Evidence for an Anti-parallel Orientation of the Ligand-activated Human Androgen Receptor Dimer*

Domain interactions of the human androgen receptor (AR) dimer were investigated using a protein-protein interaction assay in which the NH 2 - and carboxyl-termi- nal regions of human AR were fused to the Saccharomyces cerevisiae GAL4 DNA-binding domain and herpes simplex virus VP16 transactivation domain to produce chimeric proteins. Transcriptional activation of a GAL4 luciferase reporter vector up to 100-fold was greater than Fos/Jun leucine zipper binding, indicating stable AR interaction between AR NH 2 -terminal residues 1–503 and steroid-binding domain residues 624–919 that was specific for and dependent on androgen binding to the steroid-binding domain and was inhibited by anti-an- drogen binding. Deletion mutagenesis within the NH 2 terminal region indicated transactivation domain resi- dues 142–337 were not required for dimerization, whereas deletions near the NH 2 terminus ( (cid:68) 14–150) or NH 2 -terminal to the DNA-binding domain ( (cid:68) 339–499) re- duced or eliminated the AR interaction, respectively. An NH 2 -/NH 2 -terminal interaction was also observed, but no interaction was detected between ligand-free or bound steroid-binding domains. The results indicate that high affinity androgen binding promotes interactions between the NH 2 -terminal and steroid-binding do- mains of human AR, raising the possibility of an andro-gen-induced anti-parallel AR dimer.

Domain interactions of the human androgen receptor (AR) dimer were investigated using a protein-protein interaction assay in which the NH 2 -and carboxyl-terminal regions of human AR were fused to the Saccharomyces cerevisiae GAL4 DNA-binding domain and herpes simplex virus VP16 transactivation domain to produce chimeric proteins. Transcriptional activation of a GAL4 luciferase reporter vector up to 100-fold was greater than Fos/Jun leucine zipper binding, indicating stable AR interaction between AR NH 2 -terminal residues 1-503 and steroid-binding domain residues 624 -919 that was specific for and dependent on androgen binding to the steroid-binding domain and was inhibited by anti-androgen binding. Deletion mutagenesis within the NH 2terminal region indicated transactivation domain residues 142-337 were not required for dimerization, whereas deletions near the NH 2 terminus (⌬14 -150) or NH 2 -terminal to the DNA-binding domain (⌬339 -499) reduced or eliminated the AR interaction, respectively. An NH 2 -/NH 2 -terminal interaction was also observed, but no interaction was detected between ligand-free or bound steroid-binding domains. The results indicate that high affinity androgen binding promotes interactions between the NH 2 -terminal and steroid-binding domains of human AR, raising the possibility of an androgen-induced anti-parallel AR dimer.
The androgen receptor (AR) 1 is a ligand-activated transcription factor that requires high affinity androgen binding to initiate a series of molecular events leading to specific gene activation required for male sex development. In its unliganded state, AR resides in the cytoplasm (1), where it rapidly degrades (2) and is regulated by a cytoplasmic dnaJ homologue (3). High affinity androgen binding slows AR degradation in a concentration-dependent manner, accounting at least in part for the physiological differences between the biologically active androgens (4). Androgen binding activates a bipartite nuclear targeting signal (5) and triggers receptor dimerization and acquisition of DNA binding that involves distal regions of the AR (6). Once activated, AR binds androgen response elements that resemble the simple consensus glucocorticoid response element (7) or more distinct, specific complex response elements (8 -10). Little is known, however, about transcription factors that interact with AR during gene activation or the role of AR phosphorylation (11). That AR is crucial for specific gene regulation required for male sex development is demonstrated by an abundance of AR gene mutations that result in different degrees of impaired male sex development characteristic of the androgen insensitivity syndrome (12)(13)(14)(15).
