Mediator MED23 cooperates with RUNX2 to drive osteoblast differentiation and bone development

How lineage specifiers are regulated during development is an outstanding question, and the molecular regulation of osteogenic factor RUNX2 remains to be fully understood. Here we report that the Mediator subunit MED23 cooperates with RUNX2 to regulate osteoblast differentiation and bone development. Med23 deletion in mesenchymal stem cells or osteoblast precursors results in multiple bone defects similar to those observed in Runx2+/− mice. In vitro, Med23-deficient progenitor cells are refractory to osteoblast differentiation, and Med23 deficiency reduces Runx2-target gene activity without changing Runx2 expression. Mechanistically, MED23 binds to RUNX2 and modulates its transcriptional activity. Moreover, Med23 deficiency in osteoprogenitor cells exacerbates the skeletal abnormalities observed in Runx2+/− mice. Collectively, our results establish a genetic and physical interaction between RUNX2 and MED23, suggesting that MED23 constitutes a molecular node in the regulatory network of anabolic bone formation and related diseases.

M esenchymal stem cells (MSCs) are multipotent progenitors that can self-renew and differentiate into several distinct cell lineages, including osteoblasts, chondrocytes and adipocytes 1,2 . Their differential potentials and intrinsic properties make MSCs of great interest for applications in a variety of cell-based therapies [3][4][5][6] . In the past few decades, several master regulators that specify the various cell fates associated with MSCs have been identified: for example, Runx2 triggers MSCs to differentiate into osteoblasts 7,8 . The function of Runx2 in osteoblast differentiation was initially demonstrated by observations that Runx2 À / À mice lack both mature osteoblasts and a mineralized skeleton [8][9][10] . Although Runx2 þ / À mice do undergo skeletal mineralization, the process is impaired, with delayed closure of the fontanelles and clavicular hypoplasia 10 , resembling the cleidocranial dysplasia (CCD) phenotype that arises from Runx2 haploinsufficiency in humans 9 . The skeletal pathologies in humans and mice highlight the importance of the precise regulation of Runx2 activity during osteogenesis.
Apart from these key transcription factors, several cofactors have been found to be involved in the fine tuning of cell-fate determination [11][12][13] . One such cofactor is the transcriptional complex Mediator, which is a multiprotein complex that has been evolutionarily conserved from yeast to metazoans. The Mediator complex is recruited by transcription factors in response to developmental or environmental signals. Through direct protein interactions, Mediator links transcription factors to the RNA polymerase II (Pol II) machinery and modulates the transcriptional output 14 .
Our previous research demonstrated that the Mediator MED23 subunit acts as a molecular modulator to control the balance between adipogenesis and myogenesis 15,16 . However, it remained unknown whether MED23 is also involved in the development of osteoblasts, which, like adipocytes and myocytes, are also derived from mesenchymal progenitors. In this study, we provide evidence of a role for MED23 in osteogenesis. Mice deficient in Med23 in MSCs or preosteoblasts display defective bone formation and impaired osteoblast differentiation, resembling the skeletal phenotype of Runx2 þ / À mice. Further molecular studies demonstrate that MED23 cooperatively interacts with RUNX2 in the regulation of osteogenesis. Our results reveal that MED23 is an important regulator of osteogenesis and is essential for the transcriptional activity of RUNX2.

Results
Med23 deficiency in MSCs impaired bone ossification. To investigate the role of Med23 in the commitment of MSCs to an osteoblast fate, we generated a Med23 conditional knockout mouse model by crossing Med23 fl/fl mice 17 with Prx1-Cre mice (Fig. 1a). The latter is a transgenic line wherein Cre activity is largely confined to ventral and craniofacial mesenchyme 18,19 . Western blot assays verified that MED23 was largely abrogated in bone marrow-derived MSCs (Fig. 1b). MSC-specific Med23 deletion had no effect on survival or fertility. However, Med23 fl/fl ; Prx1-Cre mice (Med23 MSC À / À ) were shorter and underweight relative to their control littermates. Furthermore, as the Med23 MSC À / À mice aged, they developed a marked dwarfism phenotype with short legs that was independent of sex ( Fig. 1c,d). The heterozygous mice (Med23 MSC þ / À ) appear normal ( Supplementary Fig. 1d). Consistent with this observation, the level of MED23 was barely reduced in bone tissue samples from Med23 MSC þ / À mice ( Supplementary Fig. 1c). Notably, the membranous ossification of the skull was impaired in Med23 MSC À / À newborns compared with control littermates. The anterior and posterior fontanelles were enlarged and the cranial sutures were wider in Med23 MSC À / À mice. Defective osteogenesis was not limited to cranial tissues, as the clavicles became thinner and shorter, and the sternum xiphoid was also hypoplastic (Fig. 1e). And these defects occurred as early as E15.5 when the bone mass is formed ( Supplementary Fig. 1a). By contrast, a skeletal analysis of Med23 MSC þ / À mice confirmed that they were phenotypically normal relative to control littermates ( Supplementary Fig. 1b).
