Crystal Structure and Site-directed Mutagenesis of 3-Ketosteroid Δ1-Dehydrogenase from Rhodococcus erythropolis SQ1 Explain Its Catalytic Mechanism*

Background: 3-Ketosteroid Δ1-dehydrogenases catalyze the 1,2-desaturation of 3-ketosteroids. Results: First structures of the enzyme from Rhodococcus erythropolis SQ1, combined with site-directed mutagenesis, clarify its catalytic mechanism. Conclusion: Tyr487 and Gly491 promote keto-enol tautomerization, whereas Tyr318/Tyr119 and FAD abstract a proton and hydride ion, respectively. Significance: This study is an important step toward tailoring the enzyme for steroid biotransformation applications. 3-Ketosteroid Δ1-dehydrogenases are FAD-dependent enzymes that catalyze the 1,2-desaturation of 3-ketosteroid substrates to initiate degradation of the steroid nucleus. Here we report the 2.0 Å resolution crystal structure of the 56-kDa enzyme from Rhodococcus erythropolis SQ1 (Δ1-KSTD1). The enzyme contains two domains: an FAD-binding domain and a catalytic domain, between which the active site is situated as evidenced by the 2.3 Å resolution structure of Δ1-KSTD1 in complex with the reaction product 1,4-androstadiene-3,17-dione. The active site contains four key residues: Tyr119, Tyr318, Tyr487, and Gly491. Modeling of the substrate 4-androstene-3,17-dione at the position of the product revealed its interactions with these residues and the FAD. The C1 and C2 atoms of the substrate are at reaction distance to the N5 atom of the isoalloxazine ring of FAD and the hydroxyl group of Tyr318, respectively, whereas the C3 carbonyl group is at hydrogen bonding distance from the hydroxyl group of Tyr487 and the backbone amide of Gly491. Site-directed mutagenesis of the tyrosines to phenylalanines confirmed their importance for catalysis. The structural features and the kinetic properties of the mutants suggest a catalytic mechanism in which Tyr487 and Gly491 work in tandem to promote keto-enol tautomerization and increase the acidity of the C2 hydrogen atoms of the substrate. With assistance of Tyr119, the general base Tyr318 abstracts the axial β-hydrogen from C2 as a proton, whereas the FAD accepts the axial α-hydrogen from the C1 atom of the substrate as a hydride ion.

The dehydrogenation catalyzed by ⌬ 1 -KSTD proceeds via direct elimination of two hydrogen atoms from its substrate, instead of via hydroxylation followed by dehydration (1,7,8). As determined using isotope-labeled substrates, the stereochemistry of the dehydrogenation appeared to proceed preferentially via a trans-diaxial elimination of the ␣-hydrogen of the C1 atom and the ␤-hydrogen of the C2 atom of the 3-ketosteroid substrate (9 -11). Such a mechanism is also observed in the activity of acyl coenzyme A dehydrogenases (12,13). ⌬ 1 -KSTDs catalyze the exchange of the alkali-labile hydrogen at the C2 atom of their substrates under conditions that inhibit enzyme turnover (9 -11). This observation suggests that the enzymecatalyzed dehydrogenation more likely follows a stepwise unimolecular elimination conjugate base (E1cB) mechanism, in which departure of the first hydrogen atom precedes that of the second hydrogen accompanied by the formation of an interme- diate. A concerted bimolecular elimination (E2) mechanism, in which the two hydrogens depart simultaneously without the formation of any intermediate, was considered less likely. Therefore, catalysis by ⌬ 1 -KSTD was proposed to proceed via a two-step mechanism, initiated by the interaction of the C3 carbonyl group of the 3-ketosteroid substrate with an electrophile to promote labilization of the C2 hydrogen atoms. Subsequent abstraction of a proton from the C2 atom by a general base was proposed to result in either an enolate or a carbanionic intermediate, after which a double bond is formed between the C1 and C2 atoms in synchrony with hydride transfer from the C1 atom of the substrate to the flavin coenzyme (9 -11). This proposed mechanism is in contrast to the dehydrogenation catalyzed by acyl coenzyme A dehydrogenases, which proceeds via a concerted removal of the eliminated hydrogens (12,13). Based on chemical modification, mutagenesis, and kinetics experiments, several amino acid residues of the ⌬ 1 -KSTD from Rhodococcus rhodochrous (formerly Nocardia corallina) were identified as important for catalytic activity, including one or two histidines, an arginine, and Tyr 121 , and several other residues were shown to be important for binding of the steroid substrates, such as Tyr 104 and Tyr 116 (14,15). Moreover, mutagenesis studies on the ⌬ 1 -KSTD isoenzyme 2 from Rhodococcus erythropolis SQ1 (⌬ 1 -KSTD2) have shown that Ser 325 and Thr 503 are crucial for catalysis (5).
