Mechanistic Insights into the Retaining Glucosyl-3-phosphoglycerate Synthase from Mycobacteria*

Background: Knowledge of conformational changes occurring in glycosyltransferases is limited. Results: The active site of GpgS is essentially preformed as the protein proceeds along the catalytic cycle with the nucleotide sugar β-phosphate playing a central role in substrate binding. Conclusion: Conformational dynamics is a major determinant of GpgS activity. Significance: This model of action might be operational in other GT-A glycosyltransferases. Considerable progress has been made in recent years in our understanding of the structural basis of glycosyl transfer. Yet the nature and relevance of the conformational changes associated with substrate recognition and catalysis remain poorly understood. We have focused on the glucosyl-3-phosphoglycerate synthase (GpgS), a “retaining” enzyme, that initiates the biosynthetic pathway of methylglucose lipopolysaccharides in mycobacteria. Evidence is provided that GpgS displays an unusually broad metal ion specificity for a GT-A enzyme, with Mg2+, Mn2+, Ca2+, Co2+, and Fe2+ assisting catalysis. In the crystal structure of the apo-form of GpgS, we have observed that a flexible loop adopts a double conformation LA and LI in the active site of both monomers of the protein dimer. Notably, the LA loop geometry corresponds to an active conformation and is conserved in two other relevant states of the enzyme, namely the GpgS·metal·nucleotide sugar donor and the GpgS·metal·nucleotide·acceptor-bound complexes, indicating that GpgS is intrinsically in a catalytically active conformation. The crystal structure of GpgS in the presence of Mn2+·UDP·phosphoglyceric acid revealed an alternate conformation for the nucleotide sugar β-phosphate, which likely occurs upon sugar transfer. Structural, biochemical, and biophysical data point to a crucial role of the β-phosphate in donor and acceptor substrate binding and catalysis. Altogether, our experimental data suggest a model wherein the catalytic site is essentially preformed, with a few conformational changes of lateral chain residues as the protein proceeds along the catalytic cycle. This model of action may be applicable to a broad range of GT-A glycosyltransferases.

tion for the nucleotide sugar ␤-phosphate, which likely occurs upon sugar transfer. Structural, biochemical, and biophysical data point to a crucial role of the ␤-phosphate in donor and acceptor substrate binding and catalysis. Altogether, our experimental data suggest a model wherein the catalytic site is essentially preformed, with a few conformational changes of lateral chain residues as the protein proceeds along the catalytic cycle. This model of action may be applicable to a broad range of GT-A glycosyltransferases.
Glycosyltransferases (GTs) 4 play a central role in nature due to their exceptional capacity to synthesize a broad range of glycans. They transfer a sugar moiety from nucleotide sugar and lipid-phosphosugar donors to acceptor substrates, including mono-, oligo-, and polysaccharides, proteins, lipids, small organic molecules, and deoxyribonucleic acids (1). GTs can be classified as either "retaining" or "inverting" enzymes according to the anomeric configuration of substrates and products (2). Inverting GTs follow a direct displacement S N 2-like mechanism via a single oxocarbenium ion-like state. In contrast, the catalytic mechanism for retaining GTs remains less clear. By analogy with glycosylhydrolases, a double displacement mechanism involving a covalently bound glycosyl-enzyme intermediate was first suggested. However, in the absence of both, a clear catalytic nucleophile and structural/kinetic evidence of a viable covalent intermediate, an alternative mechanism has been proposed. In this mechanism, known as "internal return," leaving group departure and nucleophilic attack occur on the same face of the sugar (3-7) involving either a short lived oxocarbenium ion intermediate (SNi-like) (8) or an oxocarbenium ion transition state (SNi) (9). Two major structural folds have been described for the nucleotide sugar-dependent enzymes among the first 35 GT sequence-based families (CAZy, carbohydrate-active enzymes database (10)) for which three-dimensional structures have been reported. These topologies are variations of "Rossmann-like" domains and have been identified as GT-A and GT-B (11,12). Moreover, bioinformatics analysis revealed that many of the structurally uncharacterized nucleotide sugar-dependent GT families are also predicted to adopt one of these two folds. Interestingly, both inverting and retaining enzymes were found in GT-A and GT-B folds indicating that there is no correlation between the overall fold of GTs and their catalytic mechanism. The GT-A fold was first described for the 256-amino acid protein SpsA from family GT2, a putative inverting GT from Bacillus subtilis (13). It consists of two tightly associated ␤/␣/␤ Rossmann-like domains, where the N-terminal domain recognizes the nucleotide sugar donor and the C-terminal domain of the protein contains the acceptorbinding site. Most GT-A enzymes exhibit an Asp-Xaa-Asp (also known as DXD) signature in which the carboxylate groups coordinate a divalent cation and/or a ribose ring (2). Kinetic and structural studies have revealed that most GT-A enzymes follow an ordered mechanism in which the divalent cation and nucleotide sugar donor bind first, prior to binding of the acceptor (14 -16). The glycosylated acceptor is then released, followed by the nucleotide group. The divalent cation may react with the free enzyme and does not dissociate after each catalytic cycle (17)(18)(19)(20)(21)(22)(23). Often the interaction of GT enzymes with their natural substrates leads to substantial changes in the structural conformation of the proteins, compared with their free forms, with direct implications for their function (12,24,25). Specific loops, adjacent to the active site, for instance, often adopt different conformations in the presence or absence of substrates. These loops have been suggested to restrict water access to the active site and appear to play a crucial role during substrate binding and catalysis in GT-A enzymes, including inverting GnT-I (26), ␤4Gal-T1 (21), GlcAT-I (27), CstII (28), and MgS (29) and retaining ␣-(1,3)GalT (20,30) and GTA/GTB (31).
