The X-ray Structure of dTDP-4-Keto-6-deoxy-D-glucose-3,4-ketoisomerase*

The repeating unit of the glycan chain in the S-layer of the bacterium Aneurinibacillus thermoaerophilus L420-91T is composed of four α-d-rhamnose molecules and two 3-acetamido-3,6-dideoxy-α-d-galactose moieties (abbreviated as Fucp3NAc). Formation of the glycan layer requires nucleotide-activated sugars as the donor molecules. Whereas the enzymes involved in the synthesis of GDP-rhamnose have been well characterized, less is known regarding the structures and enzymatic mechanisms of the enzymes required for the production of dTDP-Fucp3NAc. One of the enzymes involved in the biosynthesis of dTDP-Fucp3NAc is a 3,4-ketoisomerase, hereafter referred to as FdtA. Here we describe the first three-dimensional structure of this sugar isomerase complexed with dTDP and solved to 1.5 Å resolution. The FdtA dimer assumes an almost jellyfish-like appearance with the sole α-helices representing the tentacles. Formation of the FdtA dimer represents a classical example of domain swapping whereby β-strands 2 and 3 from one subunit form part of a β-sheet in the second subunit. The active site architecture of FdtA is characterized by a cluster of three histidine residues, two of which, His49 and His51, appear to be strictly conserved in the amino acid sequences deposited to date. Site-directed mutagenesis experiments, enzymatic assays, and x-ray crystallographic analyses suggest that His49 functions as an active site base.

Many Gram-positive and Gram-negative bacteria, as well as some Archaea, synthesize cell surface envelopes referred to as surface or S-layers. These outermost cell layers are composed of proteins or glycoproteins that self-assemble into two-dimensional arrays. Depending upon the organism, these arrays can form hexagonal, square, trimeric, or oblique crystalline lattices (1). For those bacteria with an S-layer, it has been estimated that ϳ20% of the total cellular protein synthesis is devoted to its production (2). With respect to the level of protein glycosyla-tion, it varies from 2 to 10% (w/w) depending upon the organism (2).
Although the function of the S-layer is still not well understood, it has been postulated to be involved in bacterial virulence by facilitating invasion of host tissue and by protecting the pathogen against host defense mechanisms (3,4). Interestingly, the S-layer glycans can be lost after prolonged growth of bacteria in rich media, and thus it has been speculated that the carbohydrate components of the S-layer confer a selective advantage on the bacteria in their natural habitat (2).
In recent years, the bacteria Aneurinibacillus thermoaerophilus strains L420-91 T and DSM 10155/G ϩ and Geobacillus stearothermophilus strain NRS2004/3a have served as model systems for probing the nature of the S-layer glycans (5). In A. thermoaerophilus L420-91 T , the S-layer is composed of identical 109-kDa glycoprotein subunits arranged in a square lattice (6). The repeating unit of the glycan chain is a hexasaccharide composed of four ␣-D-rhamnose units and two 3-acetamido-3,6-dideoxy-␣-D-galactose residues linked together as indicated in Scheme 1. Formation of the glycan chains occurs in the cytoplasm and requires nucleotide-activated sugars as the donor molecules (7). For S-layer production in A. thermoaerophilus, both GDP-rhamnose and dTDP-3-acetamido-3,6-dideoxy-␣-D-galactose (abbreviated as dTDP-Fucp3NAc) are required.
The biosynthetic pathway for the production of dTDP-Fucp3NAc in A. thermoaerophilus was elucidated in 2003 and is indicated in Scheme 2. Like most of the pathways for the synthesis of 3,6-dideoxyhexoses, the formation of this unusual sugar begins with the attachment of ␣-D-glucose 1-phosphate to dTMP via the action of glucose-1-phosphate thymidylyltransferase. In the next step, the 6-hydroxyl group is removed, and the 4-hydroxyl group is oxidized to a keto-functionality yielding dTDP-4-keto-6-deoxyglucose. This reaction is catalyzed by dTDP-glucose 4,6-dehydratase. Both the thymidylyltransferase and the dehydratase have been well characterized with respect to structure and function (8,9).