In R. erythropolis SQ1, three ⌬ 1 -KSTD isoenzymes have been found: ⌬ 1 -KSTD1 (16), ⌬ 1 -KSTD2 (5), and ⌬ 1 -KSTD3 (2). Previously, we described the purification, crystallization, and preliminary x-ray crystallographic analysis of ⌬ 1 -KSTD1 (17). Here, we report crystal structures of the enzyme in the unliganded form and in complex with the reaction product 1,4-androstadiene-3,17-dione (ADD). The structures provide insight into the active site of the enzyme and allowed us to assign its catalytic residues, which, interestingly, include three tyrosines. To further investigate their role, these tyrosines were mutated to phenylalanines, and the kinetic properties of the produced mutants were studied. These results enabled us to identify the roles of the amino acid residues involved in catalysis and to clarify the catalytic mechanism of ⌬ 1 -KSTD.
The information presented here may allow manipulating the catalytic properties of the enzyme to improve it for application in the pharmaceutical steroid industry. Furthermore, the steroid catabolic activity of steroid-degrading pathogenic bacteria such as Mycobacterium tuberculosis and Rhodococcus equi is frequently associated with their pathogenicity (18,19). In particular, cholesterol degradation appeared to be essential for the survival of M. tuberculosis in the normally adverse environment of the host macrophages (19,20). For this purpose, M. tuberculosis H37Rv contains a large gene cluster coding for enzymes catalyzing cholesterol degradation, including a gene for ⌬ 1 -KSTD (Rv3537) (19) that is important for its pathogenicity (21). Thus, these first three-dimensional structures of a ⌬ 1 -KSTD may facilitate the design of inhibitors that may be developed into efficacious drugs to combat pathogenic bacteria.
Protein Crystallization-⌬ 1 -KSTD1 was crystallized using the sitting drop, vapor diffusion method at room temperature (17). Typically, 0.30-l drops comprising 0.15 l of protein sample (15 mg ml Ϫ1 protein) mixed with 0.15 l of reservoir solution (2% (v/v) polyethylene glycol 400, 0.1 M HEPES were equilibrated against wells containing 50 l of the reservoir solution. Brightly yellow, rectangular crystals grew in 5-7 days to maximal dimensions of ϳ100 ϫ 100 ϫ 300 m 3 . Because of the low solubility of ADD in aqueous solution, protein samples were incubated with ADD powder (Steraloids Inc.) for 3 h on ice, prior to the cocrystallization experiments.
X-ray Data Collection and Processing-Before x-ray diffraction data collection, ⌬ 1 -KSTD1 crystals were transferred for 5 s to a cryoprotection solution (reservoir solution with 40% (w/v) sucrose added), followed by a 1-s transfer to a 1:1 solution of paraffin oil and paratone-N and flash cooling in liquid nitrogen (17). ⌬ 1 -KSTD1 crystals with bound ADD (⌬ 1 -KSTD1⅐ADD) were cryoprotected in a similar way, except that both above solutions were saturated overnight with ADD powder. X-ray diffraction data from both ⌬ 1 -KSTD1 and ⌬ 1 -KSTD1⅐ADD crystals were collected at 100 K on beam-line ID14-1 (European Synchrotron Radiation Facility, Grenoble, France) using an ADSC Quantum Q210 detector. Data sets comprising 450 and 550 frames with an oscillation range per frame of 0.2°were recorded from ⌬ 1 -KSTD1 and ⌬ 1 -KSTD1⅐ADD, respectively, at a wavelength of 0.93340 Å. All data sets were processed and integrated using the program XDS (22) and scaled and merged with the program SCALA (23) from the CCP4 package (24).
Structure Determination and Refinement-The initial phases of ⌬ 1 -KSTD1 were obtained using multiwavelength anomalous dispersion data collected from a Pt-derivatized crystal, and a complete model for the protein was obtained (17). This model was subjected to rigid body refinement with the program Refmac5 (25) from the CCP4 package using native diffraction data of ⌬ 1 -KSTD1. The resulting model was then refined manually using the visualization program COOT (26). Water and ligand molecules were added to the model, manually checked, and refined against sigma-A-weighted 2F o Ϫ F c maps (27). The final model of ⌬ 1 -KSTD1 was obtained after several rounds of local real space refinement in COOT and overall restrained refinement in Refmac5. The same protocol was used to prepare the final model of ⌬ 1 -KSTD1⅐ADD. The atomic coordinates and experimental structure factor amplitudes have been deposited with the RCSB Protein Data Bank with Protein Data Bank codes 4c3x and 4c3y for ⌬ 1 -KSTD1 and ⌬ 1 -KSTD1⅐ADD, respectively.
