The structure of the ultraspiracle ligand-binding domain reveals a nuclear receptor locked in an inactive conformation

Purification and Crystallization.

dmUSP LBD was expressed with an N-terminal 6-His tag in Escherichia coli [strain C41(DE3)] by using an isopropyl β-D-thiogalactoside-inducible pTac promoter. The protein was purified by using Ni-nitrilotriacetic acid affinity resin (Qiagen, Chatsworth, CA) and gel filtration using a Sephacryl S100 resin (Amersham Pharmacia). The monomeric homogeneous dmUSP LBD was concentrated to 10 mg/ml for crystallization. Crystals were grown at 18°C by using standard vapor-diffusion techniques with well buffer containing 130 mM ammonium sulfate, 160 mM sodium acetate (pH 4.6), and 25% polyethylene glycol (PEG) 2000. Larger crystals (maximum dimensions of 150 × 75 × 20 μm) were obtained by using streak-seeding techniques and by addition of ethylene glycol to the drop. After transfer to cryoprotectant (13 mM Tris, pH 7.0/32.5 mM sodium chloride/17.5 mM sodium acetate, pH 4.6/19.5 mM ammonium sulfate/2.5% PEG2000/30% glycerol), crystals were frozen in a stream of N₂ at 100 K.

Data Collection and Structure Solution.

Data collection and refinement statistics are given in Table 1. Although the crystals diffract to beyond 2 Å, data were useful only to 2.4 Å because of poor data quality at higher resolution as a consequence of poor spot shape arising from crystal bending during freezing. The structure was solved by using the estrogen receptor LBD as a search model in the cns program suite (21). Iterative rounds of positional simulated annealing and B factor refinement were carried out by using CNS along with rounds of manual rebuilding in the molecular modelling program o (22).

TLC.

Protein samples (typically 4 mg) and lipid standards were extracted with chloroform/methanol [2:1 (vol/vol)]. Samples were analyzed by using 0.25-mm TLC plates precoated with silica gel 60. Two TLC solvents were used [50:50 and 95:5 chloroform/methanol (vol/vol)] according to the run. Plates were dried and stained with iodine to visualize lipid.

Matrix-Assisted Laser Desorption Ionization (MALDI)/Time-of-Flight (TOF) Mass Spectrometry.

Lipid was extracted from the protein samples with chloroform/methanol [2:1 (vol/vol)]. The matrix used for all samples was 1 or 5 mg/ml α-cyano-4-hydroxy-trans-cinnamic acid in 60% aqueous acetonitrile with 0.1% trifluoroacetic acid. The matrix (0.5 μl) mixed with 0.6 μl of the lipid extract was applied directly to the sample plate and allowed to dry at ambient temperature. MALDI/TOF spectra were acquired on a Voyager-DE STR mass spectrometer (PerSeptive Biosystems, Framingham, MA) and calibrated with respect to standards deposited adjacent to each sample.

MALDI/Postsource Decay (PSD) was used to confirm the assignment of molecular-ion-peak 563.5 as diacylglycerol. The 563.5 PSD spectrum was compared with spectra of known lipid standards. The fragmentation pattern was similar to dimyristoyl (C14:0) glycerol + Na and was assigned tentatively as 1-palmitoyl-2-myristoyl-glycerol + Na.

Structure Determination.

The dmUSP LBD (residues 230–508) was expressed and purified from E. coli. Diffraction-quality crystals were grown as described in Materials and Methods. The structure of the dmUSP LBD was solved by using molecular-replacement techniques. Initial attempts at determining the structure, by using either the apo or holo Homo sapiens RXRα (hRXRα) structure as a search model, were unsuccessful (23, 24). However, because efforts to prepare heavy atom derivatives also were unsuccessful, alternative molecular-replacement models were tested. Unexpectedly, a solution was obtained by using the liganded Homo sapiens estrogen receptor-α (holo hERα; ref. 20) as a search model.

The dmUSP LBD is packed in the P1 crystal lattice as two interdigitated trimers. Consequently, the unit cell contains six copies of the LBD, each in a unique environment within the crystal lattice. An excellent initial electron-density map was obtained by averaging according to the six-fold noncrystallographic symmetry. Strict averaging was maintained in the early stages of refinement. Subsequently, the core LBD structure was restrained according to the noncrystallographic symmetry so as to optimize the refinement.

Overall Structure.

