Interaction of β-Catenin and TIF2/GRIP1 in Transcriptional Activation by the Androgen Receptor*

The multifunctional oncoprotein β-catenin interacts with the activation function-2 domain of androgen receptor (AR) to stimulate androgen receptor transcriptional activity, increase sensitivity, and broaden specificity of ligand interactions. β-Catenin interacts with androgen receptor in close proximity to the binding groove for P160 coactivators such as transcriptional intermediary factor-2 (TIF2)/glucocorticoid receptor interacting protein-1 (GRIP1). β-Catenin can also bind directly to TIF2/GRIP1. Both N- and C-terminal regions of β-catenin are needed for optimal interaction with TIF2/GRIP1. We show that distinct residues of β-catenin are responsible for both binding and functional interactions with androgen receptor and with TCF4, thus allowing the introduction of missense mutations that selectively affect these interactions. β-Catenin and TIF2/GRIP1 are each able to mediate binding between the other and androgen receptor in functional interactions that enhance ligand-dependent transcription. The data strongly imply that AR, β-catenin, and TIF2/GRIP1 bind in a three-way interaction that mediates transcription. Lastly, we observed that a β-catenin C-terminal peptide containing 229 amino acids can bind TIF2/GRIP1 and AR but has a profound dominant inhibitory effect on ligand-dependent transcription. We propose that β-catenin may play an integral role in formation of the androgen-receptor transcriptional complex.

AR 2 activity is essential to the growth and progression of prostate cancer at all phases of the disease. A member of the steroid hormone receptor family, AR binds androgen and drives the transcription of androgen-responsive genes in a variety of tissues and cell types. Like the prostatic epithelium from which it is derived, prostate cancer cells are dependent on the presence of androgen when it first presents. Androgen ablation therapy is the first line of treatment for metastatic prostate cancer (1). The vast majority of cases show a clinical response to androgen ablation therapy (2). However, this clinical response is transient, and when the cancer recurs, it does so in the presence of castrate levels of androgen. Androgen-independent prostate cancer paradoxically still depends on AR. For example 30 -40% of patients with advanced prostate cancer show amplification of the AR gene following androgen ablation therapy (3). Other specimens have shown point mutations within the AR ligand-binding domain that broaden ligand specificity to include anti-androgens and other steroid hormones (4). Point mutations in the N terminus may affect AR interaction with other proteins of the transcription complex (5,6).
Another mechanism that affects AR signaling in advanced prostate cancer was suggested when our laboratory found that 5% of primary specimens had mutations in the ␤-catenin gene (7). Further investigation into the interaction between AR and ␤-catenin showed ␤-catenin coimmunoprecipitates with AR, activates AR signaling in the presence of androgens, increases its affinity for other steroid hormones, and diminishes the effects of anti-androgens (8). AR affects ␤-catenin translocation to the nucleus (9, 10) and specific regions ␤-catenin and AR that interact to mediate AR activation and suppression of TCF4-mediated signaling (11)(12)(13)(14).
␤-Catenin is a multifunctional protein that mediates WNT signaling and is an important oncogene in a variety of cancers (15). Oncogenic activation of ␤-catenin involves disruption of phosphorylation sites at the N terminus of the protein that interrupt signals for recognition by ubiquitin ligases and protein degradation. Recently, it was shown that activation of the protease calpain is common in advanced prostate cancer and results in cleavage of ␤-catenin near the N terminus to remove the phosphorylation sites and produce a 75-kDa protein. This finding further supports the notion that ␤-catenin plays an important role in activation of AR signaling in advanced, androgen-independent prostate cancer.
Several studies have suggested interaction of components of the TCF4 transcription factors and androgen receptor transcription complex. A number of reports have described functional and physical interaction between androgen receptor and ␤-catenin (9,10,12), cross-talk between AR-mediated and TCF4-mediated transcription (11,16), and direct interaction of ␤-catenin, TIF2/GRIP1, and AR in the region of the AR C-terminal AF-2 domain (11). Moreover, ␤-catenin can also complex with TIF2/GRIP1 itself, suggesting a three-way interaction with AR and possible regulatory role for ␤-catenin in androgen action (13).
Here we explore the structural elements of ␤-catenin that determine interactions with TCF4 and AR and find that missense mutations of critical amino acids affect these two interactions differently. We proceed to show that either TIF2/GRIP1 or ␤-catenin can each mediate binding of the other to AR and that together they enhance AR-dependent transcriptional activation. The data have implications for regulation of AR action and suggest strategies for intervention of AR signaling in androgen-independent prostate cancer.
Cell Culture and Transfection-Unless otherwise stated in the figure legends, all transfections were performed in CV-1 cells. Briefly, 24 h before transfection, cells were plated in 24-well microtiter plates (Falcon) at a cell density of 2-4 ϫ 10 4 cells per well. Transfection was performed with Lipofectamine Plus reagent (Invitrogen) with 10 ng of receptor-containing plasmid, 100 ng of reporter, 10 ng of Renilla null luciferase, and the indicated amounts of coregulators. Unless otherwise stated, transfection was routinely done with the total transfected DNA brought up to 300 ng/well with pBSKϩ DNA. In experiments with different coregulator cDNA plasmids in different vectors, equimolar amounts of the corresponding plasmid vectors were cotransfected to control for artifacts of the vector DNA. After incubation for 16 h, the cells were washed and phenol red-free medium supplemented with 5% dextran charcoal-stripped fetal calf serum containing hormones or vehicle was added. The final concentration of vehicle ethanol was 0.1%. After a further 24 h, cells were lysed in 100 l/well 1ϫ passive lysis buffer (Promega, Madison, WI), and 30 l of the cell lysates was used to assay for luciferase activity with the dual-luciferase assay system from Promega. The data were then normalized for the cotransfected Renilla activity. All samples were in triplicate, and each experiment described was repeated at least two times.
