Direct Interactions of Runx2 and Canonical Wnt Signaling Induce FGF18*

Canonical Wnt signaling is clearly required for skeletal development and bone formation. However, the targets of Wnt signaling that convert this signal into bone are unclear. Identification of these targets will yield insight into normal bone physiology and suggest new therapeutics for treatment of bone disease. Here we show that an essential regulator of bone development, FGF18, is a direct target of canonical Wnt signaling. A single DNA binding site for the Wnt-dependent transcription factors TCF/Lef accounted for the stimulation of the fgf18 promoter in response to Wnt signaling. Additionally, targeted disruption of βcat blocked fgf18 expression in vivo. Partially overlapping the TCF/Lef binding site is a Runx2 binding site and experiments showed that Runx2 and TCF/Lef work cooperatively to induce fgf18 expression. RNA interference knockdown of Runx2 inhibited and Runx2 forced expression augmented the induction of fgf18 by canonical Wnt signaling. Significantly, Runx2 formed a complex with Lef1 or TCF4 and this complex bound the composite binding site in the fgf18 promoter. These results demonstrate that two transcription pathways that are essential for bone, physically and functionally converge at the fgf18 promoter.

The signaling cascade initiated by the Wnt family of polypeptides controls normal and abnormal development in a variety of tissues (1)(2)(3). Wnt signaling is initiated by the binding of the extracellular ligand (Wnt) to a seven-transmembrane spanning receptor and coreceptor. The receptor complex is comprised of frizzled, a multiple membrane spanning protein and an low density lipoprotein-related coreceptor, LRP5 or LRP6. The receptor-ligand complex initiates a complex cascade of events. Signals downstream of the Wnt receptor complex are separated into canonical and noncanonical pathways (4). Canonical Wnt (cWnt) 2 signaling regulates the protein levels of ␤-catenin (␤cat). Within the cytosol ␤cat resides in a multisubunit complex that includes the proteins axin, APC, glycogen-synthase kinase 3 (GSK3) as well as others (5). In the absence of Wnt, GSK3 phosphorylates specific residues near the amino terminus of ␤cat, leading to its recognition and degradation by a ␤TrCP-dependent proteosome pathway. When Wnt engages its receptor-coreceptor complex, the protein disheveled is recruited to the membrane, phosphorylated, and directed to the APC-AXIN-GSK-␤cat complex where it represses GSK3 activity. Inhibition of GSK3 leads to accumulation of ␤cat and migration to the nucleus. Within the nucleus ␤cat interacts with specific DNA binding transcription factors, most notably the TCF/Lef family of transcription factors (6). When associated with TCF/Lef, ␤cat recruits other transcription cofactors and stimulates the expression of Wnt target genes.
The TCF/Lef transcription factors are a family of 4 proteins that bind to DNA through a conserved high mobility group domain (7). TCF/Lef proteins transduce cWnt signals into changes in gene expression. In the absence of Wnt, ␤cat levels are low and TCF/Lef proteins act as transcription repressors by recruiting the general transcription inhibitors Groucho/TLE and histone deacetylases (8,9). Accumulation of ␤cat in the nucleus leads to direct displacement of Groucho/TLE through the competitive binding of ␤cat to TCF/Lef (10). ␤cat activates gene expression by recruiting coactivators like p300/CBP (11) and the Mediator complex (12). cWnt signaling has profound effects on normal cell and tissue development and directly contributes to pathological conditions like neoplasia. During bone formation and remodeling, Wnt signaling is indispensable. Studies of genetically modified mice have contributed much to our understanding of the many functions of Wnt signaling in the skeleton. Gene knockouts of individual Wnts have shown requirements for Wnt10b (13), Wnt5a, -5b (14), and others during normal skeletal development. To study cWnt signaling during bone growth and remodeling, ␤cat has been conditionally inactivated in several studies. These studies have proven that Wnt signaling is required at several steps during osteoblast differentiation. Inactivation of ␤cat at early steps of osteoblast formation using cre recombinase expressed from the Dermo1, type I collagen, and Prx1 promoters caused aborted osteoblast differentiation (15)(16)(17). Early osteoblasts expressing the osteoblast transcription factor Runx2 were present, but mature osteoblasts and synthesis of bone matrix was absent. Other studies showed ␤cat requirements at later stages of osteoblastogenesis (18,19); mature osteoblasts failed to develop when ␤cat was deleted at a stage when the transcription factor osterix is expressed, indicating a requirement for cWnt downstream of or coinciding with * This work was supported by National Institutes of Health Grant AR050024 and the Paul Beeson Physician Faculty Scholars in Aging Research Program. The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked "advertisement" in accordance with 18 U.S.C. Section 1734 solely to indicate this fact. osterix. In addition, cessation of cWnt signaling is required for osteoblast maturation (18). Finally, cWnt signaling also regulates bone osteoclast differentiation and bone resorption by regulating osteoprotegerin expression (20). Naturally occurring mutations in humans have proven the relevance of Wnt signaling during osteoblast development and for controlling bone mass. A syndromic form of osteoporosis is caused by loss of function mutations in the Wnt coreceptor LPR5 (21), and a high bone mass phenotype results from mutations in LRP5 that diminish its affinity for Dkk1, an antagonist of Wnt (22,23). Thus, in accord with the animal studies, Wnt signals are anabolic for bone and decreased Wnt signaling result in clinically significant bone deficiencies.
