene. In addition, miR-30d was induced by osteo-induction (Fig. 5J), and miR-30 family recognition sites were found in the 39-UTR regions of the runx2 and nov/ ccn3 mRNAs (Fig. S2, S3). Moreover, the miR-30 family was predicted to recognize sox9, lrp6, smad2, smad1, notch1, bdnf and a number of epigenetic factors (Table 2). These findings suggest that members of the miR-30 family could play an essential role in  osteocytic differentiation. Moreover, among miRNAs that were predicted to recognize the 39-UTR of ctgf/ccn2, an osteo2, chondrogenesis factor, and miR-18ab and miR-19 are also OstemiR, which increased during osteocytic differentiation (Fig.  S4, Table 1).

miR-30 expression pattern during KUSA-A1 MSC osteocytogenesis
In the miRNA PCR array, miR-30d showed an increased expression level in osteocytogenesis of KUSA-A1. Since miR-30 family members are homologous ( Fig 6A) and possibly share targets, we further investigated the miR-30 family expression patterns at four time points with or without osteo-induction. As a result, miR-30a, miR-30c and miR-30d were highly expressed compared with miR-30b or miR-30e (Fig. 6B). All the miR-30 members once reduced during osteoblastic differentiation stage on day 2 and day 7. Among those members, miR-30a/d/e were increased on day 14 around a late osteocytic stage (Fig 6A). These data suggest that miR-30 members could be repressing targets at the MSC and osteocytic stages, while repression on target mRNA may be relieved during the intermediate osteoblastic stage.
Searching 39-UTR of putative target mRNA, targeting sequences which can make base pairing with 59 seed sequences of miR-30 were found in the 39-UTR of lifr, eed, pcgf5 and sirt1 utilizing TargetScan (Fig 7B). One miR-30 targeting sequence in the 39-UTR of ctgf/ccn2 has been reported. In addition, two putative miR-30 targeting sites on spp1/osteopontin were found. Matching around the 39 part and intermediate part of miR-30 were tested to those targets. Not only 59 seed sequences but also 39 sequences of miR-30d matched to the lifr, eed and sirt1 39-UTR. On the other hands, miR-30b/c 59 seed as well as 39 part was matched with 39-UTR sequences of spp1/opn, pcgf5, hspa5/grp78 and ctgf/ccn2. These in silico analyses suggested putative shared and distinctive target mRNA recognition by miR-30 family, the groups of miR-30a/d/e and miR-30b/c.

miR-30 targeting in mMSC line
In order to clarify the function of miR-30d on target mRNAs, qRT-PCR was carried out in stable miR-30d transfected KUSA-A1 and in control vector transfectant. In a result, hnrnpa3 variant B level in proliferating/sparse miR-30d tranfectant was around 50% lower than that in the vector transfected control (Fig 8A), while no significant change in confluent cells (Fig 8B), indicating context dependent repression of hnrnpa3 vB by miR-30d. Lin28a mRNA level in confluent miR-30d tranfectant was around 50% lower than that in the vector transfected control (Fig 8A, left), while around 50% higher in proliferating cells (Fig 8A, right), indicating context dependence as well. Ccn2/ctgf and ccn1/cyr61 mRNA levels in confluent miR-30d cells were lower than those in the control (Fig 8B), while these gene product levels in proliferating miR-30d cells were higher than those in the control (Fig 8A). Runx2 and sox9 mRNA level in miR-30d transfectants were higher than that in the control (Fig 8AB). Hspa5/grp78, lifr, eed, opn/spp1 and pcgf5 mRNA levels in miR-30 transfected cells were 20-30% lower than those in control cells in both proliferating and confluent cells (Fig 8AB), indicating direct repression of mRNA stability. These data suggested targets of miR-30d and context-dependent effect of