Three additional enzymes are ultimately required for the synthesis of dTDP-Fucp3NAc, namely an isomerase, an aminotransferase, and an acetylase (Scheme 2). These enzymes are encoded by the fdtA, fdtB, and fdtC genes, respectively (10). The isomerase, hereafter referred to as FdtA, is especially intriguing in that it catalyzes the conversion of a 4-keto substrate into a 3-keto product with accompanying epimerization about C-4 of the hexose ring. From previous biochemical studies, it appears that FdtA does not require cofactors for activity (11). Thus it can be speculated that the reaction catalyzed by FdtA proceeds via a concerted acid-base mechanism similar to that proposed for triose-phosphate isomerase, the well known enzyme of the glycolytic pathway.
To date, at least 65 ORFs 3 demonstrating homology to FdtA in the NCBI data base have been identified, but until now, no three-dimensional structures have been reported (11). These ORFs are found only in bacteria and are typically associated with additional genes thought to be involved in the production of outer membrane polysaccharides (11).
Here we report the first x-ray structure of FdtA, complexed with dTDP and determined to a nominal resolution of 1.5 Å. Each subunit of the dimeric enzyme adopts a ␤-barrel motif with the dTDP ligand positioned near the opening of the barrel. Inspection of the active site region reveals a cluster of the following histidine residues: His 49 , His 51 , and His 95 . Both His 49 and His 51 are strictly conserved among the amino acid sequences presently reported. To probe the biological role of these two residues, each was changed to an asparagine via sitedirected mutagenesis, and the mutant proteins were assayed for activity. In addition, a double mutant protein was prepared whereby both histidines were substituted with asparagines. Whereas the H51N mutant protein retained limited activity, both the H49N and the H49N/H51N proteins were catalytically inactive. X-ray crystallographic analyses of these mutant proteins revealed no significant structural perturbations of their active site regions. On the basis of both biochemical and x-ray crystallographic data, we propose that His 49 functions as an active site base to abstract the hydrogen from C-3 of the sugar and to deliver it to C-4, whereas His 51 serves to shuttle protons between the C-3 and C-4 sugar oxygens. Details concerning the overall structure and function of FdtA are presented.

EXPERIMENTAL PROCEDURES
Materials-Unless otherwise noted, all chemicals and reagents were purchased from Sigma. Platinum Pfx DNA polymerase was obtained from Invitrogen. Reagents for DNA purifications were manufactured by Qiagen.
Isolation of Genomic DNA-A. thermoaerophilus L420-91 T was obtained from ATCC (700303). The freeze-dried pellet was reconstituted in sterile SVIII media (10) and subsequently used to inoculate a small volume of SVIII media. Cells were grown overnight at 55°C with shaking and then harvested by centrifugation. Genomic DNA was isolated according to standard protocols (12).
Cloning of the fdtA Gene-The gene encoding the isomerase was PCR-amplified from genomic DNA. Forward and reverse primers containing the restriction sites for NdeI and NotI, respectively, were used to amplify the gene. The gene was subsequently ligated into a modified pET28b(ϩ) vector (Novagen) in which the thrombin cleavage site was replaced with the recognition sequence for TEV protease. Proteins expressed with this modified vector possess an N-terminal hexahistidine tag that can be released by cleavage with TEV protease. Plasmids were tested for successful ligation by digestion with NdeI and NotI.
Protein Expression and Purification-The recombinant pET28-fdta plasmid was utilized for the transformation of Escherichia coli Rosetta(DE3) cells (Novagen). A single colony was picked to inoculate an overnight starter culture of LB media. Subsequently, the starter culture was used to inoculate several large scale cultures of TB media. Cells were grown at 37°C until an absorbance of ϳ1.0 at 600 nm was reached. Flasks were transferred to a shaker held at 22°C, and cells were allowed to grow for 18 h. Protein expression was induced with isopropyl ␤-D-thiogalactopyranoside. Cells were harvested by centrifugation.