Site-directed Mutagenesis-The pET15b-kstD1 plasmid (2) was used as the template for site-directed mutagenesis to introduce Y119F, Y318F, and Y487F point mutations into ⌬ 1 -KSTD1. Forward mutagenic primer sequences, which were designed using the QuikChange primer design program, were 5Ј-gt gcg ttc ccc gac tTc tac aaa gcc gaa gg-3Ј for Y119F, 5Ј-c ctc aac gag tcg ctt ccg tTc gac cag ttc g-3Ј for Y318F, and 5Ј-tg agc ggc cgc ttc tTc ccc ggc c-3Ј for Y487F. The mutated nucleotides are shown as capital letters. Reverse mutagenic primer sequences were complementary to the forward primers and had the same lengths. The mutagenesis was performed using the QuikChange II site-directed mutagenesis kit (Stratagene) according to the instruction manual. Following mutagenesis, the plasmid template was digested using DpnI, and each of the mutants was transformed into E. coli XL1-Blue supercompetent cells (Stratagene). Individual clones of these E. coli variants were grown, and their plasmids were isolated for sequencing and transformation into the expression strain E. coli BL21(DE3). All mutant proteins were expressed and purified using the same procedures as the wild-type protein. The mutants were expressed at roughly the same level and crystallized under the same conditions as the wild-type enzyme, implying that no large scale conformational or stability changes were caused by the mutations.
Steady-state Kinetics-The steady-state kinetics parameters of ⌬ 1 -KSTD1 and its mutants were determined by measuring the dehydrogenase activity of the enzyme on the substrate 4-AD (Sigma) using dichlorophenol indophenol (Sigma) as an artificial electron acceptor. The stock solution of the substrate was prepared in ethanol. Activity assays were carried out at room temperature using 1-ml samples that had a typical composition (final concentration) of 50 mM Tris-HCl buffer, pH 8.0, 60 M dichlorophenol indophenol, 4-AD (varying concentrations), and protein (chosen concentration). For each protein variant, seven different substrate concentrations (each in triplicate) were used. The protein concentrations were chosen such that the amount of consumed substrate at the end of the measurement was less than 10%. Controls were prepared by omitting the substrate. Except for the enzyme, all sample components were preincubated for 3 min, and the reactions were initiated by adding the enzyme. The reduction of dichlorophenol indophenol was monitored by measuring the decrease of its absorbance at 605 nm using an Ultrospec 4300 Pro UV-visible spectrophotometer (Amersham Biosciences) every minute for 5 or 6 min, and the data were collected using the SWIFT II reaction kinetics software (Amersham Biosciences). The kinetics parameters were analyzed by nonlinear regression curve fitting of the data to the Michaelis-Menten equation using GraphPad Prism version 5.00 for Windows (GraphPad Software). (2). Therefore, in addition to the ⌬ 1 -KSTD1 amino acid sequence (510 residues), the expressed protein also contains a His-tagged 20-amino acid leader sequence (MGSSHHHHHHSSGLVPRGSH). The expressed protein was purified in three steps of, successively, immobilized Ni 2ϩ affinity, anion exchange, and size exclusion chromatography (17). The purified protein was colored brightly yellow and exhibited a typical absorption spectrum for a flavoprotein with maxima at 275, 377, and 464 nm (data not shown). Moreover, a Pfam search (28) using the amino acid sequence of ⌬ 1 -KSTD1 revealed that the sequence could be assigned to Pfam family PF00890, which is a protein family that binds FAD. Taken together, the expressed ⌬ 1 -KSTD1 comprises 530-amino acid Structure and Mechanism of a 3-Ketosteroid ⌬ 1 -Dehydrogenase DECEMBER (17).

Expression, Purification, and Preliminary Characterization-
Crystal Structure Determination-The ⌬ 1 -KSTD1 holoenzyme and ⌬ 1 -KSTD1⅐ADD complex crystallized under the same conditions and had the same space group P2 1 2 1 2 1 with similar unit cell parameters. The crystal structure of ⌬ 1 -KSTD1 was solved using multiwavelength anomalous dispersion data collected from a Pt-derivatized crystal (17), whereas that of ⌬ 1 -KSTD1⅐ADD was elucidated by rigid body fitting using the refined model of the holoenzyme and subsequent refinement. The structures were determined at 2.0 and 2.3 Å resolution for ⌬ 1 -KSTD1 and ⌬ 1 -KSTD1⅐ADD, respectively, with R factor and R free (for 5% of the data not included in the refinement selected in thin shells) values as expected for their resolutions. Data collection and refinement statistics are summarized in Tables 1 and 2, respectively. For each of the models reported here, there are eight ⌬ 1 -KSTD1 monomers (denoted as chains A-H in the Protein Data Bank files) in the crystallographic asymmetric unit. Except for the first 22 amino acid residues, including the 20-amino acid leader sequence and the first two residues of ⌬ 1 -KSTD1, all residues could be positioned in the electron density. However, for residues 421-423 of monomers B, C, F, and G, the electron densities are very weak, indicating conformational flexibility and/or disorder. A total of 2,970 and 1,414 water molecules could be modeled unequivocally and refined in the final model of ⌬ 1 -KSTD1 and ⌬ 1 -KSTD1⅐ADD, respectively. The electron density showed the presence of an FAD molecule in each monomer. Based on its electron density and octahedral coordination geometry, a sodium ion was identified at the interface between two monomers on a 2-fold symmetry axis. In the ⌬ 1 -KSTD1 and ⌬ 1 -KSTD1⅐ADD electron density maps, clear densities were present for a chloride ion, which most likely originated from the protein storage buffer, and an ADD ligand, respectively, in each of the eight monomers.
