Differential Recruitment of the Mammalian Mediator Subunit TRAP220 by Estrogen Receptors ERα and ERβ*

Estrogen receptors (ERs) associate with distinct transcriptional coactivators to mediate activation of target genes in response to estrogens. Previous work has provided multiple evidence for a critical role of p160 coactivators and associated histone acetyltransferases in estrogen signaling. In contrast, the involvement of the mammalian mediator complex remains to be established. Further, although the two subtypes ERα and ERβ appear to be similar in regard to principles of LXXLL-mediated coactivator binding to the AF-2 activation domain, there are indications that the context-dependent transcriptional activation profiles of the two ERs can be quite distinct. Potentially, this could be attributed to differences with regard to coregulator recruitment. We have here studied the interactions of the nuclear receptor-binding subunit of the mammalian mediator complex, referred to as TRAP220, with ERα and ERβ. In comparison to the p160 coactivator TIF2, we find that TRAP220 displays ERβ preference. Here, we show that this is a feature of the binding specificity of the TRAP220 LXXLL motifs and demonstrate that the ER subtype-specific F-domain influences TRAP220 interaction. Such differences with regard to coactivator recruitment indicate that the relative importance of individual coregulators in estrogen signaling could depend on the dominant ER subtype.

Gene regulation by estrogens and related compounds plays critical roles in cellular differentiation, homeostasis, reproduction, and cancer. Estrogen signaling is mediated by two distinct estrogen receptor (ER) 1 subtypes named ER␣ and ER␤ (1). Although they are quite similar in certain conserved aspects of structure and key functions, ER␣ and ER␤ are also substantially different with regard to expression pattern, knockout phenotypes, and ligand-binding features (2)(3)(4). Thus, they exert specific physiological functions through distinct estrogen signaling pathways (5). Recent structural and functional studies have revealed a more precise mechanistic picture of how estrogen binding causes ER activation (6 -9). In particular, the ability of the C-terminal ligand binding and transcription activation domain, referred to as activation function (AF)-2 domain, to adopt different conformations in response to ligands has been recognized as the critical feature with regard to transcriptional activation. Protein-protein interaction screenings and biochemical approaches have led to the identification of multiple nuclear receptor (NR)-associated proteins (10). Many of them represent relevant coactivators or corepressors, which may collectively act as ER-coregulators at different stages of the activation process (11,12). The majority of coactivators to date represent AF-2-binding proteins that contain a leucinerich signature motif, referred to as the LXXLL motif or the NR box (13). Apparent mechanistic simplicity of coregulator binding to the ER ligand-binding domain is opposed by enormous complexity of coregulator function, raising questions of coregulator specificity and redundancy in ER-dependent transcriptional activation (11,12). Among the best characterized ER coactivators are p160/SRC family members as well as p300/ CREB-binding protein (CBP) and p300/CBP-associated factor (P/CAF), all of which exhibit histone-acetyltransferase (HAT) activity. They form complexes with each other and may participate at distinct steps in ER-mediated transcription activation prior to association with RNA polymerase II (12). In contrast, mediator-like coactivator complexes, which are devoid of HAT activity, are believed to function at different steps by directly associating with the basal transcription machinery (14 -16). Intriguingly, one such complex, referred to as TRAP or DRIP complex, has been biochemically identified as a thyroid hormone receptor (TR)-or vitamin D receptor (VDR)-associated multiprotein complex, respectively (17,18). The 220-kDa receptor binding subunit of the complex, referred to as TRAP220/ DRIP205/Peroxisome proliferator-activated receptor-binding protein (PBP), was independently isolated by us and others in genetic two-hybrid screenings (19,20). Most current models appear to favor a sequential recruitment of distinct coactivators: first, recruitment of chromatin-modifying HATs, and second, recruitment of mediator complexes (10,14,16). However, combinatorial or even parallel pathways, which support functional independence of distinct coactivators, may exist as well (12). For example, by analyzing promoter occupancy of estrogen-regulated genes in the breast cancer cell line MCF-7, it was suggested that p160 coactivators and TRAP220 function in a rather combinatorial manner (21). Interestingly, p160 coactivators but not TRAP220 appeared to be sufficient for estrogenregulated gene activation in these studies. A critical role for p160 coactivators in estrogen signaling is further supported by knock-out studies, particularly in the case of SRC3 (22,23). In contrast, the phenotype of TRAP220/PBP knockout mouse embryos did not provide any support for functions in ER signaling but pointed at critical functions of TRAP220 in TR and VDR signaling (24,25). Thus, the relative importance of TRAP220 for estrogen-regulated gene expression remains currently unclear.
