Glycerol-3-phosphate Cytidylyltransferase

The bacterial enzyme, glycerol-3-phosphate cytidylyltransferase (GCT), is a model for mammalian cytidylyltransferases and is a member of a large superfamily of nucleotidyltransferases. Dimeric GCT from Bacillus subtilis displays unusual negative cooperativity in substrate binding and appears to form products only when both active sites are occupied by substrates. Here we describe a complex of GCT with the product, CDP-glycerol, in a crystal structure in which bound sulfate serves as a partial mimic of the second product, pyrophosphate. Binding of sulfate to form a pseudo-ternary complex is observed in three of the four chains constituting the asymmetric unit and is accompanied by a backbone rearrangement at Asp11 and ordering of the C-terminal helix. Comparison with the CTP complex of GCT, determined previously, reveals that in the product complex the active site closes around the glycerol phosphate moiety with a concerted motion of the segment 37-47 that includes helix B. This rearrangement allows lysines 44 and 46 to interact with the glycerol and cytosine phosphates of CDP-glycerol. Binding of CDP-glycerol also induces smaller movements of residues 92-100. Roles of lysines 44 and 46 in catalysis have been confirmed by mutagenesis of these residues to alanine, which decreases Vmax(app) and has profound effects on the Km(app) for glycerol-3-phosphate.

Cytidylyltransferases function predominantly in biosynthetic pathways where they are responsible for activation of metabolic intermediates. They catalyze reversible reactions in which CTP and an alcohol are substrates, and a cytidylyl ester and pyrophosphate are the products, as shown in Equation 1 (below) for glycerol-3-phosphate cytidylyltransferase (GCT), 1 a bacterial enzyme from Bacillus subtilis. GCT, which functions in the biosynthesis of teichoic acid, a component of certain bacterial cell walls, is a member of a family of cytidylyltransferases that we have been investigating. The other two principal members of this family are phosphocholine and phosphoethanolamine cytidylyltransferases, which function in eukaryotic phospholipid biosynthesis. This family ranges in complexity from multidomain eukaryotic enzymes to simple bacterial prototypes in which the entire protein corresponds to a minimal catalytic domain. The complex cytidylyltransferases have attracted considerable interest, because they are frequently regulatory in biosynthetic pathways, and the activities of some of these enzymes are modulated by membrane lipids (1). CTP ϩ glycerol-3-phosphate 7 CDP-glycerol ϩ pyrophosphate (Eq. 1) We have previously determined the structure of GCT, in complex with its substrate CTP, as a model for the catalytic domain of this family of cytidylyltransferases (2). GCT is a homodimer in which each monomer adopts an ␣/␤ fold with a central five-stranded parallel ␤ sheet. In this structure, a large pocket, termed the substrate-binding "bowl," is formed by helices and loops that extend from the central ␣/␤ core (see Fig. 5 in Ref. 2). CTP is bound at one side of the bowl, primarily by interactions with two prominent conserved motifs. The histidine side chains of the 14 HWGH motif (3) interact with the ␣ and ␤ phosphates of the CTP. Both the side chains and backbone atoms of the 113 RTEGISTT motif that characterizes this family of cytidylyltransferases (3) interact with CTP (see below).
The GCT structure is closely related to the structures of phosphopantetheine adenylyltransferase (PPAT) (4,5) and nicotinamide mononucleotide adenylyltransferase (NMNAT) (6 -8), key enzymes in cofactor biosyntheses. The GCT, PPAT, and NMNAT structures are representatives of a superfamily of nucleotidyltransferases, all of which are presumed to share a common catalytic fold (9,10). They typically incorporate Rossmann-like ␣/␤ domains that are topped by a substrate binding region, and they contain the signature HXGH sequence, which forms part of the nucleotide phosphate binding site. The HXGH sequence was first noted in class I aminoacyl-tRNA synthetases (11), which are also members of this nucleotidyltransferase superfamily. The GCT family of cytidylyltransferases is distin-guished from the other nucleotidyltransferases in the superfamily by the presence of the RTEGISTT motif, which is unique to the GCT family. The structure of the CTP complex (2) showed interactions of the Arg 113 side chain of this motif with the substrate but also revealed that specificity for the cytosine base was conferred primarily by hydrogen bonds with backbone atoms of the motif. Despite its broad distribution in nature, the GCT-like fold is not universally found in cytidylyltransferases. For example, a recent x-ray analysis of phosphocholine cytidylyltransferase from Streptococcus pneumoniae (12) demonstrates that the structure of this enzyme, which lacks the HXGH and RTEGISTT signatures, assigns it to a different family of nucleotidyltransferases, many of which catalyze the activation of sugars rather than lipid phosphates.
