The crystal structure of guinea pig 11beta-hydroxysteroid dehydrogenase type 1 provides a model for enzyme-lipid bilayer interactions.

The metabolic reduction of 11-keto groups in glucocorticoid steroids such as cortisone leads to the nuclear receptor ligand cortisol. This conversion is an example of pre-receptor regulation and constitutes a novel pharmacological target for the treatment of metabolic disorders such as insulin resistance and possibly other derangements observed in the metabolic syndrome, such as hyperlipidemia, hypertension, and lowered insulin secretion. This reaction is carried out by the NADPH-dependent type 1 11beta-hydroxysteroid dehydrogenase (11beta-HSD1), an enzyme attached through an integral N-terminal transmembrane helix to the lipid bilayer and located with its active site within the lumen of the endoplasmic reticulum. Here we report the crystal structure of recombinant guinea pig 11beta-HSD1. This variant was determined in complex with NADP at 2.5 A resolution and crystallized in the presence of detergent and guanidinium hydrochloride. The overall structure of guinea pig 11beta-HSD1 shows a clear relationship to other members of the superfamily of short-chain dehydrogenases/reductases but harbors a unique C-terminal helical segment that fulfills three essential functions and accordingly is involved in subunit interactions, contributes to active site architecture, and is necessary for lipid-membrane interactions. The structure provides a model for enzyme-lipid bilayer interactions and suggests a funneling of lipophilic substrates such as steroid hormones from the hydrophobic membrane environment to the enzyme active site.

Glucocorticoid hormones control essential functions such as development, immune response, and metabolism. The primary mode of action is accomplished through binding to intracellular glucocorticoid receptors, which, in combination with other factors, can induce or repress gene transcription. This basic principle is in a tissue-specific manner modulated by the enzyme system 11␤-hydroxysteroid dehydrogenase (11␤-HSD), 1 which interconverts the steroid C11 carbonyl/␤-hydroxyl groups in cortisone and cortisol (in humans) or dehydrocorticosterone and corticosterone (in rodents) (1)(2)(3)(4). Free plasma levels of cortisol/corticosterone are low, due to tight plasma protein (corticosteroid-binding globulin) binding, whereas cortisone/dehydrocorticosterone do not bind, resulting in high free levels of this glucocorticoid precursor, to be taken up by target cells. Cortisol/corticosterone bind to the glucocorticoid receptor and can exhibit in this receptor-ligand complex transcriptional activity; in contrast, cortisone/dehydrocorticosterone cannot. Thus 11␤-HSD is a central "cellular switch," which decides whether and how much receptor ligand is available (1)(2)(3)(4).
At present, two isozymes (11␤-HSD1 and 11␤-HSD2) are known, and both act seemingly in a mutually exclusive manner (1,3). Both are bound through N-terminal transmembrane domains to the endoplasmic reticulum (ER) (1,3). The active site of 11␤-HSD1 is located in the lumenal compartment of the ER, whereas 11␤-HSD2 faces the cytosolic space. These specific localizations explain why 11␤-HSD1 is in most instances an NADPH-dependent cortisone reductase, whereas the type 2 enzyme acts as an NAD ϩ -dependent dehydrogenase of cortisol. The cellular functions of 11␤-HSD1 (as a cortisol-producing catalyst) in insulin resistance and metabolic disorders establish this enzyme as a novel pharmacological target in noninsulin-dependent diabetes mellitus, and successful inhibitor developments have been reported (5,6).
Both isozymes belong to the superfamily of short-chain dehydrogenases/reductases (SDR) (1,3,4), a large family of enzymes with well over 3000 primary structures from all forms of life deposited in databases (7,8). SDR enzymes display a low level of residue identities (typically 10 -30%) sharing few conserved sequence motifs, which are necessary for the maintenance of fold and catalysis (8,9). About 30 three-dimensional structures of this family have been deposited in the PDB, showing a nearly superimposable ␣/␤-folding pattern with a central Rossmann fold for nucleotide cofactor binding (8,9). However, all enzymes show unique active site cavities, thus explaining the large variability in substrate specificities found in SDR enzymes. Purification and thorough characterization of the membrane-attached 11␤-HSD isozymes have been difficult, and no structure determination has been reported yet. We recently described the successful expression and kinetic characterization of transmembrane domain-deleted human and guinea pig 11␤-HSD1 variants (10,11). We now present the successful crystallization and structure determination at 2.5 Å of the guinea pig enzyme and discuss the structural implications on enzyme function and membrane attachment.