Steroid receptor dimerization is well documented (16 -19) and apparently does not have a strict requirement for ligand or DNA binding, particularly with the estrogen (ER) (20 -24) and progesterone receptors (25,26). The glucocorticoid receptor forms ligand-dependent homodimers independent of DNA binding (27). A mutant progesterone receptor lacking the steroid-binding domain fails to dimerize in solution but activates a reporter gene, suggesting that receptor dimerization mediated through the steroid-binding domain is not a requirement for DNA binding and that dimerization after DNA binding is mediated by the DNA-binding domain (28). Direct evidence for receptor dimerization was revealed in electron micrographs showing dumbbell shaped glucocorticoid receptor monomers with globular NH 2 -terminal and steroid-binding domains and four-leaf clover shaped dimers. It was not established, however, whether dimer orientation was parallel or anti-parallel (29). Additional protein-protein interactions and alterations in DNA structure are indicated by increased gene activation following cooperative dimer binding to tandem hormone response elements (30). ER dimerization involves hydrophobic interactions between the steroid-binding domains (31) and, for thyroid hormone receptor ␣, regions outside the steroid-binding domain (32). In addition, a leucine zipper-like structure in the thyroid receptor ligand-binding domain mediates heterodimerization with the retinoic acid receptor (33).
Numerous recent studies have taken advantage of a proteinprotein interaction assay developed originally in yeast (34) and later adapted for mammalian cells (35) that relies on the coexpression of two fusion proteins, each containing a protein or protein region coupled to a transcription factor functional domain. Stable protein-protein interactions bring together DNA binding and transactivation functions that regulate a reporter gene. Using this assay, we demonstrate an androgen-dependent interaction between the AR NH 2 -and carboxyl-terminal domains that raises the possibility of an anti-parallel oriented AR dimer.

EXPERIMENTAL PROCEDURES
Materials-The following reagents were purchased. Monkey kidney COS-1 and Chinese hamster ovary (CHO) cells were from the American Type Culture Collection (Rockville, MD); ␣ minimum essential medium and prestained protein molecular weight standards were from Life Technologies, Inc.; Dulbecco's modified essential medium with high glucose with or without phenol red was from JRH Biosciences; fetal calf serum was from Hyclone Laboratories (Logan, UT), and bovine calf serum was from Irvine Scientific (Santa Ana, CA); AR PG-21 rabbit polyclonal AR anti-peptide antiserum was from Affinity Bioreagents, Inc. (Neshanic Station, NJ); [ 3 H]methyltrienolone [17␣-methyl- 3 H]R1881, 80 Ci/mmol was from DuPont NEN; hydroxyflutamide was provided by R. O. Neri (Schering Corporation, Bloomfield, NJ); deep vent polymerase, T4 DNA ligase, and restriction enzymes were from New England Biolabs (Beverley, MA); Sequenase was from U. S. Biochemical Corp.; diethylaminoethyl dextran was from Pharmacia Biotech Inc.; D-luciferin was from Analytical Luminescence; cell lysis buffer was from Ligand Pharmaceuticals (San Diego, CA); phenylmethylsulfonyl fluoride and general chemicals were from Sigma; ECL Western blotting detection kit was from Amersham Corp.; and Immobilon-P was from Millipore (Bedford, MA).
