ABO(H) Blood Group A and B Glycosyltransferases Recognize Substrate via Specific Conformational Changes*

The final step in the enzymatic synthesis of the ABO(H) blood group A and B antigens is catalyzed by two closely related glycosyltransferases, an α-(1→3)-N-acetylgalactosaminyltransferase (GTA) and an α-(1→3)-galactosyltransferase (GTB). Of their 354 amino acid residues, GTA and GTB differ by only four “critical” residues. High resolution structures for GTB and the GTA/GTB chimeric enzymes GTB/G176R and GTB/G176R/G235S bound to a panel of donor and acceptor analog substrates reveal “open,” “semi-closed,” and “closed” conformations as the enzymes go from the unliganded to the liganded states. In the open form the internal polypeptide loop (amino acid residues 177-195) adjacent to the active site in the unliganded or H antigen-bound enzymes is composed of two α-helices spanning Arg180-Met186 and Arg188-Asp194, respectively. The semi-closed and closed forms of the enzymes are generated by binding of UDP or of UDP and H antigen analogs, respectively, and show that these helices merge to form a single distorted helical structure with alternating α-310-α character that partially occludes the active site. The closed form is distinguished from the semi-closed form by the ordering of the final nine C-terminal residues through the formation of hydrogen bonds to both UDP and H antigen analogs. The semi-closed forms for various mutants generally show significantly more disorder than the open forms, whereas the closed forms display little or no disorder depending strongly on the identity of residue 176. Finally, the use of synthetic analogs reveals how H antigen acceptor binding can be critical in stabilizing the closed conformation. These structures demonstrate a delicately balanced substrate recognition mechanism and give insight on critical aspects of donor and acceptor specificity, on the order of substrate binding, and on the requirements for catalysis.

Glycosyltransferases synthesize carbohydrate moieties of glycoconjugates by catalyzing the sequential addition of monosaccharides from specific donors to specific acceptors. The ubiquitous presence of glycolipids and glycoproteins in all living systems underlines the importance of the glycosyltransferases superfamily, and the DNA of all domains of life encode for a large number of these enzymes (1). To date, crystal structures of glycosyltransferases have displayed a high degree of structural similarity even when there is low sequence homology (2)(3)(4). As such, glycosyltransferases provide an excellent example of the preferential conservation of structural phenotype over the conservation of sequence identity (2), which indicates that the mechanism of glycosylation, although not yet fully understood, has been conserved. Elucidation of the details of substrate recognition would allow the development of new inhibitors for the treatment of microbial diseases (5), genetic ailments such as diabetes (6), and cancer (7). The generation of inhibitors of the blood group A and B synthesizing glycosyltransferases GTA 2 and GTB have been reported (8,9), including an inhibitor-bound structure (10).
Most glycosyltransferases are observed to lie in one of two major fold families, GT-A and GT-B (not to be confused with the GTA and GTB enzymes discussed here) (2,3,11). Structural studies have revealed that specific internal sections of polypeptide adjacent to the active site are often observed to be flexible or completely disordered. These internal loops have been suggested to restrict water access to the active site, as well as act in donor recognition and catalysis (3), including the inverting enzymes ␤4Gal-T1 (12), GnT-I (13), GlcAT-I (14), and GlcAT-P (15); the retaining enzymes EXTL2 (16) ␣-(133)-GalT (17,18), GTA, and GTB (19,20); the microbial inverting SpsA (21) and CstII (22); and the retaining microbial enzyme LgtC (23).
The retaining ␣-(133)-galactosyltransferase (␣-(133)-GalT) is the enzyme most homologous to GTA/GTB in sequence and structure, and it has been reported to display substrate-induced conformational changes (18). This enzyme transfers Gal from UDP-Gal to oligosaccharides terminating in lactose or LacNAc (␤-D-Gal-(134)-␤-D-GlcNAc) (24). Like GTA and GTB, ␣-(133)-GalT is a retaining enzyme with a GTA fold. Unlike GTA and GTB, the structure of ␣-(133)-GalT displays a completely ordered internal loop in the unliganded state, which has been reported to lie in different conformations for different mutants and in substrate-bound and unbound complexes (17,18).
