Coregulator Control of Androgen Receptor Action by a Novel Nuclear Receptor-binding Motif*

Background: The interaction surface of coactivators and the androgen receptor (AR) is an important target for prostate cancer therapeutics. Results: A new interface formed by binding of the sequence (GARRPR) and the allosteric pocket (BF-3) of the AR has been identified. Conclusion: GARRPR binding modulates AR activity. Significance: The GARRPR/BF-3 interaction is a novel regulatory hub for AR activity. The androgen receptor (AR) is a ligand-activated transcription factor that is essential for prostate cancer development. It is activated by androgens through its ligand-binding domain (LBD), which consists predominantly of 11 α-helices. Upon ligand binding, the last helix is reorganized to an agonist conformation termed activator function-2 (AF-2) for coactivator binding. Several coactivators bind to the AF-2 pocket through conserved LXXLL or FXXLF sequences to enhance the activity of the receptor. Recently, a small compound-binding surface adjacent to AF-2 has been identified as an allosteric modulator of the AF-2 activity and is termed binding function-3 (BF-3). However, the role of BF-3 in vivo is currently unknown, and little is understood about what proteins can bind to it. Here we demonstrate that a duplicated GARRPR motif at the N terminus of the cochaperone Bag-1L functions through the BF-3 pocket. These findings are supported by the fact that a selective BF-3 inhibitor or mutations within the BF-3 pocket abolish the interaction between the GARRPR motif(s) and the BF-3. Conversely, amino acid exchanges in the two GARRPR motifs of Bag-1L can impair the interaction between Bag-1L and AR without altering the ability of Bag-1L to bind to chromatin. Furthermore, the mutant Bag-1L increases androgen-dependent activation of a subset of AR targets in a genome-wide transcriptome analysis, demonstrating a repressive function of the GARRPR/BF-3 interaction. We have therefore identified GARRPR as a novel BF-3 regulatory sequence important for fine-tuning the activity of the AR.

tion function-2 (AF-2) for coactivator binding. Several coactivators bind to this hormone-induced conformation via a conserved LXXLL sequence to enhance the transcriptional activity of the receptor (5)(6)(7). In addition, the AF-2 binds strongly to a 23 FQNLF 27 motif at its N terminus (8 -11) to generate an N/Cterminal intramolecular interaction for alternative coactivator binding (4,12). More recently, a topologically distinct surface adjacent to but different from AF-2 has been identified in a screen of chemical libraries for small coactivator inhibitors. This surface has been proposed as an additional protein binding site for AR regulatory proteins and has been termed binding function-3 (BF-3) (13). However, not much is known about its in vivo function except that it may be a site for protein-protein interaction.
One of the factors known to modulate the activity of the AR is the cochaperone Bag-1L. This cochaperone belongs to a family of polypeptides translated from the same mRNA by a leaky scanning mechanism generating four isoforms in humans (Bag-1L, Bag-1M, Bag-1, and Bag-1S) (14,15). These isoforms differ in their N-terminal sequences but have a conserved C-terminal domain (otherwise known as the BAG domain) with which they bind the molecular chaperone Hsp70/Hsc70 as nucleotide exchange factors (16,17). Besides Hsp70/Hsc70 binding, the Bag-1 proteins interact with and regulate the activities of several cellular proteins ranging from Bcl-2 and steroid receptors to growth factor receptors (18). Bag-1L, the largest member of this family, possesses an N-terminal nuclear localization sequence and is therefore localized to the nucleus, where it functions as a transcriptional regulator (15,19,20).
Immunohistological studies have shown that Bag-1L is expressed in the basal cells of benign prostate tissue, but the site of its expression is changed in prostate carcinoma to the secretory epithelium, where the AR is expressed (21). The AR and Bag-1L are therefore expressed in the same cell type in prostate carcinoma, suggesting a mechanistic link between the two proteins in this disease. Furthermore, overexpression of Bag-1L potentiates the transactivation action of the AR, and domain mapping experiments have shown that the C-terminal region of Bag-1L contributes to the enhancement of the receptor activity (22,23). Although deletion or single amino acid substitutions within the C terminus abolish the ability of Bag-1L to enhance AR activity and to bind Hsp70/Hsc70 (24), our earlier studies suggest that sequences other than those in the C-terminal region may contribute to the regulation of AR activity (21).
In this report, we have identified a novel hexapeptide repeat sequence, GARRPR, in the N-terminal region of Bag-1L as an additional interaction site for AR and have shown that it functions through the BF-3 pocket in the AR-LBD. We have further demonstrated that mutations in the hexapeptide motifs impair binding of Bag-1L to the AR and differentially alter the androgen response of the receptor. The GARRPR sequence therefore defines a novel regulatory motif for modulating the activity of the AR through the BF-3 domain.
