Crystal Structure of Human Liver Δ4-3-Ketosteroid 5β-Reductase (AKR1D1) and Implications for Substrate Binding and Catalysis*♦

AKR1D1 (steroid 5β-reductase) reduces all Δ4-3-ketosteroids to form 5β-dihydrosteroids, a first step in the clearance of steroid hormones and an essential step in the synthesis of all bile acids. The reduction of the carbon-carbon double bond in an α,β-unsaturated ketone by 5β-reductase is a unique reaction in steroid enzymology because hydride transfer from NADPH to the β-face of a Δ4-3-ketosteroid yields a cis-A/B-ring configuration with an ∼90° bend in steroid structure. Here, we report the first x-ray crystal structure of a mammalian steroid hormone carbon-carbon double bond reductase, human Δ4-3-ketosteroid 5β-reductase (AKR1D1), and its complexes with intact substrates. We have determined the structures of AKR1D1 complexes with NADP+ at 1.79- and 1.35-Å resolution (HEPES bound in the active site), NADP+ and cortisone at 1.90-Å resolution, NADP+ and progesterone at 2.03-Å resolution, and NADP+ and testosterone at 1.62-Å resolution. Complexes with cortisone and progesterone reveal productive substrate binding orientations based on the proximity of each steroid carbon-carbon double bond to the re-face of the nicotinamide ring of NADP+. This orientation would permit 4-pro-(R)-hydride transfer from NADPH. Each steroid carbonyl accepts hydrogen bonds from catalytic residues Tyr58 and Glu120. The Y58F and E120A mutants are devoid of activity, supporting a role for this dyad in the catalytic mechanism. Intriguingly, testosterone binds nonproductively, thereby rationalizing the substrate inhibition observed with this particular steroid. The locations of disease-linked mutations thought to be responsible for bile acid deficiency are also revealed.

The ⌬ 4 -3-ketosteroid functionality is present in all steroid hormones except estrogens. The first step in their metabolism involves the reduction of the ⌬ 4 -ene to produce either 5␣-or 5␤-dihydrosteroids in reactions catalyzed by steroid 5␣-or 5␤-reductase, respectively (1). These two enzymes belong to distinct gene superfamilies (2,3), and structural information on these enzymes is currently lacking. ⌬ 4 -3-Ketosteroid 5␤-reductase is a soluble monomeric NADPH-dependent enzyme and a member of the aldo-keto reductase (AKR) 3 superfamily and is designated AKR1D1 in humans (3,4). In utilizing NADPH as hydride donor, the enzymatic reaction introduces a 90°bend at the steroid A/B-ring junction, which adopts a cis-A/B-or ␤-configuration ( Fig. 1) (5)(6)(7). In the absence of enzyme catalysis, this reaction is extremely difficult to perform with chemical reductants. Treatment with borohydride favors formation of the 3␤-allylic alcohol, and under harsher conditions, the 3␤,5␣-tetrahydrosteroid is formed (8,9). It is possible to generate a 5␤-cholestane from the corresponding ⌬ 4 -3-ketosteroid only if a large directing group is placed at C-4 or C-7 to direct the face of hydride addition or catalytic hydrogenation (10,11). Notably, AKR1D1 exclusively generates the reduced steroid bearing a thermodynamically less favorable 5␤-configuration. Moreover, the enzyme reduces only the ⌬ 4 -ene, leaving the 3-oxo group intact. Thus, AKR1D1 catalyzes a reaction that is unique in steroid enzymology, and it accomplishes this with ease in comparison with chemical methods.