Previous studies have shown that TRAP220 efficiently interacts with, and thus is also able to compete with, other coactivators for binding to various retinoid X receptor (RXR) heterodimers (17, 20, 26 -29). However, there are preliminary indications that TRAP220 and the entire complex binds only weakly to ER␣ (17,26,27). Here we have analyzed the interaction characteristics of ER␣ and ER␤ with TRAP220 and, for comparison, with the p160 coactivator transcription intermediary factor 2 (TIF2). Our binding studies suggest that TRAP220 displays ER␤ preference, and we have delineated specificity determinants for both TRAP220 and ERs. Differences between the ERs with regard to TRAP220 recruitment could have implications for the relative importance of distinct coregulators and coregulator complexes in estrogen signaling.

Protein-Protein Interaction Assays
Glutathione S-Transferase (GST) Pull-down Assay-GST-ER fusion proteins were purified on glutathione-Sepharose beads (Sigma) according to standard methods and incubated with in vitro translated TRAP220 proteins in incubation buffer (50 mM KP i , 100 mM NaCl, 1 mM MgCl 2 , 10% glycerol, and 0.1% Tween 20) and 1.5% serum bovine albumin. Incubation with rotation was carried out for 2 h at 4°C. The beads were washed four times with 20 volumes of incubation buffer and then boiled with SDS sample buffer. The proteins were separated by SDS-polyacrylamide gel electrophoresis and analyzed by autoradiography (Amersham Pharmacia Biotech ECL system).
Electrophoretic Mobility Shift Assay (EMSA)-EMSA assays were carried out under previously described conditions (20,32) with minor changes. Briefly, ϳ20 ng of ER proteins (in vitro translated/recombinant purified from baculovirus expression) were incubated with radioactively labeled double-stranded ERE oligonucleotides for 10 min at room temperature. Purified GST fusion proteins (ϳ500 ng) were added subsequently and binding proceeded for 30 min.
Coimmunoprecipitations-0.5 mg of COS-7 whole cell extracts coexpressing GAL4-ER DEF and TRAP220 wild-type proteins were precleared with 20 l of protein A/G-agarose (Santa Cruz Biotechnology) in IP-T150 buffer containing 0.2% Nonidet P40 (32) for 30 min. Extracts were incubated with 2 g of polyclonal rabbit GAL4 antibody (Santa Cruz Biotechnology) for 1 h. 20 l of protein A/G-agarose were added last, and incubation proceeded for 2 h before washing three times in IP buffer w/o Nonidet P40. Immunodetection of TRAP220 in the precipitates was performed using a rabbit polyclonal antiserum (20).
