X-ray Structure of Human Class IV ςς Alcohol Dehydrogenase

The structural determinants of substrate recognition in the human class IV, or ςς, alcohol dehydrogenase (ADH) isoenzyme were examined through x-ray crystallography and site-directed mutagenesis. The crystal structure of ςς ADH complexed with NAD+ and acetate was solved to 3-Å resolution. The human β1β1 and ςς ADH isoenzymes share 69% sequence identity and exhibit dramatically different kinetic properties. Differences in the amino acids at positions 57, 116, 141, 309, and 317 create a different topology within the ςς substrate-binding pocket, relative to the β1β1 isoenzyme. The nicotinamide ring of the NAD(H) molecule, in the ςς structure, appears to be twisted relative to its position in the β1β1isoenzyme. In conjunction with movements of Thr-48 and Phe-93, this twist widens the substrate pocket in the vicinity of the catalytic zinc and may contribute to this isoenzyme’s high K m for small substrates. The presence of Met-57, Met-141, and Phe-309 narrow the middle region of the ςς substrate pocket and may explain the substantially decreased K m values with increased chain length of substrates in ςς ADH. The kinetic properties of a mutant ςς enzyme (ς309L317A) suggest that widening the middle region of the substrate pocket increases K m by weakening the interactions between the enzyme and smaller substrates while not affecting the binding of longer alcohols, such as hexanol and retinol.

Human alcohol dehydrogenase (ADH) 1 isoenzymes are NAD ϩdependent, zinc metalloenzymes that catalyze the reversible oxidation of alcohols to aldehydes. The ADH system is the major pathway for the metabolism of beverage ethanol as well as biological important alcohols or aldehydes like retinol, 3␤hydroxysteroids, -hydroxy fatty acids, and 4-hydroxynonenal (1)(2)(3). Each isoenzyme in the ADH family is a dimer comprised of two 40-kDa subunits. The individual subunits are comprised of two domains, a catalytic domain and a coenzyme-binding domain (4). Seven human ADH genes (ADH1-ADH7) have been identified (1,5). The ADH1-ADH5 genes encode the ␣, ␤, ␥, , and subunits, respectively. The protein product of the ADH6 gene has not been identified in vivo. The subunit is encoded by ADH7. Polymorphism occurs at both the ADH2 (␤ 1 , ␤ 2 , and ␤ 3 ) and ADH3 (␥ 1 and ␥ 2 ) loci (6), such that nine distinct human ADH subunits have been identified. ADH isoenzymes have been assigned to five distinct classes based on their amino acid sequences as well as their electrophoretic and enzymatic properties (7). The human ␣␣, ␤␤, and ␥␥ isoenzymes comprise class I, and the , , , and ADH6 comprise classes II, III, IV, and V, respectively. All ADH isoenzymes are expressed in the liver except for ADH, which is primarily localized in epithelial tissue, such as the stomach mucosa (8,9).
The three-dimensional structures of horse and human class I ADHs have been solved by x-ray crystallography (4, 10 -12). Recently, the structure of human class III ADH was reported (13), as well as the structure of a cod liver ADH isoenzyme (14). Thus, an increasingly diverse structural data base exists from which information concerning the determinants of substrate recognition can be obtained by comparing the structures and kinetic properties of ADH isoenzymes. Important amino acids within the substrate-binding site directly affect the substrate specificity of the human ADH isoenzymes (Table I). For instance, ␥ 1 ␥ 1 ADH, which has a Ser at residue 48, is the only human isoenzyme able to bind and oxidize 3␤-hydroxysteroids (15). Amino acid substitutions within the loops comprised of residues 55-61 and of residues 113-121 in ADH cause these loops to adopt new conformations and contribute to the enzyme's inability to be saturated with ethanol (13). Mutagenesis studies on the ␤ 1 ␤ 1 isoenzyme indicate that residues 93 and 94 account for the increased catalytic efficiency toward secondary alcohols exhibited by ␣␣, relative to ␤ 1 ␤ 1 (16). Residue 116, located at the entrance to the substrate pocket, also affects the K m for alcohols by acting as a bottleneck (17).
