Axin-dependent Phosphorylation of the Adenomatous Polyposis Coli Protein Mediated by Casein Kinase 1ε*

Axin and the adenomatous polyposis coli protein (APC) interact to down-regulate the proto-oncogene β-catenin. We show that transposition of an axin-binding site can confer β-catenin regulatory activity to a fragment of APC normally lacking this activity. The fragment containing the axin-binding site also underwent hyperphosphorylation when coexpressed with axin. The phosphorylation did not require glycogen synthase kinase 3β but instead required casein kinase 1ε, which bound directly to axin. Mutation of conserved serine residues in the β-catenin regulatory motifs of APC interfered with both axin-dependent phosphorylation and phosphorylation by CKIε and impaired the ability of APC to regulate β-catenin. These results suggest that the axin-dependent phosphorylation of APC is mediated in part by CKIε and is involved in the regulation of APC function.

Mutations that interfere with the ability of cells to regulate ␤-catenin have been identified in a wide variety of human cancers (reviewed in Ref. 1). In some cases the mutations occur in the ␤-catenin gene (CTNNb1) itself, affecting specific amino acids that are integral to the ubiquitin-dependent targeted degradation of the protein. In other cases, proteins that facilitate the process of ␤-catenin degradation are instead mutated, and the wild type ␤-catenin protein assumes a prolonged halflife. In either scenario the outcome is aberrant transcription of genes that are activated by ␤-catenin signaling. Two of the proteins that facilitate the turnover of ␤-catenin are axin and the adenomatous polyposis coli (APC) 1 protein, both of which bind directly to ␤-catenin (2)(3)(4)(5)(6). The APC tumor suppressor was identified as the gene responsible for a heritable predisposition to colorectal cancer, referred to as familial polyposis coli, and is also mutated in most sporadic colorectal cancers (7)(8)(9)(10). Axin was shown to regulate negatively wnt-1 signaling, and its direct association with APC, ␤-catenin, and glycogen synthase kinase 3␤ (GSK3␤) is consistent with its involvement in the wnt-1 pathway (2,(11)(12)(13)(14). Recently, mutations that inactivate axin were identified in a subset of human hepatocellular cancers and cancer cell lines (15). Thus, APC and axin function in concert to regulate ␤-catenin, and the inactivation of either of these genes, or the mutational activation of ␤-catenin, contributes to the oncogenic state of transformed cells.
A multiprotein complex consisting minimally of APC, Dishevelled, GSK3␤, and protein phosphatase PP2A, all of which bind directly to independent sites on axin (16), regulates the stability of ␤-catenin. ␤-Catenin binds directly to both APC and axin, and its phosphorylation by GSK3␤ leads to its recognition by a ubiquitin-protein isopeptide ligase containing the F box protein ␤-TRCP (17)(18)(19)(20). In vitro reconstitution experiments have demonstrated that the association of axin with APC enhances the binding of ␤-catenin to axin and consequently facilitates the phosphorylation of ␤-catenin by GSK3␤ (21). APC is also phosphorylated by GSK3␤, which is thought to promote entry of ␤-catenin into the complex by increasing the binding affinity of ␤-catenin for APC (22). The phosphorylation of APC by GSK3␤ is also enhanced by the binding of ␤-catenin to axin. Although this multiprotein complex has the capacity to promote the degradation of ␤-catenin, it might also serve as a nexus for incoming signals that stabilize ␤-catenin, thereby activating gene transcription. The stabilization of ␤-catenin occurs in response to stimulation of the Wnt-1 receptor, which signals through Dishevelled to inhibit GSK3␤. It is not clear how the Wnt receptors communicate with Dishevelled, but the subsequent inactivation of GSK3␤ might involve the GSK3␤binding protein GBP/FRAT, which binds to Dishevelled (23)(24)(25). Additional regulatory molecules are likely involved, and recent work has implicated casein kinase 1⑀ (CKI⑀) in the propagation of the wnt signal (26,27). Overexpression of CKI⑀ in Xenopus embryos resulted in the duplication of the dorsal axis, a phenotype typically observed following overexpression of wnt-1. Additional experiments demonstrated that CKI⑀ associated with and phosphorylated Dishevelled, and expression of a kinase-inactive mutant interfered with the activation of ␤-catenin signaling by wnt-1. These results implicated CKI⑀ as a positive mediator of wnt signaling.
