Activation of AXIN2 Expression by -Catenin-T Cell Factor

The Wnt pathway regulates cell fate, proliferation, and apoptosis, and defects in the pathway play a key role in many cancers. Although Wnts act to stabilize β-catenin levels in the cytosol and nucleus, a multiprotein complex containing adenomatous polyposis coli, glycogen synthase kinase 3β, and Axin1 or its homolog Axin2/Axil/conductin promotes β-catenin phosphorylation and subsequent proteasomal degradation. We found that the rat Axil gene was strongly induced upon neoplastic transformation of RK3E cells by mutant β-catenin or γ-catenin or after ligand-induced activation of a β-catenin-estrogen receptor fusion protein. Expression of Wnt1 in murine breast epithelial cells activated the conductin gene, and human cancers with defective β-catenin regulation had elevated AXIN2 gene and protein expression. Expression ofAXIN2/Axil was strongly repressed in cancer cells by restoration of wild type adenomatous polyposis coli function or expression of a dominant negative form of T cell factor (TCF)-4. TCF binding sites in the AXIN2 promoter played a key role in the ability of β-catenin to activate AXIN2 transcription. In contrast to AXIN2/Axil, expression of human or rat Axin1 homologs was nominally affected by β-catenin-TCF. Because Axin2 can inhibit β-catenin abundance and function, the data implicate AXIN2 in a negative feedback pathway regulating Wnt signaling. Additionally, although Axin1 and Axin2 have been thought to have comparable functions, the observation that Wnt pathway activation elevates AXIN2 but not AXIN1 expression suggests that there may be potentially significant functional differences between the two proteins.

The Wnt pathway regulates cell fate, proliferation, and apoptosis, and defects in the pathway play a key role in many cancers. Although Wnts act to stabilize ␤-catenin levels in the cytosol and nucleus, a multiprotein complex containing adenomatous polyposis coli, glycogen synthase kinase 3␤, and Axin1 or its homolog Axin2/Axil/conductin promotes ␤-catenin phosphorylation and subsequent proteasomal degradation. We found that the rat Axil gene was strongly induced upon neoplastic transformation of RK3E cells by mutant ␤-catenin or ␥-catenin or after ligand-induced activation of a ␤-catenin-estrogen receptor fusion protein. Expression of Wnt1 in murine breast epithelial cells activated the conductin gene, and human cancers with defective ␤-catenin regulation had elevated AXIN2 gene and protein expression. Expression of AXIN2/Axil was strongly repressed in cancer cells by restoration of wild type adenomatous polyposis coli function or expression of a dominant negative form of T cell factor (TCF)-4. TCF binding sites in the AXIN2 promoter played a key role in the ability of ␤-catenin to activate AXIN2 transcription. In contrast to AXIN2/Axil, expression of human or rat Axin1 homologs was nominally affected by ␤-catenin-TCF. Because Axin2 can inhibit ␤-catenin abundance and function, the data implicate AXIN2 in a negative feedback pathway regulating Wnt signaling. Additionally, although Axin1 and Axin2 have been thought to have comparable functions, the observation that Wnt pathway activation elevates AXIN2 but not AXIN1 expression suggests that there may be potentially significant functional differences between the two proteins.
Defects that interfere with ␤-catenin regulation have been reported in various human cancers. In a subset of many different cancer types, mutations at or near the serine and threonine residues in the ␤-catenin N-terminal domain alter its ability to be phosphorylated by GSK3␤. In other cancers, particularly colorectal cancers, inactivation of the APC tumor suppressor gene appears to be the predominant mechanism leading to ␤-catenin deregulation (4,5,24). In yet other cancers, mutations in the genes encoding one of the two Axin proteins have been reported, including the AXIN1 gene in hepatocellular carcinomas and medulloblastomas (25,26) and the AXIN2 gene in a small fraction of colorectal cancers lacking APC or ␤-catenin mutations (27). A prime consequence of the mutational defects in ␤-catenin regulation is constitutive activation of downstream ␤-catenin-TCF-regulated target genes, particularly genes with major effects on cell growth regulation and tumorigenesis, such as c-myc, CCND1, and MMP-7 (4,5).
In an effort to understand better the effects of Wnt-␤-catenin-TCF pathway activation in cancer cells, we undertook studies to identify novel ␤-catenin-TCF-regulated target genes. We used oligonucleotide microarrays to identify transcripts with elevated expression after neoplastic transformation of the rat E1A-immortalized RK3E cell line by mutant ␤-catenin or ␥-catenin or after ligand-induced activation of a ␤-cateninestrogen receptor (ER) fusion protein. We found that expression of the rat Axil gene was strongly induced in the RK3E cell line in all three of these settings. Further studies established that the mouse and human homologs of Axil, known as conductin and AXIN2, respectively, were consistently induced by Wnt pathway activation. TCF proteins played a key role in AXIN2 induction. Unlike AXIN2, AXIN1 was not found to be a ␤-catenin-TCF-regulated gene. Prior studies have shown that the Axin1 and Axin2 proteins have roughly 45% amino acid identity and essentially identical functions in regulating ␤-catenin levels (7)(8)(9)(10)28). In addition to showing that AXIN2 functions in a feedback repressor pathway regulating Wnt signaling, our findings on the differential effects of Wnt pathway activation on AXIN2 versus AXIN1 expression suggest that potentially significant functional differences may exist between their protein products.