Oligomeric Structure-⌬ 1 -KSTD1 crystallizes with eight monomers in the asymmetric unit. These eight monomers form two independent, very similar tetramers (ABCD and EFGH), which have noncrystallographic 222 symmetry. Superposition of these tetramers gave an RMSD of 0.44 Å for 1,940 of 2,026 C␣ atoms. A calculation of the buried solvent-accessible surface areas caused by intermonomer interactions in the ABCD tetramer resulted in values of 1,622, 633, 1,161, 517, 534, and 1,662 Å 2 for the AB, AC, AD, BC, BD, and CD interactions, respectively. These values suggest that the tetramer is a dimer of dimers, i.e., AB and CD, instead of a homotetramer. A sodium ion was found in the AB dimer interface on the noncrystallographic 2-fold symmetry axis relating the A and B chains. Asp 154 , Gln 155 , and Gln 160 from each chain ligate the sodium ion. A sodium ion is also present at the corresponding position in the CD interface.
Although many hydrogen bonding interactions exist between the monomers in the dimer and tetramer, size exclusion chromatography indicated that ⌬ 1 -KSTD1 occurs as monomers in solution. Moreover, most residues involved in hydrogen bonding interactions in the dimer interface are relatively variable among ⌬ 1 -KSTDs (data not shown). Thus, the intermolecular contacts observed in the ⌬ 1 -KSTD1 crystal structure are most likely biologically insignificant.
Overall Fold and Domains-All ⌬ 1 -KSTD1 monomers exhibit a highly similar fold with an overall C␣ RMSD of 0.18 Å when all amino acid residues are structurally aligned. Each

Summary of x-ray data collection and processing
The values in parenthesis are for the highest resolution shell. I i (h) is the integrated intensity of a reflection, ϽI(h)Ͼ is the mean intensity of multiple corresponding symmetry-related reflections, and N is the multiplicity of the given reflections.  (17). a R factor ϭ ⌺ hkl ͉͉F P(obs) ͉ Ϫ ͉F P(calc) ͉͉/⌺ hkl ͉F P(obs) ͉, where ͉F P(obs) ͉ and ͉F P(calc) ͉ are observed and calculated structure factor amplitudes, respectively. b R free is R factor for 5% of the data not included in the refinement selected in thin shells. c Ramachandran statistics were obtained from COOT (26). ⌬ 1 -KSTD1 molecule has an elongated shape with dimensions of ϳ40 ϫ 45 ϫ 70 Å 3 and is organized in two domains: an FAD-binding domain and a catalytic domain (Fig. 2).

Crystal
The FAD-binding domain consists of two separate polypeptide segments (residues 3-278 and 449 -510) forming subdomains FAD-A and FAD-B. Subdomain FAD-A essentially folds into a central five-stranded parallel ␤-sheet sandwiched by a three-␣-helix bundle at one side and a twisted threestranded ␤-meander and an ␣-helix at the other side (Fig. 3a). The arrangement adopts the Rossmann fold, the well known mononucleotide-binding fold with a basic topology of a symmetrical ␣/␤ structure composed of two halves of ␤1-␣1-␤2-␣2-␤3 and ␤4-␣4-␤5-␣5-␤6 connected at the ␤3 and ␤4 strands by an ␣-helix (␣3) crossover (29,30). The first half of the Rossmann fold of ⌬ 1 -KSTD1 (␤1b-␣1-␤2-␣7-␤5) is very similar to the basic topology, whereas the second half (␤8b-␣9-␤19-␣15) is slightly different because of a missing third ␤-strand. Moreover, instead of an ␣-helix, ⌬ 1 -KSTD1 uses a three-stranded ␤-meander as a crossover to connect the two halves of its Ross-mann fold. Such a fold has been identified in the glutathione reductase structural family and can be observed in many different FAD-containing enzymes catalyzing a wide variety of reactions (29). In addition to the Rossmann fold, the FAD-A subdomain also includes several secondary structure elements inserted between ␤8b and ␣9, as well as ␤19 and ␣15. The FAD-B subdomain is inserted between ␤2 and ␣7, whereas the catalytic domain (residues 279 -448), where the catalytic residue Tyr 318 comes from, is in between ␣9 and ␤19 of the FAD-A subdomain.