Proximity Ligation Assay (PLA)-LNCaP cells were starved for 72 h and then treated with vehicle or 10 nM dihydrotestosterone (DHT) for 4 h. Cells were fixed and permeabilized using 4% paraformaldehyde and 0.2% Triton X-100. LNCaP cells were then blocked with 10% FBS and incubated with anti-AR (N-20, Santa Cruz Biotechnology), anti-Bag-1 (CC9E8, Santa Cruz Biotechnology), or anti-IgG (mouse IgG antibodies (BD Biosciences) and rabbit serum (Sigma-Aldrich)). Protein-protein interactions were analyzed using Duolink-based in situ PLA (Sigma-Aldrich), according to the manufacturer's instructions. Signals were quantified using a confocal Leica SPE microscope and the Duolink ImageTool. 300 cells were analyzed per treatment.
Gene Expression and Microarray Analysis-LNCaP cells stably expressing poZC vector or poZC-Bag-1L wild-type or mutant were starved for 72 h and then treated with vehicle or 10 nM DHT for 4 h. Subsequently, total RNA was isolated as described above. Biological triplicate RNAs were hybridized to a human U133 Plus 2.0 expression array (Affymetrix) at the Dana-Farber Cancer Institute Microarray Core Facility. Gene expression data were normalized and log-scaled using the RMA algorithm (29) and the RefSeq probe definition (30). For genes with more than one probe set, the probe set with the highest average expression was used for further analysis. Differentially expressed genes were determined by the linear models for microarray data algorithm (31). The full sequences of primers used in the RT-PCR analysis are shown in supplemental Table 2.
GST Pull-down Experiments-Expression of GST-tagged proteins and GST pull-down experiments were performed as described previously (32,33).
Chromatin Immunoprecipitation (ChIP), ChIP-re-ChIP, and Co-IP-LNCaP cells were starved for 72 h and then treated with DHT or vehicle for 4 h. Thereafter, ChIP was performed as described previously (34,35) using the following antibodies: anti-AR (N20, Santa Cruz Biotechnology), Bag-1L (36), and HA-tag (ab9110, Abcam). As a control, rabbit IgG (Santa Cruz Biotechnology) was employed. ChIP DNA was purified using the PCR purification kit (Qiagen) and quantified using the Quant-iT dsDNA HS assay kit (Invitrogen).
ChIP-re-ChIP was carried out as described for ChIP with the following alterations. After the initial round of overnight IP (with AR, Bag-1L, or IgG antibody), the magnetic beads were incubated in 10 mM DTT for 30 min at 37°C. Subsequently, the DTT was quenched with radioimmunoprecipitation assay buffer (10 mM Tris-HCl, pH 7.5, 1 mM EDTA, pH 8.0, 0.1% SDS, 1% Triton X-100, 0.1% NaDOC, and protease inhibitors) with 0.3 M NaCl. Free chromatin complexes were then subjected to a second IP using different antibodies of choice. All primers used for the ChIP and ChIP-re-ChIP experiments are listed in supplemental Table 3.
Endogenous co-IP was carried out as described previously (37) with the following alterations. After nuclear lysis and overnight incubation with anti-Bag-1L or HA-tag antibody, protein complexes were precipitated with magnetic beads. Beads were then washed five times for 10 min each, and complexes were eluted by boiling in Laemmli sample buffer (Bio-Rad). LGSR; Bag-1L(61-80), FITC-RGAAAGARRPRMKKKTRRRS) or SRC-2 peptide (FITC-HDSKGQTKLLQLLTTKSDQM) was incubated with serially diluted AR-LBD (4 to 0.4 M) in binding buffer (50 mM NaPO 4 , pH 6.5, 50 mM KCl, 1 mM DTT) and 10 M DHT. The samples were then equilibrated for 45 min at ambient temperature. For the competition experiments, 400 to 0.4 nM Bag-1L core (GARRPR) or SRC-2 peptides (HDSKGQT-KLLQLLTTKSDQM) were incubated with 4 M AR-LBD and 5 nM fluorescent Bag-1 or SRC-2 peptide. Binding was measured using polarization (excitation ϭ 485 nm, emission ϭ 530 nm) on a Synergy H1 hybrid reader (Biotek). Synthesis imidazole was synthesized in house essentially as described (38) with minor modifications in the addition of 1H-benzimidazole-2-thiol to 2-(2-bromoethoxy)-1,3-dimethylbenzene in the presence of potassium iodide and carbonate. The latter reactant was synthesized by allowing 1,2-dibromoethane to react with a commercially available 2,6-dimethylphenol in the presence of potassium carbonate as a base at 60°C, as described previously (39).