Expression Plasmids-Eukaryotic expression vectors pGALO containing the DNA-binding domain of the Saccharomyces cerevisiae GAL4 protein (amino acid residues 1-147) and pNLVP coding for the transcriptional activation domain of the herpes simplex virus VP16 protein (residues 411-456) were previously described (36) and kindly provided by Gordon Tomaselli, Johns Hopkins University. Expression vectors contain the SV40 promoter, nuclear targeting signals, and a 3Ј multiple cloning site. Control plasmid pGAL-Fos contains the Fos leucine zipper region (amino acid residues 137-216) and pVP-Jun, the Jun leucine zipper region (residues 250 -334) as described previously (36) and kindly provided by Gordon Tomaselli and Chi V. Dang, Johns Hopkins University. Full-length human AR was cloned in frame in both vectors using polymerase chain reaction (PCR) mutagenesis to place an NdeI site at the initiation codon. A triple ligation reaction was performed with the NdeI/AflII NH 2 -terminal AR PCR fragment, the AflII/XbaI carboxyl-terminal AR fragment from pCMVhAR (1), and the NdeI/XbaI fragments of the pGALO or pNLVP expression vectors. pGAL-A1 (A1 comprising human AR amino acids 1-503) and pVP-A1 were constructed by deleting sequence coding for carboxyl-terminal residues 504 -919 using KpnI/XbaI followed by ligation of the filled ends. NH 2terminal deletion and insertion mutants pCMVhAR⌬14 -150, pCMVhAR⌬142-337, pCMVhAR⌬339 -499, and pCMVhARGln66 previously described (4) were used to clone the corresponding deletions into the pGAL-A1 and pVP-A1 vectors or into the full-length pGAL-hAR and pNLVP-hAR vectors using unique sites in the AR coding sequence to exchange the deletion fragment or, in the case of the ⌬14 -150 deletion, using PCR to include the NdeI site at the initiation codon. pGALD-H (human AR amino acids 624 -919; 624 -627 from exon C, 628 -919 exons D-H) was constructed using PCR to create an NdeI site at methionine 624 and was cloned into the NdeI/XbaI sites in the polylinker of pGALO. pGAL-LNCaPD-H was constructed in the same manner using pCMV-LNCaP AR (37) as template for PCR. Plasmids were amplified using DH5␣ E. coli cells and DNA isolated by banding in CsCl. Fusion genes created using PCR were sequenced using Sequenase to verify the absence of random mutations.
Cell Culture and Transient Transfections-CHO cells were maintained in ␣ minimum essential medium containing 10% fetal calf serum, 20 mM Hepes, pH 7.2, and penicillin and streptomycin. CHO cells at 50% confluence were transiently transfected with expression and reporter plasmids using DEAE-dextran as described previously (38). Unless otherwise stated for the protein-protein interaction assay 1 g each of the GAL4 and VP16 fusion protein plasmids and 5 g of reporter plasmid G5E1bLuc (provided by Gordon Tomaselli) were cotransfected into CHO cells (0.4 ϫ 10 6 cells/60-mm dish). The reporter vector had five tandem copies of the GAL4-binding sites, the E1bTATA promoter as described previously (39), and the luciferase coding sequence. Immediately after transfection, cells were placed in 0.2% serum-containing medium and 24 h later into serum-free, phenol red-free medium in the presence or the absence of steroid. Medium with or without hormone was replaced 3 h prior to harvest. Plates were washed with phosphatebuffered saline, harvested in 0.5 ml of cell lysis buffer, and frozen at Ϫ80°C until assayed for luciferase activity. Lysates were supplemented with 8 mM MgCl 2 , 1 mM dithiothreitol, and 0.4 mM phenylmethylsulfonyl fluoride. Relative light units were determined using a monolight 2010 Analytical Luminescence Laboratory luminometer after combining 0.4 ml of reaction buffer (15 mM MgCl 2 , 5 mM ATP, 0.5 mg/ml bovine serum albumin, and 15 mM glycylglycine, pH 7.8), 0.1 ml of cell lysate, and 0.1 ml of 1 mM D-luciferin injected automatically. All experiments were performed in duplicate and repeated at least three times.
Immunoblots and Binding Assays-Expression of GAL4-and VP16-AR fusion proteins was assessed on immunoblots and by whole cell steroid binding assays. Monkey kidney COS-1 cells were grown in Dulbecco's modified essential medium supplemented with 4.5 g/liter glucose and L-glutamine, 10% fetal calf serum, 20 mM Hepes, pH 7.2, and antibiotics. For immunoblot analysis, COS cells plated in 60-mm dishes (0.45 ϫ 10 6 cells/dish) were transiently transfected with 5 g of wild-type or mutant AR chimeric expression plasmids using DEAEdextran (38). Cells were maintained in Dulbecco's modified essential medium with 4.5 g/liter glucose and L-glutamine with 10% bovine calf serum and, for cells transfected with plasmids expressing the AR steroid-binding domain, supplemented with 0.1 M dihydrotestosterone. 48 h after transfection, cells were washed in phosphate-buffered saline and harvested in 0.1 ml of 2% SDS, 10% glycerol, and 10 mM Tris, pH 6.8. ␤-Mercaptoethanol (4%) and bromphenol blue (1%) were added, and the samples were boiled for 5 min. 20-l aliquots were analyzed in SDS containing 8% acrylamide gels as described previously using AR antipeptide rabbit antiserum AR32 (40) or AR PG-21 (41) and the ECL immunoblotting detection kit.
