X-ray structure of human beta3beta3 alcohol dehydrogenase. The contribution of ionic interactions to coenzyme binding.

The three-dimensional structure of the human b3b3 dimeric alcohol dehydrogenase (b3) was determined to 2.4-Å resolution. b3 was crystallized as a ternary complex with the coenzyme NAD and the competitive inhibitor 4-iodopyrazole. b3 is a polymorphic variant at ADH2 that differs from b1 by a single amino acid substitution of Arg-369 3 Cys. The available x-ray structures of mammalian alcohol dehydrogenases show that the side chain of Arg-369 forms an ion pair with the NAD(H) pyrophosphate to stabilize the EzNAD(H) complex. The Cys-369 side chain of b3 cannot form this interaction. The three-dimensional structures of b3 and b1 are virtually identical, with the exception that Cys-369 and two water molecules in b3 occupy the position of Arg-369 in b1. The two waters occupy the same positions as two guanidino nitrogens of Arg-369. Hence, the number of hydrogen bonding interactions between the enzyme and NAD(H) are the same for both isoenzymes. However, b3 differs from b1 by the loss of the electrostatic interaction between the NAD(H) pyrophosphate and the Arg-369 side chain. The equilibrium dissociation constants of b3 for NAD 1 and NADH are 350-fold and 4000-fold higher, respectively, than those for b1. These changes correspond to binding free energy differences of 3.5 kcal/mol for NAD and 4.9 kcal/mol for NADH. Thus, the Arg-369 3 Cys substitution of b3 isoenzyme destabilizes the interaction between coenzyme and b3 alcohol dehydrogenase.

ADH subunits share greater than 93% sequence identity and form a complex group of homodimeric and heterodimeric isoenzymes. 2 These isoenzymes have relatively high catalytic efficiency for ethanol oxidation and account for the majority of ethanol oxidation by the liver (3,5,6).
Three-dimensional structures have been determined for horse liver EE ADH as an apoenzyme and complexed with a variety of substrates or inhibitors (12,13). Structures of ␤ 1 complexed with NAD(H) and cyclohexanol, ␤ 2 complexed with NAD ϩ and 4-iodopyrazole, and ␤47Gly complexed with NAD ϩ have been solved to 2.5-Å resolution (14). These structures show that residues 47 and 369 are located in the coenzyme binding cleft and provide electrostatic stabilization for coenzyme binding to both horse liver and human ␤ 1 ADH (13,14).
In this study, we report the structure of ␤ 3 to 2.4 Å and the structure of ␤ 1 to 2.2 Å in complexes with NAD ϩ and 4-iodopyrazole. Structures of the ␤ 3 and ␤ 1 ternary complexes are compared in an effort to understand the molecular basis for differences in coenzyme binding affinity.

MATERIALS AND METHODS
Site-directed Mutagenesis-Single stranded ␤ 1 cDNA in the M13HinEco1 vector (15) was used as a template for site-directed mutagenesis (Amersham) as described previously for ␤ 1 mutants (16). The codon for Arg-369 (CGT) of the ␤ 1 cDNA was mutated to TGT (Cys) using the oligonucleotide (5Ј-GGGAAAAGTATCTGTACCGTCCT-GACG-3Ј). After subcloning the ␤ 3 cDNA into the pKK223-3 expression plasmid (␤ 3 /pKK223-3), the coding sequence of the ␤ 3 cDNA insert was completely sequenced with a Sequenase kit, version 2.0 (U.S. Biochemical Corp.) to verify that only the desired change was made in the ␤ 1 cDNA sequence.
Enzyme Expression, Purification, and Crystallization-The ␤ 3 / pKK223-3 expression plasmid was transformed into competent Escherichia coli TG-1 cells, and ␤ 3 was expressed as described previously for ␤ 1 ADH (16). ␤ 3 was purified from an E. coli lysate in three chromatographic steps, DEAE-cellulose, S-Sepharose, and Affi-Gel Blue, as described for other ␤ ADH variants (17), except that zinc sulfate concentrations were increased to 0.1 mM in lysis and column buffers. Addition of zinc sulfate increased the enzyme's stability during purification (18). * This work was supported by National Institutes of Health Grants R37-AA07117 (to W. F. B.), K21-AA00148 (to C. L. S.), and K21-AA00150 (to T. D. H.). 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 Purified ␤ 3 was dialyzed into 10 mM HEPES, pH 7.5, buffer containing 2 mM dithiothreitol. Enzyme purity was evaluated by electrophoresis on sodium dodecyl sulfate-polyacrylamide gels (19), using a Hoefer Mighty Small apparatus (Hoefer Pharmacia Biotech, San Francisco, CA). Gels were stained using the alkaline silver stain method (20). Purified ␤ 3 was concentrated with a Microcon 30 concentrator (Amicon, Beverly, MA) to 10 mg of protein/ml and filtered through a sterile 0.22-m cellulose-acetate membrane. Protein concentrations were determined using the Bio-Rad dye reagent protein assay (Bio-Rad) and bovine serum albumin as a standard (21).