To define the binding site for glycerol-3-phosphate and to examine the conformational changes that accompany product formation (13,14), we have determined the structure of GCT complexed with its product, CDP-glycerol (see Fig. 1). The structure reported here is a pseudo-ternary complex, with sulfate as a partial mimic for the second product, pyrophosphate, and suggests roles for Lys 44 and Lys 46 in substrate binding and catalysis. These lysines are part of a substructure termed the "40s flap," which closes around the glycerol phosphate end of the active site in the product complex.

EXPERIMENTAL PROCEDURES
Materials-CTP, glycerol-3-phosphate, protease inhibitors, and imidazole were from Sigma. Nickel-nitrilotriacetic acid-agarose was from Qiagen. Vent DNA polymerase and restriction enzymes were from New England Biolabs. [ 14 C]Glycerol-3-phosphate was from Amersham Biosciences. Oligonucleotides were synthesized by the University of Michigan Biomedical Research Core Facility.
Protein Expression and Purification-His-tagged mutant constructs were expressed in BL21(DE3)pLysS Escherichia coli in a 4-h induction with 1 mM isopropylthiogalactoside. Cells were centrifuged at 5,000 ϫ g for 10 min and resuspended in 20 ml of Buffer A (10 mM Tris, pH 8.0, 300 mM NaCl, 10% glycerol, 2.5 g/ml leupeptin, 2 g/ml chymostatin, 2 g/ml pepstatin, 1 g/ml antipain, 10 g/ml p-aminobenzamidine, 10 g/ml benzamidine, and 0.2 mM phenylmethylsulfonylfluoride). The cells were disrupted in a French press. After centrifugation at 40,000 ϫ g for 1 h, the soluble lysate was applied to a 5-ml column of nickelnitrilotriacetic acid resin equilibrated with 10 mM Tris, pH 8.0, 300 mM NaCl, 10% glycerol. The column was washed with 20 mM Tris, pH 8.0, 300 mM NaCl, 10% glycerol, 20 mM galactose, 50 mM imidazole. GCT was eluted with by raising the imidazole concentration to 200 mM. Each protein was judged to be pure by SDS-PAGE.
The wild-type protein used for crystallization was expressed and purified by chromatography on Blue-Sepharose (Amersham Biosciences) essentially as originally described for GCT expressed in E. coli (15) except that elution from Blue-Sepharose was accomplished with 100 mM phosphate instead of CTP. The purified protein was dialyzed (or washed by ultrafiltration) versus 20 mM phosphate at pH 8.0.
Enzyme Assay-GCT was assayed as described previously (15). CTP and glycerol-3-phosphate were varied from 0 to 20 mM, except that glycerol-3-phosphate was varied from 0 to 500 mM for the mutant enzymes. Substrate dependence curves were fit to the Michaelis-Men- using Kaleidograph (Synergy software) to derive apparent V max , K m , and KЈ values. S 0.5 , the substrate concentration at half V max , was calculated as the nth root of KЈ, where n is the Hill coefficient (16).