For protein purification, cells were resuspended in lysis buffer (50 mM NaH 2 PO 4 , pH 8.0, 300 mM NaCl, and 10 mM imidazole) and maintained at 4°C for all subsequent steps. Cells were lysed by sonication, and insoluble debris was removed by centrifugation. The clarified lysate was then loaded onto a Ni 2ϩ -nitrilotriacetic acid-agarose column (Qiagen). The recombinant His 6 -tagged protein was eluted with a linear gradient of 10 -250 mM imidazole in lysis buffer. On the basis of SDS-PAGE analysis, fractions containing FdtA were pooled and dialyzed overnight against an excess of lysis buffer. The tag was removed with TEV protease, and cleaved FdtA was separated from the protease and uncut protein by running the digestion mixture over a Ni 2ϩ -nitrilotriacetic acid column. The protein was concentrated to 17 mg/ml (using an extinction coefficient of 0.91 (mg/ml) Ϫ1 ⅐cm Ϫ1 as calculated with Protean (DNAstar)) and dialyzed overnight against 1.0 liter of storage buffer (20 mM HEPES, pH 8.0, and 300 mM NaCl). Dialyzed protein was frozen in liquid nitrogen and stored at Ϫ80°C.
Construction of Site-directed Mutant Proteins-Using the pET28-fdtA construct, mutations were introduced into the gene using a QuikChange XL site-directed mutagenesis kit (Stratagene). Three mutated genes were generated as follows: H49N, H51N, and H49N/H51N. Prior to expression of the proteins, the genes were sequenced to ensure no additional muta-tions had been introduced during the mutagenesis process. The mutant proteins were expressed and purified as described above for the wild-type enzyme.
Activity Assays for Wild-type Protein and Mutant Forms of FdtA-The wild-type and mutant forms of FdtA were qualitatively assayed for activity via HPLC. Reaction mixtures contained 0.5 mol of dTDP-glucose, 10 nmol RmlB (Scheme 2), 10 mM MOPS, pH 7.5, and 10 nmol of FdtA. A reaction containing no isomerase was used as a negative control. For the reactions containing mutated FdtA, the wildtype reaction served as a positive control. Reactions were incubated at 37°C for 1 h and then boiled to denature the protein. Protein was removed by centrifugation, and the supernatant was run over a Resource Q column (Amersham Biosciences). Each sample was eluted with a linear gradient of ammonium acetate at pH 4.0. ESI mass spectrometry was used to identify the reaction products.
To verify the results of these assays, 10 nmol of the second enzyme in the N-acetylfucosamine pathway, FdtB (Scheme 2), was added to each reaction mixture. Each sample was eluted with a linear gradient of ammonium bicarbonate at pH 8.5. As before, ESI mass spectrometry was employed to analyze the eluted compounds.   Preparation of Selenomethioninelabeled Protein-A starter culture of E. coli Rosetta(DE3) cells harboring the pET28-fdtA plasmid was grown overnight at 37°C in M9 minimal media. Subsequently, the overnight culture was used to inoculate several large scale cultures of M9 minimal media supplemented with 5 mg/liter thiamine. Cultures were grown at 37°C to an absorbance of ϳ0.9 at 600 nm and then cooled on ice. Flasks were transferred to a shaker held at 16°C, and expression of selenomethionine-labeled protein was induced as described previously (13). Cells were harvested by centrifugation, and selenomethionine-labeled protein was purified to homogeneity as described for the wild-type protein.
Crystallization of Wild-type and Selenomethionine-labeled FdtA-An "in-house" sparse matrix screen consisting of 144 unique conditions was used to identify crystallization conditions. The hanging-drop method of vapor diffusion was employed, and crystallization trials were conducted at both 25 and 4°C. Preliminary crystals of the cleaved protein in complex with 10 mM dTDP and 20 mM D-fucose grew at 25°C from poly(ethylene glycol) 3400 and KCl solutions at pH 7.5. Crystallization conditions were subsequently optimized to 20 -21% poly(ethylene glycol) 3400, 150 -200 mM KCl, and 100 mM HEPES, pH 7.5. Crystals routinely grew in 1-2 days to maximum dimensions of ϳ0.4 ϫ 0.4 ϫ 1.0 mm. They belonged to the space group P4 1 2 1 2 with unit cell dimensions of a ϭ b ϭ 62.7 Å and c ϭ 201.3 Å and one dimer per asymmetric unit.
