Structure of the Human 3 (cid:1) -Hydroxysteroid Dehydrogenase Type 3 in Complex with Testosterone and NADP at 1.25-Å Resolution*

The first crystallographic structure of human type 3 3 (cid:1) -hydroxysteroid dehydrogenase (3 (cid:1) -HSD3, AKR1C2), an enzyme playing a critical role in steroid hormone metabolism, has been determined in complex with testosterone and NADP at 1.25-Å resolution. The enzyme’s 17 (cid:2) -HSD activity was studied in comparison with its 3 (cid:1) -HSD activity. The enzyme catalyzes the inactivation of dihydrotestosterone into 5 (cid:1) -androstane-3 (cid:1) ,17 (cid:2) -diol (3 (cid:1) -diol) as well as the transformation of androstenedione into testosterone. Using our homogeneous and highly active enzyme preparation, we have obtained 150-fold higher 3 (cid:1) -HSD specificity as compared with the former reports in the literature. Although the rat and the human 3 (cid:1) -HSDs share 81% sequence homology, our structure reveals significantly different geometries of the active sites. Substitution of the Ser 222 by a histidine in the human enzyme may compel the steroid to adopt a different binding to that previously described for the rat (Bennett, bottom penalty penalty-free C atoms 1 Å crystallographic (outer) shell of the amino residues belonging to the flexible shell at least one atom within 10 Å of testosterone or the cofactor. The residues in the outer shell were kept rigid during energy minimizations. vector (pivot) testosterone atoms O17 O3 the crystallographic structure the ternary the cofactor. along the pivot with a step of 1 Å, an MCM energy A position of the steroid imposed constraining atom O17 the plane normal the pivot and crossing it the level a displacement parameter s , hereafter referred the testosterone position. The displacement s (cid:2) Å (position 0) corresponds to our x-ray structure, positive values of s indicating displace- ment of the steroid toward the cofactor. It be that the atom-plane constraint one of the six positional and rota- tional degrees of freedom of the ligand. Twelve MCM trajectories were calculated for 12 positions of testosterone. In each MCM trajectory, torsion angles of amino acids and testosterone and generalized coordinates characterizing position and orientation the steroid were sampled. In the subsequent energy minimizations, all generalized coordinates of the system, including those of the cofactor, were allowed to vary. Each MCM the lowest energy structure for the given of the steroid. These structures are visualized with the help of program RASMOL. Several energetic and geometric parameters of these structures are plotted against position of the steroid. The MCM energy profile was computed with the help of the ZMM program described Ref. and

Human 3␣-hydroxysteroid dehydrogenases (3␣-HSDs; 1 EC 1.1.1.213) work in concert with the 5␣/5␤-steroid reductases to convert steroid hormones into the 3␣/5␣ and 3␣/5␤-tetrahydrosteroids. These isozymes catalyze the inactivation of androgens, estrogens, progestins, and glucocorticoids. However, the inactivation of the most potent androgen 5␣-dihydrotestosterone (5␣-DHT) to 5␣-androstane-3␣,17␤-diol (3␣-diol) is its best known function (1). These isoenzymes thus play a major role in the regulation of the intracellular concentration of 3␣-DHT in peripheral tissues, especially in the androgen-sensitive prostate that is susceptible to benign prostatic hyperplasia and prostate cancer. Testosterone, after entering prostatic cells, is transformed to 5␣-DHT by 3-oxo-5␣-steroid-4-dehydrogenase (2,3). 