Estrogen Receptors α and β Form Heterodimers on DNA*

The estrogen receptor (ER) is expressed in two forms, ERα and ERβ. Here we show that ERα and ERβ, expressed both in vitro and in vivo, form heterodimers which bind to DNA with an affinity (K d of approximately 2 nm) similar to that of ERα and greater than that of ERβ homodimers. Mutation analysis of the hormone binding domain of ERα suggests that the dimerization interface required to form heterodimers with ERβ is very similar but not identical to that required for homodimer formation. The heterodimer, like the homodimers, are capable of binding the steroid receptor coactivator-1 when bound to DNA and stimulating transcription of a reporter gene in transfected cells. Given the relative expression of ERα and ERβ in tissues and the difference in DNA binding activity between ERα/ERβ heterodimers and ERβ it seems likely that the heterodimer is functionally active in a subset of target cells.

Estrogen receptors (ER) 1 were recently shown to be encoded by two distinct genes, ER␣ and ER␤ (1,2). Reverse transcription-polymerase chain reaction (PCR) analysis indicates that ER␤ is highly expressed in prostate and ovary (1,2), but moderate expression was detected in many other tissues including testis and uterus, some of which also seem to express ER␣ (3). The two receptors which share about 95% homology in the DNA binding domain and 55% homology in the ligand binding domain, both bind to a consensus estrogen response element (ERE) (4) and exhibit similar ligand binding properties (3). They are poorly conserved in the N-terminal domain but ER␤, like ER␣, appears to contain a similar activation domain, activation function 1 (AF-1) sensitive to a mitogen-activated protein kinase pathway (4 -6). In addition, both receptors contain a second activation domain, activation function 2 (AF-2) (7,8), whose activity is enhanced by the coactivator SRC-1 (4,9,10). Thus, although the relative expression of ER␣ and ER␤ varies in cells, their ligand binding, DNA binding, and transactivation properties are rather similar to one another.
Steroid hormone receptors usually bind to inverted DNA repeats as homodimers, although the glucocorticoid and mineralocorticoid receptors have been reported to form heterodimers, at least in vitro (11,12). In the classically accepted model of steroid hormone action, the estrogen receptor is sequestered in an inactive state in a multiprotein complex in the absence of hormone (13). Upon estrogen binding, the receptor forms homodimers which then interact with response elements in the vicinity of target genes and modulate rates of gene transcription. In view of the similarity of the ligand binding domain of ER␣ and ER␤ we investigated the possibility that the two receptors may form functional heterodimers in target cells. ER␣ and ER␤ were capable of forming heterodimers on DNA that could bind the coactivator, SRC-1, and appeared to stimulate transcription of a reporter gene. Moreover, we demonstrate that while the region of ER␣ required for homodimerization overlaps with that required for heterodimerization the two regions are not coincident.

EXPERIMENTAL PROCEDURES
Plasmids-The isolation and construction of cDNA clones that encode the mouse ER␣ and a series of point mutants for analyzing receptor dimerization have been described previously (14,15). To express human ER␤ the 1.5-kilobase ER␤ cDNA (1) was subcloned into the BamHI site of pSP65 for in vitro transcription and translation and pSG5 for mammalian cell expression. The human ER␣ cDNA from pSG5HEGO, kindly provided by Pierre Chambon, was subcloned into the EcoRI site of pSP65. Glutathione S-transferase (GST)-SRC-(570 -780) was generated by subcloning a PCR fragment of SRC-1 into pGEX2TK.
In Vitro Transcription and Translation-ER␣ and ER␤ protein was synthesized in vitro using the TNT-coupled reticulocyte lysate system (Promega) in the presence of 0.1 mM methionine as described previously (16). To quantitate the relative amount of receptors produced, 1 Ci/l [ 35 S]methionine (Amersham) was included in the reaction mixture.
Transfection Experiments-For biochemical analysis, wild-type and mutant receptors were overexpressed in COS-1 cells by electroporation using a Bio-Rad gene pulser at 450 V and 250 millifarads as described previously (17). Cells were transfected with 20 g of expression plasmid, either pSG5HEGO or pSG5ER␤ as indicated. After 2 days, cells were harvested, and whole cell extracts were prepared using a high salt extraction buffer (400 mM KCl, 20 mM HEPES, pH 7.4, 1 mM dithiothreitol, 20% glycerol, plus protease inhibitors) as described (17). Their protein concentration was determined using the Bio-Rad protein assay kit (Bio-Rad). For transient transfection assays, HeLa cells were plated into 24-well microtiter plates in phenol red-free medium containing 5% charcoal/dextran-stripped fetal calf serum. Cells were transfected using a modified calcium phosphate coprecipitation method (18) with 1 g of pEREBLCAT reporter plasmid, a total of 10 ng of ER␣ and ER␤ expression plasmids as described under "Results," 150 ng of pJ7lacZ plasmid as an internal control, and pSG5 plasmid to a total of 1.5 g of DNA per well. After 24 h, the cells were washed with Dulbecco's modified Eagle's medium and then maintained in medium with or without 1 ϫ 10 Ϫ8 M 17␤-estradiol for 24 h. The cells were then washed with phosphate-buffered saline and harvested in lysis buffer (10 mM Tris-HCl, pH 8.0, 1 mM EDTA, 150 mM NaCl, and 0.65% Nonidet P-40). Extracts were assayed for CAT activity (19) or ␤-galactosidase activity using a Galacto-Light chemiluminescent assay (Tropix).
DNA Binding Assays-DNA binding was assayed using a gel shift assay. Aliquots of receptors, either translated in vitro or expressed in COS-1 cells, were incubated with a 32 P-labeled double-stranded oligonucleotide probe containing a consensus ERE sequence (5Ј-CTA-GAAAGTCAGGTCACAGTGACCTGATCAAT-3Ј) as described previously (17). The human ER␣ monoclonal antibody H226 or ligands were added as indicated. In some experiments GST-SRC-(570 -780), expressed and purified as described previously (20), was added. The * The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked "advertisement" in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.