A structural similarity search using the DALI server (31) indicated that ⌬ 1 -KSTD1 is related to many FAD-binding proteins, with the highest similarity to 3-ketosteroid ⌬ 4 -(5␣)-dehydrogenase (⌬ 4 -(5␣)-KSTD) from Rhodococcus jostii (Protein Data Bank entry 4at0; Z-score 39.4; RMSD 3.4 Å; 28% sequence identity) (32), followed by flavocytochrome c fumarate reductase from Shewanella putrefaciens (Protein Data Bank entry 1d4c; Z-score 37.2; RMSD 3.8 Å; 24% sequence identity) (33). In the FAD-binding domain, the first half, the second half, and the ␤-meander crossover of the Rossmann fold (subdomain FAD-A) are colored pale green, brown, and deep blue, respectively, whereas subdomain FAD-B is in deep teal. The side chains of the three tyrosine residues in the active site, as well as FAD, are shown in ball and stick representation with carbon atoms in blue. b, molecular surface drawing of ⌬ 1 -KSTD1 colored according to its hydrophobicity (minimum, tan; zero, white; maximum, hot pink). The electron density for the bound ADD ligand is represented as a green surface with its faces indicated as ␣ and ␤. FAD is depicted in ball and stick representation with carbon atoms in blue. The ⌬ 1 -KSTD1 molecule is in approximately the same orientation as in a. c, simplified secondary structure topology diagram of ⌬ 1 -KSTD1 colored in accordance with a. Cylinders and arrows symbolize helices and ␤-strands, respectively. Key residues for enzyme activity are indicated with red circles. Secondary structure elements were assigned using DSSP (37). Figs. 2a, 3a, and 4 were prepared with the program PyMOL (PyMOL Molecular Graphics System, Version 0.99, Schrö dinger), whereas Fig. 2b was made with the program Chimera (38). DECEMBER 6, 2013 • VOLUME 288 • NUMBER 49 FAD-binding Site-Most of the FAD coenzyme is buried in an elongated cavity in the FAD-binding domain. It adopts an extended conformation comparable to those in other proteins of the glutathione reductase family (29). Its adenine end is tethered in front of the parallel ␤-sheet of the Rossmann fold, and its isoalloxazine ring is at the edge of the FAD-binding domain pointing to the catalytic domain. The isoalloxazine ring has an almost planar conformation. Its si-face interacts with the ␣3-␣4 loop of the FAD-binding domain, its re-face faces the catalytic domain, whereas the O4, C4A, N5, and C5A atoms face the bulk solvent. In both structures, the FAD molecules are in the oxidized state, as suggested by the solution spectrum of the protein and the bright yellow color of the crystals. As observed in the majority of FAD-containing proteins of the glutathione reductase family (29), the FAD coenzyme in ⌬ 1 -KSTD1 is bound noncovalently. The different parts of the coenzyme bind to the apo-enzyme through a variety of interactions, including hydrogen bonds, van der Waals contacts, and dipole-dipole interactions (Fig. 3).

Structure and Mechanism of a 3-Ketosteroid ⌬ 1 -Dehydrogenase
The adenine moiety is nested in a hydrophobic pocket formed primarily by the side chains of Val 13 , Leu 36 , Lys 38 , Val 194 , Leu 195 , Ala 229 , and Ala 262 . The N1, N3, N6, and N7 atoms of this moiety are hydrogen-bonded to the protein. As also observed in other members of the glutathione reductase family (29), ⌬ 1 -KSTD1 contains a consensus dinucleotidebinding motif (29) of the XhXhGXGXXGXXXhXXhX 7 hXhE(D) type (where X is any residue, and h is a hydrophobic residue), i.e., 10 VLVVGSGGGALTGAYTAAAQGLTTIVLE 37 (the conserved residues are in bold), in which the third conserved glycine is replaced by alanine (Ala 19 ). These residues are part of the ␤1b-␣1-␤2 portion of the first half of the Rossmann fold. The conserved Glu 37 at the end of the motif (and at the C-terminal end of ␤2) has a bidentate hydrogen bonding interaction of its carboxylate group with the diol O2 and O3 atoms of the ribose. The pyrophosphate group is located in a site surrounded by amino acid residues from the ␣2 and 3 helices and the ␤8-␣8 and 2-␣9 loops, close to the ␣1 helix. The positive end of the helix dipole of this latter ␣-helix is in a good position to favor-  Fig. 2a. Various interdomain/subdomain residues are numbered and colored in red. FAD is represented as balls and sticks with blue carbon atoms; its electron density is contoured at 1.8 . b, schematic drawing emphasizing the hydrogen bonds between FAD and protein and water (circled W) molecules. The hydrogen bonds are shown as dashed lines, and their distances, which were adopted from chain A of ⌬ 1 -KSTD1 structure, are indicated in Ångströms.