Crystallization-Protein eluted from Blue-Sepharose columns was concentrated to 10 mg/ml by repeated ultrafiltration in 100 mM Tris, pH 8.5, 1 mM EDTA, and 1 mM dithiothreitol. CDP-glycerol was added to crystallization drops to final concentrations of 5 to 10 mM. Crystallization conditions were the same as for the monoclinic form of the CTP complex (2): reservoir solutions were 30% PEG 3350, 200 mM Li 2 SO 4 , 100 mM Tris, pH 8. 5. Initial characterization of the crystals showed that they were triclinic with a ϭ 37.80 Å, b ϭ 55.93 Å, c ϭ 63.70 Å, ␣ ϭ 88.99°, ␤ ϭ 75.03°, ␥ ϭ 82.54°with a V M of 2.13 for two dimers per asymmetric unit. For data collection crystals were transferred into holding solutions that included 20% (v/v) PEG200 as a cryoprotectant.
Structure Determination-Data were collected with an RAXIS IV imaging plate detector using a Rigaku RU200 rotating anode generator as the x-ray source, with monochromation provided by Yale mirrors. The crystal was flash-cooled to 140 K. Images were processed and intensities scaled using DENZO and SCALEPACK (17). Data collection statistics are reported in Table I. The outermost data between 1.80 and 1.87 Å were 82% complete, with a missing wedge of reflections, but the observed reflections in this annulus were included in the final refinements.
The structure was solved by molecular replacement using the GCT-CTP complex (Protein Data Bank code 1COZ) as a model. Rotation searches with the coordinates for the GCT dimer (2), but omitting CTP, were carried out with X-PLOR (18). The highest peak determined the orientation of the first dimer and fixed the origin. The second dimer, whose orientation was derived from the rotation search, was positioned in a phased translation function computation (18).
Refinements were conducted in CNS (19) with cross-validation and maximum likelihood targets (20). Initial rigid body refinement with data from 15.0 to 2.0 Å used first the dimer model and then single chains as rigid groups and was followed by minimization with noncrystallographic symmetry (NCS) restraints applied to all atoms. At this stage R work was 0.404. Rounds of minimization and simulated annealing from 2000 K in torsional space were conducted using NCS restraints. After adding bound product and sulfate to the model, R work  was 0.316. Examination of electron density and difference density maps, including maps based on coefficients from simulated annealing omit refinement (21), indicated regions where the chains differed from one another; chain C showed the largest variations in conformation. In subsequent refinements, selective NCS restraints were applied to a core of the structure (ϳ50% of the atoms) chosen on the basis of deviations among the chains. This strategy aimed to maintain reasonable data-toparameter ratios while allowing for true differences among the four chains. Subsequent protocols included scaling and overall B adjustment, minimization, and torsional simulated annealing. After R work had decreased to 0.294, waters were gradually added and were retained if they had peak heights greater than 3 and made acceptable hydrogen bonding interactions. NCS restraint weights and the number of restrained atoms were also decreased during these stages, using R free to monitor the effects of parameter changes. In the last three rounds the data beyond 2.0 Å were included to improve the data-to-parameter ratio (Table II). In these refinements the decrements in R work and R free were 0.006 and 0.0093, respectively. The final R work for a model with 518 waters was 0.2180, and R free was 0.2541. Statistics from refinement with data from 15.0 to 1.8 Å are reported in Table II. Root mean square differences among the core regions that were subjected to NCS restraints until the last stages of refinement were about 0.18 Å, whereas the overall root mean square differences from chain A were 0.301, 0.545, and 0.328 Å for main chain atoms of residues 1-115 of chains B, C, and D, respectively. The C chain, which does not bind sulfate, displays the largest deviations.
Evaluation of the model in PROCHECK (22) indicated that all residues were in allowed regions of the Ramachandran maps. The thermal factors for the C chain of the product complex are somewhat larger than for the other chains (Table II), and the C terminus of this chain, from Thr 116 onward, could not be modeled. Coordinates have been deposited in the Protein Data Bank with the accession number 1N1D. The chains of dimer I include residues A1-A129 and B1-B129; the chains of dimer II include residues C1-C115 and D1-D123.

