Differences in DNA Binding Characteristics of the Androgen and Glucocorticoid Receptors Can Determine Hormone-specific Responses*

The basis for specificity of gene regulation by steroid hormone receptors remains an important problem in the study of steroid hormone action. One possible mechanism for steroid specificity is the difference in DNA binding characteristics of the receptors, although they share a high homology in their DNA-binding domains. Indeed, the androgen-specific expression of, for exam-ple, the probasin (PB) gene can be explained by the presence of an androgen response element (ARE) in its promoter (PB-ARE-2), specifically recognized by the androgen and not by the glucocorticoid receptor. Three residues in the DNA-binding domain of the AR were identified as main determinants for its high affinity for the PB-ARE-2. In addition, the direct repeat nature of this ARE seems to prohibit high affinity binding by the glucocorticoid receptor. This is confirmed by the fact that several imperfect direct repeats of the 5 * -TGT-TCT-3 * core recognition sequence are recognized by the androgen receptor and not by the glucocorticoid receptor. Up to now, only differences between the androgen and glucocorticoid receptor in the transcription activation functions were invoked to explain the specificity of their genomic actions. In the present study, we describe the influence of the DNA-binding domain on the specificity of androgen action. The novelty of our working hypothesis resides in the demonstration of the capacity of the AR-DNA-binding domain to recognize elements with a direct repeat structure. Steroid hormones are important endocrine messengers that

The basis for specificity of gene regulation by steroid hormone receptors remains an important problem in the study of steroid hormone action. One possible mechanism for steroid specificity is the difference in DNA binding characteristics of the receptors, although they share a high homology in their DNA-binding domains. Indeed, the androgen-specific expression of, for example, the probasin (PB) gene can be explained by the presence of an androgen response element (ARE) in its promoter (PB-ARE-2), specifically recognized by the androgen and not by the glucocorticoid receptor. Three residues in the DNA-binding domain of the AR were identified as main determinants for its high affinity for the PB-ARE-2. In addition, the direct repeat nature of this ARE seems to prohibit high affinity binding by the glucocorticoid receptor. This is confirmed by the fact that several imperfect direct repeats of the 5-TGT-TCT-3 core recognition sequence are recognized by the androgen receptor and not by the glucocorticoid receptor. Up to now, only differences between the androgen and glucocorticoid receptor in the transcription activation functions were invoked to explain the specificity of their genomic actions. In the present study, we describe the influence of the DNA-binding domain on the specificity of androgen action. The novelty of our working hypothesis resides in the demonstration of the capacity of the AR-DNA-binding domain to recognize elements with a direct repeat structure.
Steroid hormones are important endocrine messengers that activate their receptors, which translocate to the cell nucleus and regulate gene expression mainly after interaction with DNA sequences, called response elements (1,2). The steroid receptors are a subfamily of the nuclear receptor superfamily, a large group of structurally homologous transcription factors. A problem with the explanation of the specificity of these hormone responses arose when several studies pointed out that the class I receptors (androgen receptor (AR), 1 glucocorticoid receptor (GR), progesterone receptor, and mineralocorticoid receptor) have identical consensus response elements (3,4) and that their DNA-binding domains were highly conserved (5). This contrasts with the fact that the in vivo expression of several genes is specifically controlled by only one steroid hormone (6).
Several possible mechanisms have been described to explain the steroid specificity of transcriptional control, e.g. steroid metabolism, tissue-specific receptor presence (7), influence of coactivator complexes (8), and chromatin structure (9,10). In addition, more recent reports indicate that the AR on the one hand and the GR, progesterone receptor, and mineralocorticoid receptor on the other exhibit different DNA binding characteristics (11)(12)(13)(14)(15). One AR-specific response element was found in the promoter of the rat probasin gene (PB-ARE-2) (12,16,17). Probasin is an androgen-regulated protein exclusively expressed in the dorsolateral epithelium of the prostate (18). Two cis-acting androgen response elements (PB-ARE-1 and PB-ARE-2) were identified in the promoter and were shown to be necessary for its androgen regulation.
