The crystallographic structure of the mannitol 2-dehydrogenase NADP+ binary complex from Agaricus bisporus.

Mannitol, an acyclic six-carbon polyol, is one of the most abundant sugar alcohols occurring in nature. In the button mushroom, Agaricus bisporus, it is synthesized from fructose by the enzyme mannitol 2-dehydrogenase (MtDH; EC ) using NADPH as a cofactor. Mannitol serves as the main storage carbon (up to 50% of the fruit body dry weight) and plays a critical role in growth, fruit body development, osmoregulation, and salt tolerance. Furthermore, mannitol dehydrogenases are being evaluated for commercial mannitol production as alternatives to the less efficient chemical reduction of fructose. Given the importance of mannitol metabolism and mannitol dehydrogenases, MtDH was cloned into the pET28 expression system and overexpressed in Escherichia coli. Kinetic and physicochemical properties of the recombinant enzyme are indistinguishable from the natural enzyme. The crystal structure of its binary complex with NADP was solved at 1.5-A resolution and refined to an R value of 19.3%. It shows MtDH to be a tetramer and a member of the short chain dehydrogenase/reductase family of enzymes. The catalytic residues forming the so-called catalytic triad can be assigned to Ser(149), Tyr(169), and Lys(173).

Furthermore, mannitol metabolism may be involved in the recycling of NADP and NADPH. NADP produced during the conversion of fructose to mannitol becomes available for the oxidative reactions of the pentose phosphate shunt, which in turn release NADPH. The interaction between these two biochemical pathways may result in a fine regulation of the cellular NADP/NADPH ratio and suggests a role for mannitol metabolism in growth regulation (8). The production of transgenic mushrooms with altered characteristics is possible, because transformation procedures have been established (9 -11). The lack of detailed information at the molecular level regarding genes and proteins involved in central metabolic processes (such as mannitol metabolism) has restricted the breeding of A. bisporus by genetic methods.
Apart from its agrotechnological importance, mannitol dehydrogenases are being evaluated as an alternative way to synthesize mannitol from fructose. The ability to produce large quantities of pure and active enzyme using microbial expression systems together with the more efficient enzymatic mannitol production process (compared with the chemical reduction of fructose) may enable commercial production of pure mannitol for use in the food industry.
Given the importance of mannitol metabolism in fungi, the MtDH 1 gene of A. bisporus has been cloned and characterized (4). In addition, the knowledge of the three-dimensional structure reported in this work makes it easier and more rational to produce mutated variants of mannitol dehydrogenase.

EXPERIMENTAL PROCEDURES
Crystallization of the MtDH-NADP ϩ Binary Complex-Expression of the A. bisporus MtDH gene within the Escherichia coli host BL21(DE3), purification of the resulting soluble protein, and production of monoclinic crystals (space group C2) have been reported previously (12). Crystals of the binary complex (MtDH and NADP ϩ ) were obtained under similar conditions using the sitting drop vapor diffusion method. NADP ϩ was added to the protein solution to a final concentration of 1 mM prior to crystallization. The drop was created by combining 5 l of the protein solution (10 -20 mg/ml and 1 mM NADP ϩ ) with 5 l of well solution (90 mM Tris-HCl, pH 7.5, 18% PEG 4K, and 9% isopropanol at 20°C).
X-ray Diffraction Data Collection-X-ray diffraction data were collected to 1.5-Å resolution at 120 K using a MAR345 image plate detector at beamline BM1A at the European Synchrotron Radiation Facility in Grenoble (wavelength ϭ 0.800 Å). The detector was placed 280 mm from the crystal. Each image was exposed in dose mode for ϳ5 min and consisted of a 0.5°rotation of the crystal. The data from one crystal (1.5 ϫ 0.2 ϫ 0.1 mm) were processed using MOSFLM (13) and SCALA (14,15). Data collection statistics are listed in Table I.