Here, we report the first x-ray crystal structure of a mammalian steroid hormone carbon-carbon double bond reductase, human ⌬ 4 -3-ketosteroid 5␤-reductase (AKR1D1). We have determined the structures of AKR1D1 complexed with NADP ϩ at 1.79-Å resolution, NADP ϩ and cortisone at 1.90-Å resolution, NADP ϩ and progesterone at 2.03-Å resolution, and NADP ϩ and testosterone at 1.62-Å resolution. Additionally, we report the crystal structures of the binary complex with NADP ϩ at 1.35-Å resolution (containing a HEPES buffer molecule bound in the active site). These structures provide valuable inferences regarding the catalytic mechanism and also illuminate the possible effects of single point mutations responsible for bile acid deficiency. Construction of Expression Vectors-Previously, we reported the purification and characterization of recombinant AKR1D1 expressed from a pET16b vector (13). This enzyme was refractory to crystallization, and therefore, a different expression vector was generated. The pET16b-AKR1D1 expression vector was digested with NdeI and BamHI to remove the cDNA for AKR1D1. The resultant fragment was subcloned into the expression vector pET28a. The AKR1D1 insert was verified by dideoxy sequencing.

EXPERIMENTAL PROCEDURES
The expression vectors for AKR1D1(Y58F) and AKR1D1 (E120A) were made using the pET16b-AKR1D1 construct as a template by conducting site-directed mutagenesis using the QuikChange method following the manufacturer's protocol. The following forward and reverse primers were used (where the boldface nucleotides indicate the mutation introduced): for the Y58F mutant, 5Ј-dGGG GCC TAC ATC TTC CAA AAT GAA CAC GAA GTT GG-3Ј and 5Ј-dCC AAC TTC GTG TTC ATT TTG GAA GAT GTA GGC CCC-3Ј; and for the E120A mutant, 5Ј-dG GAT CTT TAC ATC ATT GCA GTA CCA ATG GCC TTT AAG C-3Ј and 5Ј-dG CTT AAA GGC CAT TGG TAC TGC AAT GAT GTA AAG ATC C-3Ј. The intro- duction of these mutations into the pET16b-AKR1D1 construct was verified by dideoxy sequencing.
Expression and Purification of AKR1D1-The pET28a-AKR1D1 expression vector was used to transform competent E. coli C41(DE3) cells. The cells were grown in 4-liter cultures of Luria-Bertani medium at 37°C (containing 100 g/ml ampicillin). Upon reaching A 600 ϭ 0.6, 1 mM isopropyl 1-thio-␤-Dgalactopyranoside was added to induce enzyme expression overnight. The following day, the culture was centrifuged for 15 min at 10,000 ϫ g, and the pellets were resuspended in 20 mM Tris-HCl (pH 7.9) and 5 mM imidazole. Resuspensions were lysed by sonication and centrifuged for 15 min at 10,000 ϫ g, and the supernatant was dialyzed overnight in 20 mM Tris-HCl (pH 7.9), 20 mM imidazole, and 0.5 M NaCl. The dialyzed fraction was loaded onto a nickel-Sepharose column equilibrated with the dialysis buffer, and the column was washed with the same buffer. Bound protein was eluted with a linear gradient of 20 -400 mM imidazole. Active fractions containing AKR1D1 were identified by monitoring the conversion of [4-14 C]testosterone to 5␤-dihydrotestosterone by discontinuous assay using standard assay conditions and by visualization of the protein content of each fraction by SDS-PAGE. Peak fractions were pooled and dialyzed overnight in 20 mM Tris-HCl (pH 7.0) containing 1 mM EDTA. The final specific activity of the recombinant enzyme was 80 nmol of testosterone reduced per min/mg. The mutant proteins AKR1D1(Y58F) and AKR1D1(E120A) were purified in a similar manner.
Standard Radiometric Assay-The reduction of [4-14 C]testosterone was used to monitor 5␤-reductase activity collected during purification and for activity measurements. Reactions contained 2 M [4-14 C]testosterone (40,000 dpm), 8 M unlabeled testosterone, 5% acetonitrile, and 100 mM potassium phosphate buffer (pH 6.0). Reactions were initiated by the addition of 200 M NADPH and performed at 37°C. The substrate and product of the quenched reaction were separated by TLC and quantitated by scintillation counting.