These findings have proven the significance of cWnt signaling during osteoblast differentiation. Moreover, these studies suggest that careful stimulation of cWnt signaling or antagonism of ␤cat decay will promote bone formation. Identification of the Wnt target genes that are required for osteoblasts formation in response to cWnt signaling will also signify important pathways that can be modulated for therapeutic benefit of bone disease. In this report we show that the fibroblast growth factor 18 gene (fgf18) is a downstream target of cWnt signaling. FGF18 is expressed in a tissue domain that coincides with (i) sites of osteoblast development and (ii) regions where ␤cat levels (and therefore Wnt signaling) are increased. This implied that fgf18 is a candidate for a Wnt target gene that translates cWnt signaling into osteoblast differentiation. We find that fgf18 is directly induced by cWnt signaling. TCF/Lef proteins bind to a consensus target sequence of the fgf18 promoter and when stimulated by ␤cat induce fgf18 expression. Remarkably, the expression of fgf18 is directly coupled to another essential osteoblast transcription factor, Runx2. Partially overlapping the TCF/Lef site of the FGF18 promoter is a recognition motif for Runx2. Experiments showed that Runx2 interacts with TCF/Lef and forms a ternary complex at the FGF18 promoter. RNA interference experiments showed that Runx2 is necessary for stimulation of FGF18 expression by Wnt, demonstrating that Wnt signaling and Runx2 are both physically and functionally linked. These findings reveal exquisite control of fgf18 expression by two essential osteoblast transcription factors that combine to determine tissue specificity and stimulus dependence. These findings suggest that FGF18 may function as an essential component of the Wnt-dependent steps of osteoblastogenesis that are genetically downstream of Runx2.

MATERIALS AND METHODS
Reagents-SB216763 was from Tocris (Cookson, MO). Wnt3a conditioned medium was prepared from Wnt3a expressing L-cells according to the suppliers instructions (ATCC, Manassas, VA). Conditioned medium without Wnt3a was made using L-cells.
Plasmids-To create the Ϫ3.3-kb fgf18 reporter plasmid a 3.4-kb MscI fragment of BAC clone RPI 23323L8 was blunt-end cloned into the SmaI restriction site of pGL3basic (Promega, Madison, WI). For the Ϫ1.1-kb reporter plasmid a PvuII/KpnI fragment was excised from the Ϫ3.3-kb plasmid and blunt-end ligated into the SmaI site of pGL3basic. The Ϫ0.8-kb reporter was prepared by SacI digestion of the Ϫ3.3-kb reporter that was cloned into pGL2basic rather than pGL3basic. The released fragment was gel purified and cloned into the SacI site of pGL3basic. MluI digestion of the Ϫ3.3-kb pGL2basic reporter resulted in a DNA fragment that was cloned into the MluI site of pGL3basic. To produce the trimeric repeats, gel purified oligonucleotides or 415-bp fragments having BamHI or BglII sites at opposite ends were kinased, ligated, and fractionated by agarose gel electrophoresis. The trimer was excised from the gel and cloned into the BglII site of pGL3 TATA, a TATA box containing luciferase reporter. pGL3 TATA was created by cloning the oligonucleotide AGATCTGGGTATATAAG-GATCCGGTAAAGCTT and the reverse complement into the BglII/HindIII sites of pGL3basic. Lef1 was cloned into the StuI site of pCS2ϩ following MluI/HindIII digestion and Klenow fill-in of the image clone 6054813. Runx2 (form II) was PCR amplified from a mouse cDNA library using the primers CCGGTACCACCATGCTTCATTCGCCTCA-CAAA and TCTAGATCAATATGGCCGCCAAACAGAC-TCATCCATTCTGCCGCTAGAATTCAA and cloned into pcDNA3.1 (Invitrogen).
Site-directed Mutagenesis-Mutations of the TCF binding site were introduced into the Ϫ1.1-kb and (415) 3 reporter plasmids via QuikChange site-directed mutagenesis following the manufacturer's instructions (Stratagene, La Jolla, CA). The mutations were verified by DNA sequencing. The sequence of the mutagenesis primers is available on request.
Metatarsal Cultures-Metatarsal were dissected from E15.5 day C57Bl/6 embryos and cultured in serum-free medium as described previously (24). When treated with Wnt3a, the medium was exchanged with conditioned medium (with or without Wnt3a) and the samples were harvested after 24 h.