5␣-DHT is a more potent androgen than testosterone in stimulating prostate cancer growth (4,5) and preferentially binds to the androgen receptor (dissociation constant (K d ) for the androgen receptor of 10 Ϫ11 M) (6). Elevation of 5␣-DHT content in prostate has been associated with benign prostatic hyperplasia in humans (6,7) and with human prostate carcinoma (8,9). The action of 5␣-DHT may be terminated by 3␣-HSD, which catalyzes the inactivation of 5␣-DHT to 3␣androstanediol (a weak androgen; K d for the androgen receptor of 10 Ϫ6 M) (1). It has also been proposed that, by catalyzing the reverse reaction, 3␣-HSD may function as a molecular switch and, in this manner, may regulate the amount of 5␣-DHT available for androgen receptor binding and activation. Three isoforms of 3␣-HSD have been described in human tissues (10 -12) as playing critical roles in sex hormone metabolism and action, such as regulating the occupancy of the androgen receptor. At least two related 3␣-HSD isoforms (type 2 and type 3) have been detected in the human prostate (13,14). These reports have shown that both enzymes eliminate 5␣-DHT but that only the type 3 forms the active hormone 5␣-DHT. Thus, 3␣-HSD3 may increase the pool of active androgens in the prostate. Through its action on allopregnenolone, 3␣-HSD also regulates the GABA A receptor (K d ϭ 10 Ϫ9 M), a membranebound chloride ion-gated channel, and may have profound effects on the receptor function (15,16). The types 1 and 3 3␣-HSDs (AKR1C4 and AKR1C2, respectively) are highly homologous and catalyze the above reactions. It is reported that the type 1 enzyme is the most efficient in DHT reduction, with its substrate specificity (expressed in k cat /K m ) being 10-to 40-fold higher than the other members of the family. The type 2 enzyme (AKR1C3) shows a significantly higher activity in catalyzing the transformation of androstenedione (4-dione) to testosterone than the other two types of 3␣-HSD. In this re-* This work was supported by EndoRecherche and the Medical Research Council of Canada. Data collection was carried out at the National Synchrotron Light Source, Brookhaven National Laboratory, which is supported by the United States Department of Energy, Division of Materials Sciences and Division of Chemical Sciences. This work was also supported by a joint grant from the Natural Science and Engineering Council and Medical Research Council of Canada for a "consortium for operation of a protein crystallographic synchrotron beam line." 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 gard, it is more similar to 17␤-HSD and therefore was first named type 5 17␤-HSD (17,18). Both the type 2 and 3 enzymes showed a wide distribution of their mRNA in discrete brain regions (19).
3␣-HSDs belong to the aldo-keto-reductase superfamily (AKR) and have no significant sequence similarity to 17␤-HSDs (20 -22) or 11␤-HSD (23). The latter two HSD families belong to the short chain dehydrogenase-reductase superfamily. The human type 3 3␣-HSD, cloned in our laboratory (12), is the only member of the 3␣-HSD family that converts 3␣-diol to the most active androgen 5␣-DHT, despite the functional plasticity of members of the family (14). This enzyme shows a much higher amino acid sequence identity to human 20␣-hydroxysteroid dehydrogenase (20␣-HSD) than to the type 1 and 2 3␣-HSDs. There are only seven amino acid differences over the 323 residues between the type 3 3␣-HSD and 20␣-HSD (AKR1C1) (12). The structure of the rat 3␣-HSD complexed with NADP ϩ and testosterone has been solved at 2.5-Å resolution (24). At the difference of the rat enzyme, human 3␣-HSDs are present in different isoforms with varying steroid specificities. Our structure represents the first structure of a prostatic 3␣-HSD.
Detailed studies of prostatic 3␣-HSD3 are required to understand its role in androgen metabolism and action. In this paper, we describe the structure of the human 3␣-HSD3 at near atomic resolution, in complex with NADP and testosterone. With such resolution, we are able to describe in detail the interactions occurring between the different partners. This is critical to understand the mechanism of recognition of these multispecific enzymes. Description of the detailed interactions between the enzyme, NADP, and the steroid constitutes an important step that may lead to rational drug design. The search of inhibitors for the oxidative activity of 3␣-HSD3 based on structural evidence would be a step forward in the treatment of prostate cancer. In our study, the same preparation with high specificity was used for both kinetics and crystallization. The kinetic results are also reported and compared with the structural information.