Crystals of the selenomethioninelabeled protein in complex with dTDP grew out of conditions similar to those for the unlabeled protein (22% poly(ethylene glycol) 3400, 200 mM KCl, and 100 mM HEPES, pH 7.5). These crystals belonged to the same space group and had similar unit cell dimensions.
Crystallization of the Mutant Proteins-All crystals of the mutated proteins were grown in the pres-ence of 10 mM dTDP and 20 mM fucose and achieved similar dimensions as the wild-type protein crystals within several days. Crystals of the H51N and H49N/H51N protein grew out of 18% poly(ethylene glycol) 3400 with 100 mM CHES, pH 9.0, and 150 mM LiCl. Crystals of the H49N protein grew out of 18% poly(ethylene glycol) 8000 with 100 mM HEPPS, pH 8.5, and 100 mM KCl. The space groups for all of the mutant protein crystals were the same as that for the wild-type protein, and the unit cell dimensions were very similar.
X-ray Data Collection and Processing-Crystals of both the selenomethionine-labeled and wild-type protein were stabilized for x-ray data collection by harvesting them into a synthetic mother liquor composed of 17% poly(ethylene glycol) 3400, 150 mM KCl, 200 mM NaCl, 10 mM dTDP, 20 mM fucose, and 100 mM HEPES, pH 7.5. They were subsequently frozen after transfer into a cryoprotectant solution composed of 25% poly(ethylene glycol) 3400, 225 mM KCl, 400 mM NaCl, 10 mM dTDP, 20 mM fucose, 15% ethylene glycol, and 100 mM HEPES, pH 7.5. X-ray data from both the selenomethionine-labeled and wild-type protein crystals were collected on a CCD detector at SBC Beamline 19-BM (Advanced Photon Source, Argonne National Laboratory, Argonne, IL). These data were processed and scaled with HKL2000 (14).
Crystals of the mutant proteins were stabilized for x-ray data collection in a similar manner, and x-ray data sets from these crystals were collected at 100 K with a Bruker AXS Platinum 135 CCD detector controlled with the Proteum software suite (Bruker (2004), PROTEUM, Bruker AXS Inc., Madison, WI). The x-ray source was CuK ␣ radiation from a Rigaku RU200 x-ray generator equipped with montel optics and operated at 50 kV and 90 mA. These data were processed with SAINT (version V7.06A, Bruker AXS Inc., Madison, WI) and internally scaled with SADABS (version 2005/1, Bruker AXS Inc., Madison, WI). Relevant x-ray data collection statistics are listed in Table 1.
X-ray Structural Analysis of FdtA-The three-dimensional structure of FdtA was solved via MAD phasing. The program SOLVE was used to locate the positions of seven of the eight selenium atoms in the asymmetric unit and to calculate initial protein phases (15). Solvent flattening with RESOLVE produced a readily interpretable electron density map calculated to 2.2 Å resolution (16). The entire polypeptide chain for the FdtA subunit was traced, with the exception of residues Met 1 and Glu 136 to Gly 139 . In addition, the N-terminal Gly-His residues resulting from the tag were not visible. The FdtA model was then employed as a search probe for molecular replacement with the software package EPMR (17) against the x-ray data set collected to 1.5 Å resolution from the wild-type crystal. Alternate cycles of manual model building and least squares refinement with the program TNT (18) reduced the R-factor to 19.5% for all measured x-ray data from 50.0 to 1.5 Å.
The three-dimensional structures of the site-directed mutant proteins were solved by molecular replacement with the program EPMR (17) using the wild-type protein model as a search probe. Relevant refinement statistics for all of the protein models are given in Table 2.

RESULTS
Overall Three-dimensional Structure of FdtA-Crystals of the wild-type protein employed for the structural analysis presented here diffracted to a nominal resolution of 1.5 Å and contained a complete dimer in the asymmetric unit. The electron densities for the polypeptide chain backbones were very well ordered for both subunits with electron density missing for only Met 1 , Lys 137 , Glu 138 , and Gly 139 in subunit 1 and Met 1 , Glu 136 , Lys 137 , Glu 138 , and Gly 139 in subunit 2. The quality of the model is excellent with 90.1% of the residues adopting , angles in the "most favored" and 9.9% in the "additionally allowed" regions of the Ramachandran plot.