EXPERIMENTAL PROCEDURES
Plasmids-Expression vectors for wild type and mutant (codon 33 substitution of tyrosine for serine, S33Y) forms of ␤-catenin and dominant negative Tcf-4 (Tcf-4⌬N31) have been described previously (29). The pBabe-S33Y-ER-puro expression vector encoding a chimeric ␤-catenin-ER protein, in which full-length S33Y ␤-catenin sequences are fused in-frame to a mutated ER ligand binding domain, was generated by cloning the S33Y ␤-catenin cDNA into the BamHI and EcoRI sites of the retroviral plasmid pBabe-puro (30). The reporter constructs pTOP-FLASH, which contains three copies of an optimal TCF binding motif (CCTTTGATC), and pFOPFLASH, which contains three copies of a mutant motif (CCTTTGGCC), have been described previously (31). Plasmid pCH110 (Amersham Biosciences) contains a functional lacZ gene cloned downstream from a cytomegalovirus early region promoterenhancer element. The Axin2pcDNA3.1mycHis(Ϫ)B expression vector was a kind gift from Wanguo Liu (Mayo Clinic, Rochester, MN) (27). DNA fragments containing human AXIN2 promoter sequences cloned upstream from a luciferase reporter gene were obtained by PCR amplification of genomic DNA, using primers generated from AXIN2 sequences in GenBank (accession no. AC00485). AXIN2 genomic DNA fragments were subcloned upstream from the luciferase reporter gene in the pGL3Basic reporter vector (Promega, Madison, WI), using the KpnI and NheI sites. The reporter gene vector AX2(1078WT)/Luc contains AXIN2 sequences from Ϫ1078 to ϩ5 relative to the presumed transcription start site, and the vector AX2(181WT)/Luc contains AXIN2 sequences from Ϫ181 to ϩ5. The forward primer for generating the AX(1078WT)/Luc vector was 5Ј-CCCGTTCAGCCCCTACCCTTCT-TAG-3Ј, and the forward primer for the AX(181WT)/Luc vector was 5Ј-CAGCGCCTGATACTTAGATGAGC-3Ј; the reverse primer for generating both vectors was 5Ј-CAAGTCAGCAGGGGCTCATCTG-3Ј. Mutations in a presumptive TCF DNA binding site at bp Ϫ108 to Ϫ102 were obtained in vitro via a standard PCR-based mutagenesis strategy, generating the reporter gene vectors AX2(1078Mut)/Luc and AX2/ (181Mut)/Luc. All plasmid sequences were confirmed by automated sequencing of double-stranded DNA templates.
Cell Culture-All cell lines were obtained from American Type Culture Collection (Rockville, MD), with the exception of the following: the amphotropic Phoenix packaging cell line, which was obtained from G. Nolan (Stanford University School of Medicine); the RAC311, RAC311/ Wnt-1, RAC311/Wnt-1 9, C57/Vect, and C57/Wnt-1 lines (32), all of which were obtained from L. Howe (Weill Medical College of Cornell University); Gli-transformed RK3E cells (33), which were obtained from J. M. Ruppert (University of Alabama at Birmingham); and the HT29/ ␤-Gal and HT29/APC lines (34), which were obtained from B. Vogelstein (Johns Hopkins University School of Medicine). All cells were grown in 5% CO 2 with medium containing 10% fetal bovine serum and penicillin/ streptomycin, unless otherwise stated. HEK293, Phoenix, parental RK3E cells, RK3E cells neoplastically transformed by K-ras, Gli, ␤-catenin, and ␥-catenin (29,35), RAC311 lines, C57 lines, and all human colon cancer lines, except for HT29, LS174T, RKO, and SW48 cells, were grown in Dulbecco's modified Eagle's medium (Invitrogen). LS174T cells were grown in minimum Eagle's medium ␣ (Invitrogen), and SW48 cells were grown in L15 medium (Invitrogen) in the absence of CO 2 . RKO, HT29, HT29/␤-Gal, and HT29/APC cells were cultured in McCoy's medium (Invitrogen). Hygromycin B (Sigma) was included at a concentration of 0.6 mg/ml for the HT29/␤-Gal and HT29/APC cells. Insulin (10 g/ml; Sigma) was added to the media for the C57 lines. A clonal RK3E cell line expressing the ␤-catenin S33Y-ER fusion protein was obtained after retroviral transduction of RK3E cells with supernatants from amphotrophic Phoenix cells transfected with pBabe-S33Y-ER-puro. Drug selection on the pBabe-S33Y-ER-puro-transduced RK3E cells was carried out in puromycin (Sigma) at a concentration of 1.0 g/ml. A single resistant colony was isolated by ring cloning and expanded into a stable cell line, termed RK3E/S33Y-ER. The RK3E/ S33Y-ER line was maintained subsequently in 0.5 g/ml puromycin. To activate the S33Y-ER fusion protein, the RK3E/S33Y-ER cells were treated with medium supplemented with 0.5 M 4-hydroxytamoxifen (4-OH-T) (Sigma), made from a stock concentration of 100 M 4-OH-T in 100% ethanol. To inhibit new protein synthesis in RK3E/S33Y-ER cells, the medium was supplemented with cycloheximide (Sigma) at a concentration of 1 g/ml. To assess the effects of dominant negative TCF-4 on AXIN1 and AXIN2 gene expression, a retroviral TCF-4⌬N31 expression construct (29) was used to transduce two RK3E lines that had been transformed neoplastically by mutant ␤-catenin (RK3E/⌬N47-B and RK3E/⌬N132-A) (29) as well as the SW480 and DLD1 colon cancer lines. Empty vector (pPGS-Neo) control transductions of the two RK3E and two colon lines were carried out in parallel. The TCF4⌬N31-and empty vector-transduced cells were subsequently selected for 7-10 days in 1.0 -1.5 mg/ml G418 (Sigma). To assess the effects of wild type APC gene function on AXIN1 and AXIN2 gene expression, HT29/␤-Gal and HT29/APC cells were treated with 150 M ZnCl 2 for induction of the control lacZ and wild type APC genes, respectively.