We have investigated the interaction between APC and axin, and we show this is necessary for the down-regulation of ␤-catenin by APC in mammalian cells. APC was also phosphorylated in response to binding axin, in a manner that did not require the association of GSK3␤ with axin. Our results suggest that APC is a substrate for CKI⑀, which binds directly to axin, and that the sites phosphorylated by it are involved in the regulation of ␤-catenin.

MATERIALS AND METHODS
Cell Lines-SW480 cells were derived from a human colorectal cancer (ATCC reference CCL228); 293 cells are an immortalized embryonic kidney cell line (ATCC reference CRL1573), and SNU475 is a human hepatocellular cancer cell line (ATCC CRL2236). Growth conditions were according to ATCC guidelines.
cDNA Constructs and Site-directed Mutagenesis-The construct APC f2-7 (28) codes for amino acids 1342-2075 of the full-length APC sequence and was previously designated APC25 (4). The fbc1,2 construct * 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  Antibodies and Immunological Procedures-The polyclonal antibodies raised against the central region of APC (APC2) have been described previously (6). Polyclonal antibody to ␤-catenin was described previously (22), and monoclonal antibody to ␤-catenin was purchased from Transduction Laboratories (Lexington, KY). Polyclonal antibodies to GSK3␤ and CKI⑀ were purchased from Santa Cruz Biotechnology (Santa Cruz, CA), and polyclonal antibodies to axin were raised in rabbits as described previously (11). Monoclonal antibodies to the Myc epitope have been described elsewhere (29) and were partially purified on DE52 cellulose for use in our studies. For immunoprecipitations, cells were lysed in Triton X-100 lysis buffer (20 mM Tris-HCl, pH 8.0, 1.0% Triton X-100, 137 mM NaCl, 10% glycerol, 1 mM EGTA, 1.5 mM MgCl 2 , 1 mM dithiothreitol, 1 mM sodium vanadate, 50 mM NaF, 1 mM Pefabloc, 10 g/ml each of aprotinin, pepstatin, and leupeptin), and ϳ300 l of lysate containing 2 mg of total protein each was incubated with the appropriate antibody for 4 h at 4°C. Antibodies were recovered using protein A-or G-Sepharose, and the beads were washed three times with 1 ml each of buffer B (20 mM Tris-HCl, pH 8.0, 150 mM NaCl, 0.5% Nonidet P-40) and finally eluted with SDS-PAGE sample buffer. For Western blotting, polyclonal sera were used at 1:1000 dilution, affinity-purified anti-APC2 at 0.2 g/ml, and anti-Myc at 1 g/ml. Blots were developed using the ECL system (Amersham Pharmacia Biotech). Immunofluorescent detection of ␤-catenin in fixed whole cells was carried out as described previously (4). The transfected SW480 cells were grown on coverslips, fixed in methanol, and stained using mouse monoclonal antibodies to ␤-catenin (Transduction Laboratories). The cells were costained with 4,6-diamidino-2-phenylindole to visualize nuclei.
Reporter Assays-SW480 cells were seeded in 6-well plates (5 ϫ 10 5 cells/well) and transfected with 0.75 g of pTopflash (provided by H. Clevers), 0.25 g of Renilla luciferase (pRL SV40), and 1-3 g of indicated plasmid using Lipofectin (Life Technologies, Inc.). Media were changed at 24 h post-transfection, and cells were lysed after an additional 24 h in 300 l of lysis buffer as described above. Duplicate 10-l aliquots were assayed using the Dual-luciferase Reporter Assay System (Promega). An aliquot of each sample was frozen in sample buffer for Western analysis for the indicated transfected constructs.
In Vitro Kinase Assays-293 cells (10-cm dishes) were transfected with 3 g of plasmid encoding Myc epitope-tagged axin or axin L396P using Effectene (Qiagen) and harvested as described above. Protein extracts were immunoprecipitated with anti-Myc antibody, and the immunoprecipitated protein complex was washed twice in buffer B and then twice in kinase buffer (25 mM Hepes, pH 7.2, 10 mM MgCl 2 , 10% glycerol, 5 mM dithiothreitol). Washed immunoprecipitates were split into 4 tubes and incubated in 20 l of kinase buffer with either GSK3␤ phosphopeptide substrate, CKI phosphopeptide substrate, CKII peptide substrate (New England Biolabs), or no peptide plus 50 M [␥-32 P]ATP (ϳ10,000 cpm/pmol). Assays were performed for 20 min at 30°C and then a 10-l aliquot was spotted onto P81 paper (Whatman) which was then washed and subjected to scintillation counting.
