Cooperative Assembly of Androgen Receptor into a Nucleoprotein Complex That Regulates the Prostate-specific Antigen Enhancer*

Prostate cancer is characterized by elevated serum levels of prostate-specific antigen (PSA). PSA gene expression is controlled by an androgen-responsive transcriptional enhancer. Our study suggests that formation of a nucleoprotein complex, encompassing 170 base pairs of enhancer DNA, mediates androgen-responsive PSA enhancer activity. The complex is assembled by cooperative binding of androgen receptor to at least four tandem, nonconsensus androgen response elements (AREs). Systematic mutagenesis of the AREs demonstrated that they act synergistically to stimulate androgen receptor-responsive gene expression. We discuss a mechanism whereby a combination of high androgen receptor levels in the prostate and low affinity AREs contribute to the cell type specificity and activity of the enhancer.

Prostate cancer is characterized by elevated serum levels of prostate-specific antigen (PSA). PSA gene expression is controlled by an androgen-responsive transcriptional enhancer. Our study suggests that formation of a nucleoprotein complex, encompassing 170 base pairs of enhancer DNA, mediates androgen-responsive PSA enhancer activity. The complex is assembled by cooperative binding of androgen receptor to at least four tandem, nonconsensus androgen response elements (AREs). Systematic mutagenesis of the AREs demonstrated that they act synergistically to stimulate androgen receptor-responsive gene expression. We discuss a mechanism whereby a combination of high androgen receptor levels in the prostate and low affinity AREs contribute to the cell type specificity and activity of the enhancer.
Androgen receptor (AR) 1 plays a central role in prostate cancer progression. Depletion of androgens by surgical or chemical treatments slows cancer growth. AR activates transcription of the gene encoding the kallikrein protease prostatespecific antigen (PSA) and other genes involved in secretory epithelial cell metabolism. Serum PSA levels parallel oncogenic growth in the initial androgen-dependent phase and in early stages of androgen-independent cancer. The prostate specificity of the PSA transcriptional regulatory region has made it an ideal reagent for use in gene therapy trials (1)(2)(3)(4). We have focused on the role of AR in PSA regulation as a model for understanding how AR regulates prostate gene expression.
Biochemical and genetic studies have led to the cloning and partial genetic dissection of the PSA promoter and enhancer (3,(5)(6)(7)(8)(9). Both the enhancer and promoter display androgen responsiveness consistent with the observation that both regions contain androgen response elements (AREs) (3, 5, 7, 8, 10 -12). The proximal promoter has been delineated to an ϳ630-bp fragment containing a core TATA box (5) and two AREs, ARE I and ARE II (8). AR activates transcription synergistically from these AREs (8). Although the promoter plays an important role in PSA expression, an experiment in transgenic mice has shown that it is insufficient to confer strong androgen responsiveness and cell type specificity in vivo. In the same experiment, however, a 6-kb region encompassing the promoter and extending further upstream was able to mediate proper regulation (14).
An enhancer element, centered at approximately Ϫ4.2 kb, is located within this 6-kb region (6). The enhancer was originally identified as a 1.6-kb fragment containing, by sequence analysis, sites recognized by the androgen receptor, AP-1, cAMPresponsive element binding protein, and Fos (7). Two subsequent studies have further delineated the enhancer to an 822-bp fragment (3) and a 455-bp minimal core region encompassing an androgen-responsive DNase I-hypersensitive site (6). The role of the single identified ARE, termed ARE III, within this minimal core enhancer is not entirely clear. In one study, mutation of ARE III eliminated enhancer activity in transfection assays (6). In a different study, however, mutation of ARE III in combination with ARE I of the promoter was required to abolish androgen responsiveness (11). Although these results may reflect differences in methodology, we will show below that the enhancer is more complex than previously realized. We report that the enhancer contains multiple low affinity, non-consensus AREs, which bind AR cooperatively and act synergistically to stimulate transcription.
AR is a 110-kDa protein containing an amino-terminal transcriptional activation domain (AF-1) spanning amino acids 141-338 (16 -22), a zinc finger DNA binding domain from amino acids 556 -623, and a carboxyl-terminal hormone/ligand binding domain from amino acids 666 -918 (23). AR is a member of the nuclear receptor superfamily. The remarkable homology of the conserved domains suggests that family members will activate transcription using conceptually similar mechanisms (24). The ligand binding domain contains an additional activation domain (AF-2) by virtue of its homology to similar regions in related receptors and its ability to interact with co-activators (25). Interaction between the amino-and carboxyl-terminal activation domains is thought to be important for full activity of AR (16,26). The regulatory functions of AR in transcription have been demonstrated both in vivo and in vitro (19,(27)(28)(29)(30).
The DNA binding domain of AR is a dimer and exhibits strong sequence homology with the progesterone (PR) and glucocorticoid (GR) receptors (31)(32)(33)(34). Expression of androgenresponsive genes is regulated by binding of ligand-activated AR to androgen response elements (AREs). Characterization of different AREs indicates that there are two classes of sites (23). One group contains the 15-bp near-dyad, consensus sequence (GG(A/T)ACAnnnTGTTCT). The consensus also binds to and mediates transactivation by GR and PR. A second class of sites, however, has been identified with sequences that diverge considerably from the consensus. It is likely that these non-consensus sites contribute to specific binding by AR (8,(35)(36)(37).