In Situ Hybridization-4-m sections of paraffin-embedded tissues were processed and hybridized with riboprobes as described previously (25). The FGF18 riboprobe plasmid was a gift from D. Ornitz (St. Louis, MO).
Real Time Quantitative-PCR-Total RNA was prepared from cells using RNeasy according to the manufacturer's instructions (Qiagen). One g of total RNA was reverse transcribed as described (28). Real time PCR was done using a SYBR Green PCR mixture (Applied Biosystems) in an ABI 7000 sequence detection system (Applied Biosystems). The sequence of the primers were as reported (28). Ubiquitin was used as the normalizer.

Runx2 and Wnt Induce FGF18
RNA Interference-The small interfering RNA oligonucleotide against mouse Runx2 was targeted to GTCCTATGAC-CAGTCTTAC. A negative control shRNA oligonucleotide directed against TTCTCCGAACGTGTCACGT, unrelated to Runx2, was used. The BD Knock-out RNAi System (BD Bioscience) was used for generating small interfering RNAs. Briefly, a double-stranded DNA oligonucleotide containing the shRNA and BamHI and EcoRI overhangs was cloned into the corresponding sites of RNAi-Ready pSIREN-Shuttle vector. The construct was verified by sequencing. MC3T3E1 cells were transiently transfected with shRNA plasmids using Nucleofection (Amaxa Inc., Gaithersburg, MD). 10 6 cells were resuspended in 100 l of Nucleofector solution T and transfected with 5 g of DNA using Program T-20. Transfection efficiency using these conditions was shown to be Ͼ80%. RNA was prepared 24 h post-transfection from cells using an RNeasy (Qiagen) kit according to the manufacturer's instructions.
Chromatin Immunoprecipitation-MC3T3 cells at about 80% confluence were stimulated for 4 h with 20 M SB216763. Thereafter, protein-DNA complexes were cross-linked with 1% formaldehyde for 10 min at room temperature and cross-linking was stopped by the addition of glycine to a final concentration of 0.125 M. Cells were washed with phosphate-buffered saline containing 1 mM phenylmethylsulfonyl fluoride, pelleted at 1200 rpm, and solubilized in swelling buffer (5 mM Pipes, pH 8.0, 85 mM KCl, 0.5% Nonidet P-40, protease inhibitor mixture), incubated on ice for 20 min, and then Dounce homogenized. The nuclei were collected by microcentrifugation and then resuspended in sonication buffer (1% SDS, 10 mM EDTA, 50 mM Tris-HCl, pH 8.1, 0.5 mM phenylmethylsulfonyl fluoride, and protease inhibitor mixture) and incubated on ice for 10 min. The samples were sonicated on ice with a Branson Digital Sonifier at 50% amplitude for three 10-s pulses to an average length of ϳ1,000 bp and then microcentrifuged. The sonicated cell supernatant was diluted 4-fold with RIPA buffer (0.1% SDS, 0.1% sodium deoxycholate, 1 mM EDTA, 0.5 mM EGTA, 140 mM NaCl, 10 mM Tris, pH 8) and precleared for 2 h at 4°C with 2 mg/ml salmon sperm DNA, 10 mg/ml bovine serum albumin, and 30 l of Protein G-Sepharose. Prior to use, Protein G-Sepharose was blocked with 0.2 mg/ml of sheared salmon sperm DNA and 1 mg/ml of bovine serum albumin for at least 4 h at 4°C. Precleared chromatin was incubated with antibody Immunoprecipitation-HEK293 cells were transfected with expression plasmids using a modified calcium phosphate precipitation method (26,27). 48 h after transfection the cells were lysed with 1% deoxycholate and 1% Triton X-100, 50 mM Tris, pH 8.0, 0.15 M NaCl, 2 mM EDTA, and 50 g/ml phenylmethylsulfonyl fluoride (Sigma), 0.25 mM orthovanadate (Sigma), and protease inhibitor mixture (Sigma). Clarified lysates were precleared with Protein G-Sepharose (Amersham Biosciences), and then immunoprecipitated with anti-Lef1 (Santa Cruz; sc8591), anti-␤cat (Santa Cruz; sc7199), or anti-hemagglutinin (Sigma; H6908) absorbed to protein G-Sepharose (Amersham Biosciences). Proteins were fractionated by SDS-PAGE and transferred to polyvinylidene difluoride membranes (Milli-FIGURE 2. Regulation of the FGF18 promoter. A, luciferase assays of FGF18 reporter constructs containing the indicated lengths of the proximal promoter cotransfected into HEK293 cells with an expression plasmid for a stabilized form of ␤cat. Relative light units represent the ratio of luciferase to ␤-galactosidase activity. ␤-Galactosidase activity is derived from a cotransfected plasmid that has a constitutive promoter. B, luciferase reporter assays for the indicated FGF18 reporter plasmids transfected into MC3T3E1 cells and stimulated for 16 h with conditioned medium with or without Wnt3a. C, results of luciferase reporter assays from HEK293 cells transfected with the indicated plasmids. (415) 3 represents a head to tail trimer of the 415-base pair motif cloned into a plasmid containing a TATA box minimal promoter (vector ϭ pGL3TATA). Assays were performed in triplicate, and the error bars represent Ϯ S.E. The figures show data representative of three independent experiments. D, sequence of the 415-bp motif that is regulated by cWnt signaling. The putative TCF/Lef binding sites are enclosed by rectangles. E, sequence of the oligonucleotides used in luciferase reporter plasmids or EMSA binding experiments. The TCF/Lef site is bold type. The Runx2 site is italic. Mutations are indicated by lowercase text. pore). Blots were probed with anti-Runx2 (Santa Cruz; sc8566) and detected by chemiluminescence with the West Dura Extended substrate (Pierce).