Other Procedures-All SDS-gel electrophoresis was performed using precasted gels according to the manufacturer's instructions (NOVEX). For treatment of cells with N-(2-aminoethyl)-5-chloroisoquinoline-8-sulfonamide (CKI-7, Toronto Research Chemicals), cells were transfected, and media were changed at 24 h post-transfection and incubated for an additional 20 -24 h in 150 M CKI-7. Yeast two-hybrid analysis of protein interactions was carried out as described previously (11).

RESULTS
Mutations in the APC gene that are associated with colorectal cancer progression typically result in truncated proteins lacking all three of the axin-binding sites (30,31). The selection against the presence of axin-binding sites on APC strongly implies that the direct interaction of these two proteins is important for the regulation of ␤-catenin. Here we have employed a small fragment of APC, fbc1,2, that is incapable of regulating ␤-catenin, and we asked whether the transposition of an axin-binding site to it would restore this function (Fig.  1A). The fbc1,2 fragment contains the second and third 15amino acid ␤-catenin-binding sites and the first two 20-amino acid repeat motifs (28). The APC fragments containing each of the three axin-binding sites, but not the control fragment fbc1,2, associated with endogenous axin when expressed in SW480 cells (Fig. 1B). In this figure, differences in gel mobility between the control and AxBs fragments are attributed to a stop codon introduced 3Ј to the axin-binding sites that was not present in fbc1,2. All subsequent experiments were performed with an fbc1,2 construct containing a stop codon equivalent in position to that in the AxBs constructs. The APC fragment containing the third axin-binding site (AxBs3) exhibited the highest level of axin binding and was selected for further studies on the regulation of ␤-catenin. Expression of this fragment, as well as the f2-7 APC fragment, in SW480 cells resulted in the down-regulation of ␤-catenin-dependent transcriptional activity (Fig. 1C). Thus the AxBs3 fragment is functional in this assay, although it is less active than the larger f2-7 fragment that was expressed at lower levels than AxBs3. To determine whether AxBs3 affected ␤-catenin protein levels, SW480 cells were transiently transfected with either AxBs3 or fbc1,2, and fixed cells were stained with antibody specific to ␤-catenin and the nuclei visualized by staining with 4,6-diamidino-2-phenylindole. The presence of dark blue nuclei in the merged image indicates a loss of nuclear ␤-catenin staining, whereas blue-green color is indicative of nuclei that retain ␤-catenin. The transfection efficiency under these conditions is 50 -75% (4). Approximately half of the cells transfected with AxBs3 exhibited a loss of nuclear ␤-catenin that was not evident in cells transfected with vector control or fbc1,2 control fragment (Fig. 1D). Thus, small fragments of APC that associate with ␤-catenin are able to regulate its stability provided they possess an axin-binding site.
In the course of carrying out experiments on the regulation of ␤-catenin, we noted a difference in mobility on SDS gels between the AxBs3 and fbc1,2 APC fragments that was accentuated on coexpression with axin ( Fig. 2A). The mobility shift was eliminated by treating the immunoprecipitates with phosphatase prior to electrophoresis, demonstrating phosphorylation of the APC fragment (data not shown). The association of axin with GSK3␤ implicates this kinase as a likely candidate for mediating the axin-dependent phosphorylation of AxBs3 (2, 3, 11, 13). To examine this, we overexpressed the GSK3␤-binding protein FRAT together with AxBs3 and axin. Although FRAT is known to interact with GSK3␤ in a manner that precludes its association with axin (25, 32), it did not affect the electrophoretic mobility of AxBs3 (Fig. 2, B and C). The RGS domain fragment of axin would be expected to bind APC and thereby compete with its ability to interact with full-length axin. Accordingly, overexpression of the axin RGS domain, at levels comparable to that of FRAT, completely abrogated the ability of axin to promote the phosphorylation of AxBs3. RapGAP was expressed as a functionally irrelevant control and had no effect on AxBs3 phosphorylation. It was possible that FRAT was not expressed at levels sufficient to inhibit GSK3␤ in our assay. To address this, we introduced the mutation L396P into axin, which completely eliminated GSK3␤ binding (33). GSK3␤ was detected in the immunoprecipitates of wild type but not the mutant axin, whereas ␤-catenin coimmunoprecipitated with both wild type and mutant axins (Fig. 3A). Despite the lack of association with GSK3␤, coexpression of the axin L396P mutant with AxBs3 promoted its phosphorylation to a degree comparable with that observed with wild type axin (Fig. 3B). Axin associates with itself through a C-terminal oligomerization domain (34,35), and it was therefore possible that the axin mutant recruited endogenous wild type axin along with GSK3␤ into the APC complex. To rule this out, we tested the human hepatocellular cancer cell line SNU 475 that does not express wild type axin (15). AxBs3 was still phosphorylated in this cell line when coexpressed with the axin L396P mutant (Fig. 3B).
