Structure and mechanism of CTP:phosphocholine cytidylyltransferase (LicC) from Streptococcus pneumoniae.

Pneumococcal LicC is a member of the nucleoside triphosphate transferase superfamily and catalyzes the transfer of a cytidine monophosphate from CTP to phosphocholine to form CDP-choline. The structures of apo-LicC and the LicC-CDP-choline-Mg(2+) ternary complex were determined, and the comparison of these structures reveals a significant conformational change driven by the multivalent coordination of Mg(2+). The key event is breaking the Glu(216)-Arg(129) salt bridge, which triggers the coalescence of four individual beta-strands into two extended beta-sheets. These movements reorient the side chains of Trp(136) and Tyr(190) for the optimal binding and alignment of the phosphocholine moiety. Consistent with these conformational changes, LicC operates via a compulsory ordered kinetic mechanism. The structures explain the substrate specificity of LicC for CTP and phosphocholine and implicate a direct role for Mg(2+) in aligning phosphocholine for in-line nucleophilic attack and stabilizing the negative charge that develops in the pentacoordinate transition state. These results provide a structural basis for assigning a specific role for magnesium in the catalytic mechanism of pneumococcal LicC.

proteins that are secreted from the cells and subsequently attached to the cell surface by their homologous choline-binding domains (see Ref. 10 and references therein). These cholinebinding proteins are essential for many aspects of S. pneumoniae cell physiology including competence and stationary phase lysis.
The pathway for choline metabolism in S. pneumoniae and H. influenzae has been hypothesized to consist of a choline transport system, a choline kinase, CTP:phosphocholine cytidylyltransferase (CCT), and a choline phosphotransferase that transfers P-Cho from CDP-Cho to either lipoteichoic acid or lipopolysaccharide (5). The existence of this pathway is supported by the detection of choline kinase and CCT activity in crude extracts of S. pneumoniae (11,12). Genetic elements required for choline incorporation into the lipopolysaccharide of H. influenzae are found in the lic1 locus, which contains four open reading frames. The hypothesis drawn from the bioinformatic analysis of the lic1 locus (4) is that licA corresponds to choline kinase based on a 31% identity to the choline kinase of Saccharomyces cerevisiae over the short span of 40 amino acids between residues 222 and 262. The licB gene has several predicted transmembrane domains and is thus postulated to be a choline transporter. The hydrophilic licC gene product is a candidate for the CCT due to the resemblance of its amino terminus to the amino-terminal 60 residues of NTP transferase family members, leaving the licD gene as a candidate for the choline phosphotransferase. A homologous licC gene exists in S. pneumoniae (13), and the predicted LicC proteins of H. influenzae and S. pneumoniae are 37% identical and 60% similar.
Recently, the LicC of S. pneumoniae was purified and demonstrated to catalyze the CCT reaction (14). The enzyme possesses a high degree of selectivity for P-Cho and CTP, although phosphoethanolamine and ATP are poor substrates. LicC lacks homology to either the prototypical metazoan CCTs (15,16) or the CTP:glycerol-3-phosphate cytidylyltransferase from Bacillus subtilis (17). Rather, LicC is related to members of the NTP transferase superfamily that primarily activate sugars by transferring the NMP moiety to a phosphorylated acceptor substrate concomitant with the release of PP i (see the NCBI Conserved Domain Database, protein family 00483) and contain the N-terminal signature sequence KAX 5 GXGTRX 6 -9 K. The inclusion of LicC in this family is based on its similarity to the amino termini of Escherichia coli GalU and P. aeruginosa RmlA, two prototypical nucleotidyltransferases that form NDP-hexose derivatives that are used in the biosynthesis of bacterial cell wall components (14). Members of this superfamily are characteristically dimers or tetramers; however, LicC is a monomer (14).
