Novel UDP-GalNAc Derivative Structures Provide Insight into the Donor Specificity of Human Blood Group Glycosyltransferase

Two closely related glycosyltransferases are responsible for the final step of the biosynthesis of ABO(H) human blood group A and B antigens. The two enzymes differ by only four amino acid residues, which determine whether the enzymes transfer GalNAc from UDP-GalNAc or Gal from UDP-Gal to the H-antigen acceptor. The enzymes belong to the class of GT-A folded enzymes, grouped as GT6 in the CAZy database, and are characterized by a single domain with a metal dependent retaining reaction mechanism. However, the exact role of the four amino acid residues in the specificity of the enzymes is still unresolved. In this study, we report the first structural information of a dual specificity cis-AB blood group glycosyltransferase in complex with a synthetic UDP-GalNAc derivative. Interestingly, the GalNAc moiety adopts an unusual yet catalytically productive conformation in the binding pocket, which is different from the “tucked under” conformation previously observed for the UDP-Gal donor. In addition, we show that this UDP-GalNAc derivative in complex with the H-antigen acceptor provokes the same unusual binding pocket closure as seen for the corresponding UDP-Gal derivative. Despite this, the two derivatives show vastly different kinetic properties. Our results provide a important structural insight into the donor substrate specificity and utilization in blood group biosynthesis, which can very likely be exploited for the development of new glycosyltransferase inhibitors and probes.

chain of Arg 352 in the equally flexible C terminus. Upon binding of native donor substrates to blood group GTs, this stacking interaction stabilizes the "closed" conformation and contributes to the H-antigen acceptor (HAA) binding site, which is required for full catalytic activity. Hence, obstruction of this crucial conformational transition provides an explanation for the inhibitory activity of 1 and 2 toward AAGlyB (7,19). The structures of AAGlyB in complex with compounds 1 or 2 and the HAA substrate acceptor, ␣-L-Fucp-(132)-␤-D-Galp, in conjunction with detailed enzymological data, also provided an explanation for the different binding affinities and turnover rates of the two different UDP-Gal derivatives (12). Intriguingly, the ternary structure of the AAGlyB-1-HAA complex revealed an alternatively closed conformation of the C terminus forming hydrogen bonds with the 5-formylthienyl substituent. This alternatively closed conformation was not present in the AAGlyB-2-HAA structure presumably because of the lack of hydrogen bonding opportunities between the C terminus and the 5-phenyl substituent in 2. Therefore, the ability of AAGlyB to adopt this alternative conformation in 1 but not in 2 provided an explanation for the difference in residual rate of product formation between the two derivatives. It also suggested a reason for the elevated K m for acceptor substrate with compound 1 because a histidine residue in this alternative confor-mation of the C terminus is no longer hydrogen bonding to the acceptor.
Although several structures of human blood group GTs in complex with UDP-Gal have been reported (7,12,15), attempts to solve a structure with UDP-GalNAc have remained elusive due to the very rapid hydrolysis of the donor when soaked into the crystals. Therefore, we have expanded our ligand design to the UDP-GalNAc congener of inhibitor 1 (compound 3, Fig.  1B) assuming that this UDP-GalNAc derivative would also behave as an inhibitor with a low turnover rate.
Herein, we present the synthesis of compound 3 as well as structures of AAGlyB in complex with 3 alone, and in complex with 3 together with the HAA-acceptor. Importantly, the structure of the AAGlyB-3 complex contains the donor analogue with a partially intact GalNAc portion, which, for the first time, gives an indication of the binding mode of UDP-GalNAc to a human blood group GT. In this new structure, the GalNAc moiety is placed in the active site in a conformation, which is different from the canonical tucked under conformation previously observed in other UDP-sugar-GT complex structures (7, 12, 15-18, 20 -24).