The LBD of dmUSP has a similar overall fold to other members of the nuclear hormone receptor family (Fig. 1a) and resembles a three-layered helical sandwich. To facilitate comparison with previous structures, the nomenclature used here is based on the apo-hRXRα structure (23). Helices 1 and 3 comprise one outer layer. Helices 4, 5, 8, and 9 form the central layer. Helices 7 and 10/11 form the other outer layer. The central layer is incomplete, leaving a cavity below helices 4 and 5 between helices 3, 7, and 10/11. Acting as β-sheet S1–S2 (between helices 5 and 6), helix 6 and helix 12 serve to seal off the edges of the cavity. dmUSP contains a glycine-rich insertion following helix 5. This region generally is disordered in the structure and 25 amino acids () could not be modeled. Similarly, residues 230–237 and 503–508 at the termini were not seen in the electron-density maps.

In terms of sequence (and functional) similarity, dmUSP LBD is related most closely to the mammalian RXR. Alignment with hRXRα indicates 70% similarity (42% identity). Superposition of the dmUSP structure with those of the apo and holo forms of hRXRα (Fig. 1 b and c) give rms-deviation (RMSD) values of 1.33 and 1.22 Å (with 169 and 164 matched Cα atoms, respectively). Superposition of dmUSP on the hERα structure gives an RMSD of 1.12 Å with 172 matched Cα atoms. Thus, despite the lower sequence similarity of 58% (23% identity), the dmUSP structure is more similar to the holo hERα structure than the hRXRα structures, particularly in terms of the orientations/conformations of helices 1, 3, and 4.

USP Is Locked Into an Inactive Conformation.

Two regions of the dmUSP structure differ significantly from both the hRXRα and holo hERα structures: the loop joining helices 1 and 3; and helix 12 (shown in red and yellow in Fig. 1). This H1–H3 loop adopts an unusual position in dmUSP not seen in any previous nuclear receptor structure. It wraps over the top of helix 3 and lies between helices 3 and 11. In most of the nuclear receptor structures [e.g., apo-hRXRα, holo-H. sapiens peroxisome proliferator-activated receptor γ (18), and holo-H. sapiens thyroid receptor β (25)], this loop passes outside the β-strand (Fig. 1b, Right). An exception is the hERα structure in which the loop passes between the β-strand and helix 3 (Fig. 1 d and e).

The position of this loop in dmUSP is significant because it prevents helix 12 from adopting the position seen in either the unliganded or 9-cis retinoic acid (9cRA)-bound structures of hRXRα (Fig. 1 b and c). As a consequence, helix 12 adopts a position that closely resembles that seen in the structure of estrogen receptor bound to an antagonist (Fig. 1e; refs. 19 and 20). However, in contrast to estrogen receptor, helix 12 is locked firmly in this inactive position by making extensive contacts to the H1–H3 loop (Fig. 2 a and b).

Sequence comparison of the various USP proteins reveals a sequence motif within the H1–H3 loop that is conserved in both diptera (flies) and lepidoptera (moths) USP proteins but not in hymenoptera (honey bee, GenBank accession no. AF263459), orthoptera (migratory locust; ref. 2), or other arthropods such as crab and tick (highlighted in Fig. 2c; refs. 26 and 27). The structure shows that every residue within this conserved motif plays an architectural role in mediating interactions with helices 3, 4/5, 11, and 12 as well as the loop between helices 11 and 12.

This sequence conservation and extensive pattern of interactions clearly indicate that the position of the H1–H3 loop and of helix 12 are bona fide features of the structure that are not the result of crystal-packing interactions. This finding is supported further by the evidence that all six copies of the dmUSP in the crystal have identical conformations in this respect.

An Unexpected Ligand for Ultraspiracle.

One of the goals of determining the structure of USP was to ascertain whether it has a ligand-binding cavity and hence whether its activity is likely to be regulated by ligand. The structure reveals that dmUSP has an unusually large hydrophobic cavity bounded by helices 3, 5, 6, 7, and 11 as well as the β-sheet S1–S2 (Fig. 3a). Furthermore, in the earliest electron-density maps (averaged according to the noncrystallographic symmetry) it was apparent that this cavity contained additional electron density that could not be accounted for by the protein (Fig. 3c). This density resembled a lipid molecule with two fatty acyl chains and a relatively electron-dense phosphate head group.