Western Blotting-For detection of the expression of pcDNA3-␤-catenin and VP-16/␤-catenin mutants, SKBR3 cells were plated overnight in six-well plates. Cells were transiently cotransfected with 1 g each of wild-type pcDNA3-␤-catenin, VP-16/␤-catenin, or different mutants, together with 1 g of GFP vector with a Lipofectamine method according to the manufacturer's protocol. 48 h after transfection, cell lysates were prepared as described in the previous paragraph. Aliquots of 30 g of total proteins were resolved by SDS-PAGE, transferred onto a nitrocellulose membrane (Bio-Rad), and probed with ␤-catenin antibody (BD Biosciences Transduction Laboratories, 1:2000) and anti-GFP antibody (Clontech, 1:5000), followed by a secondary anti-mouse (␤-catenin) or anti-rabbit (GFP) immunoglobulinhorseradish peroxidase antibody (1:5000). Signal detection was performed with Supersignal West Pico chemiluminescent substrate (Pierce).
Statistical Analysis-Unless otherwise noted, values shown represent mean Ϯ S.D. The differences between groups were analyzed for statistical significance by the two-tailed Student's t test using the program GraphPad Prism software version 4.02 (GraphPad Software, San Diego, CA). p Ͻ 0.05 was considered significant. In figures the following symbols are used to represent p values: *, p Ͻ 0.05; **, p Ͻ 0.01; ***, p Ͻ 0.001; and ****, p Ͻ 0.0001.

␤-Catenin
Interactions with AR and TCF4-We showed previously that ligand-bound AR inhibited the transcriptional interaction between ␤-catenin and a TCF-responsive promoter (11). Because the interaction of ␤-catenin with both AR and TCF4 was blocked by inhibitor of ␤-catenin and TCF (17,18), we predicted that both AR and TCF4 would interact with the same region of ␤-catenin (11). To test this hypothesis we generated missense mutations in ␤-catenin to determine their effect on interactions with AR and TCF4. The location of these mutations is shown in the map in Fig. 1A. K312A and K435A disrupt each of the two lysine residue "charge buttons" that anchor the interaction with TCF proteins (19) and E-cadherin (20). The R386A mutation was chosen because Arg-386 is involved in binding to APC (21) and is also involved in the association with the Phe-21 residue of TCF4 (20). Arginine 469 and histidine 470 resides are required for the binding of TCF4/LEF1 by the formation of charge buttons with residues Lys-435, Asn-426, Lys-508, and Lys-312 (19,20).
In a mammalian two-hybrid assay, the AR ligand R1881 induced an interaction between wild-type VP-16/␤-catenin and GAL4/AR LBD as we previously showed (11). Missense mutant proteins VP-16/␤catenin(K312A), VP-16/␤-catenin(R386A), and the double mutant VP-16/␤-catenin(K312A/K386A) did not interact with GAL4/AR LBD in the presence of R1881. However, three other missense mutations, K435A, R469A, and H470A did not affect the interaction with GAL4/AR LBD to as great a degree ( Fig. 2A). These results suggest that Lys-312 and Arg-386 are required for the interaction between ␤-catenin and ligand-bound bound AR. To make sure that the variation of the binding activity of different VP-16/␤-catenin mutants was not due to the changes in protein expression levels, we did a parallel Western blot assay and showed that the expression levels of different constructs were not substantially different ( Fig. 2A).
The mammalian two-hybrid binding data correlated with transcriptional activation of an AR-responsive reporter construct. In a transient transactivation assay we examined the effect of different ␤-catenin mutant constructs on the ligand-dependent activation of an N-terminal truncated AR-(507-919). We used the truncated AR-(507-919) instead of a full-length construct, because the truncated receptor is essentially inactive in the absence of coactivators such as TIF2/GRIP1 or ␤-catenin, and therefore provides a qualitative read-out of the transcriptional interaction between AR and ␤-catenin constructs. As shown in Fig. 2B, wild-type ␤-catenin can effectively activate the truncated AR in the presence of R1881. In contrast, ␤-catenin point mutants K312A, R386A, and the double mutant K312A/R386A could not activate the receptor. The other three ␤-catenin point mutants, K435A, R469A, and H470A, could activate the receptor to the wild-type level. Western blot showed that the expression of different VP-16/␤-catenin mutants were at the comparable levels. These results are consistent with the notion binding of ␤-catenin to AR is required for coactivation as we have previously suggested.
The ␤-catenin missense mutations affecting the "charge buttons" had different effects on TCF4-mediated transactivation of the OT-luciferase reporter than they had on AR transactivation. The mutations that produced the greatest disruption of the interaction with AR affected TCF4mediated transcription to lesser degrees. ␤-Catenin point mutants K312A and R386A reduced OT-luciferase activities to 45 and 75% of wild-type activity, respectively. The K312A/R386A double mutant induced 32% of wild-type activity (Fig. 2C). In contrast, the mutant ␤-catenin constructs K435A, R469A, and H470A that retained the ability to interact with AR induced Ͻ20% of wild-type TCF4 transcriptional activity. These results with the TCF4-responsive reporter assays are similar to those reported by others in LEF-1-mediated transactivation assays (21).