RESULTS
Overview of the Product Complex-The structure of the triclinic crystal form of GCT with bound CDP-glycerol was solved by molecular replacement using the structure of the CTPbound complex as a model. Because the asymmetric unit of the triclinic crystal includes two dimers, designated I and II, there are four independent views of the fold and of the enzymeproduct interactions. Each of the monomers binds CDP-glycerol with thermal factors similar to the B-factors for the surrounding protein, consistent with high occupancy of the binding sites. Both chains of dimer I (Fig. 1A) contain strong, isolated electron density with protruding lobes characteristic of phosphate or sulfate, located close to the position corresponding to the ␤ phosphate of CTP in the substrate complex (Fig.  1B). This bound ligand was assigned as sulfate, because the crystals were grown in lithium sulfate with no added phosphate. The location of the sulfate suggests that it is a mimic of the other product, pyrophosphate. In dimer II this sulfate is slightly displaced in one monomer (chain D) and absent in the other (chain C).
Measurements of the Binding of CDP-glycerol to GCT-The affinities of the enzyme for CDP-glycerol were measured to compare the enzyme⅐CTP complex with the enzyme⅐CDP-glycerol complex. Binding to GCT was monitored by the quenching of the intrinsic fluorescence of tryptophan as described for the substrates of the forward reaction (13). Changes in fluorescence upon binding of CDP-glycerol (data not shown) were biphasic as has been observed for CTP and glycerol-3-phosphate. Dissociation constants of 0.4 Ϯ 0.1 and 16 Ϯ 9 M were obtained for the first and second steps of product binding to the GCT dimer. These affinities are similar to those determined for CTP or glycerol-3-phosphate (13,23).
Interactions of GCT with CDP-glycerol-The bound product adopts a zig-zag shape, with the ␣-(cytidine) phosphate positioned above the ribose ring and the glycerol moiety extended across the substrate binding bowl toward helix D (Fig. 1B). Interactions of the CMP moiety of the product with the fingerprint sequences 14 HWGH and 113 RTEGISTT (see Table III and Fig. 2A) are similar to those found with CTP (2). The side chains of His 14 and His 17 interact with the sulfate in the monomers that contain sulfate. As shown in the CTP complex (2), hydrogen bonding of the cytosine moiety to backbone atoms from Thr 114 and Ile 117 establishes the selectivity for the cytosine base ( Fig. 2A). However, in the CDP-glycerol complex the ribose 2Ј and 3Ј hydroxyls interact with the Asp 94 carboxylate and with the amide of Gly 92 , respectively, whereas these ribose hydroxyls have no close protein partners in the CTP complex. These differences in product and substrate interactions result from a small movement of residues 92-99 toward the product.
The structure of the complex with CDP-glycerol reveals the binding determinants for glycerol-3-phosphate (see Table III and Fig. 2B). The oxygen that bridges the ␣Ϫ and ␤-phosphates of the product hydrogen bonds with the amide of Thr 9 (not shown), and an exo-oxygen of the ␣-phosphate interacts with Lys 46 (Fig. 2B). The indole nitrogen of Trp 95 and the ⑀-amino groups of Lys 44 and Lys 46 form hydrogen bonds to the ␤-phosphate, Lys 77 binds the terminal 1Ј hydroxyl of glycerol, and Glu 71 binds both of the hydroxyl groups of the glycerol moiety (see Table III and Fig. 2B). Closure of the rims of the substrate bowl (see below) brings Lys 44 , Lys 46 , and Trp 95 within hydrogen bonding distance of the product.
The orientations of the substrate and product molecules in the respective complexes of GCT are compared in Fig. 3. The ␤and ␥phosphates from CTP and the glycerol phosphate from CDP-glycerol extend on opposite sides of the CMP phosphate. The ␣-phosphate configuration in CDP-glycerol is inverted from that in CTP. These two states represent the configurations of this phosphate prior to and after formation of the transition state in which the ␣-phosphate assumes a trigonal bipyramidal configuration.