GTA and GTB are responsible for the generation of the human ABO(H) blood group A and B antigens (25,26). GTA catalyzes the transfer of GalNAc from UDP-GalNAc to the H antigen acceptor (␣-L-Fuc-(132)-␤-D-Gal-O-R, where R is glycolipid or glycoprotein) to form the A antigen, whereas GTB catalyzes the transfer of Gal from UDP-Gal to the H antigen acceptor to form the B antigen (27,28). Initial high resolution structural studies of both GTA and GTB revealed two regions of disordered polypeptide (19). One region consisted of the last 10 residues of the C terminus, whereas the other was an internal polypeptide loop composed of residues 177-195. Subsequent studies have shown that part of the disorder of the internal loop was because of the presence of a heavy atom, and that crystals of the mutant enzyme GTB/C209A grown in the absence of heavy atoms display a smaller disordered segment of the internal loop consisting of residues 177-187 (20).
GTA and GTB are the two most homologous glycosyltransferases known that utilize different nucleotide donors and differ by only 4 of 354 amino acids as follows: Arg/Gly 176 , Gly/Ser 235 , Leu/Met 266 , and Gly/Ala 268 in GTA and GTB, respectively (29). The role of each critical residue in donor and acceptor recognition has been studied through the generation of chimeric GTA/GTB enzymes. A nomenclature based on these four critical amino acid residues has been developed to describe GTA and GTB chimera, where GTA can be referred to as AAAA and GTB as BBBB with each letter corresponding to one critical residue in increasing order, such that the ABBB chimera would correspond to the GTB/G176R mutant enzyme and AABB would correspond to the GTB/G176R/S235G mutant enzyme. Critical residues Leu/Met 266 and Gly/Ala 268 have been shown to be responsible for discrimination between the two donor molecules (30 -32), whereas Gly/Ser 235 and Leu/Met 266 significantly impact acceptor recognition (33); however, the function of the conserved mutation Arg/Gly 176 has been elusive. Structural studies in the past have been hampered by the fact that Arg/Gly 176 lies at the edge of the internal disordered loop from residues 176 -195; however, the development of crystallization conditions for BBBB (GTB), ABBB, and AABB in the absence of heavy atoms permits a structural investigation of the influence of residue 176 on loop ordering and substrate binding.

Construction of the Synthetic Glycosyltransferase Chimeric
Genes ABBB and AABB-The synthetic wild-type GTA (designated AAAA, amino acids 53-354) gene was constructed from synthetic oligonucleotides as described previously (34). The synthetic gene was designed with unique restriction sites to facilitate mutagenesis. Glycosyltransferase chimeric mutants ABBB and AABB were synthesized by digesting the AAAA gene with KpnI/SphI and ligating in the appropriate oligonucleotides to form the desired gene sequence.
The Ϫ10/ABBB and Ϫ10/AABB genes (amino acids 63-354) were made by PCR amplification using the wild-type ABBB and AABB genes as templates. The forward primer 5Ј-ATA TGA ATT CAT GGT TTC CCT GCC GCG TAT GGT TTA CCC GCA GCC GAA-3Ј (MIN2) introduced an EcoRI site in the 5Ј end, and the reverse primer 5Ј-ATA ATT AAG CTT CTA TCA CGG GTT ACG AAC AGC CTG GTG GTT TTT-3Ј (PCR-3B) introduced a HindIII site in the 3Ј end. The PCR profile used was 94°C/3 min (94°C, 30 s, 55°C, 30 s, and 72°C, 1 min) for 30 cycles. After gel purification, the PCR products were digested with EcoRI and HindIII for 2 h at 37°C and were ligated into pCW⌬lac, which had been opened with EcoRI/HindIII. Each ligation was transformed into BL21-competent cells. The DNA sequences were confirmed on both strands.