The results obtained with axin L396P suggested that a kinase other than GSK3␤ was in part responsible for the axindependent phosphorylation of APC. We therefore tested CKI⑀, which has also been implicated in the wnt-1 signaling pathway (26,27). Coexpression of a kinase-inactive mutant of CKI⑀ (kinase dead-CKI⑀) with AxBs3 partially inhibited the ability of axin to promote the phosphorylation of AxBs3 (Fig. 4A). More-over, expression of wild type CKI⑀ produced a mobility shift on SDS gels similar to that observed following coexpression of AxBs3 with axin (Fig. 4B). Despite higher levels of expression, GSK3␤ promoted mobility shifts that were far less pronounced than those seen with CKI⑀. We also tested the effect of CKI-7, a compound known to inhibit casein kinase catalytic activity (36). Treatment of cells with CKI-7 significantly inhibited the phosphorylation of AxBs3 induced on coexpression with axin ( Fig. 4C). As expected, the inhibitor also interfered with the phosphorylation of AxBs3 that occurred in response to coexpression with CKI⑀. The mobility of the larger fragment of APC, f2-7, was also increased by treatment of cells with CKI-7 (Fig. 4D).
Casein kinase 1⑀ was reported to associate with the PDZ domain of Dishevelled in vitro and with the axin-Dishevelled complex in cultured mammalian cells (26,27). We sought to map the site of interaction of CKI⑀ with axin and to determine whether CKI⑀ binding occurred independent of GSK3␤ binding. To examine this we expressed wild type and the L396P axin mutant, and we analyzed the immunoprecipitates for kinase activity (Fig. 5A). In the absence of peptide substrate, background levels of activity were detected with the immunoprecipitates. The GSK3␤-specific substrate was highly phosphorylated by the wild type axin immunoprecipitate but not by the L396P mutant. By contrast, a peptide substrate specific for CKI⑀ was phosphorylated by immunoprecipitates of both the wild type and mutant axins. As an additional control we used a substrate specific for casein kinase II, and we were unable to detect any activity above background levels. The binding site for CK1⑀ on axin also mapped to a region distinct from that which associates with GSK3␤. Axin fragments were expressed in 293 cells and assayed for CKI⑀ activity following their recovery by immunoprecipitation (Fig. 5, B and C). Casein kinase activity was specifically associated with axin sequence residing The three axin-binding sites, indicated as ovals, were fused to fbc1,2 to generate the three AxBs fragments. MCR is the mutation cluster region. B, the indicated constructs were expressed in SW480 cells, immunoprecipitated with antibody specific to the Myc epitope tag, and analyzed by immunoblotting for associated axin. C, the indicated constructs were cotransfected with the TopFlash reporter into SW480 cells, and the production of luciferase was measured. The values represent means and standard deviations derived from at least five independent experiments. The f2-7 fragment spans the region of APC containing repeat motifs 2-7 and all three of the axin-binding sites. Immunoblotting was performed on cell lysates normalized to total protein (lower panel). D, SW480 cells were transiently transfected with the indicated constructs, and ␤-catenin (␤-cat.) was visualized by immunofluorescence and nuclei by 4,6diamidino-2-phenylindole (DAPI) staining.
between the ␤-catenin-binding site and the DIX domain.