ably interact with the double negative charge of the pyrophosphate. Additionally, four of the pyrophosphate oxygen atoms of the pyrophosphate are hydrogen-bonded to the apo-enzyme or water molecules. Except for the O2Ј hydroxyl group, all hydroxyls of the ribityl moiety make hydrogen bonds to the protein. Of the isoalloxazine ring, the dimethylbenzene ring is found in a hydrophobic environment between the side chains of Tyr 48 , Ser 49 , Leu 153 , Met 252 , Phe 294 , Ile 354 , and Leu 447 , whereas the N5 atom and the pyrimidine moiety interact through hydrogen bonds with the protein backbone. In addition, the hydroquinone and semiquinone forms of the pyrimidine moiety, which is situated just near the N terminus of the ␣15 helix, may be stabilized by their interaction with the positive end of the macro-dipole moment of this helix.
Active Site Structure-The electron density map of the ⌬ 1 -KSTD1⅐ADD complex revealed the presence of electron density in its active site compatible with a bound ADD molecule (Fig. 4a). The ADD is located at the interface between the FADbinding and the catalytic domains, near the FAD-binding site. It is bound in a pocket-like cavity that is lined with hydrophobic amino acid residues originating from both domains and bordered by the re-face of the isoalloxazine ring of FAD (Figs. 2b  and 4a). The apolar nature of the residues that line this pocket is conserved among ⌬ 1 -KSTD enzymes from different bacteria (data not shown). The electron density for the two axial ␤-methyl groups at the C10 and C13 atoms of ADD clearly defined the orientation of the 3-ketosteroid molecule. The A-ring of the 3-ketosteroid is nearly planar and lies almost parallel to the plane of the isoalloxazine ring. It is buried in the active site and sandwiched between the re-face of the pyrimidine moiety of the isoalloxazine ring system on its ␣-face and residues Tyr 119 and Tyr 318 on its ␤-face, whereas the five-membered D-ring of the 3-ketosteroid occupies a solvent-accessible pocket near the active site entrance.
The ADD molecule is bound in the active site via a large number of van der Waals interactions (Fig. 4a). Phe 116 is in a good position for hydrophobic stacking interactions with the B-ring of the 3-ketosteroid. The C3 carbonyl oxygen atom of the 3-ketosteroid forms hydrogen bonds with the Tyr 487 hydroxyl group and the Gly 491 backbone amide, whereas the C1 and C2 atoms are at short distances to the N5 atom of the isoalloxazine ring and the hydroxyl group of Tyr 318 , respectively. Binding of ADD does not induce significant conformational changes, as evidenced by the low RMSD of 0.27 Å for C␣ atoms between the holoenzyme and enzyme-product complex structures.
The binding mode of ADD in the active site of ⌬ 1 -KSTD1 resembles that of 4-AD in ⌬ 4 -(5␣)-KSTD, a flavoenzyme that introduces a 4,5-double bond into 3-keto-(5␣)-steroids (32). However, a structural superposition of the proteins based on . The active site of ⌬ 1 -KSTD1. a, wall-eyed stereo view of the active site of ⌬ 1 -KSTD1 with bound ADD product. ADD is depicted with green carbon atoms together with electron density contoured at 1.5 . FAD and the side chains of amino acid residues within 5.0 Å from the ADD are shown with blue carbon atoms. b, protein superposition based on the FAD molecules of ⌬ 1 -KSTD1⅐ADD and ⌬ 4 -(5␣)-KSTD⅐4-AD. ADD is represented with green carbon atoms, whereas 4-AD is in orange. The FAD molecules and the amino acid residues within 5.0 Å from ADD and 4-AD are shown in blue and gray carbon atoms, respectively. Inset, a close-up view of the A-rings of the steroids in ⌬ 1 -KSTD1 and ⌬ 4 -(5␣)-KSTD showing the superposition of the atoms between which the double bond is formed. c, 4-AD, a ⌬ 1 -KSTD1 substrate, modeled in the active site by superposition on the product. Hydrogen bonds and close contacts between substrate (green carbon atoms) and FAD and several active site residues (blue carbon atoms) are shown as dashed lines, and their distances are indicated in Ångströms. The superposition was performed with chain A of ⌬ 1 -KSTD1⅐ADD structure. DECEMBER 6, 2013 • VOLUME 288 • NUMBER 49 their FAD coenzymes revealed that the ADD and 4-AD ligands are bound in a similar orientation but rotated by ϳ40°in the plane of the steroid ring system. Importantly, the C1, C2, and C3 carbonyl O atoms of ADD virtually superpose on the C5, C4, and C3 carbonyl O atoms of 4-AD, respectively (Fig. 4b). In addition, the hydroxyl groups of Tyr 318 and