Data Collection, Molecular Replacement, and Refinement-Diffraction data were collected from one cryo-cooled crystal at beamline X8C at the National Synchrotron Light Source while the crystal was oscillated through 1°steps. The crystal diffracts up to 1.0 Å, and a complete data set has been collected at 1.25-Å resolution. The diffraction data were processed using the HKL suite (26). The crystal belongs to the monoclinic space group P2 1 with a ϭ 55.1 Å, b ϭ 87.2 Å, c ϭ 76.9 Å, and ␤ ϭ 107.4°and contains two molecules in the asymmetric unit. The structure of the ternary complex was solved using the program EPMR; the search model used was the refined 2.5-Å resolution structure of the rat 3␣-HSD in complex with testosterone and NADP ϩ (24) (Protein Data Bank entry 1AFS (27)). The positions of the ligands in the 3␣-HSD3 active site were apparent in electron density maps calculated using 2F o Ϫ F c and F o Ϫ F c coefficients. Iterative cycles of model building using the program O (28) were followed by restrained refinements using the programs CNS (29) and refmac (30) and automated water building using the program ARP (31). The stereochemistry of the final model was verified with the program PROCHECK (32). Refinement statistics are summarized in Table I. Coordinates have been deposited in the Protein Data Bank with the accession number 1J96 (27).
Monte Carlo Minimization (MCM)-To predict the optimal position of testosterone in the active site of 3␣-HSD3, we have used a computational strategy similar to that proposed by Zhorov and Lin (33) in the modeling study of estradiol binding to type I 17␤-HSD. In addition to the energy components described in Ref. 33, the energy expression included hydration energy calculated by the method of Lazaridis and Karplus (34) and energy of deformation of bond angles in testosterone. The enzyme was represented by a two-shell model based on our crystallographic structure. The first (inner) shell of the model included amino acid residues having at least one atom within 6 Å of testosterone or the cofactor in the x-ray structure. All residues in the inner shell were treated as flexible, their torsion angles being allowed to vary during energy minimizations. C ␣ atoms of the flexible-shell residues were constrained by pins, flat bottom penalty functions that allowed penalty-free deviation of C ␣ atoms up to 1 Å from their crystallographic positions. The second (outer) shell of the model included amino acid residues not belonging to the flexible shell and having at least one atom within 10 Å of testosterone or the cofactor. The residues in the outer shell were kept rigid during energy minimizations.
A vector (pivot) was drawn between testosterone atoms O17 and O3 in the crystallographic structure of the ternary complex to specify the direction from the entrance to the steroid-binding pocket toward the cofactor. Testosterone was pulled along the pivot with a step of 1 Å, and an MCM energy profile against the testosterone position was computed as follows. A given position of the steroid was imposed by constraining atom O17 to the plane normal to the pivot and crossing it at the level characterized by a displacement parameter s, hereafter referred to as the testosterone position. The displacement s ϭ 0 Å (position 0) corresponds to our x-ray structure, positive values of s indicating displacement of the steroid toward the cofactor. It should be noted that the atom-plane constraint removes only one of the six positional and rotational degrees of freedom of the ligand. Twelve MCM trajectories were calculated for 12 positions of testosterone. In each MCM trajectory, torsion angles of amino acids and testosterone and generalized coordinates characterizing position and orientation the steroid were sampled. In the subsequent energy minimizations, all generalized coordinates of the system, including those of the cofactor, were allowed to vary. Each MCM trajectory predicted the lowest energy structure for the given displacement of the steroid. These structures are visualized with the help of program RASMOL. Several energetic and geometric parameters of these structures are plotted against position of the steroid. The MCM energy profile was computed with the help of the ZMM program described in Ref. 33 and references therein.
Radiochemical Determination of Kinetic Parameters-Unlabeled dihydrotestosterone (DHT), 4-dione, testosterone, NADP(H), K 2 HPO 4 /KH 2 PO 4 buffer, Tris-HCl buffer, bovine serum albumin, EDTA, glycerol, and phenylmethylsulfonyl fluoride were obtained from Sigma. 14 C-DHT, 14 C-4-dione, and 14 C-testosterone were purchased from PerkinElmer Life Sciences. Diethyl ether, dichloromethane, toluene, and acetone were obtained from Fisher. The Bio-Rad Protein Assay Kit was purchased from Bio-Rad. Silica-coated aluminum TLC (thin layer chromatography) plates were purchased from BDH. All other reagents were purchased from Sigma. The Storm imaging system was from Molecular Dynamics, Inc. (Sunnyvale, CA). The software used for kinetic data analysis was ENZFITTER (Biosoft, Cambridge, UK) version 1.05.