That CKI⑀ associated with axin independent of GSK3␤ was confirmed by immunoblotting for these two kinases following immunoprecipitation of various axin fragments expressed in 293 cells (Fig. 6 A). Endogenous GSK3␤ and CKI⑀ associated with independent fragments of axin, and the CKI⑀ binding was again localized to a region contained between the ␤-cateninbinding site and the DIX domain. This region of axin has been reported to bind to the catalytic subunit of protein phosphatase 2A (PP2Ac) and to Dishevelled (34,35,37). Therefore, it was possible that CKI⑀ was indirectly associated with axin though its interaction with one of these two proteins. However, we were unable to detect Dishevelled or PP2Ac in our axin immunoprecipitates, despite the presence of ample protein in the cell lysates (Fig. 6A). Moreover, the binding of CKI⑀ to axin was also detected using a yeast two-hybrid format, suggesting that the interaction between these two proteins was direct (Fig. 6B). If CKI⑀ is essential for axin-dependent phosphorylation of APC, then axin fragments capable of performing this activity must contain binding sites for both APC and CKI⑀. Accordingly, an axin fragment containing binding sites for APC, ␤-catenin, and GSK3␤ (amino acids 1-475) was inactive, whereas extension of this sequence to include the CKI⑀ interaction region (amino acids 1-712, FL) restored axin-dependent phosphorylation of APC (Fig. 7).
Numerous highly conserved serine and threonine residues are contained within the 20-amino acid repeat motifs that constitute the ␤-catenin regulatory domain of APC (38). To identify sites of phosphorylation, we mutated serine residues present in the 20-amino acid repeat motifs that fit the consensus for phosphorylation by CKI⑀. Serines and threonines phosphorylated by CKI⑀ are frequently followed by aliphatic residues such as isoleucine or leucine and are preceded by negatively charged amino acids or a phosphorylated serine or threonine residue at the Ϫ3 position (39,40). Serine 1392 in APC, contained in the second 20-amino acid repeat motif, conforms to this consensus and was therefore mutated to an alanine residue in the AxBs3 fragment. The S1392A-AxBs3 mutant was markedly resistant to phosphorylation induced by either axin or CKI⑀ coexpression (Fig. 8). Substituting a serine for aspartate will sometimes mimic the phosphorylated state, so we tested the S1392D-AxBs3 mutant. This mutation restored the phosphorylation observed on coexpression with axin or CKI⑀. If serine 1392 is a site of phosphorylation by CKI⑀ then serine 1389, at the Ϫ3 position, might represent a phosphorylated priming site. Accordingly, the S1389A-AxBs3 mutant was resistant to phosphorylation on coexpression with axin or CKI⑀. Mutation of serine 1389 to an aspartate residue partially restored the axin-dependent phosphorylation of the AxBs3 fragment. It has been reported that a single aspartate residue serves only as a weak priming site for CKI⑀, which is consistent with the partial effects of the S1389D mutant (40). We next introduced the corresponding serine mutations into the first 20-amino acid motif present in AxBs3. Although all of these mutants underwent substantial mobility shifts in response to axin or CKI⑀ coexpression, the S1276A, S1276D, and S1279A mutants exhibited slightly faster mobility on SDS gels in the basal state (Fig. 8). By contrast, the S1279D mutant appeared to migrate normally relative to wild type AxBs3. The results indicate that phosphorylation of serines residues in the second 20-amino acid repeat sequence are largely responsible for the axin-dependent mobility shifts observed with AxBs3. This was not the case for serine residues in the first 20-amino acid repeat sequence, although phosphorylation of some of these serines might influence the overall mobility of AxBs3 irrespective of axin binding.
It is possible that APC utilizes the amino acids phosphorylated by CKI⑀ for the regulation of ␤-catenin and that their phosphorylation affects this activity. We tested this by examining the ability of AxBs3, and the corresponding serine mutants, to down-regulate ␤-catenin-dependent transcription in SW480 cells. The AxBs3 reduced transcriptional activity by ϳ50%; however, the activity of the S1392A was impaired (Fig.  9A). This suggested that either the phosphorylation of this amino acid is required for full APC activity or that any modification, including phosphorylation of serine 1392, is inhibitory to APC. To discriminate between these possibilities, we tested the AxBs3 S1392D mutant, which was designed to approximate the phosphorylated state. Mutation of serine 1392 to aspartate partially restored APC function relative to the S1392A mutant. Thus the phosphorylation of serine 1392 appears to play a positive role in the down-regulation of ␤-catenin signaling by APC. Mutation of serine 1389, the presumptive priming site for serine 1392, to an alanine also impaired the activity of AxBs3. However, mutation of serine 1389 to an aspartate did not restore regulatory activity. Again, this might relate to the inability of a single aspartate residue to serve as a competent priming site. A similar pattern was observed for the corresponding mutations in the first 20-amino acid repeat motif. Mutation of serine 1279 to an alanine, which corresponds to S1392A, again impaired the activity of the AxBs3 APC fragment, whereas mutation to an aspartate restored activity. The putative priming site for phosphorylation of serine 1279 by CKI⑀ is serine 1276, and its mutation to alanine also impaired the regulatory activity of the AxBs3 fragment. As was observed for serine 1389, substitution of serine 1276 with aspartate did not restore activity to the AxBs3 fragment. Assessing the relative impact of a single serine mutation on APC activity in the context of AxBS3 could be misleading due to the limited size and structure of this fragment. Therefore we also tested the effects of the S1389A in the contest of the larger APC fragment f2-7. Although the f2-7 APC fragment was not completely inhibited by the S1389A mutation, an ϳ50% reduction was observed.
