Communication between the ERRα Homodimer Interface and the PGC-1α Binding Surface via the Helix 8–9 Loop*

Although structural studies on the ligand-binding domain (LBD) have established the general mode of nuclear receptor (NR)/coactivator interaction, determinants of binding specificity are only partially understood. The LBD of estrogen receptor-α (ERα), for example, interacts only with a region of peroxisome proliferator-activated receptor coactivator (PGC)-1α, which contains the canonical LXXLL motif (NR box2), whereas the LBD of estrogen-related receptor-α (ERRα) also binds efficiently an untypical, LXXYL-containing region (NR box3) of PGC-1α. Surprisingly, in a previous structural study, the ERα LBD has been observed to bind NR box3 of transcriptional intermediary factor (TIF)-2 untypically via LXXYL, whereas the ERRα LBD binds this region of TIF-2 only poorly. Here we present a new crystal structure of the ERRα LBD in complex with a PGC-1α box3 peptide. In this structure, residues N-terminal of the PGC-1α LXXYL motif formed contacts with helix 4, the loop connecting helices 8 and 9, and with the C terminus of the ERRα LBD. Interaction studies using wild-type and mutant PGC-1α and ERRα showed that these contacts are functionally relevant and are required for efficient ERRα/PGC-1α interaction. Furthermore, a structure comparison between ERRα and ERα and mutation analyses provided evidence that the helix 8–9 loop, which differs significantly in both nuclear receptors, is a major determinant of coactivator binding specificity. Finally, our results revealed that in ERRα the helix 8–9 loop allosterically links the LBD homodimer interface with the coactivator cleft, thus providing a plausible explanation for distinct PGC-1α binding to ERRα monomers and homodimers.

The ligand-binding domain (LBD) 4 of nuclear receptors (NRs) acts as a ligand-dependent molecular switch recruiting diverse cofactor complexes (1)(2)(3). It exhibits a canonical threelayered "␣-helical sandwich" fold generally composed of 12 ␣-helices (H1-H12) and a ␤-sheet. Binding of activating ligand (agonist) to the LBD triggers conformational changes that result in coactivator (CoA) recruitment. CoAs typically interact with the LBD via a short ␣-helix containing a LXXLL sequence motif, also termed the NR box (4). The leucine residues of this motif form hydrophobic contacts with the so-called CoA cleft of the LBD, which is constituted by H3, H4, and H12. Binding of the LXXLL helix is further stabilized by a "charge clamp" interaction with two conserved, charged residues of the LBD (lysine or arginine in H3 and glutamate in H12). Numerous proteins are recruited to the agonist-bound NR LBD including CBP/ p300, the p160 CoAs (steroid receptor coactivator-1, transcriptional intermediary factor(TIF)-2, activator of thyroid and retinoid receptor), the p220 subunit of the DRIP⅐TRAP⅐Mediator complex (5,6), and members of the family of peroxisome proliferator-activated receptor (PPAR) CoAs (PGC-1␣, PGC-1␤, and PRC) (7,8).
The large number of potentially interacting NRs and CoAs, and the relatively small interaction surface constituted by the CoA cleft and the LXXLL helix, have raised the question how selective binding is achieved (18). Several examples of preferential or selective NR/CoA interaction have been documented.
Early work showed that residues flanking the LXXLL motif can contribute to binding specificity (19,20). Furthermore, the binding preference of the LBD of the glucocorticoid receptor or constitutive androstane receptor for the third LXXLL motif (NR box3) of TIF-2 appears to result from a second charge clamp interaction between CoA cleft residues and charged residues at positions ϩ 2 and ϩ6 (relative to the first leucine residue) of the box3 motif (21,22). In comparison, multiple features account for selective repression of liver receptor homolog-1 activity by a small heterodimer partner; key residues in the CoA cleft, orientation of the main chain C-terminal of H12, and the precise positioning of H12 affecting the shape of the CoA cleft (23). The understanding of CoA binding selectivity is further complicated by the observation that some NRs, such as ERR␣ or androgen receptor, can interact with the untypical motifs LLKYL or FXXLF, respectively (17,24). Although the binding preference of FXXLF peptides for androgen receptor was reported to be determined solely by CoA cleft residues that differ between steroid hormone receptors (25), the determinants of specific ERR␣ interaction with the LLKYL region of PGC-1␣ have not been addressed in detail.