AR and TCF4 can compete for binding of ␤-catenin, and each has an inhibitory effect on the transcriptional activity of the other in transfection experiments (11). ␤-Catenin(K435A) had lost nearly all inter-FIGURE 1. Schematic maps of two proteins used in this report as expression constructs or as fusion constructs with VP16 or GAL4. A, ␤-catenin map showing the location of the missense mutations inserted individually or in pairs to affect the charge buttons in the armadillo repeat region. In addition, the deletion constructs that are studied as fusion proteins with VP-16 or GAL4 are shown. B, TIF2/GRIP1 constructs are shown in alignment with the full-length protein sequence with the functional domains indicated. NOVEMBER 11, 2005 • VOLUME 280 • NUMBER 45 action with TCF4 and retained interaction with AR, whereas ␤-catenin(K312A) had the opposite effect. We therefore compared the effects of K312A and K435A with wild-type ␤-catenin on the ligand dose dependence of AR-mediated transcription. We observed that at 10 pM DHT, which did not activate AR alone, there was activation in the presence of either wild-type or K435A mutant ␤-catenin. This result underscores the finding that ␤-catenin can support AR transcriptional activity at castrate levels of DHT. ␤-Catenin(K435A) had a more potent effect on AR activation than wild type ␤-catenin, consistent with the notion that the absence of competitive binding to TCF family proteins decreased competition for ␤-catenin (Fig. 2D). ␤-Catenin(K312A) had very little activity in the same experiment and was not differentiable from the control plasmid at DHT concentrations Ͻ100 pM and not significantly different at all concentrations. We also examined the competition for ␤-catenin between AR and TCF4 by a physical association assay. 293T cells were cotransfected with both AR and TCF4 expression vectors, and cells were cultured with or without R1881 before isolation of lysates (Fig. 3). Equal amounts of cell lysates were incubated with agarose beads bound with either GST or GST/␤-catenin fusion protein. AR and TCF4 retained on glutathione beads were assayed by Western blotting. In agreement with the mammalian two-hybrid data in Fig. 2A, we observed ligand-dependent binding of AR to GST/␤-catenin and GST/␤-catenin(K435A), GST/␤catenin(R469A), and GST/␤-catenin(H470A) (Fig. 3). In contrast much less binding was observed with GST, GST/␤-catenin(K312A), GST/␤catenin(R386A), and GST/␤-catenin(K312A/R386A) (Fig. 3). In contrast, binding of TCF4 to the ␤-catenin fusion constructs was seen best in the absence of R1881. TCF4 bound strongly to GST/␤-catenin and less well to GST/␤-catenin(K312A), GST/␤-catenin(R386A), and GST/ ␤-catenin(R469A). We observed essentially no binding of TCF4 to GST/␤-catenin(K435A), GST/␤-catenin(H470A), and the double mutant GST/␤-catenin(K312A/R386A). The absence of binding to the double mutant in the GST-pull-down assay was in contrast to the result of the transactivation assay where OT-Luc transcriptional activity was 32% of wild type, which may have been due to interaction with endogenous wild type ␤-catenin expressed in CV-1 cells (Fig. 1C).

AR, TIF2, and ␤-Catenin Interaction
The presence of R1881 and exogenous AR expression markedly reduced availability of TCF4 to bind GST/␤-catenin fusion proteins. This observation is best explained by the ligand-dependent interaction of AR and TCF4, which decreased the availability of TCF4 for binding to the GST/␤-catenin fusion constructs (22). The amounts of GST fusion proteins used in the pull-down assays are shown at the bottom of Fig. 3 and were comparable among the different lysates. Thus differences in AR or TCF4 binding to GST/␤-catenin fusion proteins was not due to the differences in fusion proteins bound to the glutathione-agarose beads.
Interaction of ␤-Catenin with TIF2/GRIP1-We previously showed that ␤-catenin and TIF2/GRIP1 synergistically activated AR-mediated transcription. Li et al. (13) obtained complementary results and further showed that the AD2 domain of GRIP1 mediated a physical interaction with ␤-catenin and was necessary for synergism with ␤-catenin in acti-vating AR. However, the region of ␤-catenin required for the interaction with TIF2/GRIP1 has not been determined. We performed mammalian two-hybrid assays with GAL4 fusion proteins harboring various regions of ␤-catenin and VP-16/GRIP1. A GAL4/␤-catenin fusion protein bound strongly to the VP-16/GRIP1 fusion protein. GAL4 fusion proteins with either N-or C-terminal ␤-catenin deletions also bound to VP-16/GRIP1. However, a GAL4/␤-catenin fusion construct with the armadillo repeat alone (141-668) showed no detectable binding (Fig.  4A). To control for differences in the basal transcriptional activities of the GAL4/␤-catenin constructs we plotted the fold induction of reporter activity induced by the interactions with VP-16/GRIP1 relative to VP-16 alone. Transcription induced by interactions of VP-16/GRIP1 with GAL4/␤-catenin-(1-668) and GAL4/␤-catenin-(141-781) was equal or greater than transcription induction with GAL4/␤-catenin (Fig. 4A, inset). VP-16/GRIP1 did not induce any increase in reporter activity mediated by GAL4/␤-catenin-(141-668) that contained the armadillo repeat region alone. The data suggest that the primary sites of interaction between ␤-catenin and TIF2/GRIP1 resided in the C and N termini of ␤-catenin. We next confirmed the result of Li et al. (13) that the AD2 domain of GRIP1 is required for the direct interaction with and ␤-catenin (Fig. 4B). They had shown that the region of GRIP1 from 1122-1462 mediated nearly complete binding to ␤-catenin. We used a slightly shorter construct and found that the interaction with ␤-catenin was attenuated ϳ50% compared with wild-type GRIP1 (see constructs in Fig. 1B).