The structure of the AKR1D1⅐NADP ϩ ⅐testosterone complex was solved by molecular replacement using one monomer of the AKR1C2⅐NADP ϩ ⅐testosterone complex (Protein Data Bank code 1J96) (27) less the atoms of NADP ϩ ⅐testosterone and solvent as an initial search probe. The program PHASER (37) was used to perform the molecular replacement calculations; the optimal solution for the positioning of the two monomers in the asymmetric unit yielded a total log likelihood gain of 1832, a rotation function Z score of 12.3, and a translational function Z score of 19.5 for the first monomer and a rotation function Z score of 13.2 and a translational function Z score of 42.5 for the second monomer using data in the 50-2.5-Å resolution range. The analysis of this solution showed reasonable packing interactions in the unit cell. The initial electron density map clearly revealed the presence of NADP ϩ bound in the active site. Using the programs O (38) and CNS (39) for model fitting and refinement, respectively, the model was manually rebuilt and refined. In the later stages of refinement with CNS, the non-crystallographic symmetry restraints were released, after which the program SHELX was used to complete the refinement (40). Additional solvent molecules were introduced during refinement with SHELX, and the electron density map of testosterone became very clearly defined. In the final stage of refinement, the atomic coordinates of testosterone were retrieved from entry TESTOM in the Cambridge Structural Database (41), built into the electron density map, and refined with full occupancy. The quality of the final model was checked with Verify3d (42); the analysis of the structure was performed with PISA (43).
The crystal structures of the AKR1D1⅐NADP ϩ ⅐cortisone, AKR1D1⅐NADP ϩ ⅐progesterone, AKR1D1⅐NADP ϩ ⅐HEPES, and AKR1D1⅐NADP ϩ complexes were solved by the difference Fourier method. Refinement was performed as described above using CNS (39).
Twenty-one residues at the N terminus (the uncleaved expression tag from the pET28a vector) were disordered and therefore excluded from the final model of each structure. The exclusion of 21 of 346 residues in each monomer (6.1% of the scattering matter) likely contributed to the slightly elevated R and R free values recorded in Table 1.

RESULTS
Crystal Structure of the AKR1D1⅐NADP ϩ Complex-The asymmetric unit of the unit cell contains two monomers of AKR1D1. Each monomer consists of a 325-residue polypeptide chain that adopts an (␣/␤) 8 -barrel fold typical of AKRs. The top of the ␤-barrel is capped by loops A (Ile 119 -Leu 147 ), B (Tyr 219 -Leu 238 ), and C (Leu 302 -Tyr 326 ), which enclose the active site (Fig. 2a). Notably, the Pro 133 -Lys 139 segment in loop A of monomer A exhibits higher thermal B-factors compared with the Pro 133 -Lys 139 segment in monomer B, where this segment appears to be stabilized in part by interlattice contacts. However, the loop A conformations in monomers A and B are essentially identical (data not shown).
Given the amino acid sequence identities of 56% between AKR1D1 and AKR1C2 (27) and 58% between AKR1D1 and AKR1C9 (33), the overall tertiary structures of these enzymes are quite similar. The root mean square deviation of 307 C-␣ atoms between AKR1D1 in complex with NADP ϩ and between AKR1C2 in complex with NADP ϩ and testosterone (Protein Data Bank Code 1J96) (27) is 1.0 Å, and that of 317 C-␣ atoms between AKR1D1 and AKR1C9 complexed with NADP ϩ and testosterone (Protein Data Bank code 1AFS) is 0.78 Å (33). However, significant conformational differences in loops A, B, and C are observed when the structures of AKR1D1, AKR1C2, and AKR1C9 are compared (Fig.  2b). The majority of amino acid substitutions between AKR1D1 and these enzymes are found in these loops. The Val 309 -Phe 322 segment of loop C and its corresponding segment in AKR1C2 exhibit significant C-␣ deviations of ϳ4 Å. Because these loops play an important role in substrate binding, it is likely that their sequence and structural differences reflect differences in steroid substrate specificity and catalysis among these enzymes.