AREs are found in promoters of genes expressed in different tissues including prostate, brain, kidney, liver, and testis (23). It is unlikely, therefore, that AR alone regulates cell-type specificity of PSA gene expression. Indeed, DNase I footprinting and gel shift studies have identified several binding sites for prostate-specific and ubiquitous transcription factors within the PSA enhancer and promoter (7,9,12). It is likely that a combination of AR and cell-specific factors confers tissue selectivity. One of the hallmarks of such regulation is the formation of nucleoprotein complexes that use cooperative DNA binding and transcriptional synergy to elicit specific patterns of gene expression (38).
In an effort to understand the regulation of the PSA gene as a model for AR-mediated gene expression during prostate cancer, we undertook a biochemical analysis of the core enhancer. We employed DNase I footprinting to demonstrate that the enhancer contains a cluster of six closely spaced putative AREs differing in affinity for recombinant AR DNA binding domain (ARDBD). We systematically mutated these sites to demonstrate that at least four of the six sites were physiologically relevant in conferring response to AR in co-transfection assays into BHK and LNCaP cells. We then generated a retroviral construct encoding a FLAG-tagged version of AR, introduced it into HeLa and LNCaP cells, and isolated the receptor by immunoaffinity chromatography. We demonstrated by transactivation, nuclear translocation, and DNA binding assays that the FLAG-tagged AR was biologically active in vivo and in vitro. Purified fAR bound cooperatively to the four core sites in the enhancer and responded to ARE mutants in a manner that roughly paralleled the in vivo transcriptional analysis. Conversely, PR did not substantially activate transcription from the core enhancer but did activate from a reporter bearing tandem AREs. We discuss how the cooperative binding to and synergistic activation of the enhancer by AR can contribute to the strong androgen responsiveness and cell selectivity of PSA gene expression.

EXPERIMENTAL PROCEDURES
Plasmids-pET11dHis 6 ARDBD was constructed by insertion of a polymerase chain reaction (PCR) DNA fragment, encoding the DNA binding domain of human AR, from amino acids 549 -650, into the BamHI site of pET11dHis 6 heart muscle kinase (provided by R. Hori). The construct generated a 127-amino acid protein with a His 6 tag on the amino terminus and heart muscle kinase phosphorylation site on the carboxyl terminus.
The wild-type PSA enhancer-E4CAT reporter vector was constructed by PCR subcloning a 496-bp PSA enhancer region from the 2.4-kb enhancer described by Belldegrun and colleagues (3) in reverse orientation into HindIII/XbaI sites upstream of E4 TATA box of pE4TCAT (39). The PSA enhancer-E4LUC reporter vector was constructed by PCR amplification and subcloning of the PSA enhancer-E4 TATA region (up to ϩ38) into SacI/XhoI sites of pGL3-Basic vector (Promega). All constructs were sequenced to confirm their integrity. The constructs contained several notable point mutations versus the published sequence by Henderson and colleagues (15). These point mutations were present in the original sequence and were not introduced inadvertently by PCR. The DNA sequence of ARE III, Ϫ4150GGAAGAtatTGTATC-4136, was altered to GGAACAtatTGTTATT, a change that brings the sequence into better alignment with the ARE consensus. Furthermore, the sequence of ARE VI, Ϫ4303GGATGCtgtGCAGAA-4289, was altered to GGATGCtgtGCACAC. Two other mutations outside of identifiable AREs were also detected within this region. We have not tested if any of the nucleotide changes affect enhancer activity.
Enhancer mutant vectors were constructed by a two-step overlap PCR method where AREs were replaced by GAL4-binding sites and restriction enzyme cleavage sites to conserve phasing of the DNA (below, the replacement sequence is in capitals).
FLAG-tagged AR was constructed in two steps. In the first step a primer encoding the FLAG peptide fused to the sequences encoding the amino-terminal 5 amino acids of AR (5Ј-GCTCTAGACCACCATGGAC-TACAAGGACGACGACGACAAGGCCGAAGTGCAGTTAGGGC-3Ј) was used in combination with an internal AR primer (5Ј-CCCTCTAG-ACGGCCGAGGGTAGACCCT-3Ј, encoding amino acids 9 -14), to obtain a DNA fragment from amino acids 1-14 of AR. This fragment was digested with XbaI and inserted into the XbaI site of pBluescriptKS (Stratagene). The resulting plasmid was digested with EagI, and an EagI fragment of the AR cDNA, encoding the remaining sequences from amino acids 10 -919, was cloned in. The FLAG-tagged full-length AR (fAR) cDNA was then excised from the pBluescriptKS vector by XbaI digestion and placed into the XbaI site of pSR␣ (40) to generate pSR␣fAR.
ARDBD Purification-The Escherichia coli strain BL21(DE3) was transformed with pET11dHis 6 ARDBD. 0.1 liter of cells was grown to an Expression and Purification of FLAG-tagged AR-20 g of pSR␣fAR was transfected with helper virus into a 10-cm dish of 293T cells at 50% confluency using calcium phosphate transfection (41). One day after transfection, virus was collected in Iscove media containing 10% FBS at 12-h intervals for 48 h. Virus was isolated by filtration of the culture supernatant through a 0.45 M filter. 2-3 ml of virus supernatant was incubated with HeLa and LNCaP cells at 10 -30% confluency (LNCaP cells 20 -30%, HeLa cells 10 -20%) in the presence of 8 g/ml Polybrene for 4 h or overnight. 10 ml of medium (DMEM for HeLa cells, RPMI 1640 for LNCaP cells) was added, and cells were passaged for 24 h. 800 g/ml G418 (Life Technologies, Inc.) was then added. After 2 weeks, individual clones were isolated and expanded, and levels of expression were compared by immunoblotting with FLAG monoclonal antibodies (Sigma). The clone expressing the highest level of fAR was chosen for large scale preparation.