To confirm the physical interaction between ␤-catenin and TIF2/ GRIP1, we performed a set of GST-pull-down assays with GST fusion proteins harboring full-length ␤-catenin. Five single missense mutations and one double mutation were introduced into the GST fusion constructs to determine whether missense mutations that affected ␤-catenin interaction with either AR or TCF4 had any effect on interaction with TIF2/GRIP1. As shown in Fig. 4C, TIF2 bound very strongly to most of the GST/␤-catenin fusion proteins. GST/␤-catenin(H470A) that bound TIF2 weakly. Exposure of the filter to an anti-GST antibody showed that the amounts of GST fusion proteins used in each interaction were comparable (Fig. 4C). Because the armadillo repeat domain is not required for the interaction between ␤-catenin and TIF2, it is not AR, TIF2, and ␤-Catenin Interaction NOVEMBER 11, 2005 • VOLUME 280 • NUMBER 45 surprising that most of the point mutants in the armadillo repeat domain did not interfere with binding of ␤-catenin to TIF2 in a GSTpull-down assay. The effect of the missense H470A mutation on TIF2 binding is not as yet explained. Importantly, the regions of ␤-catenin that interacted with AR were not the same that interacted with TIF2/ GRIP1. The data were consistent with our proposed model that the three proteins interact by direct three-way binding during AR-mediated transcription.
We also performed a competition experiment in which binding in a mammalian two-hybrid interaction was challenged with wild-type or mutant ␤-catenin expression. If wild-type or mutant ␤-catenin competed with GAL4/␤-catenin for GRIP1 binding, the reporter activity was expected to decrease. As shown in Fig. 4D, cotransfection of five of the ␤-catenin mutants induced a significant decrease in the GAL4responsive reporter activity, suggesting that these ␤-catenin mutants competed with GAL4/␤-catenin for GRIP1 binding. ␤-Catenin(R469A) induced borderline significant decrease in reporter activity and ␤-catenin(H470A) was inactive in this assay. These results correlate with the results of the GST-pull-down assay in Fig. 4C. Thus, the sequence requirements for the binding of ␤-catenin to TIF2 differ from those of TCF/LEF-family transcription factors, APC, conductin, E-cadherin, and ␤-actin.
TIF2/GRIP1 and ␤-Catenin Cooperate in AR Binding and Transactivation-␤-Catenin, TIF2/GRIP1, and AR can form a complex in the presence of androgen and ␤-catenin and TIF2/GRIP1 can synergize as coactivators of AR-mediated transcription (11,13). Therefore changes in intracellular levels of both TIF2/GRIP1 and ␤-catenin can affect the activity of AR that is modulated by the presence of the other molecule. To demonstrate that either TIF2 or ␤-catenin could mediate the binding of the other to AR, we employed ␤-catenin missense mutations that were shown in Figs. 1 and 2 to affect either association with AR or interaction with TCF4. We carried out mammalian two-hybrid assays with VP-16 fusion proteins containing full-length, wild-type VP-16/␤-catenin, VP-16/␤-catenin(K312A), an AR interaction mutant, or VP-16/␤-catenin(K435A), a TCF4 interaction mutant. We found that wild-type VP-16/␤-catenin, VP-16/␤-catenin(K435A), and TIF2 all induced reporter activity, indicating these proteins interacted with GAL4/AR LBD in a ligand-dependent manner, whereas VP-16/␤catenin(K312A) did not interact with GAL4/AR LBD due to the disruption of interaction with AR LBD by the K312A mutation (Fig. 5A, top   panel). Cotransfection of TIF2 with either wild-type VP-16/␤-catenin or VP-16/␤-catenin(K435A), enhanced reporter activity to a greater degree than with either coactivator alone. Even though VP-16/␤catenin(K312A) is deficient in AR binding, TIF2 plus VP-16/␤catenin(K312A) induced a greater degree of GAL4 reporter induction than TIF2 alone (Fig. 5A, top panel). TIF2m123 is a mutant of TIF2 deficient in AR interaction (24). TIF2m123 and VP-16/␤catenin(K312A) together had no effect on GAL4/AR LBD, because both have inactivation of AR binding. TIF2m123 reduced the binding interactions of both VP-16/␤-catenin and VP-16/␤-catenin(K435A) by approximately one-third to one-half, suggesting that binding of TIF2 to the AF-2 groove is critical for the synergism with VP-16/␤-catenin for activation of GAL4/AR LBD. In each case TIF2 enhanced the interaction of the VP-16/␤-catenin fusion construct and TIF2m123 inhibited the interaction (Fig. 5A, top panel).