Molecular Modeling-The GARRPR peptide and CPD49 were docked into the BF-3 pocket of AR using DOCK 3.5.54 (40,41). Six AR complex structures (Protein Data Bank codes 2PIT, 2PIU, 2PIV, 2PIW, 2PIX, and 2PKL) (13) were used individually in order to capture receptor conformational flexibility. For each docking experiment, the protein in one complex structure was used as the rigid receptor, with all of the waters removed. The original ligand in the corresponding complex crystal structure was used to generate the matching spheres. Multiple conformations of the peptide and CPD49 were pre-generated and placed in the binding pocket guided by the matching spheres. Ligand poses were scored through a grid-based method, calculating van der Waals forces, electrostatic interactions, and the ligand desolvation energy. The highest-ranking poses from each of the six docking experiments were combined and visually examined in the context of their respective receptor environment. The best pose was chosen based on its polar and non-polar interactions between the ligands and the receptor, as well as the shape complementarity with the binding pocket.

RESULTS
Bag-1L Is Important for AR Action-To study the significance of the regulation of AR action by Bag-1L, we first determined whether these two proteins interact in LNCaP prostate cancer cells using co-immunoprecipitation. We were able to detect AR by Western blotting when Bag-1L was immunoprecipitated, both in the absence and presence of the androgen DHT (Fig.  1A). We could additionally show that Bag-1 and AR interact in LNCaP cells by an in situ proximity ligation assay (43). This assay uses oligonucleotide-generated high intensity fluorescent signals to assess in situ whether two proteins are located in close proximity to one another. The ligation assay performed with Bag-1 (which recognizes all Bag-1 isoforms) and AR antibodies showed positive punctuated perinuclear signals in the LNCaP cells in the absence and presence of DHT. However, the number of fluorescence signals per nucleus was slightly enhanced in the presence of the hormone (Fig. 1B). Little or no fluorescent signal was observed when AR or Bag-1 antibodies were used together with an isotype IgG control antibody.
To determine the contribution of Bag-1L to AR response, we next down-regulated the expression of Bag-1 in LNCaP cells using RNA interference and determined the expression of two well studied, androgen-regulated genes, PSA and FKBP5 (44). siRNA transfection significantly reduced the level of expression of the Bag-1 isoforms ( Fig. 2A). Concomitant with this reduction was a down-regulation of the androgen-mediated transcription of the two genes, which was highly significant for FKBP5 (Fig. 2B). This shows that Bag-1 proteins contribute to an efficient, androgen-mediated transcription.
Next we investigated whether Bag-1L is recruited to AR enhancer regions of the PSA and FKPB5 genes using a Bag-1Lspecific antibody (36). We were able to identify solo occupancy of either AR (Fig. 2C) or Bag-1L (Fig. 2D) using ChIP. Further-more, we could demonstrate a partial co-occupancy of Bag-1L and AR at these enhancer elements, shown by a comparable or increased fold enrichment in the ChIP-re-ChIP relative to the solo ChIP (Fig. 2, compare C with E and compare D with F) (45). No co-occupancy of the two proteins was observed in the re-ChIP experiments when the antibody for either of the proteins was substituted with IgG (data not shown).
Bag-1L Interacts with the AR-To determine which Bag-1L domains are important for AR binding, we carried out pulldown experiments with GST-fused truncations of Bag-1L and lysate from androgen-treated LNCaP cells (Fig. 3A). We could confirm the results of Knee et al. (23), which showed that the C-terminal domain of Bag-1L binds the AR (Bag-1L⌬214, construct 5; Fig. 3, B and C). Similarly, all other Bag-1L constructs containing this part of the protein also bound the AR (Bag-1L⌬115 and Bag-1L⌬122, constructs 3 and 4; Fig. 3B). Intriguingly, a construct lacking the C terminus altogether also interacted with the AR (Bag-1L⌬C, construct 2; Fig. 3B). This binding could be traced to the first 128 N-terminal amino acids of Bag-1L (BagN128, construct 6) because amino acids 141-298 FIGURE 1. AR and Bag-1L interact in situ. A, co-immunoprecipitation of endogenous Bag-1L and AR in LNCaP cells using a Bag-1L-specific antibody for the IP and an antibody against AR to evaluate binding. An IgG IP was carried out simultaneously, as a negative control. One-tenth of the input samples are shown and confirm equal protein loading. B, PLA was performed in LNCaP cells treated with vehicle or 10 nM DHT for 4 h. Antibodies against AR, Bag-1, or IgG control were used. The quantification of the PLA dots (scored as dots/nucleus) is shown on the right. p-values were calculated using the Mann-Whitney U test; *, p Ͻ 0.05. . E and F, complex formation between AR and Bag-1L was analyzed by ChIP-re-ChIP, immunoprecipitating first with AR followed by Bag-1L ChIP (E) or by Bag-1L ChIP followed by IP with an AR antibody (F).
failed to bind the receptor (BagULD, construct 7; Fig. 3C). We have therefore identified an additional interaction site for AR on the N terminus of Bag-1L.