Flap Closure and Other Structural Changes-The overall folds of the product and CTP complexes are very similar (Fig.  1B). However, the two structures differ significantly at residues 37-47 in the vicinity of the glycerol phosphate moiety and more subtly at residues 92-99. The difference between the CTP and the product complexes at residues 37-47 is striking (Fig.  3). This piece of the structure, which includes a short helix at residues 37-43, acts as a flap or lid that closes over the glycerol phosphate. The concerted rearrangement can be approximated as a rigid body motion and has been analyzed using the program DynDom (24,25). The motion of the flap is best described as a rotation of ϳ14°about an axis passing through the hinge residues 36 and 48. The backbone of residue 45 moves about 3.0 Å in this rearrangement. The smaller displacements at residues 92-99 and the C-terminal helix (residues 119 -126) are not concerted and cannot be treated as rigid body motions.
In the triclinic crystal form, the two dimers of the asymmetric unit contact one another at an interface that includes residues 44 -48 from each chain. This raises the possibility that crystal packing might influence the position of the flap in the product complex. The chains at this interface are related by an improper rotation of about 160°and a translation of 4.2 Å, and the interchain interactions are asymmetric. For example, Gln 45 from the C chain intrudes into the substrate bowl of its neighboring A chain, approaching the sulfate ion, but the opposing Gln 45 wraps back against its own chain. Despite this asymmetry, the motion of the flap is essentially the same in all four chains. Similarity of the flap motions and the studies of mutants that establish the functional importance of lysines 44 and 46 (see below) argue that flap closure is functional and not simply an artifact of crystal packing.
Role of Lysines 44 and 46 in GCT: Analysis of Mutations-The movement of the 40s flap toward the glycerol-3-phosphate portion of the product suggested that residues in this region were important for binding the substrate, glycerol-3-phosphate, and possibly important for catalysis. The two lysines in this segment, Lys 44 and Lys 46 , are conserved in other putative bacterial glycerol-3-phosphate cytidylyltransferases (Fig. 4). Moreover, a conserved lysine is found in a similar segment of eukaryotic phosphocholine (CCT) and phosphoethanolamine (ECT) cytidylyltransferases (Fig. 4).
To assess the functional contributions of Lys 44 and Lys 46 , each residue was mutated to alanine in the context of Histagged GCT (3). The mutant proteins were purified and assayed as a function of substrate concentration to determine whether K m and V max were altered toward either substrate (Table IV). Both the K44A and K46A mutants were considerably less active than the wild-type enzyme, with the apparent V max for each mutant enzyme reduced by a factor of about 10. Each mutant displayed a small degree of positive cooperativity toward CTP; K46A was positively cooperative toward glycerol-3-phosphate, but K44A was negatively cooperative toward glycerol-3-phosphate. Because of this cooperativity, the kinetic data were fit to a Hill equation. This allowed a calculation of S 0.5 , the substrate concentration at half V max . The S 0.5 for glycerol-3-phosphate was altered dramatically, increasing about 125-fold for each of the mutants (Table IV). This change is consistent with the conclusion that each lysine plays a role in the reaction. In addition, the S 0.5 for CTP for the mutants was about 5-fold higher than for the wild-type, a result that is not surprising, because all previous mutations that increase K m for one substrate also increase K m for the other substrate (3). The Ternary Complex and Comparisons with the Sulfate-free Chain--As already noted, sulfate from the crystallization medium appears in three of the four chains. In both chains of dimer I it occupies a site very close to the ␤-phosphate of CTP, a likely position for one of the phosphates of pyrophosphate (see Table III and Fig. 1B and Fig. 3); in the D chain of dimer II the sulfate is bound at a site between the positions of the ␤and ␥-phosphates of CTP.
Chain C of dimer II is a binary complex of enzyme and product without bound sulfate (Fig. 5). Comparisons of this chain with the chains of dimer I indicate that binding of sulfate to form a pseudo-ternary complex is accompanied by significant changes in the nucleotide end of the substrate-binding bowl (Fig. 5B). One of the structural differences between chains with and without the sulfate ion is the orientation of the peptide connecting Asp 11 and Leu 12 , with the carbonyl oxygen of Asp 11 pointing toward the substrate bowl in the ternary complex with sulfate bound and away from the bowl in the binary complex. This oxygen also points away from the bowl in the CTP complex (Fig. 5B). The orientation of this peptide seems to depend on the presence and nature of the ligands that interact with His 14 , His 17 , and Thr 119 . In addition, the C-terminal region of the unique sulfate-free chain is much less ordered than the corresponding regions of the remaining chains in the structure and cannot be modeled beyond residue 115. We believe this apparent mobility of the sulfate-free chain has functional significance (see "Discussion").