To further determine the function of MED23 in the skeletal system, we utilized micro-quantitative computed tomography micro computed tomography (m-CT) to compare skeletal elements isolated from Med23 MSC À / À mice with corresponding elements from control littermates. As shown in Fig. 1f, 6-day-old Med23 MSC À / À mice displayed hypomineralization of the calvaria compared with control mice. In addition, we found that 1-monthold Med23 MSC À / À mice were osteopenic, with reduced bone mineral density (BMD) and bone volume per tissue volume) (BV/TV) in the femoral trabecular bone relative to age-matched control littermates. Further analysis showed that the reduced trabecular number (Tb.N) of Med23 MSC À / À mice was accompanied by decreased trabecular thickness (Tb.Th) and a reduction in the Tb.N compared with those of control mice (Fig. 1g,h). Consistent with the decreased BMD in Med23 MSC À / À mice, an ELISA assay of PINP (N-terminal propeptide of type I procollagen), a marker of bone formation, revealed a reduced bone formation rate in Med23 MSC À / À mice (Fig. 1i). In adults, bone is continuously remodelled by bone-resorbing osteoclasts and bone-forming osteoblasts 20 . We performed a histological analysis and in vitro osteoclastogenesis experiment to exclude the possibility that the decreased bone density in the long bones of Med23 MSC À / À mice could be attributed to increased osteoclast differentiation and/or activity. We found that osteoclast activity was comparable between Med23 MSC À / À mice and their control littermates ( Supplementary Fig. 2a,b). Moreover, the expression of osteoclastic genes did not change either ( Supplementary  Fig. 2c). Taken together, these results suggest an important role for Med 23 in bone formation.
Ablation of Med23 in preosteoblasts reduced bone formation. To further determine whether the abnormal osteogenesis in Med23 MSC À / À mice results from a primary defect in osteoblast development, we generated an osteoblast-specific Med23-deleted mouse model (Med23 ob À / À mice) by crossing Med23 fl/fl mice with osterix (Osx)-Cre mice, a line in which Cre expression is primarily restricted to osteoblast precursors 21 (Fig. 2a). Western blot assays showed that MED23 decreased in the bone of both Med23 ob þ / À and Med23 ob À / À mice ( Supplementary Fig. 3b). Consistent with decline in protein of MED23, histological analysis by alcian blue and alizarin red (ARS) staining showed defects in the skeleton occurring in both Med23 ob þ / À and Med23 ob À / À newborns, and more severe in the latter, including abnormality in calvarial and clavicle ossification ( Supplementary Fig. 3a). However, Osx-Cre transgenic mice have been found to manifest defects in bone phenotype, such as to delay bone mineralization and develop scapula calluses 22,23 . To exclude such effect of Osx-Cre, we made a comparison between Med23 ob À / À mice and Med23 ob þ / À control littermates that were both in the context of Osx-Cre. Again, Med23ob À / À mice developed runty and underweight compared with age-matched control littermates (Fig. 2b,c). Mice with Med23 deficiency in osteoblast reproduced the phenotype of Med23 MSC À / À mice, including impaired membranous ossification of calvarial bones and dysplasia of the clavicles, although the sternum appeared normal. In the Med23 ob À / À mice, the bone was obviously porous; this feature was more evident in the bones of the skull (Fig. 2d,e). mCT analysis further confirmed the osteopenic phenotype of Med23 ob À / À mice (Fig. 2f,g). Likewise, relatively lower levels of serum PINP indicated a decreased bone formation rate in Med23 ob À / À mice (Fig. 2h). Again, in vivo and in vitro osteoclastogenesis showed comparable levels of osteoclasts between Med23 ob À / À mice and control littermates ( Supplementary Fig. 4a-c). Hence, the Med23 ob À / À mice recapitulate the defects observed in Med23 MSC À / À mice with striking fidelity, supporting the conclusion that Med23 is necessary to the differentiation and function of committed osteoblast precursors.