Surface Plasmon Resonance (SPR) Analysis-SPR analysis was performed using the BIAcore 2000 system and software BIACORE 2000 Control Software 3.1.1 (BIAcore Inc). The buffer used for all experiments was 50 mM Tris-HCl, pH 7.4, 150 mM NaCl, 1 mM EDTA, 0.05% Tween 20. The experiments were performed at 25°C. The peptideprotein interactions were analyzed using the BIA evaluation program 3.1 (BIAcore Inc.). For the studies with GST fusion protein, anti-GST antibody was immobilized on research grade CM5 sensor chips (BIAcore Inc.) using the amine-coupling kit and the GST capture kit provided by the manufacturer. The immobilization procedure was carried out as described previously (20). After immobilization, GST, GST-TRAP220-(579 -718) and GST-TIF2-(594 -766) were bound to individual flow cell surfaces. Samples of ER␣ or ER␤ were injected over the three surfaces. After injection stopped, the surfaces were washed with buffer for an additional 600 s. Surfaces were then regenerated down to the anti-GST protein by two pulses of 1 min injections of 10 mM glycine, pH 2.2. For the peptide studies, research grade streptavidin (SA) chips were used. The SA chips were first treated with 3 pulses of 50 mM NaOH and 1 M NaCl at a flow rate of 5 l/min. Biotinylated peptides diluted in water were then injected over their respective surfaces to variable response (25-300 RU). ER␣ or ER␤ were then injected, and after injection was completed, the surfaces were washed with buffer for at least 600 s. The flow cell surfaces were regenerated down to captured peptide by 2 or 3 pulses of 1 min injections of 10 mM NaOH.
Affinity determination analysis was done using the BIAevaluation software 3.0 (BIAcore Inc.) The program finds values for the parameters in the rate equations that best fit the experimental data (curve fitting procedure). The program uses the Marquardt-Levenberg algorithm, which optimizes parameter values by minimizing the sum of the squared residuals. We tested different binding models (different rate equations) in the curve fitting procedure, and the model best describing the experimental data was a conformational change model according to Equation 1.
The receptor (A) first forms an unstable complex (AB) with the peptide (B) and then undergoes a conformational change that leads to a more stable complex (AB*). To keep the model simple the complex AB* can only dissociate to free A through prior conversion to AB. The apparent K D values in Table I are calculated from the different rate constants as follows: Mammalian Two-hybrid Interaction Assay-COS-7 cells were maintained in Dulbecco's modified Eagle's medium supplemented with 10% fetal bovine serum, 1% penicillin, and 1% streptomycin (Life Technologies, Inc). The cells were plated onto 6-well plates in phenol red-free Dulbecco's modified Eagle's medium (Life Technologies, Inc) supplemented with 10% charcoal-stripped fetal bovine serum, 1% penicillin, and 1% streptomycin. At 50% confluency, cells were transfected with plasmids using FuGene (Roche Molecular Biochemicals) according to the manufacturer's instructions. The transfections were performed using 500 ng of the reporter UAS-tk-luc, 50 ng pM, pM-TRAP220 or pM-TIF2, and variable amounts of pVP16-ER␣ or pVP16-ER␤ as indicated in the figure legend. Empty expression vector was added to equalize total transfected plasmid DNA concentrations. Four hours after transfection, 17␤-estradiol was added to a total concentration of 10 nM to the appropriate wells. Each type of transfection was done in triplicate, and 24 h after transfection the cells were harvested and the cell extracts were analyzed for luciferase activity.

RESULTS
Previous work has suggested that the receptor binding characteristics of TRAP220 are similar to those of other LXXLLcontaining coactivators (19,20,26,28,29,35,36). Particularly, we have shown that the binding of the NR-interacting LXXLL domains of TRAP220 and the p160 coactivator TIF2 to isolated receptor AF-2 domains as well as to full-length TR-RXR heterodimers occurred competitively in vitro, suggesting similar binding surfaces on these receptors and comparable affinities (20). When analyzing binding of GST-TRAP220 to ER␣ dimers in EMSA assays under conditions similar to those in our previous study (20), we could not observe supershifts that would have supported ternary complex formation (Fig. 1B, upper). As a control for functional TRAP220, the same purified protein easily supershifted TR/RXR homo-or heterodimers (Ref. 20 and data not shown), and for functional ERs, ternary complex formation was easily achieved using a comparable GST-TIF2 protein. Furthermore, the failure to detect high affinity TRAP220 binding was irrespective of what source of ER protein was used (in vitro-translated, mammalian cell extract, baculovirus expressed; data not shown). Surprisingly, when performing similar binding studies on ER␤ dimers, we easily observed ternary complex formation (Fig. 1B, lower). Furthermore, as previously seen on TR/RXR heterodimers (20), TRAP220 could efficiently compete with TIF2 for binding to ER␤ but not to ER␣ (data not shown).