The core DNA-binding domain (DBD) of the nuclear receptors is composed of two zinc finger modules (19). The N-terminal zinc finger is involved in specific DNA interaction, whereas the C-terminal zinc finger mainly provides DNA-dependent dimerizations (20). Our earlier results indicated that residues in the second zinc finger and a C-terminal extension (CTE) of 12 amino acids determine the difference in the PB-ARE-2 binding between the AR and the GR (21). This was the first clear indication of a direct involvement of the CTE residues in the specificity of DNA binding for steroid receptors. The CTE described here overlaps with the T-box described for other members of the nuclear receptor family (e.g. 9-cis retinoic acid receptor (RXR), thyroid receptor (TR), and nerve growth factorinducible protein B) (22)(23)(24). Structural studies showed that in these receptors the T-box residues form an ␣-helix, which interacts with the DNA phosphate backbone and which is also involved in the DNA-dependent heterodimerization. Structural studies of the estrogen receptor DBD and GR-DBD, however, did not clarify the involvement of this region in DNA binding (20,(25)(26)(27)(28). The RXR and the TR bind in a "head-to-tail" orientation on a response element with a direct repeat structure (half-site, 5Ј-AGGTCA-3Ј), whereas the steroid receptors bind in a "head-to-head" orientation on a response element organized as an inverted repeat (half-site estrogen receptor, 5Ј-AGGTCA-3Ј; class I receptors, 5Ј-TGTTCT-3Ј).
In the present study, the difference in DNA binding between the AR and the GR is analyzed in detail by determining the identity of the amino acids involved and the characteristics of the response element responsible for the exclusion of the GR from binding.
Plasmids-The cDNA encoding the rat AR was described by Chang et al. (29), and that for the rat GR was described by Hollenberg et al. (30). fAGA is derived from the full-size rat AR by swapping the AR-DBD for that of the rat GR (AR amino acids 537-619) (21). The receptor constructs pCMV5-mAR, pCMV5-rat GR, pCMV5-AGA, and pCMV5-GAG were a kind gift from Dr. D. Robbins (University of Michigan Medical School). In the latter constructs (AGA and GAG), the DBD is exchanged between the AR and the GR (31).
Purification of the DNA-binding Domains as Glutathione S-transferase Fusion Proteins-For the prokaryotic expression and purification of the steroid receptor DBDs, the corresponding receptor cDNA-fragments were amplified by polymerase chain reaction and subsequently cloned in the pGEX-2TK expression vector (Amersham Pharmacia Biotech). The fragments were expressed as glutathione S-transferase fusion proteins in the E. coli BL21 strain. After thrombin cleavage to remove the glutathione S-transferase fusion partner, the DBDs still contain a short foreign amino acid stretch at the N-terminal end (GSRRASV) as well as at the C-terminal end (IHRD). The AR1 construct consists of rat AR-DBD amino acids 533-637; the GR1 construct comprises the corresponding rat GR-DBD amino acids 432-533. The different mutated GR1 and AR1 constructs generated are described in Fig. 2. The different receptor fragments were purified as described by Schoenmakers et al. (21).