The crystal structures of three members of the NTP trans-ferase superfamily are known. These are the dimeric GlmU (18) and methylerythritol-4-phosphate cytidylyltransferase from E. coli (19) and the tetrameric RmlA from both P. aeruginosa (20) and Salmonella enterica (21). The NMP transferase domains of these enzymes have a similar overall folding, and they share several key residues involved in substrate binding and/or catalysis, such as Gly 14 , Arg 18 , and Lys 28 (LicC numbering) in the signature sequence. Magnesium is required for activity. In S. enterica RmlA and E. coli methylerythritol-4phosphate cytidylyltransferase, Mg 2ϩ is proposed to position the nucleotide substrate during the NMP transfer reaction and to stabilize the PP i leaving group after the reaction (19,21); thus, Mg 2ϩ is not thought to play a role in the catalytic mechanism. In the other two structures, the magnesium binding site is undefined (18,20). Several positively charged residues that are not conserved in all of the NTP transferase structures are proposed to be involved in catalysis of RmlA and methylerythritol-4-phosphate cytidylyltransferase by providing a favorable electrostatic environment, although their exact role is unclear (19 -21). Exceptions are Arg 18 and Lys 28 (LicC numbering) in the signature sequence that bind the triphosphate group of the nucleotide substrate to stabilize the negative charge developed on the oxygens of the ␣-phosphate in the pentacoordinate transition state and the released PP i product. We have determined the kinetic mechanism of LicC from S. pneumoniae and the structures of LicC and the LicC⅐CDP-Cho⅐Mg 2ϩ ternary complex. The structures explain the specificity of LicC for CTP and identify the active site residues involved in binding the product, CDP-choline, with magnesium. We propose the mode of substrate binding and the direct role of the magnesium in catalysis based on the kinetic data and structures.

EXPERIMENTAL PROCEDURES
Materials-Sources of supplies were Sigma (for CTP, CDP-Cho, P-Cho, and seleno-L-methionine) and Amersham Biosciences, Inc. (for phospho[methyl-14 C]choline (specific activity 55 mCi/mmol)). Histagged native LicC was expressed and purified as described previously (14). For metabolic incorporation of selenomethionine into LicC, pET-15b containing the licC gene was transformed into BL21-Codon-Plus(DE3)-RIL-X (Stratagene), and protein was expressed as described previously (22) and purified in the same manner as the His-tagged native LicC. Protein concentration was determined by the method of Bradford (23). All other chemicals were reagent grade or better.
LicC Enzyme Assay-The enzymatic activity of LicC under different conditions was studied using the assay method described previously (14). The apparent K m for CTP or P-Cho in the presence of different concentrations of the second substrate was determined. In a final volume of 50 l, the LicC assay contained the substrates, CTP and phospho[methyl-14 C]choline (specific activity 3.68 mCi/mmol), at the appropriate concentrations for each study indicated in the figure legends, and 10 mM MgCl 2 in 150 mM bis-Tris-HCl, pH 7.0. LicC was added last to start the reaction. After incubation of the reaction mixture at 37°C for 10 min, the reaction was stopped by the addition of 5 l of 0.5 M EDTA. A 40-l aliquot of the reaction mixture was applied to a preabsorbent Silica Gel G plate, which was developed with 95% ethanol/2% ammonium hydroxide (1:1, v/v). The single product was visualized with a Bioscan Imaging detector and migrated with the standard CDP-Cho. Furthermore, different concentrations of either product (CDP-Cho and PP i ) of the LicC reaction were included in the assay to determine the order of the reaction.
Crystallization-Se-Met LicC was concentrated to 23.3 mg/ml in 50 mM Tris, pH 7.5, 1 mM dithiothreitol, 1 mM EDTA. The preliminary crystals of Se-Met LicC were obtained from Hampton Research screening kits, and the refined condition contained 0.1 M sodium citrate, 1.85 M ammonium sulfate, 0.2 M sodium/potassium tartrate, pH 5.8, at 18°C. Crystals appeared after several days, which grew further for 1 week. These crystals were transferred to a cryoprotectant solution that was made by adding 30% glycerol to the mother liquor for data collection. Co-crystals of LicC and CDP-Cho were grown by adding CDP-Cho and MgCl 2 to the protein to a final concentration of 10 mM each, and the mixture was incubated overnight at 4°C. The crystallization condition contained 0.4 M sodium acetate, 21% polyethylene glycol 4000, 0.1 M sodium cacodylate, pH 6.5, at 18°C. The rectangular crystals grew in 10 days. A cryoprotectant solution for cryodata collection was made by adding 22.5% glycerol to the mother liquor.
Data Collection and Structure Determination-Multiwavelength anomalous dispersion data were measured from a single crystal of Se-Met apo-LicC at beam line X8-C at the National Synchrotron Light Source, equipped with an ADSC Quantum 4R detector operating at 100 K. Two wavelengths near the selenium absorption edge, 0.9793 Å (the peak) and 0.9795 Å (the inflection point), were chosen by measuring the x-ray absorption spectrum of the protein crystal. The data sets from both wavelengths were collected, reduced, and scaled by the program HKL 2000 (24). The crystals of apo-LicC belong to space group P2 1 2 1 2 1 with unit cell dimensions of a ϭ 42.6 Å, b ϭ 63.6 Å, c ϭ 122.4 Å. The program SOLVE was used to scale the two wavelength data sets together, to locate selenium positions, and to calculate the protein phases (25). Two of the four selenium atoms in the LicC enzyme were confirmed. The program RESOLVE was used for reciprocal space solvent flattening (26). The statistics of data collection and phasing are summarized in Table I. The atomic model was built using the programs Warp (27) and Xtalview (28), followed by crystallographic refinement in the program XPLOR (29). The refinement steps included simulated annealing, conjugated gradient minimization, and individual B-factor refinement. The refinement statistics are shown in Table II.