Med23 is required for the osteoblast differentiation of MSCs. Next, we asked whether Med23 deficiency affects the osteogenic potential of MSCs in a cell-autonomous manner. Primary MSCs were isolated from the bone marrow of control and Med23 MSC À / À littermates, and the deficiency of Med23 in the MSCs from Med23 MSC À / À mice was confirmed by western blot assay (Fig. 3a). Med23 deficiency in MSCs does not appear to alter the cell viability or growth rate, indicated by comparable numbers of colony-forming unit fibroblasts (c.f.u.-F) (Fig. 3b). However, Med23 À / À MSCs displayed markedly decreased alkaline phosphatase (ALP) activity and mineralization (Fig. 3c,d). In addition, the mRNA levels of osteogenic genes, such as osteocalcin (Ocn), were significantly downregulated in Med23 MSC À / À cells (Fig. 3e), while Runx2, a key regulator in bone development, did not change during osteogenesis To exclude developmental differences between control and Med23 À / À primary MSCs, we infected wild-type primary bone marrow MSCs with retroviruses expressing either shRNAs targeting Med23 or control shRNA. Consistent with the differentiation difference between WT and Med23 KO primary MSCs, shRNA-mediated acute knockdown of Med23 in MSCs also resulted in decreased osteogenesis, as indicated by ALP and ARS staining ( Supplementary Fig. 5a). Real-time PCR analysis showed that Med23 knockdown reduced the expression of osteogenic marker genes such as Osx, Alp, collagen type I (Col1a1) and Ocn. However, noticeably, the expression of Runx2, the master regulator for osteoblast differentiation, was unaffected by Med23 deficiency (Supplementary Fig. 5b).
Collectively, these results showed that Med23 is required for osteoblast differentiation. The key osteogenic genes downregulated in Med23-deficient cells are known targets of Runx2, suggesting that MED23 might regulate osteoblast differentiation via Runx2.
Med23 regulates the gene network of bone development. Despite our observation that the expression of Runx2, a master transcription factor for osteoblast differentiation, was not affected by Med23 deletion during osteogenesis, Med23 MSC À / À mice appeared to phenocopy the skeletal defects of Runx2 þ / À mice, including delayed closure of the fontanelles, clavicular hypoplasia and decreased bone density 10 , suggesting that Med23 and Runx2 share a functional mechanism. To compare the effects of Med23 and Runx2 on global gene expression patterns, we performed RNA-seq with total RNA extracted from the calvaria of control and Med23 MSC À / À littermates, as well as WT and Runx2 þ / À littermates. Globally, Med23 MSC À / À versus control and Runx2 þ / À versus WT revealed a significantly positive correlation in global gene expression (Fig. 4a). We next found that downregulated genes in Med23 MSC À / À set presented in a similar pattern in Runx2 þ / À set (Fig. 4b). In more details, 438 genes were downregulated 41.5-fold in the Med23 MSC À / À set compared with the corresponding control, and 504 genes were downregulated 41.5-fold in the Runx2 þ / À set compared with the control and a total of 124 genes overlapped between the 2 groups (Fig. 4c). These data further implied that these two molecules may function cooperatively. We then performed a gene ontology (GO) analysis of the set of 1.5-fold significantly downregulated genes from the Med23 MSC À / À sample. The results showed that these genes were enriched for associations with bone development and ossification (Fig. 4d). We verified the expression of osteoblast-specific genes in the calvaria from control and Med23 MSC À / À newborns by real-time PCR and found that osteogenic genes such as Osx, Alp, Col1a1 and Ocn were significantly reduced in Med23 MSC À / À mice compared with controls, while Runx2 and Atf4, another key factor involved in osteogenesis, were unchanged (Fig. 4e). Consistently, Von Kossa staining and immunohistochemical analyses of bone tissue sections from E16.5 control and Med23 MSC À / À embryos further confirmed the defects in bone formation and impaired osteoblast differentiation in vivo (Fig. 4f,g). Similarly, defects in ossification and decrease in osteogenic gene expression were observed in Med23 ob À / À mice ( Supplementary Fig. 3c-e). Both the in vitro and in vivo experiments thus showed that Med23 deletion reduced the expression of multiple Runx2-target genes while not affecting the expression of Runx2 itself. These observations implied that MED23 may participate in osteoblast differentiation and bone development by controlling RUNX2 activity.