The NADP ϩ cofactor in the AKR1D1⅐NADP ϩ complex adopts an extended anti-conformation as observed in other   AKR structures and is located in a long tunnel with the adenine group exposed to solvent and the nicotinamide ring located deep inside the tunnel oriented toward the central active site cavity (Fig. 2a). Approximately 84% of the accessible surface area of NADP ϩ is buried inside the tunnel. The NADP ϩ cofactor participates in a significant number of hydrogen bond interactions (Fig. 3). Notably, the cofactor binding mode is essentially conserved with respect to that observed in AKR1C9 (32) or AKR1C2 (27). In complex with AKR1D1, the nicotinamide headgroup of NADP ϩ is stacked against Tyr 219 , and the carboxamide group makes contacts with Asn 170 , Ser 169 , and Gln 193 . Additionally, the phosphate group of 2Ј-AMP makes an electrostatic link with Arg 279 . However, these structures do not reveal a hydrogen bond interaction between Asn 227 and Lys 30 and Lys 273 that would correspond to the salt link interaction between Asp 217 and Lys 22 and Lys 263 observed in the AKR1A1 structure (23) or the salt link interactions between Asp 216 and Lys 21 and Lys 262 observed in the AKR1B1 structure (24). These salt links create a "safety belt" that contributes to the high affinity (ϳ10 nM) of cofactor binding to AKR1A1 and AKR1B1. Because NADP ϩ binding to AKR1D1 lacks this safety belt, binding affinity should be weaker and more comparable with cofactor binding to AKRs similarly lacking a safety belt (100 -200 nM), and this is what has been observed (13).
Interestingly, when crystals of the AKR1D1⅐NADP ϩ complex were grown in the presence of 5␤-cholestan-3-one, the binding of this product-like steroid containing a cis-A/B-ring fusion was not observed. Instead, either a molecule of glycerol from the cryoprotectant solution bound in the active site of monomer A (1.79-Å resolution structure) (data not shown), or a HEPES molecule from the crystallization buffer bound in the active site of monomer B (1.35-Å resolution structure) (Fig. 3). Thus, these two structures represent those of the binary AKR1D1⅐NADP ϩ complex. Crystal Structures of the AKR1D1⅐NADP ϩ ⅐Cortisone and AKR1D1⅐NADP ϩ ⅐Progesterone Complexes-In these enzyme⅐ substrate complexes, the electron density of each steroid substrate is clear and unambiguous (Fig. 4, a and b), showing   2.2; red). The electron-rich sulfur atom of HEPES is indicated by its stronger electron density contoured at 6.7 (green). b, hydrogen bond interactions in the AKR1D1⅐NADP ϩ ⅐HEPES complex are indicated by red dashed lines. Atoms are color-coded as described for Fig. 2a except that protein carbon atoms are yellow, HEPES carbon atoms are black, and its sulfur atom is green. that each steroid lies perpendicular to the NADP ϩ cofactor. The binding of cortisone is very similar to the binding of progesterone even though these steroids bear different pendant groups on their D-rings.
Overall, the crystal structure of each ternary complex is very similar to that of the AKR1D1⅐NADP ϩ complex, with root mean square deviations of 0.26 and 0.13 Å, respectively, for 325 C-␣ atoms. However, important conformational differences are evident in loops A and B. Significantly, cortisone and progesterone binding triggers the movement of Ser 225 -Val 231 in loop B away from the active site, with C-␣ deviations ranging from 0.3 to 2 Å. Furthermore, an ϳ7-Å movement of the side chain of Trp 230 accommodates substrate binding (illustrated for cortisone binding in Fig. 5). Thus, Trp 230 appears to play a key role in packing against the ␤-face of the steroid ring system in the active site of human AKR1D1, as also observed for the corresponding tryptophan residue in the structure of the AKR1C9⅐ NADP ϩ ⅐testosterone complex (33).
