Regulation of Cyclooxygenase-2 and Periostin by Wnt-3 in Mouse Mammary Epithelial Cells*

Wnt family members are critical in developmental processes and have been shown to promote carcinogenesis when ectopically expressed in the mouse mammary gland. The gene expression pattern mediated by Wnt is pivotal for these diverse responses. The Wnt pathway has been conserved among different species. Genetic studies have shown that Wnt effects are mediated, at least in part, by β-catenin, which regulates transcription of “downstream genes.” Wnt stimulation inactivates glycogen-synthase kinase-3β (GSK-3) with subsequent stabilization of β-catenin, which after heterodimerizing with lymphocyte enhancer factor-1/T-cell factor cofactors stimulates transcription. To establish whether Wnt-stimulated transcription is mediated solely by β-catenin, a comparison was made of gene expression profiles in response to Wnt-3, overexpression of β-catenin, and inhibition of GSK-3. Infection of cells with Wnt-3 and inhibition of GSK-3 regulate a set of genes that include cyclooxygenase-2 and periostin. Interestingly, overexpression of β-catenin or reducing β-catenin levels with antisense oligonucleotide transfection did not have any effect on cyclooxygenase-2 or periostin expression, thereby defining a Wnt pathway, which cannot be mimicked by β-catenin overexpression.

The Wnt proteins are a family of secreted cysteine-rich glycoproteins that play an essential role in directing developmental processes such as cell adhesion, cell fate, and cell proliferation (1,2). The Wnt signaling pathway appears to be highly conserved across different species, and genetic and biochemical studies in Caenorhabditis elegans, Drosophila, Xenopus and mammals have contributed to increased understanding of the pathway (for review see Ref. 3). Some of the Wnt genes have been shown to promote mammalian carcinogenesis. Wnt-1 and Wnt-3 were initially identified as mouse mammary oncogenes that became tumorigenic by the insertion of a mouse mammary tumor virus (4,5). Even though a role for Wnt proteins in breast tumorigenesis has been established in mice, this link has not yet been made in human breast cancer. However, components of the Wnt signaling pathway such as adenomatous polyposis coli (APC), 1 a tumor suppressor protein, and ␤-catenin are clearly involved in other forms of human cancers including melanoma, colon, and hepatocellular cancer (6).
Stimulation of cells with Wnt-1 protein results in an increase of cytosolic levels of ␤-catenin as a consequence of the inhibition of glycogen-synthase kinase-3␤. ␤-Catenin then heterodimerizes with a member of the Lef-1/T-cell factor family of transcription factors and induces gene transcription. APC can bind to ␤-catenin and facilitate its degradation when phosphorylated by GSK-3. Deletions in the tumor suppressor protein APC found in colon cancer preclude ␤-catenin degradation, and in so doing the increased level of ␤-catenin can stimulate transcription as described above. Further, stabilizing mutations or deletions in the regulatory N-terminal domain of ␤-catenin also result in increased transcription (3). How this increase in transcription relates to cancer is unclear, but some recently identified ␤-catenin target genes cyclin D1 and the protooncogene c-myc may contribute to neoplastic transformation (7,8).
A number of potential Wnt-1 target genes have been identified in different organisms, including ultrabithorax and engrailed in Drosophila (9,10), nodal-related 3 and siamois in Xenopus (11)(12)(13), Connexin 43 (14), Wisp (15), cyclin D1 (16), and Cox-2 (17) in various mammalian cell lines or tissues. Furthermore, it is not clear whether the entire transcriptional response to the stimulation by Wnt-1 or Wnt-3 is mediated by ␤-catenin-dependent processes. To explore whether there is any evidence for Wnt-stimulated pathways that are independent of ␤-catenin, we compared gene profiles regulated by Wnt-3, ␤-catenin, and inhibition of GSK-3. Cox-2 was up-regulated and periostin (formerly named osteoblast-specific factor-2) was down-regulated by both Wnt-3 and inhibition of GSK-3. However, in contrast, ␤-catenin did not have an effect on the regulation of these two genes, even though it did affect expression of other genes. Importantly, the data show that the genes regulated by Wnt-3 involve multiple pathways downstream of GSK-3, not all of which are dependent on ␤-catenin.