Steady-state Kinetics-All reactions were performed at 37.0 Ϯ 0.5°C, and the reaction mixtures contained 0.1 M potassium phosphate buffer, pH 7.5, 0.05 mg/ml bovine serum albumin, 100 M NADP(H), and varying amounts of 14 C-labeled steroids. The final content of ethanol for each reaction was standardized to 2%. Reactions were initiated by the addition of the enzyme sample, aliquots were taken, and the reaction was then stopped by the addition of 3 volumes of diethyl ether on ice. The stopped reaction mixture was chilled using a dry ice/ethanol bath, and the aqueous phase was discarded. Diethyl ether was then evaporated, and steroids were resuspended in 60 l of dichloromethane for TLC separation. The migration solvent system consisted of a 4:1 ratio of toluene to acetone. TLCs were than exposed and quantified using a STORM device. Initial velocities were measured with less than 10% substrate consumption and were expressed as nmol/(min⅐mg). Experiments were duplicated, compiled, and then treated by the ENZFITTER program.

RESULTS AND DISCUSSION
Structure of the Human 3␣-HSD3 Ternary Complex-The human type 3 3␣-HSD, in addition to its 3␣-HSD activity, also possesses some 17␤-HSD activity, but the latter activity is much weaker than in the case of the type 2 3␣-HSD. We thus attempted to obtain the enzyme ternary complex with 4-dione and NADP. The crystal structure of the human 3␣-HSD type 3 in the ternary complex was determined by the molecular replacement method using the program EPMR (35). The quality of the refined ternary complex can be assessed from the statistics given in Table I. All data between 10 and 1.25 Å were used in the refinement, yielding a crystallographic R-factor of 18% for 180,677 reflections and a free R-factor of 20% for 8672 reflections. There are two molecules in the asymmetric unit related by a 2-fold noncrystallographic symmetry axis, although the protein is a monomer in solution. The refined model includes two complete ternary complexes, each of which contains an NADP molecule, a testosterone molecule, and an acetate molecule (Fig. 1). In addition, the model includes 858 water molecules.
The protein adopts a well known triose-phosphate isomerase barrel motif, namely an (␣/␤) 8 -barrel with two additional helices. A ternary complex exhibiting good stereochemistry was refined to 1.25 Å. Refinement of this model against the 1.25-Å data has enabled atomic positions in the model to be defined by electron density that discriminates between carbon, nitrogen, and oxygen. Out of the 323 amino acids contained in the protein, 323 were modeled into the electron density for molecule A and 318 for molecule B. In molecule B, the loop (positions 132-136) is disordered, whereas this latter is stabilized by a symmetric molecule in molecule A. Individual isotropic temperature (B) factors were also refined and resulted in the same B values in molecules A and B for the protein and the acetate. Nevertheless, the B factor for testosterone in molecule B is higher than in molecule A. The Ramachandran plot for the ternary complex positions 94.4% of the backbone dihedral angles in the core regions, as defined by Kleywegt and Jones (36). The only residue that falls into the disallowed region is the residue Ser 221 , which is also in a disallowed region in the rat structure, more precisely because the nitrogen of Ser 221 makes a hydrogen bond with the cofactor pyrophosphate group. With the exception of the N-terminal region (Asp 1 -Cys 7 , implicated in the crystal packing), Ser 320 -Tyr 323 in the C terminus (disordered in the rat structure), and the loops interacting with the steroid (see below), the structure obtained for the human 3␣-HSD3 ternary complex is highly similar to the structure reported for the rat 3␣-HSD3 ternary complex (24), as shown by the root mean square deviation of 0.678 Å for 242 C ␣ positions. However, important shifts are observed for the five loops involved in the geometry of the active site, namely loops Tyr 24 -Ser 32 , Ala 52 -Asn 57 , His 117 -Leu 144 , Gly 220 -Pro 233 , and Leu 308 -Phe 319 . Furthermore, in the human 3␣-HSD3, the crystal protein sequence begins with an Asp 1 -Asp 2 -Ser 3 sequence instead of the expected Met 1 -Asp 2 -Ser 3 sequence. This is due to the expression with the addition of the glutathione S-transferase. Since thrombin did not cut all of the linker sequence, Asp 1 was found to remain in the purified protein.