All insert and plasmid purifications were made by Qiagen plasmid purification system (Qiagen, Chatsworth, CA). All ligations were made by the use of T4 DNA ligase (Invitrogen) at room temperature for 1 h. All restriction enzymes were purchased from New England Biolabs.
Protein Purification-Mutant enzymes were purified from Escherichia coli by methods described previously (36), with the exception of R188H and R188K where cells were disrupted at 1.35 kbar with a constant system cell disrupter. Expression levels for mutants were good, and the yields of final purified proteins were ABBB 36 mg/liter, AABB 50 mg/liter, R188S 8 mg/liter, R188K 66 mg/liter, and R188H 15 mg/liter.
Kinetic Characterization-Kinetics using ␣-L-Fucp-(132)-␤-D-Galp-O-R as an acceptor were carried out with a radiochemical assay, where a Sep-Pak reverse-phase cartridge is used to isolate radiolabeled reaction products created when the label is transferred from a radioactive donor to the hydrophobic acceptor (37). Assays were performed at 37°C in a total volume of 12 l containing substrates and enzyme in 50 mM MOPS buffer, pH 7.0, 20 mM MnCl 2 , and bovine serum albumin (1 mg/ml). Seven different concentrations of donor and acceptor were employed, and initial rate conditions were linear with no more than 10% of the substrate consumed in the reaction. For the donors, the K m values were determined at 1.0 mM acceptor, and the K m for the acceptor was determined at 1.0 mM donor. The kinetic parameters k cat and K m were obtained by nonlinear regression analysis of the Michaelis-Menten equation with the Graph Pad PRISM 3.0 program (GraphPad Software, San Diego). Two-substrate kinetic analysis was performed for the AABB and ABBB mutants to obtain K A (acceptor K m ), K B (donor K m ), K ib , and K ia , as described previously (34). K ib is the apparent Michaelis constant for donor that is independent of the concentration of acceptor and thus corresponds to the dissociation constant of the enzyme⅐UDP-Gal or enzyme⅐UDP-GalNAc complexes. K ia is the dissociation constant of the enzyme⅐acceptor complex.
Crystallization-All proteins were crystallized using conditions different from those reported previously (10,19,33,35,38,39). Whereas the first crystals of GTB were grown from relatively low protein concentrations (ϳ8 -15 mg/ml) and as a mercury derivative, the crystals in this paper were initially generated from higher protein concentrations (ϳ60 -75 mg/ml). The first crystals of the ABBB and AABB mutants grew in stock solutions containing 20 mM MOPS, pH 7.0, 75 mM NaCl, 15 mM ␤-mercaptoethanol, 0.05% NaN 3 and stored at 4°C for several months. Crystals of ABBB and AABB were washed with mother liquor consisting of 7% PEG-4000, 70 mM Ada buffer (N-(2acetamindo)iminodiacetic acid), pH 7.5, 30 mM sodium acetate buffer, pH 4.6, 40 mM ammonium sulfate, and 5 mM MnCl 2 . Crystals of BBBB were obtained by the hanging drop method from 30 to 40 mg/ml fresh protein solutions containing 1% PEG, 4.5% methyl-pentanediol (MPD), 0.1 M ammonium sulfate, 0.07 M NaCl, 0.05 M Ada buffer, pH 7.5, 5 mM MnCl 2 against a reservoir containing 2.7% PEG-4000, 7% MPD, 0.32 M ammonium sulfate, 0.25 M NaCl, and 0.2 M Ada buffer, pH 7.5. Crystals of BBBB, ABBB, and AABB in complex with UDP, (41) in various combinations were obtained by soaking substrate into the unliganded crystals. Crystals were washed with mother liquor consisting of 7% PEG-4000, 70 mM Ada buffer, pH 7.5, 30 mM sodium acetate buffer, pH 4.6, 40 mM ammonium sulfate, and 5 mM MnCl 2 . The concentration of UDP was usually 25 mM, but as little as 10 mM was often sufficient, and 50 mM was used for BBBBϩUDP. The H antigen acceptor analogs HA, DA, and ADA concentrations ranged from 10 to 20 mM. The concentration of UDP-Gal ranged from 35 to 50 mM. The concentration of MnCl 2 was 5 mM. All substrates were added incrementally over a period of a few minutes to a few hours so as to prevent crystal fracture. In the case of AABBϩUDP-GalϩDA, additional UDP-Gal was added to the crystal minutes before freezing to minimize the extent of UDP-Gal hydrolysis. No UDP was added to AABBϩUDP, as the UDP appeared to follow the protein through the purification process. The UDP was removed to generate the AABBϩH structure by washing the crystal with 10 mM EDTA to remove the manganese that bound the UDP to the protein.