An assay for AR-dependent transcription was also done to determine if the binding interactions seen in the top of Fig. 5A correlated with effects on AR activity. We performed parallel reporter assays with AR-(507-919) instead of GAL4/AR LBD and used the MMTV-Luc reporter for readout of AR transcriptional activation. Whereas the ␤-catenin constructs had a stronger binding interaction with AR than TIF2, there was much less difference between VP-16/␤-catenin and VP-16/␤-catenin(K435A) and TIF2 in an AR transcriptional interaction (Fig. 5A, bottom panel). The TIF2m123 construct reduced AR-mediated transcriptional activation by VP-16/␤-catenin by 68%, and by VP-16/␤catenin(K435A), but did not abrogate either interaction. This suggests that TIF2m123 had a dominant negative effect by complexing with other members of the transcriptional complex or by interfering with the AR-VP-16/␤-catenin complex. Because TIF2m123 has an intact AD2 domain and thus binds VP-16/␤-catenin, TIF2m123 likely competed with AR for ␤-catenin binding (Fig. 5A, bottom panel). However, TIF2 synergized with both VP-16/␤-catenin and VP-16/␤-catenin(Lys-435) to enhance AR-dependent transcription. TIF2m123 and ␤-catenin(K312A) together had no effect on MMTV reporter activity due to the disruption of AR interactions of both proteins. On the other hand, TIF2 had a significant synergistic effect with VP-16/␤-catenin(K435A). In contrast, VP-16/␤-catenin(K312A) had very little effect on TIF2 interaction with AR. Even though TIF2 was able to mediate binding between GAL4/AR LBD and VP-16␤-catenin(K312A) (Fig. 5A), the effect on AR-mediated transcription was much less, suggesting further that a three-way interaction of TIF2, VP-16/␤-catenin, and AR-(507-919) was important for maximal enhancement of AR transcriptional activity. The data are consistent with an interaction model that requires TIF2/GRIP1 occupancy of the AF-2 that is enhanced by the presence of ␤-catenin bound either to AR and TIF2/GRIP1 or to TIF2/GRIP1 alone.
The interaction of TIF2 and ␤-catenin was further examined with a coactivator mutant subGRIP1, in which the three NR boxes have been changed to CoRNR nuclear receptor corepressor binding motifs (24). The subGRIP1 construct alone showed no binding interaction with GAL4/AR LBD and induced 90% inhibition of binding between GAL4/AR LBD and VP-16/␤-catenin demonstrating that subGRIP1 was able to block binding by complexing with ␤-catenin, because sub-GRIP1 cannot bind to AR LBD (24) (Fig. 5B, top panel). In contrast subGRIP1 reduced by half, but did not abrogate, the transcriptional interaction between AR-(507-919) and VP-16/␤-catenin. subGRIP1 had no effect on AR-dependent transcription but was not only unable to synergize the interaction of AR and VP-16/␤-catenin but inhibited transcription either by competing for ␤-catenin or by inhibiting AR LBD directly (Fig. 5B, bottom panel). This result was also consistent with the notion that optimal AR-mediated transcriptional activity required binding of both ␤-catenin and TIF2/GRIP1 to the AR.
Both TIF2m123 and subGRIP1 retained the ability to bind ␤-catenin as shown by a modified mammalian two-hybrid assay in which we cotransfected cells with GAL4/␤-catenin, the FR-Luc reporter plasmid, and the TIF2/GRIP1 mutants. As shown in Fig. 5C, wild-type GRIP1, subGRIP1, and TIF2m123 induced significant increases in reporter activity as compared with control vector, indicating that both subGRIP1 and TIF2m123 bound ␤-catenin.
These data thus far suggested that ␤-catenin and TIF2/GRIP1 could each mediate the binding of the other to a complex with AR, but that interaction of both with AR was required for a synergistic effect on AR-dependent transcription. To explore this interaction further we examined the activity of a GRIP1 construct that did not interact with ␤-catenin (13). We performed a modified mammalian two-hybrid assay with GAL4/AR LBD, VP-16/␤-catenin, and full-length GRIP1 or GRIP1-(5-1121) that deleted the second activating domain that is required for interaction with ␤-catenin (see Fig. 1B) (13). When cells were cotransfected with VP-16/␤-catenin and wild-type full-length GRIP1, the reporter activity was 8.5-and 11.4-fold higher than the activities induced individually by VP-16/␤-catenin and TIF2, respectively, indicating a strong synergistic binding of these two factors to agonist bound AR (Fig. 6A). When cells were cotransfected with VP-16/␤-catenin and GRIP1-(5-1121) the reporter activity was 4.3-and 2.5-fold of that seen with transfection of the individual VP-16/␤-catenin and GRIP1-(5-1121) constructs, respectively (Fig. 6A).
Loss of the second GRIP1 activation domain diminished the coactivation of AR (Fig. 6B). In this assay VP16/␤-catenin enhanced the effects of both GRIP1 and GRIP1-(5-1121) to nearly identical degrees (1.17 to 1. 18-fold) showing that the attenuation of AR binding by loss of the AD2 domain did not affect interaction with AR-mediated transactivation.
When the binding assay of Fig. 6A was repeated with the mutant VP-16/␤-catenin(K312A) that abrogated the direct ␤-catenin-AR interaction, in place of the wild type ␤-catenin construct, we observed a result qualitatively very similar (Fig. 6C). This implied that binding of VP-16/␤-catenin(K312A) to the transcription complex was mediated entirely by either GRIP1 or GRIP1-(5-1121). The data thus far suggest that TIF2/GRIP1 interaction directly with AR can facilitate the effect of ␤-catenin on AR-mediated transactivation, and that either TIF2/GRIP1 or ␤-catenin mediated binding between AR and the other protein.