Complete diffraction data from a single crystal of the LicC⅐CDP-Cho⅐Mg 2ϩ complex were measured at beamline 19-BM at the Advanced Photon Source, equipped with a SBC-2 CCD x-ray detector operating at 100 K. The program HKL 2000 (24) was used to integrate and scale the data set. The crystals of the ternary complex belonged to space group P1 with unit cell dimensions of a ϭ 48.2 Å, b ϭ 69.0 Å, c ϭ 81.6 Å, ␣ ϭ 93.4°, ␤ ϭ 92.8°, ␥ ϭ 97.1°. The data collection statistics are summarized in Table I. The structure of the CDP-choline-bound LicC was determined by the molecular replacement method using the refined apo-enzyme as a search model. The cross-rotation and translation functions and Patterson correlation refinement were calculated using the program XPLOR (29). Using the data between 15.0 and 4.0 Å, the rotation function, followed by Patterson correlation refinement, gave four outstanding solutions that correspond to four molecules of LicC. Because any point can be taken as an origin in space group P1, we rotated the search model according to the highest solution of rotation function and considered this rotated molecule as the first molecule. With fixing the first molecule, the orientations of the other three molecules were determined by applying the three independent noncrystal- is the ratio of the root mean square value of the calculated heavy atom structure factor (FH) to the root mean square value of the difference between calculated and observed derivative structure factors (E), where it is averaged not only over all reflections but over all phases for each reflection, weighted by the phase probability.
lographic symmetry operations to the first molecule, and then their positions relative to the first molecule were found by searching the x, y, and z translations in the unit cell. After the model, including all four molecules, was subjected to rigid body refinement at the resolutions between 6.0 and 4.0 Å in the program XPLOR, the resulting R factor was high (40.6%), suggesting the conformational difference of the CDPcholine-bound structure from the search model of the apo-enzyme structure. After the bulk solvent contribution was made to the intensities of reflections at low resolution, the high and low resolution limits were extended to 20 and 2.4 Å, respectively. Crystallographic refinement including simulated annealing, conjugate gradient minimization, and individual B-factor refinement was then performed at the extended resolutions. After the crystallographic refinement, the electron density map was improved by the noncrystallographic symmetry averaging method using the program DM (30). The F o Ϫ F c difference map clearly showed the electron density of CDP-Cho⅐Mg 2ϩ in all four molecules, verifying the correct molecular replacement solutions. Several iterations of model building and the crystallographic refinement gave the excellent refinement statistics (Table II). Noncrystallographic restraints were applied to the main chain of each molecule for simulated annealing refinement. The final structure contains four molecules of LicC (residues 4 -231 in each of four subunits).

RESULTS
Overview of the LicC Structure-The apo-LicC structure reveals a mixed seven-stranded ␤-sheet flanked by three ␣-helices on one side (␣1, ␣2, and ␣6) and another three ␣-helices on the other (␣3, ␣4, and ␣5) (Fig. 1). The seven-stranded sheet has six parallel strands (strands ␤1, ␤2, ␤3, ␤4, ␤5, and ␤7), and strand ␤6 is antiparallel to others, representing the typical nucleotide-binding fold (31). Four additional strands (␤5a, ␤5b, ␤5c, and ␤5d) located between strands ␤5 and ␤6, which are perpendicular to the seven-stranded ␤-sheet, and another four strands (␤1a and ␤1b, ␤4a, and ␤7a) are not part of the main ␤-sheet. Comparison of the apo-LicC structure with known structures in the Protein Data Bank using the DALI server (32) revealed that the two best matches were with two members of the NTP transferase superfamily: P. aeruginosa RmlA (Protein Data Bank code 1g0r, r.m.s. deviation 2.6 Å, C␣ atoms of 218 residues) and GlmU (Protein Data Bank code 1fwy, r.m.s. deviation 2.8 Å, C␣ atoms of 214 residues). The structures of two additional enzymes that belong to the NTP transferase superfamily were also compared using the program O (33): S. enterica RmlA (Protein Data Bank code 1iin, r.m.s. deviation 1.9 Å, C-␣ atoms of 171 residues) and E. coli methylerythritol-4phosphate cytidylyltransferase (Protein Data Bank code 1i52, r.m.s. deviation 1.7 Å, C-␣ atoms of 131 residues). The finding of these low r.m.s. deviation values indicates that LicC is structurally similar to the three members of the NTP transferase superfamily. The structural similarity along with the conservation of the N-terminal signature sequence supports the classification of LicC as a member of the NTP transferase superfamily (14). Among the three members of the NTP transferase superfamily homologous to LicC, GlmU is an uridyltransferase involved in the production of UDP-N-acetylglucosamine, an essential precursor for the cell wall and membrane biosynthesis in bacteria (18). RmlA is a thymidylyltransferase and the first enzyme in the biosynthesis of deoxy-TDP-L-rhamnose that is a component of the cell wall in many bacteria (20). E. coli methylerythritol-4-phosphate cytidylyltransferase is a cytidyltransferase involved in the mevalonate-independent pathway for isoprenoid biosynthesis (19).