Although the human and the rat models show the same overall structure, the two enzymes demonstrate important differences in the cofactor and steroid binding regions.
NADP Binding Site-The electron density map clearly shows that NADP is bound tightly to 3␣-HSD3. NADP binds at the C-terminal end of the barrel, threading through a short tunnel FIG. 1. Schematic stereoview representation of the 3␣-HSD3 ternary complex. ␣-Helices, ␤-strands, and coils are represented by helical ribbons, arrows, and ropes, respectively. The testosterone, acetate, and NADP molecules are shown as ball and stick representations with purple, yellow, and white bonds, respectively. This image was produced using MolScript (43) and Raster3D (44).  ϭ (h, k, l). c The free R-factor is calculated for a "test" set of reflections that were not included in atomic refinement. mainly formed by van der Waals contacts between the ␤1-␣1 loop (residues 26 and 27), the loop B (residues 222-226), and Lys 270 , with the adenine ring at the periphery and the nicotinamide ring toward the core of the barrel (Figs. 1 and 2). Similarly to the rat structure (37), the orientation of NADP in the cofactor binding site is determined by several hydrogen bonds and salt bridges. At the core of the barrel, the nicotinamide ring stacks against the side chain of Tyr 216 and makes three hydrogen bonds between the carboxy-amide group and the side chains of Ser 166 , Asn 167 , and Gln 190 . The ribose group is also stabilized via hydrogen bonds with the main chain of the Tyr 24 residue and the side chain of the Asp 50 residue. The pyrophosphate portion is involved in hydrogen bonding with the side chain of Ser 217 and the main chains of Leu 219 , Ser 221 , and Lys 270 . At the periphery of the barrel, the adenosine 2Ј,5Јdiphosphate portion of NADP is stabilized by hydrogen bonds and a salt bridge. Hydrogen bonds occur with side chains of Gln 279 , Ser 271 , and Arg 276 and the main chain of Tyr 272 . At the same time, Arg 276 forms a salt bridge with the 2Ј-phosphate (Fig. 2).
Although several attempts to crystallize the apoenzyme form of 3␣-HSD3 were performed, none of them were successful. In fact, whenever crystals of the enzyme were grown in the absence of any ligand, the electron density obtained from the diffraction of these crystals showed the presence of the NADP in the cofactor binding site. This observation suggests that the cofactor bound to the enzyme comes from the cell itself during the overproduction and that its affinity for the enzyme is very high. However, because the structure of the apoenzyme has been solved for rats, it is possible that its cofactor affinity is lower than that of humans. Two substitutions in the cofactor binding site of the human enzyme can be at the origin of this difference: Asn 280 (human) instead of Leu 280 (rat) and Lys 270 (human) instead of Arg 270 (rat). These two residues are local-ized in the periphery of the NADP binding site. Whereas Leu 280 (rat) does not form any interaction with the cofactor, the human homologue Asn 280 is able to form two hydrogen bonds with the adenine part of NADP. Furthermore, Lys 270 (human) forms a salt bridge with the 2Ј-phosphate of NADP, an interaction that is absent in the rat enzyme because the corresponding amino acid Arg 270 is positioned farther apart (Fig. 2). Although all of the other interactions existing between NADP and the tunnel are similar in the rat and human enzyme, these three extra interactions support the higher affinity of NADP in the human enzyme.
