The Three-dimensional Structure of the Liver X Receptor β Reveals a Flexible Ligand-binding Pocket That Can Accommodate Fundamentally Different Ligands*

The structures of the liver X receptor LXRβ (NR1H2) have been determined in complexes with two synthetic ligands, T0901317 and GW3965, to 2.1 and 2.4 Å, respectively. Together with its isoform LXRα (NR1H3) it regulates target genes involved in metabolism and transport of cholesterol and fatty acids. The two LXRβ structures reveal a flexible ligand-binding pocket that can adjust to accommodate fundamentally different ligands. The ligand-binding pocket is hydrophobic but with polar or charged residues at the two ends of the cavity. T0901317 takes advantage of this by binding to His-435 close to H12 while GW3965 orients itself with its charged group in the opposite direction. Both ligands induce a fixed “agonist conformation” of helix H12 (also called the AF-2 domain), resulting in a transcriptionally active receptor.

The structures of the liver X receptor LXR␤ (NR1H2) have been determined in complexes with two synthetic ligands, T0901317 and GW3965, to 2.1 and 2.4 Å, respectively. Together with its isoform LXR␣ (NR1H3) it regulates target genes involved in metabolism and transport of cholesterol and fatty acids. The two LXR␤ structures reveal a flexible ligand-binding pocket that can adjust to accommodate fundamentally different ligands. The ligand-binding pocket is hydrophobic but with polar or charged residues at the two ends of the cavity. T0901317 takes advantage of this by binding to His-435 close to H12 while GW3965 orients itself with its charged group in the opposite direction. Both ligands induce a fixed "agonist conformation" of helix H12 (also called the AF-2 domain), resulting in a transcriptionally active receptor.
Liver X receptors (LXR) 1 are members of the superfamily of nuclear receptors. These transcription factors regulate target genes through a dynamic series of interactions with specific DNA response elements as well as transcriptional coregulators. The binding of ligand has profound effects on these interactions and has the potential to trigger both gene activation and, in some cases, gene silencing. There are 48 sequence-related nuclear receptors in humans and the family comprises receptors that recognize hormones, both steroidal and non-steroidal, but also receptors responding to metabolic intermediates and to xenobiotics. There are also a number of so-called orphan receptors where the natural ligand is unknown. Some of the receptors show a very specific and high affinity ligand binding, like the thyroid hormone receptors, whereas others have a substantially lower affinity for their ligands and are less discriminating in their ligand selectivity. Like many of the other nonsteroid hormone receptors, LXR functions as a heterodimer with the retinoid X receptor (RXR) to regulate gene expression (1,2). Together with peroxisome proliferator-activated receptor (PPAR) and farnesoid X receptor (FXR), LXRs represent a subclass of so-called permissive RXR heterodimers. In this subclass, the RXR heterodimers can be activated independently by either the RXR ligand, the partner's ligand, or synergistically by both (3).
LXRs consist of two closely related receptor isoforms encoded by separate genes, LXR␣ (NR1H3) and LXR␤ (NR1H2). LXR␣ shows tissue-restricted expression with the highest mRNA levels in the liver and somewhat lower levels in the kidney, small intestine, spleen, and adrenal gland (4,5). In contrast, LXR␤ is ubiquitously expressed (6,7). Both LXR isoforms can be activated by specific oxysterols that are formed in vivo (2,8,9). In view of the high degree of homology between the LXR isoforms (75% identity in the ligand-binding domain (LBD), 54% identity overall), it is perhaps not surprising that few subtypespecific biological responses have been described and that information on subtype selective ligands is limited. LXRs have been shown to regulate several genes involved in cholesterol and lipid homeostasis. Prominent examples are the phospholipid/cholesteryl ester transporters ABCA1 (10 -13) and ABCG1 (14) and the sterol response element-binding protein (SREBP1c) (15) that induces fatty acid synthesizing enzymes. Increasing insight into the involvement of LXRs in cholesterol and fatty acid homeostasis has led to considerable interest in LXRs from the pharmaceutical point of view. As an example, one hallmark of atherosclerosis is the buildup of cholesteryl esters in macrophages of the arterial wall, transforming the cells into foam cells that are constituents of the atherosclerotic plaque. The potential to increase cholesterol efflux from macrophages/foam cells by inducing genes such as ABCA1 and/or G1, thereby preventing or even reversing the atherosclerotic process, makes LXRs highly interesting drug targets. This notion is supported by recent reports pointing to the importance of LXRs as inhibitors of foam cell formation and thereby also of atherosclerosis (16,17).