Importantly, this new GalNAc conformation is catalytically productive and, supported by enzyme kinetics, we suggest that donor hydrolysis along with additional conformational changes is a prerequisite for proper HAA binding. The new structures allow a direct comparison, for the first time, of the recognition and utilization of the two alternative donors UDP-Gal and UDP-GalNAc by AAGlyB, and offer a structural explanation for the dual specificity of this enzyme.

Synthesis of UDP-GalNAc Derivative
3, Preparative Chromatography-Ion-pair chromatography was performed using Lichroprep RP-18 resin gradient 0 -10% acetonitrile (or methanol) against 0.05 M triethylammonium bicarbonate over 480 ml at a flow rate of 5 ml/min. Productcontaining fractions were combined and reduced to dryness. The residue was co-evaporated repeatedly with methanol to remove residual triethylammonium bicarbonate. Anion-exchange chromatography was performed using a MacroPrep 25Q resin, gradient 0 -100% 1 M triethylammonium bicarbonate (pH 7.3) against H 2 O over 480 ml at a flow rate of 5 ml/min. Product-containing fractions were combined and reduced to dryness. The residue was co-evaporated repeatedly with methanol to remove residual triethylammonium bicarbonate. 5-Iodouridine-5␣-monophosphate 5 (292 mg, 0.65 mmol) (19) was dissolved in dry dimethyl sulfoxide and co-evaporated (3 times) with dry N,N-dimethylformamide to remove residual water. The residue was dissolved in 0.5 ml of dry dimethyl sulfoxide. Morpholine (400 l, 4.6 mmol) was added, and the mixture was stirred at room temperature for 5 min. Dipyridyl disulfide (500 mg, 2.3 mmol) and triphenylphosphine (600 mg, 2.3 mmol) were added at 5-min intervals. The reaction mixture was further stirred for 60 min at room temperature before being quenched with 0.1 M NaI in acetone until a colorless solid precipitated out of the solution. The supernatant was removed and the phosphoromorpholidate of 5 (sodium salt) was isolated by filtration as a colorless powder (354 mg, 99% yield). This material was sufficiently pure for analytical characterization and was used in the next step without further purification. 5-Iodo-UDP-GalNAc (4)-5-Iodouridine-5␣-monophosphoromorpholidate (33 mg, 0.06 mmol, 1 eq) was co-evaporated with dry pyridine (3 ϫ 5 ml). N-Acetylgalactosamine-1-phosphate in its tributylammonium salt form (ref or prep) (2.5 eq) was added to the dry morpholidate mixture and further coevaporated (3 ϫ 5 ml) with pyridine. The dry residue was dissolved in dry N,N-dimethylformamide (2 ml) under nitrogen atmosphere and tetrazole in dry acetonitrile (5 eq) was added. After 5 h, the reaction reached completion and all solvents were removed under reduced pressure. The crude residue was purified sequentially by ion-exchange and ion-pair chromatography to give the title compound in its triethylammonium salt form (1.6 eq.) as a colorless, glassy solid in 27% yield (15.7 mg). 5-(5-Formylthien-2-yl) UDP-GalNAc (3)-A 2-necked round bottom flask with 5-iodouridine-5Ј-diphosphate-␣-D-N-acetylgalactosamine 4 (12.0 mg, 16.4 mol, 1 eq), Cs 2 CO 3 (2 eq), and 5-formyl-2-thiopheneboronic acid (1.5 eq) was purged with N 2 . Triphenylphosphine-3,3Ј,3Љ-trisulfonic acid trisodium salt (0.0625 eq), Na 2 Cl 4 Pd (0.025 eq), and degassed H 2 O (4 ml) were added and the reaction was stirred under N 2 for 1 h at 50°C. Upon completion, the reaction was cooled to room temperature and the pH was adjusted to 7 with 1% HCl. The black suspension was filtered through a membrane filter (0.45 m). The filter was washed with H 2 O and the combined filtrates were evaporated under reduced pressure. The residue was sequentially purified by ion exchange and ion-pair chromatography to give the title compound in its triethylammonium salt form (1.5 eq) as a glassy solid in 76% yield (11.9 mg).

