Runx2 Protein Represses Axin2 Expression in Osteoblasts and Is Required for Craniosynostosis in Axin2-deficient Mice*

Background: Runx2 and Axin2 are required for proper skeletal development. Results: Runx2 and Hdac3 repress Axin2 transcription in osteoblasts. Runx2 insufficiency prevents craniosynostosis in Axin2-deficient mice. Conclusion: A Runx2-Axin2 regulatory mechanism controls the pace of calvarial bone formation. Significance: The molecular and functional interplay between Runx2 and Axin2 controls the rate of cranial suture closure. Similar interactions may occur during skeletal development and carcinogenesis. Runx2 and Axin2 regulate craniofacial development and skeletal maintenance. Runx2 is essential for calvarial bone development, as Runx2 haploinsufficiency causes cleidocranial dysplasia. In contrast, Axin2-deficient mice develop craniosynostosis because of high β-catenin activity. Axin2 levels are elevated in Runx2−/− calvarial cells, and Runx2 represses transcription of Axin2 mRNA, suggesting a direct relationship between these factors in vivo. Here we demonstrate that Runx2 binds several regions of the Axin2 promoter and that Runx2-mediated repression of Axin2 transcription depends on Hdac3. To determine whether Runx2 contributes to the etiology of Axin2 deficiency-induced craniosynostosis, we generated Axin2−/−:Runx2+/− mice. These double mutant mice had longer skulls than Axin2−/− mice, indicating that Runx2 haploinsufficiency rescued the craniosynostosis phenotype of Axin2−/− mice. Together, these studies identify a key mechanistic pathway for regulating intramembranous bone development within the skull that involves Runx2- and Hdac3-mediated suppression of Axin2 to prevent the untimely closure of the calvarial sutures.

Runx2 and Axin2 regulate craniofacial development and skeletal maintenance. Runx2 is essential for calvarial bone development, as Runx2 haploinsufficiency causes cleidocranial dysplasia. In contrast, Axin2-deficient mice develop craniosynostosis because of high ␤-catenin activity. Axin2 levels are elevated in Runx2 ؊/؊ calvarial cells, and Runx2 represses transcription of Axin2 mRNA, suggesting a direct relationship between these factors in vivo. Here we demonstrate that Runx2 binds several regions of the Axin2 promoter and that Runx2-mediated repression of Axin2 transcription depends on Hdac3. To determine whether Runx2 contributes to the etiology of Axin2 deficiencyinduced craniosynostosis, we generated Axin2 ؊/؊ :Runx2 ؉/؊ mice. These double mutant mice had longer skulls than Axin2 ؊/؊ mice, indicating that Runx2 haploinsufficiency rescued the craniosynostosis phenotype of Axin2 ؊/؊ mice. Together, these studies identify a key mechanistic pathway for regulating intramembranous bone development within the skull that involves Runx2-and Hdac3-mediated suppression of Axin2 to prevent the untimely closure of the calvarial sutures.
Craniosynostosis is a common birth defect, appearing in ϳ1 of 2500 live births (1,2). This condition is characterized by premature fusion of one or more cranial sutures, which causes abnormal skull shapes and cognitive disabilities if surgical reconstructions are not performed (1,2). Considerable progress has been made in understanding the genetic causes of craniosynostosis, as mutations in MSX2, TWIST1/2, FGFR1/ 2/3, and EFNB2 are linked to various craniosynostosis syndromes (3)(4)(5). Despite these advances, the etiology of ϳ85% of non-syndromic cases remains unknown (3).
Mouse models recapitulate human craniosynostosis syndromes and provide substantial mechanistic insight into disease mechanisms. For example, Twist haploinsufficiency (6,7) and introduction of activating mutations in Fgfr1 (8) and Fgfr2 (9) faithfully mimic features of human craniosynostosis in developing mice. Other mouse models reveal new candidate genes that were not previously linked to human craniosynostosis syndromes. One such gene is Axin2 (10 -12), an intracellular negative feedback regulator of Wnt, Lrp5/6, and ␤-catenin signaling (10 -12, 14, 16). Axin2-deficient mice, created by insertion of LacZ into exon 2 (hereafter referred to as Axin2 Ϫ/Ϫ mice), feature up-regulated ␤-catenin signaling in the cranial sutures (10,12,16). These mice develop craniosynostosis because of enhanced mesenchymal progenitor cell proliferation in the cranial sutures and rapid differentiation of progenitor cells into matrix-producing osteoblasts (10 -12).