MATERIALS AND METHODS
Cell Lines-The mouse mammary epithelial cell line C57MG was grown in Dulbecco's modified Eagle's medium (4.5 mg/ml D-glucose) supplemented with 10% fetal bovine serum, 100 units/ml penicillin, 100 g/ml streptomycin, 2 mM L-glutamine, and 10 g/ml insulin (Sigma). Wnt-3 was expressed in C57MG cells after infection with a murine leukemia virus-based retroviral vector (LNCX, Invitrogen). The vector expresses Wnt-3, which was cloned from a mouse brain library, under the control of the cytomegalovirus promoter and the bacterial neomycin phoshotransferase gene (neo). The same vector was used for expression of a ␤-catenin deletion mutant (⌬N89␤-catenin) that lacks the potentially destabilizing phosphorylation sites for GSK-3. Infected cell populations were selected for 3 weeks in 400 g/ml Geneticin (G418). At that time partial transformation to a more spindle-like cell shape was easily detectable in the Wnt-3 cells. About 400 G418 resistant clones designated Wnt-3, ⌬N89␤-catenin, or control (empty vector) were pooled and expanded for total RNA preparation and protein lysate preparation.
Differential Display-Total RNA was isolated from Wnt-3 or vector control cells according to manufacturer's conditions (Qiagen) and then further treated with DNase using the MessageClean (Genhunter) protocol.
Based on a method developed by Liang and Pardee (18), 20 primers provided by the Hieroglyph mRNA profile kit for differential display analysis (Genomyx Corp.) were used to amplify 12 different pools of reverse-transcribed PCR reactions. A total of 240 primer pairs were used to profile gene expression. Products from these PCR reactions were run in duplicate with two sets of independently prepared total RNA on 4.5% sequencing gels using a genomyx LR sequencer (Genomyx Corp.). The dried gels were exposed to Kodak XAR-2 film (Eastman Kodak Co.).
A total of 50 differentially expressed cDNA fragments were excised, reamplified according to the manufacturer's protocol (Genomyx Corp.), and sequenced. The length of the PCR products ranged from 0.3 to 1.0 kilobases. Only bands that were reproduced in duplicate lanes for two independent total RNA preparations were excised and reamplified.
cDNA Microarray-The cDNA microarray consisted of 26 cDNA fragments identified in the differential display analysis and control genes such as actin and tubulin.
Poly(A)ϩ RNA was purified with the Oligotex mRNA mini kit (Qiagen) from total RNA isolated from Wnt-3-expressing and control cells. 0.5 g of mRNA was oligo(dT)-and nonamer-primed at 70°C for 5 min, cooled on ice, and then reverse-transcribed with 600 units of Super-scriptII reverse transcriptase for 2 h at 42°C in 1ϫ first strand buffer, 10 mM dithiothreitol, 2 mM each dNTP, 1 mM dCTP (all reagents, Life Technologies, Inc.), and either 1 mM Cy3 or Cy5 (Amersham Pharmacia Biotech) for fluorescent labeling of the cDNA in a 20-l total volume. The mRNA was digested with 1 unit RNaseH (Life Technologies Inc.) and 0.5 unit RNase ONE (Promega) for 30 min at 37°C. Unincorporated nucleotides were removed using Qiaquick PCR clean up spin columns (Qiagen), and the volume was reduced to 9.5 l. The aminosilane-coated slides were prehybridized for 3-4 h at 42°C in 5ϫ SSC, 0.1% SDS, and 50% formamide solution after UV-cross-linking and baking of the slides for 1 h at 80°C.
The probe was made in 5ϫ SSC, 0.1% SDS, 50% formamide, 1 g COT DNA (Roche Molecular Biochemicals), 1.25 g of poly(A) oligo, and 4.25 l of the fluorescently labeled cDNA in a volume of 25 l and hybridized to the slides overnight at 42°C. The next day the slides were washed once in 1ϫ SSC, 0.2% SDS, twice in 2ϫ 0.1 SSC, 0.2% SDS, and dried after a wash in deionized water.
Imaging and Data Analysis-Fluorescence intensities of the immobilized probes were determined from images taken with an Arrayscanner GenerationII (Molecular Dynamics) confocal microscope with laser excitation sources and interference filters appropriate for Cy3 and Cy5 fluors. Separate scans were taken for each fluor at the resolution of 80 m 2 /pixel. The image intensities were quantified using ImageQuant software (Molecular Dynamics). A minimal threshold value was set, and spots with values below this threshold were considered negative. The average for the 16 repeats of immobilized probes was taken to calculate the average intensities. The ratio of the average intensities determined the differences of gene expression in the different probes. Control spots, e.g. tubulin and actin cDNA, gave ratios close to 1.