Androgen binding was determined in COS and CHO cells using a whole cell binding assay (38) for vectors comprised of exons D-H (residues 624 -919) (42). COS cells in 12-well (10 5 cells/well) or 35-mm (0.25 ϫ 10 6 cells/plate for Scatchard plot analysis) tissue culture plates were transiently transfected using DEAE-dextran with 1 g of GAL4 or VP16 fusion plasmids containing wild-type or LNCaP AR steroid-binding domain. Cells were maintained for 48 h in Dulbecco's modified essential medium with 4.5 g/liter glucose and L-glutamine media containing 10% bovine calf serum and labeled for 2 h at 37°C with 5 nM [ 3 H]R1881 in serum-free, phenol red-free medium in duplicate. For Scatchard analysis, cells were incubated with 0.05-4 nM [ 3 H]R1881 in the presence and the absence of a 100-fold excess of unlabeled R1881 for 2 h at 37°C. Aliquots of free [ 3 H]R1881 were taken, and the cells were washed in phosphate-buffered saline and collected in SDS sample buffer for scintillation counting. Nonspecific binding was determined by parallel incubations in the presence of a 100-fold excess unlabeled R1881. Labeling medium was removed, and cells were washed twice with phosphatebuffered saline and harvested in 0.2 ml of 2% SDS, 10% glycerol, and 10 mM Tris, pH 6.8. Radioactivity was determined by scintillation counting.

Expression of GAL4-AR and VP16-AR Fusion
Proteins-Intermolecular interactions within the AR dimer were investigated using a protein-protein interaction assay in mammalian cells. Fusion protein expression vectors contained partial AR sequence cloned in frame and carboxyl-terminal to the S. cerevisiae GAL4 DNA-binding domain or the herpes simplex virus VP16 protein transactivation domain (shown schematically in Fig. 1). Activation of the GAL4-luciferase reporter vector results when regions of AR interact stably after expression of the chimeras in CHO cells. GAL-Fos and VP-Jun vectors expressing Fos and Jun leucine zipper fusion proteins (36) were used as a positive control.
GALD-H and VPD-H comprise the GAL4 DNA-binding and VP16 transactivation domains, respectively, fused NH 2 -terminal to AR steroid-binding domain residues 624 -919. Epitopes for AR antibodies were not present in this region, so these fusion proteins were quantitated by ligand binding after expression in CHO and COS cells. GALD-H and VPD-H displayed high affinity (K d ϭ 0.11 Ϯ 0.02 nM) saturable binding of the synthetic androgen [ 3 H]R1881. Equilibrium binding affinity FIG. 1. Schematic representation of GALD-H and VP-A1 chimeric vectors. AR exons D-H coding for residues 624 -919 were inserted 3Ј into pGALO containing GAL4 protein DNA-binding domain amino acid residues 1-147. AR DNA sequence coding for residues 1-503 was cloned into pNLVP 3Ј of the herpes simplex virus VP16 protein transcriptional activation sequence coding for residues 411-456. Reporter vector G5E1bLuc has five GAL4-binding sites, the E1b promoter, and the luciferase coding sequence. was indistinguishable from full-length wild-type AR; however, these fragments would be expected to have increased ligand dissociation rates based on results with AR deletion mutants lacking the NH 2 -terminal domain (4). Expression levels were similar for the two constructs and were approximately 10-fold greater in COS cells than in CHO cells (data not shown).
Androgen-dependent Interaction between the AR Steroidbinding and NH 2 -terminal Domains-In the absence of androgen, luciferase activity was negligible when cotransfecting GALD-H with VP-A1 (Fig. 3). Treatment with 1 nM DHTinduced luciferase activity greater than that observed with the Fos/Jun leucine zipper chimeras relative to background activity (Fig. 3). Similar results were obtained using COS cells for transfection (data not shown); however, CHO cells were used in subsequent experiments due to lower background activity. No significant transactivation was observed in CHO cells in the presence of androgen when either plasmid was transfected with the other parent plasmid lacking AR sequence (Fig. 3).
