Crystal Structure of Δ5-3-Ketosteroid Isomerase from Pseudomonas testosteroni in Complex with Equilenin Settles the Correct Hydrogen Bonding Scheme for Transition State Stabilization*

Δ5-3-Ketosteroid isomerase from Pseudomonas testosteroni has been intensively studied as a prototype to understand an enzyme-catalyzed allylic isomerization. Asp38 (pK a ∼4.7) was identified as the general base abstracting the steroid C4β proton (pK a ∼12.7) to form a dienolate intermediate. A key and common enigmatic issue involved in the proton abstraction is the question of how the energy required for the unfavorable proton transfer can be provided at the active site of the enzyme and/or how the thermodynamic barrier can be drastically reduced. Answering this question has been hindered by the existence of two differently proposed enzyme reaction mechanisms. The 2.26 Å crystal structure of the enzyme in complex with a reaction intermediate analogue equilenin reveals clearly that both the Tyr14 OH and Asp99 COOH provide direct hydrogen bonds to the oxyanion of equilenin. The result negates the catalytic dyad mechanism in which Asp99 donates the hydrogen bond to Tyr14, which in turn is hydrogen bonded to the steroid. A theoretical calculation also favors the doubly hydrogen-bonded system over the dyad system. Proton nuclear magnetic resonance analyses of several mutant enzymes indicate that the Tyr14 OH forms a low barrier hydrogen bond with the dienolic oxyanion of the intermediate.

(3). The unfavorable proton transfer requires the energy given as 2.303RT⌬pK a (3), where ⌬pK a is the difference in pK a values between the proton donor and acceptor.
⌬ 5 -3-Ketosteroid isomerase (KSI) 1 (EC 5.3.3.1) from Pseudomonas testosteroni catalyzes the isomerization of ⌬ 5 -to ⌬ 4 -3ketosteroid by a stereospecific intramolecular transfer of the C4␤ proton to the C6␤ position (4 -6), which is also found in the synthesis of all steroid hormones in mammals. The reaction consists of enolate formation and reketonization that are involved in a wide variety of biologically important reactions of carbonyl compounds. The enzyme, a homodimer in solution, is a "perfect enzyme" enhancing the catalytic rate by a factor of 11 orders of magnitude (7). Since the discovery of this enzyme in 1955 (8), it has been intensively studied as a prototype for understanding the chemical and thermodynamic aspects of enzyme-catalyzed C-H bond cleavage. Three residues have been shown by site-directed mutagenesis and kinetic studies to be important for the catalysis: Tyr 14 (9), Asp 38 (9), and Asp 99 (10,11). Asp 99 is a newly identified catalytic residue on the basis of the solution structure of the enzyme (11) and the crystal structure of a homologous KSI from Pseudomonas putida (10). The latter enzyme shares 34% sequence identity with P. testosteroni KSI and retains the three catalytic residues identically. It is generally agreed upon that Asp 38 serves as the general base abstracting the C4␤ proton of the steroid substrate. This proton transfer poses a common major mechanistic enigma of how a strong acid enzymatic group serves as a general base to abstract proton from a much weaker acid group in a substrate. The large disparity in ⌬pK a of ϳ8 between Asp 38 (pK a ϳ4.7) (12) and the C4␤ proton of the steroid substrate (pK a ϳ12.7) (13) requires 11 kcal/mol of energy according to the equation. The energy has to be supplied in the course of the enzyme reaction and/or the ⌬pK a should be substantially reduced at the enzyme active site by the transition state stabilization to account for the high turnover rate. It has not yet been settled how Tyr 14 and Asp 99 interact with the dienolic intermediate, and thus the underlying mechanism for the enormous rate enhancement by the enzyme cannot be addressed properly. Two different hydrogen bonding schemes for the transition state stabilization have been proposed. In the first proposal, both Tyr 14 and Asp 99 directly provide hydrogen bonds (H-bonds) to the enolate oxygen of the intermediate (Fig. 1 (11) and by the 2.5 Å crystal structure of P. putida KSI in complex with equilenin (10). The second proposal involves a hydrogen-bonded catalytic dyad, Asp 99 COOH-Tyr 14 OH-Osteroid, consisting of normal hydrogen bonds in the enzymesubstrate complex. As the intermediate dienolate is formed, the H-bond between Asp 99 and Tyr 14 is strengthened to form a low barrier hydrogen bond (LBHB) that facilitates polarization of the C3-keto