Crystal Structure of Pseudomonas fluorescens Mannitol 2-Dehydrogenase Binary and Ternary Complexes SPECIFICITY AND CATALYTIC MECHANISM*

Long-chain mannitol dehydrogenases are secondary alcohol dehydrogenases that are of wide interest because of their involvement in metabolism and potential applications in agriculture, medicine, and industry. They differ from other alcohol and polyol dehydrogenases because they do not contain a conserved tyrosine and are not dependent on Zn 2 (cid:1) or other metal cofactors. The structures of the long-chain mannitol 2-dehydro-genase (54 kDa) from Pseudomonas fluorescens in a binary complex with NAD (cid:1) and ternary complex with NAD (cid:1) and D -mannitol have been determined to resolu-tions of 1.7 and 1.8 Å and R -factors of 0.171 and 0.176, respectively. These results show an N-terminal domain that includes a typical Rossmann fold. The C-terminal domain is primarily (cid:2) -helical and mediates mannitol binding. The electron lone pair of Lys-295 is steered by hydrogen-bonding interactions with the amide oxygen of Asn-300 and the main-chain carbonyl oxygen of Val-229 to act as the general base. Asn-191 and Asn-300 are involved in a web of hydrogen bonding, which precisely orients the mannitol O2 proton for abstraction. These residues also aid in stabilizing a negative charge in the intermediate state and in preventing the formation of nonproductive complexes with the substrate. The catalytic

Mannitol, a six-carbon non-cyclic polyol, is the most abundant sugar alcohol found in nature. Produced in plants, fungi, protozoa, and bacteria as a storage compound for carbon and reducing equivalents, it also functions in response to oxidative stress and as an osmoregulator (1)(2)(3)(4)(5). D-Mannitol is used extensively in the food and pharmaceutical industry because of its favorable bulking properties and the fact that it does not cause tooth decay and is safe for diabetics. The traditional method of industrially producing mannitol involves the reduction of fructose using a metal catalyst and hydrogen gas, resulting in nearly equal amounts of D-sorbitol and D-mannitol, which must then be separated.
In general, mannitol dehydrogenases (MDH) 1 catalyze the NAD(P) ϩ -dependent reversible oxidation of D-mannitol or D-mannitol-1-phosphate to the corresponding ketose, D-fructose, or D-fructose 6-phosphate. These secondary alcohol dehydrogenases are specific for the C2(R) configuration of polyhydroxylated compounds and are of interest because of their potential applications in chiral synthesis. More recently, mannitol dehydrogenases have been identified in plants that catalyze the oxidation of D-mannitol to D-mannose, an aldose (6). MDHs have been characterized from plants and fungi that are members of the medium-chain zinc-containing dehydrogenase/ reductase family (7,8). Other MDH from fungi are members of the short-chain dehydrogenase/reductase (SDR) family (3,9). Often, bacterial MDHs do not share significant similarity with either of these families (10) but instead belong to a family of long-chain MDH that so far includes 54 recognized members. These members are classified by the Protein Families Data Base as family 01232 (11). More recent work increased this number to 66. 2 Sequence identity with other long-chain dehydrogenases is low, typically around 10%.
Most members of the prokaryotic long-chain MDH family have been identified by primary sequence alone (11), and a limited number of these proteins have had their specificity characterized and activity quantitated. The proteins that have been studied are monomeric long-chain dehydrogenases of a molecular mass of ϳ54 kDa (10,13,14). This family so far includes mannitol 2-dehydrogenases, mannitol-1-phosphate 5-dehydrogenases, D-mannonate dehydrogenases, sorbitol dehydrogenases, L-sorbose reductase, fructuronate reductase, altr-onate oxidoreductases, and D-arabinitol dehydrogenases. In addition to their significance as an alcohol dehydrogenase that employs a novel mechanism, interest in long-chain MDHs originates from several potential uses: (i) transgenic expression of bacterial MDHs in plants has been tested to improve salt tolerance and resistance to oxidative stress in agricultural crops (15), (ii) quantitative analysis of mannitol concentration in serum and urine via a simple and sensitive enzymatic assay has potential clinical use (16), and (iii) enzymatic production of D-mannitol from fructose would reduce downstream purification (17).