The glucosyl-3-phosphoglycerate synthase (GpgS) is a retaining ␣-glucosyltransferase that initiates the biosynthetic pathway of the 6-O-methylglucose lipopolysaccharides (MGLPs) in mycobacteria. The enzyme transfers a Glcp moiety from UDP-Glc to the 2-position of the 3-phosphoglycerate (PGA) to form glucosyl 3-phosphoglycerate ( Fig. 1A) (31,33). MGLPs are cytoplasmic lipopolysaccharides of intermediate size containing up to 20 Glcp units, many of which are 6-O-methylated. Moreover, MGLPs can be acylated with additional acetyl, propionyl, isobutyryl, succinyl, and octanoyl groups (Fig. 1B) (34). A remarkable property associated with MGLPs is their ability to form stable 1:1 complexes with long-chain fatty acids and acylcoenzyme A derivatives in vitro (35,36). The fact that MGLPs are composed of Glcp units predominantly in ␣-(134)-linkage confers on these molecules a proclivity to assume an helical conformation. Within the complexes, the fatty acyl chain is included in the nonpolar cavity of the coiled polysaccharide chain (37). Interestingly, with an intracellular concentration of long-chain acyl-CoAs in Mycobacterium smegmatis of ϳ0.3 mM, the concentration of polymethylated polysaccharides approaching 1 mM, and the dissociation constant of the polysaccharide⅐lipid complex estimated to 0.1 M, all of the long-chain fatty acids of the cytosol may form complexes with polymethylated polysaccharides leading to the suggestion that the physiological function of these polymethylated polysaccharides may serve to as general carriers for long-chain fatty acids synthesized in the cytosol (38).
This study describes a detailed investigation of the conformational properties of MtGpgS in solution. Using a combination of x-ray crystallography, limited proteolysis, isothermal titration calorimetry (ITC), and analytical ultracentrifugation (AUC), we propose a plausible model for donor and acceptor substrate recognition and binding. The implications of this model for the understanding of the early steps of MGLPs biosynthesis and the catalytic mechanism of other members of the GT-A family are discussed.
GpgS Crystallization and Data Collection-The apo-forms of MtGpgS and MtGpgS in complex with Mg 2ϩ and uridine 5Ј-diphosphate (UDP, Fluka; MtGpgS⅐Mg 2ϩ ⅐UDP) were crystallized as described previously (39). Crystals of MtGpgS in complex with Mn 2ϩ , UDP, and PGA (MtGpgS⅐Mn 2ϩ ⅐UDP⅐PGA) were obtained by mixing 0.5 l of the protein (10 mg ml Ϫ1 ) in the presence of 5 mM MnCl 2 , 5 mM UDP, and 5 mM PGA with 0.5 l of a mother liquor of 0.4 M NH 4 dihydrogen phosphate using the sitting drop vapor diffusion method. Crystals appeared after 1-2 days and grew as rhombuses reaching 0.37 ϫ 0.30 ϫ 0.15 mm. Prior to data collection, the crystals were cryo-cooled in liquid nitrogen by using 0.5 M NH 4 dihydrogen phosphate and 30% (v/v) glycerol as cryo-protectant solution. X-ray diffraction data from single crystals of MtGpgS⅐Mn 2ϩ ⅐UDP⅐PGA were collected using synchrotron radiation in the ID-23-2 microfocus beamline ( ϭ 0.873 Å) at the European Synchrotron Radiation Facility (ESRF, Grenoble, France), and processed with the XDS program. MtGpgS⅐Mn 2ϩ ⅐UDP⅐PGA crystals belong to the I4 1 space group and diffracted to 1.98 Å and have two monomers per asymmetric unit, corresponding to a Matthews coefficient of 2.32 Å 3 and a solvent content of 47.04%. The complete data collection statistics are shown in Table 1.