It was previously established that the cofactor binding or release is rate-limiting in 3␣-HSDs (38), which could be a function of making or breaking the many interactions between the enzyme and the cofactor. Our results show that there are three additional hydrogen bonds in the human structure on the adenine, this latter being exposed to the solvent. These extra hydrogen bonds seem to be important for the stabilization of the cofactor and could contribute to the lower k cat values obtained for the human enzyme 3␣-HSD3 (AKR1C2) (k cat for the oxidation of androsterone ϭ 0.42 min Ϫ1 , and k cat for the reduction of 4-dione ϭ 1.39 min Ϫ1 (14)) than those obtained for the rat (k cat for the oxidation of androsterone ϭ 66 min Ϫ1 , and k cat for the reduction of 4-dione ϭ 19 min Ϫ1 (39)). This hypothesis is corroborated by the fact that for the other members of the human AKR family, namely 20␣-HSD (AKR1C1), 3␣-HSD2 (AKR1C3), and 3␣-HSD1 (AKR1C4), residues 270 and 280 are also substituted by Lys and Asn, respectively, and the k cat values are lower than the rat (k cat for the oxidation of androsterone ϭ 0.060 min Ϫ1 and 1.39 min Ϫ1 for human AKR1C1 and AKR1C4, respectively, and k cat for the reduction of 4-dione ϭ 3␣-HSD3, a Member of the AKR Family 0.47 min Ϫ1 , 0.28 min Ϫ1 , and 1.79 min Ϫ1 for human AKR1C1, AKR1C3, and AKR1C4, respectively (14)).
Steroid Binding Pocket-The excellent quality of the electron density map calculated at nearly atomic resolution enabled the unambiguous localization of the steroid. Surprisingly, although 4-dione was introduced before crystallization, the electron density observed in the active site of 3␣-HSD3 corresponds to testosterone (Fig. 3). Therefore, a reduction occurred before or during the crystallization process. Further kinetic investigation revealed that the overnight incubation of our enzyme preparation in the presence of 14 C-labeled 4-dione but without the addition of cofactor produced 14 C-testosterone, while the control experiment, namely the overnight incubation of 14 C-4dione in the absence of the enzyme, did not produce any 14 Ctestosterone (Fig. 4). Furthermore, and as explained previously, whenever crystals of the purified enzyme were grown in the absence of steroid and cofactor, a clear electron density for NADP was always observed in the cofactor's binding site (data not shown). This evidence supports on one hand that 3␣-HSD3 is expressed and purified with NADP(H), this latter being tightly and stably bound to 3␣-HSD3, and on the other hand that this cofactor remains in its reduced state under the experimental conditions. To date, it still remains unclear whether NADP(H) is present in its reduced state from the beginning of the expression process in Escherichia coli, this latter being stabilized in the binding tunnel (see below), or if the cofactor was reduced later (disordered in the rat structure). No detectable increase of the absorbance at 340 nm could be observed from a solution containing NADP, buffer, dithiothreitol in the presence or absence of the enzyme, this suggesting that NADP is not reduced due to small molecule chemicals present in the experimental environment (data not shown).
Interestingly, testosterone is positioned with its O3 pointing toward the nicotinamide ring, a position that is not in the expected binding mode resulting from the simple reduction of 4-dione. Indeed, because the observed density corresponds to testosterone, this means that the C17 ketone has been reduced. This observation suggests that 4-dione should enter into the active site with its C17 ketone pointing toward the nicotinamide ring of the cofactor, permitting the hydride transfer and therefore the formation of testosterone. The position of testosterone within the binding cavity obtained from the crystallographic structure, with its O3 and not O17 pointing toward the NADP, suggests that the testosterone formed from the reduction of 4-dione leaves the catalytic site and reenters in the binding site with its O3 forward.