A ribbon representation of the dimer is presented in Fig. 1a. The dimer has overall dimensions of ϳ47 ϫ 51 ϫ 64 Å and assumes an almost jellyfish-like appearance with the sole ␣-helices representing the tentacles. The subunit-subunit interface is rather extensive with a total buried surface area of ϳ3100 Å 2 . Formation of the FdtA dimer represents a classic example of domain swapping whereby ␤-strands 2 and 3 from one subunit form part of a ␤-sheet in the second subunit (19). The "hinge" loop required for the domain swapping is formed by a type II turn (Glu 22 to Lys 25 ) and a loop (Asn 26 to Lys 32 ) that together connect ␤-strands 3 and 4. This type of swapping phenomenon was first revealed in the dimeric structure of diphtheria toxin (20) and has now been observed in at least 40 structurally and biologically diverse proteins (21). It has been postulated that some domain-swapped multimeric proteins may have arisen from monomeric proteins as a result of destabilizing mutations (20,22).
The overall tertiary structure of each subunit is dominated by two layers of anti-parallel ␤-sheet that form a flattened barrel as can be seen in Fig. 1b. The two layers of sheet within the subunit each contain five ␤-strands. As indicated in the topology plot presented in Fig. 1c, one layer is composed of ␤-strands 1, 4, 6, 9, and 11, and the second layer contains ␤-strands 5, 7, 8, 10, and 12. Because of domain swapping, however, one layer of sheet contains an additional two anti-parallel ␤-strands contributed by the second subunit with ␤-strand 3 in one subunit abutting ␤-strand 4 in the second subunit. As a consequence, ␤-strands 3 from each subunit contribute significantly to the dimer interface. In addition to ␤-strand 3, the dimeric interface is also formed by the ␤-sheet composed of ␤-strands 1, 4, 6, 9 and 11, which projects primarily hydrophobic side chains into the interstitial space. This is illustrated in Fig. 2, a and b. Not surprisingly, given the hydrophobic character of the subunit-  (30). A topology drawing for one subunit of FdtA is shown in c. It was generated from the TOPS data base of protein structural topology (31). The red triangles represent strands of ␤-sheet, the green sphere corresponds to a helical turn, and the yellow sphere denotes the single ␣-helix in the subunit. C-term, C terminus; N-term, N terminus. subunit interface, there is a striking absence of ordered water molecules lining this region.
Although most of the amino acid residues lie within strands of ␤-sheet, there is one major ␣-helix that initiates with Tyr 123 at the end of ␤-strand 12 and extends to the C terminus. A variety of loops connect these secondary structural elements, including four type I turns (Asp 13 -Gly 16 , Ser 82 -Val 85 , Ser 101 -Cys 104 , and Asp 115 -Asp 118 ) and one type II turn (Glu 22 to Lys 25 ). Approximately 70% of the amino acid residues adopt , angles corresponding to classical secondary structural elements.
Recently, the biochemical properties for a similar 3,4-ketoisomerase from Streptomyces fradiae, referred to as Tyl1a, were reported (11). Given that FdtA and Tyl1a share a 42% amino acid sequence identity (supplemental Fig. 1), it can be speculated that the overall molecular architecture of Tyl1a will be similar to that described here for FdtA. Both of these enzymes belong to the RmlB-like cupin superfamily (11,23,24).
The Active Site of FdtA-Electron density corresponding to the dTDP ligand in subunit 2 is presented in Fig. 3a, and as can be seen, the nucleotide is well ordered. Unlike that observed in subunit 2, however, the electron density for the ligand in subunit 1 is indicative of multiple conformations.
Consequently, for the sake of simplicity, the following discussion only refers to subunit 2 because the nucleotide adopts a single conformation.