To show that the BagN128/AR interaction can also occur in cells (in situ), we performed a mammalian one-hybrid assay in HeLa cells, where we transfected a construct coding for a Gal4DBD-BagN128 fusion or just Gal4DBD alone together with a Gal4 binding site-luciferase reporter gene and an AR expression vector. The transfected cells were then treated with or without DHT, and the reporter gene activity was measured. A basal reporter gene activity was observed for the Gal4DBD alone or in combination with AR, independent of androgen treatment. However, when Gal4DBD-BagN128 and AR were transfected together, an increase in the reporter gene activity was observed in the presence of DHT (Fig. 3D). This demonstrates a cellular interaction of the N-terminal sequence of Bag-1L with AR in the presence of hormone.
A Duplicated Hexapeptide Motif for AR Binding-The N-terminal sequence of Bag-1L (BagN128) consists of an unidentified sequence, the nuclear localization sequence, and a T(R/ Q)SEEX repeat region (construct 1; Fig. 4A). To determine precisely which region is responsible for AR binding, we carried out GST pull-down experiments utilizing deletion mutants of BagN128 (Fig. 4A) and lysate of androgen-treated LNCaP cells. Our results show that AR binding is localized to the 1-80 but not to the 81-128 amino acids of Bag-1L (constructs 2 and 3; Fig. 4B). Further C-and N-terminal truncations of the 1-80 amino acid sequence all interacted with the AR. The smallest peptides bound by the AR were identified as amino acids 1-20 and 61-80 (constructs 6 and 9; Fig. 4B).
The only identical feature that could explain why these two peptides bind the AR is a motif, GARRPR, that is present in both sequences (Fig. 4C). We therefore postulated that the GARRPR motif is the interaction site of the AR. To validate this, we substituted the hexapeptide sequence with alanine residues and repeated the above described GST pull-down assay (Fig. 4, D-F). These alanine substitutions completely abolished the interaction between the Bag-1L peptides and the AR, confirming the importance of the GARRPR motif for AR binding (Fig. 4, E and F, compare construct 3 with construct 4 and construct 5 with construct 6). To determine the selectivity of this hexapeptide sequence for the AR compared with other nuclear receptors, we repeated the GST pull-down assays substituting LNCaP cell lysate for those from MCF7, HeLa, or T47D cells (Fig. 4G). These lysates contain high levels of estrogen (ER␣), glucocorticoid (GR), or progesterone receptor (PR), respectively. ER␣ proved similar to AR in recognizing the GARRPR motif (Fig. 4G, top row). However, although GR and PR bound both the 1-128 and the 1-80 N-terminal peptides of Bag-1L, they failed to bind the two shorter peptides (constructs 3 and 5; Fig. 4, D and G), suggesting that their binding sites are localized outside these two sequences.
The GARRPR Motif Binds the AR-LBD-To determine the region of the AR that binds the GARRPR motif, fluorescence polarization experiments were carried out with the two Bag-1L peptides (amino acids 1-20 and 61-80) labeled with FITC and purified domains of the AR covering the hinge-LBD (amino acids 646 -919), the 5 region (amino acids 360 -548), the DBD (amino acids 552-636), the hinge region (amino acids 610 -685), and, as a control, GST. Both Bag-1L peptides bound the hinge-LBD (Fig. 5A) but not the other AR domains or the GST control (data not shown). The binding affinities of the Bag-1L peptides for the AR-LBD were estimated by Sigma Plot to be K D ϭ 2.0 Ϯ 0.6 M for Bag-1L(1-20) and 3.3 Ϯ 1.3 M for Bag-1L(61-80). A positive control peptide, the high affinity nuclear receptor box of SCR-2 that contains the LXXLL core motif (46), also bound the AR-LBD with comparable affinity (K D ϭ 2.8 Ϯ 1.6 M; Fig. 5B). Given the similar affinities of the Bag-1L and SRC-2 peptides for AR-LBD, we questioned if these peptides could directly compete for binding to the AR-LBD. In line with our hypothesis, binding of the GARRPR-containing Bag-1L peptides to the AR-LBD could be outcompeted by SRC-2 (Fig. 5C), and conversely, SRC-2 binding was outcompeted by an unlabeled GARRPR core sequence (Fig. 5D). These results suggest that the binding sites of the Bag-1L and SRC-2 peptides must be similar or in close proximity to one another. Alternatively, binding of one of the peptides may allosterically hinder binding of the other.