To further characterize the interactions between AR, TIF2/GRIP1, and ␤-catenin in transcription we exploited four AR missense mutations, E8983A, M894A, E897V, and Q902R, known to affect the interaction of the AR AF2 domain with either TIF2/GRIP1 or ␤-catenin (11). AR(E893A) was generated in the laboratory and found to selectively disrupt AR binding to ␤-catenin and attenuate AR interaction with TIF2/GRIP1. AR(M894A) and AR(E897V), also laboratory-derived, and AR(Q902R), a prostate cancer mutation (25), all have substantial defects in TIF2/GRIP1 binding and varying degrees of altered ␤-catenin interactions. A Q902K mutation was found to be associated with partial androgen insensitivity in one family and was found to decrease sensitivity to R1881 and to decrease interaction with TIF2 (25). It is not clear how the Q902R mutation was selected in a patient with advanced prostate cancer, however, the patient with the Q902R mutation experienced disease progression just 1 month after orchiectomy for a recurrent Gleason score 8 stage C tumor (25). It is therefore possible that the Q902R mutation was not selected by prostate cancer treatment. The relative activities of the AR LBD mutants with ␤-catenin and TIF2/GRIP1 in mammalian two-hybrid and AR transcriptional assays are shown graphically in Fig. 7A based on data we obtained previously (11).
In mammalian two-hybrid assays GRIP1 alone did not significantly affect reporter activation by GAL4/AR LBD(M894A), E897V, or Q902R. VP-16/␤-catenin, on the other hand, interacted with these three constructs to varying degrees (Fig. 7B). VP16/␤-catenin was able to engage GRIP1 in the transcriptional complex with these AR-interacting mutants as indicated by the augmentation of the VP-16/␤-catenin interaction with the addition of GRIP1 (Fig. 7B). The enhancing effect of GRIP1 on the interaction of ␤-catenin with the GAL4/AR LBD mutant constructs was 16-fold for GAL4/AR LBD(M894A), 47-fold for GAL4/AR LBD(E897V), and 19-fold for GAL4/AR LBD(Q902R). VP-16/␤-catenin did not demonstrate an interaction with GAL4/AR LBD(E893A) and had no effect on the interaction of GAL4/AR LBD(E893A) with GRIP1 (Fig. 7B). The data show that ␤-catenin can recruit TIF2 to AR and overcome reduced affinity of TIF2 for the AF-2 binding groove. In the case of the mutant AR LBD(E893A) TIF2 had no appreciable effect on the recruitment of ␤-catenin to AR. Because of the overall attenuation of activity by the E893A mutation, a cooperative effect of ␤-catenin and TIF2/GRIP1 may not be possible with this construct.
In the case of the TIF2/GRIP1 interaction mutant AR-(507-919)(M894A) GRIP1 induced activation at a level 13% of wild type AR-(507-919) and VP-16/␤-catenin at a level of 45% (Fig. 7C). There was no cooperation of GRIP1 and VP-16/␤-catenin with AR-(507-919)(M894A). This implies that, despite the cooperative effect observed in the binding assay, the M894A mutation decreased affinity of the binding groove for ␤-catenin and TIF2/GRIP1 to a degree that prevented a cooperative effect on AR-mediated transcription. The MMTV-luciferase assays with the AR-(507-919)(E893A) mutant suggest that there was insufficient binding of TIF2/GRIP1 to recruit VP-16/ ␤-catenin to a three-way complex and enhance AR-mediated transcription. On the other hand, the AR mutants E897V and Q902R had marked attenuation of transcriptional activation by GRIP1, but with both there was cooperative interaction of VP-16/␤-catenin and GRIP1 in AR-mediated transcription (Fig. 7C). In the instance of these two mutations the data suggest that there was sufficient affinity of the AF-2 groove to allow VP-16/␤-catenin to recruit and retain TIF2/GRIP1 in a binding conformation with AR.
To confirm further the notion of a three-way interaction between AR, ␤-catenin, and TIF2/GRIP1, the GAL4/AR LBD(E897V) mutant that did not bind TIF2/GRIP1 but had the most robust interaction with a GRIP1-VP-16/␤-catenin complex was used in mammalian two-hybrid experiments that tested the effects of ␤-catenin mutants with compromised AR and TIF2/GRIP1 binding. As expected GRIP1 and GRIP1-(5-1121) had minimal interaction with GAL4/AR LBD(E897V) (Fig. 7D). VP-16/␤-catenin bound to the mutant fusion protein receptor in the presence of hormone and recruited both GRIP1 and GRIP1-(5-1121) to enhance binding and AR-mediated transcription (Fig. 7, D and E). The mutant VP-16/␤-catenin(K312A) that has minimal interaction with wild-type AR did not bind to GAL4/AR LBD(E897V). VP-16/␤catenin(K312A) had no interaction with GRIP1-(5-1121), and a low degree of interaction with GAL4/AR LBD(E897V) when GRIP1 was added. The weak effect of GRIP1 and VP-16/␤-catenin(K312A) was most likely due to the residual binding of GRIP1 to the mutant AR constructs. These data further show that the binding of VP-16/␤-catenin to AR was critical for the enhancement of the interaction with mutant GRIP1.
When AR was not able to interact efficiently with both ␤-catenin and TIF2/GRIP1 then either of the two cofactors could compete with AR for binding of the other and, paradoxically, inhibit AR-mediated transcription. This is demonstrated in Fig. 7F where VP-16/␤-catenin was able to compete GRIP1 off AR and inhibit transcription from AR-(507-919) (E893) in a dose-dependent manner. In a similar manner, GRIP1, whose binding to AR is compromised by the M894A mutation, was able to compete for VP-16/␤-catenin and reduce AR-mediated transcription in a dose-dependent manner.
