Definition of the Surface in the Thyroid Hormone Receptor Ligand Binding Domain for Association as Homodimers and Heterodimers with Retinoid X Receptor*

Thyroid hormone receptors (TRs) bind as homodimers or heterodimers with retinoid X receptors (RXRs) to DNA elements with diverse orientations of AGGTCA half-sites. We performed a comprehensive x-ray crystal structure-guided mutation analysis of the TR ligand binding domain (TR LBD) surface to map the functional interface for TR homodimers and heterodimers with RXR in the absence and/or in the presence of DNA. We also identified the molecular contacts in TR LBDs crystallized as dimers. The results show that crystal dimer contacts differ from those found in the functional studies. We found that identical TR LBD residues found in helices 10 and 11 are involved in TR homodimerization and heterodimerization with RXR. Moreover, the same TR LBD surface is operative for dimerization with direct repeats spaced by 4 base pairs (DR-4) and with the inverted palindrome spaced by 6 base pairs (F2), but not with TREpal (unspaced palindrome), where homodimers appear to be simply two monomers binding independently to DNA. We also demonstrate that interactions between the TR and RXR DNA binding do-mains stabilize TR-RXR heterodimers

from those that form heterodimers. For example, changing leucines to arginines in H11 of chicken TR␣1 (L365R and L372R) or human (h) TR␤1 (L428R) disrupted TR-RXR heterodimers, but not TR homodimers on TREpal (the unspaced palindrome), DR-4 elements (direct repeat elements spaced by 4 base pairs), and inverted palindromes (11,17). Furthermore, single mutations in arginine 316 to histidine, arginine 338 to tryptophan, and arginine 429 to glutamine in hTR␤1 disrupt homodimers but not heterodimers (18). Whereas these data indicate differences in requirements for heterodimerization and homodimerization, the usefulness of these mutations for defining the TR homodimerization or heterodimerization surfaces is limited, because these mutations are in the interior core structure of the TR.
Several investigators demonstrated that TR-RXR heterodimers interact in solution through a LBD heterodimer interface. It has been proposed that these heterodimers bind DR-4 elements by initial formation in solution of the LBD-LBD interface, which leads, upon DNA binding, to the formation of a second interface involving the second zinc finger of the RXR DBD and the first zinc finger of the TR DBD (19 -22). Indeed, crystallographic studies of isolated TR-RXR DBDs bound to DR-4 supported the existence of a DBD heterodimer interface in which the RXR occupies the 5Ј half-site, whereas TR occupies the 3Ј half-site (23). However, there is no functional evidence that DBD surfaces participate in complexes formed by full-length TR homodimers or TR-RXR heterodimers bound to DR-4 or other elements. Also, if DBD surfaces participate in dimerization, it is unclear if the LBD and DBD surfaces form sequentially, nor is it known whether a LBD surface is used by TR homodimers to bind any DNA element. To date, no x-ray crystal structural data are available for TR LBD homodimers or heterodimers bound to DNA, and no comprehensive mutation analyses are available that locate the homodimer or heterodimer TR LBD interface.
RXR␣ (24), ER␣ (25), ER␤ (26), PPAR␥ (27), and PR (28) LBDs crystallized as homodimers. Crystals of LBD heterodimers were recently reported for the RXR-RAR (29), where the interface is similar to that observed in the RXR␣, ER␣, ER␤, and PPAR␥ homodimers (24 -27) and for the PPAR␥-RXR␣ heterodimer (30), where the interface is slightly asymmetric with the PPAR LBD rotated ϳ10°relative to that expected from the C2 symmetry axis of the RXR LBD. The residues in H10 and proximal region of H11 form more than 75% of the total LBD dimer surface found in homodimers or heterodimers, whereas PR homodimers contacts are in the distal part of H11 and H12. The PR dimerization domain also covered a much smaller area (700 Å 2 ) than that for the ER␣ (1703 Å 2 ), which led the authors to question the physiological relevance of the observed dimer (31). In addition, the structures of the RXR␣ LBD bound to 9-cis retinoic acid (32) and of the PPAR␦ LBD bound to fatty acids (33) were solved and showed two monomers per asymmetric unit. However, the packing contacts found between these monomers did not resemble those present in the unliganded RXR␣ LBD or in the PPAR␥ LBD homodimers suggesting that a different dimeric conformation may be found with these receptors. Altogether, these structural data indicate that in the absence of DNA, nuclear receptors form dimers by engaging H11 and possibly other helices in diverse macromolecular interactions that may give rise to varied dimeric conformations.
The TR LBD structure reported to date is formed by TR monomers. Thus it is not apparent which TR residues participate in homodimer or heterodimer interactions and whether they vary depending on the nature of the TRE. Furthermore, structural data do not demonstrate functional TR interactions.