Among the human class I enzymes, ␤ 1 ␤ 1 exhibits the lowest K m for ethanol (18), 120-and 7-fold lower than ␣␣ and ␥ 1 ␥ 1 , respectively. From ethanol to hexanol, the catalytic efficiency (V max /K m ) of the ␣␣ isoenzyme increases 400-fold, while it increases just slightly more than 2-fold for the ␤ 1 ␤ 1 isoenzyme (18). Compared with class I ␤ 1 ␤ 1 ADH, class IV ADH also exhibits very different substrate binding characteristics. The K m for ethanol exhibited by ADH is 215-fold higher than that for hexanol (9). In addition, ADH exhibits the highest catalytic efficiency for the oxidation of all-trans-retinol to alltrans-retinal among the known human ADH isoenzymes (19). The production of all-trans-retinal from retinol is thought to be the rate-controlling step for the production of all-trans-retinoic acid (20), an important regulator of gene expression during embryonic development (21). The pathway of retinoic acid biosynthesis involves retinoid-binding proteins, which may provide a mechanism to discriminate specific dehydrogenases from nonspecific dehydrogenases (22). Retinol, in vivo, is bound to the cellular retinol-binding protein (CRBP). Evidence has been presented showing that holo-CRBP serves as substrate for microsomal dehydrogenases (22) and that CRBP may then transfer retinal to the cytosolic retinal dehydrogenase for oxidation to retinoic acid. Complete dependence on the CRBP pathway for retinoic acid production may deny accessibility of retinol to ADH in vivo. However, retinoic acid synthesis during embryogenesis was reported to correlate spatiotemporally with the expression of class IV ADH gene (23). It was proposed that competitive inhibition by ethanol consumed during pregnancy can reduce retinoic acid synthesis and may contribute to the development of fetal alcohol syndrome (24,25).
In this paper, we examine the structural basis for substrate recognition in ADH through x-ray crystallography and sitedirected mutagenesis. By comparing the structures of the known human ADH substrate-binding sites, it may be possible to gain a more complete understanding of their roles in the metabolism of endogenous and exogenous alcohols.

EXPERIMENTAL PROCEDURES
Protein Purification and Crystallization-The cDNA for the subunit (5) in M13 was subcloned into the vector pKK223-3 (Pharmacia Biotech Inc.) by site-directed mutagenesis using a commercial kit (Amersham Corp.) and expressed in Escherichia coli as described for ␤ 1 ␤ 1 ADH (10). The lysate was first mixed with DEAE-cellulose (Whatman, Maidstone, UK) in 50 mM Tris, pH 8.8, at 4°C, 1 mM benzamidine, 2 mM dithiothreitol. The unbound proteins were eluted in a batch procedure and then were buffer exchanged into 7 mM sodium phosphate, pH 6.4, 1 mM DTT using the Minitan apparatus (Millipore, Bedford, MA) and loaded onto a 5-ϫ 15-cm S-Sepharose column. The protein was eluted with a linear sodium phosphate gradient from 7 to 65 mM. The enzyme was dialyzed into 10 mM sodium phosphate, pH 6.4, 1 mM DTT and applied to a 2.5-ϫ 10-cm Affi-Gel Blue column (Bio-Rad). The enzyme was then eluted with a linear gradient from 10 mM sodium phosphate, pH 6.4, to 100 mM Tris, pH 8.8, 1 mM DTT. The purified ADH was dialyzed into 10 mM HEPES, pH 7.0, 1 mM DTT and concentrated with a Microcon 30 concentrator (Amicon, Beverly, MA) before crystallization. The sitting drop method was employed to crystallize the protein. Typically 2 l of an 8 mg/ml ADH solution was mixed with 2 l of the precipitant solution in the drop. The optimized crystallization conditions for ADH complexed with NAD ϩ were 100 mM cacodylate, pH 6.5, 50 -100 mM zinc acetate, 7.5 mM NAD ϩ , and 18% (w/v) polyethylene glycol 6000. The crystals formed as flat parallelepipeds overnight and grew to maximal size in 1 or 2 more days.