group of the steroids (Fig. 1, mechanism II). This proposal is primarily based on the detection and assignment to the Asp 99 COOH of a highly deshielded proton NMR peak at 18.15 ppm in the presence of dihydroequilenin (14). Such anomalously downfield-shifted resonances have been typically observed for a number of H-bonds between two heteroatoms with an equal pK a in an apolar environment (15) and attributed to the formation of unusually strong H-bonds (15)(16)(17)(18). The resolution (2.5 Å) of the crystal structure of P. putida KSI complexed with equilenin was criticized as being insufficient to distinguish distances shorter than 0.75 Å and to discern other possible binding modes of equilenin (19). To resolve this conflicting issue and provide a sound ground for addressing the more significant issue of how the enormous rate enhancement is achieved by the enzyme, we have determined the structure of P. testosteroni KSI in complex with equilenin (Scheme 1). The elucidation of the equilenin binding modes observed for the six KSI molecules in the asymmetric unit of the crystal, together with the previously reported structure of the uninhibited enzyme (20), settles the confusion regarding the catalytic mechanism of this heavily studied enzyme.

MATERIALS AND METHODS
Protein Purification-D38N mutant enzyme was used for the structure determination of KSI in complex with equilenin because it binds steroid analogues more tightly than the wild-type enzyme owing to the absence of the negative charge at the residue 38 (12). Furthermore, Asn 38 mimics the protonated form of Asp 38 in the transition state. D38N mutant gene was constructed by site-directed mutagenesis using a Muta-gene mutagenesis kit (Bio-Rad). For NMR studies, P. putida KSI mutant genes D38N, D38N/Y14F, and D38N/D99L were prepared similarly. Overexpression and purification of the wild-type and the mutant enzymes were carried out as described previously (21) with a slight modification. The mutant KSI genes were subcloned into the EcoRI and HindIII sites of pKK223-3 to construct a recombinant plasmids. The enzymes were expressed in the Escherichia coli strain BL21 (DE3) transformed with each of the plasmids when induced with 0.5 mM isopropyl-␤-D-thiogalactopyranoside. Unlike other mutant enzymes, the D38N/D99L mutant KSI was expressed in lower quantities and mostly as inclusion bodies. The crude lysate containing overexpressed KSI was applied directly onto a deoxycholate affinity column equilibrated with 40 mM potassium phosphate buffer (pH 7.0) containing 1 mM EDTA. After the column was washed, the bound proteins containing KSI were eluted with 50 mM Tris-HCl buffer (pH 7.0) containing 25% ethanol and finally separated on a Superose 12 column (Amersham Pharmacia Biotech). The purified enzymes, which showed a homogeneous single band on SDS-polyacrylamide gel electrophoresis, were concentrated by Centricon-10 (Amicon) for crystallization.
Crystallization-It was quite difficult to obtain suitable crystals of P. testosteroni KSI in either the uninhibited (20) or the inhibited state. The crystals of the complex were obtained by co-crystallization method. Small amount of equilenin was dissolved in dimethyl sulfoxide, and 2 l of the solution was mixed with 70 l of enzyme solution (10 mg/ml). Saturated amount of equilenin was ensured by the presence of white precipitate of the compound formed immediately after the mixing. The crystals grew in droplets containing a 2-l protein sample and an equal volume of precipitant solution containing 25% polyethylene glycol 4000, 0.1 M sodium acetate (pH 4.6), and 0.2 M ammonium sulfate at 12°C. The crystallization conditions were different from those used for the uninhibited wild-type enzyme (20), where 1.3 M ammonium sulfate was the major precipitant. Like the crystals of uninhibited enzyme, the crystals of the complex tended to grow in clusters of microcrystals under many different conditions. While refining 15 different initial crystallization conditions for a long period of time, one crystal happened to grow into a large single crystal (0.15 ϫ 0.2 ϫ 0.3 mm) with which we were able to determine the structure. The crystal belongs to the space group P2 1 with unit cell dimensions of a ϭ 73.90 Å, b ϭ 72.51 Å, c ϭ 83.83 Å, ␤ ϭ 104.29°and contains three dimers of the enzyme in the asymmetric unit. There is no local symmetry between the three dimers, and the crystal packing environment of each dimer is different from the others.