Nuclear extracts from both HeLa and LNCaP cells expressing fAR were prepared as described previously (42). Although the scale of the purification varied among our different preparations, on average, 1 ml of nuclear extract was incubated with 20 l of agarose beads conjugated with FLAG monoclonal antibodies (Sigma) at 4°C for 6 h. The beads were washed three times with 0.5 ml of Buffer D containing 0.3 M KCl, 0.05% Nonidet P-40, 0.5 mM phenylmethylsulfonyl fluoride, and 0.5 mM dithiothreitol and eluted twice, 20 min each, with 20 l of Buffer D containing 0.2 mg/ml FLAG peptide (Eastman Kodak Co.).
Western Blotting-Immunoblotting was performed according to standard procedures. Briefly, proteins were separated on a 12% SDSpolyacrylamide gel and transferred to a Hybond C Extra membrane (Amersham Pharmacia Biotech). The membrane was blocked with phosphate-buffered saline containing 5% nonfat milk for 30 min, incubated sequentially with primary and secondary antibodies for 30 min each, and then developed using ECL reagents from Amersham Pharmacia Biotech. Antibodies against the amino terminus of AR were purchased from Santa Cruz Biotechnology. Monoclonal antibodies against the FLAG tag of AR were purchased from Kodak through Sigma. Secondary rabbit and mouse antibodies were from Bio-Rad.
DNase I Footprinting-The binding reactions for DNase I footprinting were as described previously (43). Indicated amounts of recombinant AR DNA binding domain (ARDBD), FLAG-tagged AR, and crude nuclear extracts from f-LNCaP cells were incubated in 13 l of buffer containing 12.5 mM HEPES, pH 7.9, 12.5% glycerol, 5 mM MgCl 2 , 70 mM KCl, 0.2 mM EDTA, 60 mM mercaptoethanol, 0.5 mg/ml bovine serum albumin, and 200 ng of poly(dG-dC). After 45 min at 30°C, DNase I was added to the reactions. The cleavage reactions were terminated after 1 min by addition of 100 l of stop buffer containing 400 mM sodium acetate, pH 5.2, 0.2% SDS, 10 mM EDTA, 50 g/ml yeast tRNA, and 100 g/ml proteinase K. The mixtures were incubated at 50°C for 15 min, extracted with phenol/chloroform, and the products precipitated with ethanol. Precipitates were dissolved in formamide loading buffer and analyzed on 8% polyacrylamide/urea sequencing gels. The cleavage ladders were visualized by exposure to XAR-5 film or by PhosphorImager analysis.
Cell Culture and Transfections-BHK 21 cells were grown in DMEM (Life Technologies, Inc.) supplemented with 10% FBS, L-glutamine, and antibiotics (penicillin/streptomycin). 2.5 ϫ 10 5 cells per well were plated into 6-well plates in phenol red-free DMEM (Mediatech) supplemented with 10% charcoal/dextran-treated FBS (Omega Scientific) 24 h prior to transfections. Transfection of BHK 21 cells was performed using the calcium phosphate method or Tfx-10 reagent (Promega). 200 ng of reporter plasmid and 50 ng of CMV-AR or CMV-PR were used in transfections. AR and PR were induced with 1 nM R1881 and 10 nM progesterone, respectively. 48 h after transfection, the cells were harvested and CAT or luciferase assays were performed according to standard procedures.

RESULTS
The Enhancer Region-The goal of our study was to understand the molecular basis for the androgen responsiveness of the core PSA enhancer. A 496-bp DNA fragment (Fig. 1D) bearing the minimal 455-bp minimal enhancer (6) was amplified by PCR, confirmed by DNA sequencing, and cloned upstream of the adenovirus E4 core promoter (Ϫ38 to ϩ38) fused to either CAT or luciferase reporters. We chose the E4 reporter for transfection studies because it is highly inducible by activators and has been employed in numerous studies to examine the properties of mammalian enhancers (39). Furthermore, transfection experiments revealed that the E4 promoter alone did not respond to AR in the absence of upstream AREs (data not shown). The resulting PSA-E4 template was employed both in DNA binding analyses with recombinant AR and in transfection studies with endogenous AR or AR expressed from the CMV enhancer/promoter.
Identification of a Cluster of Low Affinity AREs in the PSA Enhancer-We first subjected the enhancer region to DNase I footprinting analysis using the AR DNA binding domain (ARDBD), amino acids 549 -650. The ARDBD was expressed as a His 6 fusion protein from a modified pET11d (Novagen) T7 expression vector in E. coli. The E. coli lysate was bound to a Ni-NTA resin, and ARDBD was eluted with imidazole as de-scribed under "Experimental Procedures." To purify active ARDBD away from inactive protein, we subjected the Ni-NTA eluate to an ARE-DNA affinity column. The final protein was Ͼ95% homogeneous as measured by Coomassie Blue staining of SDS-polyacrylamide gels (Fig. 1A). The peak fractions in Fig.  1A were pooled and employed in the experiment of Fig. 1B.
Our initial objective was to confirm the presence of a single published AR-binding site in the PSA enhancer using DNase I footprinting analysis. This site, termed ARE III, is centered at Ϫ4200 (6) or Ϫ4143 (7, 11) depending on the numbering system used in different laboratories. The high specific activity of our ARDBD, however, allowed us to detect five additional binding sites within the enhancer region, each footprint approximately 23 bp in size. The sites are visible on the autoradiograph in Fig.  1B. We have denoted the additional sites AREs IIIB, IIIA, IV, V, and VI with respect to their position relative to ARE III.