The Bag-1L Peptides Interact with the BF-3 Pocket of the AR-LBD-Detailed studies of the AR AF-2 have shown that it forms an interaction surface for LXXLL-binding proteins or proteins with LXXLL variants with bulky aromatic hydrophobic groups, such as FXXLF (11,47). Given the highly charged nature of the GARRPR motif, it is therefore unlikely that it interacts with the AF-2 domain. It is more likely that it interacts with the recently described BF-3 domain, which is found next to the AF-2 and which allosterically modulates the activity of this domain.
In a mammalian two-hybrid assay in which we determined binding of the GARRPR-containing Bag-1L peptides to the AR-LBD, CPD49 had no significant inhibitory effect on the activity of the LBD alone (Fig. 6B, black line) but potently inhibited the interaction of the Bag-1L peptides with the LBD with IC 50 values of 4 and 10 M for Bag-1L(1-20) and Bag-1L(61-80), respectively (Fig. 6B, blue and red lines). This compares favorably with the IC 50 values of 3.3 and 6.8 M for the inhibition of PSA secretion in LNCaP and enzalutamide-resistant LNCaP cell lines, respectively (13,38).
Our molecular modeling studies show that CPD49 binds to the BF-3 domain by making extensive non-polar contacts with amino acids Phe-673, Pro-723, and Phe-826 that form the BF-3 pocket in addition to favorable electrostatic interactions between its nitrogen atom and Glu-837 of the BF-3 (Fig. 6C). CPD49 binding would therefore probably interfere with the binding of the peptides to the BF-3 domain because the proline residue of the peptide core (GARRPR) prefers to sit in the  hydrophobic cavity of the BF-3 domain (Fig. 6D). In addition, favorable electrostatic interactions appear to occur between the G(1), A(2), R(3), R(4), and R(6) of the peptide core with amino acids Arg-726, Asn-727, Gly-724, Glu-829, and Glu-837 that form the BF-3 pocket (Fig. 6D).
To further confirm the interaction of the Bag-1L peptides with the BF-3 domain, we performed a mammalian two-hybrid assay with Gal4DBD fused to the wild-type AR-LBD or LBD domains harboring mutations in the BF-3 pocket (Fig. 6E). These mutations either line the BF-3 pocket itself (Phe-826, Leu-830, Asn-833, Arg-840) or are at the boundary of the BF-3 and AF-2 domains (Arg-726) (13,49,50). In addition, we included mutant Val-757, which is located at the end of helix 5 of the AR-LBD outside the BF-3 pocket (49). Although the Bag-1L(1-20) and Bag-1L(61-80) peptides interacted with the wild-type AR-LBD, their interaction with most of the BF-3 mutants was strongly compromised (Fig. 6E). The only exception was the interaction with Phe-826, where a substitution of an arginine suppressed but did not abolish binding of the Bag-1L peptides to the AR-LBD. This F826R mutation has pre-viously been reported to have an anomalous behavior in interaction studies with other coactivators (49). The control mutation at Val-757 did not affect binding of the Bag-1L peptides (Fig. 6E). Together, these results strongly suggest that the BF-3 pocket is an essential domain for the interaction of the Bag-1L peptides with the AR-LBD.
Identification of a Signature Motif-A systematic analysis of the contribution of the individual residues in the hexapeptide motif of Bag-1L was carried out to determine the exact binding signature for Bag-1L to the AR. Each amino acid in the GARRPR cores of Bag-1L(1-20) and Bag-1L(61-80) was consecutively substituted with alanine residues. These mutants were then fused to GST and used in a pull-down assay with lysate from LNCaP cells previously treated with or without androgen. Androgen-mediated binding was destroyed when any single amino acid of the core residues GARRPR was exchanged for alanine moieties in the Bag-1L(1-20) peptide (Fig. 7A). However, this was not the case for the Bag-1L(61-80) peptide, where only the exchange of the penultimate two amino acids XXXRPX to alanine decreased binding (Fig. 7B). In a subsequent attempt to define the contribution of the individual residues in the hexapeptide motif, a protein BLAST search was carried out to determine whether homologies of the Bag-1L GARRPR sequence exist in other proteins. Over 70 proteins with nearly complete or partial homologies were identified, and peptides encompassing the homologous regions were synthesized. These peptides were then spotted onto membranes using an automated SPOT synthesizer (51) and hybridized to (bacterially-purified) AR-LBD. A wide variety of peptides containing complete to partial homology to the hexapeptide core was found to interact with the AR-LBD (Fig.  7C and supplemental Table 1). A sequence logo derived from the validated peptide sequences identified G(1) and P(5) as the crucial residues for AR binding (Fig. 7D).