Skeletal development and bone mass maintenance are complex processes involving multiple levels of regulation by transcription factors and epigenetic modifications to control gene expression and coordinate activities of bone-residing cells. Runx2 is a master osteoblast transcription factor that is essential for proper bone development. Runx2-deficient mice die at birth because of a lack of skeletal ossification, and Runx2 haploinsufficient mice and humans develop an osteopenic phenotype and features of cleidocranial dysplasia, including delayed calvarial development (17)(18)(19). A 30% reduction in Runx2 levels is sufficient to induce a cleidocranial dysplasia phenotype in mice (19). Interestingly, although Runx2 insufficiency impairs cranial suture closure, increased Runx2 expression coincides with the increased osteoblast differentiation that causes premature suture closure in many models of craniosynostosis (20 -24), suggesting convergence on a key regulatory role for Runx2 in cranial suture biology and, in particular, in craniosynostosis etiology. This is supported by a recent finding of RUNX2 duplication in human metopic craniosynostosis (25) and studies showing that overexpression of Runx2 in the condensing calvarial mesenchyme causes craniosynostosis in mice (26).
The opposing cranial phenotypes of the Axin2-deficient (craniosynostosis) and Runx2 haploinsufficient (cranial dysplasia) mice suggest potential interaction between these two factors in vivo, an idea supported by our observations that Axin2 protein levels were up-regulated in Runx2-null osteoblasts and that Runx2 repressed Axin2 transcription (27). Runx2 represses transcription by recruiting histone deacetylases (Hdacs) 2 to gene promoters. We showed that Hdac3 binds to the N terminus of Runx2 and represses Runx2-dependent activation of Bglap (28). Although suppression of Hdac3 in committed osteoblasts accelerates differentiation (28), conditional deletion of Hdac3 in osteochondral progenitor cells (from Osterix-Cre:Hdac3 fl/fl conditional knockout mice, hereafter referred to as Hdac3 CKO Osx mice) causes calvarial, axial, and appendicular bone loss in both trabecular and cortical compartments because of reduced bone formation (29). Interestingly, Axin2 was one of the most highly expressed genes in Hdac3 CKO Osx calvaria as compared with wild-type littermates, suggesting that the deleterious osteoblast phenotype in these mice could be due, in part, to a suppression of Wnt/␤-catenin signaling via up-regulation of its inhibitor Axin2.
To determine whether Axin2, Runx2, and Hdac3 directly interact and are components of the same molecular pathway that regulates suture closure, we performed molecular experiments and crossed the Axin2 Ϫ/Ϫ and Runx2 ϩ/Ϫ mouse models to generate double mutant Axin2 Ϫ/Ϫ :Runx2 ϩ/Ϫ mice. These studies demonstrate that Runx2 haploinsufficiency rescues the craniosynostosis phenotype in Axin2 Ϫ/Ϫ mice and that Runx2 directly represses Axin2 expression in an Hdac3-dependent manner.
Transient Transfections, Adenoviral Transductions, and PCR-C2C12 cells were transfected with Lipofectamine (Invitrogen) in 6-well plates with 1 g of Runx2 expression plasmid (Runx2-I, P2 promoter-dependent isoform, beginning with the amino acids MRIPV) or 1 g of a control plasmid (pcDNA3, Invitrogen). RNA was harvested with TRIzol reagent (Invitrogen) 48 h after transfection and was reverse transcribed into cDNA using the SuperScript III first-strand synthesis system (Invitrogen). Expression levels of mRNAs for Runx2, Axin2, and Gapdh were measured by real-time PCR. Primer sequences were reported previously (32,33). Reactions were performed using 37.5 ng of cDNA/15 l with Bio-Rad iQ SYBR Green Supermix and the Bio-Rad MyiQ single color real-time PCR detection system. Transcript levels for each gene of interest were normalized to the reference gene Gapdh. Gene expression levels were quantified using the 2ˆ(-⌬⌬ Ct) method (17).
To observe the effects of reintroduction of Runx2 expression into Runx2 Ϫ/Ϫ cells, control and Runx2 adenoviruses were prepared as described previously (31). WT and Runx2 Ϫ/Ϫ cells were infected with each indicated virus at a multiplicity of infection of 100. RNA was harvested with TRIzol reagent (Invitrogen) 24 h after infection. RNA was reverse-transcribed, and gene expression levels for Runx2, Axin2, Hdac3, and Gapdh were quantified as described above.