Structure Determination and Refinement-The structures of the apo-form of MtGpgS and that of the MtGpgS⅐Mg 2ϩ ⅐UDP and MtGpgS⅐Mn 2ϩ ⅐UDP⅐PGA complexes were solved by molecular replacement with the program Phaser Version 2.1.2 (40), using the atomic coordinates of MAP2569c from Mycobacterium avium subsp. paratuberculosis as the search model (Protein Data Bank code 1CKJ, see Ref. 41). The final MtGpgS apo-form, MtGpgS⅐Mg 2ϩ ⅐UDP, and MtGpgS⅐Mn 2ϩ ⅐UDP⅐PGA models were obtained after alternate cycles of model building using the program COOT (42) and restrained refinement using the program phenix.refine (43).
N-terminal Sequence Analyses-Samples were run in a NuPAGE 4 -12% gel. The gel was then washed with NuPAGE transfer buffer (Invitrogen) during 15 min at room temperature. Proteolytic fragments were electrotransferred to a PVDF membrane using the iBlot dry blotting system (Invitrogen) during 7 min. The PVDF membrane was then washed for 10 min with Milli-Q purified water. Bands were stained with a  (66). Position 3 of the second and that of fourth ␣-D-Glcp residues (closest to the reducing end) are substituted by single ␣-D-Glcp residues. R1, R2, and R3 are acyl groups: R1, acetate, propionate, or isobutyrate; R2, octanoate; and R3, succinate. MGLPs occur as a mixture of four main components that differ in their content of esterified succinate. The names of the genes thought to be involved in the different steps of their elongation and modifications are shown (32,67). Acylation and methylation are thought to occur concurrently; the precise stage at which the two ␤-(133)-linked Glc residues are attached is not known, but the definition of early MGLP precursors suggests that they are added early during the elongation process. solution containing 0.1% Coomassie Brilliant Blue R-250, 40% methanol, and 1% acetic acid and subjected to N-terminal sequence analysis using Applied Biosystems 494 procise high throughput protein sequencer at the Biomolecular Resource Facility of the University of Texas Medical Branch.
Isothermal Titration Calorimetry-Ligand binding to MtGpgS was assayed using the VP-ITC system (MicroCal Inc.) as described previously (25,44), with the following modifications. The ITC cell (1.4 ml) contained 40 M MtGpgS in 50 mM HEPES, pH 7.5, 2 mM MnCl 2 , 150 mM NaCl, and the syringe (300 l) contained 500 M of URI, UMP, UDP, UDP-Glc, or PGA in the same buffer. Binding of PGA to MtGpgS-UDP and uridine complexes was assayed as follows. MtGpgS was first titrated with the nucleotide analogs, and the resulting solutions of the protein nucleotide complexes were then titrated with a 500 M PGA solution. Sample solutions were thoroughly degassed under vacuum, and each titration was performed at the indicated temperature by one injection of 2 l followed by 37 injections of 8 l, with 210 s between injections using a 416 rpm rotating syringe. Raw heat signal collected with a 16-s filter was corrected for the dilution heat of the ligand in the MtGpgS buffer and normalized to the concentration of ligand injected. UMP, uridine, and PGA binding isotherms were fitted to a single site bi-molecular model (45) using the Origin TM software provided by the manufacturer. Fitting the UDP-Glc and UDP binding isotherms required to develop a specific binding algorithm as described below.