RESULTS
The DNA binding activity of ER␣ and ER␤ was tested using in vitro translated receptors and a consensus estrogen response element in a gel shift assay. Both ER␣ and ER␤ bound to the element, and the mobility of ER␣, but not that of ER␤, was retarded in the presence of the hER␣ antibody, H226 (Fig. 1). When the two receptors were cotranslated we were able to detect a complex with an intermediate mobility corresponding to ER␣/ER␤ heterodimers, in addition to ER␣ homodimers ( Fig. 1, tracks 7 -12). Their mobility was retarded by H226 consistent with the presence of ER␣ in both complexes. Their relative amounts varied depending on the input ratio of the two receptors but it is noteworthy that ER␤ homodimers were barely detected even when ER␤ was expressed in 2-fold excess over ER␣ (Fig. 1, track 11). Their affinity for DNA was then determined by carrying out gel shift experiments over a wide range of probe concentrations (Fig. 2). We found that ER␣ homodimers and ER␣/ER␤ heterodimers bound to DNA with a similar K d of approximately 2 nM whereas that of ER␤ homodimers was about 4-fold greater.
We next analyzed the DNA binding activity of ER␣ and ER␤ when they were expressed in COS-1 cells. When ER␣ alone was expressed, we observed two complexes, a major upper complex, corresponding to the ER␣ homodimer, and an additional complex that is probably generated by proteolysis. It seems to lack N-terminal sequences since it is recognized by a monoclonal antibody specific for the C-terminal F region (21) (data not shown). Similar results have been reported previously (22). As expected, the mobility of ER␣ but not ER␤ was retarded in the presence of the specific hER␣ antibody, H226. When equivalent amounts of ER␣ and ER␤ expression vector were coexpressed, heterodimers were the predominant form observed, and ER␤ homodimers were not detected (Fig. 3A). We then used these extracts to compare the effect of 17␤-estradiol and 4-hydroxytamoxifen on the DNA binding activity of the three dimeric forms (Fig. 3B). As previously demonstrated for ER␣ (7), the DNA binding activity of both ER␤ and ER␣/ER␤ heterodimers was unaffected by ligand binding, but their mobilities were slightly increased in the presence of 17␤-estradiol (Fig. 3B, tracks 2, 5,  and 8). Thus, we conclude that ER␣/ER␤ heterodimers, expressed in intact cells, are capable of forming on DNA and that ER␣ homodimers and ER␣/ER␤ heterodimers are preferentially formed.
Previous work with a series of ER␣ mutants has identified a region within the ligand binding domain of the estrogen receptor which is required for both receptor dimerization and high affinity DNA binding (15). We have used these mutants to determine whether the region required to form homodimers with a truncated version of ER␣ (mouse estrogen receptor (MOR)-182-599) is similar to that required to form heterodimers with ER␤ on DNA. We find that the ability of R507A to form either ER␣ homodimers or ER␣/ER␤ heterodimers is markedly reduced (Fig. 4, compare tracks 4 and 11) while L511R and I518R, which show negligible homodimer formation, retain some ability to form ER␣/ER␤ heterodimers (Fig. 4,  compare tracks 6 with 13 and 7 with 14). In contrast, mutation of Q510A had no affect on the dimerization of either receptor. A series of other mutations in this region of the receptor (A509R, L512V, L513G, I514R, L515G, L516A, H517A, R519G) was then screened to attempt to identify additional residues which could discriminate between homo-and heterodimerization, but all the mutants retained their DNA binding activity both as ER␣ homodimers and ER␣/ER␤ heterodimers (data not shown). We conclude that the region of ER␣ required for homodimerization overlaps that required for heterodimerization, but the two regions are not coincident.
We next assessed the transcriptional activity of ER␣/ER␤ heterodimers in transiently transfected COS-1 cells using the pEREBLCAT reporter gene. ER␣ and ER␤ expression vectors FIG. 1. DNA binding activity of ER␣, ER␤, and ER␣/ER␤ heterodimers. Equivalent amounts of in vitro translated ER␣, ER␤, or different ratios of the two, as indicated, were incubated with a 32 Plabeled ERE in the presence or absence of ER antibody H226. As controls, the unprogrammed lysate was analyzed (tracks 1 and 2). The position of the ␣/␣ homodimer, ␣/␤ heterodimer, and ␤/␤ homodimer is shown on the right-hand side.