Our understanding of how nuclear receptor ligands exert their effects has been dramatically enhanced by the elucidation of the crystal structures of nuclear receptors either in their apo-state or with an agonist or antagonist bound to the LBD (18 -25). The LBDs share a common, mainly ␣-helical, fold that embeds a hydrophobic ligand-binding pocket. The structures have revealed that ligands can affect the conformation of the ligand-dependent activation function 2 residing in the C-terminal H12. An agonist allows H12 to cover the ligand-binding pocket, thereby providing one of the sides in the coactivatorbinding pocket. Conversely, an antagonist induces an H12 conformation blocking the coactivator-binding site and/or leading to recruitment of corepressors (21,25,26). The binding modes of several of these coregulators have been described in detail (26 -28).
Given its biological and pharmaceutical importance in cholesterol homeostasis, it is of great interest to determine the three-dimensional structure of LXR LBD. Synthetic, high-affinity ligands have great utility in this respect. Here we report structures of LXR␤ in complex with two very different synthetic agonist ligands, T0901317 (29) and GW3965 (30), respectively.
Crystallization and Data Collection-Crystallization was carried out using the hanging drop vapor diffusion technique. Both LXR␤-T0901317 and LXR␤-GW9365 crystals were grown from buffer containing 8.5% isopropanol, 17% polyethylene glycol 4000, 85 mM HEPES, pH 7.5, and 15% glycerol at room temperature. The first LXR␤-T0901317 crystals formed in the P6122 space group, with a ϭ b ϭ 58.7, c ϭ 293.8 and diffracted to better than 3 Å. In the same drops another crystal form was later detected belonging to the P212121 space group. Before data collection, the crystals were flashfrozen in the 100-K nitrogen gas stream of an Oxford cryostream700. Data were either collected with an MAR345 image plate detector using X-rays from a Rigaku H3R rotating anode generator ϩ Osmic Confocal Max-Flux TM optics or with an ADSC Q4R CCD at Experimental Station ID14 -4 at European Synchrotron Radiation Facility (ESRF). The observed reflections were reduced, merged, and scaled with MOSFLM (31) and Scala (32) in the CCP4 package (33). Details of data collection are summarized in Table I.
Structure Determination and Refinement-The structure was determined by molecular replacement methods with the CCP4 version of AMoRe (34), using an LXR␤ homology model based on a thyroid hormone receptor ␤ structure. 2 The molecular replacement was done on the first 3 Å data of LXR␤-T0901317 crystallized in P6122 and revealed one monomer per asymmetric unit. The crystal packing along one of the two folds revealed that the protein formed a tight homodimer, which allowed us to use the homodimer to search the second crystal form P212121 that gave two homodimers in the asymmetric unit. Electron densities for the T0901317 ligand confirmed the solutions of the molecular replacement. Model building was done with O (35) and refined with Refmac (33,36) and manual rebuilding in O. The four monomer complexes were treated as single TLS groups in Refmac, which gave more interpretable electron density maps and improved the R-factors substantially. Statistics of the data collection and refinement are summarized in Table I.