In the AKR1D1 active site, ϳ80% of the accessible surface area of each steroid substrate is buried in the hydrophobic binding pocket. Each steroid substrate is positioned with its ␤-face oriented toward the re-face of the nicotinamide ring of NADP ϩ such that the steroid carbon-carbon double bond would be adjacent to the 4-pro-(R)-hydrogen of NADPH (Fig. 4, a and b). The C-3 carbonyl oxygen of each steroid substrate accepts hydrogen bonds from the phenolic hydroxyl group of Tyr 58 and the anti-oriented conformer of the carboxylic acid side chain of Glu 120 . Notably, the C-3 carbonyl oxygen of each substrate occupies the same position as the water molecule that hydrogen bonds with Tyr 58 and Glu 120 in the AKR1D1⅐NADP ϩ complex (Figs. 2  and 3). Site-directed mutagenesis of these residues to Y58F and E120A supports their role in catalysis Atoms are color-coded as described for Fig. 2a except that protein carbon atoms are yellow. For clarity, the orientation of the protein is rotated ϳ180°horizontally relative to the orientation shown in Fig. 2a. a, difference electron density map of cortisone contoured at 2.7. b, difference electron density map of progesterone contoured at 2.6. c, difference electron density map of testosterone contoured at 2.3. Note the dramatically different, nonproductive binding mode of testosterone; in contrast with the binding of cortisone and progesterone, the A-ring of testosterone is oriented away from the catalytic tetrad and the nicotinamide ring of the cofactor.
because neither mutant has any detectable activity in the reduction of testosterone using the standard radiometric assay, where the limit of detection is 0.002 nmol/min.
Crystal Structure of the AKR1D1⅐NADP ϩ ⅐Testosterone Complex-In this enzyme⅐substrate complex, the electron density of testosterone is extremely well defined. However, the arrangement of NADP ϩ and testosterone differs from that seen in the other ternary complexes because they are parallel and not perpendicular. Testosterone also binds with a "backward" orientation in which its D-ring binds in the same region occupied by the A-rings of cortisone and progesterone in their respective complexes with AKR1D1 (Fig. 4c). Consequently, the carboncarbon double bond of testosterone is distant from the nicotinamide ring of the cofactor, and this binding orientation is judged to be nonproductive. The side chain of Trp 230 does not undergo a conformational change to accommodate testoster-one binding and instead remains in the conformation observed in the structure of the substrate-free AKR1D1⅐NADP ϩ complex (Fig. 5). Approximately 85% of the accessible surface area of testosterone is buried in the hydrophobic binding pocket. The testosterone C-3 carbonyl group accepts hydrogen bonds from Asn 227 and Ser 225 , whereas the testosterone hydroxyl group makes a hydrogen bond interaction with Tyr 132 .
The nonproductive testosterone binding modes observed with AKR1C2 and AKR1D1 suggest that this particular steroid can occupy more than one position and that at the high steroid concentrations used in the crystallization trials, the nonproductive binding mode is favored. This is reflected in substrate inhibition observed with AKR1C2 (44,45). We also observed substrate inhibition of AKR1D1 by high concentrations of testosterone. Although the K m value for this steroid is close to 2.0 M, concentrations of testosterone that exceed 10 M produced marked inhibition (Fig. 6).
Disease-linked Mutations-Four point mutations in AKR1D1 are associated with bile acid deficiency: L106F, P198L, P133R, and R261C (20 -22). As predicted from other AKR structures, these residues are not located in the cofactor-or steroid-binding sites and are not involved in catalysis. The locations of these residues in the AKR1D1 structure are shown in Fig. 7. The P133R substitution in loop A is closest to the substrate-binding site, and it is conceivable that this substitution perturbs the conformational changes in loop A that accompany steroid binding. The remaining mutations may exert a deleterious effect on catalysis by causing structural changes that propagate through the protein scaffold to perturb substrate and/or cofactor binding or by otherwise destabilizing the folded conformation of the protein.

