Hormone-dependent transactivation by estrogen receptor chimeras that do not interact with hsp90. Evidence for transcriptional repressors.

The ligand-free estrogen receptor (ER), like other steroid receptors, interacts with the 90-kDa heat shock protein hsp90 in vitro. Analysis of the effect of potential ER-hsp90 interactions in vivo on receptor function is complicated by the fact that hsp90 binds to ER domains required for hormone binding and stable DNA binding. ER chimeras were therefore created by replacing the ER DNA binding domain with that of GAL4. In addition, the N-terminal AF-1 domain of the ER was replaced with the strong constitutive activation domain of VP16 to create VP16-GAL-ERs. These chimeras bind DNA in a ligand-independent manner, but, importantly, are ligand-dependent transactivators, unlike VP16-GAL, which displays strong constitutive activity under the same conditions. Hormone induces transactivation by VP16-GAL-ERs to levels similar to the constitutive activity of VP16-GAL. Glycerol gradient and coimmunoprecipitation experiments showed that, unlike the wild-type ER, VP16-GAL-ER chimeras do not interact with hsp90. Deletion analyses indicate that a region of the ER, primarily between amino acids 370 and 470, is responsible for repressed transcription. Our results suggest that interaction with hsp90 is not necessary for controlling hormone-dependent transcription by the ER and provide evidence for repressor factors that interact with the N-terminal portion of the receptor's ligand binding domain in the absence of hormone.

The estrogen receptor (ER) 1 is a ligand-activated transcriptional regulator (1)(2)(3)(4). Ligand-free (apo) steroid receptors can be isolated from cell extracts associated with complexes composed of a number of heat shock and immunophilin proteins (5)(6)(7). The major nonreceptor constituent of these complexes is a dimer of the 90-kDa heat shock protein, hsp90. Addition of hormone leads to complex dissociation and receptor ho-modimerization. These in vitro experiments would be consistent with a model where steroid receptors are in a multisubunit cytoplasmic complex in the absence of ligand and that hormone binding leads to complex dissociation, receptor homodimerization, and transfer of cytoplasmic receptor to the nucleus.
A potential role for hsp90 in vivo in controlling ligand-inducible transactivation by the glucocorticoid receptor (GR) has been well supported by genetic studies in Saccharomyces cerevisiae (8). Reduced expression of the hsp90 gene strongly inhibited GR-dependent transactivation, suggesting that hsp90 stabilized the ligand-free GR. However, the ER was less affected in similar experiments (8), suggesting that hsp90 may not be necessary for regulating ligand-inducible transcription by ER. Moreover, several immunocytochemical studies have suggested that the hormone-free ER is at least partially nuclear (9 -12). Gene transfer experiments have shown that the receptor can be nuclear in the absence of hormone, and can bind DNA, providing evidence for the presence of ligand-free ER homodimers (13,14).
Stable ER-hsp90 interactions in vitro require portions of domains essential for ligand binding and stable DNA binding (5,15), thus complicating analysis of potential interactions in vivo. Here, we have created ER chimeras that are functional in vivo, and that do not interact with hsp90 in vitro, by replacing the ER DNA binding domain with that of the yeast transactivator GAL4. Our results suggest that interaction with hsp90 is not necessary for controlling hormone-dependent transcription by the ER and, moreover, provide evidence for transcriptional repressors that interact with the ligand binding domain of the receptor in the absence of hormone.

MATERIALS AND METHODS
Recombinants-All chimeras were constructed in the pSG5 expression vector (16) by polymerase chain reaction amplification of appropriate regions of VP16, GAL4, and the wild-type ER HEG0. Duplicates of each recombinant were tested for transactivation and verified by DNA sequencing.