The Final Model-The region between H1 and H3 (residues 243-259), part of the ␤-sheet between H6 and H7 (residues 329 -332), and the loop connecting H11 and H2 (residues 445-447) have been abol- R work is calculated from a set of reflections in which 5% of the total reflections have been randomly omitted from the refinement and used to calculate R free . c R.m.s.d., root mean square deviation.
ished or modeled as alanines in some of the protein complexes because of weak electron densities. H12 (residues 439 -460) is absent in molecule C of the GW3965-complexed protein. The geometry of the final model is good, and only one amino acid residue, Leu-330, has been found to be an outlier in the Ramachandran plot. In some of the structures where Leu-330 has an ordered conformation it has an energetically unfavorable conformation.

RESULTS
The Overall Structure of LXR␤-Attempts to crystallize LXR␣ and LXR␤ with or without coactivator peptides containing an LXXLL motif were performed. LXR␤ without peptide was the first to produce sufficiently good crystals for structural analysis. The structures of the LXR␤-T0901317 and LXR␤-GW3965 complexes were determined by molecular replacement and refined, including all diffraction data, to 2.1 and 2.4 Å, respectively. Four LXR␤ complexes form the asymmetric unit with the chain names A, B, C, and D, respectively. Ligand occupies all four LBD molecules in the respective complexes with the exception of molecules C in the GW3965 co-crystallized protein, which appears to lack most of its ligand. 185 and 167 defined water molecules, respectively, were also included in the refinement.
The nomenclature of the secondary structure elements is based on that of the thyroid hormone receptor (18) and has been applied throughout this report even when comparing to other nuclear receptors where the original authors used a different nomenclature. The overall structure of the LXR␤ LBD encompasses residues 220 -460. Regions with weak electron densities were not built (for details see "Experimental Procedures").
Overall, the structure comprises a core layer of three helices (H5/6, H9 and H10) sandwiched between two additional layers of helices (H1-4 and H7, H8, and H11, respectively) and represents a typical nuclear receptor LBD fold ( Figs. 1 and 2). This arrangement creates a wedge-shaped molecular scaffold that contains a wider upper part, which shows the highest degree of sequence conservation between different nuclear receptor LBDs. The lower, narrower, part is folded to form a hydrophobic cavity into which the ligand can bind. The remaining secondary elements, an antiparallel ␤-sheet comprising three strands and H12 (sometimes referred to as the AF-2 motif), reside on either side of the ligand-binding cavity. The crystal structures of the LXR␤ complexes contain a homodimer of the protein with H10 and H11 of the two monomer LBDs forming the dimer interface related by a C2 symmetry axis (Fig. 1). The dimer interface is consistent with the ones found for other homodimer structures such as PPAR␥ (26) and RXR␣ (19) and also with the heterodimer structure of RXR␣/RAR␣ (20), which also has the same C2 symmetry axis between the monomers. In the PPAR␥/RXR␣ structure a dimer with the same secondary elements was observed, but the dimer interface is asymmetric where the PPAR monomer is rotated about 10°from the C2 symmetry axis of the RXR monomer (22). It was postulated that this asymmetry together with an interaction of PPAR␥ H12 and H8 and H11 of RXR would be the explanation for the permissiveness of PPAR␥ and that this would apply to the permissive LXR and FXR receptors as well.
T0901317 Is Cleaved by X-rays-The ligand-binding site revealed a clear density corresponding quite well to the T0901317 ligand. The ligand could easily be fitted with its hexafluoroisopropanol aniline group close to H12, and the results gave no doubts about the orientation of the ligand. The trifluoroethyl group could also be fitted in a nice density, but the sulfophenyl group did not appear to be connected to the rest of the molecule. There was a clear gap in the electron density between the nitrogen and sulfur atoms (Fig. 3A). A more detailed analysis on a rotating anode source revealed that it is the x-rays that split the ligand. 2 The consequence of the cleavage is that the sulfophenyl group moves away from the rest of the ligand and the distance between the nitrogen and sulfur atom increases by 1 Å from what would be anticipated in an intact ligand. A data set was collected to 2.8 Å on a rotating anode for only 12 h to get data with no detectable cleavage of the ligand (Fig. 3B). These data were used to build a model of the intact ligand in the binding site and are used under the remaining "Results" and "Discussion." The shift in the position of the sulfophenyl group is the only detectable difference between structures obtained with cleaved and non-cleaved ligand.