Nuclear Extracts and EMSAs-Nuclear extracts were prepared as previously described (26,27). The protein concentration of extracts was estimated using the BCA protein assay reagent (Pierce). All oligonucleotides encompassing the desired binding site sequences were gel-purified and subsequently annealed. Binding reactions were performed at room temperature for 20 min in a 20-l total volume, which consisted of 0.1 pmol of probe (25,000 cpm), 3 g of bovine serum albumin (Promega), 1.25 g of poly(dI-dC) (Amersham Biosciences), 2-4 g of nuclear protein extract, and buffer D (20 mM HEPES, pH 7.9 (4°C), 25% glycerol, 420 mM NaCl, 1.5 mM MgCl 2 , 0.2 mM EDTA, 0.5 mM phenylmethanesulfonyl fluoride, 0.5 mM dithiothreitol). Duplex oligonucleotide competitors (specific and nonspecific) were preincubated with the nuclear extract for 5-10 min at room temperature for competition assays. For supershift assays, antibodies were added to nuclear extract and incubated on ice for 20 min, followed by incubation at room temperature for 20 min. The resulting complexes were resolved as described (26). Anti-Runx2 (sc8566X) was obtained from Santa Cruz.
Animals-The Dermo-1cre deleter strain has been described (29) and mice carrying a floxed allele of ␤cat (30) were obtained from The Jackson Laboratory (Bar Harbor, ME). The animals were maintained in accordance with protocols approved by the Animal Care and Use Committee at University of Texas Health Science Center, San Antonio.

RESULTS
We examined by indirect immunofluorescence the presence of FGF18 and ␤cat in metatarsal bone rudiments isolated from E15.5 mouse embryos. The rudiments were cultured in serumfree medium and differentiated spontaneously. Interestingly, FGF18 protein was found at sites in the perichondrium that coincide with sites of osteoblast differentiation and increased levels of ␤cat (Fig. 1, A and B, and color micrographs in supplementary data Fig. S1). We previously determined that FGF18 expression is induced following inhibition of GSK3 (24). We therefore reasoned that FGF18 is induced by cWnt signaling. In accord with this, fgf18 expression was induced in the perichondrum of metatarsals cultured in medium containing Wnt3a (Fig. 1, C and D) and in cultured cells (Fig. 1E). To further characterize the regulation of fgf18 by canonical Wnt signaling, we examined fgf18 expression in bones devoid of ␤cat. ␤cat was conditionally inactivated in the developing bones by crossing mice with floxed alleles of ␤cat (30) to the credeleter strain Dermo1-cre (29). Consistent with the view that canonical Wnt signaling directly induces FGF18 expression, FGF18 transcripts were absent in the perichondrium of bones lacking ␤cat (Fig. 1, F and G).
To identify the specific targets of Wnt signaling in the fgf18 gene we cloned various portions of the fgf18 promoter into a luciferase reporter vector. Transient transfection of these reporter constructs together with a proteosome-resistant form of ␤cat or following stimulation of the transfected cells with Wnt3a demonstrated that a 415-base pair motif between Ϫ0.8 and Ϫ0.5 kb is stimulated by ␤cat or Wnt3a (Fig. 2, A and B). This 415-bp sequence was sufficient to respond to ␤cat when cloned into a TATA box containing minimal promoter (Fig.  2C). Stimulation of this sequence by ␤cat was enhanced when the Wnt-dependent transcription factor, TCF4, was cotransfected (Fig. 2C). These data indicate that the 415-bp motif is stimulated by cWnt signaling; therefore, we examined this sequence for putative TCF/Lef recognition sites (Fig. 2D). Three potential binding sites (A, B, and C) were identified. To test the function of these we designed oligonucleotides spanning these sites and determined if 1) the DNA sequence associates with the DNA binding domain of TCF4, and 2) ␤cat-dependent transcriptional activity is apparent in a transient transfection assay. Only site A interacted specifically with the DNA binding domain of TCF4 and showed transcriptional activity when cloned as a trimeric repeat in a luciferase reporter containing a TATA box minimal promoter (Fig. 3, A and B). Moreover the activation of this sequence by ␤cat was blocked by a form of TCF4 lacking the amino-terminal domain required for interaction with ␤cat (Fig. 3B). We then mutated the TCF/ Lef site to test its significance. Mutation of the consensus site completely abolished activation by Wnt3a (Fig. 3C) and ␤cat

Runx2 and Wnt Induce FGF18
(data not shown), both of a trimer of the oligonucleotide containing site A, as well as a trimer of the entire 415-bp motif.