ChIP-Seq-MC3T3-E1 cells cultured in osteogenic medium for 0, 9, or 28 days were washed with 37 ºC PBS and then fixed at room temperature with 1% formaldehyde for 8 min. Cells were lysed according to a published protocol (35). Nuclear lysates were sonicated to generate chromatin fragments ranging from 200 -600 bp. Chromatin was then precipitated with Runx2 antibody (M-70, Santa Cruz Biotechnology, Inc.) or control normal IgG followed by protein G Dynabeads (Invitrogen). Runx2-precipitated chromatin was washed, and cross-links were reversed overnight at 65 ºC. DNA samples were subsequently treated with RNase A followed by proteinase K at 55 ºC. DNA was recovered by phenol/chloroform extraction followed by ethanol precipitation and transformed into a sequencing library according to the Illumina DNA library preparation protocol. A fraction of DNA fragments with genomic inserts at 200 Ϯ 50 bp were sent for sequencing on an Illumina Genome Analyzer II operated by the Deep Sequencing Core Facility at the University of Massachusetts Medical School. Base calls and sequence reads were generated by Illumina CASAVA software (version 1.6), and short reads were aligned to the mouse mm9 genome using Bowtie aligner (version 0.12.7) (36). Values shown for Axin2 represent two biological replicates. Runx2enriched peaks were analyzed by model-based analysis of ChIP-Seq (MACS) (37).
Construction of Plasmids and Luciferase Reporter Assays-The 5.6-kbp Axin2-luciferase construct (32) containing the mouse Axin2 promoter, exon 1, and intron 1 was obtained from Addgene (plasmid no. 21275). To focus on the Axin2 promoter regions showing Runx2 association identified in ChIP and ChIP-Seq assays, the full-length Axin2-Luc construct (hereafter referred to as Axin2(FL)) was cut with restriction enzymes (Acc65I and HindIII) to generate smaller inserts. Deletion constructs were ligated into the pGL3 basic vector (Promega). Sitedirected mutagenesis of the four Runx2 consensus sites was performed with the QuikChange II site-directed mutagenesis kit (Agilent Technologies, catalog no. 200523). The effects of Runx2 on these constructs were determined using a pCMV-HA-Runx2 expression plasmid (Runx2 isoform 1, beginning with the amino acids MASNS) that has been described previously (27,28). Mutant pCMV-HA-MASNS(M182R) was generated using the site-directed mutagenesis kit described above and gene-specific oligonucleotides containing a mutation in the DNA binding site (5Ј-CGAAATGCCTCCGCTGTT AGGAA-AAACCAAGTAGCCAGG-3Ј and 5Ј-CCTG GCTACTTG-GTTTTTCCTAACAGCGGAGGCATTTCG-3Ј).
C2C12 cells were transfected with Lipofectamine (Invitrogen) in 12-well plates with 200 ng of each Axin2-luciferase promoter construct, as indicated, and pRL-null (10 ng). Runx2 expression plasmids (300 ng) were added as indicated (28). Following a 48-h incubation at 37°C, luciferase activity in 20 l of cell lysate was measured using the dual luciferase assay system (Promega) and a Glomax 96 microplate luminometer. All transfections were performed in triplicate, and data were normalized to the activity of Renilla luciferase.
Animal Studies-All animal research was conducted according to guidelines provided by the National Institutes of Health and the Institute of Laboratory Animal Resources National Research Council. The Mayo Clinic Institutional Animal Care and Use Committee approved all animal studies. Animals were housed in an accredited facility under a 12-hour light/dark cycle and were provided with water and food (PicoLab Rodent Diet 20, LabDiet) ad libitum.
Hdac3 CKO Osx Mice-Hdac3 CKO Osx , heterozygous and wild-type littermate mice were generated and genotyped as described previously (29). Calvarial explants were removed from 5-or 6-day-old mice, dissected free of soft tissues, and incubated in collagenase digestion medium before flash-freezing in liquid nitrogen, as reported previously (38). For mRNA analyses, bone explants were homogenized in TRIzol using a high-speed disperser (Ultra-Turrax T25, IKA). RNA was extracted and purified from the ground tissue with TRIzol reagent (Invitrogen) and was reverse-transcribed into cDNA and analyzed via real-time PCR as described above. For protein isolation, calvarial explants were crushed in liquid nitrogen with a mortar and pestle and were suspended in modified radioimmune precipitation assay buffer supplemented with protease inhibitors. Protein extracts from calvarial explants were sonicated and resolved by SDS-PAGE. Immunoblotting was performed with antibodies recognizing Axin2 (1:1000, Abcam, catalog no. Ab32197) and mSin3A (1:1000, Santa Cruz Biotechnology, Inc., K-20). mSin3A is a ubiquitous global transcriptional corepressor and used here as a loading control.