Recombinant human ␤ 3 , 10 mg/ml, was crystallized as a ternary complex. Crystals were grown in 100 mM sodium phosphate, pH 7.5, 7.5 mM NAD ϩ , 1 mM 4-iodopyrazole, and 16.5% (w/v) polyethylene glycol 8000. The concentrations of NAD ϩ and 4-iodopyrazole exceeded the K m,NADϩ by 9-fold and the K i,4MP by 900-fold, respectively (Table I). ␤ 3 was crystallized using CRYSCHEM sitting drop plates (Charles Supper Co., Inc., Natick, MA) and equilibrium drop volumes of 3 l. Crystals grew to maximal size in 4 to 5 days and were stable for up to 1 month.
Recombinant human ␤ 1 was expressed and purified as described previously (14). Crystals of the ␤ 1 ternary complex with NAD ϩ and 4-iodopyrazole were grown at a protein concentration of 15 mg/ml using equilibrium drop volumes of 4 l. The crystallization buffer contained 50 mM sodium phosphate, pH 7.5, 2 mM NAD ϩ , 1 mM 4-iodopyrazole, and 13.5% (w/v) polyethylene glycol 8000.
X-ray Diffraction Data Collection and Processing-X-ray diffraction data for human ␤ 3 were collected at 25°C on a Rigaku 200HB rotating anode generator equipped with a RAXIS IIC image plate area detector at a crystal-to-detector distance of 12.5 cm. The ␤ 3 diffraction data were collected to a resolution of 2.37 Å from two orientations of a single crystal. The ␤ 1 diffraction data were collected to a resolution of 2.19 Å from two orientations of a single crystal at a crystal-to-detector distance of 11 cm. RAXIS IIC data processing software (T. Hagashi, Rigaku Corp., Japan) was used to index, merge, and scale the data sets.
Molecular Replacement Calculations and Structure Refinement-The enzyme and NAD ϩ coordinates from the human ␤ 1 ⅐NAD(H)⅐cyclohexanol ternary complex (14) were used directly to solve the structure of both the ␤ 3 and ␤ 1 ternary complexes. Refinements were performed using the program X-PLOR (22). The new data sets were isomorphous with the original ␤ 1 data (14). Therefore, each complex was oriented by rigid body refinement using the data between 8 and 2.8 Å. The resulting R-factors for the correctly oriented molecules were 21.5 and 28.7, for ␤ 1 and ␤ 3 , respectively. Cys was substituted for Arg-369 in the ␤ 3 model using CHAIN (23). The atomic positions were refined to 2.8 Å using the positional refinement protocol. 2F o Ϫ F c and F o Ϫ F c maps were generated using this coordinate set and were inspected with the program CHAIN. The position for 4-iodopyrazole was identified in the active sites and included in subsequent refinement steps. Independent refinement of each model was performed in stages with data of increasing resolution until all data between 8 Å and the limit of resolution (2.4 Å for ␤ 3 and 2.2 Å for ␤ 1 ) with F/ F greater than 1 were included. Water molecules were added independently to each model once the resolution exceeded 2.5 Å and were assigned by inspecting F o Ϫ F c and 2F o Ϫ F c maps for strong positive electron density within 3.3 Å of hydrogen bond acceptors or donors. Following each cycle of rebuilding, the coordinates were refined using the positional and individual restrained isotropic temperature factor refinement protocols. Non-crystallographic symmetry restraints were applied with a weight of 2 kcal/mol per Å 2 to all atoms during the first round of positional refinement and then only to main chain atoms for subsequent rounds of positional refinement. Noncrystallographic symmetry restraints were removed during the last two rounds of refinement. Alignments of the backbone ␣ carbons of ␤ 3 and ␤ 1 isoenzymes were performed with the programs QUANTA (Molecular Simulations Inc., Burlington, MA) and X-PLOR (22).