The crystals of LicC⅐CDP-Cho⅐Mg 2ϩ ternary complex contain four molecules in an asymetric unit, assembled to form two dimers. However, purified LicC behaves as a monomer (14), and the apo-LicC structure contains one molecule per asymmetric unit, arguing that the dimers arise from the crystallization condition for the CDP-Cho⅐Mg 2ϩ complex. The four molecules in the asymmetric unit superimpose to one another with an r.m.s. deviation of between 0.64 and 0.75 Å, illustrating that they have nearly identical conformations. The hydrophilic nature of the dimer interface (ϳ60% of interactions) is also consistent with the conclusion that the monomer is the biologically functional unit. Additionally, sequence analysis of the NPT transferase superfamily members reveals that LicC lacks the oligomerization domain at the C terminus. The structures of apo-LicC and the LicC⅐CDP-Cho⅐Mg 2ϩ complex superimpose with an r.m.s. deviation of 1.51 Å, indicating that only small structural changes occur upon CDP-Cho⅐Mg 2ϩ binding to LicC.
The LicC Active Site-CDP-Cho⅐Mg 2ϩ is bound in a deep pocket with negative electrostatic potential ( Fig. 2A). Bound Mg 2ϩ is located between the pyrophosphate group of CDP-Cho and the active site pocket, neutralizing the charge repulsion between them. The negatively charged residues lining the active site pocket of LicC include Asp 105 , Asp 107 , Glu 135 , Asp 192 , Glu 216 , and Asp 218 . Among these residues, Asp 107 and Glu 216 are conserved within the NTP transferase family members except GlmU and E. coli methylerythritol-4-phosphate cytidylyltransferase. In GlmU, while Asp 107 is conserved, Glu 216 is changed to Gly, and in E. coli methylerythritol-4-phosphate cytidylyltransferase, Asp 107 and Glu 216 are replaced by Ala and Lys, respectively. Consistent with the conserved number of negatively charged residues, the active site electrostatic poten-  1. Overall stereoview of the LicC structure. Helices are shown in blue, strands in green, and other secondary structural elements in yellow. The core seven ␤-stranded sheets and flanking ␣-helices on either sides are labeled as ␤1-7 and ␣1-6. Additional ␤-strands that are not part of the core sheet are labeled by alphabetizing after the preceding ␤-strand number. For example, the ␤-strand inserted between ␤4 and ␤5 is labeled as ␤4a. Figs. 1, 3, and 4 were produced by the program RIBBONS (40). tial of LicC ( Fig. 2A) and RmlA (Fig. 2B) are highly negative. The GlmU active site (Fig. 2C) is slightly less negatively charged compared with LicC and RmlA. In contrast, the methylerythritol-4-phosphate cytidylyltransferase active site (Fig.  2D) is positively charged.