An inducible mannitol 2-dehydrogenase belonging to the long-chain MDH family was isolated from Pseudomonas fluorescens DSM50106 (pfMDH, EC 1.1.1.67) (18). It catalyzes the reversible oxidation of D-mannitol to D-fructose, D-arabinitol to D-xylulose, and D-sorbitol to L-sorbose by transferring the C2 hydride to the pro-S position on the nicotinamide without the use of metal cations. It is specific for the C2(R) configuration of polyols with a minimum of five carbons, and no activity is measurable with mannitol 1-phosphate, fructose 6-phosphate, or 5,6-dideoxy-D-fructose. 3 How specificity for polyol substrates is achieved is not well understood. Although able to use both NADH and NADPH as cofactor, the activity with NADH is greater. Other alcohol dehydrogenases use metal ions (medium-chain dehydrogenase/reductases and some long-chain dehydrogenases) or a conserved tyrosine (SDRs) for catalysis. Neither is present in long-chain MDHs, so presumably, these alcohol dehydrogenases use a novel catalytic mechanism. Biochemical data implicate an enzyme side-chain with a pK a of 9.34 (13). This study was undertaken to gain an understanding of how specificity is achieved for substrate and cofactor as well as elucidate the mechanism for catalysis in the structurally uncharacterized family of long-chain MDHs.

MATERIALS AND METHODS
Expression and Purification-Recombinant wild-type MDH from P. fluorescens DSM 50106 was expressed in Escherichia coli and purified as previously described (13). The selenomethione-substituted protein was expressed in the presence of 60 mg/liter selenomethionine accompanied by amino acids inhibiting de novo synthesis of methionine (19). Purification was carried out using a protocol similar to the previously described wild-type preparation (13). A final gel filtration step was included to obtain highly purified protein. Gel filtration was carried out on an Ä ktaexplorer 100 system (Amersham Biosciences) using 140-ml Superdex 75 prep-grade material packed into a 1.6/70-cm column. Approximately 15 mg of protein in 50 mM Tris, pH 7.2, were applied to the column equilibrated with 50 mM Tris, 200 mM NaCl. The protein was eluted at a flow rate of 0.75 ml/min. Fractions containing enzyme activity were pooled and concentrated to ϳ11 mg/ml. The purified protein migrated as a single band in SDS-PAGE and non-denaturing anionic PAGE. Semiquantitative densitometric analysis of the Coomassie Blue-stained gels suggested that the purity of pfMDH was 99% or better. The selenium-substituted protein retains full wild-type activity.
Crystallization-Wild-type pfMDH was concentrated to 14 mg/ml, and the buffer changed to 10 mM Tris, 25 mM NaCl, pH 7.5. Hanging drop vapor diffusion experiments were duplicated at 277 and 293 K using both apoprotein and protein solution containing 5 mM NADH. Initial crystals of a binary complex with NADH took five months to appear at 293 K over a well containing 30% (w/v) polyethylene glycol 4000, 200 mM ammonium acetate, 100 mM sodium citrate, pH 5.6. A single crystal was flash-cooled in a buffer containing 75% (v/v) well solution and 25% (v/v) ethylene glycol. Diffraction intensities to 2.5 Å were collected at Stanford Synchrotron Radiation Laboratory (SSRL) beamline 9Ϫ1 and processed using the program Denzo (20). This indicated that the spacegroup was I222 with unit cell dimensions of a ϭ 102 Å, b ϭ 103 Å, c ϭ 107 Å. A Matthews' constant (V m ) of 2.58 Å 3 /dalton implied one monomer/asymmetric unit.
Because a suitable molecular replacement model could not be identified and the original protein had degraded, selenomethione-substituted protein was used to grow additional crystals. Crystallization experiments were conducted around the previously determined condi-tions by mixing 1 l of protein solution and 1 l of well solution. Crystals of size 0.1 ϫ 0.1 ϫ 0.2 mm took three months to appear from a protein solution containing 14 mg/ml selenomethionine pfMDH, 25 mM NaCl, 5 mM NADH, 50 M EDTA, 10 mM dithiothreitol, 10 mM Tris, pH 7.5, over a well of 34% (w/v) polyethylene glycol 4000, 250 mM ammonium acetate, 10 mM dithiothreitol, 100 mM sodium citrate, pH 5.0. These crystals were used for seeding, and square rods of size 0.15 ϫ 0.15 ϫ 0.5 mm were obtained within 2 days. Crystals were flash-cooled in a buffer containing 25% (v/v) glycerol, 75% (v/v) well solution. Lattice constants of a ϭ 102.23 Å, b ϭ 103.28 Å, and c ϭ 106.57 Å were observed, similar to those found in the wild-type crystals.