PGC-1␣ harbors a putative (box1) and two confirmed (box2 and box3) NR interaction motifs (17). Several NR LBDs, including that of PPAR␣ (10), HNF-4␣ (11), RXR␣ (12), and ER␣ (13), interact only with the canonical box2 region (spanning 144 LKKLL 148 ), whereas ERR␣ and ERR␥ can also bind the untypical box3 region (spanning 210 LLKYL 214 ) (17). Importantly, PGC-1␣ interacts only with ERR␣ homodimers but not with monomers (26). A previously reported crystal structure of the ERR␣ LBD in complex with a PGC-1␣ box3 peptide shows that the untypical motif binds in an ␣-helical conformation (like LXXLL peptides), with the exception that Leu(ϩ4) is replaced by tyrosine (27). Such a LXXYL binding mode has already been observed earlier in the crystal structure of the ER␣ LBD complexed with a TIF-2 box3 peptide (28). Because of a particular amino acid composition (LLRYLL), the TIF-2 box3 region has the potential to interact with NRs in either a canonical (LXXLL) or an untypical (LXXYL) manner. However, based on additional functional studies, it was concluded that untypical binding to ER␣ may be favored by the crystal packing and that in solution TIF-2 box3 likely interacts via LXXLL (28). Together, these observations raise the question as to what determines efficient binding of the LXXYL region of PGC-1␣ box3 (but not of TIF-2 box3) to ERR␣ (but not to ER␣). ERRs are known to interfere with ER signaling in tissues where the receptors are coexpressed (29 -31). Thus, the understanding of specific CoA recruitment to ERR␣ may, as an alternative to the modulation of the receptor with ligands (32), help to design specific peptide antagonists of potential therapeutic interest.
Here we present a structural and functional analysis of PGC-1␣ binding to the ERR␣ LBD, which is based on a comparison between our new crystal structure of the ERR␣ LBD complexed with a PGC-1␣ box3 peptide at 2.1 Å resolution and previously presented ERR␣/PGC-1␣ box3 and ER␣/TIF-2 box3 structures (27,28). We co-crystallized the ERR␣ LBD with a "wild-type" PGC-1␣ box3 peptide ( 198 QQQKPQRRPCSELLK-YLTTNDD 219 ), whereas in the case of the reported structure (27) a shorter, mutant peptide ( 205 RPASELLKYLTT 216 ; C207A) was used (probably to avoid potential oxidation problems during crystallization). In our crystal structure, residues N-terminal of the LXXYL motif form additional contacts with the ERR␣ surface, notably with H4, the H8-H9 loop, and the C terminus of the LBD. Through extensive functional interaction studies using wild-type and mutant forms of PGC-1␣, ERR␣, and ER␣, we provide evidence that the contacts between N-terminal flanking residues of PGC-1␣ and the H8-H9 loop of ERR␣ are a major determinant of coactivator binding specificity. Furthermore, our results show that the H8-H9 loop allosterically links the homodimer interface with the CoA cleft, providing a plausible explanation for distinct PGC-1␣ binding of ERR␣ monomers and homodimers.

EXPERIMENTAL PROCEDURES
Construction of Recombinant Plasmids-cDNA fragments encoding wild-type or mutant human ERR␣ LBD (residues 189 -423) (GenBank TM accession number NP_004442) and human ER␣ LBD (residues 307-548) were generated by PCR. The wild-type ERR␣ cDNA fragment was cloned at the BamHI/ NotI sites of a modified pET-24d expression plasmid (Novagen) in which the T7 tag had been replaced by a hexahistidine (His) tag (MRSHHHHHHGPGLVPRGS). Wild-type and mutant cDNA fragments of ERR␣ and ER␣ were also cloned at the EcoRI/BamHI sites of the eukaryotic expression plasmid pCMX-VP16, resulting in fusion with the VP16 transactivation domain. Similarly, cDNA fragments encoding wild-type or mutant NR interacting regions of human PGC-1␣ or human TIF-2 were generated by PCR and cloned at the BamHI or EcoRI/BamHI sites of pCMX-Gal4, resulting in fusion with the Gal4 DNA-binding domain (amino acids 1-147). The coding region of full-length human ERR␣ was cloned by PCR at the EcoRI/BamHI sites of pCMX-K⅐ATG containing a Kozak sequence 5Ј to an ATG start codon and the multiple cloning site. ERR␣ mutants were subcloned into pCMX-K⅐ATG from the corresponding pCMX-VP16 plasmids using an internal SpeI site and a 3Ј BamHI site present in both pCMX vectors. The coding sequence for PGC-1␣ interaction domain (ID) and PGC-1␣ ID(L144A/L210A) was subcloned from pCMX-Gal4 at the BamHI site of a modified pET15b plasmid (Novagen) containing the coding sequence for glutathione S-transferase (GST) 5Ј of the hexahistidine tag sequence. A detailed description of the cloning of all plasmids is available upon request.
Protein Production and Purification-The His-tagged ERR␣ LBD (residues 189 -423) was produced in Escherichia coli BL21(DE3). Cultures were grown in Terrific Broth medium at 37°C until an A 600 of 0.4 was reached. The temperature was then decreased to 18°C, and cultures were induced at an A 600 of 0.8 overnight with 0.1 mM isopropyl 1-thio-␤-D-galactopyranoside. Harvested bacterial pellets were resuspended in buffer containing 20 mM Tris/HCl (pH 8.0) and 200 mM NaCl. After sonication of the bacterial cells and ultracentrifugation the tagged ERR␣, LBD was purified by affinity chromatography using a NiSO 4 -loaded HiTrap chelating column (GE Healthcare) and by gel filtration using a HiLoad 16/60 Superdex 200 column (GE Healthcare). The purified protein (estimated purity, Ͼ95%) was concentrated to about 10 mg/ml in gel filtration buffer containing 20 mM Tris/HCl (pH8.0) and 150 mM NaCl. GST-PGC-1␣ ID and GST-PGC-1␣ ID(L144A/L210A) fusion proteins were produced at 37°C in E. coli BL21(DE3) using a pET15b-GST expression plasmid. The GST fusion proteins were purified using glutathione-Sepharose (GE Healthcare) according to the manufacturer's instructions and dialyzed against buffer containing 20 mM Tris (pH 8.0) 100 mM NaCl.