Cloning, Purification, and Crystallization
The AAGlyB was cloned, expressed, and purified in Escherichia coli as previously described (12,25,26) and the crystallization condition was described in Refs. 7 and 12. The AAGlyB-3 complex crystals were flash cooled in liquid N 2 after adding a cryosolution consisting of reservoir solution including 20% glycerol and 25 mM compound 3 to a crystal drop and letting it soak for 30 min at 4°C. The AAGlyB-3-HAA crystals were flash cooled in liquid N 2 as for AAGlyB-3 crystals but with an additional 25 mM ␣-Fucp-(132)-␤-Galp-O-(CH 2 ) 7 -CH 3 (HAA) acceptor added to the cryosolution.

Enzyme Kinetics
The K m , k cat , and K i values for UDP-Gal, UDP-GalNAc, 1, and HAA acceptor with AAGlyB were previously described (7). To avoid substrate depletion, the assay was only incubated for 8 min. For acceptor kinetics with 3, the following concentrations were used: 3 (33.3 M) and TMR-labeled acceptor (250 -1.9 M), incubated for 20 min. The K m and V max values were obtained by fitting the data points to a Michaelis-Menten curve using GraphPad Prism 4 (GraphPad Software). The K i value for 3 with AAGlyB competing with UDP-[6-3 H]Gal transfer was determined by a standard SepPak reversephase radiochemical assay as previously described (15). K i values were determined by linear regression of Dixon plots using 100 M HAA, 2 M UDP-Gal, 300,000 dpm of UDP-[6-3 H]Gal, and 0, 0.5, 1, and 2 M 3. Because the GalNAc moiety in compound 3 is not radiolabeled this assay set-up does not differentiate between inhibitor and substrate behavior of 3. Reduced incorporation of labeled [6-3 H]Gal from UDP-[6-3 H]Gal can be the result either of inhibition, or incorporation of unlabeled GalNAc. Therefore, the determination of K i values in this assay is compatible with the observed donor substrate behavior of 3.

Dowex Hydrolysis Assay
Hydrolysis of UDP-Gal and UDP-GalNAc by AAGlyB was followed with a radiochemical assay, as previously described (28) using the following conditions: 50 mM MOPS (pH 7), 20 mM MgCl 2 , 1 mg/ml of bovine serum albumin, 300 M UDP-Gal or UDP-GalNAc, 300,000 dpm of UDP-[6-3 H]Gal or UDP-[6-3 H]GalNAc, 350 and 35 M enzyme for the UDP-Gal and UDP-GalNAc assays, respectively. The 10-l reaction mixtures were incubated in a thermo block at 37°C for 60 min. The reaction was terminated by adding 300 l of ice-cold H 2 O and transferred to Mini-columns (5 ml, Sigma), which were packed with 0.4 g of AG1-X8 Resin (chloride 200 -400 mesh, Bio-Rad). The reaction product was washed first with 400 l and then 500 l of H 2 O by brief centrifugation and transferred to a scintillation vial. Ten milliliters of Ecolite TM ϩ scintillation fluid (MP Biomedicals) was added, and radioactivity was measured in a LS 6500 scintillation counter (Beckman Coulter TM ).

NMR Experiments
NMR was used to study the stereochemistry of the hydrolysis reaction. In these experiments the reaction medium consisted of 500 M UDP-GalNAc and 17.5 M AAGlyB or 250 M 3 and 30 M AAGlyB, in 50 mM deuterated BisTris buffer (pH 7) containing 10 mM MgCl 2 . After mixing donor and enzyme the NMR spectra were recorded at 291 K on a Bruker Advance 800 instrument at 799.96 MHz for proton using a 5-mm cryo probe. The assignment was based on Bruker standard experiments as two-dimensional DQFCOSY, NOESY, and TOCSY recorded with 4 K data points and 512 increments and an acquisition time of 0.23 s, 0.6-s NOESY mixing time, and 0.08-s spinlock in the TOCSY experiment.