Several attempts to get mannitol bound in the active site were made. Ultimately, a protein solution of 14 mg/ml selenomethionine-substituted pfMDH, 25 mM NaCl, 10 mM D-mannitol, 5 mM NAD ϩ , 50 M EDTA, 10 mM Tris, pH 7.5, was used to grow seeded crystals over identical well conditions. Initially, these crystals were flash-cooled in a buffer containing 300 mM mannitol, 1 mM NAD ϩ , 36% (w/v) polyethylene glycol 4000, 250 mM ammonium acetate, 100 mM sodium citrate, pH 5.0. Electron density maps calculated using data collected from these crystals suggested incomplete incorporation of mannitol. Data used for refinement of the ternary complex were obtained using crystals soaked in this buffer for 12 h before flash cooling.
Structure Determination-The binary complex multiwavelength anomalous dispersion (MAD) data set was collected at SSRL beamline 9-2. A three-wavelength MAD experiment was conducted with a high energy reference after performing a fluorescence scan to determine peak and inflection points (Table I). Diffraction intensities were processed with Denzo and Scalepack (20). All eleven of the expected selenium sites were determined using the program Solve that resulted in a figure of merit of 0.468. After density modification executed by Resolve, the figure of merit was 0.638 and maps were calculated to 2.3-Å resolution (21). The density for protein and cofactor was easily interpretable, and the model was built using the program O (22). A data set to 1.7-Å resolution was subsequently collected on a binary crystal at SSRL beamline 9-2. The crystals were isomorphous, and the model built from the MAD data was used as starting model. Diffraction intensities to 1.8 Å were collected on an MDH⅐NAD ϩ ⅐mannitol complex crystal at SSRL beamline 9-1. The program Denzo was used to determine a primitive orthorhombic lattice of dimensions of a ϭ 107.0 Å, b ϭ 104.5 Å, and c ϭ 101.5 Å. Systematic extinctions indicated the that the spacegroup was P2 1 2 1 2. The calculation of the Matthews' constant (V m ) of 2.6 Å 3 /dalton implied two protein molecules/asymmetric unit. Molecular replacement using the holo-MDH structure stripped of water molecules as a search model was implemented using the program EPMR (23). An initial solution was found for two molecules using data between 30 and 4 Å, which yielded a correlation coefficient of 0.603 and an initial R cryst ϭ 0.41 in this resolution range.
In both cases before refinement commenced, 5% of the data was flagged for calculation of R free . Alternating rounds of manual fitting and crystallographic refinement using the programs O and CNS (24) resulted in the final structures of which statistics appear in Table II. Ordered water molecules were picked in CNS and manually checked for appropriate hydrogen bonding. Despite the inclusion of 1 mM NAD ϩ in the mannitol soak, the density of the NAD ϩ indicated occupancy of Ͻ1. Because it is not possible to accurately refine occupancy and temperature factors at this resolution, values between 0.5 and 1.0 were tested and the occupancy for the NAD ϩ in the ternary model was set to 0.8, a value that gave both reasonable temperature factors and minimized features in difference maps.

RESULTS AND DISCUSSION
Structure Determination and Model Quality-The structure of the binary complex of pfMDH with NADH was initially determined to 2.3 Å resolution by MAD using selenomethionesubstituted protein (Table I). A data set to 1.7 Å was subsequently collected and used for refinement (Table II). Crystals of a ternary complex of pfMDH with NADH and mannitol were obtained by including 10 mM D-mannitol in the protein solution and soaking the crystals for 12 h in a solution containing 300 mM mannitol. Diffraction intensities to 1.8 Å resolution were collected on a ternary crystal (Tables I and II).
In both cases, the model for MDH includes residues 1-492. The final C-terminal residue was disordered and not fit. The holo model includes 1 pfMDH monomer, 1 NADH molecule, and cludes 2 pfMDH protein chains, 2 NAD ϩ molecules with 80% occupancy, and 630 water molecules. Root mean square deviations between holo-and ternary models are Ͻ0.8 Å for ␣-carbons and 1.0 Å for all atoms. A Ramachandran plot as implemented by Procheck for the binary model indicates that 89.6% of the residues lie in the most favored regions, 9.9% in additional allowed regions and 0.2% in generously allowed regions. A Procheck Ramachandran plot for the ternary model indicates that 90.8% of the residues are in the most favored regions and 9.0% in additional allowed regions (25). In both the binary and ternary models, residue Thr-132 was the single exception and was always located in a disallowed region. This residue is located in a loop involved in cofactor binding, and density clearly indicates it is in this conformation. Consistently low temperature factors of Thr-132 and clear 2F o Ϫ F c density provide clear evidence that it is very well ordered.