The affinity of the sites varied considerably. Among the six sites, site III had the highest affinity consistent with the study by Trapman and colleagues (6). Overall, however, the AREs were 15-60-fold lower affinity than the consensus (see Fig. 1B, legend). This observation may bear on the tissue specificity of PSA expression, an issue we will cover under "Discussion." Fig.  1C summarizes the positions of all six AR-binding sites in the context of the minimal AR-responsive PSA enhancer. The DNA sequences and relative positions of the sites are shown in Fig. 1D.
The observation that the affinities of AREs in the enhancer are significantly lower than the consensus raised the issue of their physiological relevance. Indeed, inspection of the DNA sequences within the footprinted region revealed few matches to the consensus ARE. Even ARE III diverged from the consensus by two nucleotides. To investigate this issue in more detail, we performed three experiments. First, we subjected the enhancer AREs to substitution mutagenesis to determine which sites were important in co-transfection experiments with AR into BHK cells. We then compared the key constructs by transfection into a more physiological prostate cancer cell line, LNCaP. Finally, we developed a procedure to purify intact AR to homogeneity from mammalian cells and employed the intact AR to study binding to the enhancer and its substitution mutants.
Enhancer Mutagenesis and Transfection Analysis-To establish the physiological validity of the new AR-binding sites, we mutagenized them individually and in select pairwise combinations. The AREs were replaced by GAL4-binding sites using PCR mutagenesis techniques. The substitutions were constructed based on the positioning of the footprints. In creating the substitution mutants, we attempted in some cases to remove precisely the site and in others we attempted to disrupt both the target site and putative adjacent sites (see regions marked by dark bars in Fig. 1D). The rationale for this approach will be justified below, where we show that intact AR binds over the entire region between ARE IIIA and ARE V. In all cases we preserved the original spacing by replacing the enhancer sequences with an equal number of bases. We employed the 17-bp GAL4 sites because GAL4 is a yeast transcriptional activator, and numerous studies have shown that in the absence of GAL4, or its derivatives, the sites have no endogenous transcriptional activity in mammalian cells (39). Fig. 2A shows a representative DNase I footprint demonstrating that mutants lacking either ARE III, ARE IV, or ARE V no longer bind the ARDBD but instead bind to the GAL4 DNA binding domain.
To determine whether the AREs contributed to the AR response, we first performed co-transfection assays into the baby hamster kidney (BHK) cell line. We chose BHK cells because they contained low amounts of endogenous AR (i.e. relative to LNCaP) as measured by immunoblotting (data not shown) and were unlikely to contain prostate-specific transcription factors. This experimental scenario permitted us to assess directly the contribution of co-transfected AR, and its interaction with AREs, to enhancer activity. Fig. 2B is a bar graph showing the average of three independent co-transfection experiments of an effector plasmid, expressing AR from the CMV enhancer (22), and wild-type and mutant PSA-E4-CAT enhancer constructs.
The transcriptional activity of the enhancer was dependent upon co-transfected AR and the addition of 1 nM R1881 (data not shown). Sites IIIA, III, IV, and V contributed significantly to enhancer activity because, when these sites were mutated, transcription decreased by 50 -75% relative to wild type. In contrast, mutation of sites IIIB and VI had only marginal effects. For the remainder of the paper we will focus solely on AREs IIIA, III, IV, and V. This is not to say that sites IIIB and VI are unimportant. However, the experimental design only allowed us to confidently attribute androgen responsiveness to AREs IIIA-V.
The wild-type and mutant enhancer constructs containing a luciferase reporter were transfected into an LNCaP cell line (Fig. 2C). We reasoned that by differing the concentration of ligand we might observe differences in the effects of particular mutants, i.e. different mutants might be sensitive to the concentration of active AR. At 1 nM R1881 the individual site mutants all retained low activity averaging about 20% that seen with wild type. However, the effects of S-III and S-IV mutants were consistently more severe at 0.3 nM ligand. Furthermore, the residual, but reproducible, activity of the foursite mutant (S-All) at 1 nM ligand disappeared at 0.3 nM. Remarkably, the S-IIIA and S-V mutants retained much of their activity at the lower concentration of ligand. In all cases the sites appeared to act synergistically in LNCaP cells as removal of any site elicited a disproportional decrease in activity. We noted that the transcriptional activities of the mutants  (25,50, and 100 ng, as shown by the filled triangles) were incubated with the 32 P-labeled enhancer fragment, from Ϫ4366 to Ϫ3824, to detect footprints on both strands of the enhancer. The sites identified are denoted as IIIA, IIIB, III, IV, V, and VI. Sequencing ladders are included alongside the footprints to localize the sites. A combination of footprinting (B) and gel shift studies (data not shown) revealed the following affinity order: ARE consensus Ͼ promoter ARE I Ͼ ARE III Ͼ ARE V Ͼ ARE IIIA ϭ ARE IV Ͼ ARE VI ϭ ARE IIIB. We find it likely that site IIIA is a combination of two closely spaced sites as the footprint is slightly greater than 23 bp in size. We will refer to it as a single site with the caveat that it may comprise two overlapping sites. were similar to those observed in BHK cells, although the deleterious effects of individual mutations were more pronounced in LNCaP. The more severe effect in LNCaP might be attributed to the fact that the LNCaP line expresses physiological levels of AR, whereas the CMV-driven AR is likely to accumulate at high levels in BHK cells. Alternatively, the presence of prostate-specific factors may modulate the response to AR in LNCaP.