Analysis of UDP and UDP-Glc Binding Isotherms-ITC binding isotherms observed upon UDP and UDP-Glc binding to GpgS were fitted to a model considering two equilibrium reactions, the equilibrium between two protein conformations, P 0 and a P 1 (Reaction 1), with equilibrium constant K e and thermodynamic parameters ⌬G e , ⌬H e , ⌬S e , and ⌬C p,e , and the binding equilibrium with 1:1 stoichiometry between one free protein conformation only, P 1 , and its bound complex (Reaction 2), The GpgS protein population shifts upon ligand binding from free P 0 to bound P 1 conformations with the overall equilibrium constant K ϭ K e ⅐K b . The heat absorbed or evolved upon ligand addition to the GpgS solution is the sum of the heats absorbed or evolved during equilibration of the two protein equilibriums, dQ ϭ dQ e ϩ dQ b . The experimental parameter determined in the titration calorimeter is the differential heat dQ/dL tot (actually ⌬Q/⌬L tot ), where L tot is the total ligand concentration, free plus bound (see Equations 1-3 and supplemental material) L r and r are two unitless parameters that depend on the total ligand and protein concentrations, r ϭ 1/P tot K b , L r ϭ L tot /P tot (45), and ␣ is a unitless parameter that depends on the equilibrium constant between the two protein conformations, ␣ ϭ 1 ϩ 1/K e . V 0 is the reaction cell volume. Nonlinear regression analysis of dQ/dL tot (Equation 1) allows estimation of the thermodynamic parameters of the two equilibrium reactions. Analytical Ultracentrifugation-AUC experiments were performed with a Beckman XL-1 analytical ultracentrifuge using absorbance optics. Velocity measurements utilized twosector charcoal-filled Epon centerpieces, quartz windows, 400-l sample and 420-l reference volumes in 50 mM Tris-HCl, pH 8.0, 150 mM NaCl. All samples were centrifuged in a Beckman 8-hole An50Ti rotor at 22°C at 40,000 rpm, and the data were collected at 280 nm, with a radial increment of 0.003 cm for ϳ7 h. Velocity data were edited and analyzed using the boundary analysis method of Demeler and van Holde as implemented in Ultrascan version 7.3 for Windows (46). Sedimentation coefficients (s) are reported in Svedberg units (S), where 1 S ϭ 1 ϫ 10 Ϫ13 s and were corrected to that of water at 20°C (s 20,w ). The partial specific volume of full-length MtGpgS was calculated from the amino acid sequence within Ultrascan. Modeling of hydrodynamic parameters was performed using Ultrascan. The frictional ratio (f/f 0 ) was calculated from the known molecular mass and measured sedimentation coefficient using Ultrascan.
Structural Alignment-Structural alignment of MtGpgS (Protein Data Bank codes 4DDZ, 4DE7, and 4DEC are from this study; 3E25 and 3E26 are from Ref. 47), MaGpgS (3CKJ, 3CKN, 3CKO, 3CKQ, and 3CKV are from Ref. 41), and other GT-A glycosyltransferases was performed by the distance alignment matrix method using DALI Lite. Molecular graphics and analyses were performed with the UCSF Chimera package (48).

RESULTS AND DISCUSSION
Metal Ion Promiscuity in MtGpgS-In most GT-A glycosyltransferases, metal ions play a central role in substrate recognition and catalysis (2,23). Specifically, in MtGpgS, the conserved His 258 residue and the carboxylate groups of the Asp 134 -Ser 135 -Asp 136 signature are involved in divalent metal cation coordination, which also coordinates to the ␣and ␤-phosphate moieties of the donor substrate UDP-Glc. The divalent metal cation has been proposed to play the role of the Lewis acid during catalysis in retaining GT-A enzymes and is an essential co-factor for enzymatic activity in MtGpgS (Fig. 2) (2). Therefore, we decided to investigate the influence of several metal ions on the activity of the enzyme. In contrast to previous reports (49,50), we found that MtGpgS was not only active in the presence of Mg 2ϩ but also when other metal cations were used as co-factors. As depicted in Fig. 2, the enzyme requires Mg 2ϩ for maximal activity. However, MtGpgS was enzymatically active when another group II metal ion (Ca 2ϩ ) and transition metal ions (Mn 2ϩ , Co 2ϩ , or Fe 2ϩ ) were present in the reaction mixture. MtGpgS was not active in the presence of Zn 2ϩ or Cu 2ϩ . Our results thus point to a relatively broad specificity of GpgS for metal cations.
Dynamic Loop as Key Factor in MtGpgS-mediated Catalysis-The first described crystal structure of a mycobacterial GpgS (GT81 family) was that of M. avium subsp. paratuberculosis (MaGpgS; apo-form and Mn 2ϩ ⅐UDP, Mn 2ϩ ⅐UDP-Glc, UDP-Glc and UDP-GlcNAc complexed forms; Ref. 41) followed by that of M. tuberculosis (MtGpgS; apo-form and Mg 2ϩ ⅐UDP⅐ PGA complexed form (47), and apo-form, Mg 2ϩ ⅐UDP and Mn 2ϩ ⅐UDP⅐PGA (Table 1)). MaGpgS and MtGpgS are structurally closely related enzymes (root mean square deviation value of 1.4 Å) displaying an 82% sequence identity and 95% sequence similarity. Both MaGpgS and MtGpgS are homodimers and display the characteristic two tightly bound domain organization of GT-A glycosyltransferases (Fig. 3A) (2).