It was possible that some of the AxBs3 mutants were inactive as a result of their inability to bind to ␤-catenin. This was addressed by immunoprecipitating the APC fragments from SW480 cells and examining the immune complexes for the presence of ␤-catenin. All of the APC fragments coimmunoprecipitated with amounts of ␤-catenin that exceeded background levels. The control fragment fbc1,2, which lacks the axin-binding site, and the wild type AxBs3 were associated with similar amounts of ␤-catenin (Fig. 9B). The S1276A, S1276D, and S1279A all coimmunoprecipitated with apparently lower amounts of ␤-catenin suggesting that these three mutants might be somewhat compromised in their ability to bind ␤-catenin. All constructs containing mutations in the second 20amino acid motif were associated with similar amounts of ␤-catenin.
As seen in Fig. 9B, a limited amount of axin-dependent phosphorylation of AxBs3 occurred even in the absence of ectopically expressed axin. This was most apparent when comparing fbc1,2 to wild type AxBs3. We believe that this phosphorylated form of AxBs3 is limited by the amount of endogenous axin available in the cell and reflects the steady state level of the active form of the AxBs3-axin complex. Three of the mutations in the second 20-amino acid repeat motif, S1389A, S1389D, and S1392A, eliminated this limited phosphorylation state and also inhibited ␤-catenin regulatory activity. By contrast, the fourth mutant, S1392D, retained activity and also exhibited limited phosphorylation. None of the mutations introduced into the first 20-amino acid repeat motif prevented axin-dependent phosphorylation of AxBs3, although three of the mutations, S1276A, S1276D, and S1279A, abrogated activity and increased the overall mobility of the fragments on SDS gels. The fourth mutation, S1279D, restored regulatory activity to AxBs3 and also retarded the overall mobility of the fragment on SDS gels relative to the other three mutants. Overall, the results indicate that the phosphorylation state of APC relates to its regulatory activity. In particular, the phosphorylation of serine residues 1279 and 1392 appears to directly affect activity as their mutation to aspartate restored function to AxBs3 that was lost upon their replacement with alanine.

DISCUSSION
Oncogenes and tumor suppressors that contribute to the progression of various human tumors have been identified through the elucidation of the wnt-1 signaling pathway (1,41,42). From these studies it has become clear that the regulation of ␤-catenin is central to wnt-1 signaling and represents the focal point of the genetic defects in the pathway that leads to cancer. The identification of axin and its interaction with the APC tumor suppressor protein and GSK3␤ imply the existence of a multiprotein complex that coordinates the phosphorylation and presentation of ␤-catenin for ubiquitin-dependent degradation. The importance of axin in this activity is underscored by the mutational spectrum of APC in cancer. The pattern of mutations suggests there is strong selection for the elimination of all three axin-binding sites from APC. Prior to the discovery Site directed mutagenesis was performed to generate AxBs3 constructs in which alanine or aspartate residues were substituted for the indicated serine residues. The sequences of the first and second 20amino acid ␤-catenin regulatory motifs are presented with the targeted serines underlined. The AxBs3 mutants were transiently cotransfected into 293 cells with empty vector, axin or CKI⑀ and cell lysates were analyzed by immunoblotting with antibody specific to APC. Approximately equal amounts of axin and CKI⑀ protein were detected in each of the corresponding lysates (not shown).
of axin, we had demonstrated that the pattern of mutations in APC correlated strongly with the loss of its ability to regulate ␤-catenin (28). In retrospect, it is now apparent that this loss of regulatory capacity likely relates to the loss of axin-binding sites. This has been borne out by mouse models in which the cancer phenotype is dependent upon the elimination of all axin-binding sites from the mutant truncated APC protein (43). These results are also consistent with a study showing that the limited disruption of axin-binding sites in APC abrogates its ability to regulate ␤-catenin (44). Here we have reinforced this notion by demonstrating that a small fragment of APC can be engineered to regulate ␤-catenin by the addition of a 17-amino acid sequence that confers axin binding to the fragment.