Initial Phase Determination and Model Refinement-Molecular replacement was done with the CNS software package (16). Initial mo-* This work was supported by Schweizer Nationalfond Grant 31-52398.97 (to U. B.). 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  lecular replacement calculations proved to be difficult because of the low sequence identity (30% at most) to members of the short chain dehydrogenase/reductase (SDR) family with known three-dimensional structure and because of the large unit cell content in the crystals of MtDH. Parallel attempts to determine the structure by multiple isomorphous replacement were hampered by nonisomorphism where the cell axes varied by 5-8%. Hence, a more extensive molecular replacement approach was undertaken. The tetramer of mouse lung carbonyl reductase (Protein Data Bank entry 1cyd) was used as a search model. The probe had to be modified to give significant solutions in the rotation function calculation in the following way. A sequence alignment between MtDH and mouse lung carbonyl reductase was calculated, and all nonconserved residues were changed to alanine with the SEAMAN program (17). Data from 6 to 10 Å was used in the cross-rotation search, and three tetramers could be located in the asymmetric unit. This corresponds to a Matthews coefficient (18) V M of 2.43 Å 3 /Da with ϳ47.5% solvent volume. Subsequent to rigid body refinement, noncrystallographic symmetry restraints were applied, and torsion angle refinement was performed using data in the resolution range from 40 to 3.5 Å (R factor ϭ 39%, R free ϭ 40%). In a 2Fo Ϫ Fc electron density map, 243 (of 265) residues could be identified, and 222 side chains were build into the electron density map using the program O (19). Attempts to employ noncrystallographic symmetry averaging using CNS and DM (20) failed. The electron density correlation dropped during averaging, and the quality of the map became inferior. In hindsight this can be explained by the large differences among the three individual tetramers concerning their anisotropic movement in the crystal. At this stage automatic model building was carried out with the program ARP/wARP (21) in mode warpNtrace using data in the resolution range from 40 to 1.5 Å, which led to a dramatic improvement. After 10 cycles of automated model building, 260 residues (excluding the 5 N-terminal residues) and the NADP ϩ could be identified in the resulting 2Fo Ϫ Fc electron density map. Two nickel ions per tetramer (coordinated each by the C termini of two monomers) were also added. Evidence showing that these metal ions were nickel arose from the purification by nickelchelate affinity chromatography and the lack of chelating agents in all buffers and are therefore presumed to be purification artifacts. After the addition of about 3000 water molecules, torsion angle and individual B factor refinements, the R factor dropped to 22.3% (R free ϭ 23.6%). Visual inspection of the electron density map and comparison of the three individual tetramers revealed a striking difference in quality, reflected also in the average individual B factors for each subunit (Table  II). TLS refinement with the program REFMAC (22,23) was carried out to account for this behavior. In this procedure, every monomer was treated as a rigid body, for which the three tensors T, L, and S were refined. Because of the possibility that crystal packing restricts movement of whole molecules (24), the largest movements as modeled by the TLS parameters are found in the tetramer showing the smallest number of crystal contacts. The removal of the subunit movement from the individual B factors resulted in comparable B factors for the individual subunits (Table II) and enabled the application of noncrystallographic symmetry restraints. The final model statistics after anisotropic B factor refinement for the nickel ions are listed in Table III.
Molecular Modeling-The substrate mannitol was modeled into the active site cavity with the program AUTODOCK using the genetic algorithm of the program. The results from 100 docking calculations were clustered into groups containing structures with less than 1.5-Å root mean square deviation. The clusters were sorted by energy.
Analytical Size Exclusion Chromatography and Dynamic Light Scattering-The quaternary structure of MtDH was investigated by determining the molecular weight of MtDH in solution by size exclusion chromatography and dynamic light scattering. For the size exclusion chromatography, a prepacked Amersham Pharmacia Biotech Superdex 75 HR16/60 column was equilibrated in 100 mM NaCl and 20 mM Tris-HCl, pH 7.5. The flow rate was set to 0.5 ml/min and controlled by a fast protein liquid chromatography system from Amersham Pharmacia Biotech. The column was calibrated with a set of standard proteins (1 mg each), and the molecular weight of MtDH was determined by graphical comparison of the measured elution volumes. Dynamic light scattering was performed on a single wavelength fixed angle DynaPro system (Protein Solutions). Protein samples were centrifuged at 14,000 ϫ g for 10 min and measured in 12-l quartz cuvettes. Size distribution analysis was performed with the program DYNALS.