Med23 modulates Runx2 activity as a cofactor. We next set out to investigate whether MED23 controls the transcriptional activity of Runx2. To this end, we infected murine C3H10T1/2 embryonic fibroblasts with retroviruses expressing control or Med23 knockdown shRNAs and then transfected the stable cells with a luciferase reporter driven by the Ocn promoter (OC1050-Luc), which is a prototypical Runx2-target gene promoter that is widely used as a tool to study the molecular regulation of osteoblast development 24 . In the absence of MED23, RUNX2-dependent activation of the Ocn promoter was significantly inhibited (Fig. 5a). The Ocn promoter harbours two binding sites: OSE2 and OSE1, which are specifically bound by Runx2 and ATF4, respectively 25 . To discriminate the effects of Med23 deficiency on the two key factors, we performed the luciferase reporter assay with either 6XOSE2-Luc or 6XOSE1-Luc in control and Med23 shRNA knockdown cells. Med23 deficiency specifically impaired the Runx2driven activation of 6XOSE2-Luc but did not affect the ATF4-driven activation of 6XOSE1-Luc, suggesting a specific connection between Med23 and Runx2 (Fig. 5b,c). To further verify the specific regulation of RUNX2 activity by MED23, we performed a titration experiment with low and high doses of MED23 co-transfected with the OC1050-Luc reporter, which showed that RUNX2 activity was repressed gradually with increasing dose of MED23 (Fig. 5d). The repressive effect on RUNX2 activity by overexpression of MED23 was also confirmed by 6XOSE2-Luc reporter system ( Supplementary Fig. 7a). By contrast, overexpression of MED23 did not affect ATF4 activity ( Fig. 5d and Supplementary Fig. 7b). RUNX2 activity was quenched by high level overexpression of MED23 probably in that the exogenous MED23 may compete with endogenous MED23 within the Mediator complex to prevent the transcription factor Runx2 from recruiting the Mediator complex.
Cooperative activation of the Ocn promoter by MED23 and RUNX2 led us to examine the physical interaction between the two molecules. We transfected 293T cells with tagged Runx2 and Med23 and performed a co-immunoprecipitation (co-IP) assay, which revealed that MED23 physically associated with RUNX2 (Fig. 5e). Consistent with these results, endogenous RUNX2 and MED23 interacted in differentiated MC3T3E1 osteoblastic cells (Fig. 5f). The interaction with MED23 was mediated via the Runt and PST domains of RUNX2 ( Supplementary Fig. 8a,b). GST pull-down assay implied that MED23/RUNX2 interaction was likely direct (Supplementary Fig. 8c). Moreover, immunofluorescence staining showed that these two proteins largely colocalized in the nucleus (Fig. 5g). As gene transcription involves a key step in which the activators work with the Mediator complex to recruit RNA Pol II to the gene promoters, we next used a chromatin IP (ChIP) assay to test whether Med23 deletion affected the recruitment of RNA Pol II by RUNX2 and the binding of RUNX2 to its target's promoter. The result showed that the occupancy of Pol II at the promoter of Ocn, a direct target gene of Runx2, was markedly decreased in the absence of MED23 while RUNX2 occupancy did not alter, suggesting that Med23 deficiency impaired the recruitment of Pol II to the promoter of Runx2-target osteogenic marker genes (Fig. 5h). Taken together, these results suggest that the Mediator subunit MED23 acts as a coactivator of Runx2 to regulate osteoblast differentiation. Based on the in vitro findings described above, we next questioned whether MED23 synergized with RUNX2 in vivo. We hypothesized that if MED23 acts together with RUNX2 to regulate osteoblast activity, then deletion of Med23 in parallel with Runx2 haploinsufficiency in vivo should aggravate the skeletal defects observed in Runx2 þ / À mice. To test this hypothesis and since disruption in single copy of Med23 did not reduce the protein level of MED23, we crossed Med23 MSC À / À mice with Med23 fl/fl /Runx2 þ / À mice and analysed the skeletal phenotypes of E16.5 embryos by alcian blue/ARS double staining. As shown in Fig. 6a, both Med23 fl/fl /Runx2 þ / À mice and Med23 MSC À / À mice showed the previously reported CCD-like skeletal abnormalities observed in heterozygous Runx2 mutant mice 8,10 . Runx2 þ / À /Med23 MSC À / À mice displayed significant bone defects, including delayed ossification of the fontanelles and hypoplasia of the clavicles (Fig. 6a). Furthermore, Von Kossa staining revealed reduced ossification of the long bones of E16.5 Runx2 þ / À and Med23 MSC À / À mice compared with control mice. However, Runx2 þ / À /Med23 MSC À / À compound mutant mice displayed a more severe reduction in ossification (Fig. 6b). Consistent with the reduced Von Kossa staining, the expression of Ocn was decreased significantly in the Runx2 þ / À /Med23 MSC À / À compound mutant mice compared with the Runx2 þ / À mice and Med23 MSC À / À mice ( Fig. 6c). Taken together, these data indicate a genetic interaction between Med23 and Runx2 and provide in vivo verification that MED23 can regulate osteoblast function via cooperation with RUNX2.

