The tetrameric structure of sialic acid–synthesizing UDP-GlcNAc 2-epimerase from Acinetobacter baumannii: A comparative study with human GNE

Sialic acid presentation on the cell surface by some pathogenic strains of bacteria allows their escape from the host immune system. It is one of the major virulence factors. Bacterial biosynthesis of sialic acids starts with the conversion of UDP-GlcNAc to UDP and ManNAc by a hydrolyzing 2-epimerase. Here, we present the crystal structure of this enzyme, named NeuC, from Acinetobacter baumannii. The protein folds into two Rossmann-like domains and forms dimers and tetramers as does the epimerase part of the bifunctional UDP-GlcNAc 2-epimerase/ManNAc kinase (GNE). In contrast to human GNE, which showed only the closed conformation, the NeuC crystals contained both open and closed protomers in each dimer. Substrate soaking changed the space group from C2221 to P212121. In addition to UDP, an intermediate-like ligand was seen bound to the closed protomer. The UDP-binding mode in NeuC was similar to that in GNE, although a few side chains were rotated away. NeuC lacks the CMP-Neu5Ac–binding site for allosteric inhibition of GNE. However, the two enzymes as well as other NeuC homologues (but not SiaA from Neisseria meningitidis) appear to be common in tetrameric organization. The revised two-base catalytic mechanism may involve His-125 (Glu-134 in GNE), as suggested by mutant activity analysis.

Sialic acids are a family of nine-carbon ␣-keto sugars with 1-carboxylate group. More than 50 different species of sialic acids are found in nature, among which N-acetylneuraminic acid (Neu5Ac) 2 is the most common (1). These negatively charged sugars are important in mammalian cell-cell recognition when incorporated as the terminal component of surface glycoconjugates (2). Tumor-derived sialic acids can disable the killing mechanisms of effector immune cells (3). Expression of sialic acids on the bacterial cell surface provides a way to escape the host immune system (4). A variety of virus-host interactions also involve sialic acids (5). The biosynthesis of sialic acids starts with hydrolytic epimerization of N-acetylglucosamine (GlcNAc), catalyzed by UDP-GlcNAc 2-epimerase and producing N-acetylmannosamine (ManNAc), which then reacts with phosphoenolpyruvate to form Neu5Ac (6,7). Glucose and mannose are epimers at C2, and hence the enzyme is a 2-epimerase. In mammals, ManNAc is phosphorylated at C6 by the kinase part of the bifunctional enzyme GNE, and an additional step is required to separate the 9-phosphate group form Neu5Ac. Subsequently, the sugars are activated by CMP-sialic acid synthetases, ready to be presented on the cell surface. The downstream product CMP-Neu5Ac also acts as a feedback inhibitor to regulate the GNE activity. Impaired inhibition can result in sialuria (8).
The Gram-negative coccobacillus Acinetobacter baumannii is an opportunistic pathogen that causes nosocomial (i.e. originating in the hospital) urinary tract and bloodstream infections, ventilator-associated pneumonia, and meningitis (9). The emergence of A. baumannii strains that can resist treatment with antibiotics, as a result of acquisition and expression of ␤-lactamase genes, has posed a significant problem in eliminating the bacteria (10). Secretion of capsular polysaccharide and formation of biofilm can further protect the bacteria from antibiotics and the host immune system (11). Besides, the genome of some A. baumannii strains, as well as those of other pathogenic strains of Pseudomonas aeruginosa, Escherichia coli, and Neisseria meningitidis, for example, contains genes that encode enzymes for sialic acid biosynthesis (12). The bacterial enzymes in the sialic acid pathway are named NeuC, NeuB, and NeuA, which correspond to UDP-GlcNAc 2-epimerase, sialic acid synthetase, and CMP-sialic acid synthetase (Fig. 1). The crystal structures of prokaryotic NeuB and NeuA have been solved for N. meningitidis (PDB codes 1XUU and 1EYR) (13,14). The structure of a mammalian CMP-Neu5Ac synthetase is also known (PDB code 1QWJ) (15). However, no NeuC structure has been solved until now. Only the related structure of human GNE is available (PDB code 4ZHT) (16).
Whereas the substrate for nonhydrolyzing epimerase and hydrolyzing epimerase is the same, in this case UDP-GlcNAc, the product of the former is UDP-ManNAc and that of the latter is ␣-ManNAc plus UDP. As a hydrolyzing enzyme, NeuC is different from the nonhydrolyzing epimerase for bacterial cell wall manufacturing (6). Despite the limited homology, the two classes of epimerases share a common protein fold and dimeric organization (16). Crystal structures of the nonhydrolyzing enzyme are known for at least 10 species (PDB codes 1F6D, 1O6C, 1V4V, 3BEO, 3DZC, 3OT5, 4HWG, 4NEQ, 5DLD and 5ENZ). Each protomer contains two Rossmann-like domains. The binding of UDP-GlcNAc to an allosteric site stabilizes the closed, active conformation of the enzyme, whereas the open form is not active (17). Open and closed conformations are believed to affect the hydrolyzing enzyme activity as well, but so far the precise relationship remains unclear. In addition to dimer, the epimerase part of human GNE formed a tetramer in the crystal, which contained UDP and CMP-Neu5Ac (16). The feedback inhibitor binds to the dimer-dimer interface and locks the tetramer into a tightly closed and inactive conformation. This is distinct from the activation mechanism of the nonhydrolyzing enzyme through a closed conformation.
To investigate the protein conformations of bacterial hydrolyzing UDP-GlcNAc 2-epimerase, as well as the underlying catalytic and regulatory mechanisms, we cloned, expressed, purified, and crystallized the NeuC protein from A. baumannii, and we solved its structure. Unexpectedly, not only did NeuC turn out to be a dimer, but the dimers also assemble into a tetramer, which bears striking similarity to that of human GNE. The substrate-binding mode was also investigated by soaking experiments. Analyses of the crystal structures suggest a conserved catalytic mechanism employed by NeuC and GNE.