Data Collection, Structure Determination, and Refinement-Diffraction data (Table I) were measured on a DIP2020 area detector with graphite monochromated CuK␣ x-rays generated by a MacScience M18XHF rotating anode generator operated at 90 mA and 50 kV at room temperature. Data reduction, merging, and scaling were accomplished with the programs DENZO and SCALEPACK (22). The structure was solved by the molecular replacement protocol in the X-PLOR program package (23) using the dimeric structure of the uninhibited enzyme (20) as a search model. Patterson correlation refinement followed by a rotation search yielded three refined rotation function peaks that were outstanding from noise level solutions. Translation searches along the xz plane with the three refined rotation function peaks resulted in top solutions that were about 1.2 above the next highest unrelated peaks. Translation searches along the y axis for origin correlation yielded correct translation positions as judged by no serious overlaps between symmetry mates. The rigid body refinement of the translation solutions, consisting of 20 steps of conjugate gradient minimization against reflections at 10 -4.0 Å resolution, resulted in an R-factor of 34.8%. 5% of the reflections was set aside for monitoring R free . Even at this stage, strong electron densities for equilenins bound to the six monomers were clearly visible. Several rounds of manual refitting and atomic positional refinement were carried out to yield an R working of 26.2% at 10.0 -2.26 Å resolution. Guided by the 2F o Ϫ F c and F o Ϫ F c electron density maps, equilenins were fit into the electron densities. Especially, the protruding density for the C13 methyl group of equilenin warranted unambiguous fittings. Strong electron density, observed between two symmetry related molecules in the middle of two arginine and two histidine residues, was interpreted as a sulfate ion, which was included in the crystallization conditions. The current model containing a total of 156 water molecules and one sulfate anion exhibits an R working of 19.6% (R free ϭ 25.6%). The crystallographic statistics for the final refined structure is shown in Table I. NMR Experiment-Proton NMR spectra of P. putida KSIs were collected with Bruker Avance DRX500 spectrometer at Ϫ3°C except for the spectrum of D38N/D99L mutant, which was collected at Ϫ6°C. Jump-and-return pulse sequence (24) was employed to suppress water signal. Typically, a total of 1024 transients in 32,768 data points, relaxation time of 2.0 s, and spectral width of 17,000 Hz were used. Line broadening of 10 Hz was employed in processing free induction decays with XWIN-NMR software (Bruker Instruments). The chemical shift was referenced from that of tetramethylsilane as an internal standard.
Theoretical Calculations-The energetics and reactivity of the two different mechanisms were compared using density functional calculations employing Becke-3-Lee-Yang-Parr (B3LYP) functionals and 6 -31ϩG* basis set (25). The structures of the active site having the three key catalytic residues were fully optimized without constraints. In the calculation, Asp 38 , Asp 99 , Tyr 14 , and the steroid substrate were simply modeled as formate, formic acid, phenol, and ␤,␥-enone derivative, respectively.

RESULTS AND DISCUSSION
Overall Structure and Active Site Environment-The structure of KSI is a conical closed barrel formed by a highly curved ␤-sheet and three ␣-helices (Fig. 2). The widely open end of the structure forms the entrance to the deep active site cavity tapering off to a blocked wall, which is inaccessible from the other narrow end. So far, three other proteins, bearing no sequence and functional homology to each other and to KSI, have been found to have a similar tertiary structure: scytalone dehydratase (26), nuclear transport factor 2 (27), and the ␤ subunit of naphthalene dioxygenase (28). Considering the hydrophobic nature of the substrates for the three enzymes, the observed folding pattern appears to be a general scaffold for an active site cavity that binds hydrophobic substrates.