Protein Complex Crystallization, Data Collection, and Processing-Crystallization trials were carried out in the presence of a 3-fold molar excess of PGC-1␣ box3 peptide ( 198 QQ-QKPQRRPCSELLKYLTTNDD 219 ). Crystals of the ERR␣ LBD/ PGC-1␣ box3 complex were obtained at 17°C by vapor diffusion in sitting drops. Reservoir solutions contained 0.1 M bis-Tris (pH 5.5), 15% polyethylene glycol 3350, and 0.2 M Mg(NO 3 ) 2 . Crystals were cryoprotected in reservoir solution supplemented with 20% ethylene glycol and flash-frozen in liquid nitrogen. Data collection was performed at 100 K at the European Synchrotron Radiation Facility (beamline ID 14-2) in Grenoble. Crystals belong to the monoclinic space group C2 with one homodimer per asymmetric unit. Data were integrated and scaled using HKL2000 (33).
Structure Determination, Refinement, and Comparison-The structure was solved by molecular replacement with AMoRe (34) using the published ERR␣ LBD crystal structure (27) as a model probe. Refinement involved iterative cycles of manual building and refinement calculations. The programs CNS-SOLVE (35), COOT (36), and QUANTA (37) were used throughout the structure determination and refinement. Anisotropic scaling and a bulk solvent correction were applied. Individual B-atomic factors were refined anisotropically. Solvent molecules were then placed according to unassigned peaks in the difference Fourier maps. The final model, refined at 2.1 Å with no cutoff, contains for ERR␣ (subunit A in the PDB file) 209 residues (194 -212, 223-365, 374 -420), or (subunit C) 216 residues (192-213, 223-367, 375-423), for PGC-1␣ (subunit B) 13 residues (205-217) or (subunit D) 14 residues (205-218), and 272 water molecules. According to PROCHECK (38), 91.8% of the and angles of the peptide chains are found in the most favored regions and 8.2% in additional allowed regions. Data collection and refinement statistics are summarized in Table 1. For structure comparisons, the C␣ traces of ERR␣ and ER␣ were superimposed from H3 (Val-225 in ERR␣, Met-343 in ER␣) to H11 (Leu-404 in ERR␣; Cys-530 in ER␣) using the lsq commands of "O" and default parameters. Figures  1R-D, 4B, 5B-C, and supplemental Fig.S1 were generated with PyMol (39).
Cell Culture and Transient Transfection Experiments-COS-1 and BHK cells were cultured in Dulbecco's modified Eagle's medium supplemented with 5% fetal calf serum, penicillin, streptomycin, and 1 g/liter glutamine. Transient transfection assays were carried out in 24-well plates (Greiner) with 0.5 ϫ 10 5 or 0.25 ϫ 10 5 cells/well (COS-1 or BHK cells, respectively) using the standard calcium phosphate co-precipitation technique. Cells were transfected with 250 ng of Gal4 (5x) -TATA-LUC or Gal4 (3) -TK-LUC reporter plasmid, 25 ng of the respective pCMX-Gal4 and pCMX-VP16 expression plasmid, and 100 ng of pCH110 (Amersham Biosciences) encoding ␤-galactosidase per well. Total transfected DNA was kept constant at 2 g of DNA/well using pUC19. Empty pCMX expres-sion plasmids served as controls. Experiments involving ER␣ were carried out in the absence or presence of 10 Ϫ7 M estradiol (Sigma). Cells were lysed with reporter lysis buffer (Promega). Luciferase activity was measured in a MicroLumat LB96P luminometer (EG&G Berthold) according to the manufacturer's instructions and normalized to ␤-galactosidase activity according to a standard protocol. All experiments were repeated at least three times in duplicate. Statistical significance was calculated using the unpaired t test.

Structure Comparison of ERR␣/PGC-1␣ Box3 and ER␣/ TIF-2 Box3
Complexes-We solved a new crystal structure of the ERR␣ LBD in complex with a PGC-1␣ box3 peptide ( 198 QQQK-PQRRPCSELLKYLTTNDD 219 ). In a previously reported structure of the ERR␣⅐PGC-1␣ complex (PDB ID: 1XB7) (27), a shorter mutant peptide ( 205 RPASELLKYLTT 216 ; C207A) was used. (The C207A mutation was probably introduced to avoid potential oxidation problems during crystallization.) Our crystals of the ERR␣⅐PGC-1␣ complex belong to space group C2 and contain one LBD homodimer (subunits A and C in the PDB file) per asymmetric unit. The structure was solved by molecular replacement and refined to 2.1 Å resolution. Data collection and refinement statistics are summarized in Table 1.