Overall structure and tetramer formation
The two NeuC protomers in the C222 1 crystal were modeled with 370 and 366 amino acid residues. The structure was refined to acceptable polypeptide geometry and low R-values (Table 1). Each protomer comprises two Rossmann-fold domains. The N-terminal domain contains seven parallel ␤-strands (␤1-␤7) connected by six ␣-helices (␣1-␣6) with a topology of ϩ1x/ ϩ1x/Ϫ3x/Ϫ1x/Ϫ1x/Ϫ1x. The order of strand arrangement is ␤3-␤2-␤1-␤4-␤5-␤6-␤7. The C-terminal domain contains six parallel ␤-strands (␤8 -␤13) linked by five ␣-helices (␣9 -␣13) with a similar topology of ϩ1x/ϩ1x/Ϫ3x/Ϫ1x/Ϫ1x, and it is connected to the N-terminal domain by a loop between helices ␣7 (as part of the N domain) and ␣8 (in the C domain). The C-terminal segment of the protein, which includes an additional helix ␣14, is associated with the N-terminal domain ( Fig.  2A). The loop between helices ␣13 and ␣14 in both protomers lacked clear electron density in the Fourier maps, suggesting high flexibility in this region. Moreover, the two protomers showed different overall conformation: one is open and the other is closed (Fig. 2B). If they are superimposed directly, the root-mean-square deviation (r.m.s.d.) is 2.73 Å for 359 matched pairs of C␣ atoms. However, if the N-and C-terminal domains are superimposed separately, the r.m.s.d. are significantly reduced to 0.193 and 0.466 Å for 182 and 140 pairs of C␣, respectively, indicating that the conformational change is largely the result of rigid-body domain movements.
The two NeuC protomers form a dimer in a way similar to those of bacterial nonhydrolyzing UDP-GlcNAc 2-epimerase and human GNE. The dimerization mainly involves N-terminal domain association in a highly symmetrical manner (Fig. 3A). By excluding the C-terminal domain, the NeuC dimer can turn 180°and superimpose on itself with an r.m.s.d. of 0.197 Å for 364 matched pairs of C␣ atoms (Fig. S1). It is significantly lower than the 3.47-Å r.m.s.d. for 716 pairs of C␣ if the entire dimer is compared, apparently because of the conformational difference between the open and closed protomers. Dimer formation buries 1970-Å 2 surface areas on each protomer, involving at least 40 amino acid residues. The interface is contributed mainly by helices ␣3, ␣4, and ␣5, as well as parts of helix ␣6 and the C-terminal segment (Fig. 3B). It is centered at a hydrophobic core with the side chains of Val-76, Met-79, Leu-83, Ala-87, Leu-108, Ala-113, Leu-115, Ile-116, Met-117, Tyr-134, Ile-138, Ala-141, and Met-145 (Fig. 3C). Buried deep in this core are the hydrogen-bonded side chains of Gln-112 and Gln-112* (asterisks denote residues from a different protomer). At the rim of the interface, four salt bridges are formed between two pairs of Lys-77-Asp-88* and Arg-159 -Asp-377* (Fig. 3B). Besides these major interactions, the side chain of Gln-162 is hydrogenbonded to the backbone of Tyr-376* (CO) and Leu-378* (NH), and the side chain of Phe-375 is in contact with those of Ile-129* and Leu-163*. Near the C terminus, the side chains of Leu-378 -Leu-378* are only 3.9 Å apart.
The NeuC dimer further forms a tetramer with another dimer related by a crystallographic dyad symmetry. For clarity, a model of the tetramer is constructed by renaming the symmetry-related protomers AЈ and BЈ to protomers C and D. It is interesting to note that the organization of NeuC tetramer is also similar to that of human GNE, with 222 point-group symmetry. A tetrameric model of human GNE likewise obtained via crystallographic dyad symmetry shows an r.m.s.d.