Data Collection, Reduction, and Structure Solution
X-ray diffraction data were collected to 1.9-Å resolution on the AAGlyB-3 complex at beamline I911-2 at the MAX II synchrotron in Lund, Sweden ( ϭ 1.038 Å, 100 K). X-ray diffraction data on the AAGlyB-3-HAA complex was collected to 1.65-Å resolution at beamline BL14.1 at the BESSY II synchrotron in Berlin, Germany ( ϭ 0.918 Å, 100 K). Both data sets were integrated and scaled by XDS (29). Subsequently, the structures were solved by molecular replacement using the module Phaser (30) including the solved structure of wild-type GTB (PDB entry 2RIT) as a search model. The two structures belong to space group P2 1 2 1 2 with two molecules in the asymmetric unit. Restrained refinement was initially carried out in REFMAC5 (31) before iteratively rebuilding the structures using Coot (32) and finally refining in Phenix (33) with TLS included. Ramachandran plots were calculated in PROCHECK (34) and show that all residues are within the allowed region in both structures. The crystallographic details are presented in Table 1. The structural figures were all prepared in PyMOL (DeLano Scientific LLG).

Results
Synthesis-For the preparation of the new UDP-GalNAc derivative 3 we adapted our previously developed protocol for the Suzuki-Miyaura cross-coupling of unprotected sugar-nucleotides (Scheme 1) (7, 35-37). 5-Iodo-UDP-GalNAc 4 was prepared from GalNAc-1-phosphate and 5-iodo-UMP 5 via conversion of the mononucleotide into the corresponding phosphoromorpholidate, followed by tetrazole-catalyzed pyrophosphate bond formation (38). Pleasingly, the free acetamido group in 4 did not interfere with the cross-coupling reaction in the next and final synthetic step. Thus, a water-soluble catalyst was used to successfully cross-couple the complete sugar nucleotide 4 with 5-formyl-2-thiopheneboronic acid to give the target 76% yield of UDP-GalNAc derivative 3.
Enzyme Kinetics-The kinetic parameters for the glycosyl transfer reaction of AAGlyB with UDP-Gal derivative 1 as donor and the HAA acceptor were described previously (Table  2) (7,12). These results showed that the 5-formylthienyl substituent in 1 led to efficient inhibition of transferase activity. In this study, using a capillary electrophoresis assay with TMRlabeled acceptor as previously described (12) we determined the K m and k cat values of the new UDP-GalNAc donor analogue, 3, with AAGlyB. Interestingly, the enzymological profile of 3 is markedly different from that of 1 (Table 2). Donor analogue 3, which differs from 1 only in the nature of the sugar, has a K m value similar to that of 1, but is turned over about 25-fold faster, despite the identical nature of the 5-substituent in 1 and 3. Overall, 3 has a very similar enzymological profile to its 5-unsubstituted parent UDP-GalNAc, with a K m that is about 3-fold lower and similar values for k cat and K m for acceptor. This suggests that, although 3 binds to the enzyme with similar affinity as 1, the 5-formylthienyl substituent is not sufficient in this case to prevent product formation. Interestingly, in conjunction with 3 behaving as a natural substrate the acceptor K m is also not elevated compared with the acceptor K m for the reaction using UDP-GalNAc. This is in contrast to the higher K m value for acceptor when using 1, which has a much lower turnover rate and it may suggest that donor hydrolysis is a prerequisite for proper acceptor binding. Finally, we also evaluated the ability of 3 to compete with UDP-Gal binding to AAGlyB in a radiochemical assay. As previously seen for 1, compound 3 competes for binding with a K i value, which is in the same range as K m for 3. However, because compound 3 behaves similar to the natural UDP-GalNAc substrate the reduced incorporation of labeled Crystallographic Analysis-Although a structure of a human blood group GT in complex with a deoxy-antigen acceptor (DA) and intact UDP-Gal has previously been reported (15), a structure with the alternative UDP-GalNAc donor has remained elusive. AAGlyB catalyzes the hydrolysis of UDP-GalNAc with a rate that is ϳ30 times higher than for UDP-Gal (Table 2). This is in keeping with previous findings with wildtype blood group enzymes, that the hydrolysis of UDP-GalNAc is catalyzed much more readily by GTA than the hydrolysis of UDP-Gal by GTB (39).