␤-Catenin Domains That Interact with TIF2/GRIP1 and AR-We wanted to determine the regions of ␤-catenin required for the two-way interaction with AR and TIF2/GRIP1. Loss of the C-terminal domain of ␤-catenin reduces the interaction with AR by ϳ40%, and deletion of armadillo repeats 5 and 6 completely disrupts binding of ␤-catenin to AR (12). We speculate that optimal interaction with AR requires both the charged surface of the armadillo repeat helical backbone and the C-terminal domain similar to the interaction of ␤-catenin with TATA binding protein (27). We performed mammalian two-hybrid assays with GAL4/AR LBD, GRIP1, and VP-16 fusion proteins harboring the full length, N-terminal deleted ␤-catenin(⌬15-552), or C-terminal deleted ␤-catenin (1-574) (constructs shown in Fig. 1A). The two ␤-catenin truncation mutants do not interact with AR in the presence of androgen (11). Both GRIP1 and ␤-catenin(S33A) interacted with  GAL4/AR LBD to induce GAL4-dependent transcription (Fig. 8A). Together, VP-16/␤-catenin(S33A) and GRIP1 synergized to interact with GAL4/AR in a ligand-dependent manner. When cells were cotransfected with N-or C-terminal ␤-catenin deletion mutants together with GRIP1 a much stronger reporter activity was seen, suggesting that GRIP1 was able to recruit ␤-catenin to the transcriptional complex on the androgen-responsive promoter. The reporter activity in the presence of the N-terminal deletion mutant VP-16/␤catenin(S33A)(⌬15-552) and GRIP1 was indistinguishable from the effect of VP-16/␤-catenin(S33A) and GRIP1. On the other hand, the effect of VP-16/␤-catenin(S33A)-(1-574) and GRIP1 was much less, consistent with the notion that the C terminus of ␤-catenin is important for the physical interaction with GRIP1.
The formation of a trimeric complex between AR, TIF2/GRIP1, and ␤-catenin was strongly suggested by the mammalian two-hybrid data, particularly by the experiments where binding between any two of the constituents was inactivated by missense mutation. Further confirmation of the formation of a trimer was obtained by coimmunoprecipitation of endogenous proteins from LNCaP cell nuclear extracts from cells exposed or not to R1881. First we demonstrated that three monoclonal antibodies against AR, TIF2, and ␤-catenin could be used simultaneously to detect their respective antigens (Fig. 9A). We saw no nonspecific bands, even with long Western blot exposures. Because the level of AR expression in LNCaP cells is very high, we used 1:5000 dilutions of AR antibodies in all the Western blot analysis. In the Western blot of ␤-catenin, we detected the 75-kDa ␤-catenin cleavage product induced by calpain as previously described (28). The relative densities of the input proteins are shown in the top panel of Fig. 9B.
After immunoprecipitated with polyvalent AR antiserum both TIF2 and ␤-catenin could be detected by Western blot independent of prior exposure to R1881 (Fig. 9B). This is in agreement with the fact that p160 coactivators can interact directly with the AR N terminus in a ligandindependent manner via a conserved glutamine-rich region (29). We have shown that ␤-catenin binding to AR is ligand-dependent, but we cannot say whether the ligand-independent coimmunoprecipitation of AR with ␤-catenin is mediated by an intervening TIF2/GRIP1 molecule. Densitometric analysis showed that more TIF2 and ␤-catenin were pulled down in the presence of R1881 than in its absence. Similarly, AR and ␤-catenin were pulled down by TIF2 antibody, and there was no substantial difference between the R1881 treated and untreated samples. When ␤-catenin antibody was used to isolate the immunoprecipitate, AR and TIF2 could be pulled down and more AR was detected in the presence of R1881. However, the amount of TIF2/GRIP1 pulled down by ␤-catenin antibody was comparable in R1881-treated and untreated groups, suggesting the association of TIF2 and ␤-catenin is independent of ligand binding to AR.
The formation of a three-way physical complex between AR, ␤-catenin, and TIF2/GRIP1 was also examined with exogenous mutant proteins expressed in 293T cells. Here we used AR, GRIP1, or GRIP1-(5-1121) that has attenuated ␤-catenin interaction, and VP-16/␤-catenin or VP-16/␤-catenin(K312A) that has reduced AR interaction. Both fulllength, wild-type GRIP1, and truncated GRIP1-(5-1121) could be pulled down by AR antibody in the presence of R1881 (Fig. 10A). When cells were cotransfected with both GRIP1 and VP16/␤-catenin, much more GRIP1 could be pulled down (Fig. 10A, lanes 1 and 3). This observation is in agreement with our mammalian two-hybrid and transactivation assay results showing that GRIP1 and VP-16/␤-catenin could synergistically coactivate AR-mediated reporter activity (Fig. 5A). More interestingly, when VP-16/␤-catenin(K312A), a mutant that cannot bind AR, was used, similar results were obtained (Fig. 10A, lanes 1 and  4), suggesting that ␤-catenin could be recruited to the GRIP1-AR complex by association with GRIP1 independently of AR. When GRIP1-(5-1121) was used, VP-16/␤-catenin and VP-16/␤-catenin(K312A) could still enhance the association of GRIP1 with AR to a limited degree (Fig.  10A, lanes 2, 5, and 6). This result is also in agreement with our mammalian two-hybrid and transactivation assays (Fig. 6C) and suggests that even though AD2 of TIF2/GRIP1 is the major region required for the direct association with ␤-catenin other regions in the N terminus of TIF2/GRIP1 may also be involved in the interaction (13).