At least two possibilities may account for the apparent bind-ing preference of TRAP220 to ER␤: either this is a feature of the DNA-bound ER dimer or, alternatively, might reflect distinct binding affinities intrinsic to the ER AF-2 domains in solution. To address this issue we performed pull-down assays using GST-TRAP220 and in vitro translated ERs (Fig. 1C). As the pull-downs (Fig. 1C) demonstrate, ER␤ bound equally well to GST-TRAP220 and GST-TIF2, whereas ER␣ only interacted with GST-TIF2 under our conditions. This supports the binding preferences seen in DNA-dependent assays (Fig. 1B) and indicates that the low affinity of TRAP220 to ER␣ can be observed both on and off the DNA. Interestingly, we observed similar binding preferences using GST-NR-box peptide fusion proteins as indicated (Fig. 1A), suggesting that specificity is mediated largely by the LXXLL motifs within the ER-interaction domains (see experiments in Fig. 3).
To determine whether binding preferences could be observed under in vivo conditions we performed mammalian two-hybrid assays using GAL4-tagged TRAP220 and TIF2, respectively, and VP16-tagged ER fusion proteins ( Fig. 2A). As judged from the concentration-dependent increase in reporter gene activity, ER␣ clearly showed preferential TIF2 binding, whereas ER␤ interacted equally well with TRAP220 and TIF2. Importantly, the p160 member AIB1 (SRC3) behaves as TIF2 under our conditions (data not shown), providing an additional control for the experimental set-up and indicating that the features of TIF2 may be representative also for the other p160 coactivators. These data provide support for the binding specificity observed under in vitro conditions and further indicate that interaction determinants are independent of DNA binding and reside within the ER DEF region (ligand-binding domain/AF-2 domain). Additionally, we performed co-immunoprecipitations from cell extracts expressing GAL4-ER fusion proteins and TRAP220 using an anti-GAL4 antibody (Fig. 2B). As judged from subsequent Western blot analysis using a specific TRAP220 antiserum (20), a 220-kDa protein could be precipitated by GAL4-ER␤ and, to a lesser extent, GAL4-ER␣, but not with GAL4 alone (Fig. 4A). Similar results were observed by co-precipitation of the fusion proteins used in the two-hybrid assay, i.e. GAL4-ER␤ was more potent than GAL4-ER␣ in precipitating VP16-TRAP220 fusion proteins from cell extracts,

FIG. 1. Analysis of TRAP220 and TIF2 binding to ERs on and off DNA.
A, schematic representation of TRAP220 and TIF2 LXXLL domains and peptides. B, EMSA: using purified baculovirus-expressed ER proteins and purified GST proteins as described under "Experimental Procedures." Arrowheads indicate the position of DNA-bound ER dimers and 17␤-estradiol (E 2 )-induced ternary complexes. C, GST pull-down assay: 35 S-labeled ERs were analyzed for binding to GST proteins in a pull-down assay as indicated in the presence or absence of 1 M 17␤-estradiol (E 2 ). 20% of input and eluted samples were analyzed by SDSpolyacrylamide gel electrophoresis and visualized by autoradiography.
whereas VP16-TIF2 apparently showed no ER preferences (data not shown).