Tetrameric hydrolyzing UDP-GlcNAc 2-epimerase
and 24 in B/D) are involved. The dimer-dimer interface is mainly contributed by the ␤2-␣2 loop, strand ␤3, the ␤3-␣3 loop, and helix ␣3. A central hydrophobic core is formed by the nonpolar side chains of Leu-67, Val-81, and Leu-84 from all four protomers (Fig. 3D). The other interactions are mostly polar, including a salt bridge between the Glu-61 and Arg-92* side chains and a backbone hydrogen bond between Leu-66 (CO) and Ser-68* (NH; Fig. 3E). Depending on the conformation being open or closed, the side chain of Glu-64 forms alternate hydrogen bonds with those of Ser-69*, Thr-71*, and Ser-73*. In addition, Pro-43 of protomer A (or C) is in contact with Thr-247* and Lys-251* of protomer D (or B).

Substrate soaking and ligand-binding mode
In addition to 572 water molecules, the native NeuC crystal structure contains two lithium ions and two sulfate ions in the solvent model (Fig. S3). These ions presumably come from the crystallization buffer. Each bound lithium ion is coordinated by four peptide carbonyl groups in the ␣1-␤2 loop, adjacent to the N terminus of each protomer (Fig. S3A). The sulfate ions are each bound to Ser-290 at the N terminus of helix ␣12 and the positively charged side chain of Arg-309, both in the C-terminal domain (Fig. S3B). The sulfate-binding site in NeuC correlates very well with the location of the pyrophosphate group of UDP in the active site of human GNE.
To investigate the precise substrate-binding mode of NeuC, co-crystallization experiments with UDP-GlcNAc were carried out but failed to yield suitable crystals for X-ray data collection. Neither did longer-time soaking experiments with native NeuC crystals turn out Fourier maps with significant density for the substrate. Interestingly, the space group of the crystals was transformed from C222 1 to P2 1 2 1 2 1 after the soaking. The new crystals contained one tetramer in an asymmetric unit. When the soaking time was reduced to about 1 min, a data set was obtained from a crystal with space group P2 1 2 1 2 1 , which showed densities for two bound UDPs. Finally, by limiting the soaking time to seconds, some density for the GlcNAc moiety was seen. The diffraction images could only be processed in P2 1 2 1 2 1 , but the resulting data set contained mostly zero or negative intensities when the sum of h ϩ l was odd. The electron density map based on a preliminarily refined model a Statistics for data from the highest-resolution shell are shown in parentheses. b R sym ϭ (⌺⌺͉I hkl Ϫ ͗I͉͘)/(⌺I hkl ), where the average intensity ͗I͘ is taken over all symmetry-equivalent measurements, and I hkl is the intensity for any given reflection. c Data set was collected and processed in space group P2 1 2 1 2 1 , whereas C222 1 was used in the refinement.
where F o and F c are the observed and calculated structure factor amplitudes, respectively. e R free value was calculated by using only an unrefined subset of the reflection data (5%).