To further confirm the contribution of G(1) and P(5) within the GARRPR motif for binding to the AR-LBD, some positively identified peptides from the SPOT assay were further validated in a GST pull-down experiment. These included the ribosomal RNA processing protein 36 (RRP36), retinitis pigmentosa 9 (RP), huntington-associated protein 1 (HAP1), nuclear receptor coactivator 4 (NcoA4), and p21 protein (Cdc42/Rac9-activated kinase 6 (PAK6)), the latter three of which are known to bind the AR (52)(53)(54). These peptides mostly show complete sequence identity with the core GARRPR, with the exception of NcoA4 and PAK6, which only share 2-3 homologous residues (but including G(1) and P(5)) with the GARRPR motif. We could confirm a hormonedependent binding of all of the peptides to the AR with the exception of PAK6, where binding was weaker in the presence than in the absence of hormone (Fig. 7E). Because PAK6 and NcoA4 contain only limited homologies to the GARRPR core, it is likely that only the conserved amino acids at positions 1 and 5 (GXXXPX) contribute to the interaction with the AR-LBD. In GST pull-down experiments, we could confirm this by exchanging the conserved amino acids G(1) and P(5) to alanine, which indeed reduced binding of all of the peptides to the liganded AR-LBD (Fig. 7F).
We next introduced the G6A/P10A/G66A/P70A substitutions that decreased the BagN128/AR binding into the fulllength Bag-1L; the resulting Bag-1L protein has a mutated N-terminal end but an intact C-terminal BAG-domain. We created LNCaP cells stably overexpressing HA-and FLAG-tagged versions of the wild-type or mutant Bag-1L proteins and, as a control, LNCaP cells expressing the empty vector construct. Because the GARRPR mutations are close to the nuclear localization sequence (positions 70 -76) as well as another region (amino acids 17-50) that has been reported to contribute to the nuclear localization of Bag-1L (23, 55), we carried out immunofluorescence microscopy to determine if the cellular localization of Bag-1L (or AR) is altered by the mutations. The immu- peptide sequences and sequences of other AR-binding proteins with the GARRPR sequence (or variants thereof) shown in red. D, sequence logo of the binding motif of all peptides tested in C as determined by the WebLogo website. E, the peptides shown in C were GST-tagged and incubated with LNCaP cell lysate treated with vehicle or 100 nM DHT for 1 h. Protein binding was confirmed by Western blotting using an anti-AR antibody. GST-tagged proteins were visualized using Ponceau staining. F, GST-tagged peptide versions of the sequences described in C, either in their intact form (wt) or with the G and P of their GXXXPX motifs exchanged to alanine residues (mut), were incubated with LNCaP cell lysate treated with 100 nM DHT for 1 h. Protein binding was identified by Western blotting using an anti-AR antibody. GST-tagged proteins were visualized using Ponceau staining.
nofluorescence experiments showed no change in localization of the mutant Bag-1L compared with the wild-type protein (Fig.  9A), demonstrating that the mutations do not affect the nuclear localization of Bag-1L (or the AR). We next tested the expression levels of Bag-1L and AR in the stable cell lines. Bag-1L expression levels were comparable between the wild-type and mutant Bag-1L-expressing cell lines, and Bag-1L overexpression (whether wild-type or mutant) did not change the endogenous expression levels of the AR (Fig. 9B). Last, we tested the ability of the full length and GARRPR mutant Bag-1L protein to interact with the AR by co-immunoprecipitation (Fig. 9C). We could not detect any significant interaction (either hormonedependent or -independent) between the full-length Bag-1L and AR when the GARRPR motifs were mutated, which is in agreement with the findings for the mutated Bag-1L peptides (Fig. 8A).
The N-terminal Bag-1L Mutations Cause Alterations in AR Transactivation-Having established that the GARRPR mutations impair the interaction of Bag-1L with AR (Fig. 9C), we postulated that Bag-1L-mediated regulation of AR action might also be affected by these mutations. To prove this, we performed genome-wide gene expression profiling in the LNCaP Bag-1L wild-type-and mutant-overexpressing cell lines as well as the LNCaP vector control cells. We focused entirely on ARregulated genes by culturing the cells in the absence and presence of DHT and only took into consideration differentially expressed genes with a fold change of 1.75 or higher upon hormone treatment (and a p-value of 0.05 or less). Comparison of the DHT-regulated transcriptome of the vector, wild-type, and Bag-1L mutant-transfected LNCaP cells revealed a significant number of differentially up-regulated genes that could be clustered in five different categories (Fig. 10A). The total number of down-regulated genes was modest (n ϭ 7), and therefore these genes were excluded from the analysis.