Crystallization of Guinea Pig 11␤-HSD1-Prior to crystallization, the purified protein was diluted 10 times with 20 mM Tris-HCl, pH 8.0, buffer containing 2 mM Tris(2-carboxyethyl)phosphine, 5% glycerol, 0.5 M GuHCl, 0.05% Anapoe X-100, 20 M cortisone, and 40 M NADP analogue, 3-aminopyridine adenine dinucleotide phosphate (AADP; Sigma) and concentrated back to the starting volume using the concentration device described above. This procedure was repeated three times to attempt the exchange of NADP ϩ and BVT.4584 with AADP and cortisone, respectively. The protein was finally concentrated down to around 10 mg/ml and centrifuged in a Microfuge for 30 min to remove any aggregates. Crystals were grown by hanging drop vapor diffusion at 18°C. For crystal growth, a 2-l drop of the concentrated protein solution was mixed with an equal volume of precipitating solution containing polyethylene glycol 550 mono methyl ether (Fluka)(38 -40%) and 0.1 M BisTris, pH 6.5, and equilibrated against 1 ml of the precipitating solution. The resulting tetragonal crystals containing a single guinea pig 11␤-HSD1 dimer per asymmetric unit (space group I422; a ϭ 118.3 Å, b ϭ 118.3 Å c ϭ 184.7Å, ␣ ϭ ␤ ϭ ␥ ϭ 90°) appeared within 5-6 days.
X-ray Diffraction Data Collection and Processing-Single crystals (ϳ0.25 ϫ 0.15 ϫ 0.15 mm) were transferred directly to a loop and flash-frozen in a stream of nitrogen gas at 100°K. Data were collected at the I711 beamline at MaxLAB synchrotron, Lund, Sweden, using a MAR area detector and at our home laboratory on a Rigaku RU-300 x-ray generator and Raxis IVϩϩ image plate detector system. All data were processed using the HKL and CCP4 software packages (12,13).
Structure Solution-The structure was solved by single isomorphous replacement using the autoSHARP (14). Two mercury compounds, mer-cury chloride (HgCl 2 ) and mercury cyanate (HgCN), were used to prepare heavy atom derivative crystals, both yielding the same two major binding sites. A total of 6612 reflections to 3.3 Å were phased using the HgCl2 derivative, and 15,447 reflections to 2.7 Å were phased with the HgCN derivative with phasing powers of 1.92 and 1.23, respectively. As implemented in SHARP, solvent flattening using Solomon (15) was run on experimental electron density maps from each hand, and the resulting correlation coefficients on E 2 clearly indicated the correct solution (0.189 versus 0.105). After the solvent-flattening procedure, the figure of merit was improved from 0.34 to 0.76. The resulting electron density maps were of sufficient quality to place most residues of the guinea pig 11␤-HSD1 dimer. Clear density for the cofactor was also observed. The model was built manually using O (16), and subsequent rounds of refinement were carried out using REFMAC (17). During the final refinement stages, water molecules were added to the protein model using ARP/wARP (18). The final model contained residues 24 -300 from chain A and 24 -292 from chain B, 2 NADP and 70 water molecules in the asymmetric unit with 90.3% of residues in the most favored regions of the Ramachandran plot, and 2 residues in the disallowed region. The model has root mean square deviations of 0.015 Å in bond lengths and 1.7 Å in bond angles. The R cryst and R free values of the final model were 19.2 and 26.7%, respectively (Table I). The atomic coordinates have been deposited at the Protein Data Bank with accession code 1XSE.