Tetrameric hydrolyzing UDP-GlcNAc 2-epimerase
showed two virtually identical dimers, and the crystal was considered isomorphous to the native. Consequently, those odd reflections were omitted, and the indices k and l were switched to make a C222 1 data set. The NeuC tetramer of the P2 1 2 1 2 1 crystal was compared with that of the native C222 1 crystal. The r.m.s.d. between 1103 matched pairs of C␣ atoms is 0.240 Å. Again, the C-terminal domains of protomers A and C were excluded upon structural superposition for their outstanding deviations. These protomers also adopted the open conformation in the P2 1 2 1 2 1 crystal, but they became even more open than the protomer A in the C222 1 crystal (Fig. S2B). The bound sulfate ions remained with the open protomers A and C, whereas a bound UDP molecule was observed in each active site of the closed protomers B and D (Fig. 4, A and B). The electron densities for the sugar and base of the UDP are comparatively weaker than those for the pyrophosphate group. In protomer B, the side chain of Arg-11 is in close contact with the uracil, causing disorder in this region. However, it is rotated and directed away from the active site in protomer D (Fig. S4A). In the substrate-soaked C222 1 crystal (C␣ r.m.s.d. ϭ 0.135 Å from the native) UDP was found in the active site of each protomer (Fig. 4, C and D). The binding of UDP to the open protomer A appears to be weaker, in which the Arg-11 side chain is away from the uracil base (Fig. S4B). The closed protomer B contains an additional GlcNAc moiety.
As shown in Fig. 5A, the ␤-phosphate group of UDP interacts with the protein in a virtually identical way as that for the sulfate in the native structure, making four hydrogen bonds to Ser-290 and Arg-309. The ␣-phosphate makes additional hydrogen bonds to the side chains of Arg-11 and His-210. The planar guanidinium group of Arg-11 appears to stack with the uracil base, which in turn stacks with the side chain of Tyr-275, making an arginine-uracil-tyrosine sandwich. The uracil group forms two hydrogen bonds to the backbone NH and CO groups of Phe-270. Some variations of the hydrogen bonds to the uracil, which requires tautomerization, were also observed (Fig.  S5). Most of the UDP-binding residues in human GNE except Ser-23 are found in NeuC (Fig. S6). The UDP interactions with NeuC are fewer than those observed in the GNE structure, probably because the active site of NeuC needs structural rearrangements to accommodate the substrate, but the conformational changes are limited by the crystal lattice packing. The enzyme can form similar hydrogen bonds to the ␤-phosphate and the ribose OH groups using the Arg-104 and Glu-295 side chains if they are properly reoriented. In Fig. 5B, the GlcNAc part of the substrate is disconnected from UDP. The N-acetyl Figure 3. Oligomer interface. A, two protomers in a NeuC dimer are shown as cyan and green models. The other dimer that makes a tetramer is colored magenta and yellow. B, dimer interface involves helices ␣3, ␣4, ␣5, and ␣6 and the C terminus. Some polar interactions are depicted by using sticks. C, side chains that constitute the hydrophobic core at the dimer interface are shown as sticks. Only those from the cyan protomer are labeled. D, in a tetramer, the dimer-dimer interface is centered at strand ␤3 and helix ␣3. The hydrophobic core is formed by side chains from the four protomers. E, polar interactions are also found at the tetramer interface.

Tetrameric hydrolyzing UDP-GlcNAc 2-epimerase
group makes a hydrogen bond to the ␤-phosphate of UDP and also interacts with the side chains of Leu-16 and Leu-101. The OH groups of the sugar are hydrogen-bonded to the side chains of Glu-128, Arg-139, and Ser-290, as well as the backbone CO of His-125. The side chain of Asp-103 lies adjacent to the C1 and O5 of GlcNAc, but Asp-135 does not interact directly with the sugar. Most GlcNAc-binding residues of NeuC are found in GNE as well (Fig. S7).