A close-up view of the region surrounding the dTDP is displayed in Fig. 3b. The deoxyribose adopts the C2Ј-endo pucker. The C-4 carbonyl oxygen of the thymine base is located within 2.7 Å of N 1 of Arg 33 . There are no other direct contacts between the thymine ring and the protein, but rather N-3 and the C-2 carbonyl oxygen interact with ordered water molecules. Likewise, there are no side chains or ordered water molecules located within 3.2 Å of the 3-hydroxyl group of the deoxyribose. The ␣-phosphoryl oxygens of dTDP are surrounded by three waters and the side chains of Tyr 119 and Arg 46 . There is a water molecule located within 3.1 Å of the bridging oxygen. The ␤-phosphoryl oxygens form electrostatic interactions with three water molecules and the guanidinium groups of Arg 121 and Arg 15 . Note that Arg 15 is contributed by subunit 1, and hence part of the FdtA active site is formed by the second subunit in the dimer as a result of domain swapping.
Although the crystals were grown in the presence of 20 mM D-fucose (which is 6-deoxy-D-galactose), there was no obvious electron density near the dTDP moiety that clearly indicated an ordered carbohydrate. However, there was a string of electron density near the nucleotide that was modeled as ordered water molecules, and this density could possibly correspond to a fucose molecule at very low occupancy. These solvent molecules fill in a substantial "hole" in the active site, and thus it can be speculated that this is the region where the sugar of the dTDP-4-keto-6-deoxyglucose is positioned. As displayed in Fig. 3c, there is an intriguing cluster of histidine residues located near this hole. Two of these histidine residues, 49 and 51, are absolutely conserved among the amino acid sequences reported thus far for putative sugar isomerases. To test their possible roles in catalysis, three sitedirected mutant proteins , H49N, H51N, and H49N/H51N, were subsequently constructed, and their structural and biochemical properties analyzed.
Activities of the Site-directed Mutant Proteins-To test the activity of all of the recombinant proteins, the HPLC elution profile for the sequential reactions catalyzed by the dehydratase, RmlB, and wild-type FdtA (Scheme 2) was first compared with a control experiment to which no FdtA was added. As can be seen in Fig. 4a, in the absence of FdtA, the FIGURE 2. Nature of the subunit-subunit interface in FdtA. There are two major layers of ␤-sheet that dominate the overall tertiary structure of the FdtA subunit. One side of the layer, composed of ␤-strands 4, 6, 9, and 11, is involved in the formation of the subunit-subunit interface. Shown in a is a ribbon representation of this face for subunit 1. The corresponding surface representation is given in b with carbons, oxygens, and nitrogens color-coded in gold, red, and blue, respectively. As can be seen, the face of the ␤-sheet that forms the subunit-subunit interface of the dimer is markedly hydrophobic. C-term, C terminus; N-term, N terminus.
HPLC elution profile contained two large peaks. ESI mass spectrometry confirmed that the larger peak (Fig. 4a, peak 1) corresponded to both the substrate (dTDP-glucose, 563 g/mol) and the product (dTDP-4-keto-6-deoxyglucose, 547 g/mol) of the dehydratase reaction. In Fig. 4a, the peak denoted by an asterisk corresponded to dTMP, which was present in the laboratory-prepared dTDP-glucose samples used for the assays. Addition of FdtA resulted in two new peaks (Fig.  4a). The main compound in the new peak (Fig. 4a, peak 2) had a molecular weight identical to that of the 4-keto derivative, indicating that wild-type FdtA was active. In Fig. 4a, peak 3 corresponded to dTDP. The presence of dTDP is consistent with a previous study by Melancon et al. (11), where it was noted that the 3-keto derivative is inherently unstable and readily breaks down to yield the free sugar and dTDP.