Structure of the T0901317-LXR␤ Complex-The entire ligand-binding pocket extends from H12 to the ␤-sheet lying between helices H6 and H7. Its volume varies between the four protein complexes mainly depending on the position of the ␤-hairpin loop residue Phe-329 and Arg-319. These residues are fairly disordered and show different positional preferences in the four molecules. Depending on their positions, the volume of the boot-shaped cavity ranges from 560 to 680 Å3 (Fig. 4A). The ankle shaft is buried in a cavity (C1) formed by H12 and H11 on one side and H7 and H8 on the other side. Residues from H8, H7, and H6 form the heel of the boot, (cavity 2 (C2)), while the toes are covered by the ␤-sheet and H3, forming cavity 3 (C3). In its entirety, the pocket is about 12.5 Å in length along the boot shaft down to the heel (C1-C2) and 8 Å from the heel to the toes (C2-C3). The thickness of the boot shaft down to the toes is ϳ6.5 Å.
Ligand recognition is achieved through a combination of one specific hydrogen bond and the complementarity of the binding cavity to the non-polar character of the T0901317 ligand (Figs. 4A, 5A, and 6). The hexafluoroisopropanol-substituted aniline group occupies C1, and the acidic hydroxyl oxygen atom is in good position to donate a proton to the N⑀ imidazole nitrogen of His-435. The imidazole ring is firmly positioned by a group of three water molecules stretching out to the surface between H12 and the bend between H5 and H6. Water molecule W1 is tetrahedrally coordinated by the imidazole N␦, Ser-436, the carbonyl oxygen of residue 432, and a water molecule, W2, at the surface. The bend releases the carbonyl oxygen of residue 305 and the nitrogen of residue 309 to coordinate another water GW3965 Induces Changes in the Ligand-binding Pocket-GW3965 has about the same apparent affinity for LXR␤ as T0901317. However, GW3965 is considerably larger than T0901317 and the observed ligand-binding pocket is substantially different from the one observed for T0901317. GW3965 takes advantage of the cavities used by T0901317, but in addition it forces the receptor to form an additional cavity (Fig.   4B). It also interacts with the residues at the ␤-sheet more than T0901317 does (Figs. 5B and 6). As in the case of the LXR␤-T0901317 complex, there are considerable differences between the four molecules close to the ␤-sheet, but remaining parts of the pocket are basically identical between the A, B, and D molecules. From the LXR␤-T0901317 complex boot cavity there is now a big protrusion on the right side of the heel, cavity 4 (C4) enclosed by residues from H7, H8, and the intervening loop. In addition, C3 is much larger and continues a few Å farther than in the LXR␤-T0901317 complex (Fig. 4B). All in all, these differences result in a cavity volume of about 980 -1090 Å 3 compared with the 560 -680 Å 3 observed in the case of LXR␤-T0901317 (Fig. 4, A and B). A part of the larger volume in the GW3965 complex is because the ligand-binding pocket is directly connected to another cavity (solvent cavity (SC)) filled by one isopropanol molecule. This pocket follows H1 and H3 on one side and is perpendicular to H6 and H9 on the other. It is present in all different complexes, but its shape, size, and solvent content vary due to both the ligand and crystal contacts, and it is only in the complexes with GW3965 that it is directly connected to the ligand-binding pocket.
The GW3965 ligand has a nitrogen atom with three large substituents, 1) chloro-trifluoromethyl benzyl, 2) diphenylethyl, and 3) propoxy phenylacetic acid (Fig. 5B). The chlorotrifluoromethylbenzyl group orients itself into the C1 cavity, thereby mimicking the hexafluoroisopropanol-substituted aniline group of T0901317. However, whereas T0901317 has a strong interaction through a direct hydrogen bond to His-435, the GW3965 ligand relies on hydrophobic interactions in the C1 pocket. Leu-449, Leu-453, and Trp-457 from H12, Phe-268, Thr-272, Ala-275 from H3, and Leu-345 from H7 show only minor differences compared with the LXR␤-T0901317 structure. His-435 is pushed out about 0.9 Å from the binding site, repelled by the benzyl ring and one of the fluorine atoms of the ligand. Gln-438, Val-439, and Leu-442 are shifted 1-2 Å to better accommodate the shape of the GW3965 ligand. The nitrogen at the central point of the ligand forces the Phe-271 phenyl ring to rotate 90°along the 2 angle and move away from GW3965.