Cranial Morphology and Suture Histology in Axin2 Ϫ/Ϫ : Runx2 ϩ/Ϫ Mice-Cranial morphology was assessed via microcomputed tomography and image analysis software at 4 weeks of age, when Axin2 Ϫ/Ϫ mice demonstrate a significant reduction in skull length (15), and at 6 months of age, to assess skull morphology in adult mice. Mice were euthanized and skulls were fixed in 70% ethanol. Skulls were scanned in 70% ethanol on a CT35 scanner (Scanco Medical AG, Basserdorf, Switzerland) at 20 -30 m resolution (energy settings 70 kV, 114 A) and reconstructed with the software of the manufacturer using a threshold value of 150. Skull coronal length (coronal suture to nose tip) and width (measured at the distal aspect of the zygomatic process across the parietal bones) were quantified using image analysis software (Bioquant Osteo, Nashville, TN).
For analysis of suture histology, skulls were embedded in glycolmethacrylate. Micro-CT scans of the embedded skulls were used to precisely locate the posterior frontal suture (analogous to the human metopic suture), as this suture closes prematurely in Axin2-deficient mice and contributes to their craniosynostosis phenotype (12). Thin (5-m) sections were mounted and stained with Gomorri's trichrome (39) to highlight bone and connective tissues. Cell and bone morphology were examined at ϫ600 magnification.
BMSC Growth, Mineralization, and Gene Expression-BMSC cultures were chosen as a model to study the in vitro osteoblastic differentiation patterns of progenitor cells from mice of each genotype because these cultures can be performed on tissue explants from mice at 4 weeks of age (or older), permitting tissue harvests at time points that corresponded directly with ages of morphological analyses of the skull. BMSC isolated from 4-week-old mice were seeded in osteogenic medium and cultured for 28 days to promote calcified matrix production, then fixed in 10% neutral buffered formalin and stained with 2% Alizarin red. Calcified matrix production was quantified as the percentage of Alizarin red-positive area within each well with image analysis software (Bioquant Osteo). For gene expression studies, RNA was harvested with TRIzol reagent (Invitrogen) after 7, 14, or 21 days in osteogenic culture. RNA was reverse transcribed and expression levels of Runx2 or Bglap (genes associated with osteoblast maturation) were quantified as described above and previously (32).
Statistics-Statistics were performed with JMP 9.0 statistical analysis software (SAS Institute, Inc., Cary, NC). Data were compared between groups within each experiment with Student's t-tests (when only two groups were compared) or ANOVA (when three or more groups were compared) followed by Student's t post-hoc comparisons. A significance of p Ͻ 0.05 was used for all comparisons.

Runx2 Regulates Axin2 Expression in Osteoblasts-To
understand the genetic interplay between Axin2 and Runx2, we examined Axin2 transcripts using mRNA isolated from calvarial cells of Runx2 Ϫ/Ϫ E17.5 embryos (40). Axin2 mRNA levels were increased 5.5-fold in Runx2 Ϫ/Ϫ cells as compared with wild-type primary calvarial cells (Fig. 1A). The finding that Runx2 controls Axin2 mRNA corroborates our previous study showing that Axin2 protein levels are elevated in Runx2 Ϫ/Ϫ cells (27) and indicates that Runx2 controls Axin2 protein levels through its ability to affect transcription and mRNA accumulation.
We hypothesized that Axin2 was a direct transcriptional target of Runx2 and identified four Runx2 consensus binding sequences (i.e. ACCACA or TGTGGT (41)) in the Axin2 promoter (15), which were designated as putative Runx2 binding sites 1 to 4 (Fig. 1B). These sequences were tested for Runx2 binding using an EMSA with lysates from C2C12 cells overexpressing Runx2. Runx2 bound the first site with high affinity and the fourth site with moderate affinity (Fig. 1B). Weak in vitro binding was also suggested at sites 2 and 3 in the presence of a Runx2-specific antibody (␣-AML3) that caused a supershift. However, in ChIP assays with differentiated osteoblasts, Runx2 associated with the Axin2 promoter at all four sites, and, in particular, strongly bound at sites 2 and 3 (Fig. 1C). Association of Runx2 with regions of the Axin2 promoter containing Runx2 binding sites was enriched several fold as compared with a downstream region lacking a Runx2 binding site (Fig. 1D). These data indicate that Runx2 associates with the Axin2 promoter in differentiating osteoblasts.
To further examine Runx2's association with the Axin2 promoter during osteoblast differentiation, a ChIP sequencing analysis was performed on MC3T3 cells differentiated for 0, 9, or 28 days (Fig. 2A). Association of Runx2 with the four Runx2 consensus motifs characterized above (hereafter referred to as region 1) was confirmed. This unbiased and genomic approach revealed another region of the Axin2 promoter with relatively stronger Runx2 association in differentiating osteoblasts ( Fig.  2A). This region (hereafter referred to as region 2) of Runx2 association consisted of three strong peaks located within 2 kbp of the Axin2 transcription start site. The first two of these peaks (hereafter referred to as peak i and peak ii) were validated by ChIP-PCR (Fig. 2B). Notably, a consensus Runx2 binding motif was not present in these DNA sequences, despite the strong association of Runx2 with region 2 of the Axin2 promoter. Analysis of region 2 DNA sequences with the Transcription Element Search System (42) revealed potential binding elements for other factors known to interact with Runx2 (e.g. Lef/ Tcf transcription factors, C/EBP␤). However, the cofactors with which Runx2 interacts in Region 2 remain unknown at present.