TLC of the purified USP confirmed the presence of several classes of lipid exhibiting different mobilities in the various solvents assayed. The more polar species migrates in a similar position to that of phosphatidyl glycerol, and the less polar species shows mobility consistent with diacylglycerol.

MALDI/time-of-flight mass spectra also suggested that there was a mixture of lipids bound to the protein. The observed masses are consistent with diacylglycerol, phosphatidic acid, and phosphatidyl ethanolamine with various acyl chain lengths and oxidation states. One intense peak that was seen in all samples tested (but not in the negative controls) correlates with a mass of 563.5 Da, which is consistent with diacylglycerol (C14:0/C16:0). MALDI/postsource-decay fragmentation of this species supported this assignment.

In conclusion, it seems that a mixture of lipid ligands is bound to the dmUSP LBD including diacylglycerol and various phospholipids. The most common lipids in E. coli are phosphatidyl ethanolamine, phosphatidyl glycerol, and cardiolipin (28) with acyl chains of C16:0 (43%), C16:1 (33%), and C18:1 (24%) (29). To model this mixture of lipids best, phosphatidic acid with C16 acyl chains was built into the averaged electron density (Fig. 3 a and b) and included in subsequent refinement.

Lipid Binding Affects the Dynamics of USP.

During the later stages of the refinement, it became apparent that the occupancy of the lipid is markedly lower in three of the six LBDs. This difference is clear in the 2F_o − F_c electron-density maps (Fig. 3 d and e). The differences in ligand occupancy can be explained by differences in the environment of the six LBDs in the crystal. Significantly, the lower ligand occupancy is not correlated with any major structural differences in the LBDs. The only notable difference is that two regions of the structure (magenta in Fig. 3 a and b) exhibit elevated temperature factors in those LBDs with reduced ligand occupancy (62 vs. 39 Ų and 50 vs. 34 Ų). The first region comprises the turn before helix 3 (residues 266–275) and the second region comprises part of the β-sheet and helix 6 (residues 367–379).

These regions form a gateway to the ligand-binding pocket. One might expect therefore that they are mobile in an unliganded LBD so as to facilitate the entry of a rather large ligand. Once the ligand is bound, this gateway is stabilized by hydrophobic side chains packing against the lipid. The change in mobility of these two regions would seem to be the only difference between the apo and holo forms of the LBD.

An Unusually Large Ligand-Binding Cavity.

Fig. 4 a and b shows a comparison of dmUSP bound to phosphatidic acid and hRXRα bound to 9cRA. The position of the ligand in dmUSP is notably different from that in hRXRα, and that in the ligand-binding cavity of dmUSP is far larger. These differences are the result of several concerted changes in the protein structure. In hRXRα, the amino terminus of helix 3 is curved around the ligand and in toward the main body of the receptor (Fig. 4c). In dmUSP, helix 3 is straighter and has an additional turn at the amino terminus that serves to open a channel from the core of the dmUSP ligand-binding pocket to the outside of the protein.

Whereas helix 3 is one-turn longer in dmUSP, helix 11 is shorter by one turn (Fig. 4d). The loop after helix 11 adopts an extended conformation making several backbone hydrogen bonds and many nonpolar contacts with the H1–H3 loop. Together, the H11–H12 and H1–H3 loops form a wall on one side of the channel. On the other side, the cavity is substantially open to solvent between helix 3 and the β-sheet (see Fig. 3b).

The ligand-binding cavity in dmUSP is formed from four regions of the protein (Fig. 4e): the H1–H3 loop and helix 3; helix 5; the β-sheet S1–S2 along with helices 6 and 7; and also helix 11 and the H11–H12 loop. The interior surface of the ligand-binding cavity is almost entirely nonpolar. The only polar contact with the ligand is a hydrogen bond between the amide of Q271 and a carbonyl of the lipid. After binding ligand, 31 residues become shielded from solvent (Fig. 4f). Despite this, the cavity is not filled completely by the ligand, and there remains a small unoccupied volume adjacent to helix 3.

The residues that surround the 9cRA in the holo hRXRα ligand-binding cavity are essentially all conserved in dmUSP (in size and character if not identity) with the exception of the arginine that forms a salt bridge with the carboxyl of the retinoic acid. In dmUSP, this arginine is a cysteine, and because of its shorter side chain it does not participate in the ligand-binding pocket. It is potentially possible that USP could bind 9cRA if this cysteine were mutated to an arginine, although the enlarged ligand-binding cavity would be far from filled.