In the studies reported here we performed a comprehensive x-ray crystal structure-guided mutation analysis of the TR LBD surface to map the functional interface for TR homodimers and heterodimers with RXR in the absence and/or in the presence of DNA. This approach employed the x-ray crystal structures of the TR-RXR DBDs and TR LBD (16,23) and was previously used by us for definition of a TR coactivator-binding surface (34). We also identified the molecular contacts in TR LBDs crystallized as dimers. The results show that crystal dimer contacts differ from those found in the functional studies. We found that identical TR LBD residues are involved in TR homodimerization and heterodimerization with RXR. Further, the same LBD surface is operative irrespective of the orientation and spacing of the half-sites for DR and inverted palindromic elements, but not for the TREpal homodimers, which appear to be simply two monomers binding to DNA. We also demonstrate that interactions between the DBDs stabilize TR-RXR heterodimers on DR-4. The dimer interface can be functional in the cell, because disruption of key residues impairs transcriptional activity of TRs mediated through association with RXR LBD linked to GAL4 DNA-binding domain.

MATERIALS AND METHODS
Construction of TR Mutants-TR mutants were created by ligating double-stranded oligonucleotides encoding the mutant sequence into the pCMX vector that encodes the full-length 461 amino acid hTR␤1 sequence. Some mutations were created using QuickChange site-directed mutagenesis kits (Stratagene). The mutated sequences were verified by DNA sequencing using Sequenase Kits (Stratagene). These TR mutants changed the amino acids in the hTR␤1 surface that were selected for scanning mutagenesis using the x-ray structure of the TR␣ LBD as a guide to localize the surface residues on hTR␤1 (16,34). Subsequent solution of the hTR␤ LBD structure (35) revealed that the mutations were placed correctly. Most mutations changed the resident amino acid to arginine, reasoning that bulky and charged residues on the surface were more likely to disrupt TR association with other proteins over a wider range without reducing TR solubility. Each mutant maintained the overall integrity of TR as judged by the ability to bind [ 125 I]T 3 with K d values between 4 and 174% of that for the native receptor and/or to bind the glucocorticoid receptor interacting protein 1 (GRIP-1) fused to GST as effectively as the wild-type TR. We initially tested, as reported previously, widely separated locations to probe the entire LBD surface for dimerization (34). Subsequently, a series of mutations were performed along H10 and H11 in a region where residues were identified that disrupted dimerization to define better the structural requirements for dimerization. Mutations are represented by the single-letter amino acid abbreviation of the native residue followed by its position number in the sequence and the mutated residue abbreviation. The affinity of the wild-type TR␤1 is 60 Ϯ 7 pM.
The relative affinity was determined by dividing the wild type by the mutant K d . The 62 mutants included and their relative affinities are: E217R (91%), W219K (not done (ND)), E227R (109%), K242E (68%), A TR DBD mutant was created by changing the Asp 177 , known to interact with RXR in crystallographic studies, to Ala. A RXR DBD mutant was made by changing all Arg residues previously shown in crystallographic studies to be involved in dimerization with the TR to Ala (R172A, R182A, and R186A, Ref. 23). We also found that this RXR mutant had three additional DBD changes (T168P, T170I, and Y193A) that did not interfere with RXR binding to DNA. An RXR LBD mutant was made by changing leucines 419 and 420 into arginines, positions analogous to TR mutants defective in dimerization (see "Results"). Crystallization hTR␤-Crystallization conditions for hTR␤ LBD with Triac has been described fully (35). Briefly, crystals are obtained by hanging drop vapor diffusion (12 h, 25°C) from a drop (1 l of TR␤ LBD solution at 9 mg/ml, 1 l precipitant solution) suspended above a reservoir composed of 1.0 M sodium acetate [NaH 3 OAc] and 0.1 M sodium cacodylate [NaCac], pH 7.2. Crystals were of space group P3121 (a ϭ 68.9 Å, c ϭ 131.5 Å) and contained 1 molecule of TR␤ LBD. The N-terminal His tag was not removed prior to crystallization.
rTR␣-Hexagonal crystals of rTR␣ LBD with T 3 were obtained using hanging drop vapor diffusion at 17°C from a drop containing a 1:1 mixture of ϳ10 mg/ml protein, 3 mM dithiothreitol with well solution containing 0.45 M (NH 4 )H 2 PO 4 and 0.2 M (NH 4 ) 2 SO 4 buffered with 0.1 M KH 2 PO 4 , pH 7.0. The space group was determined to be hexagonal, P6 1(5) 22 from oscillation data; the correct enantiomorph P6 5 22 was identified through molecular replacement using the model of the rTR␣ LBD (16). The crystals contain one monomer/asymmetric unit. The unit cell dimensions of these crystals are a ϭ b ϭ 108.64 Å, c ϭ 133.72 Å.