Overall Structure-The structure reveals that pfMDH folds into two main domains (Fig. 1). Overall dimensions are 65 ϫ 45 ϫ 60 Å with the active site located at the bottom of a cleft between the two domains that is 12 ϫ 11 Å wide and 23 Å deep. The structure confirms the biochemical data in which there are no metal ions involved in catalysis (13). The N-terminal domain (domain 1) includes 9 ␣-helices, 14 ␤-strands, and 3 3 10 -helices (Fig. 2). The largest ␤-sheet is a six-stranded parallel dinucleotide binding motif commonly known as a Rossmann fold. A tetra-peptide linker of sequence Thr-Asp-Asp-Val connects domain 1 and the C-terminal domain (domain 2). Domain 2 contains 11 ␣-helices, 2 3 10 -helices, and a small ␤-hairpin. The most important secondary structural elements involved in contacts between the two domains are loop regions on domain 1 and helices ␣10 and ␣15 in domain 2. In addition, helix ␣13 contributes a conserved arginine, Arg-361, which is engaged in interdomain salt bridges to Asp-230 on ␤11 and Glu-259 located on the loop at the C terminus of ␤12.
The active site is at the interface of the two domains with the majority of residues that bind NADH contributed by the N-terminal domain and residues that bind mannitol coming primarily from the C-terminal domain. Arg-373 is the only residue from the C-terminal domain that interacts directly with the bound NAD ϩ . Important secondary structural elements that contribute residues to the active site are ␣1 and loop regions at the C termini of ␤3, ␤7, and ␤10 from domain 1 and ␣10 from domain 2. A conserved lysine located on ␣10, Lys-295, is poised to act as the general acid/base in the reaction.
Domain Structure-Residues 1-285 form the NAD ϩ -binding domain, domain 1. Helix ␣1 is in the center of domain 1 and is surrounded by the six-strand parallel ␤-sheet of the dinucleotide binding motif, a three-strand ␤-sheet formed by the extension of the final strand in the Rossmann fold with two antiparallel strands, and a four-strand mixed ␤-sheet. Although the relative order of the strands in the parallel sheet are the same as a Rossmann fold (CBADEF), there are insertions within it that suggest that the additional elements are not merely appended (Fig. 2). The three-strand sheet is formed wholly from residues inserted between the last strand in the dinucleotide binding fold (strand F) and the C-terminal domain, whereas the four-strand sheet is formed from residues at the N terminus, an insertion between strands B and C and the C-terminal insertion. A small solvent-accessible ␤-hairpin is located at the C terminus of ␤7. Asn-191 and Asp-230 are the only residues from domain 1 that directly bind mannitol, whereas Glu-133 has a water-mediated interaction.
Eleven ␣-helices and a small ␤-hairpin make up the C-terminal domain, which can be further divided into two subdomains. Residues 289 -375 compose helices ␣10-␣14, forming domain 2A, an antiparallel three-helix bundle with two short connecting helices. Domain 2B, residues 378 -493, consists of ␣15-␣20 and the ␤-hairpin. Helices ␣15, ␣16, and ␣19 form a three-helix bundle of different structural arrangement with ␣15 and ␣16 antiparallel and ␣16 and ␣19 parallel. Helix ␣15 has a pronounced 40°bend because of two proline residues, Pro-383 and Pro-388, in the middle of the helix. At least one proline is found in this region in all long-chain MDH sequences examined, suggesting that the bend in ␣15 is a common feature for the family. Helix ␣15 lies antiparallel to ␣10 and contributes several contacts to domain 1 including three water-mediated hydrogen bonds and one salt link between Lys-384 and Asp-140. Also located on ␣15 is Lys-381, which hydrogen bonds to the terminal hydroxyl of mannitol away from the point of oxidation.