DISCUSSION
The x-ray crystal structure of human ⌬ 4 -3-ketosteroid 5␤-reductase (AKR1D1) is the first structure of a mammalian steroid carbon-carbon double bond reductase. As a member of the AKR superfamily, it is not surprising that the (␣/␤) 8 -barrel fold and the NADP ϩ cofactor-binding site of AKR1D1 are highly conserved with those other members of the superfamily. However, the structures of AKR1D1⅐substrate complexes reported herein illuminate new features of steroid substrate recognition, the catalytic mechanism, and the positions of natural mutations associated with bile acid deficiency.
Substrate Recognition-AKR1D1 binds ⌬ 4 -3-ketosteroids productively and nonproductively. In the productive binding mode observed with cortisone and progesterone, the two substrates are bound similarly in a pocket defined by residues from the ␤2-␣2 loop, loop A, loop B, and loop C. Alignments of pocket and loop residues of AKR1D1 with those of other AKR1C structures containing bound steroids are shown in sup- FIGURE 5. Least-squares superposition of the AKR1D1⅐NADP ؉ (yellow), AKR1D1⅐NADP ؉ ⅐cortisone (red), and AKR1D1⅐NADP ؉ ⅐testosterone (green) complexes. The distance of ϳ3.7 Å between the anomeric carbon of NADP ϩ and the olefinic C-4 of cortisone is indicated by a red dashed line, which would represent the trajectory of hydride transfer from NADPH. The position of the water molecule in the unliganded structure is indicated by a yellow sphere. Loops A, B, and C are labeled; note that loop B and Trp 230 in particular must undergo a significant conformational change to accommodate productive substrate binding as represented by cortisone (red). The indole ring of Trp 230 remains in its substrate-free conformation (yellow) to accommodate the nonproductive binding mode of testosterone (green). plemental Tables S1 and S2. Of the 10 pocket residues assigned, Tyr 58 is catalytic, and Tyr 26 and Trp 230 are highly conserved. The indole ring of Trp 230 in the AKR1D1 structures is versatile in that rotation around both side chain torsion angles and movement of the associated polypeptide backbone of loop B allow the indole ring to pack differently, i.e. against the ␤-face of a productively bound steroid substrate or against the ␣-face of a nonproductively bound steroid substrate (Fig. 5).
Interestingly, superposition of the AKR1D1⅐NADP ϩ ⅐ cortisone and AKR1C9⅐NADP ϩ ⅐testosterone complexes, in which both substrates adopt productive binding orientations, reveals that the substrate is immobilized more deeply in the active site of AKR1D1 compared with AKR1C9 (Fig. 8). In part, this appears to result from the removal of the steric bulk of the His 120 imidazole side chain by the substitution of Glu 120 in AKR1D1. As a result, the 4-pro-(R)-hydride of the NADPH cofactor would be adjacent to the substrate C-3 carbonyl group in the AKR1C9 active site, and the 4-pro-(R)-hydride of the NADPH cofactor would be adjacent to C-5 of the substrate carbon-carbon double bond in the AKR1D1 active site (Fig. 8).
Catalytic Mechanism-The four catalytic residues Asp 53 , Tyr 58 , Lys 87 , and Glu 120 are located in the center of the ␤-barrel (Fig. 2a). The side chain of Lys 87 donates hydrogen bonds to Asp 53 and Tyr 58 . In the substrate-free enzyme, the phenolic hydroxyl group of Tyr 58 donates a hydrogen bond to a water molecule, which in turn hydrogen bonds with Glu 120 . Because the C-3 carbonyl oxygen atoms of the substrates cortisone and progesterone displace this water molecule in their complexes with AKR1D1, the side chain of Glu 120 must be protonated as the anti-oriented carboxylic acid to donate a hydrogen bond to the substrate carbonyl oxygen (and possibly also the water molecule in the substrate-free enzyme). Furthermore, given that carboxylate groups that serve as general bases generally do so with their more basic, syn-oriented lone electron pairs (46), it is reasonable to conclude that protonated carboxylic acids that serve as general acids or hydrogen bond donors may do so with their more acidic (pK a ϳ0.5) anti-oriented conformers. Thus, in AKR1D1, the anti-oriented conformer of Glu 120 may serve as a superacidic hydrogen bond donor to help activate the ␣,␤-unsaturated ketone moiety of the substrate for carbon-carbon bond reduction by NADPH.