Structural Analysis
hTR␤-Structural analysis of hTR␤ LBD with Triac has been recently described fully (35). Briefly, crystals were transferred into a cryoprotectant composed of 30% glycerol, 1.4 M NaAc, and 100 mM NaCac, pH 7.2, then suspended in a nylon loop attached to a mounting pin, and flash frozen (liquid nitrogen). Diffraction data were measured to 2.4 Å using synchrotron radiation at Stanford Synchrotron Radiation Laboratory beamline 7-1 (l ϭ 1.08Å) and processed using HKL. A molecular replacement solution was found using AMORE from the model of the rTR␣ LBD, with ligand omitted (36). Refinement with CNS produced a model with an R cryst of 24% and an R free of 26%.
rTR␣-A cryoprotectant composed of 30% glycerol, 0.5 M (NH 4 )H 2 PO 4 , 0.2 M (NH 4 ) 2 SO 4 , and 0.1 M KH 2 PO 4 , pH 7.0 was used, and a complete data set measured at SSRL (l ϭ 1.08Å) and processed using HKL. A molecular replacement solution for the Form II crystal was found using AMORE from the rTR␣ LBD model with ligand removed (16). The resulting electron density maps show interpretable density for the T 3 ligand; furthermore, an anomalous Fourier map calculated from the molecular replacement phases identifies the sites for the iodine scatterers consistent with the ligand density. Refinement with CNS produced a model with an R cryst of 25% and an R free of 28%.

Glutathione S-Transferase Assay
For these experiments full-length hRXR␣ was prepared in Escherichia coli strain HB101 as a fusion protein with glutathione S-transferase (GST) as per the manufacturer's protocol (Amersham Pharmacia Biotech). The binding experiments were performed by mixing glutathione-linked-Sepharose beads containing 10 g of GST⅐hRXR␣ fusion proteins (Coomassie Plus protein assay reagent, Pierce) with 1-2 l of the 35 S-labeled wild-type or mutant hTR␤1 (30 fmols) in 150 l of binding buffer (20 mM HEPES, 150 mM KCl, 25 mM MgCl 2 , 10% glycerol, 1 mM dithiothreitol, 0.2 mM phenylmethylsulfonyl fluoride, and protease inhibitors) containing 2 g/ml bovine serum albumin for 1.5 h. Beads were washed three times with 1 ml of binding buffer, and the bound proteins were separated using 10% SDS-polyacrylamide gel electrophoresis and visualized by autoradiography.

Gel Shift Binding Assay
Binding of full-length hTR␤1 mutants to DNA was assayed by gel shift assays by mixing 30 fmols of radiolabeled TRs ([ 35 S]Met, DuPont) produced in a reticulocyte lysate system, TNT® T7 Quick (Promega), with ϳ600 fmols (10 ng) of nonradiolabeled DR-4 or F2 oligonucleotides or 120 fmols (2 ng) of the TREpal oligonucleotide and 2 g of poly(dI-dC) (Amersham Pharmacia Biotech) in a 20-l volume reaction as previously described (7). The binding buffer contained 10 mM NaPO 4, 1 mM MgCl 2 , 0.5 mM EDTA, 20 mM NaCl, 5% glycerol, 0.1% monothioglycerol. After 20 min at room temperature, the mixture was loaded onto a 5% nondenaturing polyacrylamide gel that was previously run for 30 min at 200 V. To separate the TR⅐DNA complexes, the gel was run at 4°C for 150 -240 min at 240 V, using a running buffer containing 6.7 mM Tris-base (pH 7.5 for a 10ϫ stock at room temperature), 1 mM EDTA, and 3.3 mM sodium acetate.
Cell Culture, Electroporation, and Luciferase Assays CV1 cells were maintained and subcultured in media DME H-21 4.5 g/l glucose, containing 10% newborn bovine serum, 2 mM glutamine, 50 units/ml penicillin, and 50 g/ml streptomycin. Transfection procedures were described previously (37). Briefly, cells were collected and resuspended in Dulbecco's phosphate-buffered saline (0.5 ml/1.5 ϫ 10 7 cells) containing 0.1% dextrose, and typically 10 g of reporter plasmid and 1 g of expression vector. Cells were electroporated at 350 V and 960 microfarads, transferred to fresh media, and then plated in 6-well plates. After incubation for 24 h at 37°C with ethanol, or 100 nM T 3 , the cells were collected and the pellets were solubilized by addition of 100 l of 0.25 M Tris-HCl, pH 7.6 containing 0.1% Triton X-100. For each transfection, 1 g of the pCMVhTR␤1 expression vector was cotransfected with the 2 g of the GAL⅐RXR (GAL4 DBD fused to hRXR␣ LBD) expression vector. The reporter gene contained five GAL binding sites upstream of the adenovirus E1b minimal promoter linked to luciferase coding sequences (LUC). The LUC activity was analyzed after the cells were incubated 18 -24 h with the hormone (Luciferase Assay System, Promega).

TR LBDs Form Crystals with Diverse Intermolecular Contacts-Crystal forms of LBD fragments of the TR␣ and TR␤
subtypes exhibit two distinct symmetrical intermolecular contacts. In TR␣, the individual monomers are rotated 180°with respect to each other, with an axis of symmetry parallel to H11, such that the helices cross with the junction near the mid-point of H11 (Fig. 1A). The contact buries a modest 550 Å 2 of surface area per monomer. The primary contacts are between residues in H11, with contributions from the C-terminal portion of H8. A cluster of hydrophobic residues in H11 marks the center of symmetry: Met 376 , Ile 377 , Ala 378 , and Cys 379 . At the periphery of the contact are electrostatic interactions between residues in H8 and H11: Lys 288 with Asp 326 and Glu 301 with Asp 373 .