The diphenylethyl side chain orients one of its phenyl groups into the C2 cavity, which is mostly intact, in close proximity to the C1 cavity. Ile-309, Met-312, Leu-313, and Thr-316 from H6 at the top of the C2 cavity are all unchanged relative to the T0901317 structure, but Ile-353 and Phe-349 from H8 need to change their positions slightly. These residues separate C2 from the GW3965 C4 cavity, which is filled up by the second phenyl group. Leu-345 separates the C4 cavity from the C1 cavity. The large movement of Phe-271, together with small  colored green (protein) and yellow (ligand)), and the apoform (carbon atoms colored blue), respectively. The structures were superimposed using LSQMAN (46) (0.7 Å root mean square deviation over 219 C␣ atoms using a 3.8 Å cutoff) The carbon atoms and hydrogen bonds of the residues in the T0901317 and the GW3965 LXR␤ complexes were colored gray and green, respectively. Met-312, Thr-316 form the boundary between the C2 and C4 cavities, on one hand, and C3 on the other. Although Phe-329 in the T0901317 complexes are fairly disordered, in the GW3965 complexes it is firmly positioned in the GW3965 complexes, where it is stacked on top of the propoxy phenylacetic acid group of the ligand and allows the carboxylic acid of the ligand to enter a polar environment. In all GW3965 complexes one of the carboxylic acid oxygen atoms is bound to the main-chain nitrogen of residue 330. The remaining interactions are different between the three molecules and are fairly disordered in the A and B molecule. This is in contrast to the B molecules, which reveals a well ordered, hydrogen-bonding network. The other ligand acid oxygen in the B molecule is bound to Ser-242 of H1 and a water molecule. N⑀ and one of the NH atoms of Arg-319 and the main-chain carbonyl oxygen of residue 330 coordinate that water molecule, leaving one proton of the water to donate to the carboxylic acid. Arg-319 in turn is also hydrogen-bonded to Glu-315 and another water molecule. In the other two molecules (A and D), Arg-319 remains close to the ligand but Ser-242 does not interact directly with the ligand in A or D. The H1-3 region is located on the surface and involved in crystal contacts, which could explain the different organizations of ligand binding.
The many changes in the side chains noted are mostly accommodated with little change in the supporting helix structural arrangement except for the C terminus of H1, in molecule B, where the Ser-242 interaction with the ligand moves it over 4 Å into the ligand-binding pocket. Because of this interaction, H1 is stabilized and elongated four extra residues (to Val-249).
The Empty Ligand-binding Pocket-One of the four independent molecules in the structure of LXR␤-GW3965 co-crystallized complex, the LXR molecule C, has only partial occupancy by the ligand. The ligand-binding pocket revealed only small traces of density consistent with ligand binding. The surrounding residues have considerably larger temperature factors, and many residues have weak electron densities compared with the three structures (molecules A, B, and D) where GW3965 occupies the pocket. The weak electron density of H12 is comparable with the density of the ligand, suggesting that H12 is flexible in solution in molecule C when there is no ligand bound. The crystal form has changed the cell dimensions in the GW3965-LXR crystal compared with the T0901317-LXR crystal (b decreases by about 4 Å). Analysis of crystal contacts suggests that it is only the C molecule that allows H12 to be flexible; in the other molecules H12 is involved in crystal contacts. Besides H12 many other residues surrounding the ligand-binding pocket also show weak or new positions in the putative apoform compared with the liganded structures (Fig. 6).