Interaction of the cytosine base of CDP-Cho with the enzyme explains the observed specificity of LicC for CTP (14). The O-2 oxygen of the cytosine base is hydrogen-bonded to the amide nitrogen of Ala 13 and the hydroxyl group of Ser 90 , and the N-4 nitrogen of the base is hydrogen-bonded to the carbonyl oxygens of Tyr 82 and Tyr 85 (Fig. 3A). It is clear that the N-4 nitrogen interaction excludes binding of other pyrimidine bases (i.e. thymine and uracil) because their corresponding atom (i.e. O-4 oxygen) cannot form hydrogen bonds with the carbonyl oxygens of Tyr 82 and Tyr 85 . There is no specific interaction of the C-5 and C-6 carbons of the cytosine base to the protein. The pyrimidine ring is stacked between the loop of strands ␤1 and ␤1a and the loop of strand ␤3 and helix ␣3, suggesting that larger purine bases can, if not ideally, also fit between these two loops. However, only the adenine base, and not the guanine base, is allowed, because the N-6 nitrogen of the adenine base, but not the O-6 oxygen of the guanine base, can form hydrogen bonds with Tyr 82 and Tyr 85 . Finally, the hydrogen bond interactions of 2Ј-and 3Ј-hydroxyl groups of the ribose sugar with the amide nitrogens of Gly 14 and Ala 106 and the carbonyl oxygen of Leu 11 secure the base of CTP, and potentially ATP, in the binding pocket.
It is known that the trimethylammonium group of choline is recognized by electron-rich aromatic rings (34 -36). Indeed, in the LicC⅐CDP-Cho⅐Mg 2ϩ complex, the binding pocket of the trimethylammonium group is formed by the aromatic rings of Tyr 190 and Trp 136 and the carboxyl group of Asp 192 . The side chains compensate for the positive charge of the trimethyl ammonium group through a stabilizing interaction with their electrons (Fig. 3B). We hypothesize that the negative charge of Asp 192 may also be a factor in binding of the trimethylammonium group, because an electrostatic bond is stronger than a cation-bond (37).
Magnesium is required for the catalytic activity of LicC (14). In the LicC⅐CDP-Cho⅐Mg 2ϩ structure, the magnesium atom is fixed in space by interactions with six oxygen atoms, two oxygens from each of the ␣and ␤-phosphates of CDP-choline, the carboxyl groups of Asp 107 , Glu 216 , and Asp 218 , and a water molecule, forming an octahedral geometry (Fig. 3B). This bound water molecule is further hydrogen-bonded to the carboxyl group of Asp 105 . The ␣-phosphate oxygen of CDP-Cho and the carboxyl groups of Asp 107 and Asp 218 , which bind the magnesium, form hydrogen bonds with the amino group of Lys 28 in the signature sequence (Fig. 3B), suggesting that Lys 28 participates in orienting the side chains of Asp 107 and Asp 218 for binding both magnesium and the nucleotide substrate. The interaction between the amino group of Lys 28 and the ␣-phosphate oxygen can enhance the catalytic reaction rate by stabilizing the developing negative charge in the pentacoordinate transition state. When the corresponding Lys of NTP transferases from other organisms was mutated, the catalytic activity was decreased (18,19), which is consistent with the proposed role of Lys 28 of LicC in catalysis. A ␤-phosphate oxygen Red and blue represent negative and positive electrostatic surface potential, respectively, whereas white represents neutral electrostatic surface potential. Bound ligands are shown with a ball and stick representations. A, the electrostatic potential of LicC with bound CDP-Cho⅐Mg 2ϩ . B, the electrostatic potential of RmlA from P. aeruginosa with bound dTDP-D-glucose (PDB code 1g1l). C, the electrostatic potential of GlmU with bound UDP-N-acetylglucosamine (PDB code 1fwy). D, the electrostatic potential of methylerythritol-4-phosphate cytidylyltransferase from E. coli with bound CTP (PDB code 1i52). We used the coordinates of the CTP-bound structure in D, because the product-bound structure is not available. This figure was produced with the program GRASP (41). that does not bind magnesium is hydrogen-bonded to the hydroxyl group of Tyr 190 that also interacts with the trimethylammonium group via its aromatic ring (Fig. 3B).
Structural Differences between LicC and LicC⅐CDP-Cho⅐Mg 2ϩ -The most significant change in the secondary structural elements induced by CDP-Cho⅐Mg 2ϩ binding involves the four strands (␤5a, ␤5b, ␤5c, and ␤5d) located between strands 5 and 6. In the LicC⅐CDP-Cho⅐Mg 2ϩ structure, strands ␤5a and ␤5b with their intervening loop become the continuous, long strand ␤5a, and strands ␤5c and ␤5d with their intervening loop coalesce to form strand ␤5c (Fig. 4A). These new, longer strands ␤5a and ␤5c are interconnected by a hydrogen bond network to form a two-stranded ␤ sheet.