DNA Array Expression Analysis-Trizol (Invitrogen) extraction and purification with the RNeasy Cleanup Kit (Qiagen, Chatsworth, CA) was used to prepare total RNA from five samples: parental RK3E cells; RK3E/S33Y-ER cells either mock (ethanol)-treated or 4-OH-T-treated for 24 h; a pool of equal masses of RNA from seven clonal RK3E lines transformed neoplastically by mutant ␤-catenin (29); and a pool of equal masses of RNA from five clonal RK3E lines transformed neoplastically by ␥-catenin (35). Gene expression analyses on the five samples were carried out with commercial high density oligonucleotide arrays (Affymetrix, Santa Clara, CA), using protocols and methods developed by the supplier. Arrays were scanned using the GeneArray scanner (Affymetrix), and image analysis was performed with GeneChip 4.0 software (Affymetrix), which stores the results for each feature in .CEL files. Each RG_U34A chip consists of 534 ϫ 534 probes (24 ϫ 24 m each) that are 25-base long single-stranded DNA sequences. There are typically 16 pairs of features (probe pairs) for each of the transcripts (probe sets) and a total of 8,799 probe sets. Half of the features are complementary to a specific sequence (perfect match ϭ PM features), the other half have an identical match except a central base has been altered (mismatch ϭ MM features). We have developed software to read .CEL files and perform some processing of the data, available at dot.ped.med.umich.edu:2000/ourimage/pub/shared/Affymethods.html. The chip for the parental RK3E sample was selected as a standard. Probe pairs for which PM-MM Ͻ Ϫ100 on the standard were excluded from the analysis. One-sided signed rank tests of the PM-MM values for each probe set on each chip were obtained to help judge whether transcripts were detectable. The average intensity for each probe set was computed as the mean of the PM-MM differences, after trimming away the 25% highest and lowest differences. A set of 3,692 reference probe sets was selected for use in normalization, these being the probe sets that gave p Ͻ 0.05 for all five chips for the test of detectability. A normalization factor for each chip was obtained using the reference probe sets by computing the antilogarithm of the mean log ratios of the average intensities for the selected chip divided by the standard. The average intensities were divided by this factor to obtain the normalized intensities for the probe sets. When computing fold change indices, we replaced intensities less than 10 by 10 before forming ratios to avoid negative or spuriously large fold change numbers.
Northern Blot Analysis-Total RNA was extracted from cells with Trizol, and Northern blot analysis was performed. Approximately 15-20 g of total RNA was separated on a 1.2% formaldehyde-agarose gel and transferred to Zeta-Probe GT membranes (Bio-Rad) by capillary action. cDNA probes to detect rat Axil, mouse conductin, rat Axin1, and human AXIN1 expression were generated by RT-PCR, using primers derived from sequences in GenBank. The probe to detect AXIN2 was generated by PCR using the Axin2pcDNA3myc3.1 plasmid (provided by W. Liu; Mayo Clinic). The sequences of all PCR products probes were confirmed by automated sequencing. All probes were random labeled with [␣ 32 P]dCTP using Rediprime (Invitrogen) and hybridized to the membrane with RapidHyb Buffer (Invitrogen) according to the manufacturer's protocol. All Northern blots were stripped and hybridized to a rat glyceraldehyde-3-phosphate dehydrogenase (GAPDH) cDNA probe to control for RNA loading and transfer efficiency.
Western Blot Analysis-Whole cell lysates were prepared in radioimmunoprecipitation assay buffer (Tris-buffered saline (TBS), 0.5% deoxycholic acid, 0.1% SDS, and 1% Nonidet P-40 with complete protease inhibitors (Roche Molecular Biochemicals)). Protein concentration was determined by the bicinchoninic acid assay (Pierce Biochemicals), and 50 g of total protein from each sample was separated on 10% SDSpolyacrylamide gels. Proteins were transferred to Immobilon P membranes (Millipore, Bedford, MA) by semidry electroblotting. Immuno-blot analyses were carried out with the anti-conductin (S-19) or anticonductin (M-20) goat polyclonal antibodies (Santa Cruz Biotechnology, Santa Cruz, CA) at a 1:500 dilution in 1ϫ TBS with 5% dry milk and 0.5% Tween followed by incubation with a horseradish peroxidaseconjugated donkey anti-goat antibody (Pierce Biochemicals) at a 1:10,000 dilution. To verify equal loading of the samples, membranes were incubated with a rabbit polyclonal antibody against ␤-actin (Sigma) followed by a horseradish peroxidase-conjugated donkey anti-rabbit IgG antibody (Pierce Biochemicals). Antibody complexes were detected with the ECL Western blot kit (Amersham Biosciences) and exposure to X-Omat-AR film (Eastman Kodak).