Preparation of COS-7 Nuclear Extracts Containing Full-size Receptor Constructs-The protocol for the preparation of nuclear extracts has been described by Andrews and Faller (32) and was used with some modifications. Briefly, 10 6 COS-7 cells were plated in 10 cm Petri dishes and transfected with 2.5 g of expression plasmid. The cells were incubated (1 h) before harvesting with 10 Ϫ5 M hormone, after which the medium was removed, and the cells were washed twice with 3 ml of ice-cold phosphate-buffered saline. The cells were collected in 1.5 ml of ice-cold phosphate-buffered saline per dish, transferred to an Eppendorf tube, and pelleted by centrifugation (10 s). The phosphate-buffered saline was removed, and the cells were resuspended in 400 l of ice-cold buffer containing 10 mM Hepes⅐KOH, pH 7.9, 1.5 mM MgCl 2 , 10 mM KCl, 0.5 mM dithiothreitol, and 0.2 mM phenylmethylsulfonyl fluoride. After 10 min of incubation on ice, the mixture was vortexed (30 s), and the nuclei were collected by a short spin in a microcentrifuge. The supernatant was removed, and the nuclei were resuspended in 100 l of ice-cold high salt buffer: 20 mM Hepes⅐KOH, pH 7.9, 1.5 mM MgCl 2 , 420 mM KCl, 25% glycerol, 0.5 mM dithiothreitol, 0.2 mM phenylmethylsulfonyl fluoride. After 20 min of incubation on ice and a short vortexing (10 s), the extracts were cleared by centrifugation (12,000 ϫ g) at 4°C for 2 min. The supernatant was then frozen in liquid nitrogen and stored at Ϫ80°C. Western blotting was performed on these extracts as described (33).
Gel Shift Assay-The double-stranded oligonucleotides containing the ARE sequences (described in Table II) were labeled with [␣-32 P]dATP by a fill-in reaction by the Klenow fragment of DNApolymerase to a specific activity of 5000 cpm/fmol. The response elements with a direct repeat sequence organization are derived from the DR element 5Ј-AGCTTTCATTGTTCTTGATGTTCTGAATGAGCT-3Ј (Table II).
The dissociation constants were determined by means of gel shift assays as described by Schoenmakers et al. (21). In short, a constant amount of labeled oligonucleotide was incubated with increasing amounts of purified receptor fragment until complete retardation of the probe was obtained. The K S value of each construct was calculated from the percentage of retarded probe with the formula of Hill kinetics based on at least three independent assays.
For the gel shift assays with the cell extracts containing the full-size receptor, constant amounts (20,000 cpm) of labeled double-stranded oligonucleotides were incubated (20 min on ice) with equal amounts of nuclear extracts in 20 l of binding buffer (10 mM Hepes⅐KOH, pH 7.9, 2.5 mM MgCl 2 , 0.05 mM EDTA, 10% glycerol, 1 g of poly(dI-dC), 0.05% Triton X-100, l mM dithiothreitol). Subsequently, free and bound probe were separated by electrophoresis for 120 min at 4 V/cm in a 4% nondenaturing polyacrylamide gel. In competition gel shift assays, 300fold excess of cold C3(1) ARE was incubated on ice with the receptor mixture for 10 min prior to the addition of the labeled oligonucleotide. Specific antibodies against the N-terminal part of the AR (34) and the GR (kindly provided by Prof. Gustafsson, Karolinska Institute, Stockholm, Sweden) were used as described (35) to obtain a supershift.
Transfection Assays-COS-7 cells were obtained from the American Type Culture Collection and were routinely maintained in Dulbecco's modified Eagle's medium (DMEM) (Life Technologies) containing 1 g/liter glucose, supplemented with antibiotics (penicillin, 100 IU/ml; streptomycin, 100 G/ml; Life Technologies) and 10% heat-inactivated fetal bovine serum. For the transient transfection assays, the cells were cultured in DMEM containing 5% dextran-coated, charcoal-stripped fetal bovine serum (DCC). Reporter plasmids and the expression vectors for the receptor constructs were co-transfected by means of the FuGENE 6 transfection reagent (Roche Molecular Biochemicals). A ␤-galactosidase expression plasmid (CMV-␤-galactosidase, Stratagene) was used as internal control of the transfection efficiency. For the transfection assay 15 ng of reporter plasmid, together with 5 ng of receptor expression vector, 20 ng of CMV-␤-galactosidase, and 0.1 g of carrier DNA (pGEM-7), were mixed with 20 l of DMEM containing 0.2 l of FuGENE transfection reagent. The mixture was incubated for 30 min at room temperature and then added dropwise onto the cells (96-well culture dish containing 5 ϫ 10 3 cells in 200 l of DMEM). After overnight incubation at 37°C, the medium was replaced by DMEM containing 5% DCC, and the cells were grown in the absence or presence of 1 nM R1881, dexamethasone, aldosterone, or progesterone for 48 h. Luciferase activity was measured with the luciferase assay system from Promega, and ␤-galactosidase activity was measured with the detection system from CLONTECH.