Data Collection and Reduction-X-ray diffraction data were collected at Ϫ160°C for all crystals using a CryoStream 700 crystal cooler. Each crystal was incubated with a cryoprotectant solution that consisted of mother liquor with 30% (v/v) glycerol replacing a corresponding volume of water, except AABBϩUDP-GalϩDA where a corresponding volume of MPD was used. Data were collected on a Rigaku R-AXIS IV 2ϩ area detector at distances of 72 mm and exposure times between 4.0 and 7.0 min for 0.5°oscillations. X-rays were produced by an MM-002 generator (Rigaku/MSC, College Station, TX) coupled to Osmic "Blue" confocal x-ray mirrors with power levels of 30 watts (Osmic, Auburn Hills, MI). The data were scaled, averaged, and integrated using d*trek and CrystalView (42).
Structure Determination-Although the structures were nearly isomorphous, for completeness all structures were solved by molecular replacement using the CCP4 module MOLREP (43,44) with the structure of wild-type GTB as a starting model (Protein Data Bank accession code 1LZ7), and subsequently refined using the CCP4 module REFMAC5 (45). All figures were produced using Setor (46) and SetoRibbon. 3
The primary distinguishing characteristic among the structures of the liganded and unliganded forms of BBBB, ABBB, and AABB can be found in the two regions of polypeptide observed to be completely disordered in the original structures of GTA and GTB (19). In general, the internal loop of the BBBB, ABBB, and AABB structures show fewer disordered residues than the corresponding region in the heavy atom structures (19). A summary of the observed electron density surrounding the internal loop (residues 176 -195) and the C terminus (residues 346 -354) for all structures is given in Table 3. Without exception, structures containing Arg 176 show significantly more order than the corresponding structure containing Gly 176 . All ABBB and AABB structures display large portions of the internal loop, which is observed to adopt an "open" conformation when unli-ganded, a "semi-closed" conformation when bound to UDP, and a "closed" conformation when bound to UDP or UDP-Gal and acceptor (Fig. 1, a and b). Although the structures of the BBBB enzymes display significant levels of disorder in the mobile polypeptide loops, the relative movement of the observed residues indicates that they undergo similar conformational shifts upon substrate binding.
For all structures, the internal loop itself can be divided into two portions. The first structure consists of residues 175-188 that shows significant flexibility and contains an ␣-helix consisting of residues 180 -187. The second structure consists of residues 189 -195 that adopts an ␣-helical conformation similar to that observed in a mutant GTB structure (20). The nine C-terminal residues remain disordered in the open or semi-closed states but display various levels of order in the closed conformation depending on the presence of substrate and on the identity of Arg 176 . Structures soaked with UDP sometimes show partial occupancy, whereas all structures soaked with H antigen analogs display a fully occupied acceptor binding site.
Disorder in BBBB, ABBB, and AABB-The identity of residue 176 (arginine in AXXX enzymes and glycine in BXXX enzymes) is not only strongly correlated with the level of order observed in the internal polypeptide loop of which it is a part, but with that of the C-terminal residues as well. Given that the ABBB and AABB mutants display remarkably higher levels of order and detail than the wild-type BBBB enzyme, the structures of the mutants will be discussed first and then compared with the wild type.