Real Time RT-PCR Analysis of AXIN2 Expression-Total RNA was isolated with Trizol from 42 snap frozen, primary ovarian endometrioid type adenocarcinomas (OEAs) that had been analyzed in detail previously for ␤-catenin nuclear localization and mutational defects in the ␤-catenin, APC, AXIN1, and AXIN2 genes (36). The RNA was used for real time RT-PCR studies of AXIN2 and HPRT gene expression. In brief, first strand cDNA was synthesized from DNase I-treated mRNA samples using random hexamer primers (Amersham Biosciences) and Superscript II (Invitrogen). For PCR with a Prism 7700 Sequence Detector (Applied Biosystems, Foster City, CA), 5 ng of cDNA from each tumor sample was used in each reaction. For AXIN2, PCR was performed in 96-well plates in a 25-l reaction volume containing 1ϫ TaqMan  The AXIN2 probe had a carboxyfluorescein label at its 5Ј-end, and the HPRT probe had a VIC TM label at its 5Ј-end. Both probes had carboxytetramethyl rhodamine labels at their 3Ј-ends. The AXIN2 and HPRT PCRs were performed in duplicate for each tumor sample, and AXIN2 and HPRT reactions were performed in adjacent wells. The following PCR conditions were used: 2 min at 50°C and 10 min at 95°C followed by 40 cycles of 15 s at 95°C and 1 min at 60°C. Using the software accompanying the Prism 7700 detector, the HPRT signals were used for normalization. Student's t test was used to determine the significance of differences in AXIN2 expression between the 12 OEAs with strong nuclear staining for ␤-catenin and mutations in the ␤-catenin, APC, AXIN1, or AXIN2 genes and the 30 OEAs lacking strong nuclear ␤-catenin staining and pathway mutations.
Immunohistochemical Analysis-Immunohistochemcial analysis of AXIN2 expression in OEAs was performed as described previously (36). In brief, 5-m sections of formalin-fixed, paraffin-embedded tissues were mounted on Probe-On slides (Fisher Scientific), deparaffinized in xylene, and then rehydrated into distilled water through graded alcohols. Antigen retrieval was enhanced by microwaving the slides in citrate buffer (pH 6.0; Biogenex, San Ramon, CA) for 15 min. Endogenous peroxidase activity was quenched with 6% hydrogen peroxide in methanol, and the slides were blocked with 1.5% normal horse serum for 1 h. Sections were then incubated with the anti-conductin (M-20) goat polyclonal antibody (Santa Cruz Biotechnology) at a 1:500 dilution overnight at 4°C followed by a biotinylated horse anti-goat secondary antibody at a 1:200 dilution for 30 min at room temperature. Antigenantibody complexes were detected with the avidin-biotin peroxidase method using 3,3Ј-diaminobenzidine as a chromogenic substrate (Vectastain ABC kit; Vector Laboratories, Burlingame, CA). Immunostained sections were lightly counterstained with hematoxylin and then examined by light microscopy.
Luciferase Reporter Gene Assays-For all luciferase reporter assays, cells were plated in 35-mm six-well plates 12-24 h before transfection. Transfections were performed with FuGENE 6 (Roche Molecular Biochemicals) for 24 -36 h according to the manufacturer's protocol. Lysates were collected in 1ϫ Reporter Lysis Buffer (Promega). TCF transcriptional activity was measured as the ratio of luciferase activity from the pTOPFLASH vector to the pFOPFLASH vector. All luciferase activities were normalized for transfection efficiency by cotransfection with pCH110 and measurement of ␤-galactosidase activity. To assess the effects of AXIN2 on wild type and mutant ␤-catenin-induced TCF activity, 293 cells were cotransfected with 0.25 g of pTOP-or pFOP-FLASH, 0.5 g of a pcDNA3 vector encoding wild type or S33Y mutant ␤-catenin (29), 1 g of Axin2pcDNA3.1mycHis(Ϫ)B, and 0.25 g of pCH110. To confirm stable expression of TCF-4⌬N31in ␤-catenin-transformed RK3E cells as well as SW480 and DLD1 cells, cells were cotransfected with 1 g of pTOPFLASH or pFOPFLASH and 1 g of PCH110. For reporter gene assays with AXIN2 promoter constructs, DLD1 cells were cotransfected with 1 g of AX2(1078WT)/Luc or AX2(1078Mut)/Luc and 1 g of pCH110, whereas SW480/Neo or SW480/Tcf-4⌬N31 cells were cotransfected with 1 g of AX2(181WT)/ Luc or AX2(181Mut)/Luc and 1 g of pCH110. The total mass of transfected DNA in each well was kept constant by adding empty vector plasmid DNA, when necessary. All experiments were done in triplicate, and mean Ϯ S.D. values were determined.

Induction of Axil Expression by ␤or ␥-Catenin
Deregulation or Ligand-induced Activation of a ␤-Catenin-ER Fusion Protein-Our prior studies have shown that N-terminal mutant forms of ␤-catenin akin to those found in cancers, but not wild type ␤-catenin, will promote neoplastic transformation of RK3E cells, a rat E1A-immortalized epithelial cell line (29). Unlike ␤-catenin, its close functional relative ␥-catenin (also known as plakoglobin), will promote neoplastic transformation of RK3E cells when overexpressed, without a need for N-terminal mutations in the presumptive GSK3␤ phosphorylation consensus sites to activate ␥-catenin's transforming potential (35). In an effort to define novel downstream target genes in the Wnt pathway, we used commercial oligonucleotide microarrays to identify genes with elevated expression in ␤and ␥-catenintransformed RK3E cells compared with parental RK3E cells. Because some observed changes in gene expression in individual ␤and/or ␥-catenin-transformed RK3E lines might simply reflect the clonal origin of the transformed line under study, for the array analysis, we pooled equal masses of RNA from seven independent ␤-catenin-transformed lines and five independent ␥-catenin-transformed lines. In addition, because some changes in gene expression after transformation of RK3E cells by ␤or ␥-catenin might be the result of alterations in signaling pathways unrelated to catenin-TCF deregulation, we also assessed the consequences of transient activation of ␤-catenin. Transient activation of ␤-catenin in RK3E cells was achieved by treatment of an RK3E cell line expressing a chimeric ␤-catenin-ER fusion protein (RK3E/S33Y-ER) with the ligand 4-OH-T for 24 h. For our studies, we used the Affymetrix U34A rat GeneChip array, which contains roughly 8,000 known genes and expressed sequence tags. By comparing gene expression in parental RK3E cells and mock-treated RK3E/S33Y-ER cells (i.e. the two control cell populations) with gene expression in ␤and ␥-catenin-transformed RK3E and 4-OH-T-treated RK3E/S33Y-ER cells, we identified only 14 genes predicted to have greater than 2-fold increases in expression over control levels after catenin-TCF activation.