Cell Culture-COS-7 cells were grown in 3.5-cm dishes in Dulbecco's modified Eagle's medium containing charcoal stripped 5% fetal bovine serum. Lipofections were performed according to manufacturer's instructions (Life Technologies, Inc.). For luciferase assays, 100 ng of chimera expression vector was used along with 500 ng of 17M5TATAluc and 1 g of p610AZ ␤-galactosidase expression vector for standardization. Cells were lysed in 250 l of lysis buffer (Promega). 50-and 45-l aliquots were used for ␤-galactosidase and luciferase assays, respectively. For Western and gel retardation analyses, 1.0 g of ER expression vector was lipofected along with 1.0 g of p610AZ. For glycerol gradients and coimmunoprecipitations, 10 g of ER expression plasmid was lipofected into COS-7 cells in 10-cm dishes.
Gel Retardation Assays-Cells were harvested and gel retardation assays were performed as described (17) except that cells were resuspended in 30 l of high salt buffer. Estradiol (20 nM) was added during extraction and incubation as indicated.
Immunoprecipitation-Immunoprecipitations were performed essentially as described by Scherrer et al. (18). Cytosol was prepared from transfected or untransfected COS-7 cells in HEPES buffer (10 mM HEPES, 1 mM EDTA, 20 mM sodium molybdate, 50 mM NaCl, and protease inhibitors). Clarified lysates were diluted in TEGM buffer (20 mM Tris, 4 mM EDTA, 10% glycerol, 20 mM sodium molybdate, 50 mM NaCl, and protease inhibitors) and incubated overnight on ice with anti-GAL4 antibodies 2GV3 and 3GV2 (19) or with anti-ER antibody F3 (20). Glycerol gradient fractions from 4 or 8 S peaks were pooled and diluted in TEGM prior to Immunoprecipitation. Immune complexes were absorbed to protein G-Sepharose, washed four times, and analyzed by Western blotting.
Western Analysis-48 h after transfection, cells were harvested in phosphate-buffered saline and divided in half. One aliquot was lysed in lysis buffer (Promega), and ␤-galactosidase assays were performed (21) to assess transfection efficiency. The other aliquot was lysed directly in SDS-polyacrylamide gel electrophoresis sample buffer and used for Western analysis. Blots were incubated with a combination of anti-GAL4 DBD monoclonal antibodies 2GV3 and 3GV2 (19) diluted 1/1000 in Tris-buffered saline-Tween and 1% milk powder. Western analysis of hsp90 was performed using anti-hsp90 monoclonal antibody SPA-835 (Stressgen). Blots were developed using the ECL detection system (DuPont NEN). Glycerol Gradient Analysis-Gradients were performed as described by Chambraud et al. (5). Extracts were made in the presence of phenylmethanesulfonyl fluoride and protease inhibitors and stabilized with 30 mM sodium molybdate. Fractions (150 l) were collected from the top of the gradient and assayed by scintillation counting.

RESULTS
Experimental Design-We were interested in constructing functional ER derivatives that would not interact with hsp90 in vitro to analyze the potential role of heat shock proteins in controlling ER activity. Region D of the receptor has been shown to participate in ER-hsp90 interactions in vitro (5). Since mutation of region D would disrupt stable DNA binding (15), the ER DBD was replaced with that of GAL4 by fusing the GAL4 DBD to the C-terminal portion of the ER from aa 251 to 300 (Fig. 1A).
Any hormone-free GAL-ER chimeras, which do not interact with hsp90, would be expected to be nuclear and possibly DNA-bound. Thus, it would be useful to use transactivation as an assay for nuclear localization and DNA binding. However, wild-type ER bound to DNA in the absence of ligand would be hormone-dependent for transactivation, since the transactivating domain AF-1 (Fig. 1A) has been shown to synergize strongly with ligand-bound AF-2 (21). Therefore, the N-terminal A/B region containing AF-1 was replaced with the strong hormoneindependent activating domain of the herpes simplex virus activator VP16 (Fig. 1A), which does not synergize with AF-2 (21). In addition, we used a synthetic promoter containing five GAL4 17-mer binding sites (17M5-TATA-luc, Fig. 1A), since VP16 functions highly synergistically when bound to multiple sites.