One consequence of axin binding is that of phosphorylation of APC. This has been noted previously and largely attributed to GSK3␤, which is known to associate with axin (11,23,45,46). However, our data indicate that the axin-dependent phosphorylation of APC is not entirely due to GSK3␤. A mutant form of axin that did not bind to GSK3␤ nevertheless promoted the phosphorylation of APC fragments when coexpressed with them. This also occurred in a cell line that lacked wild type axin and could not then be attributed to oligomerization of wild type with mutant axin. Moreover, the overexpression of FRAT, which competitively binds GSK3␤ with respect to axin (32), had no apparent effect on the phosphorylation of APC fragments that was induced on coexpression with axin. Although APC is a good substrate for GSK3␤, and its affinity for ␤-catenin is affected by this phosphorylation (22,23), our data suggest that other kinases might also be involved, particularly in the axindependent phosphorylation of APC. Based on recent reports of the involvement of CKI⑀ in the wnt-1 signaling pathway, we tested this kinase for phosphorylation of APC (26,27). Overexpression of CKI⑀ resulted in the phosphorylation of APC fragments in a manner similar to that observed in response to axin overexpression, and interference with endogenous CKI⑀ activity interfered with the axin-dependent phosphorylation of the APC fragments. Also, the association of CKI⑀ with the axin complex is consistent with a role for this kinase in axin-dependent phosphorylation of APC.
Although our data implicate CKI⑀ as responsible for the mobility shifts in APC observed on coexpression with axin, it does not exclude GSK3␤ from the process of APC phosphorylation. We believe that the phosphorylation events that produce an observable shift in mobility of APC on SDS gels are in part due to CKI⑀. However, the 20-amino acid repeat motifs in APC are very rich in serine and threonine residues, and the phos-phorylation of some of these residues might not be observable as mobility shifts on a background of multiple additional phosphorylations. Accordingly, we have expressed smaller fragments of APC, containing only a single repeat motif, and we have noted mobility shifts in response to treatment of cells with lithium, a known inhibitor of GSK3␤. 2 Also, substrate recognition by GSK3␤ sometimes requires prior phosphorylation of a proximal serine/threonine. Thus, it is possible that additional kinases such as CKI⑀ are required for the priming of APC for phosphorylation by GSK3␤, and inhibition of phosphorylation by CKI⑀ could inhibit subsequent processive phosphorylations by GSK3␤. The results obtained with the S1389A-AxBs3 mutant, for example, suggest that a kinase in addition to CKI⑀ might be involved in the axin-dependent phosphorylation of APC. This mutation completely blocked the phosphorylation of AxBs3 by CKI⑀, yet some phosphorylation of the mutant was still observed on coexpression with axin.
Two serine residues that conform to a consensus site for phosphorylation by CKI⑀ were identified as serines 1279 and 1392, in the first and second 20-amino acid repeat motifs, respectively. These serine residues correspond to each other when the two repeat motifs are aligned. Mutation of either of these serines to alanine compromised the regulatory activity of AxBs3, which was partially recovered when these positions were mutated to aspartate residues. Our interpretation is that serines 1279 and 1392 are phosphorylated directly by CKI⑀ in an axin-dependent manner and that their phosphorylation is essential for the regulation of ␤-catenin by APC. Mutation of serines 1276 and 1389, which occupy the Ϫ3 position relative to serines 1279 and 1392, respectively, also abrogated the regulatory activity of AxBs3. Therefore, when phosphorylated, these two amino acids might serve as priming sites for the phosphorylation of serines 1279 and 1392 by CKI⑀. Substitution of serines 1276 and 1389 for aspartate residues did not fulfill the presumptive priming requirement; however, it has been reported previously (40) that a single aspartate is not an adequate surrogate for this role. Although mutations of corresponding serine residues in the first and second 20-amno acid motifs have similar effects on regulatory activity, their effect on the phosphorylation of AxBs3 differed qualitatively. Mutations in the first motif affected the overall mobility of the AxBs3 fragment on SDS gels but did not abrogate the mobility shifts resulting from axin or CKI⑀ coexpression. By contrast, substi-2 B. Rubinfeld, D. A. Tice, and P. Polakis, unpublished data.