Discussion
The present study provides several lines of evidence that Med23 is required for osteoblast differentiation and bone development. Homozygous loss of Med23 in mesenchymal progenitors resulted in prominent defects in bone development, including hypoplasia of the clavicles, retarded ossification of the cranial bones and reductions in the trabecular bone and overall bone mass. Furthermore, Med23 conditional knockout in preosteoblasts largely recapitulated the characteristics of Med23 MSC À / À mice. Phenotypic analysis showed that Med23 deficiency affected both intramembranous and endochondral bone formation. In vitro differentiation experiments indicated that the defective osteogenesis happened in a cell-autonomous manner. In contrast to the impaired osteoblast development, the presence of multinucleated osteoclasts in the mineralized cartilage matrix of Med23-deficient mice indicated that osteoclast differentiation occurred normally.
MSCs can differentiate into different lineages, including osteoblasts, adipocytes, chondrocytes and myocytes 1  To the best of our knowledge, this is the first report of a direct role for the Mediator complex in modulating RUNX2 function. We provide evidence that the Mediator complex subunit MED23 partially regulates the transcriptional activity of RUNX2 and is required for osteoblast differentiation and bone development. First, homozygous loss of Med23 in mesenchymal progenitors resembled the bone phenotype of Runx2 heterozygous mice, resulting in prominent defects in bone development, with hypoplasia of the clavicles, retarded ossification of the cranial bones and reduction in the trabecular bone and overall bone mass. Moreover, Med23 conditional knockout in preosteoblasts largely recapitulated the characteristics of Med23 MSC À / À mice, with abnormalities in multiple skeletal elements. Interestingly, Med23 deficiency did not alter RUNX2 expression but instead caused the downregulation of multiple Runx2-regulated genes, such as Osx and Ocn. Reporter assays revealed that Med23 deficiency impaired Runx2 transcriptional activity but did not affect the activity of ATF4, another key regulator of osteoblast differentiation. Biochemical analyses further established the physical association between RUNX2 and MED23. Finally, Med23 deletion further aggravated the defective skeletal phenotype of Runx2 þ / À mice. In addition to the function in osteoblasts, Runx2 is also found to control the maturation of chondrocytes through Runx2 conditional knockout in chondrocytes by Col2a1-Cre 30,31 . Histological analysis of tibia showed that chondrocytes were arrested during terminal maturation in Med23 MSC À / À mice compared with control littermates as evidenced by less endochondral ossification and smaller hypertrophic chondrocytes in Med23 MSC À / À mice ( Supplementary Fig. 10a-c). All the results described above provide in vivo genetic evidence that MED23 is a cofactor of RUNX2.
Runx2 mutations are known to underlie human CCD, an autosomal-dominant heritable skeletal disease that is typically characterized by open or delayed closure of calvarial fontanelles and clavicle hypoplasia 9,32,33 . This phenotype can be reproduced in heterozygous Runx2 mutant mice 10 . More precisely, a 70% reduction in Runx2 levels also generates the CCD phenotype in mice 34 . Indeed, human bone marrow-derived MSCs with Med23 deficiency were retarded to differentiate into osteoblasts, which might imply that MED23 plays a role in human skeletal development and its mutation may relate to human bone diseases ( Supplementary Fig. 9). In addition, the dysregulation of Runx2 activity through manipulation of numerous nuclear factors, for example, TAZ, MAF and Satb2, causes abnormal osteoblast differentiation and bone development 11,35,36 . These studies indicate that the Runx2-dependent transcriptional output is under fine surveillance via the cooperation of many factors. Our study demonstrated that the Mediator subunit MED23 acts as a cofactor of RUNX2 in the regulation of osteoblast differentiation and bone development, providing an insight into the regulation of RUNX2 activity and skeletal dysplasias such as CCD.