The two critical catalytic residues, Tyr 14 and Asp 99 , are close to each other and located deeply in the active site cavity, whereas Asp 38 is located in the midway of the cavity (Fig. 2). The most prominent feature of the active site is the highly apolar characteristics as a result of the lining by 22 nonpolar amino acid side chains (20), which constitute a favorable environment for the binding of the steroid substrates. The side chain of Asp 99 , surrounded by apolar side chains of Phe 80 , Phe 82 , Pro 97 , Phe 101 , Met 112 , and Ala 114 , is believed to have an unusually elevated pK a of 9.5 to account for the sharp decrease in the enzyme activity at pH 9.5 (29) and for the presence of a group with pK a of 9.5 near Tyr 14 as indicated by fluorescence titration (30).
Inhibitor Binding Modes-In the structure of P. testosteroni KSI complexed with equilenin, two different binding modes of equilenin were observed because of different crystal packing interactions. Although equilenins bound to four KSI molecules in the asymmetric unit of the crystal are exposed to the bulk solvent, equilenins bound to the remaining two KSI molecules participate in the crystal packing interactions with neighboring molecules. In the four KSI molecules, equilenin shows the normal binding mode, which explains much of the biochemical and mutational data piled on this enzyme (Fig. 3). The side chain of Asn 38 is located distinctively above the ␤-face of the steroid with its N ␦2 atom within 3.6 Å of both the C4 and C6 of equilenin. This explains the stereospecific proton transfer between the two positions. The side chains of both Tyr 14 and Asp 99 are within H-bond distance from the C3 oxygen of the inhibitor. Thus, the two residues should be able to provide H-bonds for the stabilization of the developing negative charge on the C3 oxygen of the steroid substrate upon loosing the C4␤ proton. Mutations of Y14F and D99A result in 50,000-fold (31) and 5,000-fold (11) decrease in k cat , respectively. In the other two KSI molecules, the C17-keto oxygen of equilenin is involved in H-bonds via a water molecule with the side chains of Arg 13 and Asp 22 from a symmetry mate. In addition, the steroid C ring is in close contact with Ala 20 from the same symmetry mate. These interactions resulted in a nonproductive binding, the best description of which would be a rotation of the productive binding by ϳ180°about the C3-O bond of equilenin (Fig.  3). The nonproductive binding observed in the crystal structure is not relevant in solution. Regardless of the binding modes, the C3 oxygen is involved in direct H-bonds with the side chains of both Tyr 14 and Asp 99 (Table II). The high pK a of Asp 99 and its direct interaction with the inhibitor indicate that the bound inhibitor should be in the anion form (pK a ϳ 9.0) (32), whereas Tyr 14 OH (pK a ϳ11.6) (30) and Asp 99 COOH are in the protonated states for favorable interactions. Therefore, it is apparent that the high pK a of Asp 99 is to insure the stabilization of the developing negative charge on the C3 oxygen of the dienolate intermediate together with the Tyr 14 OH by forming a charged H-bond.
More than 90 different steroid compounds are known to inhibit KSI (33). Some compounds may bind to KSI "backwards" with the steroid D ring approaching the catalytic residues. This has been suggested by the covalent modification of Asp 38 by oxyrane compounds containing an epoxide ring either at the steroid A ring or at the D ring (34). The large active site cavity of KSI allowing two different equilenin binding modes explains the backward binding, although it is not observed in the crystal structure. It was possible to dock equilenin backwards into the active site without any steric or electrostatic violation so that the C17-keto group of equilenin interacted with Tyr 14 and Asp 99 (with the aid of solid docking module in the program Quanta). The ionization of the C3-hydroxyl group should favor the normal binding mode over the backward bind- ing where the C17-keto oxygen would form weaker neutral H-bonds with Tyr 14 and Asp 99 . We speculate that the steroid compounds containing a keto oxygen at both the C3 and C17 positions would bind to KSI normally or backward with similar affinities.