The results indicate an androgen-dependent stable interaction between the AR steroid-binding domain and the NH 2 -terminal region.
Only a 2-fold induction of luciferase activity was observed using the reciprocal chimeras GAL-A1 and VPD-H (Fig. 3), suggesting a preferential orientation for fusion protein interaction. GAL-A1 alone activated luciferase expression (Fig. 3) presumably resulting from linking the AR transactivation domain to the GAL-4 DNA-binding domain, thus increasing background activity in the NH 2 -/carboxyl-terminal interaction using this construct. Cotransfecting GALD-H and VPD-H both containing the AR steroid-binding domain failed to activate the reporter vector in the presence or absence of androgen (Fig. 3). Transfecting GAL-A1 with VP-A1 increased activation 4 -5-fold over the activity of GAL-A1 alone (Fig. 3), indicating a ligandindependent interaction between the AR NH 2 -terminal domains. The high transcriptional activity induced by GAL-A1 with VP-A1 was likely due in part to the presence of three transactivation domains, two from the AR NH 2 -terminal domains and one from VP-16. The lower fold induction by the NH 2 -/NH 2 -terminal interaction may therefore reflect a weaker interaction than that observed for the androgen-induced NH 2 -/carboxyl-terminal interaction (4 -5-versus 59-fold; Fig. 3).
Steroid Specificity and Concentration Dependence-Steroid specificity of the GALD-H and VP-A1 interaction was investigated by determining transcriptional activation in the presence of androgens, the antiandrogens hydroxyflutamide and cyproterone acetate, and other steroids. The strongest transcriptional activation was observed at 1 nM DHT and between 1 (Fig.  4) and 10 nM R1881 or testosterone. Hydroxyflutamide up to 1 M failed to activate luciferase activity (Fig. 4).
Because hydroxyflutamide is a potent antiandrogen, we investigated whether it would disrupt the NH 2 -/carboxyl-terminal interaction in the presence of androgen. Increasing concentrations of hydroxyflutamide between 0.2-1 M inhibited androgen-induced gene activation (Fig. 4). Estradiol, progesterone, and cyproterone acetate failed to induce luciferase activity and at 0.5 M inhibited transcriptional activation induced by 1 nM DHT (Fig. 5). The results indicate that the NH 2 -/carboxylterminal interaction induced by androgens is blocked by moderate affinity ligands such as hydroxyflutamide, progesterone, and estradiol, paralleling the activation and inhibition properties of these ligands with wild-type, full-length AR.
Activity of the LNCaP AR Mutant-It was shown previously that the mutant AR in the androgen-dependent human prostate cancer cell line, LNCaP, contains a single base mutation in the steroid-binding domain that changes threonine 877 to alanine that results in increased affinity for hydroxyflutamide with concomitantly increased agonist activity (43)(44)(45). We inserted the LNCaP Thr-877 3 Ala mutation into GALD-H and tested for the NH 2 -terminal/steroid-binding domain interaction. Whole cell binding assays using [ 3 H]R1881 showed similar binding affinities for GALD-H and GAL-LNCaPD-H (data not shown). In the presence of 1 nM DHT, greater than 100-fold induction of luciferase activity indicated an NH 2 -/carboxyl-terminal interaction similar to wild-type AR (Fig. 4). At 1 M, hydroxyflutamide-induced luciferase activity almost 10-fold and was a less active inhibitor of androgen-induced complex formation, with luciferase activity remaining approximately 10-fold (Fig. 4). This result is in agreement with the known agonist activity of hydroxyflutamide acquired by the LNCaP mutation (43)(44)(45). Estradiol and progesterone similarly induced luciferase activity by GAL-LNCaPD-H and VP-A1, whereas no detectable activity was observed with these steroids with the wild-type GALD-H and VP-A1 chimeras (Fig. 5). Cyproterone acetate did not induce luciferase