Flexibility and conformational heterogeneity of a loop connecting the acceptor binding domain and the C-terminal extension (residues 253-262, linking ␤8 and ␣9 in MtGpgS; the numbering system for MtGpgS is used for both MaGpgS and MtGpgS throughout this paper, unless stated otherwise) appears to be critical during substrate binding and catalysis in GpgS enzymes (Fig. 3, A and B). In the apo-form of MtGpgS, we observed that this flexible loop adopts a double conformation in both monomers of the protein dimer (r.m.s.d. value of 1.88 for six residues; Fig. 3, C and D), a property that was not seen in the previous reported apo crystal structures of GpgS. These conformations correspond to a catalytically active (L A , relative occupancy of 60%) and inactive (L I , relative occupancy of 40%) states of the loop in the active site. L A and L I most likely represent distinct energy states of the L loop rather than crystallization artifacts because they do not participate in crystal packing interactions. A detailed analysis of intermolecular interactions showed four residues, Arg 256 , Ala 257 , His 258 , and Arg 261 in the L loop, to be of particular importance. In the inactive L I conformation, the Arg 256 makes hydrogen bonds with the side chain OH of Asp 136 , which is part of the Asp-Xaa-Asp motif, and Glu 212 . In addition, Arg 261 is hydrogen bonded with the lateral chain of Tyr 229 . Although in the active L A conformation the two arginine residues conserved the described binding motif, a new hydrogen bond was formed between the main chains of Ala 257 and Ile 138 . Importantly, also found within this loop was His 258 , which plays a fundamental role in metal coordination in GpgS and other GT-A enzymes, including the UDP-GalNAc:polypeptide ␣-N-acetylgalactosaminyltransferase T1 (GT27 family) and the mannosylglycerate synthase (GT78 family) (41,51,52). The side chain of His 258 in the L A conformation is in an optimal position to readily coordinate with metal ions, although it is far away from the nucleotide-binding site in the L I conformation making van der Waals interaction with the side chain of Tyr 165 .
Notably, the active L A conformation of the L loop observed in the apo-form of MtGpgS is conserved in two other relevant structural states of the enzyme, the metal⅐nucleotide or metal⅐nucleotide sugar donor-bound complexes (Mg 2ϩ ⅐UDP, Mn 2ϩ ⅐UDP, and Mn 2ϩ ⅐UDP-Glc), and the metal⅐nucleotide acceptor-bound complex (Mn 2ϩ ⅐UDP⅐PGA) (Tables 2 and 3 for r.m.s.d. and // torsion angle values respectively; Fig. 3D). In all protein⅐substrate complexes, Arg 256 and Arg 261 slightly change their side chain traces to allow for metal ion coordination and substrate binding. The side chain of Arg 256 participates in Mg 2ϩ or Mn 2ϩ coordination and makes hydrogen bonding with Glu 212 , whereas Arg 261 makes electrostatic interaction with the ␣-phosphate of UDP-Glc. Importantly, the orientation of the key His 258 residue found in the L A conformation is preserved in the complexes and coordinates Mg 2ϩ or Mn 2ϩ ions. In addition, the hydrogen bonding between Ala 257 and Ile 138 is also conserved. Altogether, the structural data indicated that the conformation of the L loop detected during catalysis is already present in the free enzyme, suggesting that the protein conformation necessary for catalysis is an intrinsic property of GpgS.
Donor Recognition Site, Two Conformations for the ␤-Phosphate-The uridine moiety binds to a pocket in the N-terminal domain mainly defined by the connecting loops ␤2-␣2 (residues 80 -87), ␤3-␣3 (residues 50 -56), ␤5-␣6 (residues 133-143), ␣8-␣9 (residues 221-230), and L A , where it makes a number of hydrophobic and hydrophilic contacts with the protein. Of particular relevance, Ser 81 makes hydrogen bond with the uridyl O 2 providing the basis for the nucleoside specificity. Moreover, the side chain of Tyr 229 makes an important hydrogen bond with the O 5 of the ␣-phosphate of UDP. Its side chain conformation is also conserved in the apo-form and other ligand-bound forms of the enzyme. Interestingly, the ␤-phosphate of UDP binds to the enzyme into two different conformations (Fig. 4A). In the first conformation, the ␤-phosphate is oriented toward the ␣-face of the ribose ring and in close proximity to the catalytic center. Consequently, the sugar moiety is favorably positioned for its transfer to the acceptor substrate PGA. This conformation has been observed in MaGpgS⅐Mn 2ϩ ⅐UDP, MaGpgS⅐Mn 2ϩ ⅐UDP-Glc, MtGpgS⅐Mg 2ϩ ⅐UDP⅐PGA, and other GT-A homologous enzymes, including MgS from Rhodothermus marinus, which catalyzes the synthesis of ␣-mannosyl-D-glycerate using GDP-Man as donor sugar ( Fig. 4B; family 78) (29,41,47,52). In contrast, we found that in the second conformation, the ␤-phosphate, is on the contrary oriented toward the ␤-face of the ribose ring, solvent-exposed, and away from the catalytic site. Specifically, the ␤-phosphate rotates 240°, and its O 2 makes a new hydrogen bond with the side chain of Glu 54 . This conformer has been observed in MtGpgS⅐Mn 2ϩ ⅐UDP⅐PGA (this study) and in the homologous enzyme MgS from R. marinus, and it likely corresponds to the nucleoside diphosphate moiety of UDP-Glc leaving the catalytic site after sugar moiety transfer (52).