ABBB and AABB Structures-A comparison of electron density observed in the different complexes of ABBB reveals that the level of order in the internal and C-terminal loops changes significantly with different substrate groupings ( Table 3). The unliganded ABBB structure displays excellent electron density along almost the entire length of the polypeptide (including almost the entire internal loop), with the nine C-terminal residues completely disordered. The ABBBϩUDP and AABBϩUDP structures show electron density corresponding to a partially occupied UDP molecule and a level of order comparable with the unliganded structure; however, there is clear evidence for two alternative conformations of approximately equal weight for residues 176 -188 (Fig. 1, c and d). In one conformation the loop follows the path observed in the unliganded structure, whereas in the second conformation these same residues  move toward the UDP molecule to partially occlude the active site to form the semi-closed state. The AABBϩHA structure displays a similar degree of order as the ABBB unliganded structure (having complete disorder only for Ala 177 and Lys 179 in the internal loop); however, there is no evidence of order in the C-terminal region. Interestingly, attempts were made to crystallize ABBB in the presence of UDP-Gal; however, the structure displays only low UDP occupancy in the active site and no conformational shift, indicating that the Gal moiety is disordered or that the UDP-Gal has hydrolyzed (structure not included). The most striking structures in this series are ABBBϩUDPϩHA and AABBϩUDP-GalϩDA, which both display excellent electron density for almost the entire polypeptide chain (including the C-terminal residues) and unambiguous electron density for both the UDP and the HA. The structure of AABBϩUDP-GalϩDA displays electron density corresponding to UDP-Gal and a fully occupied DA. The internal loop is not disordered over two conformations but shows a 100% conformational change in that it corresponds to the semiclosed state in ABBBϩUDP. Together, the conformational shift in the internal loop and ordering of the C terminus result in the completely occluded active site of the closed conformation of the enzyme (Table 3 and Fig. 1a). The C-terminal residue His 348 forms unambiguous hydrogen bonds with the O-2-and O-3hydroxyl groups of the ␣-L-Fucp moiety on the acceptor molecule (Fig. 2a). In contrast, the structure of ABBB in complex with ADA (remembering that ADA is the H antigen acceptor analog that lacks the O-2-hydroxyl group) shows complete electron density only for C-terminal residues Lys 346 and Asn 347 , main chain density only for His 348 to Arg 352 , and complete disorder for residues His 348 , Asn 353 , and Pro 354 (Table 3 and Fig. 2b).
ABBBϩUDPϩHA shows a fully occupied glycerol molecule (cryoprotectant) in the donor binding site of the enzyme (Fig.  2c). A comparison of this structure with AABBϩUDP-GalϩDA shows that the glycerol molecule is positioned to mimic the interaction of the galactosyl residue with Arg 188 from the internal loop (Fig. 2, c and d).
BBBB Structures-The BBBB structures show the same two major regions of disorder. Unlike the ABBB and AABB structures, the internal polypeptide loop cannot be clearly divided

in BBBB, ABBB, and AABB
Black one-letter amino acid codes correspond to unambiguous electron density for main chain and side chain atoms; green letters correspond to unambiguous electron density for main chain atoms only; red letters correspond to weak or ambiguous electron density for main chain and side chain atoms. The internal loops of ABBBϩUDP and AABBϩUDP are disordered over two conformations corresponding to the open (1) and closed (2) forms. Residues involved in helices are underlined. Residues observed to move from the open, to semi-closed, or closed form are outlined in yellow. Residues with one-letter amino acid codes in lowercase have not been included in the refined models. Substrate moieties that exhibit partial occupancy are shown in green.