activity of the wild-type or LNCaP mutant chimeras. All three ligands, estradiol, progesterone, and cyproterone acetate, at 0.5 M were less active inhibitors of DHT-induced luciferase activity of GAL-LNCaPD-H and VP-A1 relative to wild-type GALD-H and VP-A1 chimeras (Fig. 5). Thus, although the magnitude of the interaction was 5-10-fold less than that observed with DHT, the NH 2 -/carboxyl-terminal interaction induced by hydroxyflutamide, estradiol, and progesterone with the threonine 877 to alanine mutation correlated with the agonist activities of these ligands with the full-length LNCaP mutant AR. for cotransfection with GALD-H, GAL-LNCaPD-H, and GAL-A1. Expression levels of the deletion mutants were similar as determined by immunoblotting (see Fig. 2). Deletion of the AR transactivation region (VP-A1⌬142-337) resulted in 14-fold stimulation of luciferase activity (Fig. 6), representing a significant decrease relative to the wild-type sequence. Because the AR transactivation domain likely contributes to luciferase induction, it is possible that the decrease reflects deletion of the AR activating region rather than a decrease in dimerization. Deletion immediately NH 2 -terminal to the DNA-binding domain (VP-A1⌬339 -499) abolished luciferase activity (Fig. 6) even after a 5-fold increase in plasmid concentration (data not shown). Deletion near the NH 2 terminus (VP-A1⌬14 -150) also inhibited transcriptional activation, with 2-5-fold induction of luciferase activity using a 5-fold higher plasmid DNA concentration (Fig. 6). Expansion of the NH 2 -terminal polymorphic glutamine repeat from 21 (residues 58 -78 in wild-type AR) to 66 glutamine residues (VP-A1Gln66) had little effect on transcriptional activation of luciferase (Fig. 6). Expansion of the glutamine repeat is associated with spinal/bulbar muscular atrophy (Kennedy's disease) (46), and thus, the amplified repeat does not interfere with this aspect of AR dimerization, in agreement with its wild-type level of activation. 2 Results using the NH 2 -terminal deletion and insertion mutants with the GAL-LNCaPD-H mutant were essentially identical to those with wild-type GALD-H fusion protein (data not shown).
The NH 2 -/NH 2 -terminal interaction was investigated using the VP-A1 deletion mutants described above. Inhibition was observed after deleting transactivation domain residues 142-337 and to a lower extent by deletion of residues 339 -499 or 14 -150 (Fig. 7). Expanding the glutamine repeat to 66 residues as described above slightly enhanced the interaction. Thus, different regions appear to be involved in the NH 2 -/NH 2 -terminal interaction than in NH 2 -terminal interaction with the androgen-bound steroid-binding domain.
Smaller Domains within the Interacting Regions-Insertion of shorter regions of AR coding sequence into GAL4 and VP16, such as NH 2  activation. The results suggest that the sites of interaction localize to larger domains that may involve concerted actions within the NH 2 -terminal and entire steroid-binding domains, with the latter required for androgen binding.

DISCUSSION
The objective of the present study was to establish whether a direct interaction occurs between the NH 2 -terminal and steroid-binding domains in androgen-induced AR dimer formation. Domain interactions were analyzed by reporter gene activation using fusion proteins that linked the NH 2 -and carboxyl-terminal regions of human AR to the GAL4 DNAbinding or VP16 transactivation domains. The results support previous evidence that in vivo dimerization of human AR is mediated through direct intermolecular interactions between the androgen-bound steroid-binding domain and the NH 2 -terminal region. The dependence on androgen binding and inhibition by an antiandrogen and other steroids parallels properties of native AR and raises the possibility of an androgenactivated, anti-parallel AR dimer.