To determine whether site A can account for activation of the endogenous fgf18 promoter, we introduced the same mutation into the TCF/Lef binding site of the 1.1-kb fgf18 promoter. This mutation also completely abolished the stimulation of the FGF18 promoter by Wnt3a (Fig. 3C), supporting the view that this site is fully responsible for the activation of the fgf18 gene by Wnt. To determine whether TCF4 and ␤cat interact with the FGF18 promoter in vivo we performed chromatin immunoprecipitation. Following stimulation of MC3T3E1 cells with or without the GSK3 antagonist SB216763 to mimic cWnt signaling, chromatin immunoprecipitation showed ␤cat associated with the FGF18 promoter (Fig. 3D, upper). Similarly, TCF4 were associated with the FGF18 promoter in HEK293 cells transfected with HA-tagged TCF4 (Fig. 3D, lower).
We identified a sequence that regulates the expression of fgf18 in response to Wnt. However, Wnts are widely expressed, yet fgf18 expression is much more restricted. This suggested that additional transcription regulators contribute to the expression of fgf18, and that these regulators may functionally cooperate with ␤cat and TCF/Lef to induce fgf18 expression. Significantly, fgf18 expression is greatest in the perichondrium. We then questioned if transactivators expressed in perichondrial cells also regulate fgf18 expression. We searched for clues concerning fgf18 expression by examining the DNA sequence surrounding the TCF/Lef site. Interestingly, a putative Runx2 binding site (TGTGG) partially overlaps the TCF/Lef site (Fig. 2E). Notably, Runx2 is expressed in the perichondrium, and therefore is a good candidate for a bone cellspecific transactivator that may restrict fgf18 expression to the perichondrium.
To examine effects of Runx2 on fgf18 expression during Wnt activation, we transfected MC3T3E1 osteoblasts with Runx2 and simulated cWnt signaling by inhibiting GSK3 with the pharmacological antagonist SB216763. We chose a concentration of SB216763 that produced submaximal induction of fgf18. Significantly, the combination of Runx2 and SB216763 induced fgf18 to a greater degree (9-fold) than Runx2 in the absence of SB216763 (2.8-fold; Fig. 4A). Similar results were obtained with Wnt3a (Fig. 4A, right). Fig. 4B shows that Runx2 did not stimulate the expression of Axin2, a gene that is strongly induced by canonical Wnt signaling (31). These data therefore suggest a cooperative relationship between Runx2 and TCF/Lef at the FGF18 promoter. This cooperation does not, however, extend to all Wnt-regulated genes, as evidenced by the absence of an effect on Axin2. In fact as shown in Fig. 4B, Runx2 may repress Wnt-dependent induction of Axin2. To further analyze the interaction of Wnt-dependent transcription and Runx2, we diminished Runx2 protein levels by RNA inter-   ). B, luciferase assays of the Topflash reporter cotransfected in HEK293 cells with the indicated plasmids (␤cat, 0.05 g; Runx2, 0.3 g). C, luciferase reporter assay of the trimer of the A oligonucleotide with a mutation of the Runx2 binding site. HEK293 cells were cotransfected with or without ␤cat (0.05 g) and Runx2 (0.05, 0.1, or 0.3 g). Relative light units are the ratio of luciferase to ␤-galactosidase activity. ␤-Galactosidase activity is derived from a cotransfected plasmid with constitutive expression. Assays were performed in triplicate, and the error bars represent Ϯ S.E. D, PCR for the fgf18 promoter sequence following chromatin immunoprecipitation of Runx2 from MC3T3E1 cells stimulated for 4 h with 20 M SB216763. Input represents a positive control using the starting material prior to immunoprecipitation. IgG represents immunoprecipitation using purified rabbit immunoglobulin. FEBRUARY 9, 2007 • VOLUME 282 • NUMBER 6 ference knockdown. Knockdown of Runx2 suppressed induction of FGF18 by the GSK3 antagonist and Wnt3a (Fig. 4C). In contrast, the induction of Axin2 was only modestly effected (Fig. 4D). These data support a functional interdependence of Runx2 and Wnt-dependent