into a flexible and an ordered region for many of the structures, as all the residues from 176 to 195 can display disorder (Table  3). Notably, only the unliganded BBBB and fully liganded BBBBϩUDPϩHA structure show significant order, and the binding of a single ligand, either UDP or the H antigen, results in appreciably more disorder. There are only 4 residues (Ala 177 to Arg 180 ) disordered in the internal loop of BBBB, whereas BBBBϩUDP and BBBBϩHA show disorder in 9 and 11 residues, respectively (Ala 177 to Ser 185 in BBBBϩUDP and Gly 176 to Met 186 in BBBBϩHA). The region 189 -195 observed to be an ␣-helix in ABBB and AABB, as well as the GTB/C209A mutant (20), also displays somewhat more disorder upon the binding of either substrate alone. Binding of both substrates in the BBBBϩUDPϩHA complex causes most of the polypeptide main chain of the internal and C-terminal loops to become ordered and so form the closed state. The main chain carbonyls of Val 351 and Arg 352 interact with the O-4-carbonyl of the uracil moiety of the UDP through a bridging water molecule, and the side chain of Arg 352 forms salt bridges with both phosphate moieties of UDP; however, electron density corresponding to the last three C-terminal residues is absent. There is no evidence of glycerol in the donor binding site of BBBBϩUDPϩHA.
As found in the ABBB structures, increased disorder in the C-terminal residues of the corresponding BBBB structures is observed when the ADA is substituted for the HA or DA acceptor analogs. Although BBBBϩUDPϩHA shows the closed form of the enzyme with electron density corresponding to Lys 346 -Val 351 , the BBBBϩUDPϩADA structure displays significantly more disorder in the C-terminal region with electron density corresponding only to residues Lys 346 and Asn 347 (Fig. 2, e and f). It is important to note the large degree of movement of polypeptide that was permitted in the crystalline state of BBBB, ABBB, and AABB and that great care had to be taken to add substrate slowly to prevent crystal cracking.
Kinetic Parameters-Kinetic constants for wild-type GTA, GTB and mutant enzymes are given in Table  4. The chimeric enzymes ABBB and AABB show tight binding of UDP-Gal with dissociation constants (K ib ) of 1.6 and 1.1 M, respectively. Acceptor binding is also tighter for these mutants with K ia values of 3.8 and 0.94 M. The k cat value for UDP-Gal for ABBB is comparable with that of GTB but is reduced from 5.1 to 2.2 s Ϫ1 for AABB. There was a marginal increase in k cat for UDP-GalNAc from 0.41 for GTB to 0.65 and 0.60 s Ϫ1 for ABBB and AABB confirming the dominance of Leu/Met 266 and Gly/Ala 268 in donor discrimination. The binding of the alternate donor UDP-GalNAc was also tighter for the chimeric enzymes than for GTB with dissociation constants of 9.2 and 44 M compared with 69 M; however, donor binding is weaker than that of GTA, which has a K ib of 3 M. The importance of the interaction between Arg 188 and donor is evident from the dramatic reduction in k cat for the R188S and R188K mutants.

DISCUSSION
The Open Conformation for the Enzymes-In the absence of donor or acceptor, BBBB and ABBB crystallize in the open form, where the nine C-terminal residues are disordered, and a major portion of the internal loop is disordered or lies in a conformation that leaves the donor and acceptor binding sites exposed to solvent. The effect of Arg 176 on internal loop structure is clearly evident, as a substantial portion of the internal loop is disordered in BBBB whereas most of the loop is ordered in ABBB, where it consists of two helical segments joined at Arg 187 (Table 3 and Fig.  1d). This open form is likely because of the mutual repulsion of many positively charged residues, such as internal loop residues Lys 179 , Arg 180 , and Arg 188 as well as C-terminal residues Arg 352 and Lys 346 (Fig. 3a).