Previous evidence has indicated that TRAP220 binds RXR via the first LXXLL motif (NR-Box1) whereas TR, VDR, and the peroxisome proliferator-activated receptor (PPAR) prefer the second LXXLL motif (NR-Box2) (26,28,36). To address which of the two motifs contribute to ER specificity, we next analyzed interactions using GST-ERs and in vitro-translated TRAP220 (Fig. 3) and TRAP220 variants carrying AF-2 binding-deficient point mutations where critical leucine residues at ϩ1 and ϩ4 positions were changed to alanine (31). As a control, we included either the GST-TIF2 NR-interaction domain (20,30) or the GST-TIF2 NR-Box2 peptide, which are known from previous studies to bind with high affinity to ERs (8,31). In the reverse pull-downs (Fig. 3A) and in the EMSA (Fig. 3B), the inclusion of NR box mutants confirmed that at least one of the two motifs is required for interaction with ER␤. These experiments also indicated a higher affinity of the first motif. In this respect, ER␤ displays features probably similar to RXR (28) but different from TR and VDR (26,28). Because none of the LXXLL peptides nor mutants bound efficiently to ER␣, we conclude that the ER␤ preference is specified by both, or perhaps the first LXXLL motifs within the central NR interaction domain. It further indicated that regions of the TRAP220 protein situated outside the LXXLL motifs are not sufficient for binding to ERs.
To more quantitatively examine TRAP220 and TIF2 interactions to ERs, SPR analysis was performed using a BIAcore instrument (see "Experimental Procedures."). Fig. 4 illustrates the binding of purified ER␤ in the presence or absence of estradiol to the NR box domains of TRAP220 (A) and TIF2 (B) (identical to those used before in EMSA and pull-down analysis, see Fig. 1). Estradiol treatment clearly increased the binding affinities of ER␤ to both TRAP220 and TIF2, indicating that the purified proteins are functional and that the liganddependent association of ER␤ and coactivators can be seen in this kind of experiment. To study the NR interacting parts of TRAP220 and TIF2 in more detail we used an approach where biotinylated 14-mer peptides containing NR box motifs (LXXLL) from TIF2 or TRAP220 were captured via streptavidin to the chip surface. Fig. 5A demonstrates overlaid sensograms of injections of unliganded ER␣ or ER␣ liganded with estradiol and raloxifene, respectively, over TIF2 NR-Box2 peptide. Thus, a 14-mer peptide containing the NR box motif is sufficient for binding to ER␣. Further, the result shows that binding of agonist to ER␣ enhances the affinity of ER␣-peptide interaction, whereas the partial agonist/antagonist raloxifene decreases the affinity. Similar differences in binding, dependent on ligand status, were seen with ER␤ (data not shown). In a control study, no ER binding over background was observed using a non-LXXLL peptide with a randomized sequence (data not shown). In Fig. 5B the concentration-dependent association of estradiol-bound ER␣ to TRAP220 NR-Box2 is shown together with the best calculated fit. Similar binding studies were made for all the different TIF2 and TRAP220 NR box peptides using 5-6 different ER concentrations ranging from 2.5 to 200 nM. Affinity determination analysis was then performed using BIAevaluation software. The curve fitting analysis showed that the data could not be described using a simple 1:1 binding model. The best fit for all peptide-ER interactions FIG. 3. Analysis of TRAP220 LXXLL motifs. TRAP220 interacts predominantly with ER␤ through both NR boxes. A, schematic representation of TRAP220 NR box wild-type and mutant proteins. B, 35 Slabeled TRAP220 proteins were incubated at 4°C for 2 h with GST, GST-hER␣EF, or GST-hER␤EF coupled to glutathione-Sepharose beads in the presence or absence of 1 M 17␤-estradiol (E 2 ). 20% of input and eluted samples were analyzed by SDS-PAGE and visualized by autoradiography. C, EMSA analyzing ternary complex formation of ER␤ and purified TRAP220 proteins was performed under conditions essentially identical to those described in the legend to Fig. 1 and under "Experimental Procedures. "   FIG. 2. Analysis of TRAP220 and TIF2 binding to ERs in mammalian cells. A, mammalian two-hybrid assay: COS-7 cells cotransfected with 200 ng of GAL4-TRAP220 or GAL4-TIF2 and the indicated amounts of VP16-ER expression plasmids were cultured in serum-free medium in the presence or absence of 17␤estradiol (E 2 ) for 24 h and after lysis analyzed for luciferase reporter gene activity. B, whole-cell extract from COS-7 cells transfected with GAL4-ER and fulllength TRAP220 expression plasmids were precipitated with an ␣-GAL4 antibody and subsequently analyzed for the presence of coprecipitated TRAP220 by Western blot (see "Experimental Procedures").