Runx2 Represses Axin2 in Osteoblasts
Runx2 Represses Axin2 Transcription-To test the role of Runx2 in regulating Axin2 transcription, an Axin2-luciferase reporter construct containing the 5.6-kbp region described above (15) was cotransfected with Runx2 expression plasmids into C2C12 cells. Runx2 repressed the basal activity of the fulllength Axin2 promoter construct by 60 -70% (Fig. 2C), consistent with our previous observations. Runx2 also repressed luciferase reporters driven by either region 1 or region 2 of the Axin2 promoter (Fig. 2C).
We next sought to determine whether inducing Runx2 expression would suppress endogenous Axin2 levels. Overexpression of Runx2 in C2C12 cells reduced Axin2 transcript levels by ϳ50% as compared with cells transfected with a control plasmid (Fig. 2D). Moreover, upon induction of osteoblastic differentiation, Runx2 expression levels increased 3.5-fold in primary mouse bone marrow stromal cells and coincided with 80% reductions in endogenous Axin2 expression (Fig. 2E). Together, these data demonstrate that Runx2's suppression of Axin2 may play a role in the early stages of osteoblastic lineage commitment and differentiation.
To determine whether DNA binding was required for repression of Axin2 by Runx2, we expressed a mutant Runx2 protein that contains a point mutation in the Runt domain, M182R, that is linked to human cleidocranial dysplasia and prevents DNA binding (43,44). Runx2-M182R was less effective than Runx2 at repressing the FL and region 1 constructs and completely failed to repress region 2, confirming that the Runt domain contributes to the repression of Axin2 by Runx2 in both regions of the promoter and is required for repression in region 2 (Fig. 2C). Interestingly, Runx2 and Runx2-M182R repressed the Axin2-region 1 reporter construct even when the consensus Runx2 binding sites were mutated (Axin2 (mutReg1)-Luc), suggesting that repression of Axin2 by Runx2 may require the cooperation of other transcription factors.
Runx2 Represses Axin2 Transcription in an Hdac3-dependent Manner-We and others demonstrated previously a direct interaction between Hdac3 and Runx2 in osteoblasts (28,45). To test the role of Hdac3 in Runx2-directed repression of Axin2 transcription, we collected protein and mRNA from calvarial bones or osteoblasts of Hdac3 CKO Osx animals or control littermates. Axin2 mRNA was increased 3.8-fold in Hdac3 CKO Osx calvarial osteoblasts (Fig. 3A) and was elevated in calvarial explants (B) as compared with wild-type littermates. Axin2 protein levels were also elevated in Hdac3 CKO Osx cal-  FEBRUARY 22, 2013 • VOLUME 288 • NUMBER 8 varial explants (Fig. 3C). These experiments demonstrate that Runx2 and Hdac3 actively repress Axin2 expression.

Runx2 Represses Axin2 in Osteoblasts
In an attempt to rescue the high Axin2 expression levels in Runx2 Ϫ/Ϫ calvarial osteoblasts, we transduced WT and Runx2 Ϫ/Ϫ cells with either control or Runx2 adenovirus. As expected, Axin2 levels were up-regulated in Runx2 Ϫ/Ϫ cells infected with a control adenovirus. Unexpectedly, although the transduction was successful at increasing Runx2 expression levels, Axin2 levels remained high in Runx2 Ϫ/Ϫ cells treated with Runx2 adenovirus (Fig. 3D). Further analyses revealed that

Runx2 Represses Axin2 in Osteoblasts
Hdac3 levels were suppressed in Runx2 Ϫ/Ϫ cells and remained low in Runx2 Ϫ/Ϫ cells infected with Runx2 adenovirus (Fig.  3D), indicating that Hdac3 is essential to the mechanism by which Runx2 represses Axin2 transcription.