Comparison with the Structure of Uninhibited KSI- Table II shows the H-bond distances from Tyr 14 OH and Asp 99 COOH to the C3 oxygen of equilenin and those between the two residues in the six molecules of KSI. Because the electron densities for the two residues and equilenin are well defined, the H-bond distances should be well defined within the coordinates errors less than 0.3 Å for all atoms as estimated by Luzzati plot (35). Clearly, Tyr 14 and Asp 99 are beyond H-bond distance (Ͼ4 Å) regardless of the inhibitor binding modes, negating the catalytic dyad mechanism. On the basis of a comparison of the solution structures of the enzyme and that in complex with 19-nortestosterone hemisuccinate , it was noted by other investigators that the binding of 19-NTHS to the active site resulted in the movement of the three ␣-helices, with the Tyr 14 OH approaching the Asp 99 COOH. This provided a physical ground for the direct H-bond between the two residues (19). In contrast, our comparison of the crystal structures of the uninhibited KSI and the enzyme complexed with equilenin shows no noticeable movement of the ␣-helices (Fig. 4). The rmsd after superposition of the backbone atoms of the two structures is less than 0.22 Å for any of the six monomers, whereas the rmsd values after superposition of the backbone atoms of the solution structures of TI and TI/19-NTHS are 1.26 Ϯ 0.11 and 1.30 Ϯ 0.13 Å for the two subunits. Probably, the large rmsd values of the assemble NMR structures (TI, 1.54 Ϯ 0.24 Å; TI/19-NTHS, 1.58 Ϯ 0.41 Å for dimer backbone atoms) would have led to the conjecture that the inhibitor binding causes the movement of the ␣-helices.
Involvement of a LBHB-It has been a controversial issue (36 -39) that special types of short, strong H-bonds, termed LBHBs, play an important role in the catalytic reaction of many enzymes by providing a large amount of stabilization energy (more than 10 kcal/mol) (17,18). Here, we confine the term to the H-bonds that display highly deshielded 1 H NMR resonances but may not necessarily be as strong as proposed. A characteristic resonance is detectable at 18.15 ppm in the complex of P. testosteroni KSI with dihydroequilenin (32) that contains a hydroxyl group at the C17 position instead of a keto oxygen compared with equilenin. The resonance disappears upon substitution of either Y14F or D99A (14). Assignment of the 18.15 ppm resonance to a LBHB between Asp 99 COOH and Tyr 14 OH in mechanism II provides an explanation for the disappearance of the resonance by either of the two substitutions. We suggest that the resonance arises from the LBHB between the Tyr 14 OH and the oxyanion of the bound inhibitor and that the disappearance of the peak by the D99A mutation can be explained by the solvent effect. A void space generated by the alanine substitution is expected to increase the access of   the bulk solvent, resulting in a decrease in the strength of the H-bond between Tyr 14 OH and the bound inhibitor with an increase of dielectric constant. In small molecular systems, unusually downfield-shifted resonances were observed only in the absence of a water molecule (18). Our proposal can be tested by recording a NMR spectrum of the D38N/D99L double mutant of P. testosteroni KSI in which Leu 99 would prevent the access of the bulk solvent. Although we could not obtain this mutant because of a very low expression level and insoluble nature, we were able to obtain the D38N/D99L mutant of P. putida KSI in sufficient quantity for NMR analysis. Like P. testosteroni KSI, the D38N mutant of P. putida KSI also exhibits a highly deshielded resonance at 16.8 ppm in the presence of equilenin (Fig. 5B). A slightly deshielded peak at 13.1 ppm also appears in the spectrum of this mutant. In the spectrum of the D38N/D99L mutant, the 13.1 ppm peak disappears, but the 16.8 ppm is retained and further downfield shifted by 0.9 ppm (Fig. 5C). The 16.8 ppm peak disappears in the D38N/Y14F mutant (Fig. 5D). Therefore, the 16.8 ppm peak can be assigned to the Tyr 14 OH, and the 13.1 ppm peak can be assigned to the Asp 99 COOH. The 0.9 ppm downfield shift, upon the substitution of the D99L, is most likely because of an increase in H-bond strength between the Tyr 14 OH and the bound inhibitor caused by a decrease in dielectric constant at the immediate vicinity of Tyr 14 . In the D38N/Y14F mutant, the 13.1 ppm peak also disappears. This can be explained by a fast exchange of the inhibitor because of the loss of the H-bond of Tyr 14 . The line width of the 17.7 ppm peak in the D38N/D99L mutant (Fig. 5C) broader than that of the corresponding peak in the D38N mutant (Fig. 5B) can also be explained by an increased exchange rate of the inhibitor because of the loss of the H-bond of Asp 99 . The observations support that the LBHB of Tyr 14 is stronger than the H-bond of Asp 99 . We have confirmed by crystal structure determinations that equilenin binds to the D38N/D99L mutant in the same manner as to the D38N mutant of P. putida KSI (data not shown). The highly homologous tertiary, quaternary, and active site environment between the two KSIs strongly suggests that the same spectroscopic results would be obtained for the D38N/D99L double mutant of P. testosteroni KSI. Including the LBHB observed for KSIs, the highly deshielded 1 H resonances have been detected only for charged H-bonds that should be stronger than neutral H-bonds. These observations, the disappearance of the 18.5 ppm resonance upon D99A mutation in P. testosteroni KSI, and the upfield shift of such resonances with increase in ⌬pK a between two heteroatoms (40) suggest that the anomalously shifted resonances do indeed correlate with unusual strengths of LBHBs.