RESULTS AND DISCUSSION
Crystallization of Guinea Pig 11␤-HSD1-11␤-HSD1 is an N-linked glycosylated protein bound to the membrane of the endoplasmic reticulum, and its highly hydrophobic nature has made structural studies challenging. The variant crystallized in this study is devoid of the membrane-anchoring N-terminal portion; still, it is only moderately soluble and in addition has a tendency to form large aggregates (11). To define conditions under which 11␤-HSD1 exists as a monodisperse element in solution, we screened combinations of different detergents with varying subdenaturating concentrations of GuHCl. This screening resulted in a crystallization solution containing 0.05% Triton X-100 and 0.5 M GuHCl. It should be noted that the 11␤-HSD1 retains full dehydrogenase and reductase activity after a 16-h incubation under these conditions (11). Up to now, only few proteins have been crystallized in the presence of GuHCl, and 11␤-HSD1 is the first example of a protein crystallized in the presence of both GuHCl and a detergent. Our preliminary results indicate that combinations of detergent and denaturants could be a general route to monodispersity and therefore a possible way of improving the odds for obtaining high quality crystals of low solubility proteins. Although the cofactor could be unambiguously modeled in the structure, the continuous positive difference density present in the sub- Overall Structure of Guinea Pig 11␤-HSD1-The crystal structure reveals one dimer per asymmetric unit, most probably representing the physiologically active entity (11). The monomeric subunit of 11␤-HSD1 shows the typical SDR architecture consisting of a central seven-stranded parallel ␤-sheet (␤C-␤A, ␤D-␤G), twisted over all by 90°, sandwiched between three helices on each side (␣D-␣F and ␣C-␣G) (Fig. 1). Strands ␤A to ␤F constitute the central part of the Rossman fold associated with NAD/NADP binding. The segment connecting ␤F and ␣G, residues 217-237, forms a lid over the substratebinding pocket and contains a double turn ␣-helix (␣1, residues 220 -228). The C-terminal segment following ␤G, residues 260 -297, protrudes from the domain and forms a long helical structure (␣2 and ␣3) observed to pack against the C terminus of its dimer partner in an antiparallel manner. This segment interacts mainly with the second subunit and extends the dimerization interface, which for SDR enzymes is usually limited to helices E and F, resulting in a total buried surface of 3308 Å 2 for each monomer. Notably, ␣3 and the loop preceding it interact with the substrate-binding loop of the second subunit and form parts of the substrate-binding pocket of that subunit. This C-terminal arrangement is reminiscent of dimeric Drosophila alcohol dehydrogenase, in which the C terminus serves to cover part of the active site of the adjacent subunit and furthermore is involved in subunit dimerization ((19) PDB accession code 1A4U). This structural feature is even more pronounced in the dimeric medium-chain dehydrogenase/ reductase enzymes in which the two substrate pockets are lined with residues from both subunits (20). Together with a pro-posed role in membrane interactions (discussed below), the C-terminal regions thus fulfill essential roles in subunit interactions, membrane topology, and active site architecture.
At present, primary structures of 11␤-HSD1 have been determined from 13 different species, and a pattern of conserved and variable segments of 11␤-HSD1s has been established. 2 Six variable segments (V1-V6) can be defined, and they all cluster together to form parts of the substrate-binding pocket and the dimer interface (Fig. 1). Regions V1 (residues 122-131), V4 (residues 225-236), V5 (residues 259 -269), and V6 (residues 280 -300) of the second subunit form the substratebinding pocket, whereas V2 (residues 176 -182) form parts of the dimer interface interacting with V3 (residues 198 -205) and V6 of the second subunit. The low homology around the substrate-binding pocket is manifested in the different substrate specificities (cortisol/corticosterone, 7␣/7␤-hydroxycholesterol) and the fact that species-specific inhibitors are frequently found. 2 The guinea pig 11␤-HSD1 has an 8-residue extension in the C terminus when compared with the other orthologue sequences determined.