In TR␤, the individual monomers are again rotated 180°with respect to each other, but the axis of symmetry is perpendicular to H11, producing an antiparallel LBD dimer (Fig. 1B Scanning Mutations Define a LBD Interface for TR-RXR Association in Solution-Because the hydrophobic contacts described above may reflect forces operative during crystallization and not represent a functional dimerization surface, we mapped the TR LBD surface that participates in dimer formation by mutating the TR LBD surface. These mutations were introduced ϳ7 Å apart to scan the TR LBD area involved in protein-protein interactions (34). As described under the "Materials and Methods," these mutations generally maintained overall receptor integrity and functions largely unrelated to dimer or heterodimer formation such as ligand and coactivator binding.
Sites important for TR-RXR interactions were first deter-mined by examining binding of TRs to GST⅐RXR in pull-down assays. Wild-type hTR␤1 bound to GST⅐RXR in a T 3 -independent manner ( Fig. 2A), whereas binding of wild-type or mutant TRs to GST alone was negligible (data not shown). Replacing amino acids in H10 (Glu 393 and Leu 400 ) and H11 (Pro 419 , Leu 422 , Met 423 , Thr 426 , and Met 430 ) with Arg greatly diminished TR interactions with RXR ( Fig. 2A). This was not observed with the mutants E390R in H10 and H416R and C434R in H11 ( Fig. 2A) or 62 other mutants described under "Materials and Methods" and spanning helices 1, 3, 5, 8, 9, and 12 (data not shown). Fig. 2B shows the location of the amino acids involved for TR association with RXR in solution, from the GST⅐RXR interactions with TR mutants. All of the TR mutants defective in forming heterodimers with GST⅐RXR bound T 3 and interacted with GST⅐GRIP1 as well as the wild-type TR (data not shown). Taken together, these results define a relatively small surface in H10 and in the amino-terminal portion of H11 of the TR LBD that is necessary for binding RXR in solution.

The LBD Surface That Forms TR-RXR Heterodimers in Solution also Forms TR Homodimers on DR-4 and F2 but Not on
TREpal-Because homodimers with full-length TRs are not observed in solution, we next sought to define the TR surface used by TR homodimers on DNA. We first used the DR-4 binding site, because this is the most commonly described TRE. As shown in Fig. 3A, we found that a subset of the mutants that inhibited formation of TR-RXR heterodimers in solution also blocked TR homodimer formation on DR-4. Thus, E393R and L400R in H10 (lanes 5 and 7, respectively) and P419R, L422R, and M423R in H11 (lanes 13, 15, and 17) nearly abolished formation of TR homodimers on DR-4. The mutants T426R and M430R in H11 that impaired formation of TR-RXR heterodimers in solution had minimal influences on TR homodimer formation (Fig. 3A, lanes 19 and 21). No other TR mutants of the 50 tested spanning helices 1, 3, 5, 8, 9, and 12 were defective in homodimer assembly on DR-4 (data not shown).
Because inverted palindrome DNA sequences have the highest capacity for binding TR homodimers (6), we used the F2 palindrome to examine whether the TR LBD surface that forms homodimers on F2 coincides with that mapped for homodimers on DR-4 and for TR-RXR heterodimers in solution. Fig. 3B shows that this is the case because L400R in H10 and L422R and M423R in H11 (lanes 7, 15, and 17) disrupted homodimers with F2. However, in contrast to DR-4, the E393R and P419R mutants (lanes 5 and 13) partially disrupt homodimers on F2.
These results demonstrate that the essential residues required for forming TR LBD homodimers with DR-4 or F2 are Leu 400 in H10 and Leu 422 and Met 423 in H11. However, a larger group of residues flanking this critical core of amino acids is required for forming TR homodimers with DR-4 compared with F2.
We also asked if the surface used for TR homodimers on DR-4 and F2 and for TR-RXR heterodimers in solution is also operative for forming TR homodimers on TREpal. None of these mutants (Fig. 3C) or other surface TR mutants tested (data not shown) blocked formation of TR homodimers on TREpal. These data suggest that TRs bound to TREpal are simply two monomers rather than a homodimer.
Taken together, these results show that the TR LBD surface involved in formation of TR-RXR heterodimers in solution and TR homodimers with DR-4 or F2 is similar and is not involved in TRs bound to TREpal. This surface contains a critical core of hydrophobic amino acids comprised of Leu 400 in H10 and Leu 422 and Met 423 in H11 that are required for all types of TR dimeric interactions. In H11, the residues flanking Leu 422 and Met 423 appear to be important for homodimers on DR-4 and for heterodimers in solution. These findings suggest that the dimer surface is centered on a set of critical hydrophobic residues, but that it is flexible and extends to adjacent residues (Fig. 3D).