FIG. 2. DNA binding affinity of ER␣, ER␤, and ER␣/ER␤ heterodimers.
In vitro translated ER␣, ER␤, or both (at a ratio of 2:1 in favor of ER␤) were analyzed for DNA binding activity in a gel shift assay over the range of DNA concentrations indicated. Radiolabeled probe alone was used up to a concentration of 6 nM (2 ng), and higher concentrations were achieved by diluting the labeled probe with nonlabeled oligonucleotide, and the resultant reduction in specific activity was taken into account in the calculations. Gels were then scanned with a PhosphorImager, and the relative binding of the different dimeric states was plotted against the concentration of ERE. Scatchard analysis was then used to determine the K d for the different forms of receptor dimer.
were tested individually or in combination at a ratio of 1:1 or 1:2 to minimize the relative amount of ER homodimers formed (see Fig. 3). The ability of ER␣ to stimulate transcription was slightly greater than that of ER␤ (Fig. 5A), as previously reported for this reporter (4). Coexpression of ER␣ and ER␤ resulted in an intermediate level of transcription that was blocked by the addition of the antiestrogens, 4-hydroxytamoxifen and ICI 182780. Similar results were obtained in HeLa cells (Fig. 5B). Therefore, since the heterodimer is the major dimeric form of the receptor under these conditions, it appears to retain its ability to stimulate transcription.
To obtain additional evidence to support our suggestion that ER␣/ER␤ heterodimers are capable of stimulating transcription we tested whether they were able to bind the coactivator, SRC-1, as previously demonstrated for ER␣ (10) and ER␤ homodimers (4). This was achieved by analyzing the ability of the receptors, bound to DNA, to interact with a fragment of SRC-1, residues 570 -780, that binds nuclear receptors in a ligand-dependent manner. 2 As shown in Fig. 6, when ER␣, ER␤, or both were incubated with increasing amounts of GST-SRC-(570 -780) in the presence and absence of ligand we could detect additional complexes in the gel shift assay. The interaction of SRC-(570 -780) with ER␣ was dependent on the presence of ligand (Fig. 6, compare tracks 7 and 8) whereas there was an appreciable interaction with ER␤ in the absence of ligand (Fig.  6, compare tracks 13 and 14). The interaction of SRC-1 with the heterodimer was enhanced in the presence of ligand (Fig. 6,  compare tracks 19 and 20). As a control, we showed that these retarded complexes were not due to the binding of SRC-1 directly to an ERE (Fig. 6, tracks 1 and 2) but required the presence of receptor. Thus ER␣/ER␤ heterodimers, bound to DNA, are capable of recruiting SRC-1.

DISCUSSION
The main conclusion from our study is that human ER␣ and ER␤ are capable of forming functional heterodimers on DNA. The relative distribution of ER homodimers and heterodimers will, at least in part, be dependent on the relative expression of the two receptors which varies widely in different cell types. Both ER␣ and ER␤ have been detected in many tissues by reverse transcription-PCR or in situ hybridization (3, 23) but the relative amounts of receptor protein in specific cell types have yet to be determined. Nevertheless this preliminary analysis suggests that the expression of ER␣ may be greater than that of ER␤ in epididymis, testis, pituitary ovary, uterus, adrenals, and heart. Given that ER␣ homodimers and ER␣/ER␤ heterodimers are preferentially formed over ER␤ homodimers it seems that heterodimers are more likely to be formed than ER␤ homodimers in these tissues. On the other hand, ER␤ is expressed at higher levels in prostate, bladder, lung, thymus, and certain hypothalamic cells (3,23), and so ER␤ homodimers may be formed in these tissues.
The molecular basis for the reduced DNA binding activity of ER␤ compared with that of ER␣ and ER␣/ER␤ heterodimers is unclear, but recent work indicates that the mouse ER␤ also binds to an ERE less strongly than ER␣ (4). Differences in the DNA binding domains of the two receptors are unlikely to account for the variation since they differ by only two residues (1, 2), neither of which seems to be in a position that is likely to affect its DNA binding properties (24). An alternative possibility is that the receptors differ in their ability to dimerize. The major dimer interface in ER␣ has been mapped to a region of the hormone binding domain (15,25) which is conserved and likely to correspond to helix 10 in nuclear receptors (26). The 2 E. Kalkhoven, data not shown.