FIG. 9. Effect of serine mutations on regulatory activity of APC. A, the indicated constructs were cotransfected with the TopFlash reporter into SW480 cells and luciferase activity present in the cell lysates was determined. The values represent the means and standard deviations of at least five independent experiments. B, the indicated APC constructs were expressed in SW480 cells, immunoprecipitated from cell lysates with anti-Myc and the immune complexes analyzed for the presence of axin, ␤-catenin or APC. tutions in the second motif did not affect overall mobility but rather prevented the limited phosphorylation observed in the absence of axin or CKI⑀ overexpression and interfered with the stoichiometric bandshift that occurred upon their overexpression. One interpretation is that phosphorylations in the first motif are constitutive while those in the second are transient and depend on the immediate interaction of AxBs3 with axin. The remaining repeats in the full-length APC molecule also contain putative sites of phosphorylation but were not tested here. However, in a previous study (28) we had characterized the relative importance the seven repeat motifs for APC activity and concluded that the second repeat was critical. These results are consistent with the impact of serine mutations in the second repeat motif on axin-dependent phosphorylation of APC.
We have extended the previous findings on the association of CKI⑀ with the axin complex by identifying a CKI⑀-binding site located between the ␤-catenin-binding site and the DIX domain of axin. This region of axin has been reported to bind to the catalytic domain of PP2A (35) and, according to some reports, to the Dishevelled protein (34,37). We believe the binding of CKI⑀ to axin is independent of PP2Ac and Dishevelled, as we were unable to detect either of these proteins in immunoprecipitates of axin and axin fragments, whereas CKI⑀ was detected. Moreover, yeast two-hybrid analysis demonstrated that CKI⑀ bound directly to axin. Peters et al. (26) have reported previously the binding of CKI⑀ to the PDZ domain of Dvl. Therefore, it is conceivable that Dishevelled and axin bind CKI⑀ simultaneously. In any case, the interaction of CKI⑀ with axin appears to be physically independent of its association with GSK3␤, as CKI⑀ kinase activity and protein were associated with a axin mutant incapable of binding GSK3␤.
Casein kinase 1⑀ was recently proposed as a positive regulator of wnt signaling (26,27). These studies showed that CKI⑀ promoted dorsal axis duplication when overexpressed in Xenopus embryos. It should be noted, however, that this has also been demonstrated for the expression of wild type APC, which is clearly a negative regulator of ␤-catenin in other genetic and cell biological settings, including human cancer. Nevertheless, these papers (26,27) provided additional data including biochemical studies in mammalian cells that presented compelling evidence for a positive role for CKI⑀ in wnt-1 signaling. Our proposal for the involvement of CKI⑀ in the down-regulation of ␤-catenin departs somewhat from this interpretation. However, a positive role for CKI⑀ in wnt-1 signaling is not necessarily inconsistent with its potential ability to enable APC in the down-regulation of ␤-catenin. As Dishevelled is also a substrate for CKI⑀, its phosphorylation by this kinase might be needed to affect the inactivation of GSK3␤ in response to a wnt-1 signal. Under these circumstances, CKI⑀ would not facilitate the down-regulation of ␤-catenin by APC, as GSK3␤ is also required for this activity. However, in the absence of wnt-1 signaling and in the presence of functional GSK3␤, the conditions under which we have analyzed APC activity, CKI⑀ could support APC function in the down-regulation of ␤-catenin. Such a model is consistent with CKI⑀ in propagating the wnt-1 signal while readying APC for signal termination once active GSK3␤ again becomes available. This proposal is also in accord with the ability of GSK3␤, when overexpressed, to overcome the wnt signaling elicited by the expression of CKI⑀ (26).
In conclusion, we have demonstrated that the binding of axin to APC is important for its ability to regulate ␤-catenin. This binding event results in the hyper-phosphorylation of APC that is mediated in part by a kinase associated with the axin complex. We were unable to account for the axin-dependent phosphorylation of APC by considering GSK3␤ alone and therefore propose that at least one additional kinase is involved. Our data implicate CKI⑀ as a kinase that mediates the axin-dependent phosphorylation of APC and, therefore, its ability to down-regulate ␤-catenin.