Although the overall interaction of the LXXYL core region of PGC-1␣ with the ERR␣ LBD is as described (27), we observed additional electron density for amino acids Arg-205, Pro-206, and Cys-207 ( Fig. 1, A and B), which were not included in the final model of the reported structure (Fig. 1C). These residues are in contact with H4, the H8-H9 loop, and H12. Notably, Arg-205 of PGC-1␣ forms a hydrogen bond with Gln-262 (H4), and water-mediated hydrogen bonds with Ser-337 (H8-H9 loop) and the main chain carbonyl oxygen of Ala-420 (H12). Furthermore, we could trace the C-terminal residue of the LBD (Asp-423) that interacts with His-341 (H8-H9 loop) and with Arg-205 of PGC-1␣.
These observations suggested that a network of interacting amino acids from H4, the H8 -H9 loop, and H12 of ERR␣ and N-terminal flanking residues of PGC-1␣ (comprising Arg-205, Pro-206, and Cys-207) contribute to specific ERR␣/PGC-1␣ interaction. A potential contribution of residues N-terminal of the LXXYL region has not been addressed by Kallen et al. (27), who explained efficient PGC-1␣ binding by favorable van der Waals contacts of Leu-211 and Tyr-213 with the ERR␣ CoA cleft and by intramolecular interactions involving Glu-209, Lys-212, Thr-215, and Thr-216, which stabilize the CoA helix.
A favorable contribution of Arg-205, Pro-206, and Cys-207 to the ERR␣/PGC-1␣ interaction is also interesting with respect to the crystal structure of the ER␣ LBD bound to a TIF-2 box3 peptide ( 740 KENALLKYLLDK 751 ; PDB ID: 1GWR) (28). In the crystal, the TIF-2 box3 region interacts with the receptor via LXXYL (Fig. 1D). As in the case of the reported ERR␣/PGC-1␣ structure (27), the TIF-2 helix is short and N-terminal flanking residues have not been included in the model. The residue of TIF-2 that corresponds to Cys-207 of PGC-1␣ is Glu-741 (not present in the model).
This structure comparison indicated that a cysteine at position 207 of PGC-1␣ may play a role in efficient ERR␣ binding and that other residues, such as alanine, serine, or glutamic acid, would negatively affect ERR␣/PGC-1␣ interaction. We further hypothesized that Cys-207, for example through a defined main chain conformation, helps to correctly position Pro-206 and Arg-205. Consequently, we addressed the importance of individual residues of ERR␣ and PGC-1␣ for efficient interaction in subsequent functional assays.
It must be noted that in the second molecule (subunit A) of the ERR␣ homodimer Arg-205 of PGC-1␣ does not contact the receptor surface (supplemental Fig. S1). Furthermore, there is no clear electron density for the C terminus of the ERR␣ LBD in this subunit. To date, the reason for this "asymmetry" within the ERR␣ homodimer is unclear. We speculate that distinct coactivator binding to the ERR␣ subunits is either due to the crystallization conditions or the result of an intramolecular communication of the subunits via the homodimer interface.
Interaction of Wild-type or Mutant PGC-1␣ Fragments with the ERR␣ LBD-To initiate functional studies that address the relative importance of PGC-1␣ residues, we performed mammalian two-hybrid interaction studies in transiently transfected eukaryotic cell lines. Wild-type and mutant NR interacting regions of PGC-1␣ were expressed in fusion with the Gal4

Dimer Interface and PGC-1␣ Binding Surface in ERR␣
DNA-binding domain (Fig. 2, left panel) and the ERR␣ LBD in fusion with the VP16 activation domain. A construct spanning the NR ID of PGC-1␣, including the box2 and box3 motifs, strongly interacted with the ERR␣ LBD, thus resulting in a dramatic increase in luciferase expression in COS-1 cells (Fig. 2,   right panel). Relative to the ID, the interaction of constructs containing a single point mutation in the box2 or box3 motif (ID(L144A) or ID(L210A), respectively) was reduced but still robustly above base levels. In contrast, the ERR␣ LBD no longer interacted with the double point mutant ID(L144A/ L210A). In agreement with these observations, the interaction of the smaller constructs, PGC-1␣ box2 or PGC-1␣ box3, with ERR␣ was comparable with the corresponding ID single point mutants and depended on an intact LXXLL or LXXYL motif (see mutants box2(L144A) and box3(L210A)).
Although mutation of the first leucine residue (Leu-210) of the LLKYL motif abolished interaction of PGC-1␣ box3 with the ERR␣ LBD, the single point mutations L211A or Y213A had a less dramatic effect on ERR␣ binding. In comparison, a C207S or a C207E mutation in PGC-1␣ box3 affected the interaction more strongly. In agreement with a previous report (17), these observations showed that PGC-1␣ efficiently interacts with ERR␣ via a canonical LXXLL (box2) and an untypical LLKYL (box3) motif. Although Leu-144 of box2 and Leu-210 of box3 are essential for the binding of PGC-1␣ to ERR␣, Leu-211 or Tyr-213 do not play a crucial role in the interaction of the box3 region. Furthermore, Cys-207 at position Ϫ3 relative to LLKYL contributes importantly to efficient binding.