Western Blotting-Control cells were treated with increasing concentrations of LiCl (0.1-20 mM) for 7 h. Lysates were prepared by treating cells with lysis buffer, consisting of 120 mM Tris/HCl, pH 6.8, 5% SDS (w/v), 20% glycerin (v/v), and 1 mM dithiothreitol. The lysates were run through Qiashredder membranes (Qiagen) and stored at Ϫ20°C until used. For cytosolic fractions, cells were lysed in hypotonic buffer (10 mM Hepes, pH 7.2, 1.2 mM EGTA, 1.5 mM MgCl 2 , 10 mM KCl, and protease inhibitor mixture (Roche Molecular Biochemicals)) by passing through 27G1/2 gauge needles. After centrifugation at 6000 rpm for 10 min, supernatants were spun in a Ultracentrifuge (Optima TLX, Beckman) at 100,000 rpm for 35 min at 4°C. The supernatant, which contained the cytosolic fraction, was collected, and equal amounts were prepared for SDS-polyacrylamide gel electrophoresis.
SDS-polyacrylamide gel electrophoresis was performed under reducing conditions on 10 -12% polyacrylamide gels or using NuPAGE polyacrylamide gels with Mes-based running buffer. Resolved proteins were transferred from the gel onto nitrocellulose (Amersham Pharmacia Biotech).
After blocking, membranes were incubated with ␤-catenin mouse monoclonal antibody (Transduction Laboratories), Cox-2 goat polyclonal antibody (Santa Cruz Biotechnology), and 14-3-3 rabbit polyclonal antibody (Santa Cruz Biotechnology). The membranes were then incubated with a peroxidase-conjugated secondary antibody followed by development with an enhanced chemiluminescence (ECL plus) mixture (Amersham Pharmacia Biotech).
Northern Blot Analysis-Total RNA was isolated from transduced C57MG cells using the RNeasy kit (Qiagen). 10 -15 g of each RNA sample was fractionated by agarose/formaldehyde gel electrophoresis, transferred to nylon membranes, and cross-linked by exposure to UV light.
Cox-2 and periostin-specific probes corresponding to nucleotides 2760 -3102 and 651-1026, respectively, were amplified by PCR and labeled with biotinylated dNTPs according to the manufacturer's protocol (North2South Biotin Random Prime DNA Labeling kit, Pierce). The hybridization and detection was done by following the protocol for the North2South chemiluminescent hybridization and detection kit (Pierce), which uses a streptavidin horseradish peroxidase-coupled antibody for detection. Detection of GAPDH mRNA helped to determine differences in RNA loading for each lane.
GSK-3 Kinase Assay-To measure GSK-3 kinase activity in vitro, recombinant GSK-3␤ was incubated in 50 mM Tris, pH 7.5, 1 mM dithiothreitol, and 10 mM MgCl 2 with 0.4 g of prephosphorylated P-cAMP-response element-binding protein peptide (SGSGKRREILSR-RPSpYR) with varying concentrations of LiCl (1-20 mM). The kinase reaction was started by adding an ATP mixture including [␥-33 P]ATP. After a 20-min incubation at room temperature, the reaction mixture was pipetted onto phosphocellulose filters. The filters were washed four times in 75 mM H 3 PO 4 and air-dried, and radioactivity was measured in a scintillation counter. Each kinase reaction was done in triplicate.
Quantitative RT-PCR-Using a RT reaction kit (Perkin-Elmer), 1 g of total RNA from different antisense oligonucleotide transfection experiments was reverse transcribed. After 10% of the RT reaction was amplified in a quantitative PCR using the LightCycler System (Roche), the product was quantified using a standard curve that correlated each cycle number at which the amplification of the product was in the linear phase with a value. This value was then normalized to the value of the internal standard GAPDH or actin for each probe.
Prostaglandin Assay-4 ϫ 10 4 cells were grown for 48 h in 24-well plates. Conditioned medium was centrifuged for 5 min at 6000 rpm. 25-50 l of the supernatant was assayed for prostaglandin E 2 (PGE 2 ) in an enzyme immunoassay according to the manufacturer's protocol (Amersham Pharmacia Biotech).

RESULTS
To identify Wnt-3 target genes, a reverse transcriptase-PCRbased differential display analysis was used to compare mRNA levels from two C57MG cell lines that were transduced with either Wnt-3 or control vector. The mammary epithelial cell line C57MG displays a morphological response to expression of Wnt-3 and several other Wnt proteins (20) and has been used extensively for studies on the Wnt pathway. The cells acquire an elongated spindle-like cell shape (Fig. 1A) and show increased levels of cytosolic ␤-catenin (Fig. 1B) in response to Wnt-3 expression.