Northern blotting was used to assess the 14 candidate genes identified by the array analysis, and the most promising data were obtained for the rat Axil gene. Compared with parental RK3E cell lines or RK3E lines transformed by mutated K-ras or Gli1, marked increases in Axil expression were seen in 8 of 10 independent ␤-catenin-transformed RK3E lines (Fig. 1A) and all eight ␥-catenin-transformed RK3E lines studied (Fig. 1C). In contrast to the strong induction of Axil, expression of the rat Axin1 gene was not altered in RK3E lines transformed by ␤-catenin (Fig. 1B). Confirmation that induction of Axil gene expression resulted in changes in expression of the Axil protein was documented in selected ␤-catenin-transformed RK3E lines (Fig. 1D). Rapid induction of Axil expression was also seen in the RK3E/S33Y-ER cell line after treatment with the 4-OH-T ligand ( Fig. 2A). The observation that the protein synthesis inhibitor cycloheximide did not block 4-OH-T-mediated induction of Axil expression in RK3E/S33Y-ER cells ( Fig. 2A) indicates Axil is very likely to be a gene activated directly by ␤-catenin accumulation in the nucleus. Consistent with the view that Axil is induced via ␤-catenin interaction with TCF transcription factors, expression of a dominant negative form of AXIN2 Is a ␤-Catenin-TCF-regulated Target Gene TCF-4 in RK3E cell lines stably transformed by mutant ␤-catenin inhibited TCF transcriptional activity (Fig. 2B) inhibited Axil expression (Fig. 2C). Northern blot analysis of Axil (A) and rAxin1 (B) was carried out on total RNA isolated from parental rat E1A-immortalized RK3E epithelial cells (RK3E), RK3E cells transformed by codon 12 mutant c-K-ras (RK3E/Kras) and Gli1 (RK3E/Gli), as well as 10 independent clonal RK3E lines transformed by mutant ␤-catenin proteins (29). In C, Northern blot analysis of Axil was carried out on parental RK3E cells and eight independent clonal ␥-catenin-transformed RK3E cell lines (35). Northern blots in A-C were stripped and rehybridized to a rat GAPDH probe to control for loading. In D, Western blot analyses of Axil protein expression were carried out on parental RK3E cells and two independent clonal ␤-catenintransformed RK3E lines, using the S-19 (left) and M-20 (right) goat polyclonal antibodies raised against N-terminal mouse conductin sequences. A single band of roughly 97 kDa was detected with the S-19 antibody (left), and the larger of the two bands detected with the M-20 antibody (right) in the ␤-catenin-transformed RK3E lines (right) migrated at ϳ97 kDa. Blots were stripped and incubated with an anti-actin antibody to confirm equal loading of protein samples. Luciferase activities were measured in triplicate, and the ratio of luciferase activities in TOPFLASH-transfected versus FOPFLASH-transfected cells was determined and reported as the relative TCF activity. The control ␤-galactosidase-expressing vector pCH110 was used to correct for differences in transfection efficiency. In C, Axil expression was analyzed by Northern blot analysis of total RNA from the cell lines, and a rat GAPDH probe was used to control for RNA loading and transfer. systems and settings, we analyzed expression of conductin, the mouse homolog of Axil, in breast epithelial cells expressing high levels of the Wnt-1 protein. In both RAC311 and C57 cells, Wnt-1 expression was associated with high levels of conductin expression (Fig. 3).

Axil Homologs in Mouse and
As noted above, inactivating mutations in the APC gene are common in human colon cancers, and a subset of the 20 -25% of colon cancers that lack APC mutations have gain-of-function mutations in the ␤-catenin N terminus (4,5). Both the inactivating mutations in APC and the activating mutations in ␤-catenin led to ␤-catenin deregulation and constitutive activation of ␤-catenin-TCF transcripton. Consistent with the notion that the human AXIN2 gene might also be a target of the Wnt pathway, we found variable but readily detectable expression of AXIN2 in all 12 colon cancer cell lines studied (Fig. 4A and data not shown). To explore further the relationship between ␤-catenin deregulation and AXIN2 expression in colon cancers, we took advantage of a colon cancer cell line with regulated expression of the wild type APC gene. The HT29 colon cancer line has truncating mutations in both of its APC alleles. Morin et al. (34) generated a variant HT29 line (HT29/ APC) in which, after zinc treatment, expression of an exogenous wild type APC protein is induced rapidly to roughly the same level as that of the endogenous truncated APC proteins. Using HT29/APC cells and a matched control line (i.e. HT29/ ␤-Gal), we found that AXIN2 expression was strongly downregulated after APC induction. Zinc treatment of the control HT29/␤-Gal cell line had no detectable effect on AXIN2 expression. In contrast to the AXIN2 results, restoration of APC function in HT29 cells had only modest effects on AXIN1 expression. Furthermore, expression of a dominant negative form of TCF-4 in DLD1 and SW480 colon cancer cells strongly inhibited TCF transcriptional activity and AXIN2 expression but had at best minimal effects on AXIN1 expression (Fig. 5).