To further confirm the above results, the previous experiments were repeated but this time in the presence of FdtB. As noted in Scheme 2, FdtB is a pyridoxal phosphate-dependent aminotransferase that functions on the 3-keto position of the sugar. The HPLC elution profile for the reaction lacking FdtA consisted of two distinct peaks, a sharper, larger peak and a second broader peak (Fig.  4b). On the basis of ESI mass spectrometry, Fig. 4b, peak 1, contained both dTDP-glucose and the 4-keto-6deoxy derivative. The second peak (Fig. 4b, peak 3) contained a compound with a molecular weight of 546.1 g/mol, which is consistent with the replacement of the keto group with an amine and indicates that the aminotransferase reaction is not strictly specific for the 3-position of the sugar. Addition of FdtA resulted in a new peak (Fig. 4b, peak 2) and a decrease in the area of peaks 1 and 3. The molecular weight of the compound eluted in the new peak was consistent with that of dTDP-3amino-3,6-dideoxy-␣-D-galactose, the product of the FdtB reaction.
(a) (b) 10   The relative activities of the mutant forms of FdtA, as compared with the wild-type enzyme, were determined via the same assays. Addition of the H49N mutant form of FdtA resulted in an HPLC elution profile identical to that of the control reaction lacking FdtA, indicating a complete loss of activity (Fig. 4, a and  b). The H51N FdtA mutant protein retained some activity as indicated by the HPLC elution profile. As expected, the H51N/ H49N double mutant protein exhibited no activity within the limits of the assay.
Three-dimensional Structures of the Mutant Proteins-The activity assays clearly indicated a reduction in the catalytic activity of the H51N mutant protein and a loss of activity with the H49N mutation. To ensure that these changes in activities were not because of major perturbations of the active site geometries of the respective proteins, their three-dimensional structures were determined and refined to ϳ2.5 Å resolution. For each structure, the electron density maps were well ordered with only minor changes in side chain orientations. Importantly, the active site geometries for the mutant proteins and the wild-type enzyme were nearly identical (supplemental Fig.  2, a-c). The ␣-carbons for the wild-type dimer and the H49N, H51N, and H49N/H51N proteins superimpose with root mean square deviations of 0.25, 0.26, and 0.27 Å, respectively. Hence the loss of catalytic activity with the H49N mutant protein is not a result of gross structural perturbations.

DISCUSSION
Sugars represent the most abundant biomolecules in the Earth's biomass and are involved in such important physiological processes as the immune response, cell-cell interactions, and fertilization. In recent years, there has been considerable research interest in the more unusual di-and tri-deoxysugars that possess antigenic or antibiotic activities (25). In general, the 2,6-dideoxy-and 2,3(4),6-trideoxysugars are involved in antibiotic biosynthesis (9,26), whereas the 3,6-dideoxysugars are found predominantly on the outer surface of the cell membranes of both Gram-positive and Gram-negative bacteria (27). In Gram-negative bacteria, these modified sugars are found in the O-antigen, the outermost portion of the lipopolysaccharide (28). In Gram-positive bacteria, which do not synthesize lipopolysaccharide, these sugars are found attached to S-layer proteins (29).
The pathways involved in the production of these unusual sugars represent a rich source of intriguing enzymes. FdtA, the focus of this study, represents a novel type of isomerase first characterized by Pfoestl et al. in 2003 (10). Strikingly, as indicated in Scheme 2, whereas glucose 1-phosphate serves as the starting material for the production of dTDP-Fucp3NAc, the stereochemistry about C-4 of the sugar changes to that of galactose in the reaction catalyzed by FdtA. As noted previously, the biochemical properties for a similar 3,4-ketoisomerase from the D-mycaminose biosynthetic pathway of S. fradiae have recently been reported (11). In Tyl1a, however, the stereochemistry about C-4 is retained.
From the x-ray crystallographic and biochemical data presented here, it is now known that both His 49 and His 51 in FdtA play critical roles in catalysis. On the basis of these data, we have modeled the substrate of FdtA into the active site cleft as presented in Fig. 5a. This modeling was accomplished by first anchoring the nucleoside portion of the substrate into a similar position as that observed for dTDP. By a series of rotations about the phosphoryl groups of the substrate, it was possible to position C-3 of the sugar near the imidazole nitrogen of His 49 . As can be seen, this model also places His 51 near the sugar C-3 oxygen and His 95 near the sugar C-4 oxygen.