Furthermore, mRNA expression pattern of miR-30d targets during osteogenesis of KUSA cells were quantified. Lin28a mRNA was remarkably induced during the osteogenesis, and reached around 200-fold on day 14 (osteocytic stage) compared with the day 0 (Fig 9C), indicating its essential role and possible release of repression during osteogenesis. Pcgf5 mRNA levels decreased, and reached to the half on day 14 compared with the level on day 0 ( Fig 9C). Hnrnpa3 variant c, eed and hspa5 mRNA were immediately induced by the osteo-inductive stimulation by 13fold, 8.1-fold and 4.4-fold respectively, and thereafter kept around half levels of the maximum levels ( Fig 9D). These immediate early induction followed by quick attenuation patterns were shared with those of CCN gene family shown in Fig 2A, indicating these 6 kinds of transcripts are under the control of same factors and the miR-30 family. Distinctively from the variant C, hnrnpa3 variant B, the major variant, mRNA was gradually reduced upon osteocytogenesis (Fig 9F), indicating osteo-induction controls splicing of hnRNPA3 mRNA. Both runx2 and lifr mRNA were immediately reduced at 4 hours after the stimulation, and thereafter recovered in 2 days (Fig 9E). Then lifr mRNA level was reduced to the day 14, while runx2 mRNA did not decrease and kept the expression level comparable to the initial level. Together with the data of expression patterns in Fig 9 and Fig 2, miR-30 targets were classified into several groups; immediate induction followed by rapid attenuation group (ccn1/2/3, hnrnpa3 vC, eed, hspa5/grp78), immediate reduction and rapid recovery group (runx2 and lifr), the constant induction group (lin28a and opn/spp1) and the constant reduction group (pcgf5 and hnrnpa3 vB).

Human miR-541 and miR-155 function and expression pattern in hMSC/hMBSC osteoblastic differentiation
In order to investigate OstemiR expression in hMSC osteoblastic differentiation, qRT-PCR was carried out. miR-541 and miR-155 were induced in 4 hours after the osteo-stimulation to KUSA-A1, but not on the day 14 ( Fig 3B, Fig 4). In addition, miR-155 and miR-541 were gradually reduced during osteoblast differentiation of hMSC (Fig 1H), while miR-541 level was higher in osteo-induced cells on day 4 and day 14 compared with the control (Fig 10A). In attempt to clarify a function of miR-541 and miR-155 during osteogenesis of hMSC, we transiently transfected these cells with antagonists/anti-miR targeting these two types of miRNAs. The efficiency of anti-miR-155 or anti-miR-541 knockdown was of approximately 20-40% compared with the control siRNA transfections (Fig 10B). Despite this apparently low efficiency, a significant increase by 1.8-fold in the osteoblastic marker OPN/SPP1 mRNA level was observed by miR-541 knock down (Fig 10C). ALP mRNA level in anti-miR-541 transfectants was higher than that in the control. In accordance, calcium deposition on the anti-miR541-treated cells was more rich than those of controls in a result of alizarin red S staining on day 7, comparing center regions in each wells (Fig 10D). There were no significant changes in ALP staining of upon miR-155 or miR-541 knock down. Taken together, these data indicate that miR-541 is a negative regulator of osteoblast differentiation of hMSC.

Tuning model of osteogenic factors by OstemiRs during MSC osteogenesis
Together with these results and data interpretations, we propose the tuning model of canonical and novel osteogenic factors by the OstemiRs including miR-30 family and miR-541. In this model, miR-30b/c represses hspa5, eed, ccn1/2/3, hnrnpa3 vC (Fig 11A), opn/spp1, lin28a (Fig 11B), lifr and runx2 (Fig 11C) at the MSC stage. This repression is released during osteogenesis upon reduction of miR-30b/c, a change especially significantly in increase in opn/spp1, lin28a (Fig 11B), lifr and runx2 (Fig 11C). Osteo-inductive stimulation transiently induces hspa5, eed, ccn1/2/3 and hnrnpa3 vC, but thereafter those transcripts are attenuated by miR-30b/c at the early stage and by miR-30a/d/e during the osteocytic stage (Fig 11A). miR-30a/d/e targets pcgf5 and hnrnpa3 vB as well (Fig 11D). Human OPN/SPP1 in hBMSC/MSC is attenuated by miR-541. Osteogenic differentiation of an mMSC line and hMSCs The speed of mineralization was different among the cell types tested: KUSA.hMSC.MC3T3 as shown in Fig 1. In addition, Comparing expression patterns of opn/spp1 between KUSA and hMSC/hBMSC, it was suggested that KUSA-A1 is most committed to osteoblast lineage, or has a faster differentiation process than hBMSC, which would contain more undifferentiated cells and differentiation potential to other lineage. Thus, further understanding and sorting MSC and BMSC are needed for clinical application for bone regeneration, because those populations have potentials of aging/senescence, adipogenic and chondrogenic differentiation as well.
