Crystal Structure of Human Estrogen-related Receptor α in Complex with a Synthetic Inverse Agonist Reveals Its Novel Molecular Mechanism*

Inverse agonists of the constitutively active human estrogen-related receptorα (ERRα, NR3B1) are of potential interest for several disease indications (e.g. breast cancer, metabolic diseases, or osteoporosis). ERRα is constitutively active, because its ligand binding pocket (LBP) is practically filled with side chains (in particular with Phe328, which is replaced by Ala in ERRβ and ERRγ). We present here the crystal structure of the ligand binding domain of ERRα (containing the mutation C325S) in complex with the inverse agonist cyclohexylmethyl-(1-p-tolyl-1H-indol-3-ylmethyl)-amine (compound 1a), to a resolution of 2.3Å. The structure reveals the dramatic multiple conformational changes in the LBP, which create the necessary space for the ligand. As a consequence of the new side chain conformation of Phe328 (on helix H3), Phe510(H12) has to move away, and thus the activation helix H12 is displaced from its agonist position. This is a novel mechanism of H12 inactivation, different from ERRγ, estrogen receptor (ER) α, and ERβ. H12 binds (with a surprising binding mode) in the coactivator groove of its ligand binding domain, at a similar place as a coactivator peptide. This is in contrast to ERRγ but resembles the situation for ERα (raloxifene or 4-hydroxytamoxifen complexes). Our results explain the novel molecular mechanism of an inverse agonist for ERRα and provide the basis for rational drug design to obtain isotype-specific inverse agonists of this potential new drug target. Despite a practically filled LBP, the finding that a suitable ligand can induce an opening of the cavity also has broad implications for other orphan nuclear hormone receptors (e.g. the NGFI-B subfamily).

In mammals, the nuclear hormone receptor (NR) 2 superfamily consists of 48 transcription factors that control essential developmental and physiological pathways (1). Although the transcriptional activity of NRs is often regulated by specific ligands, several members of the superfamily have no known natural ligands and are therefore referred to as orphan NRs (2). Estrogen-related receptor ␣ (ERR␣; NR3B1) was the first orphan NR to be identified on the basis of its similarity with estrogen receptor ␣ (ER␣; NR3A1) (3). ERR␣ and its relatives ERR␤ (NR3B2) and ERR␥ (NR3B3) form a small family of orphan NRs that are evolutionarily related to the estrogen receptors ER␣ and ER␤. ERRs preferentially bind to DNA sites composed of a single half-site preceded by three nucleotides with the consensus sequence TNAAGGTCA, referred to as an ERR-response element. It has been shown that ERR␣ also efficiently binds to estrogen-response elements and that these receptors share common target genes (4). This observation was further supported by studies demonstrating cross-talk between the ER and ERR pathways (reviewed in Ref. 5). Several studies have highlighted ERR␣ as a main player in mitochondrial biogenesis and oxidative metabolism (6 -8), suggesting that ERR␣ could be used for therapeutic intervention in diabetes or metabolic diseases. In addition to this central role in metabolism, ERR␣ is also now accepted as an emerging target in cancer (9,10). Finally, few papers have suggested a role of ERR␣ in bone metabolism (reviewed in Ref. 11). The importance of ERR␣ as a drug target has been recently reviewed (12), which further re-emphasizes the urge for new synthetic ligands in the ERR subfamily.