Nearly all candidate ␤-catenin-TCF-regulated genes described in the literature have been proposed based on data from in vitro and/or animal model studies. Thus far, few studies have evaluated expression of presumptive ␤-catenin-TCF target genes in primary human tumors that have been characterized thoroughly for mutational defects in ␤-catenin regulation. We chose to assess AXIN2 expression in primary OEAs because although OEAs share similar histological features, only about 30 -40% of the lesions have mutational defects affecting ␤-catenin regulation (36 -38). This contrasts with the picture in pri-mary colorectal carcinomas, which almost uniformly carry mutational defects in ␤-catenin regulation (4,5). Hence, comparison of gene expression in OEAs with intact ␤-catenin regulation versus OEAs with defective ␤-catenin regulation should permit a more definitive assessment to be made about the relationship between ␤-catenin regulatory defects and expression of candidate ␤-catenin-TCF target genes. Using real time RT-PCR assays to assess AXIN2 expression in a panel of 42 OEAs characterized previously for ␤-catenin nuclear localization and mutations in the ␤-catenin, APC, AXIN1, and AXIN2 genes (36), we found that AXIN2 expression was on average roughly 20-fold higher in the OEAs with ␤-catenin regulatory defects than in OEAs with apparently intact ␤-catenin regulation (Fig. 6). To confirm that induction of AXIN2 gene expression resulted in demonstrable changes in AXIN2 protein expression in primary tumors with ␤-catenin defects, we performed immunohistochemistry studies on a subset of the OEAs analyzed in the real time RT-PCR studies. AXIN2 expression was found to be increased in the majority of OEAs with ␤-catenin regulatory defects compared with those OEAs with intact ␤-catenin regulation (examples in Fig. 7).
Critical Role of TCF Binding Sites in AXIN2 Proximal Promoter in ␤-Catenin-mediated Induction-To establish further the role of TCFs in regulating AXIN2 expression, we examined AXIN2 genomic sequences for candidate TCF binding sites. The only consensus TCF binding site identified in a search of sequences located from -1500 to ϩ500 bp (relative to the presumed transcriptional start site) was found at -108 to -102 (i.e. CTTTGAT; Fig. 8A). Luciferase reporter gene constructs containing this element as well as reporter gene constructs in which the element was mutated to CTTTGGC were generated. In DLD1 and SW480 colon cancer cells, we found that mutation of the consensus TCF site in the AXIN2 promoter markedly decreased the activity of a reporter construct containing roughly 1.0 kb of AXIN2 promoter sequence ( Fig. 8B and data not shown). Similar results were obtained in DLD1 and SW480 colon cancer cells with wild type and mutant reporter gene constructs containing only 181 bp of AXIN2 promoter sequences ( Fig. 8B and data not shown). Moreover, although expression of a dominant negative TCF-4 mutant protein (dnTCF-4) inhibited the activity of the wild type AXIN2 reporter construct, the dnTCF-4 protein had no major effect on the activity of the reporter gene construct harboring mutations in the consensus TCF binding site (Fig. 8B). Interestingly, cotransfection experiments in HEK293, COS, and HeLa cells, in which an expression vector encoding the S33Y mutant ␤-catenin protein was cotransfected with the AXIN2 reporter gene constructs, revealed that ␤-catenin was not sufficient on its own for activation of AXIN2 transcription via the proximal TCF element. 2 The differing results in colon cancer versus other cell lines suggest that cellular context, perhaps including the expression of other cellular proteins, may play a role in the ability of ␤-catenin to activate AXIN2 transcription (see "Discussion").
The Axin2 Protein Regulates Wild Type but Not Mutant ␤-Catenin-Prior work has shown the rat Axil and mouse conductin proteins can negatively regulate Wnt signaling perhaps in large part as a result of the ability of Axil/conductin to serve as a "scaffold" for efficient coordination of the interactions of GSK3␤, APC, and ␤-catenin, resulting in the phosphorylation of ␤-catenin at critical N-terminal sites (10,28). To confirm that the human Axin2 protein had an analogous function, we assessed its ability to antagonize ␤-catenin effects on TCF transcription. As shown in Fig. 9, whereas the ability of wild FIG. 3. Wnt-1 induced activation of conductin in murine breast epithelial cells. Northern analysis of conductin was carried out on total RNA isolated from the following: parental RAC311 cells, polyclonal populations of RAC311 cells transduced with empty retroviral expression vector (RAC311/Vect only) or a vector encoding Wnt-1 (RAC311/Wnt-1), a clonal RAC311 line selected for morphological transformation and high Wnt-1 expression (RAC311/Wnt-1 no. 9), and polyclonal populations of C57MG cells transduced with an empty retroviral expression vector (C57/Vect only) or a vector encoding Wnt-1 (C57/Wnt-1). The blot was stripped and rehybridized to a control GAPDH probe to control for loading and transfer efficiency. All cell lines have been described previously (32).