tested was obtained using a conformational change model, where the receptor first forms an unstable complex with the peptide and then undergoes a conformational change that leads to a more stable complex. The apparent affinities listed in Table I show that the TIF2 peptides have higher affinity for ER than the TRAP220 peptides, with the TIF2 NR-Box2 having the highest affinity. Furthermore, the SPR analysis shows that both TRAP220 and TIF2 have binding preferences for ER␤, but the data also show that the affinity differences between ER␣ and ER␤ binding are higher in the case of TRAP220 peptides than for the TIF2 peptides. Thus, the BIAcore studies quantitatively support the low affinity of TRAP220 NR box peptides to ER␣. In particular, we were not able to detect any binding of ER␣ to TRAP220 NR-Box1 peptide (Table I), which is interesting because Box1 seems to be critical for binding to TR, VDR, and PPAR (26,28,36). However, we could demonstrate interaction with that peptide to ER␤, which would imply that ER␤ binds to a slightly different binding surface than ER␣ or that ER␤ actually binds much better to that part of the coactivator than ER␣.
To directly prove that the ER␤ preference of TRAP220 is largely, if not exclusively, determined by its LXXLL motifs, we replaced the first TRAP220 motif (14 mer) with the second motif of TIF2, which displays the highest ER affinity among all three motifs (see Fig. 5). When analyzing the corresponding GST-TRAP220 proteins carrying the substituted TIF2 motif in EMSA, ternary complex formation with ER␣ could now be observed (Fig. 6). Supershifts were induced only in the presence of ER agonists (estradiol and genistein) but not in the presence of the antagonist tamoxifen (data not shown). We conclude that a single NR box substitution, more specifically substitution of 9 residues adjacent to the conserved ILXXLL core, is sufficient to convert the TRAP220 NR interaction domain into a domain resembling the ER binding features of TIF2.
The obvious existence of ER subtype-specific LXXLL motifs suggested slightly different conformations within the C-terminal region of ER␣ and ER␤, respectively. Sequence comparison reveals that the most obvious difference between the two ERs is not the AF-2 domains (E-region) but the C-terminal F-domains. Intriguingly, most LXXLL binding experiments described here and in related studies include this ER domain. To analyze whether the F-domain plays a role in determining coactivator specificity, we performed pull-down and EMSA experiments as described above but under inclusion of C-terminal mutated ER proteins (Fig. 7). As shown in the pull-down assay (Fig. 7A), the expected TRAP220 binding profile was observed with the EF domains, whereas the ER␣ E-domain alone apparently interacted well with GST-TRAP220, indicating that deletion of the ER␣ F-domain was sufficient to increase the binding affinity to TRAP220. Under these conditions, no binding differences were seen with regard to TIF2 binding, and no differences were seen with ER␤ in the presence or absence of the ER␤ F-domain. To analyze the issue under EMSA conditions, we made a chimeric ER␣ carrying the ER␤ F-domain and compared its coactivator binding ability to wild-type ER␣ (Fig. 7B). Clearly, substitution of the F-domains was sufficient to increase the affinity of ER␣ to TRAP220.