Acceptor Recognition Site-The acceptor-binding pocket in MtGpgS is located on the top of ␣-helix 7, which is also involved in protein dimerization. The carboxyl group of PGA makes a hydrogen bond with the side chain of Thr 187 , whereas Arg 185 positions its guanidinium group in close contact with the phosphate moiety (2.88 Å ArgN⑀ and the PO4 O3; Fig. 4). A conformational change was observed in the connecting loop ␤6-␣7, which presents an intrinsic flexibility provided by three consecutive glycine residues. In the MaGpgS⅐Mn 2ϩ ⅐UPD-Glc complex, which does not contain the acceptor substrate, Gly 183 -Gly 184 occupy the acceptor-binding pocket, whereas in MtGpgS⅐Mn 2ϩ ⅐UDP⅐PGA, the equivalent residues are oriented in opposite direction and away from the binding site, allowing for PGA binding. This places the accepting OH2 group of PGA at ϳ2.4 Å from the anomeric C1 of the modeled glucose moiety of UDP-Glc. This result is in contrast with a previous report in which the predicted distance between OH2 atom of PGA and C1 of the glucose moiety was ϳ5.48 Å, clearly not compatible for the glucose transfer to take place (47). Interestingly, the PGA and D-glycerate groups lie in equivalent positions in MtGpgS and MgS glycosyltransferases (MgS⅐D-glycerate complex; see Ref. 29), where the accepting hydroxyl group is located at ϳ2.3 Å from the anomeric C1 of the sugar ring. Furthermore, mutation of Thr 139 in MgS (which is equivalent to Thr 187 in MtGpgS) resulted in a 1500-fold increase in the K m for D-glycerate (52), highlighting its role in acceptor binding and suggesting a common acceptor recognition mechanism for GpgS and MgS.
Thermodynamics of MtGpgS-Substrate Interactions-To investigate further the molecular mechanism of donor and acceptor substrates binding to MtGpgS, binding reactions were studied in solution by isothermal titration calorimetry. First, binding of the donor substrate was studied in the presence of manganese ions. It is worth noting that in the absence of metal ions binding isotherms showed weak affinity confirming the requirement of divalent cations for UDP-Glc binding (data not shown) as observed previously with bovine ␣-1,3-galactosyltransferase and more recently with MgS (20,52). UDP-Glc bound to MtGpgS with an apparent stoichiometry of one ligand molecule per protein monomer (Fig. 5). However, the binding isotherm was atypical revealing two clearly detectable reactions, one with decreasing heats of reaction at low ligand to protein molar ratios the other with increasing binding heats at high molar ratios, revealing a complex binding process (Fig. 5A). This binding isotherm could not be fitted to a bimolecular association model. This peculiar binding process was even better observed with UDP as the heat contribution of the reaction at low UDP to protein molar ratios was greater than with UDP-Glc binding (Fig. 5A). The observation that MtGpgS bound both UDP-Glc and UDP with a similar binding process and overall stoichiometries of one nucleotide per protein monomer together with the observation that the heat of the low molar ratio reaction decreased with increasing temperatures from 15 to 35°C (supplemental Fig. S1) brought us to consider MtGpgS protein as being in equilibrium between two conformations, P 0 and P 1 , with only one conformation, P 1 , binding the nucleotide diphosphate ligand ("Experimental Procedure" and Table 4) as shown in Reaction 3, Within this model, nucleotide diphosphate (L) binding to P 1 conformation shifts the P 0 P 1 conformational equilibrium of

onformational Dynamics in GT-A Glycosyltransferases
the protein toward the bound P 1 conformation. This model allowed precise predictions of the observed binding isotherms ( Fig. 5 and supplemental Fig. S1). UDP-Glc bound MtGpgS with high affinity and a largely exothermic and enthalpy-driven reaction with a large heat capacity change on binding (K d ϭ 4 M, ⌬H/⌬G ϭ 172%, ⌬C p ϭ Ϫ368 cal⅐(mol⅐K) Ϫ1 ; Fig. 5A and Table 4), a set of binding parameters in agreement with the MaGpgS/Mn 2ϩ ⅐UDP-Glc crystal structure and the involvement of hydrophilic interactions in sugar donor association (41). With respect to UDP-Glc binding, UDP bound to MtGpgS with a 4-fold lower affinity, a 4 kcal/mol smaller enthalpy, and a 45 cal⅐(mol⅐K) Ϫ1 binding heat capacity reduction, revealing the contribution of the glucose moiety to the binding process (Table 