Effects of site-specific mutation on catalytic activity
The catalytic mechanism of hydrolyzing UDP-GlcNAc 2-epimerase has been extensively studied for human GNE (18), and the biochemical characterization has also been reported for two bacterial enzymes from E. coli (NeuC) and N. meningitidis (SiaA) (19,20). Strictly speaking, these hydrolyzing enzymes are not true epimerases as are the nonhydrolyzing enzymes that catalyze reversible epimerization of UDP-GlcNAc and UDP-ManNAc. The hydrolyzing epimerization is believed to proceed by a two-base mechanism. The substrate first undergoes anti elimination in which the first base deprotonates the C2 carbon of the sugar, forming the 2-acetamidoglucal intermediate and UDP. It is followed by syn hydration, catalyzed by the second base, on the other side of the sugar ring (18). The GlcNAc-binding mode in A. baumannii NeuC suggests that the ␤-phosphate of UDP may act as the first base, which is probably assisted by the invariant Ser-290. Conserved in all hydrolyzing UDP-GlcNAc 2-epimerases but not found in the nonhydrolyzing enzymes, Arg-104 may bind to the ␤-phosphate and serve to stabilize the leaving UDP (16). The reduction and lack of catalytic activity in the human GNE mutants S302A and R113A underscore the functional importance of the equivalent Ser-290 and Arg-104 in NeuC. From the decrease of k cat value by 150-fold in the GNE mutant D112A, it is likely that Asp-103 of NeuC plays an important role in catalysis, probably as the second base.
However, the catalytic role of Asp-135 in NeuC is not clear, although the corresponding GNE mutant D143A completely lost activity. The NeuC-GlcNAc-binding mode cannot explain another inactive GNE mutant E134A, as the equivalent residue is His-125 in NeuC. In a previous model, the Glu-134 side chain of GNE made a hydrogen bond to the 4-OH group of ManNAc  Tetrameric hydrolyzing UDP-GlcNAc 2-epimerase (16). However, in the complex structure of NeuC, the side chain of His-125 is not in direct contact with GlcNAc, but the backbone CO makes a hydrogen bond to the sugar's 6-OH group. The 4-OH is hydrogen-bonded to the Glu-128 side chain instead. To investigate the roles of these two residues in substrate binding and catalysis, two NeuC mutants H125A and E128A were produced and their catalytic parameters measured, along with the WT enzyme ( Table 2). Compared with human GNE, the bacterial enzyme showed a lower k cat by about 4-fold and a much higher K m by about 30-fold, under the experimental conditions. The lower activity of NeuC than GNE might imply that sialic acid biosynthesis is not an essential pathway for bacterial growth, but the nine-carbon sugar is an important metabolite in mammals, where the production is also stringently controlled by feedback inhibition. Compared with WT NeuC, the mutant H125A showed a reduced k cat value to nearly one-half and a 2.7-fold higher K m . The overall reduction of activity to about 20% suggests that His-125 plays an essential role in both substrate binding and catalysis, and it does so not only by making a backbone hydrogen bond to the sugar. In contrast, the mutant E128A showed a 14-fold increase in k cat despite the 4-fold increase in K m . These results suggest that the hydrogen bond to the sugar's 4-OH group contributes to substrate affinity but is not required for catalysis.