As observed in the rat ternary complex, the testosterone molecule is bound in a cylindrical cavity formed by five loops at the C-terminal end of the barrel, namely loop A (residues 117-143), loop B (residues 217-238), the C-terminal tail (residues 299 -323), and the shorter connections ␤1-␣1 (residues 23-32) and ␤2-␣2 (residues 51-57). The A and B rings of the testosterone are in stacking with Trp 227 (Fig. 5) and interact with Tyr 24 , Trp 86 , and Ile 129 , while the ␤-face interacts mainly with Val 54 . If the A and B rings interact extensively with the human 3␣-HSD3, the C and D rings make only a few van der Waals contacts, principally with Val 128 and Ile 129 . In such binding, the C18 and C19 angular methyl groups are directed toward the side of the cavity containing residue Val 54 , whereas in the rat structure the C18 and C19 angular methyl groups were directed toward the side of the cavity containing Trp 227 . FIG. 5. Superposition of testosterone binding in the human and rat 3␣-HSDs. Stereoview of the active site residues of the human 3␣-HSD3 (yellow) and rat 3␣-HSD (blue). The testosterone molecules in the human and rat complexes are colored in purple and green, respectively. The enzyme residue numbers (numbered according to the human 3␣-HSD3) are indicated. This image was produced using MolScript (43) and Raster3D (44). Comparison of the binding of testosterone in the human and the rat structures reveals that the positions and orientations of testosterone in the binding pocket are different in 3␣-HSD3 from the two sources (Fig. 5). Whereas the testosterone in the human enzyme is bound at the entrance of the active site, the steroid is deeply engulfed in the catalytic site of the rat structure. Although different residues implicated in the binding of the testosterone molecule in the rat structure are substituted in the human sequence (Tyr 24 , Val 54 , Ile 129 , Pro 226 , and Ile 310 for the human enzyme instead of Thr 24 , Leu 54 , Phe 129 , Thr 226 , and Tyr 310 for rat enzyme), they cannot by themselves explain this difference in the steroid binding mode.
Close inspection of the binding site of testosterone in both enzymes shows that a large shift is observed for the important Trp 227 residue. Indeed, although Trp 227 is conserved in both species, this amino acid adopts a very different position in the two models; in the human enzyme, this amino acid is oriented at 134 degrees (C ␣ to C ␤ ) from the position observed in the rat enzyme (Fig. 5). The change in the position of the Trp 227 side chain is caused by the substitution of Ser 222 (rat) by His 222 (human). His 222 belongs to the loop B that is implicated in the binding of the cofactor and the steroid. In the human enzyme, the side chain of His 222 penetrates deeply to the enzyme's active site and compels the Trp 227 side chain to generate a 134°F IG. 7. Superposition of 12 Monte Carlo-minimized structures of the ternary complex involving the double-shell model 3␣-HSD3, NADPH, and testosterone. Thick sticks show the steroid and cofactor in positions 0 and 3 of testosterone that correspond to the minima at the MCM energy profile (Fig. 6). Fragments of amino acids within 6 Å from testosterone are shown. These residues are labeled if their C ␤ atom occurs within 6 Å from the steroid. A, view along the pivot drawn via atoms C3 and C17 of the steroid. B, view at the ␤-face of the steroid. rotation of the latter. In fact, this orientation of Trp 227 disallows the binding of the testosterone molecule in the same position, because otherwise the C18-methyl group would be at less than 1 Å from Trp 227 . As a result, testosterone cannot occupy the same position in both mammalian enzymes.
In such binding, the distance between the C3 atom of the steroid and the C4 of the nicotinamide ring is about 6 Å, prohibiting the hydride transfer. The crystallographic structure reveals the presence of an acetate molecule that sits between the nicotinamide and the testosterone, being 3.3 Å apart from the nicotinamide ring on one side and 4.3 Å apart from the C3 testosterone on the other side (Fig. 1). The acetate molecule occupies exactly the same position on the C2-C3-C4-O4 from the D-ring of testosterone as positioned in the rat structure. This acetate molecule is very stable in the active site, with B-values of 29.2 Å 2 for molecule A and 26.8 Å 2 for molecule B, which are equivalent to the B-values of testosterone (24.8 Å 2 for molecule A and 34.1 Å 2 for molecule B). The ketone group of the acetate molecule mimics the O3 ketone of the steroid by making hydrogen bonds with His 117 and Tyr 55 . A catalytic mechanism has been proposed for 3␣-HSDs that involves three residues near the nicotinamide ring. In this mechanism, Tyr 55 acts as the general acid, Lys 84 decreases the pK a of the tyrosine by hydrogen bonding to it, and Asp 50 is salt-bridged to the lysine (40,41). His 117 has also been implicated in catalysis in aldo-keto reductase on the basis of its position and pK a , which is near a pH optimum of the reaction (42). Taken together, these findings suggest that acetate, originating from the crystallization and present in large excess as compared with the steroid, can occupy the catalytic binding site, whereas testosterone binds farther apart, at the entry of the active site.