Tyr 487 of ⌬ 1 -KSTD1 are in very similar positions to those of Ser 468 and Tyr 466 of ⌬ 4 -(5␣)-KSTD, respectively. Whereas in ⌬ 4 -(5␣)-KSTD, the Ser 468 hydroxyl group abstracts a proton from the C4 carbon atom of the substrate, and in ⌬ 1 -KSTD1, the Tyr 318 hydroxyl group is in a position to abstract a proton from C2. Similarly, the N5 atom of FAD can accept a hydride ion from C5 (⌬ 4 -(5␣)-KSTD) or C1 (⌬ 1 -KSTD1). Finally, Tyr 487 of ⌬ 1 -KSTD1 can promote keto-enol tautomerization at C3 ketogroup, similar to Tyr 466 of ⌬ 4 -(5␣)-KSTD. Therefore, this binding mode is in agreement with ⌬ 1 -KSTD1 abstracting a hydride ion and a proton from the C1 and C2 atoms of its 4-AD substrate, respectively, and the C3 carbonyl group of the substrate is involved in keto-enol tautomerization. Thus, whereas the nature of the catalytic residues is the same in ⌬ 1 -KSTD1 and ⌬ 4 -(5␣)-KSTD, the substrate is bound in a different orientation, exposing either the C1-C2 bond (⌬ 1 -KSTD1) or the C4-C5 bond (⌬ 4 -(5␣)-KSTD) to the catalytic residues. This different orientation fully explains the different regiospecificity of ⌬ 1 -KSTD1 and ⌬ 4 -(5␣)-KSTD.

Structure and Mechanism of a 3-Ketosteroid ⌬ 1 -Dehydrogenase
Active Site Residues and Mutational Studies-An amino acid sequence alignment of the available ⌬ 1 -KSTDs from different bacteria revealed that, among the amino acids lining the active site of ⌬ 1 -KSTD1, four residues are strictly conserved (data not shown). These residues are Tyr 119 , Tyr 487 , and Gly 491 from the FAD-binding domain and Tyr 318 from the catalytic domain. These four residues are spatially close to the bound ADD molecule. To assess their importance for ⌬ 1 -KSTD1 activity, the tyrosine residues were mutated into phenylalanines (Y119F, Y318F, and Y487F). All mutants were expressed, purified, and crystallized, employing the same protocol as used for the wildtype enzyme, and their enzymatic properties were compared.
The steady-state kinetic parameters of wild-type and mutant enzymes are shown in Table 3. A substantial reduction in dehydrogenase activity was observed for all three mutants, verifying their importance for activity of ⌬ 1 -KSTD1. The Y318F mutation had the largest effect; it completely abolished the dehydrogenase activity of the enzyme. As discussed above, the hydroxyl group of Tyr 318 is situated close to the C2 atom of the bound ADD product, a position compatible with a role as catalytic base that abstracts a proton from the C2 atom of 4-AD. Surprisingly, the Y119F mutation also showed a drastic reduction of the dehydrogenase activity, with only ϳ0.05% of the wild-type specific activity remaining. Tyr 119 has no close contacts with the bound ADD in the ⌬ 1 -KSTD1⅐ADD complex structure, but its hydroxyl group is at hydrogen-bonding distance from the hydroxyl group of Tyr 318 and thus may serve to increase the basic character of the catalytic base and facilitate proton relay. On the other hand, the Y487F mutation had only a relatively modest effect with a residual activity of ϳ2.6% of that of the wild-type enzyme, despite the hydrogen bonding interaction of the hydroxyl group of Tyr 487 with the C3 carbonyl oxygen of ADD. An explanation for this observation could be that the backbone amide of Gly 491 has also a hydrogen bonding interaction with the C3 carbonyl oxygen. Apparently, these two active site residues may act in tandem, and therefore the absence of Tyr 487 does not completely abolish the dehydrogenase activity of ⌬ 1 -KSTD1. Together they may promote keto-enol tautomerization and labilization of the C2 hydrogen atoms.
Catalytic Mechanism-The catalytic cycle of a typical flavoenzyme includes two half-reactions, i.e., a reductive half-reaction and an oxidative half-reaction (34). In the reductive halfreaction, the flavin coenzyme is reduced by a substrate, whereas in the oxidative half-reaction, the reduced coenzyme is reoxidized by an electron acceptor (34). For ⌬ 1 -KSTDs, the physiological electron acceptor is not known, although vitamin K 2 (35) has been proposed as a natural cosubstrate for the enzyme (35,36). In contrast, the reductive half-reaction has been intensively studied (1,2). The enzyme is able to use a variety of 3-ketosteroid substrates, with a preference for those possessing a double bond at the C4-C5 position (1,2). Based on our structural and mutational studies described above, a detailed catalytic mechanism of the reductive half-reaction can now be proposed (Fig. 5).