Despite their overall sequence similarity with the ERs, ERRs seem to regulate transcription in the absence of known natural agonist ligands. The presence of a phenylalanine residue (Phe 328 on helix H3) in the ligand binding pocket (LBP) has been found to be essential for the constitutive activity of ERR␣, because its mutation to alanine abolishes constitutive activity (13). The x-ray structure of apoERR␣, in complex with a coactivator peptide containing the L3 site of peroxisome proliferator-activated receptor coactivator-1␣ (PGC-1␣), had revealed that the LBP is practically filled with side chains (14). The unoccupied volume was found to be only ϳ100 Å 3 . It was predicted that a ligand with an equivalent of more than ϳ4 -5 carbon atoms could only bind if Phe 328 would drastically change its conformation. This would also require a displacement of Phe 510 and thus of H12, making the ligand an inverse agonist. The term "inverse agonist" (instead of "antagonist") is used because such a ligand would display an intrinsic activity, namely an inhibition of the constitutively active ERR␣. In contrast a neutral antago-* The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked "advertisement" in accordance with 18 U.S.C. Section 1734 solely to indicate this fact. The atomic coordinates and structure factors ( nist does not have an intrinsic activity by itself, it just counteracts the binding of an agonist. Indeed, searches for ERR␣ ligands have so far mostly identified inverse agonists. Few synthetic ligands, classically linked to the ER pathway, have been shown to modulate ERR␣ activity, namely 4-hydroxytamoxifen (4-OHT) and diethylstilbestrol (DES) (reviewed in Ref. 15). In addition, a new synthetic ligand acting as an inverse agonist has been identified for ERR␣. It was shown that it acts at submicromolar activities in a cell-based assay and that this compound inhibits ERR␣ induction of an ERR␣ target gene, monoamine oxidase B (16,17).
For ERR␥, the crystal structures for the complexes of its ligand binding domain (LBD) with the inverse agonists DES and 4-OHT (18), a 4-OHT derivative (19), and 4-OHT together with a corepressor peptide (20) have been published. The cavity volume of the LBP for unliganded ERR␥ was reported as 220 Å 3 (21), which is small but still considerably larger than the 100 Å 3 described for apoERR␣. This difference is mainly explained by the substitution of Phe 328 (ERR␣) with Ala 350 (ERR␥). In all reported complexes of ERR␥ with inverse agonists, the activation helix H12 was found to be completely dissociated from the LBD body (18 -20) and also partially disordered (20). The ERR␥ structures are thus in contrast to selective ER modulator (SERM) complexes for ER␣, where binding of raloxifene or 4-hydroxytamoxifen revealed H12 in the coactivator groove. In the latter ER␣ complexes, accompanying structural adaptations of the C-terminal end of H11 and of the H11/H12 loop (22,23) were observed. For ER␤ LBD complexed with the full antagonist ICI 164,384 (ICI), H12 was not visible because of high mobility (24).
For ERR␣ it was difficult to predict the exact details of how (and whether at all) a ligand (most likely acting as an inverse agonist) would bind in the LBP, because of the multiple conformational changes required to create the necessary space and interactions. Similarly, it was not clear what exact consequences ligand binding would have on H12, except that it would probably be displaced from the agonist position.
Here we report the crystal structure of the human ERR␣ LBD in complex with the synthetic inverse agonist cyclohexylmethyl-(1-p-tolyl-1H-indol-3-ylmethyl)-amine (Fig. 1C, compound 1a), at 2.3 Å resolution. Compound 1a is a derivative of compound 1b (Fig. 1C), which lacks the methyl group and originally was discovered by high throughput screening. We have introduced the mutation C325S in ERR␣ in order to reduce biochemical instability problems during purification and crystallization, associated with cysteine oxidation. Isothermal titration calorimetry (ITC) and differential scanning calorimetry (DSC) measurements have demonstrated that there are no significant differences in the thermodynamic binding parameters between wild type ERR␣ LBD and the C325S mutant for compound 1a (or 1b). Also binding studies by NMR spectroscopy of compound 1b to wild type ERR␣ in solution (with and without a spin label attached to Cys 325 ) revealed no violations of the observed relaxation effects with the crystal structure of the compound 1a complex.
Our results explain the novel molecular mechanism of an inverse agonist for ERR␣ and provide the basis for rational drug design to obtain selective inverse agonists of this potential new drug target.