Structural Features of Guinea Pig 11␤-HSD1 and Relationship to the Short-chain Dehydrogenase/Reductase Family-The closest structural neighbors to guinea pig 11␤-HSD1 are SDR enzymes such as porcine carbonyl reductase (PDB accession code 1CYD), trihydroxynaphthalene reductase from Magnaporthe grisea (PDB accession code 1DOH), human estradiol 17␤-hydroxysteroid dehydrogenase type 1 (PDB accession code 1BHS), and E. coli 7␣-hydroxysteroid dehydrogenase (PDB accession code 1AHI), showing high similarity (about 2 Å root mean square deviation over 240 residues) in the central scaffold. In particular, the essential sequence motifs of the SDR family are found at homologous positions, such as the N-terminal sequence TGX 3 GXG (positions 17-24), necessary for nucleotide binding, and the active site tetrad consisting of Asn-119, Ser-170, Tyr-183, and Lys-187 (21-24). Major differences between these enzymes are seen in their substrate-binding loop, probably reflecting the differences in substrate specificity, as well as in the previously mentioned C-terminal fragment, reflecting differences in quaternary structure arrangement.
A segment of IRVN residues, with Asn constituting an Nlinked glycosylation site in 11␤-HSD1, is highly conserved in SDR enzymes. Mutational studies on this residue in a related steroid dehydrogenase showed the essential nature for catalysis, and sequence comparisons suggested a homologous role for Asn-207 found in 11␤-HSD1 (9). Mutational analysis of this residue in rat 11␤-HSD1 indeed revealed that it is essential for activity (25). Later experiments, however, suggested that the glycan moiety itself is not a prerequisite for activity (26). In the guinea pig 11␤-HSD1 structure, the conserved Asn-207 is located at the enzyme surface in a loop connecting helix F and strand F, which is consistent with glycan modification in a eukaryotic cellular ER environment. Despite an apparent homology to the conserved Asn residue in the IRVN sequence of SDRs (9), Asn-207 in guinea pig 11␤-HSD1 does not connect the active site and the C-terminal part of SDRs; instead, this stabilizing function is fulfilled by Thr-211. Its side-chain OH connects through a water molecule to the OH of Ser-164, located in strand E close to the active site, and to the carbonyl of the C-terminal residue Gln-253. Thus the role of Asn-207 in catalysis remains unknown; however, it is possible that Asn-207 is necessary for glycan modification and subsequent folding in the ER of mammalian cells.

FIG. 1. Structural organization of the guinea pig 11␤-HSD1 subunit.
A ribbons representation of the 11␤-HSD1 with secondary structural elements annotated is shown. The subunit interface is formed by helices ␣F/␣2 and segments V3/V6/␣3 of the two monomers. Segments with low sequence identity between orthologues are shown in red (V1-V6). The guinea pig orthologue-specific C-terminal extension is shown in purple. Selected parts of the second subunit are shown in light gray and pink (V3 and V6). NADP is in ball-and-stick representation with carbon, oxygen, nitrogen, and phosphor atoms colored in green, red, blue, and purple, respectively.
Cys-241, are found in all sequenced 11␤-HSD1 variants, whereas only the porcine orthologue shows variation at position 78 (Ser instead of Cys). A disulfide bond involving Cys-78 and Cys-213 has been suggested to be present in the rabbit orthologue (27).
This hypothesis is apparently supported by the subcellular localization of the catalytic domain of 11␤-HSD1 within the ER lumen, which provides an oxidative environment suitable for disul- fide bond formation. However, examination of the guinea pig structure raises serious doubts about the possibility of an intramolecular disulfide bond to be formed in 11␤-HSD1 variants. Although the distance between the ␣-carbons of Cys-78 and Cys-213 is about 20 Å, the corresponding distance between the closest cysteine pair, Cys-213 and Cys-241, is almost 10 Å, thus excluding intrasubunit disulfide formation.