Enzyme Kinetics-Kinetic measurements were done with a SPEC-TRAmax 250 microplate spectrophotometer (Molecular Devices). The assay buffer containing the substrate (mannitol or fructose) at different concentrations from 10 to 500 mM, and the cofactor (NADP ϩ or NADPH) in the range of 40 -300 M was incubated at 30°C. After the addition of MtDH (4 -10 nM), the absorption of NADPH at 340 nm was measured for 30 min, and the initial rate constants were calculated. K m values and V max for the substrates and the cofactor were calculated with the program LEONORA (25).

TABLE II TLS refinement
Comparison of average B factors before and after TLS refinement. The difference between the individual subunits can be modeled by refining the three parameters (T, L, and S) for each subunit, which corresponds to a movement of the individual subunits as rigid bodies. After this treatment of the data, the remaining B factors consist solely of the individual movement of each atom, and the average B factors are identical. The number of crystal contacts correlates to the value of the average B factors for the individual subunits before TLS refinement.

RESULTS
Overall Tertiary and Quaternary Structure of MtDH-MtDH is a member of the SDR family of enzymes, also known as the tyrosine-dependent oxidoreductase family (reviewed in Ref.   tween strands F and G. Secondary structure elements were assigned according to the convention described by Ghosh et al. (29). The loop between ␤F and ␣G contains two small helices (␣FG1, residues 204 -209 and ␣FG2, residues 211-220) and one ␤-strand (␤FG1, residues 227-229), which form one side of the active site cavity. Within the SDR family this loop is responsible for the substrate specificity among the different enzymes (30 -35). The other side of the active site is lined by the insertion loop between ␤E and ␣F, which in the case of MtDH contains a small antiparallel ␤-sheet, ␤EF1 (residues 157-159) and ␤EF2 (residues 162-164) (Fig. 1A).
Size exclusion chromatography and dynamic light scattering showed that MtDH is a tetramer in solution. Our kinetic data showed cooperativity in neither the oxidative nor the reductive reaction of MtDH. The crystal structure revealed that the asymmetric unit contains three tetramers. Each tetramer possesses internal 222 symmetry. The interaction surface is 1460 Å 2 between monomers 1 (cyan) and 2 (red), 660 Å 2 between monomers 1 and 3 (green), and 1740 Å 2 between monomers 1 and 4 (magenta) (color coding according to Fig. 1B; calculations were done with the CNS software package (16,36)). The total surface area buried in each monomer therefore is 3850 Å 2 or ϳ32%.
The C termini of monomer 1s and 3 and the C termini of monomers 2 and 4 coordinate one metal ion each. The absence of anomalous scattering for this metal ion using Cu-K␣ radiation (1.54 Å) combined with the presence of anomalous scattering peaks using synchrotron radiation (0.80 Å) give good evidence that this metal ion indeed is nickel, which is probably a purification artifact. Also, the trigonal prismatic coordination sphere of each metal ion (two C termini from two individual subunits of one tetramer and four water molecules) is not uncommon for nickel under conditions at which sterical hindrance prevents octahedral coordination.
Location and Conformation of the NADP ϩ Substrate-The NADP ϩ binding site is located at the C-terminal end of the ␤-sheet forming the Rossmann fold (Fig. 1A). The substrate binding crevice forms an oval-shaped cavity with dimensions of ϳ20 ϫ 12 ϫ 7 Å with the cofactor at the bottom of the cavity. Both riboses of the NADP ϩ have 2 E (C 2Ј -endo) puckering, the nicotinamide ring is in the syn conformation, and the adenine in the anti conformation (nomenclature according to International Union of Pure and Applied Chemistry-International Union of Biochemistry nomenclature for polynucleotides (37)). This conformation classifies MtDH as B-stereospecific and is typical for SDR enzymes (38). The cofactor is bound to the enzyme by an extensive network of hydrogen bonds, which also involves seven water molecules (Fig. 2). The loops at the Cterminal end of ␤-strands A, B, and C are responsible for the binding of the adenosine moiety. In the first loop, Asn 20 and Arg 21 form hydrogen bonds to the phosphate group of the ribose and the O 3Ј of the ribose, respectively. The main-chain amide groups of Arg 43 , Ser 44 , and Ala 45 also form hydrogen bonds to the phosphate group. In addition, the side chains of Arg 43 and Ser 44 coordinate two water molecules together with the phosphate group. The Asn 69 hydrogen bonds to the adenosyl amino group, whereas the main chain amide of Val 70 interacts with the N1 ring nitrogen. Gln 206 and Ile 23 form direct hydrogen bonds to the pyrophosphate moiety, whereas Gly 18 , Arg 21 , Gly 24 , Asn 96 , Gly 98 , and Thr 204 form hydrogen bonds mediated by water molecules.