VP16-GAL-ER Chimeras Are Hormone-inducible Transactivators That Do Not Interact with hsp90 in Vitro-VP16-GAL, which lacks the ER ligand binding domain (LBD), is a strong constitutive activator of 17M5-TATA-luc (Fig. 1B, V-G). This activity is 200-fold higher than that from a promoter containing a single 17-mer binding site (data not shown) and 50-fold higher than the hormone-dependent activity of GAL-ER (Fig.  1B, G-ER), demonstrating the powerful transcriptional activity of VP16. Transactivation by VP16-GAL-ER chimeras on 17M5-TATA-luc in the absence of hormone was 15-35% of that observed with VP16-GAL. Estradiol consistently stimulated transactivation by chimeras 4-fold (with the exception of VP16-GAL-ER 281; 3-fold), to levels similar to those observed with VP16-GAL. While this induction is less than the 6 -9-fold stimulation observed with the wild-type ER (data not shown), it is striking given that VP16 is a strong constitutive activator. Several Western analyses showed that all recombinants were expressed at similar levels ( Fig. 1C and data not shown) and that expression levels were not dependent on the presence of estradiol (Fig. 1B, inset), thus indicating that the reduced level of transactivation by VP16-GAL-ER chimeras in the absence of hormone relative to VP16-GAL was not due to low levels of expression.
Glycerol gradients were used to analyze molybdate-stabilized extracts of cells transfected with the wild-type-ER or VP16-GAL-ER chimeras for interaction with hsp90. The wildtype ER displayed a characteristic salt-dependent shift in sedimentation coefficient (5) from the 8 S hsp90-containing complex, to the hsp90-free 4 S form ( Fig. 2A). In contrast, 4 S, but no 8 S complex formation, was observed in extracts of cells transfected with VP16-GAL-ER 258 (Fig. 2B) or VP16-GAL-ER 300 (not shown), indicating that they did not interact with hsp90. Western analysis of gradient fractions showed that peak hormone binding corresponds to peaks of intact protein, suggesting that no significant proteolysis of chimeras occurred (Fig. 2B). No interaction of VP16-GAL-ER 258 with hsp90 was detected by immunoprecipitation with anti-GAL antibodies of molybdate-stabilized whole cell extracts of transiently transfected COS-7 cells or of the 4 S peak from a glycerol gradient (Fig. 2C, lanes 1-6). Identical results were obtained using anti-ER antibody F3 (not shown). In contrast, immunoprecipitation of HEG0 expressed in COS-7 cells with F3 led to increased coimmunoprecipitation of hsp90 (Fig. 2C, lanes 7-9). These results suggest that interaction with hsp90 does not control hormone-dependent transactivation by the chimeras. Gel retardation assays were also performed to test for hormonedependent DNA binding. No significant effect of hormone on DNA binding by VP16-GAL-ER 258 or VP16-GAL-ER 300 was detected in assays performed with extracts made in the presence or absence of estradiol ( Fig. 1D and data not shown). In this respect, the chimeras functioned similarly to the wild-type ER (22). 2 Deletion Analysis of the ER Ligand Binding Domain-Hor-

Transactivation by Estrogen Receptor Chimeras 25728
mone dependence of VP16-GAL-ER chimeras may be due to interaction with factors, other than hsp90, which repress transcription in the absence of hormone. A series of deletion mutants were created to test which portions of the ER LBD are responsible for the reduced transactivation in the absence of estradiol (Fig. 3A). C-terminal truncations beyond ER aa 553, which disrupt the integrity of the LBD, generated chimeras displaying low levels of constitutive transactivation (Fig. 3B). Significant constitutive activity is recovered, however, with C-terminal deletions beyond aa 430 to aa 370 and further to aa 302. Disruption of the LBD N terminus by deletion past aa 302 to aa 340, 370, or 430 generated chimeras exhibiting low levels of constitutive activity (Fig. 3B). Significant constitutive activation was only recovered by deletion of the N terminus to aa 470. Taken together, these data indicate that a portion of the LBD, primarily sequences between aa 370 and 470, is required for repressed transcription observed in the absence of hormone. DISCUSSION Our results suggest that the mechanism of action of the ER is intermediate between that of the GR subfamily of steroid receptors and those of the thyroid hormone/retinoid/vitamin D 3 nuclear receptors. The apo-GR is cytoplasmic, and ligand binding leads to its translocation to the nucleus where it binds palindromic DNA sequences as a homodimer (23). The cytoplasmic location of the aporeceptor would be consistent with its interaction in vitro with hsp90, which is predominantly cytoplasmic. In contrast, the thyroid hormone and related receptors do not interact with hsp90 in vitro (24), are nuclear in the absence of hormone and bind to response elements composed of directly repeated motifs as heterodimers with retinoid X receptors (25,26).