MATERIALS AND METHODS
Protein Cloning, Expression, and Purification-The mutation C325S was introduced into the plasmid pXI392 (wild type ERR␣ LBD, amino acids 290 -519, numbering according to Swiss-Prot P11474 (14)) by site-specific mutagenesis using the QuikChange mutagenesis kit from Stratagene. The correctness of the construction was verified by DNA sequence analysis, and the correct clone was called pXI392b. Generation of recombinant baculovirus by transfection with BacPAK8 TM , plaque cloning, and amplification was done as described (14). For production of ERR␣(C325S) LBD-expressing cells, 1-liter shake flask cultures with Sf9 cells adapted to SF900 II (Invitrogen) at a density of 1.5 ϫ 10 6 c/ml were infected with baculovirus at a multiplicity of infection ϭ 1 and were cultured for 3 days at 28°C and 90 rpm, as described previously (14). Harvest of the cells was carried out as described (14). SDS-PAGE analysis (Coomassie staining and Western blotting with anti-penta-His antibody from Qiagen) revealed good and soluble expression of ERR␣(C325S) LBD. The purification of the frozen Sf9 cell pellets was done by three-step chromatography as described previously (14). The purified ERR␣ LBD protein in 50 mM Tris-HCl, pH 7.5, 100 mM NaCl, 5 mM DTT was concentrated to 10.4 mg/ml prior to crystallization. The resulting protein was estimated to be Ͼ95% pure and homogeneous by SDS-PAGE and reverse phase high pressure liquid chromatography coupled to mass spectrometry. The measured molecular mass of 27,025.3 Da of the protein corresponded to the acetylated des-Met form of the ERR␣ (290 -519; C325S) LBD.
Crystallization, Data Collection, and Structure Determination-Compounds 1a and 1b were synthesized in-house. For cocrystallization, a 5:1 stoichiometric excess of compound 1a from a 100 mM stock solution in Me 2 SO was added to the protein solution. Crystals were obtained at 4°C by the vapor diffusion method in 1-l hanging drops containing equal volumes of protein (10.4 mg/ml ERR␣(C325S) LBD, 100 mM NaCl, 5 mM DTT, 50 mM Tris-HCl, pH 7.5) and crystallization buffer (0.2 M MgCl 2 , 12% PEG400, 0.1 M HEPES, pH 7.5). The initial crystallization hit was obtained with Crystal Screen I from Hampton Research. Crystals were directly mounted in cryoloops and flash-frozen in the nitrogen stream. Diffraction data at 100 K were collected at the Swiss Light Source (beamline X10SA), using a Marresearch CCD detector and an incident monochromatic x-ray beam with 0.8 Å wavelength. Raw diffraction data were processed and scaled with the HKL program suite version 1.96.1 (26). The estimated B-factor by Wilson plot analysis is 48.6 Å 2 . The structure was determined by molecular replacement with MOLREP (27, 28) using as starting model the coordinates of ERR␣ LBD (Protein Data Bank access code 1XB7) refined to 2.5 Å resolution (14), with H12 removed. The program REFMAC version 5.0 (28,29) was used for refinement. Bulk solvent correction, an initial anisotropic B-factor correction, and restrained isotropic atomic B-factor refinement were applied. The refinement target was the maximum likelihood target using amplitudes. No cut-off was applied on the structure factor amplitudes. Cross-validation was used throughout refinement using a test set comprising 5.1% (1056) of the unique reflections. Water molecules were identified with the program ARP/wARP (28,30) and selected based on difference peak height (greater than 3.0) and distance criteria. Water molecules with temperature factors greater than 60 Å 2 were rejected. The program O version 7.0 (31) was used for model rebuilding, and the quality of the final refined model was assessed with the programs PROCHECK version 3.3 (32) and REFMAC version 5.0 (28,29). Crystal data, data collection, and refinement statistics are shown in Table 1.