The nicotinamide moiety also shows several residues in hydrogen bonding distance. The active site Lys 173 forms a bifurcate hydrogen bond to the O 2Ј and O 3Ј of the ribose; the active site tyrosine also forms a hydrogen bond to the O 2Ј . The amide group of the nicotinamide is kept in place by a contact to Val 202 and Thr 204 and an additional water molecule.
Location of the Mannitol Substrate and Mechanism of Catalysis-All attempts to obtain a structure of MtDH with various combinations of NADP, NADPH, mannitol, and fructose to date have failed. For that reason mannitol was modeled into the active site cavity with the program AUTODOCK. The cluster with the lowest energy contained 38 of 100 docked structures. This orientation was assumed to be the best orientation possible for mannitol. In our model, the substrate can form seven hydrogen bonds to the protein. The active site serine (Ser 149 ) is in hydrogen-bonding distance to O 1Ј and O 2 ; Ser 151 also interacts with O 2Ј . Further contacts are possible between O 3Ј and the amide group of the nicotinamide and between O 5Ј and Gln 166 . This orientation places the C 2Ј at a distance of 3.4 Å to the C4 of the nicotinamide, to which the hydrogen is transferred during catalysis (Fig. 3). Structural analysis of other members of the SDR family has shown that three residues are crucial for catalysis, forming the so-called catalytic triad (30, 31, 33, 39 -45). The SDR family catalytic triad of MtDH is Ser 149 -Tyr 169 -Lys 173 . The role of the lysine is to orient the nicotinamide moiety of the NADP by forming a bifurcate hydrogen bond to the O 2Ј and O 3Ј of the ribose (46). The proposed reaction mechanism suggests Tyr 169 to be deprotonated. This is facilitated by the positively charged environment provided by Lys 173 and the NADP ϩ itself. During catalysis, the tyrosine directly interacts with the substrate hydroxyl group. In the structure presented here, the putative position of the substrate is occupied by a water molecule, which is hydrogen-bonded to both of the active site residues participating in catalysis, Ser 149 and Tyr 169 , and to Ser 151 . This water molecule lies within a distance of 0.88 Å to the proposed position of the O 2Ј of the mannitol obtained by modeling. A schematic drawing of the reaction mechanism is shown in Fig. 4.
Enzymatic Activity of Recombinant MtDH-The enzymatic  parameters of MtDH in steady-state kinetics have been determined for recombinant MtDH (Table IV). The results are comparable with the data determined for the native enzyme isolated from A. bisporus sporocarps (47).

DISCUSSION
Structure Determination-Three steps proved to be crucial for fast structure determination and refinement in the case of MtDH. First, different members of the SDR family, for which the three-dimensional structure was available, were used as probes in the molecular replacement calculations. Of those, only one (mouse lung carbonyl reductase; Protein Data Bank entry 1cyd) gave interpretable solutions for the cross-rotation and translation function. In addition, the probe had to be modified. On the basis of a sequence alignment and modeling with the program SWISSMODEL (48), all nonconserved residues were changed to alanine (except for the glycines, which were kept), and differing loops were deleted. Second, automatic model building with ARP/wARP made fast model building possible, keeping in mind that the asymmetric unit of the crystal consists of three tetramers that are not similar enough to apply real-space averaging. Third, describing the difference between the three tetramers as independent rigid body movements and the separate treatment of this behavior with the three parameters T, L, and S for each individual subunit led to remaining B factors that were very similar between the subunits and allowed for subsequent application of noncrystallographic symmetry restraints. This procedure dropped the R factor and R free by about 3%. The movement of the three tetramers can be described with a set of three nonintersecting screw axes for FIG. 5. Rigid body movement of the three different tetramers in the asymmetric unit cell. The three tetramers forming the asymmetric unit of the crystal and the three nonintersecting screw axes for each monomer are shown. The length of the axes is proportional to the amplitude of the movement. The axes were calculated with the program TLSANL (23).