transcription at certain target loci like fgf18. The proximity of the TCF/Lef and Runx2 sites of the fgf18 promoter suggests that these transcription factors may interact directly. To examine this more closely we asked if Runx2 stimulates the FGF18 luciferase reporter plasmids. Fig.  4E shows that Runx2 alone did not stimulate the 3.3-kb FGF18 luciferase reporter (nor fragments thereof; data not show). However, expression of Runx2 together with ␤cat led to dosedependent stimulation of the 3.3-kb fgf18 promoter (Fig. 4F). At submaximal levels of ␤cat the 3.3-kb FGF18 reporter was stimulated 5.3-fold, whereas ␤cat plus Runx2 stimulated the promoter to 16.6-fold. The 1.1-kb FGF18 reporter was similarly induced by the combination of ␤cat and Runx2 (Fig. 4G). The costimulation by ␤cat and Runx2 required an intact TCF/Lef site. Mutation of this site abolished stimulation by the combination of Runx2 and ␤cat (Fig. 4H). The trimer of the A oligonucleotide and the 417-bp motif ( Fig. 5A and data not shown) were also coactivated by ␤cat and Runx2. Fig. 5A shows that whereas Runx2 alone did not stimulate the reporter, Runx2 combined with ␤cat caused a 40-fold activation. This compares with a 9.8-fold stimulation by the indicated amount of ␤cat. Significantly, a TCF/Lef binding site was insufficient for coactivation by Runx2 and ␤cat. In fact, the Topflash reporter, which contains repeats of an optimum TCF/Lef binding site was inhibited rather than activated by the combination of ␤cat and Runx2 (Fig. 5B). These data predict that Runx2 interacts specifically with the sequence of fgf18. To examine this, we constructed a reporter wherein the potential Runx2 binding site was mutated (A_Runx mut, see Fig. 2E). Supporting a direct interaction with the Runx2 site, the mutated reporter was not stimulated by Runx2, yet remained inducible by ␤cat, albeit less so (Fig. 5C). Chromatin immunoprecipitation yields additional evidence that Runx2 interacts with fgf18. Fig. 5D shows that Runx2 is immunoprecipitated with the endogenous fgf18 promoter at the expected site.

Runx2 and Wnt Induce FGF18
These data (Runx2 overexpression, Runx2 shRNA, cotransfection of Runx2 and ␤cat) demonstrate a close functional relationship of Runx2 and the TCF/Lef transcription factors on fgf18. Because the TCF/Lef and Runx2 sites partially overlap, the respective transcription factors may physically interact to regulate the expression of FGF18. To examine this we performed co-immunoprecipitation experiments. First, we examined if ␤cat interacts with Runx2. If ␤cat binds to Runx2, this could explain our finding that ␤cat is required for stimulation of the fgf18 reporters by Runx2. HEK293 cells were transfected with Runx2 or Runx2 and ␤cat. ␤cat was immunoprecipitated and immunoblots for Runx2 were done. The results did not reveal interactions of ␤cat and Runx2 (Fig. 6A). However, Runx2 binds to both Lef1 and TCF4. Fig. 6, B and C, shows that Runx2 was co-immunoprecipitated with either Lef1 or HAtagged TCF4. These results demonstrated association of TCF4 or Lef1 with Runx2 in the absence of DNA binding. We next asked whether the proteins also assemble on DNA from the fgf18 gene. EMSA experiments were done using oligonucleo-tide A and in vitro transcribed and translated Runx2, ␤cat, and Lef1. Fig. 7A shows that Runx2, ␤cat, or combined Runx2 and ␤cat did not form complexes with radiolabeled oligonucleotide A. As expected, Lef1 bound significantly to oligonucleotide A. The combination of ␤cat and Lef1 showed complexes consistent with Lef1 and Lef1-␤cat. Addition of Runx2 to these components led to DNA complexes consistent with Lef1-Runx2 and Lef1-Runx2-␤cat (Fig. 7, A and B). Addition of an unlabeled oligonucleotide competitor showed these complexes were specific, given that a wild-type DNA sequence was an effective competitor, but an oligonucleotide with a mutation of the TCF/Lef site was not (Fig. 7A). Therefore these data are consistent with the formation of a complex containing: Runx2-␤cat-Lef1-DNA.