Residues 289 -313 compose ␣10, the first helix in the C-terminal domain that lies near the cleft between domains 1 and 2 and provides several interdomain contacts. It also contributes three residues important in substrate binding, Asn-300, His-303, and the apparent catalytic acid/base Lys-295. Arg-361 located on ␣13 is involved in salt links across Although there is a lack of extensive interaction between the NAD ϩ and domain 2, a closer inspection shows that the loops from domain 1 that recognize the NAD ϩ provide ϳ40% of the interdomain contacts. Therefore, a conformational change upon binding NAD ϩ is possible and is predicted from the kinetic mechanism (13). NAD ϩ Binding and Specificity-The NAD ϩ is bound with the adenine anti and the nicotinamide syn between the two domains. The sugar pucker of the adenosine ribose is C 2Ј -endo, and the nicotinamide ribose is C 3Ј -endo. There is an intramolecular hydrogen bond between the nicotinamide N7 and pyrophosphate oxygen NO1. Using the method of Lee and Richards (26) with a probe of radius 1.4 Å, 80% of the accessible surface of NAD ϩ is buried in the complex with pfMDH (26). Cofactor contacts are mainly from domain 1 and primarily from the canonical dinucleotide binding motif. As is commonly found in enzymes with a Rossmann fold, the pyrophosphate moiety is situated at the N terminus of ␣1. The glycine-rich turn in the sequence 31 HIGVGGFHR 39 precedes ␣1. HXGXGXXXR is the conserved fingerprint motif for the long chain MDH family with the exception of altronate dehydrogenases that have a glutamine at position 31. A hydrophobic Ile, Leu, or Phe follows this residue. The N⑀ of invariant Arg-39 orients the amide of Gly-36 to interact with the pyrophosphate by hydrogen bonding to the carbonyl oxygen of Gly-35. Phe-37 stacks against the A side of the nicotinamide, making the B side accessible to substrate and promoting transfer of the 4-pro-S hydride. A Phe or Ile is present at this position for all members of the long-chain MDH family and probably stacks against the pyridine ring. Thr-233, located on the loop following ␤11, has main-chain hydrogen bonds to the cofactor amide oxygen and nitrogen (Fig. 3).
The adenine packs against Ile-131 and is shielded from solvent by Arg-66. Isomerization of the loop containing residues 65-69 is required for cofactor binding and release and could explain why the release of NADH is the rate-limiting step in the direction of oxidation of mannitol (13). Loops at the C termini of strands B, D, and E contribute residues that hydrogen bond with the ribose hydroxyls. Although significant activ-FIG. 1. A, stereoview of the ␣-carbon trace of the MDH molecule taken from the binary complex. The N and C termini are labeled, and every tenth residue is marked with a sphere and is numbered. Color ranges from blue at the N terminus to red at the C terminus. B, stereoview of the holo-MDH molecule indicating secondary structure and colored by primary sequence as in A. This figure and Figs. 5A and 6 were prepared using Molscript (33) and Raster3D (34). ity with NADPH is still observed, pfMDH exhibits a strong preference for NADH over NADPH (13). Asp-69, which hydrogen bonds to both adenosine ribose hydroxyls, almost certainly contributes toward specificity for NAD ϩ over NADP ϩ by discouraging phosphate binding. However, it is located on a mobile loop, allowing Asp-69 to have a different conformation with NADP ϩ bound. Arg-66 on the same loop could make favorable contacts with a phosphate moiety and partially compensate for the negative Asp-69 interaction. Other members of this family showing a preference for NADP ϩ have a loop containing an aspartate and several lysines that could occupy this position. However, the arginine is less well conserved.
The single residue found in a disallowed region of the Ramachandran plot, Thr-132, is located between Ile-131 (mentioned above) and Glu-133, which binds a nicotinamide ribose hydroxyl. The strained conformation observed in Thr-132 is probably dictated by the required arrangement of this loop.
Mannitol Binding and Specificity-Mannitol binds 16 -23 Å deep in the active site cleft with C2 (or C5) above the nicotinamide C4 (Fig. 4A). Each half of D-mannitol is equivalent to the other because of the fact that the chirality of C5(R) is the same as C2(R) and C4(R) is the same as C3(R). Consequently, the numbering choice for mannitol is arbitrary. This becomes important when considering alternate substrates such as D-sorbitol, D-mannonate, or mannitol 1-phosphate. The mannitol is bound in an extended conformation with a pseudo-2-fold axis of rotation between C3 and C4 (Fig. 4B). Ten direct polar interactions are made between the protein and the polyol as well as one additional water-mediated interaction (Fig. 4C). Three of the direct and the water-mediated interactions are with residues from the N-terminal domain; the remaining seven are with residues from the C-terminal domain. The substrate C2 is located within hydride transfer distance 2.9 Å from the nicotinamide C4.