Discussion
Because most active-site residues are conserved in NeuC and GNE, these two enzymes as well as other homologues are supposed to share a common catalytic mechanism. The observation of a bound GlcNAc in the active site of NeuC allows a revised mechanism to be proposed (Fig. 6). First, because the ␤-phosphate of UDP is located above the sugar plane and closer to C2 than C1, upon Arg-104 -assisted bond breaking of UDP from GlcNAc, the phosphate may serve as the first base to abstract the C2 proton from the sugar, probably assisted by the nearby Ser-290. Second, Asp-103 may have another role than serving as the second base. Because the side-chain oxygen atoms of Asp-103 are in contact with the O5 of GlcNAc at distances of 2.9 and 3.1 Å, the negatively charged side chain can stabilize the positive charge of an oxocarbenium intermediate that may be transiently present after UDP elimination. Third, although the His-125 side chain of NeuC does not interact directly with the sugar, when it is rotated to a similar conformation as Glu-134 in GNE (Fig. S7), the imidazole (or carboxyl) group will be positioned below the sugar ring, where few polar residues are found, and serve as a suitable base for C2 re-protonation on the other side. Finally, according to the observed GlcNAc-binding mode, the base that activates a water molecule to attack C1 in the last step can be Asp-103 or His-125, rather than UDP and Ser-290. The involvement of oxocarbenium intermediate is the most likely mechanism of retaining glycosyltransferases (21). The hydrolyzing epimerases are similar to the retaining enzymes in some aspects, as the hydration step can be regarded as transfer of the sugar residue to a water molecule.
The similarity between NeuC and GNE is not limited to the active-site configuration but extends further to the conserved way of dimer and tetramer formations. Even the dispositions of the C-terminal segments are similar (Fig. S8). However, in a NeuC dimer, the loop beyond helix ␣14 turns inward and interacts with other residues near the protomer-protomer interface. The average B-value of the last 12 amino acid residues is 43.7 Å 2 , lower than the overall B-value of 47.8 Å 2 for the protein. In comparison, the corresponding loop in GNE has a much higher B-value than the overall average, indicating significant flexibility (16). It lacks specific interactions with the epimerase part and is supposed to swing out in order to connect with the kinase part, which also forms a dimer. Regarding the tetramer, the three nonpolar amino acids Leu-67, Val-81, and Leu-84 in the ␤3-␣3 loop and helix ␣3 of NeuC, which make up the hydrophobic core of the dimer-dimer interface, are equivalent to Val-76, Leu-90, and Val-93 in GNE (Fig. 7). Although not identical, these side chains pack snugly against one another in a similar way at the interface, which is strengthened by a number of polar interactions. A previous study showed that rat GNE

Tetrameric hydrolyzing UDP-GlcNAc 2-epimerase
protomers can reversibly associate into dimers and tetramers, whereas the presence of UDP-GlcNAc promoted tetramer formation (22). Crystallization of the epimerase part of human GNE as a tetramer further required CMP-Neu5Ac, which participated in the dimer-dimer interactions (16). In the absence of ligands, it existed predominantly as a dimer in solution.
Interestingly, in this study, NeuC was purified as a tetramer in the absence of UDP-GlcNAc, as shown by the gel-filtration profile (Fig. S9). Because the three nonpolar residues are found in the NeuC proteins from other species, including E. coli, presumably they function as similar tetramers as well. However, a sequence comparison of N. meningitidis SiaA with GNE and NeuC shows that the tetramer-forming interface may be absent in SiaA. In addition to a gap near the N terminus of helix ␣3 in the aligned sequence (Fig. 7), the two nonpolar residues in the middle of this helix are replaced by asparagine and threonine. Although an equivalent isoleucine residue to Leu-67 of NeuC (Val-76 of GNE) is found in SiaA, it may be insufficient to constitute a hydrophobic core between the two dimers. Consequently, SiaA is more likely to function as a dimer. In contrast to the negative cooperativity of the mammalian GNE, the strong positive cooperativity of UDP-GlcNAc binding to SiaA with a Hill coefficient of 1.94 suggests the presence of two substrate-binding sites and similar allosteric regulation as seen in the nonhydrolyzing enzyme (20). Even so, the two residues for interactions with the ␤-phosphate of UDP to facilitate substrate hydrolysis, equivalent to Arg-104 and Ser-290 in NeuC, are nevertheless found in SiaA (Fig. 7).
Despite the similarity in oligomerization and catalytic mechanism, there are a few significant structural and functional differences between NeuC and GNE. The NeuC protomers showed both open and closed conformations, whereas the GNE protomers were all closed (16). Because the N-terminal domain is arranged much the same way in both NeuC and GNE tetramers, apparently, formation of tetramer does not restrict the movements of the C-terminal domain for substrate binding and product release. A previous study showed that the E. coli enzyme was not inhibited by CMP-Neu5Ac (19). In human GNE, the side chains of three positively charged residues Arg-263, Arg-266, and Lys-267 from helix ␣10 at the dimer-dimer interface bind to the negatively charged CMP-Neu5Ac. The equivalents are Lys-251, Leu-254, and Asp-255 in NeuC, among which only one is positively charged. In GNE, the guanidine group of Arg-263 is salt-bridged to both the 1-carboxylate and the phosphate groups of the inhibitor. The single amino group of Lys-251 in NeuC would make the bonds much weaker, which alone is probably unable to lock the tetramer in a tightly closed conformation. Inspection of the pocket for sugar mononucleotide binding at the dimer-dimer interface of NeuC suggests a less polar environment for the ribose phosphate moiety, which fits well with a hydantoin ring, and a larger space for the cytosine base, likely sufficient for a purine group. Thus the cleft at the dimer-dimer interface may accommodate two cross-linked hydantoin molecules, each with an indole group attached. Hopefully, NeuC can be locked into an inactive closed tetramer like that of GNE but by less polar interactions than salt bridges.