To verify further that AR binding to individual AREs is responsible for induction of the enhancer activity, we created three point mutations targeted to site V, a novel site identified in this study. We chose site V because its sequence matched the consensus more clearly than sites IIIA and IV, and site III has already been validated as an authentic ARE by point mutagen-esis in other studies. Furthermore, AR apparently binds sites III and V in footprinting experiments with LNCaP extracts (data to be presented below). Among the three point mutants, mutants, MV1 and MV3 were inactive for AR binding whereas mutant MV2 retained AR binding activity comparable with that of wild-type site V as determined by DNase I footprinting (Fig. 3A, compare lanes 3, 6, 9, and 12) and gel shift assays (data not shown). When tested for enhancer activity by transfection into in LNCaP cells, the activity of these mutants correlated with their AR binding activity. As shown in Fig. 3B, the activity of MV1 and MV3 is significantly reduced while the activity of MV2 is comparable with that of the wild-type enhancer.
Since hormone-responsive elements are conserved among different nuclear receptors, it is possible that the PSA enhancer activity might not be AR-specific. To address the receptor specificity issue, we tested the responsiveness of this enhancer to progesterone receptor (PR). We performed the experiment in BHK cells as the AR in LNCaP cells contains a mutation in AF-2 that renders it activable by numerous ligands. Such an effect would preclude us from measuring the specificity and contributions of PR to enhancer activity. Both AR and PR, when co-transfected with a reporter template containing four ARE sites upstream of the luciferase gene, were able to strongly induce activity in a ligand-dependent fashion (data not shown). However, co-transfection of AR but not PR had a significant effect on PSA enhancer activity (Fig. 4).
Synthesis and Purification of Intact AR-The tandem arrangement of low affinity AREs was reminiscent of many eukaryotic promoters where cooperative DNA binding by activators is an important mechanism for ensuring specificity in gene expression. The DNA binding experiments of Figs. 1-3 were performed with ARDBD which, we reasoned based on previous studies of GR and AR, could be missing domains necessary for cooperativity. Indeed Fig. 1B showed that ARDBD filled the enhancer sites at widely varying concentrations, whereas cooperative binding is generally characterized by simultaneous site occupancy.
To understand the binding profile of intact AR it was necessary to purify AR to homogeneity. Although AR had been purified previously from the insect baculovirus system (44), we chose to express and purify AR from mammalian cells in the event that AR was post-translationally modified in subtle ways that might influence its activity. AR had also previously been expressed in the mammalian vaccinia system (28). We initially tested this system but felt that the high level of overexpression and cytotoxicity of the virus might also preclude important modifications, and prevent AR from being used for certain applications.
We therefore chose to stably express AR in both HeLa cells, where AR has been shown to be active in co-transfection as-says, and in LNCaP cells, where AR is normally expressed and is essential for normal metabolic growth. We employed an approach previously used in the purification of the thyroid receptor (45). We first generated a subclone of AR containing the 8-amino acid FLAG epitope at its amino terminus. The FLAG-tagged AR (fAR) was then introduced into an amphotropic retroviral vector pSR␣, packaged into virus, and used to stably transform HeLa and LNCaP cells (40,41).
Stable HeLa cell lines were selected by resistance to G418. Clones expressing fAR were identified by immunoblotting cell extracts against the FLAG epitope (Fig. 5A). Clone 2, which expressed the highest level of fAR, was chosen for further analysis and scale-up. One caveat of our approach is that fAR may not behave like the endogenous AR due to the addition of the FLAG immunotag. To address this issue we subjected fAR to several biological and biochemical tests. First, we transfected a plasmid expressing fAR into BHK cells and showed that it stimulated PSA enhancer activity to the same extent as native AR (data not shown). Although this test confirmed the efficacy of fAR, we thought it was important to determine whether the stably expressed fAR was functional because we would be purifying it from the cell lines for biochemical analysis. To address this issue we transfected a luciferase reporter containing the entire PSA promoter/enhancer region, previously shown to be highly AR-responsive (3), into clone 2 HeLa cells. HeLa cells are an ideal assay system because they do not contain appreciable amounts of endogenous AR and therefore allow us to measure directly the effects of the fAR on transcription.
The bar graph of Fig. 5B demonstrates that the PSA transcriptional activity was 5-fold higher in the fAR-HeLa cells in the presence of 1 nM of the synthetic androgen agonist R1881 than in its absence. In contrast, R1881-activable enhancermediated transcription was undetectable in the parental HeLa cell line. Although the absolute level of stimulation was 5-fold lower than that observed in LNCaP cells, this might be explained by the fact that fAR-HeLa cells do not contain prostate-specific factors that might augment enhancer activity. Nevertheless, the data indicate that fAR is functional for transcriptional activation in vivo. We also compared the fAR levels in nuclear extracts prepared from clone 2 HeLa cells in the absence or presence of R1881 (Fig. 5C). As expected, R1881 was able to stimulate fAR nuclear translocation resulting in approximately 10-fold more fAR in extracts from R1881-treated clone 2 cells versus untreated cells.
To test further the in vivo function of LNCaP fAR, we performed an immunofluorescence assay using FLAG monoclonal antibodies. Fig. 6A shows that fAR localizes to the cytoplasm in steroid-depleted medium. However, after treatment with 1 nM R1881, the majority of fAR translocated into the nucleus. Because LNCaP cells contain endogenous AR, we wished to determine the ratio of fAR to the endogenous AR. We subjected similar amounts of nuclear extract from the parental and fAR-LNCaP cells to immunoblot analysis using antibodies against AR. There appeared to be approximately twice as much AR in fAR-LNCaP cells as in the parental cell line (Fig. 6B, upper panel, compare lane fLNCaP NE with lane LNCaP NE), indicating that the expression levels of fAR and endogenous AR are approximately equal. This result was further verified by subjecting the fAR-LNCaP nuclear extracts to immunoprecipitation with anti-FLAG antibody to remove the fAR. Only half of the total AR-specific signal was depleted under conditions where all of the fAR was removed (Fig. 6B, lower panel).