Steady-state and Stopped-flow Kinetics-Steady-state kinetic constants were evaluated in 0.1 M sodium phosphate buffer at pH 7.5 and 25°C, utilizing a Perkin Elmer Lambda 6 double beam spectrophotometer. Ethanol oxidation was followed by monitoring the production of NADH at 340 nm (molar extinction coefficient: 6.22 mM Ϫ1 cm Ϫ1 ). The inhibition constant for 4-methylpyrazole (K i,4MP ) exhibited by recombinant ␤ 3 was calculated with Cleland's programs (24). The inhibition kinetics were analyzed using both competitive and noncompetitive models. Since K ii was 10 times K is in the noncompetitive fit, the competitive fit is reported in Table I.
Stopped-flow kinetics were evaluated in 15 mM PIPES, 15 mM Bicine, pH 7.5, at 25°C. Data were collected on a HITECH SF-51 instrument (17). Exponential traces were analyzed by HITECH software. Linear regressions were analyzed with SAS (Cary, NC) and evaluations for goodness-of-fit were performed by comparing F-statistics and correla-tion coefficients.
Apparent NADH association (k on,NADH ) and dissociation (k off,NADH ) rate constants were determined for recombinant ␤ 3 as described (17) Apparent NAD ϩ association (k on,NADϩ ) and dissociation (k off,NADϩ ) rate constants were obtained by monitoring the quench of intrinsic tryptophan fluorescence intensity of the enzyme that is associated with NAD ϩ binding. Emission intensities above 320 nm (320G filter) were obtained while exciting at 280 nm (slit width 5 nm). In these reactions, enzyme was mixed with NAD ϩ concentrations at final concentrations of 1.5 M enzyme and 39 to 1670 M NAD ϩ . Estimated apparent k on,NADϩ and k off,NADϩ rate constants were evaluated by linear regression; data best fit a linear model within this NAD ϩ concentration range (r 2 ϭ 0.89).
Apparent equilibrium NAD ϩ and NADH dissociation constants (K d,NADϩ , and K d,NADH ) for ␤ 3 were estimated from the ratio of apparent k off and k on rate constants for NAD ϩ and NADH.
Free Energy Associated with Coenzyme Binding and Calculations of the Electrostatic Contributions to Coenzyme Binding-Coenzyme binding free energies (⌬G 0 ) were calculated from equilibrium dissociation constants (K d ) using the equation ⌬G ϭ ϪRTln(1/K d ) (Equation 1). The difference in coenzyme binding free energies (⌬(⌬G 0 )) between the ␤ 1 and ␤ 3 isoenzymes was calculated using the following equation: Coulomb's Law, ⌬G E ϭ q 1 q 2 /4r⑀ o D, was used to estimate the electrostatic contribution to the coenzyme binding energy in ␤ 3 and ␤ 1 isoenzymes. Calculations for NAD(H) binding energy differences included the electrostatic interactions between the positive charge shared by the terminal nitrogens of Arg-369 and two negative charges shared by the free oxygens of the NAD(H). For calculation of NAD ϩ binding differences between ␤ 3 and ␤ 1 isoenzymes, the contribution of the interaction between the positively charged Arg-369 nitrogens with a positive charge on the nitrogen of the nicotinamide ring of the NAD ϩ was also included. The average crystallographic distances between the terminal nitrogens of Arg-369 and the free oxygens of the coenzyme's nicotinamide phosphate, the free oxygens of the adenosine phosphate, and the nicotinamide nitrogen are 3.9 Å, 6.7 Å, and 7.3 Å, respectively. Solvent Accessibility Calculations-Solvent accessibility calculations for amino acid side chains were performed with QUANTA (Molecular Simulations Inc., Burlington, MA) using a spherical probe 2.8 Å in diameter (25).
Reagents-NAD ϩ , grade I was purchased from Boehringer Mannheim, trans-4-N,N-dimethylaminocinnamaldehyde and 4-iodopyrazole were purchased from Aldrich, and ethanol was purchased from Midwest Grain (Pekin, IL). All other reagents were from Sigma.

RESULTS
Purified recombinant ␤ 3 exhibited a single band on a sodium dodecyl sulfate-polyacrylamide gel at the expected subunit molecular mass of 40 kDa (data not shown). Eighty milligrams of purified ␤ 3 were routinely obtained from 9.8-liter cultures. These yields are similar to those obtained for ␤ 1 (14). The specific activity of purified recombinant ␤ 3 at 2.5 mM NAD ϩ and 66 mM ethanol, 3.8 units/mg, was higher than that of purified ␤ 3 from human liver obtained at autopsy, 2.8 units/mg (11). The K m,NADϩ , K i,4MP , V max /K m,ethanol , V max for ethanol oxidation, and pH optimum of recombinant ␤ 3 are similar to   (Table I).