The Same LBD Interface Mediates Formation of TR Homodimers and TR-RXR Heterodimers on DNA, but Heterodimers Are More Stable because of DBD-DBD Interactions-
The mutations were examined for effects on TR-RXR heterodimer formation on DNA. None of the TR LBD single mutants that impaired formation of TR homodimers on DR-4 (see Fig. 3A) or an additional double mutant L422R/M423R containing two residues on the dimerization surface (data not shown) disrupted TR-RXR heterodimers on DR-4. However, L422R weakened formation of TR-RXR heterodimers on F2 (Fig. 3B, lane 16) and abolished them on TREpal (Fig. 3C, lane  16). Because Leu 422 in the center of the surface is critical for TR-RXR heterodimers in solution and TR homodimers on DR-4 and F2, it is likely that the TR-RXR heterodimer surface on F2, TREpal, and even DR-4 is also similar but that TR-RXR heterodimers are more stable.
The greater stability of TR-RXR heterodimers on DR-4 could be in part because of DBD-DBD contacts as observed in the x-ray crystal structure of TR-RXR isolated DBDs (without LBDs) on DR-4 (23). However, the greater resistance to mutational disruption of the DR-4 bound TR-RXR heterodimers as compared with TR-TR homodimers could be because of the fact that only the TR subunit of the heterodimer pair contained a mutation whereas with TR-TR homodimers both subunits are mutated. To address this issue, we assayed TR-RXR heterodimer formation where both the TR and RXR subunits are mutated. For these experiments, we replaced two leucines, Leu 419 and Leu 420 , with arginines in the RXR LBD, which are residues analogous to the TR double mutant L422R/M423R that did not disrupt TR-RXR heterodimers on DR-4. As shown in Fig. 4A, the RXR L419R/L420R mutant markedly inhibited heterodimer formation with wild-type TR on DR-4 (lane 4). Mixing RXR L419R/L420R with TR L422R (Fig. 4A, lane 8) or L422R/M423R (data not shown) did not further disrupt heterodimers on DR-4. These findings indicate that those mutations in the LBDs of both subunits can reduce heterodimer formation on DR-4 and suggest that the RXR LBD dimer interface is similar to the one in the TR LBD. These findings together with the known structures of the TR (27)  allowed us to propose a model for the TR-RXR heterodimer, which is shown in Fig. 4B, depicting the major interactions between the hydrophobic residues in H10 and H11 of TR with H11 of RXR.
We also investigated if DBD-DBD interactions participate in TR-RXR heterodimers on DR-4 by using TR and RXR DBD mutations based on the crystal structure of the TR-RXR DBD dimers (23). The TR DBD mutant replaced the aspartic acid 177 with alanine and the RXR DBD mutant replaced the three arginines 172, 182, and 186 with alanines. The TR D177A formed efficient homodimers and heterodimers with either the wild-type RXR or the RXR DBD mutant on DR-4 (Fig. 4A, compare lanes 1-3 to 9 -11) indicating that these DBD mutations per se do not disrupt formation of homodimers or heterodimers on DR-4. In contrast, heterodimer formation was completely abolished by mixing the TR D177A with the RXR L419R/L420R LBD mutant (Fig. 4A, lane 12). Similarly, TR-RXR heterodimers did not form (Fig. 4A, lane 7) when the RXR DBD mutant was mixed with the TR LBD mutants L422R (that formed efficient heterodimers with the wild-type RXR; Figs. 3A, lane 16 and 4A, lane 6) or L422R/M423R (data not shown). By contrast, the RXR DBD mutant did not disrupt heterodimers with L422R on the F2 DNA (data not shown). Collectively, the data show that LBD interactions are dominant in formation of TR homodimers and heterodimers on DR-4 because disruptions in LBD, but not in DBD interactions, impair TR dimer formation. These results also indicate that DBD interactions stabilize TR-RXR heterodimers on DR-4, because DBD mutations can enhance the effects of LBD mutations. The data provide additional evidence that the LBD surfaces that form TR-RXR heterodimers on DR-4 are similar to those utilized by TR-RXR heterodimers in solution and by TR homodimers on DR-4.

Mutation of TR Hydrophobic Residues in Helix 11 Impairs
Transactivation in Vivo When RXR Is Linked to a Heterologous DBD-The TR LBD single mutants that impair assembly of TR homodimers on DNA and TR-RXR heterodimers in solution were assessed for their abilities to induce ligand-mediated transcription in cultured cells in a context where the interactions occur only through the LBDs, and the TR is not bound to DNA. We cotransfected GAL⅐RXR with TRs in CV-1 cells. In the presence of wild-type TR or any of the mutants that failed to impair the TR-RXR association in solution (E390R, H416R), T 3 stimulated 25-to 48-fold the reporter gene driven by five GAL binding sites. In contrast, transcriptional activity induced by either P419R or L422R mutants was completely absent in this context, and E393R and L400R displayed 10 and 35% activation relative to the wild-type TR, respectively. By contrast, the TR mutants T426R and M430R, which impaired TR-RXR association in solution, were able to activate transcription through GAL⅐RXR (Fig. 5). These data indicate that hydrophobic residues at the center of the TR LBD dimer interface can abolish transcriptional activities of TR linked to RXR LBD by presumably preventing their interactions. The data also suggest that other proteins in the cell stabilize the TR-RXR interactions for some TR mutants relative to those observed under cell-free conditions.