FIG. 3. ER␣ and ER␤, expressed in COS-1 cells, bind as heterodimers to DNA.
Equal amounts of whole cell extracts, prepared from cells expressing ER␣, ER␤, or both, were incubated with a 32 Plabeled ERE in the presence or absence of H226 antibody (A) or different ligands (B). 17␤-Estradiol or 4-hydroxytamoxifen were tested at 2 ϫ 10 Ϫ8 M and 1 ϫ 10 Ϫ6 M, respectively. When ER␣ alone was expressed, we observed two complexes, a major upper complex, corresponding to the ER␣ homodimer, and an additional complex that is probably generated by proteolysis. It seems to lack N-terminal sequences since it is recognized by a monoclonal antibody specific for the C-terminal F region (21). The position of the ␣/␣ homodimer, ␣/␤ heterodimer, and ␤/␤ homodimer is shown on the right-hand side. crystal structure of the ligand binding domain of RXR also indicates that the dimer interface comprises this helix and, to a lesser extent, helix 9 and the loop between helix 7 and 8 (27). Comparison of the sequence of the corresponding region in ER␣ and ER␤ indicates that helix 10 is similar (13/18 residues are identical) but the loop region and helix 9 are poorly conserved in the two receptors. Thus, it seems likely that the residues required to form the dimer interface in ER␣ and ER␤ homodimers are distinct. Although the precise dimerization interfaces in these receptors have yet to be identified, functional analysis of a series of mutations in helix 10 of ER␣ indicates that the residues required for the formation of ER␣ homodimers are similar but not identical to those required for ER␣/ER␤ heterodimerization.
Both ER␣ and ER␤ are capable of stimulating the transcription of reporter genes in transfected cells and the activation functions, AF-1 and AF-2 characterized in ER␣ (5, 6, 8, 28), appear to be conserved in ER␤ (4). When ER␣ and ER␤ are expressed, under conditions when heterodimers are the pre-dominant dimeric species, transcription of an ERE reporter gene is stimulated to an intermediate level compared with that of either homodimer suggesting that ER␣/ER␤ heterodimers are transcriptionally active. This is supported by our observation that ER␣/ER␤ heterodimers are capable of binding the coactivator SRC-1 (9,29). We found that SRC-1 interacts with all three dimeric states of ER bound to DNA although its interaction with heterodimers was less dependent on ligand than that observed with ER␣ homodimers. The interaction of ER␤ homodimers with SRC-1 was dependent on ligand in solution (4) 3 but not on DNA and, consistent with this, SRC-1 was found to augment the transcriptional activity of ER␤ in the absence of ligand (4). Finally, the ability of the heterodimer to stimulate transcription was blocked by the antiestrogens, 4-hydroxytamoxifen and ICI 182780, as previously demonstrated for ER homodimers (4,7,8,30).
The discovery of a second estrogen receptor raises many questions, most notably relating to their respective functions. The ability of ER␣ and ER␤ to form heterodimers suggests that estrogen receptor may function in different dimeric states, and it is possible that they could be activated by selective ligands. In view of the similarity of their DNA binding domains it is doubtful whether different forms bind to distinct response elements, but they could activate different genes in different target cells given their distinct expression patterns.

FIG. 5. Transcriptional activity of coexpressed ER␣ and ER␤.
The ability of ER␣ and ER␤, expressed individually or in combination, to stimulate transcription of a reporter gene pEREBLCAT was tested by transient transfection in COS-1 (A) and HeLa cells (B). Transcriptional activity was determined in the absence or presence of 1 ϫ 10 Ϫ8 M 17␤-estradiol (E 2 ), 1 ϫ 10 Ϫ7 M 4-hydroxytamoxifen, and 1 ϫ 10 Ϫ7 M ICI 182780 and corrected for the activity of the internal control pJ7lacZ. Activity of the reporter was also analyzed in the absence of transfected ER as a control (C). The error bars represent standard errors determined from two transfection experiments each carried out in duplicate.
FIG. 6. SRC-1 binds to ER␣/ER␤ heterodimers bound to DNA. Equal amounts of ER␣, ER␤, and ER␣/ER␤ heterodimers were incubated with increasing amounts of GST-SRC-(570 -780) in the presence or absence of 2 ϫ 10 Ϫ8 M 17␤-estradiol. As controls, GST-SRC-(570 -780) was tested for its DNA binding activity in the absence of receptor. The position of receptors complexed with GST-SRC is indicated by solid arrowheads.