The crystallization conditions for ␤ 3 were similar to those for the ␤ 1 , ␤47Gly, and ␤ 2 isoenzymes (14,26). ␤ 1 and ␤ 3 were crystallized in the presence of NAD ϩ and 4-iodopyrazole. Large ␤ 3 crystals, which diffracted to 2.4 Å, could be obtained from high specific activity enzyme within 4 days of enzyme preparation. X-ray diffraction data were collected to 2.37 Å for ␤ 3 and 2.19 Å for ␤ 1 . Unit cell information and data collection statistics are reported in Table II. The completeness for the ␤ 3 data set was 86% to 2.4 Å and 70% within the highest resolution shell (2.51-2.40 Å), and that for ␤ 1 was 83% complete to 2.2 Å and 70% complete between 2.3 and 2.2 Å. The ␤ 3 cell dimensions and angles are virtually identical to those of ␤ 1 (Table II). The final refined ␤ 3 ⅐NAD ϩ ⅐4-iodopyrazole model, including all non-hydrogen atoms and 142 water molecules, possesses a R-factor of 17.5% for the data between 8 and 2.4 Å and a free R-factor of 24.3% (Table III). The ␤ 1 model contains 314 water molecules and possesses a R-factor of 18.0% and a free R-factor of 26.0% for the data between 8 and 2.2 Å (Table III). The electron density in the vicinity of the active site zinc and bound NAD ϩ for these structures are shown in Fig. 1. Estimates of the average coordinate error, ϳ0.23 Å, were obtained from an analysis of the R-factor versus resolution (27). Root-mean-square differences from ideal bond angles and from ideal bond lengths were 1.7°and 0.008 Å, respectively (Table III).
The ␤ 3 and ␤ 1 ternary complex structures are very similar. Alignment of the ␣-carbons in the ␤ 3 and ␤ 1 structures indicates that these structures have identical domain arrangements and orientations. Alignment of ␤ 3 and ␤ 1 reveals that the only differences in these structures occur at the site of the Arg-369 3 Cys substitution (Fig. 2). Cys-369 and two water molecules in the ␤ 3 coenzyme binding site occupy the position of Arg-369 in the ␤ 1 coenzyme binding site (Fig. 2). The two water molecules occupy similar positions as two of the guanidino nitrogens of Arg-369 and form similar hydrogen bonds to the coenzyme molecule (Fig. 2). Based on these structures, the only residue that contributes differently to coenzyme binding is 369, Cys in ␤ 3 and Arg in ␤ 1 . At pH 7.5, the Arg-369 3 Cys substitution in ␤ 3 dramatically affects the apparent association and dissociation constants for coenzyme binding, relative to ␤ 1 (Table IV). The k on,NADϩ and k on,NADH rate constants with ␤ 3 are 19-and 12-fold slower, respectively, than with ␤ 1 . The k off,NADϩ and k off,NADH rate constants with ␤ 3 are 19-and 330-fold faster, respectively, than with ␤ 1 . These differences result in 350-and 4000-fold increases in the K d,NADϩ and K d,NADH values for ␤ 3 relative to ␤ 1 . The equilibrium binding differences were used to calculate the free energy differences (⌬(⌬G)) in coenzyme binding. These calculations show that NAD ϩ and NADH bind to ␤ 3 3.5 and 4.9 kcal/mol, respectively, less favorably than to ␤ 1 .

DISCUSSION
The structure of ␤ 3 was determined to 2.4 Å resolution and that of ␤ 1 was extended to 2.2 Å. The final ␤ 3 model, containing 142 ordered water molecules and individual restrained isotropic temperature factors for all atoms, possesses a R-factor of 17.5%, a free R-factor of 24.3%, and good stereochemistry (Table III). The corresponding structure of ␤ 1 contains 314 water molecules, individual restrained isotropic temperature factors for all atoms, and a R-factor of 18.0% with a free R-factor of 26.0%. The structure of the ␤ 1 ternary complex with NAD ϩ and 4-iodopyrazole was determined in order to ensure that any differences between the two enzyme structures were not due to differences in the type of ligands bound in the substrate binding site.