Mutagenesis and Kinetic Studies-Single-stranded cDNA in the M13HinEco1 vector (26) was used as the template for site-directed mutagenesis. A single oligonucleotide, 45 bases in length, was used to mutate residues Phe-309 and Cys-317 to Leu and Ala, respectively. Following identification of the correct mutant clone by DNA sequencing, the mutant cDNA was subcloned into pKK223-3 and completely sequenced, prior to expression, to ensure that no unwanted mutations were present in the cDNA sequence. The mutant enzyme, 309L317A, was expressed and purified using the same procedure as described for the wild-type enzyme. The kinetic measurements were evaluated at 25°C in 100 mM sodium phosphate, pH 7.5, on a Beckman DU-640 spectrophotometer. Enzyme activity was monitored by following the production of NADH at 340 nm using an extinction coefficient of 6.22 mM Ϫ1 cm Ϫ1 . V max values were converted to turnover numbers assuming a molecular mass of 40 kDa per subunit. The K m values for substrates were determined at a fixed NAD ϩ concentration of 2.5 mM, except those for 1-butanol which were determined both at 2.5 mM NAD ϩ and by co-variation of NAD ϩ and 1-butanol. All kinetic experiments were evaluated using the kinetic programs of Cleland (27). All reported values are expressed as the means of at least three separate experiments with their associated standard deviations.
X-ray Diffraction Data Collection-X-ray diffraction data were collected to 3 Å. Higher resolution data were observed initially (2.6 Å), but severe radiation decay and the inability to flash-cool these crystals prevented collection of the higher resolution data. Four crystals (approximate dimensions, 0.3 ϫ 0.15 ϫ 0.07 mm 3 ) were used to collect the native data set at room temperature on a Rigaku 200HB rotating anode generator equipped with an RAXIS IIC image plate area detector with a crystal-to-detector distance of 145 mm. The data collection statistics are listed in Table II. All crystals exhibited radiation decay and were replaced every 12 h. The diffraction data were indexed, merged, and scaled using the RAXIS IIC data processing software (38).
Molecular Replacement and Crystallographic Refinement-The structure was solved by molecular replacement using the program package AMoRe (28) and the data between 15.0 and 4.0 Å. The ␤ 1 ␤ 1 ADH ternary complex dimer with NAD ϩ and the inhibitor 4-iodopyrazole (Protein Data Bank code 1DEH (13)) served as the search model for these calculations. The correlation coefficients for the top two rotation solutions were 25.7 and 19.7, respectively. After the positions for the two dimers in the asymmetric unit were found, the starting model possessed a correlation coefficient of 54.7 and an R value of 39.9%. All subsequent model refinement was performed using the program package X-PLOR (version 3.1) (29). Rigid-body refinement of the initial model structure with the data between 8.0 and 3.5 Å brought R work from 40.2 to 37.7% and R free from 40.5 to 37.3%. The atomic positions were refined to 3 Å using the positional refinement protocol in X-PLOR (30) and an overall temperature factor of 25 Å 2 . The resulting structure was inspected using 2F o Ϫ F c and F o Ϫ F c maps in CHAIN (31). Amino acid substitutions were introduced as their positions were identified during refinement. Additional solvent zinc cations and solvent acetate molecules were added when strong positive F o Ϫ F c electron density indicated their presence. In the last refinement procedure, an overall temperature factor for each subunit was refined, and the noncrystallographic symmetry restraints were removed. The final model possesses an average r.m.s.d. of 0.2 Å for the main chain atoms in the four subunits of the asymmetric unit. C␣ alignments between and ␤ 1 ␤ 1 isoenzymes were performed using LSQKAB (32) in CCP4 (1994) suite and displayed using QUANTA (Molecular Simulations Inc., Burlington, MA).
Reagents-NAD ϩ , grade I and DTT were purchased from Boehringer Mannhein, and PEG 6000 was purchased from Hampton Research; ethanol was purchased from Midwest Grain (Pekin, IL). All other reagents were from Sigma and were of the highest grade available.

RESULTS
Structure Determination-The structure of the human class IV, or , ADH isoenzyme was solved to 3.0 Å by molecular replacement using the 2.2-Å structure of the class I human ␤ 1 ␤ 1 isoenzyme (12) as the starting model. The final refined structure possesses an R work of 22.5% with an R free of 30.5% (Table II). The stereochemistry of this model was inspected using the program package PROCHECK (33). The Ramachadram plot showed that 98.6%, or 1471, of the 1492 residues were in the preferred and allowed regions, and 1.4%, or 21, of the residues were in the generously allowed region. No nonglycine residues were found in the disallowed region. Due to the presence of zinc acetate in the mother liquor, 8 solvent zinc cations and 10 acetate ions were identified as bound to the enzyme. There was an acetate ion present in the substrate pocket in all four subunits.