FIG. 6. Sequence alignment of SDR family members with respect to cofactor specificity. Sequences of proteins with known threedimensional structure and less than 1.9 Å root mean square deviation compared with MtDH were aligned based on the three-dimensional structure. Amino acids are colored as follows: dark gray, glycine; light gray, apolar; black background, negatively charged; gray background, positively charged. The consensus sequence is based on a minimum plurality of four. *, MtDH is listed here for comparison. each subunit (Fig. 5). A comparison with Table II clearly shows that tetramer 1 (monomers A, B, C, and D), having the highest B factors, also undergoes the largest movements in the crystal. An analysis of the crystal contacts formed by the three tetramers is in accordance to the hypothesis of Hery et al. (24), who theorized that crystal contacts can restrict the movement of whole molecules. The number of contacts is 38 for tetramer 1 compared with 50 and 49 for tetramers 2 and 3, respectively. The treatment of the data as rigid body movement of whole monomers significantly improved the quality of the electron density in tetramer 1, but it is still worse than for the two other tetramers.
Enzymatic Activity and Mechanism of MtDH-The kinetic constants of recombinant MtDH are comparable with those of the native enzyme. The very low affinity for the substrates has been reported for mannitol dehydrogenases from other organisms (47). MtDH can also to some extent use sorbitol or arabitol as a substrate. This is also demonstrated by our modeling studies. Mannitol can form seven hydrogen bonds to MtDH (see Table V and Fig. 3), whereas sorbitol and arabinol are only able to form six hydrogen bonds. The high K m values may explain why we have not succeeded until now to co-crystallize the binary complex of MtDH and NADP ϩ with either mannitol or fructose. The high K m value for fructose is probably because of the fact that it is actually the open and not the hemiketal form of the sugar that undergoes the reaction. The equilibrium between the open and ring forms leaves very little substrate in the open form. The ring opening is base-catalyzed, and Tyr 169 may act as a catalytic base for both the ring opening and the reduction of the ketone. The pK a of Tyr 169 is lowered by the positively charged environment formed by Lys 173 and the oxidized NADP ϩ in the case of mannitol oxidation, leading to a pH optimum of about 9.0 in the case of MtDH. Therefore the distance between the catalytic tyrosine and the lysine can be correlated to the pH dependence of the different enzymes in the SDR family (49).
Cofactor Specificity-An interesting possibility is to change the preference for the cofactor NADP to the much cheaper NAD by site-specific mutagenesis. The ability to change the cofactor specificity of NAD(P) utilizing enzymes has been demonstrated previously in the case of glutathione reductase (50). A structure alignment was produced with the programs DEJAVU and LSQMAN (51). Structures with less than 1.9 Å root mean square deviation for the C␣-carbons were compared. An alignment based on this superimposition gives suggestions of mutations necessary to change the cofactor specificity from NADP to NAD (Fig. 6). Two amino acid stretches are responsible for the cofactor specificity, the loop between ␤A and ␣B, and the loop between ␤B and ␣C. The first loop includes the GXXXGXG motif, where X can be any amino acid. In enzymes using NADP, the position in front of the second glycine (Arg 21 in the case of MtDH) is occupied by a positively charged amino acid (arginine or lysine), whereas enzymes using NAD prefer small or even negatively charged amino acids (serine, threonine, or aspartate). A second, highly conserved arginine (Arg 43 in MtDH) is found for all proteins using NADP. This arginine is located in the loop between ␤B and ␣C. For NAD utilizing enzymes, different modes of interaction with the cofactor can be seen in this position. In the case of Protein Data Bank entries 1eny and 1eno (enoyl-acyl carrier protein reductase from Mycobacterium tuberculosis and Brassica napus), aromatic residues are making stacking interactions with the adenine ring, whereas a glutamine can be found in enoyl-acyl carrier protein reductase of E. coli (Protein Data Bank entry 1qg6) and an isoleucine in 7-␣-hydroxysteroid dehydrogenase from E. coli (Protein Data Bank entry 1fmc). The first steps in changing the cofactor specificity are therefore mutations of Arg 21 to threonine, serine, or aspartic acid and of Arg 43 to either an aromatic or apolar residue.