Preparations of total RNA from Wnt-3 transduced cells and from control cells were reverse-transcribed and amplified in PCR reactions with 240 primer pairs to profile gene expression. Out of the 26 sequences that were differentially expressed, 19 were up-regulated and 7 were down-regulated in Wnt-3 cells versus control cells. From those 26 genes, 14 sequences were known, 11 were ESTs and one sequence was unknown. Sequenced fragments were considered as known if they showed a sequence identity greater than 90% with any gene in the NCBI nucleotide data base.
The differential expression pattern was validated by microarray and Northern blot analyses. For the microarray analysis, the 26 cDNA fragments were cross-linked to a glass slide. Using labeled Cy3/Cy5 cDNA probes from Wnt-3 and control cells, respectively, Cox-2 gave a 3-fold up-regulated and periostin a 10-fold down-regulated difference in signal (Fig. 2). These data were confirmed by Northern blot analysis of mRNA from Wnt-3 and vector-transduced C57MG cells (Fig. 3, A and  B). Unexpectedly however, periostin mRNA levels did not change in response to ⌬N89␤-catenin overexpression (Fig. 3C).
Whether the observed levels of mRNA were reflected in protein levels was assessed for Cox-2 in lysates prepared from Wnt-3-expressing cells when compared with lysates from vector control cells (Fig. 4A). Levels of periostin protein could not be measured, because no periostin-specific antibodies were available. In another unexpected result, C57MG cells ectopically expressing ⌬N89␤-catenin did not have increased Cox-2 protein levels (Fig. 4A). The presence of ⌬N89␤-catenin in the cytosol was confirmed by Western blot analysis (Fig. 4D), and its transcriptional activity was verified as a 54-fold increase in the ␤-catenin-Lef-1-dependent reporter gene expression (Fig. 4C).
Conditioned medium from cells transduced with either Wnt-3, ⌬N89␤-catenin, or the empty vector, were tested for PGE 2 , which is one of the main Cox-2 products. Correlating with the Cox-2 protein expression levels (Fig. 4A), Wnt-3-expressing cells secreted 7-10-fold more PGE 2 compared with the control cells and ⌬N89␤-catenin-expressing cells (Fig. 4B). These data clearly show that Wnt-3 regulates Cox-2 and periostin expression levels in C57MG cells, most likely via a pathway that does not involve ␤-catenin.
Another component of the Wnt-3 signaling pathway is GSK-3. We tested whether Li ϩ , an inhibitor of GSK-3 (21), alters Cox-2 and periostin expression in response to GSK-3 inhibition. Treatment of C57MG cells with increasing concentrations of LiCl resulted in a 3-fold increase of Cox-2 protein

FIG. 2. Fluorescent image of a microarray slide in grayscale.
Each gene fragment was linked in quadruplet to the slide surface and hybridized to labeled reverse-transcribed mRNA from Wnt-3 and control cells. Differences in the intensities for Cox-2 (f) and periostin (E) can be seen when the upper panel (subset of probes from control cells) is compared with the lower panel (subset of probes from Wnt-3 cells) at the positions indicated. Tubulin (OE) and actin (q) are indicated as controls.

FIG. 3. Northern blot analysis confirms regulation of Cox-2 and periostin in Wnt-3 expressing cells.
Total RNA was prepared from cells, and 10 -15 g of each RNA sample was analyzed by Northern blotting as described under "Materials and Methods." The blot was cut, the upper part was incubated with Cox-2 (A) or murine periostin (B and C) probes, and the lower part with a murine GAPDH probe. C, periostin mRNA level is unchanged in ⌬N89␤-catenin expressing clones. after 7 h (Fig. 5A). The same LiCl concentration (20 mM) that mediated increased Cox-2 levels resulted in a 50% reduced kinase activity in an in vitro kinase assay (Fig. 5C). It is technically difficult to measure the Li ϩ -induced inhibition of endogenous GSK-3 directly, because after immunoprecipitation, Li ϩ is removed during the wash steps, and thus the inhibition of the GSK-3 enzyme activity is relieved. In the case of periostin, incubation of cells with LiCl for 7 h resulted in a dose-dependent reduction of periostin mRNA levels (Fig. 5B). These data suggest that the Wnt-3-mediated suppression of periostin mRNA is regulated at the level of GSK-3, but ⌬N89␤catenin alone fails to elicit this response.