We next employed anti-FLAG immunoaffinity chromatography to purify fAR from the extracts. The major advantage of the FLAG approach versus standard immunoprecipitation is that the AR can be eluted from the resin with FLAG peptide for subsequent biochemical analyses. Fig. 7A shows a silverstained gel of the purified FLAG-tagged AR from both fAR-LNCaP and fAR-HeLa cells (Fig. 7A, lanes f-LNCaP and f-HeLa). A prominent band at 110 kDa, the size predicted for AR, was observed. To ensure that the band was indeed fAR, we subjected extracts from the parental LNCaP and HeLa cell lines to a mock immunopurification (Fig. 7A, lanes LNCaP and  HeLa). The 110-kDa band was not observed, and only a low background binding of contaminant proteins was evident on the gel. Furthermore, the highly pure fAR is immunoreactive to polyclonal AR antibodies in blotting analysis (data not shown).
Purified fAR was then subjected to DNA binding assays to examine its ability to bind the ARE sequence. The gel shift experiment of Fig. 7B shows that 10 ng of fAR, purified from nuclear extracts of fAR-HeLa and fAR-LNCaP lines treated with R1881, gives rise to specific gel shift complexes on a 32 P-labeled consensus ARE oligonucleotide (lanes 5 and 6) but not on a mutant ARE containing a scrambled arrangement of bases in one of the dyad half-sites (lanes 17 and 18). Further addition of 100 nM R1881 had little effect on the binding activity either in this experiment (lanes 11 and 12) or in a parallel experiment, where more limiting amounts of AR were added (data not shown). Mock protein preparations from wild-type HeLa (HeLa Nuc) and LNCaP (LNCaP Nuc) nuclear extracts did not generate shifted complexes nor did protein prepared from cytoplasmic extracts of the FLAG-tagged cells (f-HeLa Cyt). Cooperative Binding of fAR to the PSA Enhancer-In cases such as the mammalian interferon-␤ promoter (46), the yeast GAL4-controlled GAL1-10 promoters (47), and others, it has been found that small changes in the concentration of limiting activators can cause large changes in transcriptional output. Occupancy of those promoters by their respective activators occurs in a cooperative fashion. The cooperative occupancy is a result of protein-protein interactions between activators bound to adjacent sites. Cooperative binding of AR has previously been observed on the rat probasin and the mouse sex-limited regulatory regions (36,37). These studies prompted us to investigate whether AR binds cooperatively to the sites within the PSA enhancer.
A 32 P-labeled PCR fragment from Ϫ4309 to Ϫ4011 of the enhancer was subjected to DNase I footprinting studies. Increasing concentrations of ARDBD led to gradual protection of the AREs such that 75% occupancy of ARE III and ARE V occurred at 350 nM and 1.4 M protein, respectively (Fig. 8,  lanes 10 -15). In contrast, LNCaP fAR occupied several of the sites simultaneously. A 2-fold increase in concentration from 7 to 14 nM led to 60% protection of AREs III, IV, and V (Fig. 8,  lanes 4 and 5). ARE IIIA filled at the slightly higher concentration of 56 nM AR and coincided with greater than 80% occupancy of sites III, IV, and V. More remarkably, fAR protected not only AREs IIIA, III, IV, and V from DNase I cleavage but also the regions between the sites (Fig. 8, lane 6). The binding is specific because mutation of all four AREs abolished the entire DNase I footprint (Fig. 8, lanes 8 and 9). Furthermore, one striking similarity between the protection patterns of AR and ARDBD is a distinct DNase I-hypersensitive site upon protection of site V (indicated by an arrow in Fig. 8). This observation indicated that AR was binding in a manner similar to that of ARDBD. We also noticed a weak footprint below site V that we had not detected with ARDBD in earlier experiments. This footprint may represent binding of another molecule of AR, an issue we will return to below.
It has been reported that regions outside of the DNA binding domain of AR can contribute to its affinity, and possibly specificity, although the mechanism has not been determined (48). To eliminate the possibility that intact AR has different specificity requirements than the ARDBD alone, we compared by gel shift the affinities of fAR for AREs III, IV, and V. We found that although the affinity of ARDBD was Ͼ10-fold lower than that of AR, both proteins exhibited the same relative affinities for different individual sites (data not shown). Taken together, the data suggest that the ability of fAR to bind simultaneously to the AREs within the enhancer is an indication that AR binding is cooperative.
We further investigated how each individual ARE contributes to the recruitment of fAR using the enhancer mutants from Fig. 2. Fig. 9A is a DNase I footprint comparing binding of FIG. 4. Receptor specificity of the enhancer. A luciferase reporter containing the enhancer fragment was co-transfected with either AR or PR expression vector into BHK cells. Ligand treatment was with either 1 nM R1881 or 10 nM progesterone. RLU, relative light units.
ARDBD to individual mutants in AREs IIIA, III, IV, and V and a mutant removing all four sites (S-All). The individual mutants revealed a significant reduction of ARDBD binding only to the mutated site, whereas binding of ARDBD to other sites was unaffected. S-All, on the other hand, completely abolished binding to all four sites.