To further examine the enzymological profile of the UDP-GalNAc derivative, 3, and in particular its unexpected donor substrate activity, we solved two crystal structures of AAGlyB in complex with 3. One complex is with 3 alone and one complex is with the HAA acceptor in the binding site. The statistical details for the x-ray data collection and refinement of the AAGlyB structures are shown in Table 1. The data diffracted to a maximum resolution of 1.75 and 1.65 Å with final R free values of 18.43 and 19.4% for AAGlyB-3 and AAGlyB-3-HAA, respectively. Both structures belong to space group P2 1 2 1 2 containing 2 molecules in the asymmetric unit and with similar unit cell parameters. The two structures are highly isomorphous except for the last 10 residues of the C terminus, which are only visible in chain A of AAGlyB-3-HAA (residues 345-354).
AAGlyB Structure in Complex with 3-By carrying out the crystal soaks at 4°C we ultimately obtained a structure of the AAGlyB-3 complex including a visible GalNAc in chain B of the two chains in the asymmetric unit ( Fig. 2A). A difference SCHEME 1. Synthesis of UDP-GalNAc derivative 3. Reagents and conditions: (i) morpholine, 2,2Ј-dipyridyldisulfide, PPh 3 , dimethyl sulfoxide, room temperature, 1 h; (ii) ␣-D-GalNAc-1-phosphate, tetrazole, MeCN, N,N-dimethylformamide, room temperature, 5 h; (iii) 5-formyl-2-thiopheneboronic acid, Cs 2 CO 3 , triphenylphosphine-3,3Ј,3Љ-trisulfonic acid trisodium salt, Na 2 Cl 4 Pd, H 2 O, 50°C, 1 h. map of the AAGlyB structure also shows hints of a GalNAc in two different conformations in the active site of chain A, but the quality of the map in this area did not permit building a plausible model of the two GalNAc conformations. In the AAGlyB-3 structure the internal loop (residues 173-188) in both chains is well defined both in the presence and absence of acceptor. However, in contrast to the previously solved structures of AAGlyB-1 and AAGlyB-2 (12), the loop in AAGlyB-3 is in the closed conformation upon binding of the donor analogue alone (Fig. 2B). AAGlyB Structure in Complex with 3 and HAA-Soaking crystals with both 3 and HAA resulted in a structure with donor and acceptor binding sites occupied by a 5-formylthienyl UDP and a HAA, respectively (Fig. 2C). Binding of 3 and HAA triggers the C terminus of chain A to fold over the active site in the same alternative conformation (Fig. 2D) as previously seen in the AAGlyB-1-HAA structure (12), in which the donor analogue also has the 5-formylthienyl substituent. Similar to AAGlyB-1-HAA the hydrogen bond between HAA and His 348 in AAGlyB-3-HAA is disrupted and is pointing away from the active site. In fact, the two structures are almost identical, as the only distinguishing factor, the sugar moiety, is not seen in either of the two complexes. The similarity to the binding of 1 is supported by our kinetic results, which show comparable K m values for both 1 and 3 (Table 2). However, the similar alternative binding mode for 1 and 3 suggests that the unusual conformation of the C terminus cannot be the reason for the much higher K m value for acceptor binding in the AAGlyB-1-HAA complex.
The GalNAc Conformation-Interestingly, the AAGlyB-3 structure shows the GalNAc ring in a conformation clearly different from the canonical tucked under conformation we have seen in previous blood group enzyme structures with UDP-Gal, 1, or 2 and which is the conformation generally adopted by both UDP-Gal and UDP-GalNAc at a GT binding site (Fig. 3) (7, 12, 15-18, 20, 22-24). Additionally, this GalNAc conformation in AAGlyB is also different from the GalNAc conformation recently described in the similar, but metal-independent GT (CAZy group GT6) from Bacteroides ovatus (21).