To further confirm that Wnt-3 and the inhibition of GSK-3 but not ␤-catenin can regulate periostin and Cox-2, antisense oligonucleotides against ␤-catenin were transfected to reduce ␤-catenin expression levels. In cells with a 50 -80% reduction of ␤-catenin mRNA levels (measured by quantitative RT-PCR, Figs. 6B and 7), Cox-2 (Fig. 6A) and periostin levels (Fig. 7) can still be regulated by LiCl to the same extent as wild type cells or cells transfected with the reverse control oligonucleotide. However, cells with the same reduction of ␤-catenin mRNA had a 90% reduction in Lef-1 reporter gene activity when treated with Wnt-3a conditioned medium or Li ϩ for 7 h compared to cells transfected with the ␤-catenin reverse control oligonucleotides. The relative RNA expression levels of ␤-catenin are shown in Fig. 8B. Similarly the periostin and Cox-2 mRNA levels remain regulated by Wnt-3 and are independent of ␤-catenin antisense oligonucleotide transfection (Fig. 9).
To our knowledge, this is the first time that reduction in ␤-catenin levels using antisense oligonucleotides (with resultant inhibition of Lef-1 transcriptional activity in response to Wnt and Li ϩ treatment) has allowed the identification of a pool of Wnt target genes whose expression may be independent of ␤-catenin.

DISCUSSION
The present study presents evidence that Cox-2 and periostin are two genes whose expression is regulated by Wnt-3 in a pathway that is independent of ␤-catenin/TCF-mediated transcription. We have shown that Cox-2 and periostin are regulated by the inhibition of GSK-3 by Li ϩ but not by ␤-catenin. This suggests that a pool of Wnt target genes exists that is independent of ␤-catenin signaling.
Studies performed to date have shown that ␤-catenin stabilization seems to be required for Wnt-induced transformation. These data are consistent with studies showing that mutations in ␤-catenin, present in human tumors and cancer cell lines, result in a stabilized cytoplasmic pool (22,23). Cytosolic ␤-catenin then complexes with the Lef-1/TCF transcription factor(s) and activates gene transcription. Interestingly, there is no FIG. 4. Cox-2 protein levels and prostaglandin E 2 secretion in ⌬N89␤catenin or Wnt-3-expressing clones and vector control cells. A, lysates were prepared and analyzed for Cox-2 protein by Western blotting as described under "Materials and Methods" using a Cox-2 goat polyclonal antibody. As a loading control, the lower part of the blot was probed for 14-3-3 using a 14-3-3 rabbit polyclonal antibody. The position of the 31-and 52-kDa protein markers is shown. B, secretion of PGE 2 into medium of Wnt-3-expressing cells is increased. 25 l of conditioned medium of ⌬N89␤-catenin or Wnt-3-expressing clones and vector control cells was tested for PGE 2 1-20 mM). A, lysates were prepared and analyzed for Cox-2 protein by Western blotting as described under "Materials and Methods" using Cox-2 antibody. The lower part of the blot was probed with 14-3-3 antibody as a loading control. B, total RNA was prepared from cells that were incubated at two concentrations of LiCl for 7 h and analyzed by Northern blot analysis as described under "Materials and Methods." The blot was cut, the upper part was probed with murine periostin, and the lower part was incubated with a murine GAPDH probe. C, inhibition of GSK-3 kinase activity by LiCl. Recombinant GSK-3 was treated with increasing concentrations of LiCl (1-20 mM) in the presence of [␥-33 P]ATP and the P-cAMP-response element-binding protein peptide, which was then counted for radioactivity. change in the morphology of C57MG cells when they are transduced with ⌬N89␤-catenin or incubated with LiCl in contrast to transduction by Wnt-3, which promotes a spindle-shaped morphology ( Fig. 1A and data not shown). These differences in morphology suggest that even though, as previously reported, there is overlap between genes regulated by Wnt and by ␤-catenin as well as by Li ϩ inhibition of GSK-3, there must also be significant differences.