Of the seven residues whose side chains are involved in specific interactions with mannitol, Asp-230 and Lys-295 (dis-cussed below) are invariant throughout the long-chain MDH family. Asp-230 hydrogen bonds to the C1 hydroxyl, preventing a phosphate or carboxylate from binding at that position. It is also involved in an important interdomain salt-link with Arg-361. Asn-300, His-303, and Lys-381 are all conserved or have conservative replacements. Asn-191 is replaced by leucine in the altronate dehydrogenases but is conserved in all other cases. The most variable residue is Arg-373, which may be replaced by Gln, Ser, or Asp. It is impractical at this time to make predictions regarding whether substitutions in the active site are involved in specificity for alternate substrates such as mannitol 1-phosphate, arabinitol, mannonate, or altronate, because few proteins of the long-chain MDH family have been enzymatically characterized.
pfMDH is completely inactive with phosphorylated substrates D-mannitol 1-phosphate and D-fructose 6-phosphate. In the reaction catalyzed by mannitol-1-phosphate 5-dehydrogenase, a phosphate moiety replaces the hydroxyl distal to catalysis on the substrate. Although Lys-381, which interacts with O6, could potentially bind a phosphate at that position in pfMDH, there is no steric room for this additional group. The lack of steric volume near O1 and the presence of Asp-230 prohibit binding of a phosphate or carboxylate moiety directly adjacent to the catalytic site. D-Arabinitol and D-sorbitol are alternate polyol substrates of pfMDH with K m ϭ 14 and 460 mM, respectively. The specificity constant of pfMDH for Darabinitol (k cat /K m ϭ 3.4 mM Ϫ1 s Ϫ1 ) is 5-fold lower than that for D-mannitol (K m ϭ 1.2 mM, k cat /K m ϭ 18 mM Ϫ1 s Ϫ1 ) (13). Arabinitol, a five-carbon polyol, could potentially make 10 of the 11 hydrogen bonds that mannitol makes. In addition, mannitol can fit into the active site with either C2 or C5 in position to be reduced, whereas D-arabinitol can only bind productively in one orientation. Both of these factors would increase the affinity of the enzyme for D-mannitol relative to D-arabinitol. The 600-fold reduced catalytic efficiency with D-sorbitol (k cat /K m ϭ 0.03 mM Ϫ1 s Ϫ1 ) is less easily understood. D-Mannitol and D-sorbitol differ only in that the configuration of C2 is S in D-sorbitol. For D-sorbitol, it is C5(R) that is oxidized to produce L-sorbose, such that the C2 is distal to the active site. Modeling sorbitol into the active site in a conformation equivalent to that of mannitol revealed no major steric clashes. The C2 hydroxyl of sorbitol, which is equivalent to O5 of mannitol, could no longer hydrogen bond to Asn-300 but would be located between His-303 and Arg-373 in a suboptimal orientation for hydrogen bonding. In the binary complex of pfMDH with NAD ϩ , a water molecule is found at a position equivalent to O5 of mannitol and makes hydrogen bonds to His-303 and Asn-300. Presumably, sorbitol would displace this water because C2 would be too close. This would leave the hydrogen bonds to His-303 and Asn-300 unfulfilled, a thermodynamically unfavorable situation. In fact, a loss of binding energy of ϳ3.5 kcal/mol for sorbitol relative to mannitol is observed (13) and may be attributed to the loss of one or two hydrogen bonds. More importantly, the perturbation of the precise hydrogen-bonding network surrounding the Asn-300 side chain could disrupt catalysis (see below).
Catalytic Mechanism-In converting mannitol to fructose, the hydroxyl at C2 is oxidized. This is done by the sequential abstraction of a proton from the sugar O2, the transfer of this proton to bulk solvent followed by the transfer of a hydride from the sugar C2 to the nicotinamide C4. The three hydrogen bond O2 makes to the protein identify residues involved in the catalytic mechanism. Lys-295 is positioned to act as the proton acceptor. The observed inflection point in catalytic efficiency at pH 9.3, therefore, is probably the result of the titration of this side chain (13). Hydrogens on the amide nitrogens of Asn-191 and Asn-300 would stabilize the partial negative charge on the O2 in the transition state (Fig. 5). In functioning as hydrogen bond donors to the two lone pairs of the substrate oxygen, the asparagines also direct the O2 proton toward Lys-295. Carbonyls of Val-229 and Asn-300 accept hydrogen bonds from Lys-295 and direct the lone pair of electrons on the lysine to accept a hydrogen bond from the substrate O2. The side-chain conformation of Asn-300 is therefore critical for catalysis because it functions in orienting both enzyme and substrate groups. The Asn-300 side chain is additionally oriented by a hydrogenbonding interaction with the mannitol O5. Polyols must have five or six carbons to be substrates of pfMDH (13). The fact that no activity has been detected with four-carbon polyols is presumably because this interaction with the substrate O5 is necessary. A lack of measurable activity with 5,6-dideoxy-Dfructose 3 supports this hypothesis.