The AR-DBD Is the Main Determinant for the Specific PB-ARE-2 Recognition-
The difference in PB-ARE-2 recognition between the AR and GR was first observed using only their DBDs (named AR1 and GR1) (12,21). To analyze the influence of the other receptor domains on the difference in DNA binding, chimerical full-size AR and GR constructs with swapped DBDs (AGA, GR-DBD residues 449 -533; GAG, AR-DBD residues 551-637) (31), were used. The DNA binding capacities of these chimerical receptors were compared with these of the AR and GR in gel shift assays with nuclear extracts of transfected COS-7 cells. The C3(1) ARE was recognized by all receptor constructs with similar affinity (Fig. 1A). The specificity of DNA binding was verified by competition assays using cold C3(1) ARE and by the appearance of a supershifted complex in the presence of a specific anti-AR or anti-GR antibody. The PB-ARE-2 was bound with high affinity by both AR-DBDcontaining proteins AR and GAG. In contrast, neither the GR nor the AGA bound this ARE with high affinity, albeit that the antibodies induce a small amount of supershifted binding complexes.
In transient cotransfection assays, the pC3AREluc reporter construct is induced 4 -5-fold by all the chimerical receptors, indicating the functionality of the receptor proteins (Fig. 1B). Most remarkable is the transactivation of the PB-ARE-2-con-taining reporter construct by GAG to a comparable level as the wild type AR, whereas the wild type GR does not enhance the transcription. The AGA chimera is still able to transactivate pPBARE2luc in response to androgens, although the induction factor (2.5) is very low. Western blot analysis of the chimeras (Fig. 1C) shows a slightly higher level of expression of AGA and GAG compared with respectively AR and GR.
Identification of the Residues Determining the Difference in DNA Sequence Specificity between the AR and GR-In our earlier study, we demonstrated that the first zinc finger of the AR is not involved in the specific recognition of the PB-ARE-2 (21). However, experiments with the chimerical DBD constructs G/A and A/G ( Fig. 2 and Ref. 21) demonstrate that for the AR, residues of the hinge region as well as of the Nterminal part of the second zinc finger are involved. We therefore analyzed the importance of all residues that differ between the AR-DBD and the GR-DBD in these two regions.
We have already described the necessity of a C-terminal extension of 12 amino acids for specific and high affinity binding of the AR to the PB-ARE-2 (21). With the constructs AR28.1 and 28.2 (Table I), we could confirm the involvement of the 12 CTE residues in high affinity DNA binding. These constructs are the result of a deletion of residues forming the nuclear localization signal, which partially overlaps with the CTE (36). Note that AR28.1 and 28.2, kindly provided by Dr. A. O. Brinkmann, are derived from the human AR, which differs from the rat AR at three residues in the extreme C terminus of the receptor fragment. Deletion constructs showed, however, that deletion of this part of the protein had no influence on the DNA binding by the rat AR-DBD (21). Six of the 12 CTE amino acids differ between the AR and the GR. Systematic mutations of the AR-specific residues to the GR homologues resulted in a num- ber of AR1 mutants, schematically illustrated in Table I. The apparent dissociation constants of these constructs for the PB-ARE-2 and C3(1) ARE indicate that mainly the mutation of Gly-610 (and to a lower extent, mutation of Leu-617) abrogates the PB-ARE-2 binding.