Runx2 Haploinsufficiency Rescues Several Aspects of the Craniosynostosis Phenotype of Axin2-deficient Mice-To determine whether Axin2 and Runx2 are components of the same molecular pathway in vivo, Runx2 ϩ/Ϫ and Axin2 Ϫ/Ϫ mice were crossed to generate double-mutant Axin2 Ϫ/Ϫ :Runx2 ϩ/Ϫ mice. We limited studies to four genotypes of interest (WT mice, Runx2 ϩ/Ϫ and Axin2 Ϫ/Ϫ single mutants, and double mutant Axin2 Ϫ/Ϫ :Runx2 ϩ/Ϫ mice) because Axin2 haploinsufficient mice do not present a craniosynostosis phenotype (12). Skull morphology was quantified from micro-CT reconstructions at 4 weeks and 6 months of age. At 4 weeks of age, Axin2 Ϫ/Ϫ single mutants demonstrated reduced skull length, measured from the coronal suture to nose tip (-17%), as compared with wildtype animals (17)(18)(19) (Fig. 4, A, C, and D), which is consistent with previous studies (12). Axin2 Ϫ/Ϫ mice also featured an increase in skull width across the parietal bones (Fig. 4B). Micro-CT and histological analyses revealed a near absence of the jugum limitans and fully fused posterior frontal suture in the Axin2 Ϫ/Ϫ mice at 4 weeks of age (Fig. 4, D and E), indicative of craniosynostosis. Skull lengths and widths were statistically similar in 4 week-old Runx2 ϩ/Ϫ mice as compared with wildtype mice. Interestingly, 4-week-old double mutant Axin2 Ϫ/Ϫ : Runx2 ϩ/Ϫ mice had skulls that were longer and narrower than Axin2 Ϫ/Ϫ mice. The average length of the skull was significantly increased by 12%, and skull width was significantly decreased by 5% in the Axin2 Ϫ/Ϫ :Runx2 ϩ/Ϫ mice as compared with the Axin2 Ϫ/Ϫ mice (Fig. 4, A-C). These measurements in the double mutant animals were statistically comparable with Runx2 ϩ/Ϫ and WT mice. Micro-CT and histological analyses revealed that the posterior frontal suture of the Axin2 Ϫ/Ϫ : Runx2 ϩ/Ϫ mice remained patent and morphologically mimicked that of the Runx2 ϩ/Ϫ mice (Fig. 4, D and E), providing a FIGURE 2. Runx2 binds and represses two distinct regions of the Axin2 promoter. A, chromatin immunoprecipitations were performed with a Runx2 antibody in MC3T3-E1 cells at various phases of osteoblastic differentiation: day 0 (undifferentiated/proliferation), day 9 (matrix deposition), and day 28 (matrix mineralization). Immunoprecipitated DNA was analyzed with ChIP-Seq. Region 1 contains the four consensus Runx2 binding motifs characterized in EMSA and ChIP assays. Three strong peaks of Runx2 association were located in region 2 despite the fact that this segment of the Axin2 promoter does not contain any consensus Runx2 binding motifs. B, real-time PCR results from ChIP assays within region 2. Association of Runx2 with DNA in peaks i and ii is enriched severalfold compared with a downstream region (ϳ1 kbp past peak ii) lacking Runx2 association in the ChIP-Seq assay (No ChIP-Seq association). Runx2 immunoprecipitations (IP) were performed with the D130-3 antibody (MBL). Isotype control IgG2b immunoprecipitations are negative controls. C, constructs encoding full-length Runx2 (Runx2-II, starting with the amino acids MASNS) or a mutated form of Runx2-II (Runx2-M182R) containing a point mutation in the runt domain that prevents Runx2 from binding DNA) were transfected into C2C12 cells along with the full-length Axin2-luciferase reporter construct (Axin2(FL)) or truncated constructs featuring region 1 or region 2 of the Axin2 promoter. Bars with different superscript letters are statistically different from one another. Luc, luciferase. D, Runx2 overexpression suppressed endogenous expression levels of Axin2 by ϳ50% as compared with control C2C12 cells transfected with a pcDNA3 plasmid (Control). *, p Ͻ 0.05; **, p Ͻ 0.08 versus control. SEM, standard error of the mean. E, primary mouse BMSC were cultured in osteogenic medium for 3 days to induce osteoblastic differentiation. Endogenous Runx2 and Axin2 expression were quantified and compared with undifferentiated (Day 0) cells. *, p Ͻ 0.05 versus day 0 cells.  FEBRUARY 22, 2013 • VOLUME 288 • NUMBER 8 mechanism by which the gross morphology of the craniosynostosis phenotype of the Axin2 Ϫ/Ϫ single mutant animals was rescued. Taken together, these data indicate that Runx2 is essential for Axin2-deficiency induced premature fusion of the posterior frontal suture. Of note, the jugum limitans was absent in the double mutant mice, suggesting that not all of the morphological features of Axin2 deficiency induced craniosynostosis are rescued by reducing Runx2 levels. At 6 months of age, Axin2 Ϫ/Ϫ mice retained the shortened skull morphology that was first observed at 4 weeks of age. Surprisingly, skull lengths also were shorter in both Runx2 ϩ/Ϫ and in Axin2 Ϫ/Ϫ :Runx2 ϩ/Ϫ as compared with WT mice, possibly as a consequence of the cleidocranial dysplasia phenotype in these mice.