DISCUSSION
The Role of TRAP220 LXXLL Motifs in Mediating ER Subtype Selectivity-We have shown using multiple experimental approaches that the coactivator TRAP220 displays binding preferences to ER␤ over ER␣, and we have provided evidence that the specificity is largely determined by the TRAP220 LXXLL motifs. Our findings are clearly supported by recent LXXLL peptide selection and binding studies using ERs (37). These have indicated the existence of three classes of motifs with distinct binding preferences to nuclear receptors including ER␣ and ER␤. Intriguingly, p160s and TRAP220 contain motifs with homology to class I and II peptides, respectively. In support of this a related study (38) shows that many TR␤selected LXXLL peptides had proline at the Ϫ2 position and a hydrophobic residue at Ϫ1 position. Notably, these class II peptides bound poorly to ER␣ but well to ER␤. These features are characteristic for the TRAP220 motifs and, interestingly, also for another recently cloned coactivator called RAP250/ ASC-2/AIB3 (Ref. 39 and data not shown). We have confirmed these preferences by increasing the affinity of TRAP220 to ER␣ by replacing the first TRAP220 motif with the second TIF2 motif, which is known to be a high affinity binding motif (see Fig. 6 and Table I).
Differences between ER␣ and ER␤ TRAP220 Binding Surfaces-We have provided evidence that TRAP220 preference is retained within the ER EF domains. This is not unexpected because the majority of the known coregulators interact predominantly with this regulatory ligand binding and activation domain. Further, TRAP220, unlike p160 members and p300/ CBP, apparently has no additional LXXLL-independent binding sites on the receptor N-terminal AF-1 regions (Ref. 40 and data not shown). Although AF-2 residues that are predicted to directly contact the LXXLL motifs are identical in the two ERs (6 -8), their ligand-binding domains (E-domain) substantially differ in primary sequence. As seen in the crystal structures of the E-domains, agonist-bound ER␣ and ER␤ have a similar fold and display no obvious differences with regard to the coactivator binding groove (6 -8). Therefore, the specificity constraints might involve contacts between residues adjacent to the LXXLL core motif and possibly additional receptor regions (41)(42)(43). Although it has been suggested that the coactivator binding groove may be functionally different between the two ERs (37), our findings indicate that it is the ER␣ F-domain that somehow interferes with TRAP220 binding (see Fig. 7). This is consistent with the suggestion that the F-domain could have a role in modulating transcriptional activity via cofactor binding (44 -46). Because all current ER structures lack the F-domain (6 -9), the precise orientation of the F-domain in relation to helix 12 and the entire AF-2 surface remains enigmatic. In support of a modulatory function of the F-domain is our failure to observe direct binding of TRAP220 or other cofactors to the isolated F-domain (data not shown). It is thus likely that intramolecular as well as intermolecular domain communication may modulate the activities of the two ERs and that cofactor preferences are in part dictated by their unique F-domains.
Implications of Differential TRAP220 Recruitment to ER␣ and ER␤-Previous studies clearly indicated that the activation domains of ER␣ and ER␤ are not identical (47)(48)(49). Although distinct transcriptional profiles may in part be mediated by differences in their N-terminal AF-1 domains (48,50), the ER␤ AF-2 has been shown to be extremely active in cell types (e.g. HeLa cells) in which the ER␣ AF-2 was essentially transcriptionally inactive (47,51). Thus, it is possible that certain AF-2 selective coactivators will have more impact on ER␤ function than on ER␣. Because p160 coactivators as well as p160-derived LXXLL peptides do not display subtype selectivity with estradiol-bound ERs (Refs. 40, 52, and this study), AF-2 selectivity may depend on coactivators such as TRAP220 or RAP250, which carry a different class of LXXLL motifs. Additionally, there is an increasing number of corepressors that contain LXXLL motifs that antagonize ER activation (31,32,34,53). For example, the orphan receptor and AF-2 corepressor SHP seemed to inhibit ER␤ more efficiently than ER␣ (31,34). Because net transcriptional activation depends on the ratio between competing coactivators and corepressors, it is conceivable that transcriptional responses of ER␣ and ER␤ may vary in different cellular environments and may not simply correlate to affinity differences seen with individual cofactors. However, the relative contribution of such individual cofactors may increase under conditions of altered expression. This could be important in ER-containing reproductive cancers where gene amplification and protein overexpression of TRAP220 and AIB-1/SRC3 have been found (54,55). Interestingly, whereas AIB-1 overexpression correlated with ER␣ expression, no such correlation was found in tumor cells where TRAP220 was overexpressed. In light of our findings it should be interesting to determine the ER␤ expression status in those cells.