Flanking Residues at Positions Ϫ6 to Ϫ3 of the PGC-1␣ LLKYL Motif Determine Efficient Binding to ERR␣-Next we asked whether mutation of residues N-terminal of Cys-207 of PGC-1␣ also affects ERR␣ interactions. Deletion of 13 N-terminal amino acids from PGC-1␣ box3 (resulting in the mutant PGC-1␣ box3 202 ) reduced the interaction with the ERR␣ LBD by only ϳ2-fold (Fig. 3A). More importantly, deletion of four more residues ( 202 PQRR 205 ) dramatically reduced binding of the resulting mutant PGC-1␣ box3 206 to almost base-line levels. Thus PGC-1␣ box3 202 defines a minimal high affinity box3 interacting region. Point mutation of Pro-202 and Gln-203 to  alanine resulting in PGC-1␣ box3 202 (PQ) did not influence interaction with ERR␣, whereas replacement of Arg-204 and Arg-205 with alanine (mutant PGC-1␣ box3 202 (RR)) reduced binding to minimal levels. Together with the observations from the crystal structure (Fig. 1A), these results revealed that amino acids at positions Ϫ6 to Ϫ3 (Arg-204/Arg-205 and Cys-207) determine high affinity binding of the PGC-1␣ box3 region to the ERR␣ LBD.
Next, we asked whether the TIF-2 box3 could interact with ERR␣ via LXXYL or whether swapping of the N-terminal flanking residues of PGC-1␣ box3 into TIF-2 box3 could enable or enhance such an interaction. As explained, the TIF-2 box3 region (containing 744 LLRYLL 749 ) has the potential to interact with ERR␣ either in the canonical manner via LXXLL or untypically via LXXYL. In mammalian two-hybrid assays, Gal4-TIF-2 box3 interacted only weakly with VP16-ERR␣ LBD, resulting in a 2-fold stimulation of luciferase activity (Fig. 3B). This interaction probably occurs via the canonical LXXLL motif, because mutation of Leu-745 (the first leucine residue of this motif) to alanine abolished the weak binding. Importantly, swapping of residues 202-209 of PGC-1␣ into Gal4-TIF-2 box3(L745A) resulted in a strong increase in ERR␣ binding via LXXYL. Our experiments thus show that N-terminal flanking residues determine whether the LXXYL region of PGC-1␣ or TIF-2 interacts efficiently with ERR␣.
Residues Outside of the ERR␣ CoA Cleft Contribute to Selective PGC-1␣ Binding-In the next set of experiments we addressed the question as to which parts of the ERR␣ surface are involved in specific binding of PGC-1␣, in particular of the box3 region. We first considered that the distinct CoA binding properties of ERR␣ and ER␣ may result from amino acid exchanges in the CoA cleft, where ERR␣ differs from ER␣ by two residues, Met-258 and Gln-262 of ERR␣, corresponding to Val-376 and Glu-380 of ER␣, respectively (see Fig. 1). Exchange of these residues strongly reduced the interaction of the resulting mutant ERR␣(M258V/Q262E) with PGC-1␣ ID, box2, and box3 (Fig. 4A). The dramatic loss of box2 binding was unexpected, as the ER␣ LBD has been reported to interact efficiently with the canonical LXXLL region of PGC-1␣ (13), and ERR␣(M258V/Q262E) was therefore expected to gain ER␣-like properties. In the case of box3, we note that Arg-205, Pro-206, and Cys-207 of PGC-1␣ are located in the vicinity of Gln-262 of ERR␣ and that Arg-205 forms a hydrogen bond with Gln-262 (Fig. 4B). Mutation of Gln-262 to glutamate changes the electrostatic potential and solvation properties of the ERR␣ surface and may thereby perturb binding of the N-terminal flanking residues of PGC-1␣. However, because the interaction with both PGC-1␣ box2 and box3 was strongly reduced, our observations suggest that in the context of ERR␣, Met-258 and Gln-262 account for overall binding affinity but do not determine CoA binding selectivity.
Next, we investigated the functional relevance of contacts between N-terminal flanking residues of PGC-1␣ and H12 or the H8 -H9 loop of ERR␣. The ERR␣ C terminus (Met-421, Met-422, Asp-423) differs from the corresponding region of ER␣ (see Fig. 5A), and in the published ERR␣ crystal structure (27) it has neither been modeled nor has its contribution to PGC-1␣ binding been evaluated. Deletion of the three C-terminal amino acids of ERR␣ (Met-421 to Asp-423) significantly reduced the interaction of the resulting mutant ERR␣ ⌬3C with PGC-1␣ box3, whereas binding to the box2 region was less strongly affected (Fig. 4A). Thus, the box3 region of PGC-1␣ requires the ERR␣ C terminus for maximal interaction. In comparison, point mutation of His-341 (H8 -H9 loop) or of Asp-423 (C terminus), which form a hydrogen bond in one subunit of the asymmetric unit (Fig. 4B), affected PGC-1␣ box3 binding only moderately.