The proposed mechanism for mannitol oxidation requires Lys-295 to be unprotonated. Substrate inhibition patterns suggest that isomerization of the pfMDH-NAD ϩ complex is required before it can productively bind mannitol (13). Based upon the hydrogen-bonding network surrounding Lys-295, it would seem probable that dissociation of the proton from this lysine producing the active form of the enzyme is this necessary step. Because the mannitol O2 accepts two hydrogen bonds from protein side chains, its proton will be directed toward the N of Lys-295. If the lysine is protonated, a productive complex will not form easily because there would be unfavorable contacts between the O2 proton and the Lys-295 proton.
In the binary complex, there is a well ordered water molecule 3.15 Å from N of Lys-295 on the opposite side from the active site. This water is held in place by interactions with the main- presumably has a lower pK a (Fig. 5). Strand F (containing residues 222-230) is shifted 0.56 Å toward the active sight in the ternary complex, presumably because of interactions between substrate O1 and residues 230 and 231. The water molecule is shifted away from the lysine to a distance of 3.76 Å. A solvent-lined channel is also observed in the ternary complex. These conformational differences and the well ordered water molecules observed in the binary and ternary complexes suggest two possible mechanisms by which the proton could be shuttled to bulk solvent.
In the first mechanism, Glu-292 could function as the proton shuttle. In its solvent-accessible conformation, it is located within 4.6 Å of Glu-293 and has a water-mediated interaction with this residue. Repelled by this negative charge, Glu-292 swings down toward the active site. After returning to the hydrophobic pocket, it neutralizes the charge by abstracting a proton from the water molecule. This water then abstracts a proton from Lys-295. The binding of the mannitol substrate brings strand F (residues 222-230) closer to the active site. This includes the carbonyl oxygen of Val-229, which is now within 2.34 Å of the water. The water moves back to a new position 1.15 Å away, hydrogen bonded to the main-chain nitrogen of Val-229, and is now 3.76 Å from Lys-295. The new water position is 1.46 Å from Glu-292, and Glu-292 swings out and loses the proton to the bulk solvent. Lys-295 abstracts a proton from the mannitol, which is then kinetically observed to be transferred to bulk solvent before hydride transfer occurs (27). The hydride is transferred, and products are released. Strand F readjusts to the position observed in the binary complex once NAD ϩ is bound, the water moves back in close to Lys-295, Glu-292 swings down, and the cycle is repeated.
The second possible mechanism utilizes Glu-292 in a slightly different manner. In the binary complex, Glu-292 polarizes the water molecule, which abstracts a proton from Lys-295. The binding of polyol substrate brings residues 222-230 closer, causing the protonated water to move away from Lys-295. The new water position is within 1.46 Å of Glu-292, and Glu-292 is ejected from the pocket. Glu-292 moves to its solvent-accessible conformation, opening a channel by which the proton is shuttled to bulk solvent along a chain of three water molecules. Lys-295 picks up a proton from substrate and shuttles it again to the solvent, hydride transfer occurs, products are released, and the strand containing residues 222-230 moves back out. This allows the water to again move close to Lys-295 and Glu-292 to swing back in. In this mechanism, Glu-292 acts as a gate opening and closing the solvent channel. Although further studies are necessary to elucidate the exact mechanism of proton transfer to bulk solvent, a direct loss of the proton from the lysine to the solvent is very unlikely because of the seques-tered nature of the active site and the kinetically observed deprotonation of the lysine prior to hydride transfer (27).