AXIN2 Is a ␤-Catenin-TCF-regulated Target Gene type ␤-catenin to activate TCF transcription was strongly inhibited by Axin2, the S33Y mutant form of ␤-catenin was not significantly inhibited by Axin2. These findings indicate that the Axin2 protein is, as expected, a negative regulator of wild type ␤-catenin and Wnt signaling. DISCUSSION The critical role of the Wnt pathway in development has long been appreciated. Nevertheless, only in the recent past has it become abundantly clear that mutations in Wnt pathway components play a prominent role in the pathogenesis of a rather broad array of human cancers (3)(4)(5). A principal effect of the loss-of-function mutations in APC or the gain-of-function mutations in ␤-catenin is to elevate ␤-catenin levels in the cytoplasm and nucleus. As a result of its deregulation, the ability of ␤-catenin to complex with TCFs is enhanced and altered transcription of TCF-regulated genes ensues. Thus far, it appears that activation of ␤-catenin-TCF-regulated target genes is a major consequence of Wnt pathway deregulation in cancer. Candidate ␤-catenin-TCF target genes described in the literature include c-myc, CCND1, MMP-7, Tcf-1, PPAR␦, PEA3, ENC1, c-ETS2, c-myb, and c-kit (15)(16)(17)(18)(19)(20)(21)(22)(23).
In this paper, we have presented a substantial body of data implicating the human AXIN2 gene (and the rat Axil and mouse conductin genes) as a downstream target of the Wnt-␤catenin-TCF pathway. Findings consistent with those we report here on AXIN2/conductin/Axil expression and its regula-tion by the Wnt pathway were recently published by others (39 -41). We initially found that the rat Axil gene was strongly induced upon neoplastic transformation of RK3E cells by mutant ␤-catenin or ␥-catenin or after 4-OH-T-induced activation of a ␤-catenin-ER fusion protein. In murine breast epithelial cells, we found that overexpression of Wnt1 strongly activated the conductin gene. Human colon cancer cell lines had elevated AXIN2 expression, and restoration of APC function or expression of a dominant negative form of TCF-4 in the cells strongly inhibited AXIN2 expression. Primary ovarian carcinomas with defective ␤-catenin regulation were found to have elevated AXIN2 gene and protein expression compared with a similar cohort of ovarian carcinomas with intact ␤-catenin regulation.
Consistent with the notion that the AXIN2/Axil/conduction genes are activated directly as a result of binding of the ␤-catenin-TCF protein complex to regulatory elements within or nearby the genes, we found that Axil was robustly activated by ␤-catenin in the absence of new protein synthesis. Use of reporter gene constructs containing proximal promoter sequences from the AXIN2 gene established the ability of ␤-catenin to activate AXIN2 transcription as well as the key role of TCFs in AXIN2 activation. Although our findings indicate that ␤-catenin and TCFs play a vital role in the activation of AXIN2 expression in colon and ovarian cancer cells, our observation that the activity of the AXIN2 proximal promoter was not demonstrably affected by ␤-catenin in several other epithelial The control ␤-galactosidase-expressing vector pCH110 was used to correct for differences in transfection efficiency. B, AXIN2 and AXIN1 expression was analyzed by Northern blot analysis of total RNA from the cell lines, and GAPDH probe was used to control for RNA loading and transfer.

FIG. 4. Expression of AXIN2 in human colon cancer cells and its regulation by APC function.
A, AXIN2 expression in the indicated human colon cancer cell lines was assessed by Northern blot analysis. B, restoration of wild type APC expression in the HT29 colon cancer cell line represses AXIN2 expression but not AXIN1 expression. Northern blot analysis was performed on RNA isolated from an HT29 cell line that displays ZnCl 2 -inducible wild type APC expression (HT29/APC) and a control HT29 line with ZnCl 2 -inducible ␤-galactosidase expression (HT29/␤-Gal). The RNA was isolated prior to ZnCl 2 treatment (0 h) or after various exposure times (6, 12, and 18 h). After hybridization to the AXIN2 and AXIN1 probes, the blots were stripped and rehybridized to a GAPDH probe to control for loading and transfer. cell types, namely HEK293, COS, and HeLa cells, 2 suggests that the regulation of AXIN2 transcription by ␤-catenin-TCF may be complex. For example, it is possible that only certain TCF isoforms may bind to and regulate the AXIN2 promoter, and these TCF isoforms display tissue-or cell type-restricted patterns of expression. Alternatively, other transcription factors that bind to specific sites in the AXIN2 promoter may play a key role in cooperating with ␤-catenin-TCF to activate TCF transcription. Prior studies have suggested that cooperation between ␤-catenin-TCF and other transcription factors may be important for activation of certain genes, such as the cooperation between ␤-catenin-TCF and PEA3 in the activation of MMP-7 (42).
In light of prior data in the literature and the data presented here showing that Axin2 can negatively regulate ␤-catenin function, our findings imply that AXIN2 is a negative feedback regulator of the Wnt pathway. Interestingly, even though the Axin1 and Axin2 proteins appear to have similar functions in negatively regulating ␤-catenin levels via the ability of the Axins to complex GSK3␤, APC, and ␤-catenin, we obtained no clear-cut evidence that the human AXIN1 gene or its rat homolog rAxin1 was induced by Wnt pathway activation. The differential effects of the Wnt pathway on AXIN1 and AXIN2 suggest that there may be potentially important functional differences between the proteins. For instance, although the two proteins share roughly 45% amino acid identity, they may differ in their ability to interact with other cellular proteins. Thus far, the Axin1 protein has been shown to bind to multiple other proteins besides Wnt pathway factors (i.e. APC, GSK3␤, ␤-catenin, disheveled). These other Axin1-interacting proteins include the following: the mitogen activated protein kinase (MAPK) kinase kinase (MEKK1) protein (43); the GSK3␤-binding protein (44); the PR61␤ and PR61␥ regulatory subunits of protein phosphatase 2A (45,46); the low density lipoprotein receptor-related protein-5 (47), which function as a Wnt coreceptor; the transforming growth factor-␤ pathway transcription factor Smad3 (48); and a novel protein termed Axam (49). Given the apparently large number of Axin1-interacting proteins, if Axin1 expression was strongly induced by Wnt pathway activation, there might be significant effects on many other signaling pathways besides the Wnt pathway. Thus far, it is not clear whether the Axin2 protein binds any or all of these other Axin1-interacting proteins. However, some of the interactions between Axin1 and the non-Wnt pathway-interacting proteins are mediated via regions that are not highly conserved between Axin1 and Axin2. As such, perhaps the differential interactions of Axin1 and Axin2 with certain of the non-Wnt pathway proteins accounts for why Axin2 functions as a major negative feedback regulator of Wnt signaling, and Axin1 does not.