Kinetics of the ADH Mutant-A ADH mutant, 309L317A, was prepared by site-directed mutagenesis. Two residues in the isoenzyme, Phe-309 and Cys-317, were mutated to Leu and Ala, respectively. The choice of these two positions for mutagenesis was based on their unique characteristics compared with class I enzymes (Table I). Residue 309 is in the substrate-binding pocket, and residue 317 is behind the nicotinamide ring of NAD ϩ . The substrate specificity of this mutant was studied and compared with wild-type ADH (Table III). Mutations at these two residues dramatically increase the K m values toward small alcohol substrates. For instance, the mutant enzyme exhibits a K m for ethanol that is 100-fold higher than the wild-type ADH. Interestingly, the K m values for substrates with five or more carbons are less affected. The K m values for hexanol and retinol are essentially unaffected by these mutations. As the chain length for straight alcohols increases, the V max /K m of the mutant enzyme increases to a greater extent than does the wild-type enzyme. The K m value for NAD ϩ was approximately 2 times higher than the wild-type enzyme, whereas the K i (NAD ϩ ) value obtained from experiments varying both 1-butanol and NAD ϩ concentrations was identical with the wild-type enzyme (0.75 Ϯ 0.03 mM).
Structural Comparison with Other ADH Structures-An alignment of the C␣ atoms, excluding residues 113-120 and 244 -262, in the dimeric and ␤ 1 ␤ 1 isoenzymes gives a r.m.s.d. of 0.60 Å. Alignment of the individual domains within each subunit yields similar results, with r.m.s.d. values of 0.49 Å for the catalytic domain and 0.61 Å for the coenzyme-binding domain. C␣ alignment of ADH with horse and other human ADHs shows that its structure is most similar to the human and horse liver class I ADH isoenzymes. Unlike the recently reported structures of the human class III isoenzyme (13) and the ADH isoenzyme from cod liver (14), both of which exhibited semi-open domain structures, the human isoenzyme exhibits a fully closed conformation of the catalytic and coenzyme-binding domains when NAD(H) is bound. The alignments reveal that, relative to ␤ 1 , there are two major structural differences in each domain of the subunit (Fig. 1). In the coenzyme binding domain, the largest difference occurs at the C terminus of an ␣-helix comprised of residues 251-258 and the following turn. This structural change results from the substi-tutions of the two Gly residues at positions 260 and 261 in ␤ 1 ␤ 1 by two Asn residues in ADH. The main chain conformations of Gly-260 and -261 in the ␤ 1 ␤ 1 isoenzyme are incompatible with the presence of side chains, thus the main chain path must shift dramatically to accommodate Asn residues at positions 260 and 261. Relative to ␤ 1 ␤ 1 ADH, the C␣ atoms of residues 244 -262 in ADH shift, on average, by 1.8 Å, with a maximum of 4.8 Å at residue 261. The other structural difference in the coenzyme-binding domain involves residues 297-309, which are located at the dimer interface and within the substrate-binding pocket of the neighboring subunit. One structural difference in the catalytic domain involves residues 17-25 and is not likely to affect enzymatic activity or subunit interactions. The other structural difference within the catalytic domain is due to the deletion of residue 117, which shortens the loop at the entrance to the substrate-binding pocket in ADH. C␣ alignment between and , with each domain aligned separately and then the results combined, reveals that in addition to those differences between and ␤ 1 ␤ 1 , and differ substantially at both N and C termini (Fig. 1). Moreover, great differences exist at residues 55-61 and 112-120. Both regions adopt new conformations in ADH and contribute to FIG. 1. Comparison of , , and ␤ 1 ␤ 1 dimers by alignment of their C␣ atoms. All C␣ atoms were used in the alignment. The result of the alignment using the CD dimer of ADH is shown. The other dimer of ADH in the asymmetric unit gives similar results.