The VP-16/␤-catenin(⌬15-552) construct demonstrated strong inhibition of AR-dependent transcription at low levels of transfected plasmid (Fig. 11A). Moreover, there was a direct effect on AR independent of the presence of exogenous GRIP1, because a transcriptionally competent VP-16/AR-(507-919) fusion protein was inhibited by low levels of VP-16/␤-catenin(⌬15-552) (Fig. 11B). Neither exogenous excess GRIP1 nor ␤-catenin substantially reversed the inhibition imposed by VP-16/␤-catenin(⌬15-552), although both demonstrated dose-dependent effects (Fig. 11, C and D). The C-terminal domain of ␤-catenin is responsible for interaction with CBP and p300 (30) and therefore could potentially interfere with formation of an intact transcriptional complex by binding to one or the other of these components of the transcriptional complex. In exploratory experiments we pursued this notion. However, neither excess p300 nor CBP were able to abrogate the inhibitory effect of VP-16/␤-catenin(⌬15-552) (Fig. 11, E and F).

DISCUSSION
Our findings demonstrate that ␤-catenin interacts with AR via different amino acid residues than it interacts with TCF4, but we confirm that the two interactions compete for ␤-catenin binding. Based on our results and others', there appears to be a complex interaction between TCF4 and AR transcription factors. Indeed, the following binding interactions have been demonstrated: AR-␤-catenin (8 -12); AR-TCF4 (22); ␤-catenin-TIF2 (11,13) in addition to the well established interactions of AR with p160 coactivators like TIF2 and the obligate interaction of ␤-catenin and TCF4 in the initiation of TCF-4-dependent transcription (15). Our data also show, through targeted mutational analysis, that the three-way complex of AR, ␤-catenin, and TIF2 supports a higher level of transcriptional activation than the complex of AR with either of the two alone. Furthermore, either the two coactivators can tether the other to AR resulting, in most instances, in enhanced transcriptional activation.
The physical interaction of AR and peptide binding motifs of nuclear receptor coactivators is mediated by a shallow cleft in the ligand-bound AR (31). This implies that the binding between AR and TIF2/GRIP1 may need to be secured by additional molecular interactions. The interaction of AR and TIF2/GRIP1 may be mediated in part by interactions of TIF2/GRIP1 domains outside the NR box with regions of AR outside the AF-2 domain. We propose that ␤-catenin has the potential to mediate this interaction by binding simultaneously to AR and TIF2/GRIP1. Moreover, ␤-catenin by itself can activate AR-mediated transcription.
These findings have potential significance for AR-mediated signaling in prostate cancer progression. In the rare cancers with ␤-catenin mutations there is a direct effect on AR action early in the disease (7,32). In other instances APC may undergo gene methylation or mutation, resulting in ␤-catenin activation (33). Other components of the ␤-catenin regulatory pathway have also been disrupted in prostate cancer models (33). Inherited APC mutations may predispose to prostate cancer (34). And ␤-catenin is cleaved by calpain activation in advanced prostate cancer and may be activated as a result (28). Interestingly, reduced ␤-catenin expression in localized prostate cancer is associated with a shorter interval to biochemical relapse and may indicate that alternative pathways for AR activation result in more aggressive disease (35).
Because androgen receptor signaling is essential through all phases of prostate cancer development and progression, one might expect that the molecular pathology of prostate cancer might include genetic alterations of AR coactivators. Thus far this has not been demonstrated, although some have found that expression of p160 steroid receptor coactivator family members is enhanced in advanced prostate cancer. Overexpression of coactivators is expected to enhance AR action, broaden ligand specificity, and mediate AR agonism by the anti-androgen hydroxyflutamide (36). However, in other cohorts different coactivators have been implicated (37). Moreover, to our knowledge genetic change of p160 coactivator genes has not been found. ␤-Catenin is therefore one of the few AR coactivators that is implicated in prostate cancer pathogenesis both by mutation and by post-translational modification (7,28,32).
Lastly, we encountered a surprising inhibition of AR-mediated transcription by the C-terminal domain of ␤-catenin. The precise mechanism of this inhibitory effect was not elucidated, but the data imply that the C-terminal domain fragment of ␤-catenin has a profound inhibitory interaction with AR. The inhibitory effect was not reversed to a significant degree by supplementation with either CBP or p300, two components of the transcriptional complex that bind to the ␤-catenin C terminus. If the mechanism of inhibition is due to the binding of the ␤-catenin C terminus to AR, as implied by the data in Fig. 8, the binding AR, TIF2, and ␤-Catenin Interaction NOVEMBER 11, 2005 • VOLUME 280 • NUMBER 45 must have quite high affinity, as it was incompletely competed by excess full-length ␤-catenin. The competition by full-length ␤-catenin did reflect a dose response. On the other hand, the effect of the ␤-catenin C terminus on AR-mediated transcription was only partially reversed by TIF2/GRIP1 in a manner that suggested a noncompetitive interaction (Fig. 9, A and C). Further experiments will identify the minimal inhibitory region of ␤-catenin and will elucidate the mechanism of this profound AR inhibition. The mechanism of this inhibition may have relevance in instructing the design of a new class of anti-androgens.