CDP-Cho⅐Mg 2ϩ binding also alters the side chain positions of several residues. Side chains of Glu 216 and Asp 218 move toward bound magnesium by 7.4 and 3.8 Å, respectively, to coordinate the magnesium (Fig. 4B). Since Glu 216 in the apoenzyme structure is salt-bridged to Arg 129 , the movement of Glu 216 disrupts the salt bridge interaction with Arg 129 , displacing the guanidinium group of Arg 129 , breaking the hydrogen bond interaction of Arg 129 with the amide nitrogen of Gly 211 . The displaced guanidinium group of Arg 129 is hydrogen-bonded to the side chain of Glu 135 . Also, the side chains of Trp 136 and Tyr 190 move toward the trimethylammonium group of CDP-choline. Arg 129 , Glu 135 , and Trp 136 are all located in the newly formed strand ␤5a.
Arg 18 (LicC numbering) in the signature sequence of NTP transferase structures interacts with the nucleotide triphosphate (19 -21). In the apo-LicC structure, the guanidinium group of Arg 18 points down to the active site pocket, forming a hydrogen bond network with the carboxyl group of Asp 107 and the carbonyl oxygen of Ile 217 (Fig. 4C). In contrast, the guani-dinium group is flipped out and exposed to solvent in the LicC⅐CDP-Cho⅐Mg 2ϩ complex (Fig. 4C). Since the LicC⅐CDP-Cho⅐Mg 2ϩ complex is the structure of the enzyme intermediate immediately after the release of PP i (see below), the solventexposed guanidinium group of Arg 18 may correspond to the conformational change that facilitates the release of PP i to the  (34 -36). The closest distances between the side chains of three residues and the trimethylammonium groups are 3.5, 3.7, and 3.2 Å, respectively.

FIG. 4. Structural changes in the LicC⅐CDP-Cho⅐Mg 2؉ ternary complex.
A, when CDP-Cho⅐Mg 2ϩ binds to the apo-LicC (cyan), conformational changes in four strands (␤5a, ␤5b, ␤5c, and ␤5d) occur, resulting in the formation of two strands (␤5a and ␤5c, shown in red). First two strands (␤5a and ␤5b) are merged to be a new long strand ␤5a and the last two strands (␤5c and ␤5d) to be a new long strand, ␤5c. Additional conformational change occurs in ␤7a and its C-terminal loop (red). The secondary structural elements of the apo-LicC structure are labeled as in Fig. 1, and the new long strands ␤5a and ␤5c of the LicC⅐CDP-Cho⅐Mg 2ϩ ternary complex (red) are labeled and highlighted in an open rectangular box. Carbons of CDP-Cho are shown in yellow, and Mg 2ϩ is shown in magenta (ball and stick representations). B, the side chain conformations of several residues (gray in apo-LicC and blue in the LicC⅐CDP-Cho⅐Mg 2ϩ ternary complex) affected by binding of CDP-Cho⅐Mg 2ϩ are shown in ball and stick representations. The salt bridge between Arg 129 and Glu 216 in apo-LicC is broken in the LicC⅐CDP-Cho⅐Mg 2ϩ ternary complex due to the movements of Glu 216 along with Asp 218 for interacting with Mg 2ϩ , which are illustrated by arrows. Note that upon binding of CDP-Cho⅐Mg 2ϩ , the side chains of Trp 136 and Tyr 190 move inward to optimize the fit for the trimethylammonium group of Cho as it is shown in Fig. 3. C, Arg 18 in apo-LicC is anchored to the bottom of the active site by forming hydrogen bonds with the side chain of Asp 107 and the carbonyl oxygen of Ile 217 , and in the LicC⅐CDP-Cho⅐Mg 2ϩ ternary complex, the same residue swings out and is exposed to solvent. solvent. The side chain of the corresponding Arg in E. coli methylerythritol-4-phosphate cytidylyltransferase forms hydrogen bonds with the ␣and ␥-phosphate oxygens of CTP (19). In LicC, after adjusting the torsion angles of Arg 18 , similar interactions can be formed between the side chain of Arg 18 and the ␣and ␥-phosphate oxygens of CTP modeled using the CTP coordinates of E. coli methylerythritol-4-phosphate cytidylyltransferase (Protein Data Bank code 1i52). This finding suggests that Arg 18 may be involved in the release of PP i and offset the negative charge developed on the ␣-phosphate oxygen of CTP during the catalytic reaction.