DISCUSSION
In the current studies, a structurally guided scanning mutagenesis of the TR and RXR LBD and DBD surfaces was used to infer the LBD surfaces that form TR homodimers and TR-RXR heterodimers. This approach has the advantage over studies using mutagenesis based on the linear receptor sequence in that mutations placed on the outside surface of the receptor are likely to disrupt local interactions with proteins and much less likely to disrupt receptor folding, which complicates interpretations of results. In support of this we found that the muta- FIG. 4. Characterization of heterodimer formation between TR and RXR on DR-4. A, gel shift assays contained 40 fmols (usually 1-3 l) of the in vitro translated 35 S-labeled wild-type hTR␤1 or LBD (L422R) or DBD (D177A) mutants mixed respectively with 3 l of either unlabeled unprogrammed lysate (lanes 1, 5,9) or with hRXR␣ wild type (lanes 2, 6, 10), hRXR␣ DBD mutant (lanes 3, 7, 11), or hRXR␣ LBD mutant L419R/L420R (lanes 4, 8, 12). B, model for the TR␤1-RXR␣ LBD heterodimer interface. The inset highlights the interactions between the hydrophobic residues Leu 400 in H10 (green) of TR with Leu 420 of H11 (dark blue) of RXR and Leu 422 and Met 423 in H11 of TR with Leu 419 in H11 of RXR.
tions affecting dimerization and/or heterodimerization retained other receptor functions such as ligand and coactivator binding. An additional advantage of the surface scanning approach is that analyses of multiple mutations demonstrate patterns and minimize erroneous conclusions derived from interpretations of results from a single mutant. These functional data can also determine whether the way receptors crystallize with purified proteins reflects the functional interactions in other conditions. Thus, we compared the interfaces found in two TR LBDs that crystallized as dimers with those determined from the functional studies. The data provide the first comprehensive analysis directed at mapping the LBD interface for forming TR homodimers and heterodimers with RXR. They suggest that a similar surface participates in formation of TR homodimers with DR-4 and F2 DNA and TR-RXR heterodimers in solution and with DR-4, F2, and TREpal.
The TR LBD crystals were obtained with both the rTR␣ and hTR␤ LBDs. In both structures, the TR monomers were rotated 180°with respect to each other. With the rTR␣, the helices crossed with a junction near the middle of H11, whereas with the hTR␤, the axis of symmetry is perpendicular to H11, producing an antiparallel LBD dimer in which one monomer lies between the end of H8 (opposite to that with the rTR␣) and H11 of the other monomer. The contacts bury a modest 550 Å 2 and 580 Å 2 of surface area per monomer for the rTR␣ and hTR␤ LBDs, respectively. These observations could imply a weak interaction because the bona fide protein-protein interface typically buries Ͼ 700 Å 2 of solvent-exposed area (41). In support of this notion, none of the mutations of residues of either H8 or H11 involved in these contacts, such as Asp 355 , Ala 433 , Cys 434 , and Ser 437 , disrupted TR-TR dimer formation on DNA, whereas mutations of other residues of H10 and H11 did affect these interactions. These observations, taken together, suggest that the contacts observed in the TR crystal structures are not involved in formation of TR-TR homodimers on DNA and also differ from the contacts reported in the structures of RXR and PR homodimers (24,28).
The scanning mutations were used first to define a TR-RXR LBD heterodimer interface when the TRs are soluble and RXRs are linked to GST (Fig. 2A). This method likely reflects interactions of the two receptor types in solution. TR homodimer interactions were not tested in this context, because they have not been observed in solution. Mutations in the TR LBD that blocked GST⅐RXR-TR interactions were located on a small surface comprising H10 and H11, (Fig. 2, A and B), whereas 30 mutations on other parts of the TR LBD surface failed to inhibit the reaction. Although this heterodimer surface differed from that observed in our TR-TR LBD homodimer crystals or of the PR homodimer (28), it resembles that observed in the structures for the RXR␣ (24), the ER␣ (25), the ER␤ (26), and the PPAR␥ (27) homodimers. In other studies the functional protein-protein interfaces derived from mutational data have been found to differ from those that form in the crystal. For example, about 30 side chains from growth hormone and its receptors make contact, but individual replacement of contact residues with alanine showed that only a central hydrophobic region accounts for more than three-quarters of the binding free energy (42). Furthermore, mutations in interface residues cause structural changes that redistribute the particular contributions of individual residues to the free energy of binding.