The binding profiles of fAR in Fig. 9B were far more complex. Each of the individual site mutants abolished tight binding to that site, although a weaker nonspecific protection over the mutated site was observed in some cases. For example, mutation of AREs III, IV, and V led to a decrease in protection of the site from 75 to 25%. Mutations of ARE IIIA or V had only a small quantitative effect on binding to sites III and IV. The mutation to site V did, however, abolish the additional footprint below site V again arguing for it relevance as an ARbinding site. In contrast, mutations in ARE III and IV clearly diminished cooperative binding to all sites at lower concentrations of fAR (Fig. 9B, compare intensity of protection in lanes 8 and 9, 11 and 12 with lanes 2 and 3). The ARE IV result was surprising because individually it is a much weaker site for AR than ARE III. This observation suggests that the intrinsic affinity of individual AREs does not necessarily correlate with their role in cooperative binding. However, when all four sites were mutagenized, AR binding to the enhancer was largely abolished (Fig. 9B, lanes 17-18). The disappearance of the footprint in the S-All mutant suggests that the partial protection seen in the individual site mutants, even over the mutant sites themselves, may be due to AR binding nonspecifically through interactions with AR bound specifically at intact neighboring sites.
To demonstrate further the physiological relevance of AR binding to this region, we employed nuclear extracts prepared from LNCaP cells with or without androgen treatment to perform footprinting assays on the enhancer. As shown in Fig. 10, the footprint of LNCaP extracts treated with R1881 was remarkably similar in some respects to that obtained with intact fAR. First, binding to ARE III and V is evidently hormone-dependent although the footprints are not as distinct as those generated by purified fAR (Fig. 10, compare lane 3 with lane 5). We also noticed hormone-dependent weak protection below site V as with the purified receptor in Fig. 8. However, although proteins in the extract apparently bound to sites IV and IIIA, and generated footprints similar to that of intact fAR, the binding was not dependent on hormone. Other factors in the nuclear extracts might bind to these sites, masking the effect of AR binding. Alternatively, the extracts from untreated cells still contain low amounts of AR, and it is plausible that the footprints are due to binding by AR. It is also evident, however, that there are other regions bound by proteins other than AR in the nuclear extract in a hormone-independent fashion (Fig. 10, see regions indicated by asterisks). Such footprints may represent factors that cooperate with AR in regulation of PSA gene expression. The relevance of these binding activities will require further effort to understand. DISCUSSION We have employed the PSA gene as a model to understand how AR binds to and activates transcription of a natural target gene expressed during prostate cancer. Although previous studies had revealed the existence of two ARE sites within the proximal PSA promoter, the effect of the proximal promoter alone on the activity and tissue specificity of PSA gene expression is minor (3,6,7,15). More recent studies led to the view that the major control region is an enhancer centered ϳ4.2 kb upstream of the PSA gene. Deletion mutagenesis around this region revealed that the androgen responsiveness of the enhancer is conferred by a 455-bp fragment bearing a single, moderate affinity ARE referred to as ARE III (centered at Ϫ4143). We subcloned this region upstream of a reporter gene and studied its response to AR. Our DNase I footprinting results, using both AR DNA binding domain and full-length AR, revealed that there are multiple AREs with varying but low affinities when compared with the consensus ARE. These sites were termed ARE IIIB, IIIA, III, IV, V, and VI and were We demonstrated that four of these nonconsensus AREs (IIIA, III, IV, and V), regardless of their low intrinsic affinity for AR, contribute significantly to the androgen-responsive transcriptional activity. Removal of these four sites strongly negated enhancer activity in LNCaP and in BHK cells, although the effects were less evident in BHK cells possibly due to high level overexpression of AR from the co-transfected effector plasmid or a lack of cell-specific factors. We found that fAR binds cooperatively to these sites in vitro. We have not excluded roles for the other two sites because mutagenesis of ARE IIIB and VI also decreases transcription in LNCaP cells (data not shown). It is clear from our study and others that mutation of many regions of the enhancer decreases its activity, and further work will be required to identify the proteins that bind these regions.
The large difference in the size of the protected region by AR versus ARDBD was surprising. This effect could have multiple explanations including the following: (i) the existence of additional AREs that bind cooperatively to intact AR (i.e. such as the weakly protected site below site V); (ii) nonspecific interactions of AR with the intervening DNA, again mediated by cooperative interactions; (iii) finally, either a general distortion of the enhancer fragment or the larger mass of AR versus ARDBD might simply lead to steric inhibition of DNase I cleavage between the sites. The mutagenesis data support the idea that multiple AR molecules are cooperatively binding both specifically and nonspecifically along the DNA to provide a stable androgen-responsive nucleoprotein complex. This oligomerization, however, is dependent upon the identified AREs because mutagenesis of the four main AREs abolishes both specific and nonspecific binding.
Cooperative binding of activators to multiple, adjacent low affinity sites and the resulting synergistic effects on gene expression are a common mechanism for ensuring specificity in the transcriptional response (38). In the case of the PSA enhancer the low affinity AREs may ensure that binding occurs only at a higher concentration of AR. In situ hybridization studies on rat tissues, for example, have shown that prostatic secretory epithelial cells are among two or three cell types expressing the highest levels of AR in the body (23). The combination of the low affinity AREs and the high concentrations of AR may therefore provide a partial explanation for the cell type specificity of PSA expression.
The presence of multiple AREs also augments the androgen responsiveness of the enhancer through synergy. Synergy is defined as the greater-than-additive transcriptional response to increasing numbers of bound activators. The current view is that synergy results from simultaneous interaction of multiple activators with the transcriptional machinery. These interactions lead to cooperative recruitment of the general machinery and the ensuing synergistic transcriptional response. In the case of the PSA enhancer, we imagine that simultaneous binding of multiple AR molecules is in part responsible for the synergistic androgen-dependent effect on PSA gene expression (8). This effect, when superimposed with a parallel synergistic effect of AR on AREs I and II within the proximal promoter (13) would, in principle, generate a sensitive and potent response of the gene to androgens.