The biggest gene cluster we identified, defined as "all common genes," consists of a number of genes that were all highly expressed (Ͼ1.75-fold) in all three data sets. This is in contrast to the next cluster ("WT ϩ NMut only"), which only contains genes that were up-regulated in the Bag-1L wild-type-and mutant-overexpressing cells but showed low (Ͻ1.75-fold) or no differential expression in the vector-expressing cells. This finding is in agreement with previous studies that showed that Bag-1L overexpression potentiates the transactivation action of the AR (21,22,24). Correspondingly, gene annotation analysis revealed that the genes uniquely up-regulated in the "WT ϩ NMut" cell lines are associated with changes in cell signaling and proliferation.
We also identified a significant number of genes that were differentially transcribed across all three cell lines and were therefore categorized as being uniquely expressed; the exception is the vector category, which includes two genes that are also expressed in the wild-type cells. The largest and most significant of the unique categories is that of the N-terminal mutant-overexpressing cells, which points toward an exclusive function of the GARRPR mutations in the modulation of the  AR-mediated transcriptome. The majority (53.6%) of the genes identified within this group are associated with metabolic processes.
We validated the expression of a select number of genes by quantitative RT-PCR, which we had defined as being unique to the Bag-1L GARRPR mutant cell line (based on the microarray data). As expected, the expression of all tested genes was significantly higher in the Bag-1L mutant cell line compared with their respective expression levels in the vector-and Bag-1Loverexpressing cells (Fig. 10B). Concomitantly, AR binding at regulatory regions of these genes, as determined by AR ChIP, was higher in the Bag-1L mutant than in the vector or Bag-1L wild-type cells (Fig. 10C). In comparison, Bag-1L binding remained unaltered at the same AR regulatory sites across the three different cell lines (data not shown). Thus, the mutation of the GARRPR motif of Bag-1L, which causes the inhibition of the Bag-1L/AR interaction, triggers an increase in binding and activity of the AR at select target genes, without altering the direct binding of Bag-1L to chromatin. Together, our findings indicate that binding of Bag-1L to the AR through the (unmutated) GARRPR motif contributes to the suppression of a subset of AR-regulated genes.

DISCUSSION
In this work, we have analyzed the regulation of AR by the cochaperone Bag-1L. We could show that N-terminal amino acid sequences of Bag-1L bind to the AR and contribute to the transactivation function of the receptor. Furthermore, we identified a repeat hexapeptide motif GARRPR in Bag-1L as an essential sequence for binding to the AR-LBD. This motif functions through sequences that constitute the newly identified allosteric regulator BF-3 in the AR-LBD, and mutations in this motif cause the loss of Bag-1L/AR interaction and alter the transactivation function of the AR.
Previous studies to identify sequences on coactivators that bind the LBD of nuclear receptors using phage display libraries have only identified variants of the LXXLL motif (11,47). For example, peptides bound to the AR coactivator binding domain include a combination of aromatic residues and the canonical leucine-rich sequence of LXXLL such as FXXLF, FXXLW, WXXLF, WXXVW, FXXFF or FXXYF (11,47). Similarly, peptides that interact with the coactivator binding domain of another steroid receptor, the ER␣, can be categorized into three classes of LXXLL motifs that differ in their flanking sequences but not in the canonical motifs themselves (56).
Well characterized AR coactivators include the p160/SRC family of proteins that bind to the 5 of the AR through a glutamine-rich sequence independent of their intrinsic LXXLL motifs (57)(58)(59). A coactivator MED1 has also been shown to bind to the 1 region at the N terminus through two noncanonical ␣-helical motifs but independent of its LXXLL motifs (60). The GARRPR sequence we have identified in the present study presents yet another mode of AR regulation and differs both structurally and functionally from the LXXLL motif. First, it is highly charged compared with the LXXLL motif, which is highly hydrophobic. Second, it exhibits target specificity, because it binds the AR and the ER␣ but not the GR or the PR. This is therefore unlike the LXXLL motif, which interacts with the coactivator pocket, AF-2, of several nuclear receptors, including the AR. However, like the LXXLL motif, the GARRPR motif binds to the AR-LBD. Compellingly, our studies show that the interaction of the AR-LBD with the GARRPR-containing Bag-1L peptides can be outcompeted with an LXXLL-containing SRC-2 peptide. Similarly, the AR-LBD/SRC-2 interaction can be competed off by a GARRPR core peptide. This might suggest that the GARRPR and LXXLL motifs bind to closely related sequences on the AR. Alternatively, binding by one peptide may produce allosteric changes in the flanking region of AR that may affect binding by the other. However, it is more plausible to us that the GARRPR sequence binds to the newly identified allosteric region of the AR-LBD termed BF-3 (13), rather than the AF-2 pocket. We present three pieces of evidence that support this possibility. First, the use of a small, specific, molecule inhibitor of the BF-3 destroys binding of the Bag-1L peptides to the AR-LBD. Second, molecular modeling shows that the GARRPR fits into the BF-3 pocket with the proline residue localized in the hydrophobic cavity of this pocket. Third, single amino acid exchanges in the BF-3 pocket, but not an outside sequence, destroy binding of the Bag-1L peptides to the LBD in a mammalian two-hybrid assay. Together, these results strongly suggest that the Bag-1L peptides bind to the BF-3 pocket of the AR.