TABLE II
Data collection and molecular refinement statistics R merge ϭ ⌺ hkl ͉I i Ϫ I n ͉/⌺I n where I i is an observed intensity and I n is the average of the observed equivalents. R work ϭ ⌺ hkl ʈF obs ͉Ϫ͉F cal ʈ/⌺ hkl ͉F obs ͉ where ͉F obs ͉ and ͉F cal ͉ are the observed and calculated structure factor amplitudes of a reflection hkl, respectively. R free is the same as R work except that the summation is over the portion of data (7%) that is not included in the refinement.  a larger active site in ADH (13). Our structural comparisons also reveal that there is no evidence that the interaction between coenzyme and residue 223 is weakened in ADH, as was suggested by a modeling study (34). In fact, the hydrogen-bonding distances between adenosine ribose oxygens and ␥ oxygens of Asp-223 are within the range of 2.6 -2.7 Å in both and ␤ 1 ␤ 1 ADHs. DISCUSSION The alcohol binding pocket is an extension of the coenzyme binding site (4) and is fully formed only after coenzyme binding has occurred. In ADH, the substrate pocket is a cylinder having dimensions of approximately 16 by 7 by 6 Å. The substrate specificity for this enzyme is determined by surface complementarity between the enzyme and the substrates throughout this cylinder. Changes in the K m values are related to the effective concentration of the ES complex. Mutations can affect K m by changing the ratio of productive versus non-productive encounters with the enzyme. In ADH, these changes are brought about either through steric exclusion (preventing productive binding), as seen for the binding of secondary alcohols to ␤ 1 ␤ 1 ADH (16), or by changing the number of nonproductive conformations permitted by altering the accessible volume of the active site (13). The inner part of the alcohol site (near the catalytic zinc) includes residues 48 and 93 and the nicotinamide ring of NAD ϩ . ADH has a K m value for ethanol that is 560-fold higher than ␤ 1 ␤ 1 ADH (Table III). One possible cause for this difference may be the substitution of Cys for Ala-317 near the nicotinamide ring. To accommodate its longer side chain in ADH, the main chain atoms of residue 317 move ϳ1 Å away from Thr-186, toward the carboxamide group relative to ␤ 1 ␤ 1 ADH (Fig. 2). To avoid unfavorable contacts with the Cys-317 carbonyl oxygen, the plane of nicotinamide ring appears to twist in ADH, relative to its position in the ␤ 1 ␤ 1 isoenzyme (Figs. 2 and 3), This twist creates more space between the nicotinamide ring and the catalytic zinc in ADH (Fig. 3). In addition to these changes, Thr-48 and Phe-93 also shift away from the catalytic zinc. The distance between the C␣ atoms of these two residues is 0.9 Å longer in ADH. Consequently, smaller substrates, such as ethanol, are not as conformationally constrained in this active site as in the ␤ 1 ␤ 1 ADH. Thus, a higher concentration of ethanol is required to produce an equivalent number of productively bound conformations.
The structural differences near the catalytic zinc in these two isoenzymes may also explain the weak binding of the inhibitor pyrazole to ADH (K i values of 0.60 M for ␤ 1 ␤ 1 and 350 M for at pH 7.5 (5,12)). Pyrazole and its 4-substituted derivatives competitively inhibit the binding of alcohol substrates through the formation of a tight enzyme⅐NAD ϩ ⅐inhibitor complex (35), in which pyrazole nitrogens interact with both zinc and NAD ϩ . We speculate the bond between the pyrazole nitrogen atom and the C-4 atom on the nicotinamide ring may be distorted due to the twist of the nicotinamide ring in ADH. In fact, if the active sites of the and ␤ 1 ␤ 1 structures are aligned and the position of 4-iodopyrazole in the ␤ 1 ␤ 1 active site structure is used to examine the geometric constraints on pyrazole binding to the structure, the corresponding angle between the C-4 -N-1 and the N-1-N-2 bond is 133°, while it is close to 120°in the ␤ 1 ␤ 1 structure, corresponding to a low energy, stable complex. This angular difference would undoubtedly represent a higher energy conformation and could account for up to 2.7 kcal/mol of the observed difference (3.8 kcal/mol) using a harmonic potential with a force constant of 0.27 kcal/(mol⅐degree) (36). In addition, the increased distances between the N-1 of pyrazole and the C-4 of the nicotinamide ring (by 0.3 Å) and between residues 48 and 93, where pyrazole is held, could contribute to lowering the affinity for 4-methylpyrazole.