In the tucked under conformation in human blood group GTs, the hydroxyl groups of the Gal moiety are forming an elaborate network of hydrogen bonds to residues Arg 188 , Asp 211 , His 301 , and Asp 302 (Fig. 4A, Table 3). In the AAGlyB-3 structure the network of hydrogen bonds to the sugar is entirely different. Although O4Ј of the GalNAc forms hydrogen bonds to Glu 303 and His 233 , O3Ј forms hydrogen bonds to the backbone of Gly 267 and Ala 268 (Fig. 4B, Table 3). Furthermore, in contrast to UDP-Gal in the tucked under conformation, several water molecules are bridging contact between the GalNAc in 3 and His 301 , Ser 185 , and Asp 211 .
The occupancy of the UDP-part of 3 in chain B is 100%, whereas the GalNAc moiety only has ϳ60% occupancy. A simulated annealing F o Ϫ F c omit map of the active site of chain B shows a small gap between the GalNAc and the ␤-phosphate ( Fig. 2A). In addition, a simulated annealing F o Ϫ F c map of the final refined structure with 60% occupancy of intact 3 and 40% of 3 without the GalNAc shows a small blob of negative density at the position of the glycosidic C1-O3B bond between the UDP ␤-phosphate and GalNAc. These electron density maps show that, for at least a fraction of the compound 3 molecules bound to enzyme in the crystal, the glycosidic bond is broken. Importantly, this implies that even in a "non-tucked under" conformation, the AAGlyB in the crystal can still hydrolyze the glycosidic bond of 3. Also, a difference electron density map of the AAGlyB-3 structure indicates that there are two slightly differ-   DECEMBER 25, 2015 • VOLUME 290 • NUMBER 52 ent positions of Asp 213 in the DXD motif and the UDP phosphates (not shown). Furthermore, the strong electron density peak of the Mn 2ϩ ion is also slightly elongated, indicating that there may be an alternative slightly shifted position of the metal. The two slightly different conformations could be a result of the cleavage of the glycosidic bond, which then releases the strain on the nucleotide and causes a shift of the ␣and ␤-phosphates, the Mn 2ϩ and Asp 213 . We chose to model an intact UDP-Gal-NAc donor analogue because this could be done without distortion of the stereochemistry around the glycosidic bond. However, assuming the bond is cleaved and the last refinement job is run again this time with a separate GalNAc moiety, the R free only goes up by ϳ0.1% and the negative density disappears from the C1-O3B bond. Furthermore, C1 is pulled away from the ␤-phosphate, which seems to make the pyranose ring adopt a flattened geometry (Fig. 5). Because the electron density map shows no sign of an O1 attached to C1 it is possible that the cleaved GalNAc exists as the C2 deprotonation product with a C1-C2 double bond, i.e. as a glycal. Although the quality of the electron density for the GalNAc does not justify modeling such an intermediate, this type of stabilized intermediate in the absence of acceptor has previously been suggested for the retaining GT-B-folded enzymes E. coli glycogen synthase (40) and Arabidopsis thaliana sucrose synthase-1 (41).