In the C-terminal part of the second zinc finger of the GR-DBD, an ␣-helix (amino acids 492-503) was described (20,26). In the ARHm construct, all the nonhomologous residues of the ␣-helical structure in the C-terminal side in the AR-DBD (amino acids 594 -605) were swapped for those of the GR (Fig. 2). This construct had DNA binding capacities similar to those of the wild type AR1, indicating that the putative helix in the C-terminal region of the second zinc finger had no influence on the difference in DNA sequence recognition between the AR and the GR. However, for high affinity DNA binding to PB-ARE-2, a correct conformation of the helix is necessary. This is illustrated by the mutation of Leu-599 of the AR to the GR homologue Tyr-497 (AR599Y) (Fig. 2). In the x-ray structure of the GR-DBD, Tyr-497 and Leu-501 are oriented toward each other. When a tyrosine is present at both places, as in AR599Y, steric hindrance might prevent the correct folding of the helix, influencing the overall DBD-conformation. We tested this hypothesis by combining the Leu-599 to Tyr with the Tyr-603 to Leu mutation in AR1. As expected, the AR599Y/603L regained its affinity for the PB-ARE-2 leading to the conclusion that Tyr-497 and Leu-501 in the GR are two complementary residues. This is in contrast with the mutated AR construct, AR599Y/610E, which has lost its affinity for PB-ARE-2. The exchange of another nonconserved residue in the putative ␣-helix of AR1, namely Glu-604, for that of the GR homologue glutamine had no effect on the DNA specificity.
To identify the other residues in the second zinc finger necessary for the AR-specific PB-ARE-2 binding, we introduced the GR homologues in the AR1. However, none of the mutated AR1 constructs displayed a dramatic decrease in affinity for PB-ARE-2 or C3(1) ARE (data not shown). Therefore, we took an alternative approach based on the observation that the G/A chimera, containing only part of the second zinc finger and the hinge region of the AR, has a low affinity for the PB-ARE-2. In an attempt to restore the binding to the PB-ARE-2, mutations were introduced in G/A (Fig. 2). The apparent dissociation constants of these constructs indicate that Thr-585 (position 483 in GR1) might be involved in high affinity binding of the AR to PB-ARE-2. Table II, synthetic oligonucleotides containing either inverted or direct repeats of the left or right half-site of the PB-ARE-2 are listed (PB2-IR1, PB2-IR2, and PB2-DR2). The apparent dissociation constants of DBD binding to these synthetic elements demonstrate that the left half-site of the PB-ARE-2 sequence determines the AR-specific binding and prevents GR-binding. PB2-IR1, which is recognized with high affinity by both DBDs (Fig. 3), can be considered as an inverted as well as a direct repeat. The PB-IR2 and PB-DR2 are not recognized by the GR-DBD as a dimer, whereas the AR-DBD recognizes these sequences, although with lower affinity (Fig. 3 and Table II). Gel shift assays with COS-7 nuclear extracts containing the full-size AR or GR confirm the results of the DBD constructs (data not shown).
Direct Repeat Binding by Chimerical AR and GR Constructs-The importance of the AR-DBD for the recognition of response elements with a direct repeat configuration was further illustrated with the chimerical full-size AR and GR, GAG and fAGA. From gel shift assays with COS-7 nuclear extracts containing these chimeras, it became clear that it is mainly the AR-DBD that determines the high affinity binding to direct repeat elements (Fig. 5). Indeed, the fAGA construct binds the C3(1) ARE well but has a low affinity for the direct repeat sequences. The GAG construct, however, is able to bind the PB-ARE-2 and the elements DRI and DR.