UDP Binding Induces a Semi-closed Conformation-The
ABBBϩUDP and AABBϩUDP structures reveal a fascinating transition between the open and semi-closed states as both structures display clear evidence that residues 176 -188 are disordered over both conformations (Fig. 1c). The semi-closed state has the helix formed by residues 176 -188 moving as much as 6 Å toward the UDP molecule to partially occlude the active site without forming any new hydrogen bonds to the UDP moiety ( Fig. 1, a and b). The change to the semi-closed form results in the first helix of the internal loop moving into alignment with the second (Fig. 1d). In this shift a new main chain hydrogen bond forms between Arg 187 (N) and Asp 183 (O) and a main chain hydrogen bond between Glu 190 (N) and Met 186 (O) transfers to between Met 189 (N) and Met 186 (O), such that two helices become linked by a single turn of a 3 10 helix (Fig. 1d). The result is a distorted helical structure with mixed ␣-3 10 -␣ character that partially occludes the active site. The mutual repulsion of positively charged residues Lys 179 , Arg 180 , and Arg 188 that held the enzyme in the open state are likely overcome to form the semi-closed conformation through electrostatic interactions with the negatively charged pyrophosphate moiety of bound UDP (20). Interestingly, despite high concentrations of UDP, both the ABBBϩUDP and AABBϩUDP structures show electron density corresponding to ϳ50% occupancy, which correlates to the occupancy in the two observed conformations of the internal loop. In contrast, BBBBϩUDP does not show clear evidence of a split between its open and semi-closed states. Indeed, there is significantly higher thermal motion in the semi-closed form seen in BBBBϩUDP compared with the unliganded form. Although only a few residues in the internal loop of BBBBϩUDP can be seen in the electron density maps, it is clear that these at least have moved to positions that correspond to the semi-closed conformation; however, the remainder of the internal loop in BBBB displays a great number or even a continuum of conformations between the two states.
ABBBϩUDPϩHA and BBBBϩUDPϩHA Display a Closed Conformation-The fully liganded ABBB enzyme shows ordering of almost all previously disordered residues in both the C terminus and internal loop. This closed conformation in ABBB and BBBB is achieved only in the presence of both UDP and acceptor. Those residues of the internal loop that are ordered are in the same conformation as observed in the semi-closed state of ABBBϩUDP, and the mutual repulsion observed among the positively charged residues in the internal loop has been fully overcome by the combination of UDP bind-ing and the interaction of the UDP with the newly ordered C terminus (Fig. 3b).
Significantly, six of the nine C-terminal residues of the protein form a short ␣-helix (residues 347-352) that makes contact with residues in the active site, with UDP, with the ␣-L-Fucp moiety of the acceptor, and completes the sequestration of the substrates from solvent. The side chain of Lys 346 extends into the active site to form a salt bridge with the ␤-phosphate of UDP and the side chain of the third residue (Asp 213 ) of the DXD motif.
Although relative levels of disorder and thermal motion clearly show that the internal loop is stabilized by the ordering of the C-terminal loop, there are no direct hydrogen bonds between these two flexible regions. Instead this stabilization occurs though a number of bridging interactions moderated by UDP moiety and three water molecules. The only direct contact between the internal loop and the C terminus occurs through a stacking interaction between Trp 181 and Arg 352 (Fig. 3c).
Effect of Cryoprotectant-A fully occupied glycerol molecule is seen in the acceptor binding site of each structure in the absence of acceptor; however, given that this molecule does not contact either mobile polypeptide loop, and given that it is displaced by even modest concentrations of acceptor, it is unlikely to influence the conformation of these loops.
ABBBϩUDPϩHA is the only structure to display a glycerol molecule in the donor binding site, where it may contribute to the observed formation of the closed state mimicking (through a bridging water molecule) the interaction of the galactosyl moiety in AABBϩUDP-GalϩDA with Arg 188 of the internal mobile loop (Fig. 2, c and d). Both BBBBϩUDPϩHA and ABBBϩUDPϩADA form the closed conformation without any indication of glycerol in the donor binding site.