DISCUSSION
Plasticity of the Binding Pocket-The LXR␤ ligand-binding pocket shows a remarkable capability to accommodate ligands of different structures. The large variation in both shape and volume of the pocket follows from adjustments of the rotational conformers of several residues. These movements occur without significant effects on the overall structure of the protein (Fig. 2). When bound to either of the ligands described here, the receptor appears to have a relatively rigid region in proximity to H12. The area includes Trp-457, Leu-453, and Met-312. The three waters that connect Trp-457 and His-435 give His-435 some torsional freedom to accept hydrogen bonding to a ligand. Four phenylalanines (349, 340, 271, and 329) together with the volume around the ␤-hairpin loop and Arg-319 provide the flexibility of the pocket. The phenylalanines allow considerable flexibility in the middle of the receptor. The side chains can slide along each other to fit one of the phenyl groups of GW3965 and allow the carboxyl group of the same ligand to reach the ␤-hairpin loop. The former residues show virtually identical and ordered conformations with the same ligand in the 3-4 different liganded molecules in the crystal. In contrast, the ␤-hairpin loop area displays disorder and differences even between receptor molecules bound to the same ligand. By reaching farther out into the C3 cavity and binding directly to the ␤-hairpin loop and indirectly to Arg-319, GW3965 stabilizes this region more than T0901317 does.
Comparisons to Other Receptors-The closest homolog to the two LXR isoforms appears to be FXR with a sequence identity in the LBD of 29 and 27% of the two isoforms, respectively. Nine of the 31 residues in the ligand-binding pocket are conserved between FXR and LXR. Thyroid hormone receptor (TR), vitamin D receptor, PPAR, and retinoic acid receptor have 6 -8 residues of these 31 conserved, and the identities range from 20 -25% in the LBD. The volume of the LXR␤ ligand-binding pocket nearly doubles, going from 560 to 1090 Å 3 in the LXR␤ complexes. This can be compared with the small cavity in TR of about 550 Å 3 or the large ligand-binding pocket in PPAR␥ of about 1400 Å 3 (22). LXR appears to have evolved to recognize different ligands by being able to tune its structure, allowing structurally different ligands to activate it. In this sense it is similar to PPARs and pregnane X receptor.
The volume of the ligand-binding pocket of PPAR␥ is reported to be about 1400 Å3 when complexed to high affinity agonists like rosiglitazone or GI262570 (22). The PPAR␥ residues Phe-363 and Phe-282 change rotamer to accommodate these two different ligands in a similar way to what the four phenylalanines in LXR␤ do. In these structures, the ligand occupies about 25 or 40% of the available volume, respectively (22). In the LXR␤ structures presented here, the ligand occupies about 70% (T0901317), or just over 50% (GW3965) of the available volume.
Recently the structure of pregnane X receptor in complex with hyperforin was reported, showing a similar capability to tune the ligand-binding pocket to different ligands. The authors compare the SR12813 complex and the apoform of pregnane X receptor (40) to the hyperforin pregnane X receptor complex (41) and show that the cavity increases its size from about 1280 Å 3 in the apo and SR12813 structures to 1544 Å 3 . The increase in size is because of the mobility of His-407 (corresponding to His-435 in LXR␤) and Leu-209 in concert with the region before H3, which is absent in LXR␤. The increase in LXR is even larger, 500 -600 Å 3 between the two complexes described in this report. In LXR, the increase is brought about by the movements of His-435, the four phenylalanines, and the rearrangements at the ␤-hairpin loop and Arg-319 that open up into the solvent cavity, allowing it to become part of the ligand-binding pocket.