Although the BF-3 pocket has been postulated to bind regulatory proteins, it has thus far been difficult to demonstrate factor-specific binding to this site because of its weak affinity for proteins (13). So far, only FKBP52, a cochaperone of Hsp90 that binds the AR⅐Hsp90 complex and regulates ligand binding to the AR, has been shown to function through the BF-3 pocket. However, a direct demonstration of binding is lacking (61). Studies on a possible interaction sequence of FKBP52 with the BF-3 pocket suggested the involvement of alanine and proline residues in the sequence GSAGSPP at the N terminus of FKBP52 (61,62). This sequence is strikingly similar to the GARRPR motif we identified, but it is completely different from the LXXLL coactivator motif that binds the AF-2 domain of the AR. Nonetheless, there are differences in the functional consequences of the binding of FKBP52 and Bag-1L to the AR-LBD. The former binds through its N-terminal FK506 binding domain and is required for hormone binding, whereas the latter interaction is through a so far uncharacterized N-terminal Bag-1L sequence that seems to control the transactivation function of the AR. Future structural experiments will be needed to distinguish between the interactions of these two proteins with the BF-3 of the AR. For now, our results with Bag-1L agree with the notion that agents that bind to the BF-3 pocket allosterically regulate the transactivation function of the AR (13).
Studies on how BF-3-lining residues affect AR activity show that mutations in the BF-3 pocket alter the ability of AF2-binding coactivators to modulate the transactivation function of the AR (49). Residues in the BF-3 pocket are also targets for prostate cancer and androgen insensitivity syndrome mutations (63)(64)(65)(66)(67). Our finding that the GARRPR motifs of Bag-1L bind to the BF-3 would therefore strongly implicate Bag-1L in the development or progression of androgen insensitivity syn-drome and/or prostate cancer. In our study, we could show that mutations in the GARRPR motif that destroy binding to the BF-3 pocket did not inhibit AR-mediated gene expression but rather increased the expression of a subset of androgen-regulated genes mediating metabolic processes. Previous studies on AR-mediated up-regulation of genes involved in metabolism have pointed to a role of the AR in the normal prostate epithelium in driving the synthesis of seminal fluid (68,69). Therefore, our finding that the mutant Bag-1L increases the expression of AR-regulated genes involved in metabolism indicates a shift from the involvement of Bag-1L in promoting AR action in prostate cancer toward a normal prostate function.
The N terminus is not the only part of Bag-1L that binds the AR. The C terminus of Bag-1L has been shown to bind the AR in vitro, as identified by GST pull-down studies (21,23). However, our in vivo studies show a complete impairment of binding of Bag-1L to the AR after mutating the GARRPR sequence at the N terminus of Bag-1L. Thus, we can conclude that in the cellular context, the N-terminal sequence of Bag-1L is very important for AR binding. In addition, the N terminus of Bag-1L also binds DNA through lysine-and arginine-rich sequences between amino acids 68 and 81 at its N terminus (42,70,71). We therefore postulate that the association of Bag-1L with the AR, as demonstrated in this work, allows the (intact) AR⅐Bag-1L complexes to bind to enhancer elements of AR target genes (with both proteins binding to chromatin). However, when the Bag-1L/AR interaction is disrupted by mutations in the GARRPR motifs, although the interaction of Bag-1L and chromatin still persists, the AR is now free to bind more avidly to distinct enhancer sequences. This would explain the increased AR binding and androgen response of a subset of genes following the overexpression of the mutated Bag-1L in LNCaP cells, as shown in our experiments. Our findings therefore offer an opportunity for studying the mechanism of repression and derepression in the control of AR transactivation in prostate cancer cells.