Methods
Mice. Med23-floxed mice were generated by homologous recombination 16,17 . Briefly, the exons 5-7 were flanked by two loxP sites. After targeting vector delivery, ES cells were screened by PCR. Med23-floxed mice were backcrossed with C57/BL6J mice for at least six generations. Prx1-Cre and Osx-Cre-ER mice were purchased from the Jackson Laboratory. Runx2 þ / À mice were kindly provided by Professor Laurie Glimcher's lab. All mice were maintained under specific pathogen-free conditions. All animal experimental procedures were approved by the Institutional Animal Care and Research Advisory Committee of the Shanghai Institute of Biochemistry and Cell Biology.
Analysis of bone phenotypes. Skeletal preparations were double stained with alcian blue and ARS 8,10 . Briefly, embryos or newborns were eviscerated and the skin was removed. After fixation with 95% ethanol for 3 days, embryos or newborns were stained for 3 days in Alcian blue solution. Then they were fixed and cleared with 95% ethanol for three times and each 1.5 h, followed by treatment of 2% KOH for 3-4 h. After stained with ARS solution for 3-4 h, skeletons were cleared in 1%KOH/20% glycerol. For histological analysis, bone tissues were fixed in 4% paraformaldehyde (PFA) and then embedded in paraffin. For embryonic mice, 4 mm tissue sections were used for Von Kossa staining, DIG labelled in situ hybridization (Roche) and immunohistochemical staining (Dako). For postnatal mice, bone tissues were fixed in 4% PFA and decalcified for 2 weeks prior to paraffin embedding. Tissue sections (4 mm) were used for TRAP staining according to the standard protocol.
Measurement of PINP concentrations. We determined serum concentrations of PINP using the mouse PINP EIA kit (Immunodiagnostic Systems) according to the instructions provided.
Micro CT analysis. Mouse hind limbs were harvested, soft tissues were removed and the remaining tissues were stored in 70% ethanol. Scanning was performed with a Skyscan1076 instrument, and 36 slides (18 mm each) immediately below the growth plate in the distal metaphysis of the femur were used for quantification of the bone parameters.
Cell culture. C310T1/2 (ATCC) and 293T (ATCC) cells were maintained in DMEM containing 10% FBS. MC3T3E1 (ATCC) cells were maintained in Minimum Essential Medium a (a-MEM) containing 10% FBS. MSCs were isolated from bone marrow samples of 6-to 8-week-old mice and cultured in a-MEM containing 15% FBS. All cells were cultured in a 5% CO 2 humidified incubator at 37°C. For the c.f.u.-F assay, 2 Â 10 6 bone marrow-derived MSCs or 1 Â 10 2 osteoblast cells from calvaria were plated on each well of a six-well plate, cultured for 2 weeks and then fixed in 4% PFA and stained with 0.5% crystal violet, followed by counting the stained colonies. For in vitro osteoblast differentiation, MSCs were cultured in osteogenic medium (10% FBS with 50 mg ml À 1 ascorbic acid, 10 nM dexamethasone and 10 mM b-glycerophosphate) and subjected to ALP staining on day 7 and ARS staining on day 21. Human bone marrow-derived MSCs were cultured and induced to differentiate into osteoblasts according to the protocol from ScienCell.
Transient transfection and luciferase reporter assay. C3H10T1/2 cells were seeded overnight at 4 Â 10 4 cells per well into a 24-well plate and transfected by Lipofectamine 2000 (Life Technologies) with a luciferase reporter plasmid and pRL-TK (Promega) along with various expression constructs, as indicated. All wells were supplemented with control empty expression vector plasmids to keep the total amount of DNA constant. At 36-48 h post transfection, the cells were harvested and subjected to dual-luciferase reporter assays according to the manufacturer's protocol (Promega).