Cloning, expression, and purification of NeuC
The neuC gene fragment of A. baumannii was amplified and cloned into pET21b (ϩ) vector, for the optimized expression of recombinant protein with an N-terminal His tag. Target protein was thereafter produced in E. coli BL21 (DE3) cell culture with 0.5 mM isopropyl 1-thio-␤-D-galactopyranoside induction for 16 h at 293 K. Cell pellet was harvested by centrifugation and then resuspended in a binding buffer of 50 mM Tris-HCl (pH 8.0), 500 mM NaCl, and 5 mM imidazole.
After disruption of the cell suspension by sonication, the clarified supernatant was purified by Ni 2ϩ -affinity and size-exclusion chromatography. The purified NeuC protein was stored in a buffer of 50 mM Tris-HCl (pH 8.0), 200 mM NaCl, 5% glycerol, and 2 mM tris(2-carboxyethyl)phosphine, and concentrated to 5 mg ml Ϫ1 as determined by the method of Bradford for crystallization experiments.

Crystallization, soaking, and data collection
Crystals of recombinant NeuC protein were grown from hanging drops containing equal volumes of protein and reservoir solution, using the vapor diffusion method. The optimized reservoir consisted of 0.2 M lithium sulfate, 0.1 M Tris (pH 8.5), and 25% PEG 3350. Native crystals appeared within 3-4 weeks

Tetrameric hydrolyzing UDP-GlcNAc 2-epimerase
at 277 K. The NeuC-UDP complex crystals were obtained by soaking the native crystals with 0.2 mM UDP-GlcNAc for 10 -20 min. The crystals were then picked up and immersed briefly in the reservoir solution supplemented with 20% glycerol before flash-cooling in liquid nitrogen. X-ray diffraction data collection was performed at beamlines BL13B1 and BL13C1 in NSRRC (Hsinchu, Taiwan). The diffraction data were properly processed using the HKL2000 program suite (23) with data statistics as summarized in Table 1.

Structural determination and refinement
The native NeuC structure of A. baumannii was solved by molecular replacement with the PHENIX AutoMR (24) using the previously published human homologue structure (PDB code 4ZHT (16)) as a search model. The substrate-soaked NeuC complex structure was determined by CNS (25,26), using a tetramer from the native NeuC structure as the search model. The UDP molecule was built into the extra electron density by using Coot (27). The crystal structures then underwent several rounds of manual model rebuilding and computational refinement with Coot and PHENIX. The figures were generated by PyMOL (28). Some refinement statistics and the PDB accession numbers are listed in Table 1.

Site-specific mutagenesis and activity measurement
The neuC gene from A. baumannii (ABneuC) was constructed into the pET28a expression vector and then used as a template for site-directed mutagenesis to generate its derivatives through PCR with primer pairs specific for substituting the residues His-125 and Glu-128 with Ala. The pET28a-AB-neuC construct and its mutant derivatives were each transformed into E. coli BL21 for heterologous expression. Activity of UDP-GlcNAc 2-epimerase was detected via the procedures as described before (16,29). Briefly, the reaction mixtures contained 50 mM Tris buffer (pH 7.5), 10 mM MgCl 2 , 0.2 mM NADH, 2 mM phosphoenolpyruvate (PEP), 2 units of pyruvate kinase, 2 units of lactic dehydrogenase, and various concentrations of UDP-GlcNAc. The final volume of each reaction was 100 l. The reaction was initiated by adding either 100 g of ABNeuC, 100 g of ABNeuC-H125A, or 40 g of ABNeuC-E128A. The initial reaction velocities were measured at 37°C for absorbance changes at 340 nm per min. Each data point was fitted into a Lineweaver-Burk plot to generate the kinetic parameters. The specific activity in this study was defined as the micro-molarity of product obtained per s/mg of enzyme.