The AR-DBD Dimer Involves Only One CTE for High Affinity DNA Binding-The DR sequence recognition by AR1 and the different functions of the CTEs between the AR and the GR in specific and high affinity DNA binding led to the hypothesis of a head-to-tail configuration of the AR on DR sequences, rather than a head-to-head configuration as seen for the GR-DBD (20). This would imply that only one of the two DBDs needs to have its CTE making contacts with the other DBD. We tested this hypothesis by analyzing the DNA binding of mixtures of AR1 with AR(607) (AR-DBD lacking the CTE) or AR(611) (AR-DBD with 4 CTE residues), as well as GR1 with GR(505) (GR-DBD lacking the CTE) or GR(509) (GR-DBD with four CTE residues) ( Fig. 6). At low concentrations, at which AR(607) did not bind the C3(1) ARE by itself, it was able to bind as a heterodimer with AR1. In addition, although AR(611) can bind as a dimer to C3(1) ARE, it preferentially binds as a heterodimer with AR1. Under similar conditions, GR1/GR(505) heterodimers are not detectable on the C3(1) ARE, and only a very weak GR1/ GR(509) heterodimeric complex appears. For PB-ARE-2, only the AR(611) forms a heterodimeric complex with AR1 in conditions under which neither AR(607) nor AR(611) alone can bind PB-ARE-2.

DISCUSSION
The Androgen Receptor DNA-binding Domain Determines the Affinity for PB-ARE-2-The difference in binding of PB-ARE-2 between the AR and the GR has been demonstrated in gel shift assays, as well as in transient transfection assays (21). The involvement of the different androgen receptor domains on the specific binding to the PB-ARE-2 was analyzed by the use of chimerical receptor constructs AGA and GAG. In gel shift assays, we demonstrated that the presence of the AR-DBD determines the high affinity for the PB-ARE-2 (Fig. 1). In transfection assays, the GAG, in contrast with the wild type GR, was also able to transactivate a PB-ARE-2-based reporter construct. An unexpected result was seen for the AGA con-  (Table II). Free probe was separated from the DNA-protein complexes by nondenaturating polyacrylamide gel electrophoresis as described under "Experimental Procedures." The monomeric and dimeric complexes are indicated with arrowheads.

FIG. 4. Gel shift assays of the C3(1) ARE, PB-ARE-2, and synthetic
AREs with the full-size AR and GR. DNA elements, as described in Table II, were analyzed for binding by AR and GR, as explained in Fig. 1A. Lanes 1 contain no nuclear extract; in lanes 2-4, equal amounts of COS-7 nuclear extracts containing full-size AR or GR were added. In lanes 3, a 300-fold excess of unlabeled C3(1) ARE was added, and in lanes 4, an AR-or GR-specific antibody was added. *, specific binding complex; S, supershifted binding complex.  struct, for which only a weak decrease in transactivation capacities was observed as compared with the wild type AR, whereas no clear binding of AGA to the PB-ARE-2 was observed in gel shift assays (Fig. 1). The chimerical constructs AGA and GAG seemed to have a slightly higher expression level as compared with AR and GR. However, another possible explanation for the weak induction of pPBARE2luc by AGA involves the AR-specific interactions of the ligand-binding and N-terminal domains with each other, as well as with coactivators (37)(38)(39)(40)(41)(42). These differences in receptor-specific functions of the ligand-binding domain and the N-terminal domain are without a doubt involved in hormone-specific expression of genes containing response elements recognized by all four class I receptors (31). However, in the case of the androgen-specific probasin gene, the changed DNA binding capacities of GAG demonstrate that the AR-DBD is the main determinant for this specificity.
Amino Acids Determining the AR-specific Interaction with PB-ARE-2-In a detailed mutation analysis of both the ARand the GR-DBD, we have identified three amino acids, Thr-585, Gly-610, and Leu-617 (AR-relative numbering) that determine the difference in DNA binding specificity between the AR and the GR. None of these residues is located in the first zinc finger of the AR-DBD, a surprising observation because the DNA-recognition helix (20) with the three P-box residues determining the difference in DNA sequence recognition between the estrogen receptor and the GR, is situated in the first zinc finger (43). Another important observation is that the replacement of Thr-585, Gly-610, or Leu-617 in the AR-DBD by their GR homologues affected only the PB-ARE-2 binding and did not result in a dramatic decrease in the binding affinity for the nonspecific C3(1) ARE. This strongly supports the hypothesis that different binding configurations of the AR exist on these two AREs.