AABB and Binding of UDP-Gal-Crystals of AABB soaked with UDP-Gal and DA revealed a highly occupied donor and acceptor in the active site cleft with the enzyme in a closed conformation. This represents the fully liganded state required for turnover of the enzyme and, in combination with other structures that bind the active acceptor disaccharide analog HA, a complete schematic of substrate recognition can be drawn (Fig. 4). The donor sugar is bound in the classic "folded back" conformation observed for other glycosyltransferases (Fig. 3d). The shift to the closed conformation does bring Ser 185 and Arg 188 into the donor sugar binding site, but the donor displays a somewhat different hydrogen bond pattern for the ␣-Gal moiety than predicted (47).
Although the hydrogen bonds between Asp 211 and the O-3hydroxyl group and between Asp 302 and the O-4-hydroxyl group are observed, the predicted interaction between Ser 185 and the O-6-hydroxyl group is not observed. Instead, hydrogen bonds are found between Arg 188 and the O-3-hydroxyl group, and between His 301 and the O-6-hydroxyl group (Fig. 3e).
Although it does not participate in the active recognition of the donor sugar galactosyl residue, Ser 185 is positioned to provide a steric barrier to the binding of UDP-Glc and accounts for this aspect of donor specificity. The ability of Ser 185 to exclude UDP-Glc had been predicted, leading to speculation that an appropriate mutation at position 185 could allow the GTB to transfer glucose to the H antigen (48). Furthermore, the ␣-L-Fuc moiety, and that the ␣-L-Fuc moiety is required for efficient catalysis, it can be concluded that closed conformation is likely required for efficient catalysis. The stabilizing effect that O-2-hydroxyl group of the ␣-L-Fucp residue imparts on the C-terminal region can be seen in structures of BBBB and ABBB in the presence of UDP and the ADA acceptor analog, which lacks this important hydroxyl group. In general, the C-terminal residues in both ADA structures display considerably higher levels of disorder than in the analogous structures soaked with HA, having complete main chain and side chain electron density corresponding only to residues 346 in BBBB and residues 346 and 347 in ABBB (Table  3). Significantly, both structures displayed complete disorder for the side chains of His 348 (involved in fucose recognition) and Arg 352 (involved in UDP stabilization) (Fig. 2, b and f).
Effects of Loop Mutations on Enzyme Activity-Kinetic studies completed on mutants of internal loop residue 188 of GTA and GTB can now be rationalized on the basis of the current structures (Table 4 and Fig. 4). For example, the BBBB/R188S and BBBB/R188K mutants demonstrate increases in K m values for the donor and large decreases in k cat , which is consistent with its role in donor sugar recognition and turnover. BBBB/ R188H had a specific activity 3% that of BBBB/R188K and was not further characterized.
Comparison with ␣-(133)-GalT-The most closely related CAZy family 6 glycosyltransferase to GTA and GTB that has been structurally characterized is bovine ␣-(133)-GalT. A mutant of ␣-(133)-GalT has recently been crystallized in the presence of the donor analog UDP-2-fluoro-Gal (18). With a sequence similarity of only 45%, it is not surprising that bovine ␣-(133)-GalT displays significant differences with GTB and the GTB/GTA chimera. First, unlike BBBB, the corresponding internal loop in ␣-(133)-GalT has always been observed to be ordered in the wild-type enzyme. Furthermore, whereas the generation of semi-closed form involves a complete ordering of the internal loop in BBBB and a conformational change in the loop in ABBB, the binding of UDP in wild-type ␣-(133)-GalT or of UDP-2-fluoro-Gal in its R365K mutant results in five residues of the internal loop changing from random coil to an extra turn of a helix. Also unlike BBBB, ABBB, and AABB, the C-terminal region of the R365K mutant structure bound to UDP-2fluoro-Gal does not close but is observed to curve away from the active site. Like UDP-Gal in AABBϩUDP-GalϩDA, the donor sugar nucleotide UDP-2-fluoro-Gal lies in a "tucked-un-