The difference in speed or stage of differentiation may result in difference in expression signature of miRNA, e.g., mouse miR-30d was induced on 4 h or 14 days after the osteo-induction compared with the control, while human miR-30d showed waving induction and reduction during osteogenesis (Fig S2). Besides, mouse miR-541 was strongly induced in 4 hours of initial induction, and then reduced in the later stage (Fig 3B and Fig 4), while in hMSC, miR-541 was gradually reduced in long culture, suggesting some role for miR-541 in osteogenesis. In fact, knockdown of miR-541 upregulated OPN/SPP1 and mineralization. However, a direct target site of miR-541 on OPN/SPP1 has not been identified yet, indicating an unknown indirect mechanism. Putative roles of novel key factors in osteogenesis: Lin28A, hnRNPA3, Eed and Pcgf5 As targets of miR-30, we found novel key factors in osteogenesis including Lin28, hnRNPA3, Eed, Pcgf5 and HspA5/Grp78. Here we discuss about roles of these factors in bone formation as well as canonical osteogenic factors including Runx2, LifR, Opn/Spp1 and the CCN family, which are the targets of miR-30d.
Heterogeneous nuclear ribonucleoprotein (hnRNP) is another crucial RNA regulator in RNA nuclear export and splicing. hnRNPA1 directly associates with miR-18a stem-loop as well as pri-miR-17/18a/19a, and then export pri-miR-17/18a/19a in the exportin-independent manner [42]. Thereafter, hnRNPA1 promotes cleavage of the miRNA by Drosha and Dicer. Furthermore, hnRNPA1 inhibits processing of pri-let-7a by competing to KSRP, which promote Drosha processing let-7a [43]. Nuclear export of hnRNPA1 is promoted by phosphorylation in cells stressed by osmic shock [44]. The role of hnRNPA3 is presumably similar to that in hnRNPA1. In our study, only hnrnpa3 variant C was induced upon osteo-induction, but not variant B, and context-dependent effect of miR-30d on hnRNPA3 variants was suggested. Thus, context-or osteoinduction-dependent hnRNP variants presumably control mRNA splicing as well as the processing of miRNAs.
EED, named after embryonic ectoderm development, is another novel target of miR-30. Eed is one of the main components of polycomb repressive complex 2 (PRC2), which induces histone H3K9 and K27 methylation leading to gene repression. Recent studies revealed crucial roles of Eed in ESC self-renewal through interaction with Stat3, Oct-3/4 and Sox2 [45,46]. Eed was also induced during osteo-induction in our study. However, sox2, an ESC marker as well as an iPSC inducer, was undetected in KUSA cells. Therefore, MSC was distinguished from ESC or iPSC in this point, although our data suggested that Eed-mediated silencing through histone methylation has still some role in osteogenesis. PCGF5, a polycomb group ring finger protein is involved in gene repression as well. This reduction of polycomb factor PCGF5 during the osteogenesis may release expression of osteocytogenic factors and miRNAs in an epigenetic manner. miR-30 controls expression of LifR and Runx2, the known regulators for osteoblasts LIFR, the leukemia inhibitory factor receptor, is essential in ESC self-renewal and in bone marrow stromal osteoblast differentiation. Under physiological conditions, LIF is produced from articular and growth plate cartilage, promoting proliferation and differentiation of chondrocytes [47]. LIF from chondrocytes influences vascularization during bone growth through promotion of chondroclasts and osteoclasts. LIF produced in the marrow, and by osteoblasts on the bone surface acts on stromal precursors to inhibit adipogenesis, and stimulates osteoblasts on bone remodeling surfaces. LIF also acts on osteocytes to inhibit production of SOST/sclerostin. In pathology, LIF released by inflamed synovium contributes to cartilage destruction by altering MMP production, and is likely to induce osteoclastic bone erosions in rheumatoid arthritis. As observed in Fig 11C, suppression of lifr expression by miR-30 may control osteoblast and osteocyte differentiation leading to attenuation of Lif/LifR/Jak-Stat signal.