Although the present data do not rule out that the ligandinduced NH 2 -/carboxyl-terminal interaction occurs intramolecularly, previous studies using baculovirus-expressed AR fragments support a ligand-induced intermolecular interac-tion. Androgen-dependent dimerization was observed between NH 2 -terminal plus DNA-binding domain and DNA plus steroid-binding domain fragments (6). However, as the DNA-binding domain is implicated in receptor dimerization through the so-called D box region (47)(48)(49) with the other monomer (50,51), the DNA-binding domain could have accounted for the observed AR dimerization. In the present study, the AR DNAbinding domain, which itself dimerizes, 3 was excluded from the chimeric proteins, indicating an additional dimerization interface between the NH 2 -terminal and androgen-bound steroidbinding domain of AR. Further support for AR NH 2 -and carboxyl-terminal interactions comes from kinetic studies where the dissociation rate of bound androgen slows about 5-fold by the presence of the NH 2 -terminal domain despite no change in equilibrium dissociation constant (4). Crystallographic data of glucocorticoid receptor/DNA interactions (50) and the asymmetric dimer proposed for the vitamin D receptor (53) suggest a ligand-activated anti-parallel dimer may be the active conformation for other members of the steroid receptor family.
The two regions of the AR NH 2 -terminal domain required most for carboxyl-terminal interaction were immediately NH 2 - terminal to the DNA-binding domain and near the NH 2 terminus. Lack of a direct role of the more centrally positioned transactivation domain might allow this region to remain accessible for transcription factor interaction. In our unpublished studies and the work of others (54), deletion of the transactivation domain creates a strong dominant negative AR inhibitor, suggesting that loss of the transactivation domain does not interfere with receptor dimerization. It is interesting, therefore that this region or the region NH 2 -terminal to the DNA-binding domain (residues 142-337 and 339 -499, respectively) may be involved in an interaction between the NH 2 -terminal domains and may reflect an association that occurs in the unliganded receptor that could contribute to suppression of activation in the absence of ligand.
The affinity of the NH 2 -/carboxyl-terminal interaction appears to be similar to that observed for Fos-Jun leucine zipper binding. When the Fos/Jun leucine zipper regions were fused to progesterone receptor, agonist-induced progesterone receptor dimerization persisted through the receptor dimerization domain, suggesting that ligand-induced receptor dimerization was of equal or greater affinity than the Fos/Jun leucine zipper interaction (55). Leucine zipper motifs are often involved in transcription factor dimerization resulting in efficient DNA binding (56). A heptad repeat of hydrophobic amino acid residues in the steroid-binding domain of mouse ER resembles a leucine zipper, is conserved among the family of steroid recep-tors, and is implicated in dimerization and high affinity estrogen binding (20). Like ER (57), other steroid receptors appear to have two dimerization interfaces: a constitutive region in the DNA-binding domain and a stronger, hormone-dependent region in the hormone-binding domain that may be involved in stable dimer formation required for high affinity DNA binding. The carboxyl-terminal end of the thyroid hormone receptor was also implicated in receptor dimerization (58,59).
Human AR and ER differ in the length of their NH 2 -terminal domains, i.e. 559 amino acid residues in AR versus 185 in ER. It is noteworthy, therefore, that a transcriptionally inactive AR deletion mutant AR507-919, lacking all but 52 NH 2 -terminal amino acid residues (4), dimerizes and binds DNA independent of ligand binding (6) as observed with full-length ER (20 -23) but not full-length AR (6). The androgen-independent dimerization of this AR deletion mutant suggests two forms of DNAbinding homodimers: one for the AR deletion mutant AR507-919 and perhaps ligand free ER and another for androgenbound full-length AR and perhaps ligand-bound ER. A parallel dimer capable of binding DNA may form constitutively through interactions between the DNA-binding domains if no extended NH 2 -terminal region interferes, in the case of ER and the AR deletion mutant. However, the active configuration requiring ligand binding for full-length AR might be anti-parallel and depends on the presence of the NH 2 -terminal domain. This hypothesis is supported by studies using glucocorticoid recep- tor deletion mutants, where deletion of the NH 2 -terminal domain changed the contact points within the dimer in crosslinking studies and reduced the specificity of DNA binding (60).
Androgen-induced conformational effects on full-length AR that might establish the anti-parallel dimer may be required for DNA binding that results in transcriptional activation.  6 were cotransfected with GAL-A1, and luciferase activity was determined as described in the legend to Fig. 6. The data shown are representative of three experiments.