Isothermal Titration Calorimetry, Differential Scanning Calorimetry, and Fluorescence Energy Transfer Measurements-The ITC experiments were performed using a Microcal VP-ITC instrument (Microcal, Inc., Northampton, MA). The sample cell of the calorimeter was loaded with ERR␣ LBD (wild type and C325S mutant at 40 and 30 M) in 50 mM Tris, pH 7.5, 100 mM NaCl, and 0.5 mM TCEP. The syringe was loaded with compound 1b (400 M) in the same buffer. All solutions were degassed for 10 min. Titrations were performed at 25°C with injection volumes of 8 l and a spacing of 240 s. The base line was set to zero assuming that the final injections of each titration represent only the heat of dilution. The data were fit using a one-site binding model available in the Origin ITC data analysis software (version 7.0). DSC analysis was performed on buffer solutions (50 mM Tris-HCl, pH 7.5, 100 mM NaCl, 1 mM TCEP) containing the ERR␣ LBD-compound 1a complex (1:10 molar ratio). The ERR␣ LBD concentration was 25 M (both for wild type or the C325S mutant), and 300 l of protein complex solution were injected. DSC scans were obtained using a MicroCal VP-capDSC system (MicroCal, LLC, Northampton, MA) at a scan rate of 200°C/h. The fluorescence energy transfer (FRET) measurements were conducted as time-resolved measurements (TR-FRET) in a miniaturized 1536 well plate format. Typically, 100 nl of compound solution (2 mM in 90% Me 2 SO) were transferred to the assay plate. 3 l of assay buffer (50 mM Tris-Cl, pH 7.5, 100 mM NaCl, 0.05% bovine serum albumin, 5 mM DTT, 0.1% Pluronic F-127) containing LANCE EU-W1024-labeled anti-His 6 antibody (PerkinElmer Life Sciences; 1 nM final assay concentration) were added. 1 l of His-ERR␣ LBD (50 nM final assay concentration) was added, and the plates were left for incubation at 20°C for 30 min. As FRET acceptor, 1 l of an N-terminally Cy5-labeled PGC-1␣ derived peptide (Cy5-RPCSELLKYLTT, 50 nM final assay concentration) was added. After 20 min of incubation the TR-FRET readout was performed on an Envision 2102 multilabel reader (PerkinElmer Life Sciences) with an excitation at 350 nm, first emission (donor) at 615 nm, and second emission (acceptor) at 665 nm (30 flashes, 100-s delay time). The readout was calculated according to X ϭ donor/ acceptor ϫ 1000. Data were fitted with XLfit4 after plate normalization of the derived data.
NMR Spectroscopy-Ligand binding studies were performed with wild type and spin-labeled ERR␣ LBD. The spin-labeled protein was obtained by addition of a maleimide-substituted tetramethyl pyrrolidine N-oxide (TEMPO) paramagnetic tag (Aldrich) that reacts selectively with the only cysteine residue present (Cys 325 ). Selective reaction was completed within minutes at room temperature prior to the measurement. T 1 relaxation and water-LOGSY experiments were performed at 600 MHz using a room temperature probe.

RESULTS AND DISCUSSION
The Overall Structure of the Complex between ERR␣ LBD and the Inverse Agonist Compound 1a Reveals H12 in a Position Similar to PGC-1␣-We have solved the crystal structure of ERR␣ LBD (containing the mutation C325S) in complex with the inverse agonist cyclohexylmethyl-(1-p-tolyl-1H-indol-3-ylmethyl)-amine (compound 1a) (Fig. 1C), at 2.3 Å resolution (space group H3). Compound 1a is a derivative of compound 1b, which originally was discovered by high throughput screening. The asymmetric unit contains a homodimer of ERR␣ complexes, i.e. ligand binding did not interfere with dimer formation. The results of the crystallographic refinement are summarized in Table 1. In general the electron density for the ERR␣ LBD is well defined, except for the His tag, residues Pro 309 -His 317 (H2/H3 loop), residues Arg 462 -Glu 470 (H9/H10 loop), and the C-terminal residues Met 518 -Asp 519 (all of which were not modeled). The protein part of the refined model consists of the PreScission TM site (LEVLFQGP) followed by amino acids Val 290 -Met 308 , Leu 318 -Gly 461 , and Arg 471 -Met 517 of the ERR␣ LBD. The refined model also contains 236 water molecules and two compound 1a molecules. The two complexes in the asymmetric unit are very similar, with a root mean square deviation (r.m.s.d.) of 0.13 Å for 210 C ␣ atoms of residues 291-308, 332-461, and 471-498.