miR-30 controls CCN family gene expression during MSC osteogenesis
Physiological production of CCN2/CTGF is more abundant from chondrocytes in cartilage than those in other tissues, while CCN1/2/3, the prototype members of CCN family, control both chondrocytic and osteoblastic differentiation [57,58). The transient induction of ccn1/2/3 observed in this study could be necessary and essential in initiation of MSC differentiation. A cisacting element of structure-anchored post-transcriptional repression (CAESAR) was identified in the 39-UTR of ccn2 in human [59,60,61], then similar elements was found in mouse [60] and in chicken [62,63]. In addition, miR-26ab/1297, miR-132/212, miR-133, miR-18ab and miR-19, members of OstemiR, were predicted to recognize the 39-UTR of ccn2/ctgf. Among this group, miR-18 has been reported to control ctgf/ccn2 gene expression in chondrocytic cells [64]. Dexamethasone has been shown to be a strong inducer of ctgf/ccn2 [65,66]. In our study, ctgf/ccn2 was immediately triggered after the osteo-induction treatment that included dexamethasone, and then rapidly repressed in 2 days. Therefore, immediate induction and subsequent rapid repression of ctgf/ccn2 could be controlled by fluctuations in these miRNAs including the miR-30 family. In myocardial cells CTGF/CCN2 is regulated by miR-133 and miR-30c [67] and the 39-UTR of ctgf and miR-30c are basepairing by 9 bases at 59 seed of miR-30c and 11 bases at 39 part including one of each GU non-Watson-Crick base pairing (Fig 7A). In a result of direct analysis of ctgf/ccn2 mRNA, miR-30d reduced ctgf/ccn2 mRNA levels in confluent KUSA-A1, while not in proliferating cells (Fig 8), indicating that miR-30d attenuate basal ctgf/ccn2 level in idling MSCs.

Possible application of OstemiR for osteoporosis, periodontitis and peri-implantitis
The OstemiR revealed in this study will be further clarified in studies aimed at understanding and controlling distinct regulation of MSC differentiation into not only osteocyte but also adipocytes, chondrocytes, myoblasts and tendon/ligament cells. In fact, runx2 as well as sox9 a master transcription factor for chondrogenesis was upregulated in mRNA level by miR-30d, indicating miR-30 could direct differentiation of MSC. We focused on the miR-30 family and miR-541 in this study, while still further analyzing roles of OstemiR in MSC differentiation. Kawashima et al. analysed the gene expression signature of KUSA-A1 cells using cDNA microarray [33]. Comparing the protein coding gene and miRNA expression signatures in future investigations could potentially clarify target mRNAs of OstemiR. Combination of several OstemiR or anti-OstemiR might regulate firmer cell differentiation. miRNAs have the advantage of being smaller molecules than proteins or antibodies and more easily synthesized or quantified compared to the other molecules. Moreover, recent studies have demonstrated that many miRNAs can be secreted [10]. Application of OstemiRs could be useful for treatment of bone-related diseases such as periodontitis, peri-implantitis and osteoporosis as well as diagnosis.  (DOCX)