In the present study, a RT-PCR based differential display approach was used to examine the gene profiles regulated by Wnt-3. Cox-2 expression increased and periostin expression decreased in response to activation of the Wnt pathway(s). These genes were also regulated by inhibition of GSK-3 but overexpression of ⌬N89␤-catenin failed to mediate the same changes. In addition to overexpression of ⌬N89␤-catenin, the reduction of ␤-catenin expression levels by antisense oligonucleotide transfection was a technology used to address the divergence of the Wnt pathway downstream of GSK-3. The regulation of Cox-2 and periostin by Wnt-3 and LiCl was not altered by the reduction of ␤-catenin expression levels (Figs. 6, 7, and 9). These data suggest a pool of genes exists downstream of GSK-3 that is regulated by Wnt-3 in a ␤-catenin-independent pathway. LiCl inhibits GSK-3 activity (Fig. 5C) and can mimic Wnt activation in vivo in Xenopus laevis oocytes (24) and in the Lef-1 reporter gene assay, whereas the induction of the Lef-1 reporter gene is almost completely abolished by the use of specific ␤-catenin antisense oligonucleotides (Fig. 8A). Both Wnt-3 and inhibition of GSK-3 by LiCl suppress the mRNA levels of periostin. However, neither cells overexpressing ⌬N89␤-catenin nor the reduction of ␤-catenin expression levels by antisense oligonucleotide transfection resulted in a change of periostin mRNA levels (Figs. 3C and 7).
Our findings suggest that the transcriptionally active ␤-catenin-Lef-1 complex is not sufficient to repress periostin levels or to induce Cox-2 and that other or additional factors are employed to mediate transcriptional responses to GSK-3 inhibition and Wnt-3 signaling. This is supported by data obtained from the antisense oligonucleotide transfection experiments. The fact that GSK-3 inhibition is a common regulatory mechanism in a number of ligand-stimulated pathways, for example by insulin and epidermal growth factor (25,26), which do not increase cytosolic levels of ␤-catenin, indicates that there may also exist additional Wnt pathway(s) that are ␤-catenin-independent. A possible way by which the transcriptional effects of these growth factors could be mediated is via the transcription factor c-jun that can be inhibited by GSK-3 phosphorylation in vitro (27). X. laevis oocytes treated with LiCl show increased activation of an AP-1 reporter plasmid, consistent with the inhibitory effect of GSK-3 on c-jun activity (24).
The increase in Cox-2 protein correlates with an increase in PGE 2 synthesis and was seen after Wnt-1 expression, as observed recently by Howe et al. (17) and in this study for Wnt-3expressing C57MG cells (Fig. 4B). Considerable evidence exists for a role of Cox-2 in human intestinal carcinogenesis. Cox-2 mRNA levels and consequently prostaglandin synthesis are markedly increased in over 80% of human colorectal cancers and human breast cancers (28,29). The precise role for Cox-2 in intestinal tumorigenesis is not fully understood, but the variety of effects mediated by prostaglandins include the negative regulation of apoptosis in cells that are otherwise supposed to undergo cell death and induced release of angiogenic factors in endothelial cells (30,31).
An indirect link has been made between Cox-2 and the Wnt pathway in a study where Cox-2 inhibitors displayed antineoplastic activities in a mouse model for familial adenomatous polyposis. In this model, loss of APC causes an increase of polyps in the colon of mice, and a Cox-2 inhibitor was able to attenuate this effect (32). Interestingly, despite the elevation of Cox-2 expression by LiCl, no link has been established in patients treated with Li ϩ for mental disorders and the development of cancer (33).
That periostin is a gene regulated by Wnt-3 signaling is described here for the first time. However, its function as a downstream target of the Wnt pathway still needs to be clarified. Initially, periostin was identified as a secreted factor in a screen of a mouse osteoblastic library. Its potential function in bone adhesion is based on its sequence similarity to the adhesion protein fasciclin 1 (34,35) and because it is expressed in specialized connective tissues that form and support mineralized tissue (36). Interestingly, periostin was also identified in a subtractive hybridization experiment comparing embryonic rhabdomyosarcoma cells and primary lung fibroblasts (37). Periostin was down-regulated in the tumor cells compared with normal lung fibroblasts. The loss of periostin expression correlated with the transformed phenotype. This is directionally consistent with our findings, if one assumes that Wnt can promote transformation.
The data in the present study suggest that even though there is overlap between genes regulated by Wnt-3 and ␤-catenin, there are also significant differences. Distinguishing between pools of Wnt-regulated genes that can be induced via different pathways may be important with respect to oncogenic ␤-catenin mutations that would specifically induce only a subset of Wnt-regulated genes as well as an independent pool of genes not regulated by Wnt. Because mutations in both ␤-catenin and APC are found in a high percentage of tumors, this knowledge could have important consequences for the identification of therapeutic targets.