One of the residues that determine the high affinity for the PB-ARE-2 in the AR-DBD, Thr-585, is located in the second zinc finger. Its GR homologue (Ile-483) was shown to be involved in the dimerization by interaction with the conserved Ala-476 in the other monomer (20). This is the first indication of a difference in dimerization interaction between the AR and the GR. The other residues involved in the DNA specificity (Gly-610 and Leu-617) are located in a 12-amino acid-long CTE of the AR-DBD. The AR-DBD amino acid sequence resembles most closely the GR-DBD and, based on the three-dimensional structure of the latter, a model of the AR-DBD structure was proposed (44), but the structure of the CTE in the GR-DBD is not well defined (20,26,27). For other nuclear receptors (e.g. TR and RXR) this region was shown to be involved in dimerization, as well as in making contacts with the phosphate backbone (22)(23)(24). However, these nuclear receptors bind response elements of a direct repeat nature.
In the case the AR the involvement of the 12 CTE amino acids in high affinity DNA binding also depends on the nature of the response element (21). Whereas for the GR-DBD the complete CTE is always necessary for high affinity binding, the AR only requires four residues of the CTE for recognition of a nonspecific response element (C3(1) ARE) and the complete CTE for binding to the PB-ARE-2. This difference is confirmed by the heterodimer complexes formed by AR1 with the AR constructs without and with four residues of the CTE (Fig. 6). We therefore postulate the existence of different binding conformations of the AR-DBD on these two AREs. These het-  Table II. Lanes 1 contain no nuclear extract; in lanes 2-4, equal amounts of COS-7 nuclear extracts containing chimerical full-size AGA or GAG were added. *, specific binding complex; S, supershifted binding complex.
FIG. 6. Heterodimer formation of DBD constructs with a different length of the CTE on the C3(1) ARE. The gel shift assay was performed as described under "Experimental Procedures" with some modifications: the receptor fragments were separated on a 7.5% polyacrylamide. In each lane, 10 ng of the indicated receptor constructs were applied. The number in the construct names refers to the last C-terminal amino acid of these constructs (described in the text). erodimeric complexes are not detected for the GR. The involvement of only one complete CTE in stable DNA binding also indicates that the AR-DBD is able to bind DNA in a head-to-tail configuration, a binding configuration in which one CTE can provide necessary dimerization interactions. This would be in analogy with the heterodimer complex described for, for example, the DNA-binding domains of the TR and the RXR (22).
Mutation and deletion of the amino acids, which constitute the NLS, resulted in a decreased or even a loss of DNA binding (Fig. 2, AR28.1 and AR28.2). This result indicates that the NLS is an integral part of the DBD. Earlier reports studying the NLS with identical mutations in full-length human AR mention decreased transcription activation as a result of a loss of nuclear import (36). Our results indicate that loss of transcription could also be the result of a lack of DNA binding. Finally, two mutations in the CTE were identified in patients with prostate cancer: Arg-612 to Gly (45) and Lys-613 to Thr (46), indicating that the CTE region could be involved in the correct functioning of the androgen receptor.