The ER, like the GR, recognizes palindromic response elements as a homodimer (1-4). Numerous studies have indicated that the full-length ER interacts with hsp90 in vitro, suggesting that similar interactions may occur in vivo. Immunopre-

Transactivation by Estrogen Receptor Chimeras 25729
cipitation experiments by others have also indicated that the isolated ER LBD interacts weakly with hsp90 (18). However, the LBD used contains a Gly 400 3 Val mutation, which destabilizes its structure (22,27). Our glycerol gradient analyses with VP16-GAL-ER chimeras containing a Val 400 mutation have shown that, unlike their Gly 400 counterparts, these chimeras form salt-sensitive 8 S complexes in vitro (not shown). In vivo studies, including immunocytochemistry and gene transfer experiments, have provided evidence for the presence of homodimers of apo-ER in the nucleus (13,14), which would be inconsistent with stable interaction with hsp90. Taken together, the above results suggest that if the ER interacts with hsp90 in vivo, this interaction is transient, and that a significant concentration of homodimeric aporeceptor is present in the nucleus.
Our results show that ER derivatives, which do not interact with hsp90 in vitro, can function as ligand-dependent transactivators in vivo. This occurs in spite of the fact that the VP16-GAL-ER chimeras contain an acidic activating domain which is strongly constitutively active when not tethered to the ER LBD. These results suggest that interaction with hsp90 in vivo is not essential for controlling ligand-dependent transactivation by the ER. It appears that the chimeras tested here are maintained in a transcriptionally repressed state in the absence of ligand, given that ligand induces transactivation to levels similar to those seen with the constitutive activator VP16-GAL. There are several potential candidates for a repressor. In yeast, HSP70 acts downstream of hsp90 to control ER and GR function (28). It is possible that molecules analogous to those that repress the thyroid hormone and retinoic acid receptors (29,30) also act on the ER. A repressor may be specific or have a broad spectrum of effects on transcription similar to the yeast factor SSN6 (31). Deletion analyses of the ER suggest that a putative repressor would interact with a region of the LBD between aa 370 and 470 (Fig. 3).
The obvious function of a putative repressor would be to maintain DNA-bound apo-ER in a transcriptionally silent state. Hormone would stimulate dissociation of bound repressor, freeing the LBD for binding of transcriptional intermediary factors or coactivators (32)(33)(34). In this model, antagonists would either bind the ER and not stimulate repressor dissociation, or bind, stimulate repressor dissociation, but maintain the LBD in a conformation not recognized by transcriptional intermediary factors. Repression of the ER would not only act to block the activity of the LBD but also the AF-1 domain in the N terminus, which can be activated by phosphorylation in the presence of hormone (35) and which has the capacity to synergize with other classes of transactivators under certain conditions (21). The potential activity of AF-1 would be dampened in cells where truncated receptors lacking the LBD display significant levels of constitutive activity (36). It is also noteworthy that aa 370 -470 are adjacent to the region of the LBD (aa 300 -330) identified as being important for binding of TATA box-binding protein-associated factor TAFII30, which is required for transactivation by the ER (37).