4). Based on best fits of UDP binding isotherms at various temperatures, the apparent amounts of P 0 and free P 1 protein conformations varied from 31 and 69% at 15°C to 10 and 90% at 35°C, respectively. At 15°C, the P 0 P 1 transition was largely endothermic (⌬H e ϭ 7.1 kcal/mol) and entropy-driven (supplemental Fig. S1 and Table 4). The nucleotide binding process was further investigated by testing UMP and uridine binding. Both UMP and uridine bound to MtGpgS, however, with one main difference, i.e. binding isotherms exhibited one binding transition only that could be precisely fitted to a simple bimolecular association model with a binding stoichiometry of one ligand per protein monomer. Binding was enthalpy-driven with binding parameters   JULY 13, 2012 • VOLUME 287 • NUMBER 29  (45)) was found in an inactive conformation or partially unmodeled, respectively. similar to those with UDP binding (UDP and uridine binding affinities were 4-and 6-fold lower than UDP-Glc binding affinity, respectively; Fig. 5A and Table 4). An observation clearly indicated that UMP and uridine bound to the two MtGpgS P 0 and P 1 protein conformations with equal affinities, respectively. Taken together, these results emphasize the important role of the ␤-PO 4 in stabilizing the donor substrate or product MtGpgS complexes in the P 1 conformation, in agreement with the crystal data.

Conformational Dynamics in GT-A Glycosyltransferases
Second, binding of the acceptor substrate was studied by ITC. PGA bound to the UDP MtGpgS complex with a 1:1 stoichiometry with respect to protein monomer, a high affinity with an exothermic and enthalpy-driven reaction (K d ϭ 4 M, ⌬H/⌬G ϭ 85%) with a large heat capacity change on binding (⌬C p ϭ Ϫ475 kcal/mol) ( Fig. 5B and Table 4). PGA also bound to the UMP MtGpgS complex with a 1:1 stoichiometry, however, with a completely different binding process. Binding was endothermic and largely entropy-driven with a 2-fold lower binding affinity (K d ϭ 9 M, ⌬H/⌬G ϭ Ϫ23%; Fig. 1B and Table  1). Although tested at different temperatures, in the presence or absence of metal cations, PGA binding to the uridine MtGpgS complex or to free MtGpgS could not be detected (Fig. 5B).
Results of the ITC study clearly demonstrated that binding of the donor and acceptor substrates was sequential. Furthermore, UDP and UMP binding to MtGpgS leads to the formation of different PGA protein complexes. Altogether, the results of the ITC and x-ray studies make it tempting to hypothesize that P 0 and P 1 MtGpgS protein conformations could correspond to the catalytically inactive (L I ) and active (L A ) states of the protein, respectively, as observed in the protein crystals.
Overall Conformational Flexibility of MtGpgS-To further characterize the effect of substrate binding on the conformation of MtGpgS, we performed limited proteolysis experiments. When incubated with trypsin, the enzyme was rapidly degraded (Fig. 6A). Similar profiles to that observed with the unliganded enzyme were obtained when UDP-Glc, UDP, UMP, URI, or PGA alone were present in the reaction mixture. As shown in Fig. 6A, the presence of both UDP and PGA substrates slightly  Ⅺ). B, MtGpgS was first titrated with a nucleotide as described in A, and the protein⅐nucleotide complex formed was then titrated with PGA. Thermodynamic data are reported in Table 4. protected MtGpgS from degradation by the protease. Sedimentation velocity AUC studies of pure MtGpgS were in agreement with the proteolysis experiments (Fig. 6B). The nearly vertical distribution s indicates that MtGpgS sedimented as a single homogeneous species with an average sedimentation coefficient of 4.42 s, which is consistent with a dimeric protein (71,311 Da). Upon addition of equimolar UDP and PGA, the sedimentation coefficients increased slightly to 4.50 s, whereas the presence of UDP-Glc or its derivatives did not significantly affect the s values of MtGpgS. Taking into account the apparent 1:1 stoichiometry of binding and the relatively minor increase in the molecular weight of MtGpgS upon ligand binding, this change in the sedimentation coefficient indicates the formation of a slightly less compact structure. Altogether, our results suggest that although the conformation of the catalytic loop L plays a central role during the catalytic cycle, the overall structure of the enzyme remains unchanged.