Interestingly, the hydrophobicity of residues central to the TR homodimer and TR-RXR heterodimer interactions, Leu 422 and Met 423 , is conserved in all the receptors, which exhibit that interaction in the crystal (RXR␣ (24), ER␣ (25), ER␤ (26), PPAR␥ (27)). By contrast, in the PR and the other steroid receptors with the exception of ER, these residues are hydrophilic or charged, resembling the mutations introduced in the TR that disrupt the interaction. Thus, the PR homodimer interaction could be representative of one class of LBD interactions, the classic "homodimer" receptors, whereas the interactions examined here are representative of the "heterodimer" receptors.
Data in support of the notion that the TR utilizes the same LBD surface for forming homodimers with DR-4 and F2 DNA as it uses for heterodimer formation on GST is based on the observation that some of the same mutations in this surface disrupted all of these interactions. Variations in the effects of the mutations in the three contexts were observed, and some of the mutations that affected the TR-RXR⅐GST interaction failed to affect the homodimer interactions. However, the overall pattern suggests at the least an overlapping surface, and the same core of critical hydrophobic residues is employed by TR to form homodimers and heterodimers with RXR. This pattern is similar to the one found in the functional interface of growth hormone-growth hormone receptor where a small and complementary set of contact hydrophobic residues maintains binding affinity, a property that may be general to protein-protein interfaces (42). The surface centers on the Leu 422 residue in TR H11, which disrupted TR homodimers on DNA and TR heterodimers formed in solution. Whereas the overall surface ex- FIG. 5. Specific effects of mutations on hTR␤1 transcriptional activation in CV-1 cells. Cells were cotransfected with 1 g each of expression vectors encoding a GAL⅐RXR LBD fusion protein, and wild-type or mutant hTR␤1, and 5 g of a reporter gene containing 5 GAL binding sites and encoding luciferase. After culturing with or without 100 nM T 3 , cell extracts were assayed for luciferase activity. T 3 activation was expressed as the percentage of wild-type TR. The wild-type or different hTR␤1 mutants are indicated. The data show a representative experiment, which was repeated 3-5 times for each TR mutant. tends from residues Pro 419 to Met 430 in H11 and from Glu 393 to Leu 400 in H10 (Fig. 2B), the individual contribution of these residues varies depending on the context in which the interactions take place. Thus, for TR homodimers bound to F2, only Leu 400 in H10 and Leu 422 and Met 423 in H11 appear to be the critical residues under the conditions studied, whereas flanking residues such as Glu 393 in H10 or Pro 419 in H11 are critical for TR homodimer formation on DR-4. Finally, an even larger set of flanking residues such as Thr 426 and Met 430 are critical for TR-RXR heterodimers in solution. That similar homodimerizing and heterodimerizing interfaces are used for TR LBDs to bind to DR-4 and F2 elements provides support for the model that LBD dimer interfaces are independent of orientation of TRE half-sites. Another interesting point derived from these studies is that contrary to the prevailing model, the receptors do not need to form dimers in solution before DNA binding. Thus, TR mutants incapable of interacting with RXR in solution, such as E393R, L400R, P419R, M423R, T426R, and M430R formed stable TR-RXR heterodimers with DNA.
Whereas mutations readily blocked TR binding to DR-4 and F2 elements, no mutation, including L422R, disrupted homodimers on TREpal. These data suggest that the "dimers" on TREpal consist of independent binding of two monomers. This notion is supported by the observation that "homodimers" on TREpal are formed very weakly as compared with those on DR-4 and F2 (6,7). Another indication that TR "dimers" on TREpal are really two monomers is based on the fact that T 3 increases TR binding as monomers with several TREs and as "homodimers" on TREpal (6, 7), whereas T 3 decreases TR binding as homodimers with DR-4 or F2 elements (6,7,43,44). The data indicate that at least part of the TR surface that mediates both dimerization on DR-4 and F2 and heterodimerization on GST⅐RXR also mediates TR-RXR heterodimerization on DR-4, F2, and TREpal elements. The first evidence supporting this conclusion is that L422R in the center of the surface weakens formation of TR-RXR heterodimers on F2 (Fig. 3B, lane 16) and abolishes them on TREpal (Fig. 3C, lane 16).
The finding that none of the TR LBD single mutants blocked TR association with wild-type RXR with DR-4 suggested the possibility that DNA may be stabilizing such interactions. This is supported by the observation that heterodimerization on DR-4 was impaired when the TR L422R mutant was incubated with the RXR DBD mutant. These findings argue that interactions among TR and RXR DBDs play a role stabilizing heterodimer formation on DR-4 and indicate a dominant role for the LBDs relative to DBDs in stabilization of both homodimers and heterodimers on DR-4. Our results further suggest that the analogous surface on the RXR mediates the interaction of this receptor with the TR. Thus, a double mutation in the RXR LBD, L419R/L420R greatly reduced formation of heterodimers with wild-type TR on DR-4. An analogous double mutation on the TR LBD, L422R/L423R, did not affect heterodimer formation with DR-4 to the same extent (data not shown), suggesting that RXR has a more dominant role in TR-RXR heterodimer formation and explain why, in contrast to TR homodimers, TR-RXR heterodimers are formed in solution and are not disrupted by ligand binding.