Although bi-allelic inactivation of AXIN1 has been seen in some hepatocellular carcinomas and medulloblastomas (25,26), indicating that AXIN1 functions as a tumor suppressor gene, bi-allelic inactivation of AXIN2 in cancers has not yet been noted. To date, mutations in the AXIN2 gene appear to be restricted to colon and perhaps other cancers with mismatch repair pathway defects (27,36). The truncated AXIN2 alleles seen in cancers with mismatch repair defects have been proposed to encode proteins that function in a dominant negative fashion to interfere with ␤-catenin regulation (27). Because the ability of Axin2 to regulate ␤-catenin appears to depend upon intact APC function and wild type ␤-catenin N-terminal sequences, in those cancers with inactivating mutations in APC or oncogenic mutations in ␤-catenin, elevated Axin2 expression is quite unlikely to have any major inhibitory effect on ␤-catenin levels and function. Even in cancers with AXIN1 or AXIN2 mutations, because the Axin proteins have been suggested to FIG. 6. AXIN2 expression is increased markedly in OEAs with nuclear ␤-catenin localization compared with OEAs with nonnuclear ␤-catenin localization. cDNA preparations from 42 snap frozen OEA specimens that had been studied previously for ␤-catenin immunohistochemistry and mutations in critical Wnt pathway components (␤-catenin, APC, AXIN1, and AXIN2) (36) were subjected to quantitative real time (TaqMan) analysis of AXIN2 expression, using primer pairs and fluorescent probes for AXIN2 and HPRT described under "Experimental Procedures." Using HPRT to normalize, the relative AXIN2 fluorescence of the 12 samples with strong nuclear staining for ␤-catenin and mutations in the ␤-catenin, APC, AXIN1, or AXIN2 genes was compared with the relative fluorescence of the 30 OEAs lacking strong nuclear ␤-catenin staining and pathway mutations. Student's t test was used to determine the significance of differences in AXIN2 expression between the two groups.
FIG. 7. Immunohistochemical staining reveals elevated AXIN2 expression in OEAs with ␤-catenin regulatory defects compared with OEAs with intact ␤-catenin regulation. OEA specimens that had been studied previously for ␤-catenin immunohistochemistry and mutations in critical Wnt pathway components (␤-catenin, APC, AXIN1, and AXIN2) (36) were used for AXIN2 immunohistochemistry studies. Representative photomicrographs of the staining seen in OEA specimens with intact ␤-catenin regulation (A-D) and OEAs with defective ␤-catenin regulation (E-H) are shown. dimerize (28,50), it is possible that ␤-catenin cannot be downregulated by induction of AXIN2 because wild type function of both the Axin1 and Axin2 proteins is required. In light of the observations indicating that the elevated expression of AXIN2 in cancers with Wnt pathway defects is not sufficient to downregulate ␤-catenin levels and function, it is possible that Axin2 induction might have other important effects in cancer cells, potentially even growth promoting effects. Further studies of the interactions between Axin2 and other cellular proteins should offer insights into Axin2 function as well as the consequences of its induction by Wnt pathway activation in normal and cancer cells. A, schematic representation of AXIN2 reporter gene constructs. The AX2(1078WT)/Luc reporter vector contains AXIN2 sequences from Ϫ1078 to ϩ5 relative to the presumptive transcription start site, and the vector AX2(181WT)/Luc contains AXIN2 sequences from Ϫ181 to ϩ5. The AX2(1078Mut)/Luc and AX2(181Mut)/Luc vectors carry mutations in the TCF consensus element. B, effects of mutations and dominant negative TCF-4 (dnTCF-4) on the activity of AXIN2 reporter gene vectors in colon cancer cells. DLD1 (left) and SW480 (right) cells were transfected with the indicated reporter gene constructs, and luciferase activity was measured. In the case of experiments with SW480 cells, the luciferase constructs were cotransfected with either an empty expression vector or the vector encoding dnTCF-4. The assays were performed in triplicate, mean Ϯ S.D. values are shown, and the control ␤-galactosidase-expressing vector pCH110 was used to correct for differences in transfection efficiency.
FIG. 9. Axin2 can inhibit the activity of wild type ␤-catenin but not an N-terminal mutant (S33Y) form of ␤-catenin. HEK293 cells were cotransfected with the indicated pcDNA3 expression vectors and the TOPFLASH or FOPFLASH flash reporter construct. Equal masses of DNA were used in each transfection, luciferase activities were measured in triplicate, and the ratio of luciferase activities in TOPFLASHtransfected versus FOPFLASH-transfected cells was determined and reported as the relative TCF activity. The control ␤-galactosidase-expressing vector pCH110 was used to correct for differences in transfection efficiency.