Runx2 Represses Axin2 in Osteoblasts
Bone Marrow-derived Osteoblasts from Axin2 Ϫ/Ϫ :Runx2 ϩ/Ϫ Mice Produce a Mineralized Matrix but Show Decreased Osteoblastic Gene Expression-BMSCs were obtained from femurs and tibias of 4-week-old mice and cultured in osteogenic medium to investigate cellular differentiation patterns. Mineralization assays revealed a severe reduction in the calcified matrix produced by osteoblast-differentiated BMSC from Runx2 ϩ/Ϫ mice as compared with WT animals (Fig. 5A). Neither osteoblasts derived from Axin2 Ϫ/Ϫ nor Axin2 Ϫ/Ϫ : Runx2 ϩ/Ϫ mice were impaired in their ability to produce a calcified matrix (Fig. 5A). Gene profiling showed that osteoblasts derived from Axin2 Ϫ/Ϫ :Runx2 ϩ/Ϫ mice produced substantially less Runx2 and Bglap as compared with osteoblasts from Axin2 Ϫ/Ϫ single mutant mice and instead mimicked temporal gene expression patterns of cultures from WT mice (Fig. 5B). This result suggests that Runx2 insufficiency relieves the craniosynostosis phenotype in the Axin2 Ϫ/Ϫ mice by reducing osteoblastic gene expression to normal levels in progenitor cells.

DISCUSSION
Runx2 and Axin2 are key regulators of craniofacial development and skeletal maintenance with physiologically opposing functions. Runx2 is a positive master regulator of osteoblastspecific transcription that is required for commitment of mesenchymal progenitors to the osteoblast lineage and for osteoblastic differentiation (17,18,46). Axin2 is a negative intracellular feedback inhibitor of Wnt/␤-catenin signaling (15), a pathway that stimulates proliferation and survival of osteoblast progenitors (47) and facilitates the coupling of bone formation to bone resorption via regulation of osteoprotegerin (Opg) expression in osteoblasts (48). Our data demonstrate that the opposing activities of Runx2 and Axin2 are functionally linked in osteoblasts through a Runx2-and Hdac3-mediated transcriptional mechanism. Both Runx2 Ϫ/Ϫ and Hdac3 CKO Osx calvarial osteoblasts express more Axin2 than WT cells, which may be explained by a loss of the repressive effects of Runx2 and Hdac3 on Axin2 transcription. In vivo, introduction of Runx2 haploinsufficiency onto the Axin2 Ϫ/Ϫ background rescued the craniosynostosis phenotype of the Axin2deficient animals. In vitro studies with BMSCs cultured in osteogenic media suggest that the skull morphology rescue in Axin2 Ϫ/Ϫ :Runx2 ϩ/Ϫ mice may be due to reduced osteoblastic activity, as expression levels of characteristic osteoblastic genes were lower in cells from Axin2 Ϫ/Ϫ :Runx2 ϩ/Ϫ mice as compared with Axin2 Ϫ/Ϫ single mutants. On the basis of these data, we propose a model where Axin2 and Wnt/␤-catenin signaling are upstream of Runx2 in mesenchymal progenitor cells but Runx2 directly represses Axin2 transcription during the early stages of progenitor cell commitment to the osteoblastic lineage and to osteoblastic differentiation, reflecting developmental feedback regulation (Fig. 6). This model is compatible with the finding that Runx2 expression is induced by canonical Wnt signaling (49) and is supported by our data demonstrating that Runx2 represses Axin2 in a manner dependent on Hdac3 in osteoblasts. Our data are consistent with the idea that Runx2 plays a central role in the etiologies of several different forms of craniosynostosis. For example, Runx2 is downstream of the zinc finger transcription factor GLI3, another factor recently linked to craniosynostosis in mice. Gli3-deficient mice develop craniosynostosis in the interfrontal and lambdoid sutures, and removing one allele of Runx2 or increasing Twist1 expression (which subsequently represses Runx2) rescued this phenotype (22,50).