The formation of TR homodimers on DR-4 might be because of DBD-DBD interactions, or might simply result from placement of the LBDs together after independent binding of two receptor monomers to DNA. Our findings that isolated TR LBD mutations inhibited TR homodimers, and that TR DBD mutations did not, suggest that DBD-DBD interactions in TR homodimers, if they exist, are not functionally important for processes requiring dimers. These observations are consistent with results of other investigators that failed to detect forma-tion of TR DBD dimers on DR-4 (20). If the DBDs do not interact in the TR homodimer, this could explain the greater stability of the TR-RXR heterodimer on DR-4. Interestingly, TR-RXR heterodimers are not stabilized by such DBD-DBD interactions on F2, because, in contrast to what is seen on DR-4, the weakened interaction between the TR L422R and wild-type RXR on F2 is not further disrupted by mixing the TR L422R with the RXR DBD mutant (data not shown). However, TR-RXR heterodimers do form on TREpal, where they appear to be weaker than the ones formed on F2 and DR-4, because L422R disrupts them completely. We interpret that in the case of TREpal, TR monomers recruit RXR using the same LBD dimer interface it utilizes to bind RXR in DR-4 and F2, but that the resultant TR-RXR complex bound to TREpal is less stable.
RXR serves as a heterodimer partner to VDR and RAR and it was predicted that DBD-DBD contacts would stabilize VDR-RXR and RAR-RXR heterodimers (23). However, we previously found that a single LBD mutant in the VDR analogous to the L422R in the TR impaired the formation of heterodimers with RXR on DR-3 (45) suggesting that DBD-DBD contacts do not contribute to VDR-RXR heterodimers to the extent that they do with the TR-RXR heterodimers. It is noteworthy in this respect that the DR-4 configuration places the center of the DBDs with a 10 bp separation, which is approximately one turn of the DNA (10.5 bp) and aligns the DBDs in a way that could optimize contacts between the DBDs. It is possible that the shorter separation and 36°rotation expected between the RXR and the VDR twists the orientation of the DBDs to eliminate direct contact. In this context, VDR does not form homodimers on DNA (45). This again points to properties of the RXR that are dominant in forming the complex.
The current studies argue that TRs use a similar LBD surface for homodimerization and heterodimerization. However, there may be differences in the contacts made, even though they involve overlapping surfaces. In this regard, it is known that proteins may possess consensus sites with favorable intrinsic physiochemical properties that are dominant for protein-protein interactions with multiple partners (46). Modest adjustments occur within this flexible consensus site to accommodate binding to different partners and the crucial properties for this convergent binding surface are its accessibility, hydrophobicity, and limited charge interactions at its periphery (46). Such properties are indeed analogous to the ones we found in the dimer surface of TR and RXR, where a patch of hydrophobic residues are readily accessible for protein interactions. Crystal structures of heterodimeric complexes of RAR␣ and RXR␣ LBDs (29) and PPAR␥ and RXR␣ LBDs (30) revealed that the helices of these receptors, which correspond to H10 and H11 of TR, contribute more than 75% of the dimer surface. A hydrophobic core of residues analogous to the ones described here for the TR-RXR LBD interactions is at their centers. Our model for the heterodimeric interface of TR and RXR LBDs based on functional studies closely matches the surfaces reported for the crystal structures of RAR␣-RXR␣ and PPAR␥-RXR␣ LBD heterodimers. This finding supports the contention that the conservation of interface residues of this subgroup of nuclear receptors, which is significantly higher than that of the entire LBD, enables these receptors to engage in heterodimer formation with RXR (29) and predicts that the LBD surfaces for heterodimerization are likely to be similar within this subgroup of nuclear receptors.
We addressed whether the mutations could block functional receptor-receptor interactions in vivo, by using a heterologous system that employed the GAL-DBD linked to the RXR LBD and TR mutants with a reporter gene containing GAL DNA binding sites. Several mutants that blocked formation of GST⅐RXR-TR heterodimers under cell-free conditions also blocked TR action with the heterologous system indicating that the preferred functional TR-RXR interface in vivo is similar to that found in vitro. However, the mutants were overall less effective at disrupting these in vivo interactions than those observed in cell-free conditions, and therefore defined a more restricted area of required tight contact. This could imply stabilization because of interactions with endogenous factors, such as those recruited because of tethering the TR and RXR in higher order complexes with coactivators and other accessory proteins (47).
In summary, the data define a common domain of the TR and RXR that is involved in TR-TR dimerization and TR-RXR heterodimerization. The strength of the interaction varies and can be influenced by the partner, DNA, DBD-DBD interactions and probably other proteins that form higher order complexes in the cell. In some, but not all cases, interactions between this domain on the TR are important for TR action.