Kinetic Mechanism for LicC-A series of kinetic experiments were performed to define the kinetic mechanism for LicC. Double reciprocal plots of 1/v versus 1/[CTP] at different concentrations of P-Cho (Fig. 5A) intersected far to the left of the 1/v axis and far below the 1/[CTP] axis. The same pattern is observed in case of 1/v versus 1/[P-Cho] in the presence of different concentrations of CTP (Fig. 5B). A cursory inspection of the graphs leaves the impression that the lines are parallel, which would indicate a ping-pong mechanism. However, they are not, and there are many cases of sequential enzymes where lines appear to be parallel. In these cases, the enzyme has a relatively high affinity for the leading substrate, but because of the relative values of the other rate constants the apparent K m value is much larger than the equilibrium binding constant (38). Product inhibition studies were performed to verify that LicC is a sequential enzyme and to establish the order of addition of substrates. CDP-Cho is a competitive inhibitor with respect to CTP, whereas PP i is a mixed-type inhibitor with respect to CTP (Fig. 6). These data show that CTP and CDP-Cho bind to the same form of the enzyme. On the other hand, both PP i and CDP-Cho are mixed-type inhibitors with respect to P-Cho (Fig.  7). This pattern of the product inhibition is characteristic of a sequential bi bi reaction with CTP as the leading substrate and CDP-Cho as the last product to leave, which is the same as the catalytic mechanism of E. coli glucose-1-phosphate thymidylyltransferase, a homologue of P. aeruginosa RmlA (39). DISCUSSION We propose a model for the catalytic cycle of LicC that is consistent with the kinetic and structural data (Fig. 8). CTP⅐Mg 2ϩ is the leading substrate. The LicC nucleotide specificity for CTP is conferred by specific interactions that include the N-4 nitrogen of the cytosine base and the carbonyl oxygens of Tyr 82 and Tyr 85 , the CTP O 2 oxygen with the amide nitrogen of Ala 13  that in turn forms hydrogen bond to Glu 135 . This induces four ␤-strands in LicC into coalesce into two long ␤-strands. This rearrangement causes the repositioning of Trp 136 and Tyr 190 to optimize the binding of the second substrate P-Cho and position the bound P-Cho for the CMP transfer reaction. The structure is consistent with the reaction occurring by in-line nucleophilic attack of the phosphate oxygen of P-Cho on the ␣-phosphate of CTP, similar to other members of the NTP transferase superfamily. The first product released from the active site is PP i , which may be facilitated by its interaction with Arg 18 , which points away from the active site in the product structure. CDP-Cho⅐Mg 2ϩ is the final product released from the enzyme, allowing the Glu 216 ⅐Arg 129 salt bridge to reform and regeneration of the apo-enzyme structure.
The structural changes in the LicC⅐CDP-Cho⅐Mg 2ϩ ternary complex are confined to the residues and secondary structural elements that create the binding sites for Mg 2ϩ and the trimethylammonium group of Cho. These important conformational changes initiate at the Mg 2ϩ binding site and are propagated to the binding site of the trimethylammonium group via four strands (␤5a, ␤5b, ␤5c, and ␤5d). LicC binds Mg 2ϩ by breaking the salt bridge between Glu 216 and Arg 129 in the apo-enzyme to release Glu 216 to coordinate the Mg 2ϩ . The side chain of Glu 135 forms a hydrogen bond with the side chain of Arg 129 , compensating for the absence of Glu 216 . Breaking the salt bridge triggers the rearrangement of four individual strands (␤5a, ␤5b, ␤5c, and ␤5d) into two longer strands (␤5a and ␤5c) (Fig. 4A). This reorganization displaces the side chains of Trp 136 on strand ␤5a and Tyr 190 on the loop of helices ␣4 and ␣5 to position P-Cho for an in-line nucleophilic attack of the phosphate oxygen of P-Cho on ␣-phosphorus of CTP. Phosphoethanolamine lacks the trimethylammonium group of P-Cho and therefore can fit in the active site. However, the positioning of phosphoethanolamine within the active site pocket for the CMP transfer reaction is predicted to be less efficient than P-Cho, accounting for the lower catalytic activity of LicC when phosphoethanolamine is the substrate (14).
Our structures expand the understanding of the catalytic mechanism of members of the NTP transferase superfamily by implicating Mg 2ϩ as an integral component in catalysis. The Mg 2ϩ in previously determined NTP transferase structures is postulated to stabilize the PP i leaving group, and thus plays an indirect role in catalysis (19,21). However, the Mg 2ϩ bound at the active site of the LicC⅐CDP-Cho⅐Mg 2ϩ structure suggests the direct involvement of the Mg 2ϩ in the catalytic mechanism of LicC. We propose two direct roles for Mg 2ϩ in NTP transferases. First, the bound Mg 2ϩ accelerates catalysis by positioning the phosphate oxygen of P-Cho adjacent to the ␣-phosphate of CTP to align the incoming second substrate of in-line nucleophilic attack. Second, the interactions of the positively charged Mg 2ϩ works in concert with the side chains of Arg 18 and Lys 28 to stabilize the developing negative charge in the pentacoordinate transition state during CMP transfer.