In vitro interaction assay. GST or GST-RUNX2 recombined proteins were expressed in Escherichia coli BL21, followed by purification according to the instructions of the manufacturer (GE). His-Flag-MED23 was expressed using the Bac-to-Bac baculovirus expression system and purified by Ni-NTA agarose beads (Invitrogen). Purified GST or GST-Runx2 was incubated with purified MED23 and subjected to GST pull-down experiment . The bound proteins were analysed by SDS-10% PAGE and immunoblotting with indicated antibodies (Anti-GST, 1:1,000, Santa Cruz; anti-MED23, 1:1,000).
Immunofluorescence staining and co-localization analysis. Primary osteoblast cells were plated on cover slips for 24 h in a-MEM with 10% FBS with BMP2 (100 ng ml À 1 ) prior to staining. After removal of culture media, cells were washed with PBS, fixed in 4% PFA, permeabilized with 0.2% Triton X-100 for 10 min and then blocked with 2% BSA in PBS for 1 h. Anti-MED23 (1:100) and anti-RUNX2 (1:200) primary antibodies were diluted in the blocking solution and applied overnight at 4°C. After PBS wash for three times, secondary antibodies (Jackson Laboratory) were diluted (1:100) in blocking buffer and applied for 1 h at room temperature. After PBS wash for three times, nuclei were stained with DAPI (1:4,000) for 4 min. Co-localization analysis were performed using Volocity Software (PerkinElmer).
Real-time PCR analysis. Total RNA was isolated from bone tissue or cells with TRIzol reagent (Life Technologies). Complementary DNA was generated using M-MLV reverse transcription kit (Promega). Real-time PCR was conducted in triplicate with SYBR Premix Ex Taq (Takara). The level of the endogenous mRNA was normalized to the level of Gapdh mRNA using the 2 À DCT method. Specific primer sequences are listed in the Supplementary Table 1.
ChIP assay. Cells were cross-linked with 1% formaldehyde in culture media for 10 min at room temperature, followed by adding 0.125 M glycine to quench the cross-linking. Cells were washed and scraped in cold PBS and harvested in ChIP lysis buffer (50 mM Tris-HCl, pH 7.4, 1% SDS, 10 mM EDTA). Cell lysates were then sonicated to shear chromatin. After that, cell lysates were centrifuged and 100 ml supernatant was diluted with 900 ml dilution buffer (16.7 mM Tris-HCl, pH 8.0, 0.01% SDS, 1.1% Triton X-100, 1.2 mM EDTA, 167 mM NaCl) and then incubated with 2 mg anti-RUNX2 or 2 mg anti-Pol II (Santa Cruz) or non-immune Rabbit IgG overnight at 4°C. About 20 ml Protein G magnetic beads (Life technologies) were added for another 4 h on a rotating wheel. Chromatin then was immunoprecipitated, decrosslinked at 65°C and treated with Proteinase K. Precipitated DNA was extracted and quantified by real-time PCR. All values were normalized to input. Specific primer sequences were listed in Supplementary  Table 1.
Retrovirus infection. Retroviruses expressing RNA interference (RNAi) oligonucleotides were generated by transfection of recombinant pSiren-RetroQ and pCL10A1 helper plasmids into 293T cells using Lipofectamine 2000. Culture media were changed 24 h after transfection and the supernatants containing retroviruses were harvested 48 h later and passed through a 0.45-mm filter. Filtered supernatants were supplemented with 6 mg ml À 1 polybrene and added to the cells grown in sixwell plates for centrifugation at 2,500 r.p.m. for 1.5 h at 30°C. About 24 h after spin infection, infected cells were then selected in the presence of puromycin: primary MSCs at 5 mg ml À 1 and C3H10T1/2 at 8 mg ml À 1 . The RNAi oligo sequences are listed in Supplementary Table 2. RNA-Seq and GO analysis. Total RNA was isolated with TRIzol from the parietal bone of P2 control mice (n ¼ 3) and Med23 MSC À / À littermates (n ¼ 3) and from the parietal bone of P2 WT mice (n ¼ 3) and Runx2 þ / À mice (n ¼ 3). Complementary DNA library preparation and sequencing were performed according to the Illumina standard protocol. GO analysis was performed with the DAVID online tool. Top GO categories were selected according to the P values. RNA-seq data in this study have been deposited at Gene Expression Omnibus (GEO) (http:// www.ncbi.nlm.nih.gov/gen/) under accession ID GSE 77007.