Structure of the Dimer Interfaces: Comparisons with the CTP Complex-Negative cooperativity (13) leads one to expect that substrate binding will induce structural changes that propagate to the dimer interface. The interfaces were examined to look for differential effects of CTP and CDP-glycerol. Comparisons of dimer I from the product complex with the CTP complex reveal small differences in the overall orientations of the monomer chains relative to one another; one chain rotates 4.2°a nd translates by 0.6 Å. Many of the direct interactions between chains, such as the stacking of Trp 15 of chain A with  Table III. B, the interactions between GCT and the glycerol phosphate moiety of CDP-glycerol. Residues that contact or interact with the glycerol phosphate are drawn in full-atom representation and coloring. Selected hydrogen bonds between the glycerol phosphate and protein residues are shown as dashed lines. See the text for descriptions of the tightly bound solvents and Table III  His 54 of chain B (Fig. 6), are retained in both structures, but several local differences between CTP and CDP-glycerol complexes are observed. Hydrogen bonding of the interface residue, His 50 , to its own chain is disrupted by the peptide flip at Asp 11 that occurs in chains binding product and sulfate (Fig. 6). Small differences at the interface also occur in the vicinity of residues 62-66, part of a sequence that is conserved among members of the GCT family of cytidylyltransferases (2). In the CDP-glycerol complex the side chains of Arg 63 , which approach one another at the local dimer axis, adopt several different conformations. NMR studies have demonstrated that 15 N chemical shifts of this side chain are altered by CTP binding (14). In dimer II additional interchain contacts are perturbed because of disorder beyond residue 114 in the sulfate-free C chain, and the hydrogen bond between Tyr 49 of the D chain and Lys 125 of the C chain, found in the interfaces of dimer I and the CTP complex, is presumably broken.
Solvents in the Active Site-Well ordered solvent molecules, found in both the CTP-and product complexes, appear to be important in stabilizing the interactions of the ligands (see Figs. 2 and 3 and Table III). Three such waters, neighboring the cytosine, the ribose 3Ј oxygen, and the phosphate or sulfate, appear in the same positions in both structures. Three other waters, one interacting with the loop carrying Lys 44 and Lys 46 , and two adjoining Glu 71 and Lys 77 , shift positions along with changes in the protein. A solvent that interacts with Asp 38 in the CTP complex moves about 3 Å in the product complex and forms a hydrogen bond to the backbone of the displaced Lys 46 .
Another well ordered solvent occupies a site that may also be able to bind Mg 2ϩ . Mg 2ϩ is known to be required in reactions of GCT (15), in particular for binding of glycerol-3-phosphate (3, 13) 2 and is presumed to bridge the phosphate groups of sub-strates and products (6,26). The water that is a likely metal surrogate (W1 in Fig. 2A) is found in the product complex at a position close to the location of the Mg 2ϩ /Mn 2ϩ site observed in the ATP complexes of PPAT (5) and in NMNAT from Methanococcus jannaschii (6). In these latter structures Mn 2ϩ is bound to oxygens from each of the three phosphates of the ATP, which curl around the cation. Experiments with Mg 2ϩ or Mn 2ϩ will be required to confirm that this is a binding site for a divalent ion in the complex of CDP-glycerol with GCT.