An alignment of the ␤ 3 structure to the ␤ 1 structure indicates that they are virtually identical. The r.m.s. deviation of the alignments were similar to the Luzatti estimate of the coordinate error (Fig. 2 and Table III). The only substantial difference between the structures is found in the pyrophosphate anion binding site where the Arg-369 3 Cys substitution occurs (Fig. 2). In ␤ 3 , the Cys-369 side chain and two ordered water molecules occupy the space vacated by the side chain of Arg-369. These water molecules occupy similar positions and form the same hydrogen bonds as two of the guanidino nitrogens of Arg-369 (Fig. 2). The data suggest that the large changes in NAD ϩ and NADH binding affinity for ␤ 3 are not due to extensive structural rearrangements of local side chain conformations or movements of whole domains (Table III and Fig.  2), as was observed for ␤47Gly (14). ␤47Gly has no side chain at position 47 as a result of the Arg-47 3 Gly substitution. Domain closure may be favorable in ␤47Gly since Gly-47 cannot form electrostatic or hydrogen bonding interactions. The Arg-47 3 His substitution in ␤ 2 and the Arg-369 3 Cys sub- stitution in ␤ 3 are structurally more conservative than the Arg-47 3 Gly substitution in ␤47Gly because they do not cause major differences in the overall structure (Table III). Yet, from a functional viewpoint, the Arg-47 3 Gly substitution appears to be a more conservative substitution because it has only a 2-fold effect on coenzyme binding, relative to ␤ 1 (29).
In the ␤ 1 structure, the positively charged side chain of Arg-369 interacts with the negatively charged coenzyme pyrophosphate. The neutral thiol of Cys-369 of ␤ 3 cannot form a similar ion pair with a pyrophosphate oxygen. Thus, the difference in the coenzyme binding free energies between ␤ 3 and ␤ 1 isoenzymes may be due to the loss of the electrostatic interaction between Arg-369 and the coenzyme molecule. The difference in the free energy for NAD ϩ binding to ␤ 3 versus ␤ 1 (3.5 kcal/mol) is within the range of values calculated from the electrostatic differences in the enzyme⅐NAD(H) complexes alone (2.2-8.9 kcal/mol), using Coulomb's Law and a dielectric constant of either 40 or 10 (28). Solvent accessibility calculations show that Arg-369 of ␤ 1 and Cys-369 of ␤ 3 are not accessible to bulk solvent in ternary complexes. The extent to which these residues are exposed to solvent in their respective free enzyme structures and the free energy of solvation (⌬G solv ) of Arg or Cys will influence the magnitude of the electrostatic contribution to coenzyme binding. Thus, the structures of substrate-free ␤ 3 and ␤ 1 enzymes are required before a complete thermodynamic analysis of coenzyme binding can be performed.
The substitution of Arg-369 3 Cys affects NADH binding to a greater extent than NAD ϩ binding. The K d,NADH exhibited by ␤ 3 is 4000-fold higher than that for ␤ 1 , while the K d,NADϩ is only 350-fold higher (Table IV). This difference in affinity corresponds to a ⌬G of 1.4 kcal/mol. We propose that the differential effects on NAD ϩ versus NADH binding could be due to an electrostatic repulsion between the positively charged side chain of Arg-369 and the positively charged nicotinamide ring of NAD ϩ . This unfavorable interaction is missing in the ␤ 3 isoenzyme. Thus, the dissociation constants for NAD ϩ versus NADH differ by 14-fold in ␤ 3 , but the binding of NADH is favored by 150-fold over NAD ϩ in the ␤ 1 isoenzyme (Table IV).
An analysis of the changes in the apparent rate constants associated with coenzyme binding suggests that position 369 may be an important factor controlling the initial contact between the coenzyme and enzyme. This is supported by the 12-19-fold decrease in apparent coenzyme association rate constants of ␤ 3 versus ␤ 1 (k on in Table IV). The Arg-47 3 His substitution in ␤ 2 and the Arg-47 3 Gly substitution in ␤47Gly result in only 2-fold decreases in the apparent k on,NADH rates (29). Hence, Arg-47 may not be as important a primary determinant of initial contact with coenzyme as Arg-369. However, the loss of a positively charged residue at either position 47 or 369 is associated with large increases in the dissociation rate constant for NAD(H). The replacement of Arg-47 with Gln results in a 150-fold increase in the apparent dissociation rate constant for NADH (17), and in this study we show the replacement of Arg-369 with Cys results in a 330-fold increase in apparent dissociation rate constant for NADH. This data would suggest that once coenzyme is bound to the enzyme, Arg-47 contributes to the rate of coenzyme dissociation, most likely after the conformational change which occurs with coenzyme binding. The good correlation of k off,NADH with V max for ethanol oxidation in ␤ 1 and ␤ 2 (17), and not ␤ 3 (23 versus 7 s Ϫ1 in Tables I and IV) indicates that the His substitution at position 47 does not alter the rate-limiting step from that of coenzyme dissociation, but the Cys substitution at position 369 does alter the rate-limiting step. For ␤ 3 , the rate of coenzyme dissociation is approximately 3-fold faster than the V max value. Detailed studies of the individual rate constants along the ␤ 3 reaction pathway are necessary to further define which step(s) control the overall rate of alcohol oxidation.