Using the methodology described under "Experimental Procedures," we have built a Monte Carlo-minimized energy plot of testosterone pulled along the specified direction in the active site of 3␣-HSD3. Fig. 6 shows the total energy of the ternary complex, the energy of interaction of the substrate with the rest of the system, and the distance between atom C3 of the substrate and the catalytically active hydrogen atoms of the cofactor. In the position Ϫ5, testosterone is farther from the cofactor as compared with the x-ray structure. Position 0 corresponds to the x-ray structure, and positions 1-6 correspond to testosterone approaching the cofactor. Both the total energy (Fig. 6A) and energy of interaction of the steroid with the rest of the system (Fig. 6B) have two minima. The minimum corresponding to the x-ray structure (position 0) is slightly deeper and wider than the second minimum at position 3, where testosterone approaches the cofactor. Fig. 7 shows 12 superimposed MCM structures of the ternary complex; steroid and cofactor in the minimum energy positions 0 and 3 are shown by thick sticks. The results of this modeling study suggest that testosterone can bind in the active site in two positions: catalytically active position 3, where the hydride transfer can occur, and position 0 corresponding to the present crystal structure, where testosterone is farther from the cofactor. The noncatalytic position 0 may represent a higher affinity binding site, which competes for testosterone with the catalytically active binding site (position 3). The competition between the two positions may result in a slower hydride transfer and, as a result, in a lower k cat value if only one site had existed. The physiological role of the predicted higher affinity position 0 is unknown at present. One possibility is that in this position the steroid may be preoriented for subsequent catalysis at position 3.
Comparison of 3␣-HSD and 17␤-HSD Activities-The 3␣-HSD activity toward DHT of the enzyme was compared with its 17␤-HSD activity. Similar K m values were obtained for all androgens within the experimental error, namely the K m for the 3␣-reduction of DHT ϭ 1.1 Ϯ 0.2 M, K m for the 17␤oxidation of testosterone ϭ 0.67 Ϯ 0.18 M, and K m for the 17␤-reduction of 4-dione ϭ 1.38 Ϯ 0.31 M. Nevertheless, very different catalytic constants were obtained: k cat for the 3␣reduction of DHT ϭ 1.5 min Ϫ1 , k cat for the 17␤-oxidation of testosterone ϭ 0.0071 min Ϫ1 , and k cat for the 17␤-reduction of 4-dione ϭ 0.067 min Ϫ1 . As can be seen, the observed differences in specific activities (k cat /K m for the 3␣-reduction of DHT ϭ 1.3 M Ϫ1 ⅐min Ϫ1 , k cat /K m for the 17␤-reduction of 4-dione ϭ 0.049 M Ϫ1 ⅐min Ϫ1 , k cat /K m for the 17␤-oxidation of testosterone ϭ 0.011 M Ϫ1 ⅐min Ϫ1 ) are a repercussion of the differences between catalytic constants. It should be noted that the 3␣reduction of 4-dione was not observed kinetically. From these data, we know that the 3␣-reduction of steroids is more favored than the 17␤-oxidation and 17-reduction reactions, and second, among the 17␤ reactions, the reduction of 4-dione into testosterone is favored, being 5-fold more efficient than the 17␤oxidation of testosterone. This is in good agreement with the observed reductive nature of the enzyme.
When we compare the results we obtained for human 3␣-HSD3 (3␣-reduction of DHT K m ϭ 1.1 M, k cat ϭ 1.5 min Ϫ1 , k cat /K m ϭ 1.3 M Ϫ1 min Ϫ1 ) with formerly published results (e.g. 3␣-reduction of DHT K m ϭ 26 M, k cat ϭ 0.23 min Ϫ1 , k cat /K m ϭ 0.0088 M Ϫ1 min Ϫ1 from Penning et al. (14)), our enzyme preparation shows a higher k cat and a lower K m ; thus, the specificity is nearly 150-fold higher than formerly reported results, in agreement with the high resolution crystals obtained from the same preparation.