Finally, we addressed the role of the H8 -H9 loop in CoA interaction. The loop is part of a hydrogen bonding network that centers on Ser-337 and Asp-338 and links the ERR␣ homodimer interface with the CoA cleft (Fig. 4B). This observation is intriguing because it indicates possible allosteric interactions between both surface regions. Furthermore, the H8 -H9 loop differs significantly in ERR␣ and ER␣, and Asp-338 is not conserved in ER␣ (see Fig. 5, A-C). Because this region might account for distinct CoA binding, we evaluated the effect of mutations aimed at perturbing the overall conformation of the H8 -H9 loop in mammalian two-hybrid assays.
A complete exchange of the H8 -H9 loop of ERR␣ (amino acids 338 -341) for that of ER␣ (amino acids 457-468) abolished interaction of the resulting mutant ERR␣(SwapH8 -H9) with PGC-1␣ (Fig. 4C) providing evidence for the involvement of the loop in CoA binding. Mutation of Ser-259 (H4) to histidine (the corresponding residue in ER␣) had a small effect on PGC-1␣ or box2 but a larger effect on box3 binding. In comparison, PGC-1␣ interaction of ERR␣(E343A) remained almost unchanged. Importantly, however, mutation of Asp-338 (H8 -H9 loop) or Arg-315 (H7) in the homodimer interface to alanine strongly reduced PGC-1␣ binding. Together, these observations suggested that changes that significantly perturb the conformation of the H8 -H9 loop influence CoA interaction.
The H8 -H9 Loop Plays a Role in Efficient and Selective PGC-1␣ Box3 Binding-As mentioned, the H8 -H9 loop region differs significantly in ERR␣ and ER␣ (Fig. 5A). Interestingly, this region has only been modeled in some ER␣ crystal structures, possibly explaining why potential functions have not been addressed to date. Superimposition of an ER␣ LBD crystal structure in which the H8 -H9 loop has been modeled (PDB ID: 1QKU) with the ERR␣ LBD showed that in ER␣ H9 is longer and the H8 -H9 loop adopts a distinct conformation (Fig. 5B). The superimposition also suggested that in ERR␣(SwapH8 -H9) the hydrogen bonding network connecting the homodimer interface with the CoA cleft is particularly perturbed by the bulky side chain of Tyr-459 (Fig. 5C). On the other hand, some amino acids that contribute to the hydrogen bonding network amino acids that differ between the CoA cleft of ERR␣ and ER␣ are underlined. Regions that differ significantly between the two receptors (H8 -H9 loop and C terminus) and that have been exchanged in the respective swap mutants are colored red. B, superimposition of the CoA cleft and the H8 -H9 loop region of ERR␣ and ER␣ (PDP ID: 1QKU). H9 is longer in ER␣, and the H8 -H9 loop adopts a different conformation. Only selected residues that have been mutated are depicted. C, superimposition of the H8 -H9 loop region of ER␣ with the corresponding part of ERR␣ (molecule coloring as in B). The representation illustrates that swapping of the H8 -H9 loop of ER␣ into ERR␣ perturbs the homodimer interface and requires conformational adaptations of side chains, mainly because of the presence of the bulky Tyr-459. D, ER␣ mutants probing the contribution of residues to PGC-1␣ binding (E2, estradiol). In ER␣(V376M-E380Q), residues that differ in the CoA of ER␣ and ERR␣ have been exchanged. In ER␣(SwapH8 -H9), the H8 -H9 loop region (amino acids 457-468) has been replaced with the corresponding region (amino acids 338 -341) of ERR␣. ER␣(3mut) corresponds to V376M-E380Q-SwapH8 -H9-H373S-H377S. In ER␣(4mut), His-547-Arg-548 of ER␣ have, in addition, been replaced with Met-421-Met-422-Asp-423 of ERR␣. Mammalian two-hybrid interaction assays were done in BHK cells using Gal4 (5) -TATA-LUC, pCMX-Gal4-PGC-1␣ constructs, and wild-type or mutant pCMX-VP16-ERR␣ LBD. Normalized luciferase activity observed with Gal4-PGC-1␣ ID and VP16-ERR␣ LBD served as the reference and was set to 100%. Bars represent the mean Ϯ S.D. (n Ն 6). p values were calculated for ERR␣ mutants relative to the corresponding values of wild-type ERR␣ (**, p Յ 0.0001; *, p Յ 0.001).
centered around Ser-337 and Asp-338 in ERR␣ are conserved in ER␣, e.g. Arg-315 (H7) and Arg-389 (H10) of ERR␣ corresponding to Arg-434 and Arg-515 of ER␣, respectively (see Fig.  4B and data not shown). Amino acids of potential functional importance that are not conserved comprise Ser-263 (H4) and Gln-311 (H7) of ERR␣ corresponding to Cys-381 and Ala-430 of ER␣, respectively. In the ER␣ homodimer interface, Arg-434 and Arg-515 adopt distinct side chain conformations to accommodate the bulky side chain of Tyr-459 (data not shown). However, upon swapping of the H8 -H9 loop of ERR␣ into ER␣, the arginine side chains might contribute to the allosteric control of CoA binding as suggested for ERR␣.