The role of Arg 18 and Lys 28 in the in-line nucleophilic substitution mechanism proposed for LicC is the same as proposed for the structurally known members of the NTP transferase family, GlmU, RmlA, and E. coli methylerythritol-4-phosphate cytidylyltransferase. When the nucleotide and phosphosugar substrates bind to GlmU to form the ternary complex, the proximity of the substrates and the catalytic Arg 18 (also Arg 18 in LicC) aids the reaction by stabilizing a charged transition state. In RmlA, three residues, Arg 15 , Lys 162 , and Arg 194 , appear important in the catalytic mechanism. Arg 15 (Arg 18 in LicC) also functions by stabilizing the charged transition state, Kinetic data were obtained using the assay described under "Experimental Procedures," and the error bars present the range of data.
FIG. 8. The LicC catalytic cycle. LicC operates by a sequential ordered mechanism that is interpreted based on the enzyme kinetics and the conformational differences between the structure of apo-LicC and the LicC⅐CDP-Cho⅐Mg 2ϩ ternary complex. LicC first binds CTP⅐Mg 2ϩ , which triggers a conformation change that breaks the salt bridge between Glu 216 and Arg 129 . This action reorganizes the ␤-sheet structure of the protein to allow Glu 216 to directly interact with the incoming Mg 2ϩ (see "Discussion"). The subsequent binding of P-Cho is facilitated by the movement of Trp 136 and Tyr 190 to optimize the interaction of the enzyme with the incoming trimethylammonium group of P-Cho. Catalysis occurs in the ternary complex, and PP i is released first, facilitated by the movement of Arg 18 . CDP-Cho⅐Mg 2ϩ is the last product released from the enzyme, which allows the Glu 216 ⅐Arg 129 salt bridge to reform and regenerate the apo-enzyme structure. and Lys 162 (not conserved in LicC) interacts with the phosphate of the phosphosugar substrate to increase its nucleophilicity. The function of Arg 194 (not conserved in LicC) is less clear because the side chain conformation is different in two RmlA structures such that the side chain of Arg 194 in S. enterica RmlA interacts with the phosphate of the phosphosugar substrate, whereas its side chain in P. aeruginosa RmlA does not. In E. coli methylerythritol-4-phosphate cytidylyltransferase, Arg 20 and Lys 27 (Arg 18 and Lys 28 in LicC, respectively) interact with the ␣-phosphate of CTP and are considered to function in transition state stabilization. Arg 157 from the second subunit of the dimer (not conserved in LicC) and Lys 213 (Glu 216 in LicC that coordinates Mg 2ϩ ) interact with the phosphate of the phosphoalcohol substrate, increasing the nucleophilicity of the phosphoalcohol during the catalytic reaction. Taken together, these results suggest that Arg 18 and Lys 28 (LicC numbering) in the signature sequence play a common role in transition state stabilization and substrate activation in catalytic mechanism for the NTP transferase family members.
The active sites of GlmU and RmlA have negative electrostatic potential that is unfavorable for binding of the negatively charged substrates. Mg 2ϩ -coordinating Asp 107 and Glu 216 in LicC correspond to Asp 110 and Asp 225 in RmlA and Asp 105 and Gly 225 in GlmU. In RmlA, these conserved Asp residues contribute a putative, LicC-like metal binding site, but metal binding was not observed in the structures of RmlA (19,21). Mg 2ϩ may bind to Asp 110 and Asp 225 of RmlA to function as catalytic metal. The side chain of Asp 110 in P. aeruginosa RmlA is within hydrogen bond distance to the ␣-phosphate oxygen of the nucleotide substrate, and the charge repulsion between Asp 110 and the ␣-phosphate of the nucleotide substrate must be compensated during the catalytic reaction. In LicC, Mg 2ϩ bound by Asp 107 and Glu 216 neutralizes the negative charges of the substrates and increasing the catalytic activity by stabilizing the transition state. Interestingly, Mg 2ϩ -coordinating Asp 107 and Glu 216 in LicC are replaced by Ala 108 and Lys 213 in E. coli methylerythritol-4-phosphate cytidylyltransferase, respectively. These two replacements not only explain why the active site of E. coli methylerythritol-4-phosphate cytidylyltransferase has positive electrostatic potential but also suggest that this enzyme may not require the direct involvement of Mg 2ϩ for binding substrates and catalysis.