These proposed mechanisms cannot apply to the whole family, because Glu-292 is conserved only among a subgroup of the long-chain MDHs. This subgroup of 15 members includes mannitol 2-dehydrogenases, D-arabinitol 4-dehydrogenases, and mannonate dehydrogenases. The altronate dehydrogenase sequences have a lysine or arginine at this position, whereas the mannitol-1-phosphate 5-dehydrogenases have an isoleucine or valine. These other groups necessarily would use a different mechanism to return the lysine to the unprotonated state.
Structural Neighbors-A search for structural neighbors was conducted using the program Dali (28). As expected, many proteins showed similarity to the NAD ϩ -binding domain. Absent was the SDR family member tetrameric NADP ϩ -dependent mannitol 2-dehydrogenase from Agaricus bisporus (3). A number were identified that also had similarity in the C-terminal domain. The top three, 6-phosphogluconate dehydrogenase (29) (6PGDH), UDP-glucose dehydrogenase (30) (UD-PGDH), and N-(1-D-carboxylethyl)-L-norvaline dehydrogenase (31) (CENDH), were chosen for closer inspection. Overall root mean square deviations were 3.8 -4.2 Å with 7-10% sequence identity. All of them catalyze more complex reactions: NADP ϩdependent oxidative decarboxylation in the case of 6PGDH, 2-fold NAD ϩ -dependent oxidation for UDPGDH, and NADHdependent reductive condensation between an ␣-keto acid and an amino acid for CENDH. When the N-and C-terminal domains were searched separately, a similarity in both was also identified with glycerol-3-phosphate dehydrogenase (32) (G3PDH). Because the C-terminal domain of G3PDH was rotated ϳ50°compared with pfMDH, this relationship was initially missed. The domains in G3PDH are proposed to come closer upon substrate binding, explaining the difference in domain arrangement when compared with pfMDH. G3PDH catalyzes the NADH-dependent interconversion of dihydroxyacetone phosphate and L-glycerol 3-phosphate, a reaction similar to that catalyzed by pfMDH. A superposition of pfMDH with these structural neighbors is shown in Fig. 6.
Each of these enzymes has an eight-strand ␤-sheet in the N-terminal domain that consists of the canonical six parallel strands with an additional two antiparallel strands. UDPGDH and 6PGDH also have a conserved Lys-X 3 -Asn on the central ␣-helix homologous to Lys-295 and Asn-300 in pfMDH. The role of the lysine as the general base in these reactions has been proposed. Although CENDH and G3PDH do not contain a comparable lysine, they do share a homologous asparagine. The conservation of secondary structural elements as well as residues at equivalent positions that are contributed to the active FIG. 6. Overlay of pfMDH with structural relatives G3PDH, 6PGDH, UDPGDH, and CENDH. MDH is represented in white, G3PDH is represented in red, 6PGDH is represented in green, UD-PGDH is represented in magenta, and CENDH is represented in blue. This figure was made by overlapping the central six-strand parallel ␤-sheet in the N-terminal domains and equivalent secondary structural elements in the C-terminal domains separately. Homologous residues in the active site are rendered as ball and stick. sites implies that these enzymes are related by divergent evolution.
Conclusion-The structure presented here is the first crystal structure of an enzyme from the long-chain mannitol dehydrogenase family. It reveals that pfMDH is a monomer composed of two domains with the catalytic site 20 Å deep in a cleft between the two domains. The N-terminal nucleotide-binding domain includes a Rossmann fold that is extended by two antiparallel ␤-strands. Helix ␣1 has the HXGXGXXXR fingerprint motif located at its N terminus. The complexes with NAD ϩ and mannitol allow the identification of residues involved in substrate and cofactor binding. Conservation of many of these residues suggests that other long-chain MDH family members recognize their substrates similarly. Unlike other alcohol and polyol dehydrogenases that contain a catalytic tyrosine or metal ion, Lys-295 is poised to act as the general base in the reaction. Asn-191 and Asn-300 are positioned to direct the O2 proton toward Lys-295 and stabilize a negative charge on the substrate oxygen in the transition state. Asn-300 is involved in a precise hydrogen-bonding network with Lys-295 and substrate O2 and O5 that appears to be essential for efficient catalysis. An unusual proton tunnel/shuttle utilizing the mobile Glu-292 side chain could be involved in returning the lysine to its uncharged state in preparation for the next catalytic cycle. Even though pfMDH is a monomer, the closest structural neighbors are dimeric with the C-terminal domain involved in dimerization as well as substrate binding. Conserved secondary structural elements and conservation of residues contributed to the active site suggest a common ancestor for this heterogeneous group.