The existence of TRAP220 binding preferences of ER␣ and ER␤ has implications considering their ability to form homo-as well as heterodimers (56,57). In analogy to recent studies  6. Replacement of TRAP220 NR-Box1 increases the affinity to ER␣. A, schematic structure of the wild-type TRAP220 NR box domain and the mutated version carrying the TIF2 NR-Box2 motif. B, EMSA analyzing ternary complex formation of ER␣ with GST-TRAP220 wild-type and mutated protein under conditions essentially identical to those described in the legend to Fig. 1 and under "Experimental Procedures." FIG. 7. Interference of the ER␣ F-domain with TRAP220 binding. A, pull-down assays using 35 S-labeled ER EF, or E domains and GST-TIF2 or TRAP220 proteins were essentially as described in the legends to Figs. 1 and 3. B, EMSA analyzing GST-TRAP220 or TIF2 ternary complex formation on ER␣ or chimeric ER␣/␤ carrying the ER␤ F-domain under conditions described in the legend to Fig. 1. comparing p160 and TRAP220 interactions on several RXR heterodimers (20,28,36,58,59), each of the ER subunits may be bound to individual coactivator molecules and thus possibly link subunits of distinct coactivator complexes. Potentially, ER heterodimers might be more flexible in responding to changes in the cofactor environment, i.e. in cancer cells that overexpress individual coactivators. The poor in vitro binding capability of TRAP220 (this study) or of the entire TRAP/DRIP complex (17,27, 60) to ER␣ does not necessarily contradict functional studies that indicate a requirement of TRAP220 in estrogen-induced gene activation (21,35). Because these studies only assess ER␣-dependent activities, future tasks will quantitatively re-evaluate these with inclusion of the ER␤. Experimental difficulties currently exist for quantitatively assessing the relative importance of individual coactivators for each of the two ERs in vivo. The fact that we observed binding of TRAP220 to ER␣ in the mammalian two-hybrid assay is not contradictory to the more pronounced quantitative differences seen in the in vitro binding assays. Two-hybrid assays are very sensitive but have a number of disadvantages when compared with biochemical protein-protein interaction assays. First, the actual protein concentration in the nucleus is unknown and could be subject to variation because of cellular factors. Second, twohybrid assays rely on transcriptional activation. Because both ERs and TRAP220 are transcription factors that communicate with the cellular transcriptional machinery, reporter gene activation may not always directly correlate with protein-protein interaction strength. However, promising attempts to gain more insights into the in vivo situation have been described very recently in two studies that analyze the temporal recruitment of distinct coregulators (61,62). Intriguingly, both studies show that TRAP220 and TRAP220-related LXXLL class II peptides were recruited to nuclear receptors significantly later than p160 coactivators and class I/III peptides after hormone treatment. These data support our finding that TRAP220 and TIF2 are differentially recruited by ERs. With regard to the complexity of the in vivo situation, it is interesting to recall that we have previously demonstrated binding of TRAP220 to CBP, perhaps indicating that TRAP220 has additional functions independent of the TRAP complex (20). Alternatively, TRAP220 could be indirectly recruited via binding of other activation domain binding TRAP subunits, for example via TRAP170 or TRAP80 (63,64).
Various factors have been suggested to contribute to the remarkable tissue-selective action of estrogens and selective ER modulators (SERMs) (65): first, different tissue distribution and ligand selectivity of the two ER subtypes; second, through different sets and levels of coregulators; and third, through different stabilizing effects of coregulators on ligand binding. In light of the findings presented here and in previous peptideselection studies (37,38,66,67) we suggest that coregulator selectivity of the ER subtypes is an additional layer of specificity that influences the transcriptional response in estrogen target cells.