Ab Initio Calculations-In addition to the x-ray and NMR analyses, ab initio theoretical calculations were employed for the comparison of the energetics and catalytic reactivity of the doubly H-bonded oxyanion system (mechanism I) with those of the catalytic dyad system (mechanism II). In both mechanisms I and II, each whole reaction path from the enzyme-substrate reactant to the product was found to have two rate-determining steps that are almost symmetric with respect to the enzymesubstrate intermediate, although the enzyme product is ϳ6 kcal/mol lower in energy than enzyme-substrate reactant. The predicted energy profile including the transition state of the first rate-determining step responsible for k cat is shown in Fig.  6. The molecular system in mechanism I is more stabilized than that in mechanism II. The energies of the enzyme-substrate reactant, transition state, and enzyme-substrate intermediate in mechanism I are lower than those of the three species in mechanism II by 0.96, 2.84, and 3.55 kcal/mol, respectively. In particular, the enzyme-substrate intermediate of mechanism II is less stable than that of mechanism I, because in the former mechanism the hydroxyl proton of phenol as a mimic of Tyr 14 transfers partially to the steroid C3 oxygen whose proton-donating strength is greater than that of phenol. The extra stabilization of the intermediate state in mechanism I is reflected in its activation barrier of 5.53 kcal/mol in comparison with 7.41 kcal/mol in mechanism II. The lower activation energy barrier in mechanism I is more effective in stability and reactivity by 25-fold than in mechanism II. Thus, the ab which equilenin was predissolved. Peak c, which changes position with pH change, is assigned to an imidazole NH proton of a histidine residue. At pH 7 the peak is indistinguishable from peak b. Although not shown here, we recorded an NMR spectrum of D38N mutant P. testosteroni KSI in the presence of equilenin. We detected a similar deshielded peak at 17.51 ppm instead of 18.15 ppm (see text). Two other groups have detected similar peaks at 17.4 (11) and 17.49 ppm (40) in the presence of equilenin. Therefore, ϳ17.5 ppm should be the correct chemical shift value for this resonance when referenced to tetramethylsilane. The hydroxyl oxygen on the D ring of dihydroequilenin instead of a keto group would not affect the chemical shift of the peak. initio theoretical results clearly favor the doubly charged Hbonds over the catalytic dyad, and this is in agreement with the crystal structures. KSI should have evolved to choose the more efficient low lying energy pathway. A full description of the calculations will be reported elsewhere.
In conclusion, the 2.26 Å resolution structure of KSI in complex with equilenin, the NMR spectroscopic evidences, and the theoretical calculations indicate that both Tyr 14 and Asp 99 should be directly involved in the stabilization of the dienolate intermediate and that Tyr 14 provides a LBHB to the intermediate in the highly apolar milieu of the active site. Owing to the simplicity of the reaction mechanism that is settled now, KSI is an excellent system for answering the contentious issue of how enzymes achieve rate accelerations of 10 8 -10 15 . Furthermore, the availability of a wide variety of KSI inhibitors should allow probing detailed nature of LBHBs by spectroscopic methods directly at the enzyme active site with the clear picture of the equilenin binding mode.