Sequence Requirements for Specific Recognition by the AR-The difference in DNA binding characteristics between the AR and the GR is clearly illustrated by the binding experiments with the direct repeat of the consensus half-site 5Ј-TGTTCT-3Ј (Tables II and III and Fig. 3). However, cooperativity on a perfect direct repeat seems to be inefficient, because the monomeric and dimeric binding complexes appear with similar kinetics. The mutation analysis of the DR sequence suggests that a hierarchy regulates the DBD binding to the two core halfsites. The right half-site must be occupied first, upon which conformational changes in the AR-DBD render it possible for the left half-site to be recognized cooperatively by the second DBD. Our results suggest that this can only occur when the left half-site of the response element has equal or lesser affinity for the DBD as compared with the right half-site. This hypothesis is confirmed by the mutations introduced in the left half-site, creating a lower affinity half-site (3,6,47) and increasing the affinity of the AR-DBD because only dimeric complexes are observed (e.g. DRI in Table II and Fig. 3). If the left half-site has a markedly higher affinity than the right half-site, the left site will be occupied first, and conformational changes will be unsuitable for recognition of the lower affinity right half-site. This phenomenon would explain the formation of only a monomeric AR-DBD DNA complex on DRI2, DRI3 and DRI5 (Table  II and Fig. 3). For the GR, a binding mechanism similar to inverted repeats in a head-to-head orientation has been put forward (48,49). This is in agreement with structural differences between the monomers in the GR DBD-DNA crystal structure and between DNA-bound GR-DBDs versus DBDs in solution (20,26).
The mutation analysis of the perfect direct repeat of 5Ј TGT-TCT-3Ј further indicates that the mutation of thymine at position -4 to adenine (DRI4) has a positive effect on both the AR and GR binding, although the AR specificity remains (Table II). This is not surprising because adenine is also present at position -4 in the consensus GRE, and mutation to thymine resulted in a loss of affinity of the GR (3). From the co-crystal structure of the GR-DBD bound to an inverted repeat element, it is known that thymine at position -4 of the consensus GRE directly contacts Val-462 (20), and this valine and Ser-459 were shown to restrict the DNA binding when adenine is present at position -4. The high affinity of the AR-DBD for DRI, DR and PB-ARE-2 is therefore somewhat surprising because both Val-462 and Ser-459 are conserved in the AR. However, the GRE in the co-crystal is bound by two GR-DBD monomers in a headto-head configuration. One possible explanation for the high affinity AR binding to elements with thymine at position -4 might be that the AR binds in a head-to-tail configuration (Fig.  7). In this concept, the thymine at position -4 becomes the equivalent of the thymine at position ϩ5. Results reported by other groups indicate that nucleotide variations at position ϩ5 have little effect on DNA binding by class I receptors (3,6,20,47). This equivalence would also explain why the mutation to adenine in DRI5 has no effect (Table II).
In the accompanying paper by Verrijdt et al. (50), androgenspecific response elements in the promoters of the slp-gene and of the sc-gene were shown also to contain a thymine at position -4. Mutation of this nucleotide to adenine resulted in promoter constructs inducible by both androgens and glucocorticoids (14,50). Taken together, these data indicate that thymine at position -4 is an important determinant for the difference in DNA binding between GR and AR.
The involvement of nucleotides and amino acids, other than those described here, in AR-specific DNA binding to the PB-ARE-2 cannot be excluded because combinations of different nucleotides in response elements will lead to subtle changes in DNA backbone structure as well as result in other possibilities for sequence-specific interactions. The variability of the natural AREs certainly emphasizes this point. All of our data suggest that the AR binds the nonspecific AREs, organized as inverted repeats, in a head-to-head orientation, as described for the estrogen receptor and GR. Specific AREs, organized as direct repeats, however, seem to be recognized by the AR in a head-to-tail orientation (Fig. 7). Because other androgen-specific response elements with a direct repeat nature have now been identified (12,14,15,50), we propose that this difference in DNA binding could be a more general mechanism of steroid specificity. FIG. 7. Schematic models of the AR-DBD binding to an inverted repeat and to a direct repeat. The AR-DBD models are based on the structural coordinates of the GR-DBD determined by Luisi et al. (20). The N-terminal and C-terminal ends are indicated, and the zinc ions are depicted as gray dots. The locations of Thr-585 and Gly-610, which determine the DNA specificity of the AR, are shown in black; Leu-617 is not indicated because it is not included in the GR-DBD crystal. The arrows indicate the orientations of the core hexamers in the consensus and in the direct repeat element.