Transcriptional Activation Function-In complex with the two synthetic ligands, the H12 residues Trp-457 and Leu-453 are detected in very similar, if not identical, positions (Fig. 6). H12 contains the ligand-dependent activation function (AF2) and appears to be in the active conformation that would allow coactivator binding in both the T0901317 and the GW3965 complexes. No other observed differences inside the ligandbinding pocket translate into any detectable changes in the coactivator interacting area formed on the receptor surface between H3, -4, -5, and 12. The stabilization of H12 involves direct hydrophobic interactions with the H12 residues Leu-453 and Trp-457 but also indirect interactions via Phe-268. The mutation F268A affects the ligand binding severely and also decreases transcriptional activity (42). Both structures involve three water molecules that connect His-435 and Trp-457 to the bend between H5 and H6 described earlier and may be essen-tial for proper position of H12. It appears that stabilization of H12 comes about by simply having an appropriate group (the hexafluoroisopropanol-substituted aniline group in T0901317 or the chloro trifluoromethyl benzyl group in GW3965 complex) occupying the C1 cavity. The remaining cavities do not appear to be essential for the transactivation function of the receptor but mainly responsible for ligand recognition. They provide potential interaction surfaces that, when utilized, can increase ligand affinity. H12 is destabilized in the absence of ligand, as seen in molecule C of the GW3965 structure that lacks ligand. The absence of H12 in the coactivator-binding pocket removes one side of the charge clamp (Glu-455) and presumably prevents effective binding of coactivator proteins. This is analogous to what is found in the RXR apo structure where H12 is unconnected to its own molecule but instead binds to a coactivator pocket of another RXR molecule in the crystal lattice (19).
Comparisons to LXR␣-The two LXR isoforms are very similar (77% identity in the LBD), and most differences are located far away from the ligand-binding pocket. The 31 residues closest to the ligands are all identical, and one has to reach out into the secondary layer of residues to find any differences. The closest differences are located more than 6 Å away from any of the ligands. Despite the high similarity in the ligand-binding pocket, there have been reports of subtype-selective ligands (9,43), suggesting that subtype-specific residues located farther out from the ligand-binding pocket play a role for ligand binding specificity.
How Does an Endogenous Ligand Bind?-A number of endogenous and synthetic cholesterol derivatives have been shown to bind to LXR, and certain oxysterols are widely accepted as the biologically active LXR ligands (2) (44). From the two structures described here, it is not evident how a steroid would orient itself in the ligand-binding pocket. Pharmacophore studies of LXR␣ (45) suggested that the sterol side chain would form a hydrogen bond to Trp-457 and that the A-ring hydroxyl group would interact with Arg-319. The conformation taken by Trp-457 in complex with T0901317 and GW3965 would not allow a hydrogen bond. A considerable conformational change, which probably would distort the orientation of H12, would be required for Trp-457 to form a hydrogen bond to a steroid ligand. His-435 appears to be the most likely hydrogen-bonding partner, either directly or indirectly. Another possibility would be Thr-272, but mutational analysis of that residue shows that at least 24-S-hydroxycholesterol binding to LXR is unaffected by T272A mutations (42). A number of hydrogen bond donors and acceptors could be involved in binding to the A-ring hydroxyl. Among those are Arg-319 but also Ser-242, Glu-281, and perhaps more importantly the mainchain donors and acceptors of the ␤-hairpin loop around residue 330. Considering the high flexibility of these residues it is somewhat surprising that most LXR binding steroids have identical structures at this site. LXR accepts negatively charged ligands such as GW3965 (neutralized by Arg-319) but may as well accept neutral hydrogen bond donors or acceptors such as the cholesterol derivatives or even positively charged ligands (neutralized by, for instance, the glutamates 281 or 315). Ser-242, the main-chain nitrogen and oxygen of residue 330 or water molecules, could easily assist many different types of interactions with the only common feature being that they are mainly of polar character. The structures presented here may, in other words, suggest that novel endogenous ligands could be found.
The present study reveals two different liganded structures of the LXR␤ ligand-binding domain. The two different protein ligand complexes highlight the large plasticity of this receptor and explain the ability of LXR to allow fundamentally different ligands to bind and be an agonist for the receptor. Considering the importance of LXR in cholesterol homeostasis these structures provide useful tools for designing better ligands to treat diseases associated with imbalance in cholesterol metabolism and excretion.