A possible mechanism for FdtA can be envisioned whereby His 49 removes the hydrogen from the sugar C-3 and shuttles it to the sugar C-4 on the same side of the glycosyl group. This would result in inversion of configuration about C-4. His 51 might function in catalysis by shuttling protons between the C-3 and C-4 oxygens. The postulated role for His 49 is consistent with the lack of measured enzymatic activity when it is substituted with an asparagine. The fact that the enzymatic activity of FdtA is considerably reduced when His 51 is changed to an asparagine is also consistent . Activity assays for wild-type and mutant forms of FdtA. Shown in a are the HPLC elution profiles for the reactions catalyzed by FdtA. Compounds were identified by mass spectrometry. Small amounts of dTMP were present in the dTDP-glucose used as the substrate (denoted by the asterisk). Peak 1 corresponds to dTDP-glucose and dTDP-4-keto-6-deoxyglucose; peak 2 corresponds to dTDP-3-keto-6-deoxygalactose; peak 3 corresponds to dTDP. Shown in b are the HPLC elution profiles for the reactions catalyzed by FdtA followed by the aminotransferase. Compounds were identified by mass spectrometry. Peak 1 corresponds to dTDP-glucose and dTDP-4-keto-6-deoxyglucose; peak 2 corresponds to dTDP-3-amino-3,6-dideoxygalactose; peak 3 corresponds to dTDP-4-amino-4,6-dideoxygalactose. with its proposed role as a proton shuttle. The residual activity in the H51N mutant protein might result from another residue, such as His 95 , fulfilling the proton transfer role, albeit at a less-thanoptimal rate. Additional site-directed mutagenesis and x-ray crystallographic analyses are presently underway to further clarify the catalytic mechanism of FdtA.
Both FdtA and Tyl1a function on the same substrate and catalyze similar isomerization reactions, although as noted in Tyl1a, the configuration about C-4 is retained. It can thus be speculated that in Tyl1a there must be two active site bases situated on opposite faces of the sugar. One base would be required for abstracting the hydrogen from the sugar C-3 and donating it to the C-4 oxygen, whereas the other would be required to abstract the hydroxyl group hydrogen on C-3 and transfer it to C-4. On the basis of an amino acid sequence alignment, it is known that Tyl1a contains two histidines that are homologous to His 49 and His 51 in FdtA, but the similarities between these two proteins extend far beyond these two residues (supplemental Fig. 1). An expanded view of the FdtA active site is given in Fig. 5b. The only differences are at Ile 11 , Phe 48 , His 95 , and Met 97 in FdtA, which are His 25 , Gly 62 , Arg 109 , and Leu 111 , respectively, in Tyl1a. Of particular interest is the His 95 to Arg 109 substitution in Tyl1a. It is possible that this substitution results in a modified active site and/or sugar binding mode in Tyl1a that allows His 63 (His 49 in FdtA) to function as the base to abstract the hydrogen from C-3 and to transfer it to the C-4 oxygen. Likewise, His 65 (His 51 in FdtA) would function as a base to abstract the hydrogen from the C-3 hydroxyl group and to transfer it to C-4. In combination, this would lead to retention of configuration about C-4. A structural analysis of Tyl1a is clearly necessary to address the intriguing question regarding the manner in which two isomerases using the same substrate catalyze reactions that either invert or retain the configuration about C-4.
In summary, the model of FdtA described here serves as a structural paradigm for a new family of sugar isomerases. The formation of the dimer results from classical domain swapping, and two strictly conserved histidine residues, His 49 and His 51 in FdtA, play key roles in catalysis. On the basis of amino acid sequence alignments with putatively annotated sugar isomerases (supplemental Fig.  3), we have found the following characteristic signature sequence, RGX-HAH(K/R)X(L/I)XQX 6 GS, where X can be any amino acid. In FdtA, this sequence, which contains the two conserved histidine residues, forms part of the active site. This sequence will serve as a hallmark for these types of sugar isomerases, and it will be of value in assigning function to additional uncharacterized ORFs as they become available. Furthermore, it can be speculated that those enzymes that retain configuration about C-4 of the sugar, such as Tyl1a, contain an arginine at the homologous position of His 95 in FdtA, whereas those that are inverting contain a histidine residue.