A comparison of apoERR␣ (bound to a PGC-1␣ peptide (14)) with the inverse agonist complex (Fig. 1) revealed no major conformational changes for the main chain atoms from H1 to the middle of H11 (r.m.s.d. of 0.55 Å for 210 C ␣ atoms as above), with the exception of H3. The N-terminal part of H3 moves away from the LBP, in order to create necessary space for the ligand (Fig. 1). In particular the C ␣ s of Val 321 and Leu 324 move by 4.3 and 3.2 Å, respectively. Important structural perturba-tions of the main chain due to compound 1a binding also occur at the C terminus of H11 (unwinding of the last turn of H11, i.e. after Lys 499 ), the H11/H12 loop (reorganization of its conformation, so that Pro 505 is now at the beginning of H12), and H12 (new location in the coactivator binding groove) (Fig. 1). In the apoERR␣ complex with PGC-1␣ (14), H12 is formed between His 507 and Ala 516 (i.e. Pro 505 and Met 506 are in the H11/H12 loop), whereas in the inverse agonist complex H12 already starts at Pro 505 and is visible until Met 517 . In the apoERR␣ structure, the PGC-1␣ peptide (which adopts an ␣-helical conformation) utilizes two Leu side chains (Leu 210 and Leu 214 of its inverted LXXLL-motif) to make hydrophobic interactions in the coactivator groove. Unexpectedly, similar interactions are made by two Met side chains (Met 513 and Met 517 ) of H12 in the compound 1a complex. The hydrogen bond made between Lys 340 (H3) and the PGC-1␣ main chain (contributing to the "charge clamp" interaction) is replaced by a hydrogen bond between Lys 340 and the carbonyl oxygen of Ala 516 (H12). H12 in the compound 1a complex is making further stabilizing interactions via a salt bridge between the side chains of Lys 508 (H12) and Asp 329 (H3) and via a hydrogen bond between the carbonyl oxygen of Met 506 (H12) and the side chain of Gln 358 (H4). Finally, in the inverse agonist complex Leu 509 (H12) is located in the same hydrophobic pocket (with Trp 361 at the bottom) as Met 513 (H12) in the apoERR␣ structure. In the crystal lattice, Phe 510 makes an aromatic stacking interaction with Phe 510 of a neighboring molecule, but these packing interactions seem to be of less importance than the interactions in the coactivator groove itself (described above).
The finding that for the ERR␣-compound 1a complex H12 adopts a well defined position in the coactivator groove is in contrast to ERR␥, where the binding of diethylstilbestrol or 4-hydroxytamoxifen led to complete dissociation of H12 from the LBD (18 -20). On the other hand, the ERR␣ structure resembles the situation for ER␣, where binding of the SERMs raloxifene or 4-hydroxytamoxifen revealed H12 in the coactivator groove, with accompanying structural adaptations of the C-terminal end of H11 and of the H11/H12 loop (22,23). For ER␤ LBD complexed with the full antagonist ICI, H12 was not visible because of its high mobility (24).
Compound 1a Binds in the LBP of ERR␣ by Displacing Phe 328 (H3) and Phe 510 (H12); the Novel Mechanism of an ERR␣ Inverse Agonist-The empty cavity of the LBP in apoERR␣ has a volume of only ϳ100 Å 3 (14), so that multiple structural adaptations are required to enable ligand binding. In apoERR␣ the aromatic side chains of Phe 328 (H3), Phe 495 (H11), and Phe 510 (H12) cluster ( Fig. 2A) and in particular the presence of Phe 328 (which is replaced by Ala for ERR␤ and ERR␥) leads to an almost complete occupation of the LBP with side chains. Compound 1a binding induces a new side chain conformation for Phe 328 (H3) and as a consequence a displacement of Phe 510 (to avoid a steric clash with Phe 328 ), which triggers a dramatic movement of H12 (Fig. 2B). Phe 495 (H11), on the other hand, has almost the same position in both structures. In the compound 1a complex, Phe 495 makes an aromatic stacking interaction with Phe 328 , and they form together a hydrophobic lid on top of the ligand (Fig. 2B). In addition to the reorientation of Phe 328 , also the N-terminal part of H3 moves away from the LBP, in order to create space for the ligand via removal of Val 321 and Leu 324 (Fig. 3B). Taken together, Phe 510 could be regarded as a sensor that transduces the conformational state of Phe 328 (which depends on the ligand occupation of the LBP) into a position of H12 (turning the transcriptional activity of ERR␣ "on" or "off" (Fig. 1)).  (Fig. 3A). All residues in the LBP have well defined electron density. The amino nitrogen of compound 1a makes a direct salt bridge and a water-mediated hydrogen bond interaction with Glu 331 (Fig. 3A). The latter water molecule is well ordered (B-factor ϭ 25 Å 2 ) and is further stabilized by a hydrogen bond with the carbonyl oxygen of Phe 382 . Glu 331 of ERR␣ corresponds to Glu 353 of ER␣, for which it forms e.g. a hydrogen bond with the hydroxyl of 4-OHT (23). Importantly, the side chain of Glu 331 reorients toward the ligand in order to form interactions with the amino nitrogen (Fig. 3B). This reorienta-tion is made possible because of the side chain movement of Phe 328 (Fig.  3B), which creates space for Glu 331 (and the ligand). The cyclohexyl moiety of compound 1a is deeply embedded inside ERR␣ in a mainly hydrophobic pocket. There is some space left for substituents on the cyclohexyl pointing in the direction toward Arg 372 . The indole moiety of compound 1a is nicely fitting in a hydrophobic pocket, which was occupied by Leu 324 in the apoERR␣ structure (Fig. 3B). Leu 324 now contributes to one side of the indole pocket. Finally, the tolyl group is the moiety that is closest to the solvent, with the methyl group at the entrance of the LBP. The carbonyl oxygens of Ala 396 and Gly 397 on the H6/H7 loop are potential targets to obtain additional hydrogen bond interactions.

The Amino Nitrogen of Compound 1a Makes a Salt
In summary, the LBP has adapted via multiple conformational changes to the presence of compound 1a. The complementarity of fit in the hydrophobic regions is good, and the salt bridge interaction with Glu 331 is likely to contribute substantially to the binding affinity.
ITC Measurements Reveal That Binding Is Enthalpically Driven-ITC measurements have revealed that the binding of compound 1b to wild type ERR␣ LBD is enthalpically driven with ⌬H 0 ϭ Ϫ12.7 kcal/mol (Fig. 4A). A negative enthalpic contribution is compatible with the formation of an important salt bridge (if desolvation of the partners is not too unfavorable). Of course, because ⌬H 0 is composed of many contributions, it is not possible to strictly deduce the formation of a salt bridge. The entropic contribution to binding is unfavorable (ϪT⌬S 0 ϭ 4.6 kcal/mol). This is evidence for the hypothesis that the apo-form is more flexible than the ligandbound form and that H12 does not become more disordered upon inverse agonist binding. Indeed the x-ray structure shows a defined position of H12 in the coactivator groove. In addition, the unfavorable entropic term indicates that desolvation (e.g. for the salt bridge partners) does not play a dominant role. Importantly the K d values, as well as the separate enthalpic and entropic contributions, are very similar for ERR␣ wild type and the C325S mutant (Fig. 4) (K d values of 770 and 930 nM, respectively). This indicates that the binding mode has not been modified by the introduction of the mutation C325S. DSC measurements (data not shown), comparing compound 1a binding to ERR␣ wild type and the C325S mutant, also confirmed very similar affinities (⌬T m values of ϳ1.7-1.8°C in both cases). The IC 50 value for compound 1b binding as determined in the FRET assay is 700 nM, which is close to the K d of 770 nM determined by ITC. Compound 1a has an improved IC 50 value of 190 nM, which can be explained by additional van der Waals interactions (with the side chains of Val 321 , Val 498 , and Leu 500) contributed by its methyl group.