The use of nonconsensus sites to regulate gene expression has been widely observed in nature. The consensus is an artificial amalgam of sites from various steroid receptor-responsive elements and, as such, binds and mediates response to several related steroid receptors (i.e. AR, GR, mineral corricoid receptor, and PR). The specificity for a particular receptor is likely conferred by contextual interactions between a single site or combinations of sites and the receptor. Results shown in Fig. 4 support this notion. In line with previous observations that AR and PR share near identical consensus hormone response elements, we demonstrated that both AR and PR were able to efficiently induce the activity of a construct containing four consensus AREs. However, when tested on a construct containing the PSA enhancer, only AR was able to stimulate the activity. This leads us to conclude that nonconsensus sequences may play a pivotal role in conferring receptor specificity. Studies on GR, for example, have also established that the DNA sequence of the site influences nuclear receptor activity, possibly through conformational changes in the ligand binding domain (49,50). Furthermore, natural low affinity sites that bind AR are commonplace. In one case, the ARE in the rat probasin gene was shown to display an inherent preference for AR over GR suggesting that nonconsensus changes in DNA sequence may represent a basic mechanism for conferring receptor specificity (35).
Contextual cooperative interactions involving interplay between multiple AREs and domains of AR also regulate specificity. In an elegant study that emphasized the complex nature of receptor DNA binding specificity, Robins and colleagues (36) studied the DNA binding properties of domain-swap chimeras of AR and GR. The target promoters included both a standard array of consensus steroid (hormone) response elements and the natural AR-responsive enhancer of the sex-limited protein (slp) gene of mice. Whereas the different chimeras bound to and functioned from standard steroid response elements (36), AR bound tightly and cooperatively to the nonconsensus slp enhancer AREs. The most efficient cooperativity required the AR DNA binding domain and its natural ligand and amino-termi- FIG. 8. Cooperative binding of fAR to the PSA enhancer. A, DNase I footprinting was performed on the wild-type enhancer fragment using increasing amounts of fAR (5,10,20,40, and 80 ng from lanes 2-6) and ARDBD (6, 12.5, 25, 50, and 100 ng from lanes [11][12][13][14][15]. Footprints are identified to the right. Two high doses of fAR and ARDBD were used to perform control footprint studies on the mutant enhancers. The DNase I-hypersensitive site on ARE V is indicated by an arrow.
nal domains. The ligand and amino-terminal domains could not be substituted by the analogous domains of GR, which appeared to, in fact, suppress specific enhancer binding. The study implied that contextual AR-AR and AR-DNA interactions occurring within the nucleoprotein complex on the slp enhancer are the key determinants of specificity. Such contextual protein-protein interactions may also explain the inability of PR to activate the PSA enhancer.
Although contextual, cooperative AR-AR interactions are likely to contribute to the specificity of PSA gene expression, other mechanisms must exist to ensure an accurate, cell-specific transcriptional response. Indeed, DNase I footprinting studies have identified binding sites for prostate-specific and ubiquitous transcription factors within the PSA enhancer and promoter (7,9,12). Henderson and colleagues (7) used gel shift analysis to identify several complexes, which were specific to extracts prepared from LNCaP cells. It was speculated that some of these complexes located between Ϫ4150 and Ϫ4000 were due to binding of AR, a finding supported by our data since this region contains site IIIA. However, other complexes within the enhancer appeared unique and may represent binding of prostate-specific factors. Indeed, Farmer and Freedman 2 recently identified a prostate-specific factor, which appears to bind adjacent to our ARE IV. Our footprint assay using crude LNCaP nuclear extracts also demonstrates that in addition to regions around ARE IV, other regions indicated by asterisks in Fig. 10 are protected by extracts from both R1881-treated and -untreated cells. Our data combined with others strongly suggest that additional factors exist to modulate the enhancer activity. These factors may function by interacting with AR in the context of the nucleoprotein complex, an issue we are currently investigating.
Mechanistic studies on nuclear receptors suggest that they stimulate transcription by direct or indirect communication with the general factors (51). Direct interactions between the general factors and several nuclear receptors including AR have been detected biochemically (52)(53)(54)(55). In addition to general factor interactions the current view is that the receptor also interacts with ancillary factors called co-activators to increase the recruitment of the transcriptional machinery. For example, CBP/p300 has been found to support activation by many nuclear receptors including AR (56). In addition, coactivators for individual nuclear receptors have been isolated using both biochemical assays and yeast two-hybrid screens. Examples include the TRAPs for the thyroid hormone receptor (45), ARA70 for the androgen receptor (57), SRC-1/p160 (58) and RIP 140 (59) for the progesterone, estrogen, and other steroid receptors, and GRIP1 for the glucocorticoid and androgen receptors (18). The yeast two-hybrid systems have identified individual interacting molecules, whereas biochemical studies have shown that several co-activators may simultaneously associate with a receptor to form a large multiprotein complex (56). For example, the thyroid receptor-associated proteins or TRAPs contain at least 9 polypeptides, which form a transcriptionally active complex with thyroid hormone receptor and play a role as positive co-activators in vitro (45). Vitamin D receptor has also recently been shown to interact with multiple polypeptides, the DRIPs, which are very similar in mass and may be identical to the TRAPs (60). Other systems have revealed that co-activators may also be gene-and site-specific (61,62). It is therefore likely that co-activators will play a role in gene-and tissue-specific transcription by AR. The ability to isolate and study AR from natural sources should now permit us to examine whether AR is naturally associated with many of the co-activators identified in two-hybrid screens and other analyses.