DISCUSSION
The Role of the 40s Flap in Binding and Reactivity: Comparisons with Other Nucleotidyltransferases-For the large family of transferases bearing the HXGH signature, including the aminoacyl-tRNA synthetases, studies of mechanism converge on a scheme in which catalysis results from stabilization of the highly charged pentavalent transition state (27) by countercharges provided by basic residues and/or metal ions. 3 The role of basic residues has been thoroughly documented for the aminoacyl-tRNA synthetases by single and multiple mutations (28). Substitutions of alanine for the lysine residues in the KMSKS signature sequences of aminoacyl-tRNA synthetases reduce k cat /K m by factors ranging from 100 to 5000 (29 -33). The KMSKS signature of aminoacyl-tRNA synthetases corresponds approximately to 116 GISTT, located near the start of the E helix in the structure of GCT (2). The Ile and Thr residues of this sequence interact with CTP and CDP-glycerol ( Fig. 2A), but the lysines that are important for stabilization of the transition state in the synthetases are missing from the GCT signature sequence (Fig. 1C). Two downstream lysines, 121 and 123, are found in GCT but are not conserved and do not interact   (14). b Hill coefficient. When the data for the wild-type enzyme were fit to a Hill equation, the Hill coefficient was close to 1.0, so that K m essentially was the same as S 0.5 .
with phosphates of the substrate or product. The functional roles of the lysines from the KMSKS motif in aminoacyl-tRNA synthetases seem to be assumed by lysines 44 and 46 from the flap in GCT. Mutation of GCT residues Lys 44 or Lys 46 to alanine reduces V max /K m for glycerol-3-phosphate by factors of ϳ1000, an effect of the same order of magnitude as observed for the critical lysines in aminoacyl-tRNA synthetases.
The present study thus reveals surprising variability in the positions of the basic residues that are presumed to stabilize the transition state. Another variation is found in some of the NMN adenylyltransferase enzymes (see below), which appear to utilize arginine residues from yet a third sequence for transition state stabilization. As a result of these variations in the positions of critical basic residues, a signature for essential basic groups is difficult to trace through the entire nucleotidyltransferase superfamily. Nevertheless the conservation of basic residues in the 40s can be detected within the GCT subfamily of cytidylyltransferase sequences as shown in Fig. 4. Indeed, substitution of alanine for the single lysine in this segment of rat choline phosphate cytidylyltransferase reduces V max /K m by a factor of 240,000 (34), indicating an important functional role for the conserved lysine from the 40s region in this member of the GCT family. In chain C there is no density at the sulfate position. B, a view in the same orientation as Fig. 2A, showing the differences between the chain that binds only CDP-glycerol (silver) and the chains that also bind sulfate. In the sulfate-free chain there is no significant density corresponding to W1, and the structure cannot be modeled beyond residue 114.
FIG. 6. Dimer interfaces in the CTP complex and in the product (ternary) complex. A view of the dimer interface in the vicinity of His 50 showing the backbone rearrangement at Asp 11 . The structure of the CTP complex is in silver; the blue and gold chains are from dimer I of the product complex, in which both chains bind sulfate. Reorientation of Asp 11 is the major difference that is evident in this superposition of the two structures. Side chains of Leu 12 , Leu 13 , Ser 51 , and Glu 53 have been omitted for clarity.
Structures have been determined for four nucleotidyltransferases with folds similar to GCT, PPAT from E. coli (4,5) and NMNATs from E. coli and two Archaea (6 -8). As in GCT, the 40s regions of NMNAT and PPAT from E. coli incorporate basic residues that bind product. The effects of mutations in GCT suggest that these residues in NMNAT and PPAT are likely to play a catalytic role. As shown in Fig. 7, the 40s region of the E. coli NMNAT enzyme moves significantly relative to apoenzyme when NAD ϩ binds, bringing His 45 and Arg 46 close to the NAD phosphates (8). A smaller displacement of the loop carrying Lys 42 is observed in comparisons of free and product-bound PPAT (4). In contrast, in the archaeal NMNAT enzymes the segment equivalent to the flap of GCT is shortened, and primary interactions with the product phosphates are provided by arginine side chains at positions corresponding to Thr 9 and near Lys 123 of GCT, rather than by residues in the 40s region.