That Glu 120 appears in AKR1D1 represents a unique exception for AKRs having an (␣/␤) 8 -barrel fold. All other AKRs, including AKR1C9 (33) and AKR1C2 (26), have a histidine residue at the corresponding position. The side chain conformations of Asp 53 , Tyr 58 , and Lys 87 are mostly conserved with respect to the conformations of corresponding residues in the AKR1C2⅐NADP ϩ ⅐testosterone complex; the side chain of Glu 120 is coplanar with respect to the orientation of the corre-  sponding histidine residue in this complex (27). Interestingly, the hydrogen-bonded water molecule between Tyr 58 and Glu 120 occupies a position close to that of an acetate oxygen atom observed in the AKR1C2⅐NADP ϩ ⅐testosterone complex (27), and a corresponding water molecule is also observed in the crystal structure of AKR1C9 (32). It is proposed that the binding of this water molecule in AKR1C9 mimics the binding of the carbonyl oxygen of a 3-ketosteroid substrate (33).
In AKR1D1, the interactions of Tyr 58 and Glu 120 with the carbonyl groups of steroid substrates suggest that both residues participate in the catalytic mechanism of carbon-carbon double bond reduction. In AKR1B1 and AKR1C9, site-directed mutagenesis studies provide compelling evidence that the catalytic tyrosine acts as the general acid-base (47,48). Moreover, site-directed mutagenesis studies of AKR1C9 also show that the corresponding H117E substitution introduces 5␤-reductase activity into this enzyme (34). Based on k cat and k cat /K m pH-rate profiles for wild-type AKR1C9 and AKR1C9(H117E) enzymes, a facilitatory role for the glutamic acid residue is proposed in which it alters the pK of Tyr 55 , which functions as the general acid. This facilitatory role is based on an acid shift in the pK b of Tyr 55 in the H117E mutant. These data are interpreted to indicate that Tyr 55 has more TyrOH 2 ϩ character (i.e. it may behave as a "superacid"), which in turn polarizes the ␣,␤-unsaturated ketone so that hydride transfer can occur at C-5.
The carboxylic acid side chain of Glu 120 may also serve as a superacid to help polarize the C-3 ketone of a productively bound substrate such as cortisone or progesterone (Fig. 4, a and  b). The non-enzymatic mechanism of carbon-carbon double bond reduction in an ␣,␤-unsaturated ketone requires a superacid such as HF-SbF 5 (49), and it is possible that AKR1D1 adopts a similar strategy for catalysis. Generation of the AKR1D1(Y58F) and AKR1D1(E120A) mutants abolishes all enzyme activity, consistent with the crucial role of these residues in catalysis. Once 4-pro-(R)-hydride transfer from the re-face of NADPH to the carbon-carbon double bond of the substrate is achieved, the substrate will adopt a bent, cis-A/B-ring conformation. Either an enzymebound residue or a solvent molecule will protonate the C-4 atom of the substrate in the final step of the steroid reduction reaction. A proposed mechanism that incorporates these structural data in light of available enzymological measurements is presented in Fig. 9.
Disease-linked Mutations-The four point mutations associated with inherited bile acid deficiency are located in regions independent of substrate recognition and catalysis. If these mutations are responsible for bile acid deficiency, then it is likely that they affect conformational changes in loop A that accompany substrate binding (P133R) and/or otherwise compromise enzyme stability. Future studies of these mutant recombinant proteins and their expression in mammalian cells will allow us to probe the structural and functional consequences of these mutations.