The middle region of the substrate pocket, which plays an important role in the interactions with the aliphatic tail of longer substrates, such as butanol and pentanol, includes residues 57, 141, 294, and 309. Like many other ADH isoenzymes, ADH exhibits K m values for primary straight chain alcohols, which decrease with increasing chain length, whereas the V max values remain relatively constant. Thus, the catalytic efficiency (V max /K m ) increases with increasing chain length. For example, the V max /K m for hexanol is 138-fold higher than that for ethanol in ADH (Table III). In contrast, the V max /K m values for ␤ 1 ␤ 1 ADH vary only 2-to 3-fold for substrates from ethanol to hexanol. These different characteristics can be explained by differences in the amino acids within the middle region of the substrate pocket. The key amino acids within this region are residues 57, 141, and 309. ␤ 1 ␤ 1 ADH possesses Leu residues at all these positions, and their side chains do not appear to create new productive interactions as substrates get longer (Table  III). The presence of Phe at position 309, Met at position 141, and Met at position 57, in ADH, narrows the middle region of the substrate pocket compared with ␤ 1 ␤ 1 (Fig. 3). The twist on the nicotinamide ring also contributes to a shift in Phe-309, to avoid unfavorable contact with NAD(H), which further narrows the channel leading to the catalytic zinc. Although this narrowing does not appear to directly aid the binding of ethanol, it can explain the decreased K m values for propanol, butanol, and pentanol in the isoenzyme relative to ␤ 1 ␤ 1 ADH.
Our modeling indicated that the side chain of residues 57 and 309 would interact with the substrates at the carbon 4, 5, and 6 positions, whereas the side chain of residue 141 interacts with carbon 3 and 4. In the 309L317A mutant, the K m values for propanol, butanol, and pentanol are increased by 240-, 60-, and 7-fold, confirming the role of residues 309 and 317 in stabilizing the binding of these substrates. Moreover, the increase in V max /K m versus the chain length of the substrates is greater for the mutant than for the wild-type enzyme. This behavior in the mutant enzyme is due to a much lower V max /K m for small substrates, since the V max /K m values for hexanol approach those of the wild-type enzyme. The substitutions in the mutant enzyme thus appear to further widen the substratebinding site, relative to ADH, resulting in a greater number of permissible non-productive ES complexes. These changes in the mutant only affect substrates where binding is dependent on the local topology, such as ethanol, but do not significantly affect the catalytic efficiency toward hexanol or retinol.
The outer part of the substrate-binding pocket is exposed to solvent and includes the loop comprised of residues 114 -120 and residue 306 from the other subunit within the dimer. The deletion of residue 117 in ADH shortens the loop comprised of residues 114 -120 (Fig. 3) and widens the entrance to the substrate-binding site. In ␤ 1 ␤ 1 ADH, Leu-116 appears to function as a door, opening to allow substrates in or out, but then closing to help keep in bound substrate (17). Consistent with this function, its side chain was found to occupy different conformations in binary and ternary complexes (10 -12). In ADH, the shift in the position of residue 116 due to the deletion of residue 117 does not permit its side chain to function in this manner, leaving an open substrate-binding site (Fig. 3). Widening of the bottleneck at position 116 in ␤ 1 ␤ 1 ADH by mutagenesis dramatically increased the apparent K m values for primary and secondary alcohols (17). Consistent with these observations, ADH has higher K m values than the ␤ 1 ␤ 1 isoenzyme for straight chain alcohols and very poor efficiency toward all secondary alcohols (5). The K m for hexanol and retinol are virtually identical in the wild-type and mutant enzymes. With the enlarged entrance to the substrate pocket in the isoenzyme, the structure of the middle and inner regions of the substrate pocket would seem to be best suited for the oxidation of long chain aliphatic alcohols, such as -hydroxy fatty acids, farnesyl alcohols, and retinol. To examine long chain alcohol binding, all-trans-retinol was docked into the active site using program AUTODOCK (37). The results of this simulation confirm our previous results based on modeling studies (5), the ␤-ionone ring of retinol binds at the widened entrance of the substrate-binding pocket, such that an ex- FIG. 3. Comparison of the active site in and ␤ 1 ␤ 1 ADH isoenzymes. a, superposition of the active site between two isoenzymes. The alignment is performed on the C␣ of the coenzyme binding domain excluding residues 244 -262. b, the accessible surfaces of the substrate pocket in (in dark dots) and ␤ 1 ␤ 1 ADHs. This orientation is approximately perpendicular to that in a, with the catalytic zinc at the left and the entrance at the right. A hexanol molecule was introduced artificially to indicate the approximate size of the pocket. The approximate side chain positions of two residues are also indicated. tended conformation of retinol can be adopted. Thus, the ability to bind retinol in a more extended and, presumably, lower energy conformation in ADH could account for its higher catalytic efficiency.