Runx2 did not appear to bind to the DNA sequence of oligo A in the absence of Lef1. Therefore it was not clear if the affinity of Runx2 for DNA was augmented through an association with Lef1 or if Runx2 assembled in a DNA-containing complex solely through protein-protein interactions with Lef1. To determine whether Runx2 binds to oligonucleotide A we did additional competition experiments. For these experiments we prepared nuclear extracts from cells transfected with vector DNA or an expression clone for Runx2. These extracts were incubated with a radiolabeled oligonucleotide containing the Runx2 binding site of the osteocalcin promoter, OSE (32). Addi-FIGURE 6. Runx2 associates with TCF and Lef1. A, immunoprecipitation of ␤cat followed by Runx2 immunoblotting. Lysates prepared from HEK293 cells transfected with the indicated plasmids were immunoprecipitated (IP) with anti-␤cat. Lysate and Runx2 immunoprecipitation demonstrate the mobility and immunodetection of Runx2. The lower band in the immunoprecipitated samples represents Ig heavy chain. B, immunoprecipitation with anti-Lef1 followed by Runx2 immunoblotting. Lysates prepared from HEK293 cells transfected with the indicated plasmids were immunoprecipitated with anti-LEF1 antibody. Note that Runx2 is co-immunoprecipitated with Lef1. The lower band in the immunoprecipitated samples represents Ig heavy chain. C, immunoprecipitation of Runx2 with TCF4. Immunoprecipitation with anti-HA followed by Runx2 immunoblot. Lysates were prepared from HEK293 cells transfected with the indicated plasmids. Note that Runx2 is co-immunoprecipitated with HA-TCF4. The lower band in the immunoprecipitated samples represents Ig heavy chain.
tionally, in the indicated samples we included an anti-Runx2 antibody that produced slower migrating complexes that were more evident on the nondenaturing gels. Fig. 7C shows that oligo A competed for the binding of Runx2 to OSE; both supershifted and nonsupershifted species were disrupted. Competition required an intact Runx2 binding site, as an oligonucleotide containing a mutation in the Runx2 binding site (A_Runx mut ) did not interfere with binding of Runx2 to OSE. Additional control lanes show that unlabeled OSE, but not OSE with a mutated Runx2 site acted as an effective competitor. These data show that Runx2 binds directly to DNA at the FGF18 promoter and support the view that interactions with Lef1 facilitate Runx2 binding to DNA.

DISCUSSION
We showed that the expression of fgf18 is regulated by the combined actions of Wnt-dependent transcription factors TCF/Lef and Runx2. This reflects functional and physical inter-actions of TCF/Lef and Runx2 at a composite binding site for both transcription factors in the FGF18 promoter. We found that Runx2 and TCF4 or Lef1 form a complex in the absence of DNA and that this complex promotes the binding of Runx2 to its recognition motif in the FGF18 promoter. It is likely that TCF/Lef induces a conformational change in Runx2 that allosterically activates Runx2 DNA binding activity. This is analogous to the interactions of Runx1 and core binding factor ␤ (CBF␤). Binding studies of the runt domain of Runx1 and CBF␤ showed a 5-fold decrease in the Runx1-DNA dissociation constant when Runx1 is bound to CBF␤. NMR studies demonstrated changes in the amide bond backbone of Runx1 upon binding to CBF␤ and the crystal structure show that Runx-CBF␤ has stabilized loop conformations at the site of DNA contact. CBF␤ interacts with the runt domain of Runx1 principally through two surface domains that are opposite the DNA binding interface. This same surface may interact with non-DNA binding domains of TCF4 or Lef1. The DNA bending induced by Lef1 and TCF4 at its recognition site may permit simultaneous protein-protein and DNA-protein interactions, and enhanced affinity of Runx2 for DNA. Whether the high mobility group domain of TCF/Lef and the runt domain of Runx2 are sufficient for this complex is unclear. The amino-terminal glutamine/ alanine-rich region of Runx2 appears to prevent interactions with the Runx1 binding partner, CBF␤ (33). Perhaps in addition to altering affinity for CBF␤, this domain may selectively contribute to binding with TCF/Lef proteins.
After a Runx2-TCF/Lef complex is bound to DNA how does this complex stimulate the expression of FGF18? Clearly, ␤cat is a key coactivator. The interactions of ␤cat with TCF/Lef are well recognized and in this way histone acetyltransferases like p300 (11) are recruited to the promoter. If ␤cat binds to Runx2, then the activator properties of Runx2 at the FGF18 promoter could be explained by the additional binding energy for ␤cat that is contributed by Runx2. However, ␤cat did not interact with Runx2 by coimmunoprecipitation and gel-shift experiments did not demonstrate disproportionate binding of ␤cat to Lef1-Runx2 complexes. Therefore, we speculate that Runx2 activates the FGF18 promoter through mechanisms distinct from the recruitment of ␤cat. Runx2 may activate expression FIGURE 7. Assembly of a Lef1-Runx2-␤cat-DNA complex. A, EMSA with radiolabeled oligonucleotide A. The indicated in vitro transcribed and translated proteins were incubated with labeled oligonucleotide and then fractionated on a non-denaturing gel. Where indicated a 100-fold molar excess of unlabeled competitor was added. B, EMSA showing distinct Lef1/Runx2 and Lef1-␤cat-Runx2 complexes. C, EMSA using nuclear extract prepared from cells transfected with Runx2 or a control plasmid and radiolabeled OSE oligonucleotide from the osteocalcin promoter. Where indicated, anti-Runx2 antibody was included to produce supershifted complexes. Where indicated, a 100-fold excess of unlabeled competitor oligonucleotide was added.
through actions such as recruitment of histone acetyltransferases like MOZ and MORF (34) or binding to other transcription factors like AP-1 or Smad (35,36). Alternatively, Runx2 may stimulate fgf18 expression through pathways not yet described.