NMR Spectroscopy-Before inverse agonist cocrystals could be obtained, the binding of compound 1b was studied by NMR spectroscopy. Using wild type ERR␣ LBD, the binding was confirmed by T 1 relaxation and water-LOGSY experiments (33). The T 1 relaxation spectra of the compound alone and in the presence of ERR␣ (10-fold excess of compound) are shown in Fig. 5, A and B. Apart from strong T 1 relaxation effects observed in the presence of the protein, significant line broadening is also observed. This indicates that intermediate exchange processes occur during ligand binding. As we know from the crystal structure presented in this study, significant conformational rearrangements within ERR␣ are required for ligand binding. The line broadening is compatible with millisecond time scale motions, as is observed by the strong increase of T 2 relaxation of the ligand.
Because the expected LBP is close to Cys 325 (the only cysteine residue of ERR␣ LBD), we tried to obtain spatial information utilizing sitespecific spin labeling in conjunction with T 1 relaxation measurements (34). TEMPO-labeled ERR␣ was obtained after reaction with maleimide TEMPO, which reacts specifically with freely accessible cysteines. Significant relaxation effects induced by the paramagnetic center attached to Cys 325 were measured on the individual resonances of the ligand (Fig. 5, C and D). Strong effects were observed for the resonances of the indole and phenyl groups, and weak effects were observed for the aliphatic ring, indicating that the ligand binds in proximity of Cys 325 (with the cyclohexyl ring farthest away). The crystal structure of the compound 1a com-  plex shows that the phenyl and indole groups are about 6 -10 Å away from Cys 325 , whereas the aliphatic ring is about 13 Å away. The paramagnetic relaxation measured at the resonances of compound 1b bound to wild type ERR␣ in solution thus is compatible with the binding mode observed in the crystal structure of compound 1a in complex with the C325S mutant ERR␣. The reactivity of Cys 325 with maleimide TEMPO demonstrates the accessibility of this residue. In the crystal structure of apoERR␣, however, this residue is not accessible, because of the position of H12 and the H11/H12 loop. The fast reaction indicates that apoERR␣ is in a dynamic equilibrium between several states in which H12 can move away, which is important for opening up the LBP to enable ligand binding.
Comparison of Inverse Agonist Effects on the Conformations of ERR␣ and ERR␥; H3 Behaves Differently and Different Mechanisms of H12 Displacement-For the 4-OHT and DES complexes of ERR␥ (18 -20), the N-terminal part of H3 was not displaced with respect to apoERR␥. It also has a very similar position as in apoERR␣ (14). In particular Leu 268 (H3) in the inverse agonist ERR␥ complexes has practically the same position as the corresponding Leu 324 (H3) for apoERR␣. In contrast, for the compound 1a complex of ERR␣, Leu 324 is displaced by the indole moiety, and Val 321 is displaced by the tolyl moiety (Fig. 3B). Compound 1a binding thus induces in ERR␣ a different shape and location of the LBP (generated by a movement of the N-terminal part of H3), compared with the 4-OHT and DES complexes of ERR␥.
For ERR␥, DES-and 4-OHT-mediated antagonism is caused by the rotation of the side chain of Phe 435 (H11), which upon ligand binding flips out of the LBP and sterically interferes with H12 in the agonist position (18 -20). In contrast, for ERR␣ the side chain of Phe 495 (which corresponds to Phe 435 of ERR␥) practically does not move upon ligand binding (Fig. 3B). Compound 1a does not make a steric clash with Phe 495 , rather there are favorable interactions between this side chain and the ligand. For ERR␣, it is the new side chain conformation of Phe 328 that acts as a trigger for H12 displacement, removing it from its agonist position (Fig. 2). For ERR␥, the corresponding residue Ala 272 does not have to move, because its small side chain does not lead to steric clashes with DES or 4-OHT. Docking shows also that compound 1a could fit in the ERR␥ LBP without movement of Ala 272 (the N-terminal part of H3 would have to bend away from the LBP similarly as for ERR␣). On the other hand, the hydrophobic lid generated by Phe 328 and Phe 495 of ERR␣ on top of compound 1a (Fig. 3A) is not possible for ERR␥. This leads to the prediction that compound 1a should bind with a reduced affinity to ERR␥ (and ERR␤), compared with ERR␣.