A structural superposition of the substrate 4-AD on the product ADD as bound in the ⌬ 1 -KSTD1⅐ADD complex structure revealed that 1) the C1 atom of 4-AD would be at ϳ2.6 Å from the N5 atom of the FAD isoalloxazine ring; 2) the C2 atom of 4-AD would be at ϳ3.0 Å from the hydroxyl group of Tyr 318 ; 3) the isoalloxazine ring and Tyr 318 would be on opposite faces of the A-ring of the substrate, with the isoalloxazine ring at the ␣-face and Tyr 318 at the ␤-face; and 4) the C3 carbonyl group of the substrate would be at hydrogen-bonding distance to the hydroxyl group of Tyr 487 (2.4 Å) and the backbone amide of Gly 491 (3.0 Å). Thus, the substrate would be bound in the active site such that its C1 and C2 atoms are positioned appropriately for hydride and proton abstraction, respectively (Fig. 4c).
The position of Tyr 487 and Gly 491 and their hydrogen bonding interactions with the C3 keto group of 4-AD suggest that they may function in promoting keto-enol tautomerization and labilization of the C2 hydrogen atoms. Whether during the reaction a proton is transferred from the Tyr 487 hydroxyl group to the enolate oxygen atom is currently unknown, but given the significant residual activity of the Y487F mutant protein (Table  3), such a transfer appears not to be absolutely necessary. The Tyr 318 hydroxyl group is at reaction distance to the axial ␤-hydrogen of the C2 atom (2␤-hydrogen) and is in the right position to serve as a general base that abstracts the 2␤-hydrogen. It has a hydrogen bonding interaction with the hydroxyl group of Tyr 119 , which may enhance the basic character of the Tyr 318 hydroxyl group. Moreover, Tyr 318 is positioned at the N-terminal end of ␣-helix ␣10, and the positive helix macro-dipole may provide additional stabilization to the (transient) phenolate form of Tyr 318 . How the abstracted proton is transferred to the solvent is less clear; Tyr 119 may have a role in the proton relay, but the possible proton acceptors, a water molecule or the conserved Tyr 414 hydroxyl group, are too far away (at 5 Å). However, breathing motions of the protein could reduce this distance or allow water to enter and accept the proton from Tyr 119 . Indeed, the electron density suggests that several residues in this region are flexible (e.g., Phe 116 ). As a result of proton abstraction by Tyr 318 , a transient carbanionic species is formed, which is most likely stabilized by keto-enol tautomerization. The negative charge of this species is then transferred to the C1 atom to form a double bond between the C1 and C2 atoms. In concert with this, the axial ␣-hydrogen of the C1 atom (1␣-hydrogen), which is in close proximity to the re-face of the N5 of the isoalloxazine ring, is transferred as a hydride ion to the N5 atom, generating a semireduced anionic FAD. The negative charge of this anion is likely delocalized over the pyrimidine moiety of the isoalloxazine ring system, which is stabilized by hydrogen bonds with the carbonyl and amide backbone of Ser 52 and the amide backbone of Leu 494 as well as by the helix macro-dipole interaction with the N-terminal end of helix ␣15. Reoxidation of the FAD in the subsequent oxidative half-reaction will make the enzyme ready for another cycle of catalysis.
The above mechanism is in full agreement with the stereochemistry of the dehydrogenation reaction catalyzed by ⌬ 1 -KSTDs, which preferentially proceeds via a trans-diaxial elimination of the 1␣ and 2␤-hydrogens (9 -11). However, whereas the 1␣-hydrogen release is an absolute requirement, the 2␤-hydrogen loss is not obligatory (10). A dehydrogenation experiment with the ⌬ 1 -KSTD from Bacillus sphaericus using 2␤-deuterio-5␣-androstane-3,17-dione as substrate showed only 86% loss of the deuterium (10). Although reportedly the deuterated substrate was not the pure 2␤-isomer (10), the incomplete loss of the 2␤-hydrogen could also be due to the formation of a reactive transient species that may undergo fast hydrogen exchange at the C2 atom (9 -11). Such a hydrogen exchange may decrease the stereoselectivity of 2␤-hydrogen elimination. The reactive species was suggested to be either a keto-enolate (10, 11) or a carbanion (9). Both suggestions are based on the fact that ⌬ 1 -KSTD may catalyze hydrogen exchange at the C2 atom of its substrate, even when 1␣-hydrogen expulsion was prevented by the absence of an electron acceptor for the oxidative half-reaction (10,11) or by keeping the flavin coenzyme in the reduced state (9). Indeed, our structures and the observed binding mode of 4-AD show that the active site of the enzyme architecture can stabilize a transient keto-enolate during the reaction, supporting a catalytic mechanism that makes use of keto-enol tautomerization similar to the mechanism of ⌬ 4 -(5␣)-KSTD (32).