Structural Comparison with GT-A GTs-To date, the threedimensional crystal structures of GT-A enzymes in all three relevant functional states of their catalytic cycles (i.e. the ligandfree form, the binary complex with bound nucleotide (NDP), or nucleotide sugar (NDP-sugar) donor, and the ternary complex with bound nucleotide (NDP) and acceptor substrates/derivatives) have been reported for families GT6, GT7, GT8, GT13, GT14, GT15, GT29, GT43, and GT64 (2,11). Interestingly, a careful inspection of all available structures revealed examples wherein very few differences are found between the conformations of the apo-and complexed forms suggesting that, similar to GpgS, the catalytic site of some GT-A enzymes might be preformed before donor and acceptor binding. In the ST3Gal-I sialyltransferase (GT29), there are no significant structural changes of the protein main chain in the active site upon Gal␤1,3-GalNAc-␣-PhNO 2 or CMP/Gal␤1,3-GalNAc-␣-PhNO 2 binding, consistent with the random order mechanism determined for this enzyme (53,54). The ␣1,2-mannosyltransferase Kre2p/Mnt1p from Saccharomyces cerevisiae (GT15), involved in both N-linked outer chain and O-linked oligosaccharide biosynthesis, displays an r.m.s.d. between the ligand-free form and its binary and ternary complexes of 0.17 and 0.19 Å, respectively (55). Only a limited number of side chain conformational changes occur in the enzyme upon binding of the donor and acceptor substrates. The fact that an acceptor substrate⅐enzyme binary complex could not be obtained in this case suggests that the acceptor-binding site may only be available after binding of the donor substrate, which is consistent with a sequential ordered mechanism determined for other retaining GT-A GTs. Similarly, the overall structure of the apo-form of the UDP-GlcA:galactosylgalactosylxylosylprotein 3-␤-glucuronosyltransferase (GT43) is almost similar to the ternary complex with Mn 2ϩ ⅐UDP⅐N-acetyllactosamine (r.m.s.d. of 0.38 Å) (56). Interestingly, only the side chains of three basic residues (Lys 153 , Arg 165 , and Arg 313 ) undergo conformational changes upon UDP binding. In contrast, a large conformational change induced after donor and/or acceptor binding has been observed in several GT-A enzymes, including the UDP-Gal:␤-galactoside ␣-1,3-galactosyltransferase (GT6) (20), the UDP-Gal: ␤-GlcNAc ␤-1,4-galactosyltransferase T1 (GT7) (23), glycogenin (GT8) (57), and the ␣1,4-N-acetylhexosaminyltransferase EXTL2 (GT64) (58).
Concluding Remarks-As highlighted by the structural and biophysical evidence presented herein, the intrinsic flexibility of an accurate region of the active site of GpgS plays a central role during the donor and acceptor substrates recognition that seems to be of significant relevance during the glycosyl transfer. The crystal structure of the apo-form of MtGpgS revealed two distinct conformational states of the protein characterized by a highly dynamic nature of the L loop. The conformation of the L loop in the binary Mn 2ϩ ⅐UDP-Glc and the ternary Mn 2ϩ ⅐UDP⅐PGA complexes displays minor structural rearrangements when compared with the L A state in the free enzyme. Essentially the very few differences involve the lateral chains of two basic residues, Arg 256 and Arg 261 , suggesting that the conformation necessary for catalysis is an intrinsic property of GpgS. The crystallographic snapshots of GpgS during its reaction cycle and calorimetric data strongly support a prominent influence of the nucleotide ␣and ␤-phosphates in substrate binding and catalysis. Whereas the ␣-phosphate is stabilized by a stacking interaction with the conserved Tyr 229 , the ␤-phosphate seems to alternate between two conformations, which likely correspond to the pre-and post-sugar transfer states. Intriguingly, GpgS shows uncommon metal ion preferences for a GT-A enzyme with a broad range of metal cations capable of assisting catalysis.
Recent reports show a remarkable role for protein conformational dynamics in substrate recognition and product release and enzyme catalysis (59 -64). These conformational dynamics seem to act locally and allosterically to modulate the affinity and selectivity of enzymes, signaling proteins, and receptors (65). The current scenario shows the conformational dynamics of the L loop of GpgS as a major determinant in metal/substrate association and catalysis and opens the debate of whether a "conformational selection rather than an "induced-fit" mechanism might govern substrate recognition. Nevertheless, further studies would be required to confirm this hypothesis and the occurrence of a similar model in other GT-A glycosyltransferases.