Cooperativity-Dimeric GCT binds its substrates CTP and glycerol-3-phosphate in random fashion (15), but it displays an intriguing form of negative cooperativity in substrate binding (13). At the first step in binding of either substrate, the apparent K d is about 0.2 M; at the second step K d increases to ϳ30 M. As reported here, the substrate for the reverse reaction, CDP-glycerol, similarly displays negative cooperativity in binding. Kinetic analyses and simulations have suggested that the enzyme turns over only when both active sites are occupied by substrates (13). Cooperativity has also been inferred for PPAT, where structures provide evidence for half-of-the-sites reactivity (4,5). However, cooperativity may not be a universal feature of the GCT-like enzymes. A very recent study of the properties of a tetrameric GCT from Staphylococcus aureus has failed to find any evidence for cooperative behavior in that species of GCT (35).
The structural bases for negative cooperativity in B. subtilis GCT and for the requirement that the enzyme be fully loaded to turn over are incompletely understood. NMR studies of CTP binding to GCT support a sequential model for negative cooperativity (14). Measurements of Arg 15 N and backbone 15 NH resonances as a function of increasing CTP concentration revealed changes in chemical shifts that were linked to binding of the second CTP and consistent with conformational changes near the dimer interface. These studies identified residues that might be important for cooperativity. Residues displaying chemical shift changes during the titration included Arg 55 and Arg 63 ; there were also changes in the C-terminal helix and the conserved region following Arg 113 . The NMR experiments compared free enzyme with CTP-bound species and were therefore expected to detect conformational changes associated with cooperativity in binding of CTP, whereas the available x-ray structures only permit comparison of the saturated CTP-and CDP-glycerol complexes. Nevertheless the x-ray structures of these latter two complexes display some structural differences in the same regions that are associated with the chemical shift changes observed in the NMR experiments. Arg 63 appears flexible and adopts multiple conformations. Rearrangements are seen near His 50 , a neighbor of the Arg 55 that was tracked in the NMR experiments. Evidence from the x-ray structures suggests that Asp 11 and His 50 sense the presence of ligands in the pyrophosphate/sulfate site. However, comparisons of the x-ray structures of substrate and product complexes do not reveal how the changes at one active site are communicated to the second chain of the dimer.
Both the x-ray experiments and NMR data indicate disorder and flexibility in GCT. In particular, the apparent disorder of the C terminus of the sulfate-free chain, where there is no ligand in the pyrophosphate subsite, strongly correlates with the flexibility of the C terminus of substrate-free GCT dis-cerned by NMR (14). Measurements of relaxation times by NMR imply that substrate binding in the CTP subsite is associated with an increase in the ordering of the chains and the subunit interface and suggest a possible role for mobility in the cooperative behavior of this enzyme. The loss of entropy resulting from the reduced flexibility associated with substrate binding may contribute to the free energy of negative cooperativity in GCT (14).
Changes Accompanying Formation of a Ternary Complex-An intriguing property of GCT is that rapid turnover occurs only when both substrates are bound to both chains (13). Comparisons of the sulfate-free chain with the chains of dimer I, described under "Results," reveal some structural changes that are associated with formation of the pseudo-ternary complex, GCT⅐CDP-glycerol⅐sulfate, from the binary complex with CDP-glycerol (Fig. 5B).
However, the structural changes that elicit full activity in true ternary complexes of GCT (13) are probably incompletely expressed in the complexes with CDP-glycerol and sulfate. Attempts to obtain fully occupied ternary complexes by cocrystallizing GCT with glycerol-3-phosphate and the nucleotide analog CPNPP, 4 in which the ␣-␤ bridge oxygen is replaced by an amide (36), have met with limited success. Occupancies of the glycerol-3-phosphate site are low, and crystals are disordered by further soaking with very high concentrations of the ligands. 5 Similarly, despite considerable effort, it has been not been possible to obtain crystals of ternary complexes of PPAT (5). These observations suggest that binding of Mg 2ϩ and the highly charged pyrophosphate to form the ternary complex for the reverse reaction may be associated with much larger structural rearrangements than we see in comparisons of the sulfate-bound and sulfate-free chains. From the many studies of reactivity in the enzymes related to GCT it is evident that electrostatic interactions play a dominant role in both substrate binding and activity of nucleotidylyltransferases (27,28).