Consequently, to provide further evidence for an allosteric role of the H8 -H9 loop in PGC-1␣ binding, we asked whether replacement of the H8 -H9 loop alone or in concert with other mutations would enable more efficient interaction of ER␣ with PGC-1␣, in particular with the box3 region. In mammalian two-hybrid assays, the wild-type ER␣ LBD interacted in an estradiol (E2)-dependent manner, detectable only with the PGC-1␣ ID and box2 but not with box3 (Fig. 5D). The overall response was only about 10% of that obtained with the ERR␣ LBD. Relative to wild-type ER␣, mutation of CoA cleft residues to the corresponding ones found in ERR␣ (resulting in mutant ER␣(V376M/E380Q)) enhanced the interaction with PGC-1␣ ID and box2. In contrast, no binding to box3 was detected. PGC-1␣ binding was similarly enhanced in the case of the mutant ER␣(SwapH8 -H9) containing the H8 -H9 loop of ERR␣. Importantly, this mutant gained significant interaction with PGC-1␣ box3. PGC-1␣ interaction was even further enhanced in ER␣(3mut), comprising the double point mutation in the CoA cleft, the swapped H8 -H9 loop, and H373S/H377S to avoid potentially reduced CoA binding as observed for ERR␣(S259H) (see Fig. 4C). Finally, however, when we additionally swapped the three C-terminal residues of ERR␣ into ER␣ (resulting in mutant ER␣(4mut)), PGC-1␣ interaction was reduced. We hypothesize that in the context of ER␣ the bulky side chains of Met-421 and/or Met-422 cannot be accurately accommodated, such that the dynamic behavior of H12 is perturbed. In summary, our data show that regions outside the CoA cleft such as the H8 -H9 loop and, to a lesser extent, C-terminal residues critically contribute to efficient and selective PGC-1␣ binding, in particular to binding of the box3 region.
Mutations in the H8 -H9 Loop Affect DNA and PGC-1␣ Binding of ERR␣-In the final set of experiments, we asked whether mutations in the H8 -H9 loop region influence DNA or CoA binding of full-length ERR␣. To address this question, we assayed the binding of in vitro translated wild-type and mutant receptors (supplemental Fig. S2A) to double-stranded oligonucleotides containing either two direct AGGTCA repeats without spacing (DR0) or two inverted AGGTCA repeats with three-base pair spacing (ERE) in EMSAs (Fig. 6). In agreement with previous studies (26,40), we observed efficient homodimeric binding of ERR␣ to both elements (Fig. 6A). Importantly, mutation of residues in the H8 -H9 loop that contribute to the homodimer interface (mutants R315A and D338A; see Fig. 4B), or exchange of the H8 -H9 loop for that of ER␣, resulted in partial monomer binding to the DR0 element. Monomer binding was observed only on the DR0 but not on the ERE element, as only the DR0 encompasses an extended halfsite sequence (TCAAGGTCA), which allows stabilizing interactions of the T/A box (adjacent to the DNA-binding domain) of ERR␣ with the minor groove of the TCA extension (41). Compared with R315A, D338A, and SwapH8 -H9, the E343A mutation, which is located outside the homodimer interface, did not significantly alter DNA binding. Furthermore, homodimeric DNA binding of the mutant SwapH8 -H9 to both the DR0 and the ERE element was drastically reduced. These observations showed that an intact H8 -H9 loop is required for optimal DNA binding and that single point mutations such as R315A and D338A can affect homodimerization of full-length ERR␣ on DNA.
Next, we tested whether the interaction of PGC-1␣ with DNA-bound ERR␣ was also affected by exchange or mutation of the H8 -H9 loop. On the DR0 and the ERE, the addition of small amounts of E. coli-expressed, partially purified GST-PGC-1␣ ID fusion protein resulted in an upshift of DNA-bound ERR␣ (Fig. 6, B and C). This upshift was not observed when GST-PGC-1␣ ID(L144A/L210A) was added. Compared with the wild-type receptor, all DNA-bound mutants were less efficiently shifted by GST-PGC-1␣ ID. Furthermore, DNA-bound ERR␣(SwapH8 -H9) did not interact with GST-PGC-1␣ ID, even at the highest CoA concentration tested (supplemental Fig. S2B and data not shown). Thus, mutations in the H8 -H9 loop such as R315A or D338A influence not only homodimerization but also CoA binding of DNA-bound, full-length ERR␣. Together, these data provide further evidence for the role of H8 -H9 loop in ERR␣ function and show that it is an important discriminatory feature between ERR␣ and ER␣.