Differences in H12 Positions and Interactions for the Inverse Agonist Complexes of ERR␣, ERR␥, and ER␣-In the crystal structure of the ERR␣-compound 1a complex H12 adopts a well defined position in the coactivator groove of its LBD. This is in contrast to ERR␥, where the binding of DES or 4-OHT led to complete dissociation of H12 from the LBD (18 -20). This dissociation was explained by the absence of an LXXLL "coactivator" motif in H12 of ERR␥ (18). By comparison, H12 of ER␣ does contain an LXXLL motif, for which Leu 536 and Leu 540 , i.e. the first and last Leu of the motif, mediate anchoring of H12 in the coactivator cleft for the raloxifene and 4-OHT complexes (22,23). For ERR␥, the corresponding residues Met 446 and Phe 450 were thought not to allow similar stabilizing interactions of H12 with the coactivator groove (18), and thus to be an explanation for H12 dissociation. In contradiction to this hypothesis, ERR␣ has the same residues as ERR␥ in the above motif, namely Met 506 and Phe 510 , but nevertheless displays H12 in the coactivator cleft of its LBD for the compound 1a complex. Actually, there is no sequence difference between ERR␣ and ERR␥ in the whole region between the ERR␣ residues Lys 499 (C terminus of H11) and Ala 516 (close to C terminus of H12). The first difference occurs at the C terminus of H12, with Met 517 of ERR␣ replacing Lys 457 of ERR␥. Strikingly, in the ERR␣ complex with compound 1a, Met 517 is well ordered and is practically at the same position as Leu 214 of PGC-1␣ bound FIGURE 5. T 1 relaxation experiments by NMR of compound 1b binding to wild type ERR␣ LBD. Spectra of the aromatic protons are shown. A, absence of ERR␣. B, presence of ERR␣ (T 1 times are 10 ms (black) and 210 ms (red)). C, same spectrum as B but with T 1 times of 10 ms (black) and 40 ms (red). D, same spectrum as C after introduction of the paramagnetic spin label bound to Cys 325 . T 1 times are 10 ms (black) and 40 ms (red).
to ERR␣ (14). Lys 457 of ERR␥ cannot make similar hydrophobic interactions in the coactivator groove, so this substitution is a possible explanation for the differences in H12 positions.
In addition to Leu 536 and Leu 540 of the LXXLL motif, ER␣ also utilizes Leu 544 of H12 for interactions in the coactivator cleft. According to the sequence alignment, these residues would correspond to Met 506 , Phe 510 , and Leu 514 of ERR␣, but interestingly they are not pointing into the groove. Instead, Leu 509 , Met 513 , and Met 517 of ERR␣ take up this role, enabled by a lengthening (compared with ER␣) of the H11/H12 loop structure by three residues. Knowledge of these unexpected interactions can further help in the design of peptide or nonpeptide antagonists, which bind in the coactivator groove (instead of the LBP), but do not necessarily contain an LXXLL motif. Recently, ERR␣-selective peptide antagonists containing an LXXLL motif were described, which were identified by screening combinatorial phage libraries (37).

CONCLUSION
In this study, we report the first x-ray structure of ERR␣ LBD in complex with an inverse agonist, cyclohexylmethyl-(1-p-tolyl-1H-indol-3-ylmethyl)-amine (compound 1a), to a resolution of 2.3 Å. Our data reveal for ERR␣ the multiple conformational changes in the LBP induced by the binding of compound 1a. These changes are the trigger for a novel mechanism of H12 displacement. In addition, they show an unexpected movement of H3 leading to an enlargement of the LBP and a surprising mode of H12 interaction in the coactivator binding groove. The structure also provides the basis for rational drug design to obtain isotype-specific inverse agonists of this potential new drug target. The finding that, despite a practically filled LBP in apoERR␣, a suitable ligand can induce an opening of the cavity has broad implications for other orphan NRs. In particular, it is of interest for the recently published LBD crystal structures in the NGFI-B family (38 -40), for which the LBP is completely filled with four aromatic residues conserved within the subfamily. Given the dramatic structural modifications that can occur for ERR␣, it is tempting to speculate that synthetic ligands (probably acting as inverse agonists) might also be found for members of the NGFI-B subfamily.