Although Axin2 has not yet been linked to a specific craniosynostosis syndrome in humans, other clinical observations support an inverse link between Runx2 and Axin2 in human craniofacial development. RUNX2 haploinsufficiency in humans causes the formation of supernumerary teeth (46), whereas AXIN2 mutations are associated with tooth agenesis in humans (51). Biological interaction between Axin2, Runx2, and Hdac3 may also factor into disease mechanisms in cancer. Inactivating mutations in AXIN2 are associated with an increased incidence of several types of cancer because of unchecked Wnt signaling (51)(52)(53)(54), whereas RUNX2 is highly expressed in breast and prostate cancers predisposed to skeletal metastasis (55,56), and HDAC3 is one of the most commonly up-regulated genes in human solid tumors (57). These observations are consistent with our data showing that Runx2 represses Axin2 in an Hdac3dependent manner in vivo and suggest that this interaction may have biological relevance with respect to modulating Wnt/␤catenin signaling in both skeletal development and cancer.
The inhibitory effects of Axin 2 on ␤-catenin signaling modulate bone formation in the cranial sutures and prevent premature suture closure (10 -12, 21). In mice, the posterior frontal suture (analogous to the human metopic suture) is the only cranial suture that fuses naturally (58). Axin2 expression in this suture diminishes over time prior to the onset of fusion (11,12) and is inversely correlated with the amount of activated ␤-catenin present in the suture mesenchyme (11). Constitutive activation of ␤-catenin in Axin2-expressing cells is sufficient to recapitulate the craniosynostosis phenotype of the Axin2-deficient mice (21). It is possible that Runx2 regulates suture fusion by repressing Axin2 expression to facilitate a temporal increase in ␤-catenin signaling that promotes bone formation (Fig. 6). Hdac3 also interacts with the canonical Wnt/␤-catenin signaling pathway in that suppression of Hdac3 prevents nuclear translocation of ␤-catenin (59). Canonical Wnt signaling was one of the most differentially regulated pathways in Hdac3 CKO Osx mice calvaria as compared with WT littermates, and several Wnt inhibitors (Axin2, Ror2, Sfrp4) were among the most up-regulated genes (29). Our data suggest that Hdac3 is essential for the mechanism by which Runx2 represses Axin2, as Axin2 levels were strongly transcribed in Hdac3-insufficient mice.
Runx2 haploinsufficient mice mimic features of human cleidocranial dysplasia in that they present with hypoplastic clavicles and delayed cranial suture development with persistent fontanels in the calvaria (17)(18)(19). Runx2 is a stimulated by canonical Wnt/␤-catenin signaling (49), and ␤-catenin signaling is enhanced by Axin2 deficiency (10, 12, 16). Our gene expression data from BMSC suggest that double mutant Axin2 Ϫ/Ϫ :Runx2 ϩ/Ϫ osteoblasts may trend toward expressing higher levels of Runx2 as compared with osteoblasts from Runx2 ϩ/Ϫ single mutants (Fig. 5B). Inhibition of Gsk-3␤ (the kinase that targets ␤-catenin for proteasomal degradation) also increases ␤-catenin signaling (60). Interestingly, mice heterozygous for both Runx2 and Gsk-3␤ had larger clavicles and improved calvarial development as compared with the Runx2 heterozygous single mutant animals (61). However, despite the important role of Gsk-3␤ in canonical Wnt signaling, the rescue of the Runx2 ϩ/Ϫ cleidocranial dysplasia phenotype in these mice was explained by reduction of the kinase activity of Gsk-3␤ toward Runx2 and not because of increases in Wnt signaling (61). Thus, it remains unclear whether increasing ␤-catenin signaling could rescue the cleidocranial dysplasia phenotype of the Runx2 heterozygous animals. Future studies will address whether osteoblasts in the Axin2 Ϫ/Ϫ :Runx2 ϩ/Ϫ mice demonstrate increased ␤-catenin signaling and whether this increase could be sufficient to rescue the cleidocranial dysplasia phenotype of the Runx2 ϩ/Ϫ mice.
In conclusion, a novel relationship exists between Runx2, Axin2, and Hdac3 that regulates osteoblast activity in the cranial sutures. Runx2 binds and represses Axin2 expression in an Hdac3-dependent manner in osteoblasts. As a consequence of this relationship, reducing levels of either Runx2 or Hdac3 increases Axin2 expression. Runx2 is essential for Axin2 deficiency-induced craniosynostosis, as introduction of Runx2 haploinsufficiency into the Axin2 Ϫ/Ϫ background rescues the craniosynostosis phenotype of the Axin2 Ϫ/Ϫ single mutant mice. These data shed light on new mechanisms involved in cranial suture biology and may have further-reaching implications in the fields of skeletal and cancer biology. Additional studies will be required to determine the role of ␤-catenin signaling in this model and to quantify the effects of introducing Axin2 deficiency into the Runx2 ϩ/Ϫ background with respect to measures of the cleidocranial dysplasia phenotype of Runx2 haploinsufficient mice.