DISCUSSION
Previous work has established that binding and coactivation of several NRs such as PPAR␣, HNF-4␣, RXR␣, and ER␣ by PGC-1␣ depends on the canonical LXXLL motif (NR box2) (10 -13). In comparison, an untypical LXXYL-containing region (NR box3) contributes to the interaction of PGC-1␣ with ERR␣ or ERR␥ (17). In agreement with a previous structural study (27), we observed that PGC-1␣ box3 binding to the ERR␣ LBD resembles that of canonical peptides, i.e. Tyr(ϩ4) occupies the position of Leu(ϩ4) of LXXLL peptides (Fig. 1). Interestingly, LXXYL binding has also been observed earlier in the crystal structure of the ER␣ LBD complexed with a TIF-2 box3 peptide (28). TIF-2 box3 contains the sequence LLRYLL, which potentially allows canonical LXXLL as well as untypical LXXYL binding. However, despite untypical LXXYL interactions in the crystal, it was concluded from functional studies that, in solution, binding was probably canonical. Our new ERR␣/PGC-1␣ box3 crystal structure revealed contacts between N-terminal flanking residues of PGC-1␣ and the ERR␣ surface (Fig. 1, A  and B). This observation suggested that selective ERR␣/ PGC-1␣ box3 interactions are not an inherent characteristic of the LXXYL core region but depend rather on N-terminal flanking residues of the CoA.
In functional studies we found, in accordance with other reports (17,27), that binding of PGC-1␣ box2 and box3 to ERR␣ critically depends on the ϩ1 leucine residue (Leu-144 and Leu-210, respectively), as it is abolished by leucine-to-alanine mutations (Fig. 2). In comparison, Leu-211 and Tyr-213 play a less crucial role and stabilize rather than determine the interaction. Importantly, in accordance with our structural data, we observed that residues at positions Ϫ6 to Ϫ3 of the LLKYL motif contribute to efficient ERR␣ binding, as the interaction is strongly reduced in the case of the PGC-1␣ mutants C207S, C207E, and R204A/R205A (Figs. 2 and 3A). The role of these N-terminal flanking residues is further underlined by TIF-2 box3(L745A-Swap), which gains interaction with ERR␣ via LXXYL (Fig. 3B). Because of the comparable size of cysteine and serine, we did not anticipate that the C207S mutation would severely reduce interaction with ERR␣. We speculate that this effect is due to the distinct solvation properties of these residues, which in the case of cysteine results in a distinct conformation of the PGC-1␣ main chain allowing proper positioning and optimal binding of Arg-205.
In a recent report, high-affinity interacting ERR␣ peptides with N-terminal flanking regions that differ from that of PGC-1␣ have been identified (42). It will be interesting to analyze the exact binding mode of these peptides in future studies.
To investigate the contribution of the LBD to CoA binding selectivity, we first considered amino acid differences between the CoA clefts of ERR␣ and ER␣. This approach was chosen because few amino acids in the CoA cleft have previously been implicated in the specific CoA binding properties of androgen receptor (25), although partially conflicting data are presented in other reports (43,44). The exchange of residues that differ in the CoA cleft of ERR␣ and ER␣ had significant effects on PGC-1␣ binding, resulting in strongly decreased interaction with ERR␣(M258V/ Q262E) and slightly increased interaction with ER␣(V376M/E380Q) (Figs. 4A and 5D). However, these effects do not explain CoA binding selectivity, as ERR␣(M258V/ Q262E) unexpectedly lost almost all binding to PGC-1␣ box2, whereas ER␣(V376M/E380Q) did not gain interaction with PGC-1␣ box3. Therefore we examined, in mammalian two-hybrid assays, whether the interaction between N-terminal flanking residues of PGC-1␣ and the H8 -H9 loop of ERR␣ observed in the crystal structure was functionally rel- A, binding of in vitro translated, wild-type and mutant ERR␣ to double-stranded, 32 P-labeled oligonucleotides containing two direct AGGTCA repeats without spacing (DR0) or two inverted AGGTCA repeats with three-base pair spacing (ERE) in EMSAs. Specific binding was verified by adding a 100-fold excess of specific (DR0 or ERE) or unspecific competitor (Gal4-responsive element (GalRE)). Mutations in the H8 -H9 loop that perturb the homodimer interface (R315A, D338A, but not E343A) or exchange of the H8 -H9 loop for that of ER␣ result in partial monomer binding of the ERR␣ mutant to the DR0 element. B and C, interaction of E. coli expressed, partially purified GST-PGC-1␣ ID or GST-PGC-1␣ ID(L144A/L210A) fusion protein with DNA-bound wild-type or mutant ERR␣ in EMSAs. The indicated amounts (in g) of GST-PGC-1␣ fusion were added. DNA complexes of wild-type ERR␣ are more efficiently supershifted by GST-PGC-1␣ ID than the point-mutated receptors. DNAbound ERR␣(SwapH8 -H9) does not interact with GST-PGC-1␣ ID. Ctr, unprimed reticulocyte lysate used as control.
sterically linking the LBD homodimer interface with the CoA cleft.