Significantly, a previous report showed that the human FGF18 promoter is stimulated by a ␤cat-TCF complex (37) and that FGF18 contributed to colon cancer cell growth. The sequence of the mouse and human promoters at the TCF/Lef binding site is 100% conserved supporting the significance of this site for FGF18 gene expression. This implies that FGF18 expression is normally targeted to sites where cWnt signaling and Runx2 (and perhaps other members of the Runx family) coincide. We have shown that ␤cat is required for the expression of fgf18 in the perichondrium of long bones during skeletal development. This suggests that both Runx2 and cWnt signaling are required for fgf18 expression under normal physiological states; the absence of either may result in the loss of expression. That is, the physical coupling of TCF/Lef and Runx2 and the juxtaposition of the DNA binding sites specify a cooperative partnership for fgf18 expression. Accordingly, recent data show that Runx2 is essential for fgf18 expression and that Runx2 directly regulates the FGF18 promoter (38). During pathological conditions like colon cancer, fgf18 may be expressed abnormally and independently of Runx2 as a consequence of excess levels of ␤cat.
The interactions of Runx2 and TCF/Lef do not augment the expression of all Wnt target genes. For example, Axin2 is strongly induced by canonical Wnt signaling, yet its expression is not stimulated by Runx2 (Fig. 4B). Also, the prototypic Wntdependent reporter plasmid, Topflash, is repressed rather than stimulated by the combination of Runx2 and ␤cat (Fig. 5B). Kahler and Westendorf (39) also showed that Runx2 binds to Lef1 and this association represses the osteocalcin promoter. Therefore, it is likely that a set of genes are repressed as a consequence of Runx2-TCF/Lef interactions and a different set of genes including fgf18 are induced by the combination. The outcome will be determined by protein-nucleic acid interactions that are promoter-specific. Given the pivotal importance of Runx2 and Wnt-dependent transcription during osteoblast differentiation, the decision to repress or induce gene expression in response to their combined actions will be paramount for temporal and spatial regulation of osteoblast differentiation. For example, recent data show that cessation of Wnt/␤cat activity is required for full maturation of osteoblasts into osteocalcin expressing cells (18). Coactivation of FGF18 expression by Runx2 and Wnt signaling may support early osteoblast differentiation, but suppress later events. In support of this, we have shown that FGF18 can suppress osteoblast differentiation in cultured metatarsals (24). Alternatively, strong Wnt signaling may induce TCF1 and/or Lef1 expression that lead to dimerization with Runx2 and dissociation from Runx2dependent promoters like osteocalcin (39). In either case, these possibilities imply that a Runx2-TCF/Lef complex will act decisively at specific points during osteoblast differentiation.
We showed that fgf18 is induced via an osteoblast-specific, Wnt-dependent pathway. Thus, fgf18 is a Wnt-target gene that may be essential for translating Wnt signaling into osteoblast development. It is notable that normal osteogenesis is disrupted in fgf18 null embryos (40,41). If fgf18 is responsible for the action of Wnt in bone, then based on our understanding of Wnt signaling in bone fgf18 must both stimulate and suppress osteoblast differentiation at different stages of development. How this is achieved will require further investigation; however, expansion of an early osteoblast population through stimulation of cell proliferation will almost certainly be a key attribute of FGF18.
What other genes are induced by the Runx2-TCF/Lef combination? These genes are likely to have both TCF/Lef sites and Runx2 within the promoter; however, given the substantial DNA bending that is induced by the high mobility group domain of Lef (42), we predict that these binding sites need not be directly adjacent or overlapping. The dramatic change in DNA conformation caused by Lef1 can bring a distant Runx2 site into close spatial proximity. We predict that certain of these genes have expression patterns that match closely that of FGF18. Interestingly, Dkk1, tcf1, and Runx2 are induced by Wnt signaling (43)(44)(45) and are expressed in the perichondrium similar to FGF18. Moreover, Dkk1 and tcf1 expression is lost in Runx2 null embryos at sites where Runx2 and Dkk1 or tcf1 are normally coexpressed (46). Whether these and other genes are coactivated by Runx2 and Wnt-dependent transcription demands further investigation. These genes will surely be vital during osteoblastogenesis and bone formation given that they lay at the convergence of stimulus-dependent and osteoblastspecific transcription pathways that are anabolic for bone.