Active Site Topography and Catalysis-The general oxidoreductase mechanism of SDRs is reflected in a common active site configuration. The binary 11␤-HSD1-NADP ϩ structure fulfills the requirements for a catalytically competent complex. In general, hydride transfer in SDR enzymes is carried through by the C4-H from the S-face of the coenzyme (28), in which Tyr-183 acts as an acid-base catalyst, Lys-187 lowers the pK a of the tyrosine hydroxyl, and Ser-170 stabilizes the substrate carbonyl group (28). In addition, the Tyr and Lys residues form together with the 2Ј-OH of the nicotinamide ribose, a water molecule, and the main-chain carbonyl of the conserved Asn-143 residue, a hydrogen-bonding network, suggesting a proton relay from the active site to the bulk solvent (  (29); accordingly, in guinea pig 11␤-HSD1, the carbonyl of Gly-193 is oriented toward the C4h of the R-side of NADP. The pyrophosphate and adenine ribose moieties are bound in a cavity built up mainly by the conserved residues Gly-41-Ile-46 and the segment Ile-220 -Thr-222. A mostly apolar surface for the adenine ring is created by segments Gly-91-Met-93 and residues His-120, Val-121, and Val-142, whereas the other side of the ring is dominated by the adenine ribose. The 2Ј-OH phosphate of NADP is bound to Ser-67 and Arg-66, but not to Lys-44, as observed in other NADP(H)-dependent SDRs, such as mouse lung carbonyl reductase (30). Accordingly, NAD(P)(H) specificity versus NAD(H) is achieved through Arg-66. This residue is not present in NAD(H)-dependent SDRs and is also preceded in these enzymes by an Asp residue conferring electrostatic repulsion to a 2Ј-phosphate (30,31).
The shape of the substrate-binding pocket is compatible with its function to accommodate cortisol/cortisone and 7␤hydroxycholesterol/7-ketocholesterol. The substrate-accessible void is 640 Å 3 and is predominantly lined by non-polar residues (Fig. 2a). To gain insights into the determinants for substrate binding, cortisol was manually docked into the present structure (Fig. 2b). Carbon 11 was fixed within hydride transfer distance to C4 and the C11 hydroxyl group within hydrogen-bonding distance to Ser-170 and Tyr-183. These experiments reveal that cortisol fits well into the active site, provided that a side-chain movement of Tyr-123 occurs for accommodation of the substrate. Residues in ␣1, the nicotinamide ring of NADP, and residues in variable segment 1 constitute the lower part of the pocket, whereas Ile-180 and Leu-217 are crucial for the middle part. The upper part, which according to our modeling should accommodate the A-ring of the substrate, offers several possible hydrogen bond donors to the 3-keto group of cortisol, e.g. the side-chain hydroxyls of Tyr-177 and Tyr-231 and the backbone amide nitrogen of Leu-217. Notably, Ser-280 and Tyr-284 of the second subunit are lining the upper part of the substrate-binding pocket. The pocket stretches further up at this end, and the substrate-binding loop, the loop connecting ␤F and ␣2, together with ␣2 of the second subunit form a possible entrance.
Orientation of 11␤-HSD1 with Respect to the Lipid Membrane-The location of the N termini together with the amphipathic nature of the C-terminal helices in the 11␤-HSD1 dimer suggest a possible orientation of the molecule with respect to the membrane surface (Figs. 3 and 4). The two Cterminal helical segments from the dimer form a non-polar plateau (ϳ3000 Å 2 ), which is encircled by a ring of positively charged residues (Fig. 3). This arrangement suggests that the plateau is located in the non-polar center of the membrane with the charged residues forming salt bridges with the displaced phospholipid and sulfolipid head groups. This model characterizes monotopic membrane proteins, and such a non-polar surface contact formed at a dimer interface has been previously observed for squalene cyclase (32) and prostaglandin-H2synthase (33).
A possible entrance to the substrate-binding pocket is formed by the substrate-binding loop, the loop connecting ␤F, and helices ␣2 and ␣2 of the second subunit. With the suggested membrane orientation, this entrance makes the catalytic site accessible for lipophilic substrates such as cortisol or 7␤-hydroxycholesterol, enriched in the lipid membrane. Accordingly, this model intuitively suggests a funneling of hydrophobic steroids from the membrane into the active site cleft. This hypothesis further explains the discrepancies between efficient steroid conversion at low concentrations and the somewhat higher apparent K m values observed in vitro (10,34).
Taken together, this study provides detailed insight into the structural features and active site architecture of a physiologically important hormone-activating enzyme and drug target. Furthermore, the observed hydrophobic surface and the proposed membrane association might constitute a general model for lipid interactions of similar membrane-bound SDR enzymes such as 17␤-HSD types 2 and 3 or enzymes involved in retinoic acid metabolism such as retinol dehydrogenases.