UDP-GalNAc Donor Analogue Structures and Kinetic Analysis
Donor hydrolysis can be considered as transfer of the sugar to a water molecule, which is a poor acceptor substrate. In the refined position of the separate GalNAc there is a water molecule (H 2 O 786 ) coordinated by the nearby glycerol, which seems to be in a perfect position to attack C1 from the opposite side of the UDP leaving group (Fig. 5). Such an attack would likely result in an inversion of configuration of the anomeric carbon during hydrolysis of the GalNAc in the absence of HAA acceptor. We followed the enzymatic hydrolysis of both the UDP-GalNAc donor substrate and compound 3 catalyzed by AAGlyB by 1 H NMR (Fig. 6). The 1 H NMR spectra show that enzymatic hydrolysis releases ␣-N-acetyl-D-galactosamine as a product that is subsequently transformed into ␤-N-acetyl-Dgalactosamine via mutarotation. This is in accordance with previous experiments showing that hydrolysis of the UDP-Gal donor substrate by GTB also occurs with retention of configuration (28). However, these results indicate that this nearby H 2 O 786 molecule seen in the AAGlyB-3 structure does not attack the C1 atom from the opposite side of the leaving group.   When comparing the binding site of the AAGlyB-1, AAGlyB-2, and AAGlyB-3 and AAGlyB-3-HAA structures it can be seen that the HAA acceptor would not be able to bind to the active site without clashing significantly with the Gal-NAc seen in AAGlyB-3 (Fig. 7A). Therefore, it is likely that the active site undergoes additional conformational changes to put the HAA and the GalNAc in a position suitably coordinated for reaction. In contrast, in the AAGlyB-1 and -2 structures with the galactose in the tucked under conformation the superimposed HAA appears to be able to fit simultaneously together with the donor (Fig. 7B). Here the HAA is potentially forming hydrogen bonds to O1 of the donor galactose as well as to a ␤-phosphate oxygen. Furthermore, O3 of the HAA galactose is ϳ3 Å away from C1 of the donor galactose. However, because the anomeric carbon is retaining its conformation after transfer, here too, additional con-formational changes are needed for obtaining a coordination suitable for the reaction.

Discussion
Our results provide new structural insights into the factors that govern substrate specificity in GTs. AAGlyB is unusual in that this GT can use two different UDP-sugars as donor, with near equal efficiency. The new results help refine our understanding of the structural basis for this dual specificity of AAGlyB. Previous kinetic studies using GTA/GTB chimeric enzymes have shown that residues 266 and 268 are the key residues that determine A/B substrate specificity (25,42). In GTB, residues 266 and 268 are an alanine and methionine, respectively, which readily accommodate the Gal moiety in the tucked under conformation as illustrated in the previously solved GTB-specific AABB-UDP-Gal-DA structure (Fig. 8A). In GTA, Leu 266 and Gly 268 seem to leave sufficient space to accommodate the N-acetyl group of the UDP-GalNAc donor when superimposing a GalNAc moiety onto the Gal in the tucked under conformation (Fig. 8B). This is in contrast to both GTB and AAGlyB, both of which have an alanine residue in position 268. From the new AAGlyB-3 structure it appears that Ala 268 is at least partly responsible for preventing the GalNAc ring in 3 from adopting the tucked under conformation because the N-acetyl moiety of 3 would potentially be clashing with the short side chain of this alanine (Fig. 8C). When superimposing the GTB residues, Met 266 and Ala 268 , together with the GalNAc from the AAGlyB-3 structure and the GalNAc modeled in the tucked under conformation, the Met 266 is clashing severely with both GalNAc conformations (Fig. 8D). In addition, when adding hydrogen atoms to the molecules using the validation program MolProbity (43) it is evident that not only Met 266 but also Ala 268 sterically overlaps with the N-acetyl group (results not shown). Hence, GTB residues at position 266 and 268 are preventing GalNAc from binding in any of the two conformations. In AAGlyB, residue 266 is a glycine, which allows, in conjunction with Ala 268 , the accommodation of the GalNAc residue in an alternative conformation. Although the presence of Ala 268 therefore seems to preclude the GalNAc moiety of 3 from adopting the tucked under conformation in AAGlyB, it does permit, in conjunction with Gly 266 , the orientation of the donor in the alternative and productive conformation, despite the presence of the sterically demanding N-acetyl group.
Taken together, these results therefore provide an explanation for the dual specificity of the AAGlyB mutant, and may also offer an explanation for the different donor substrate specifici-  ties of the wild-type blood group GTs, GTA and GTB. Although we cannot rule out a tucked under conformation of the GalNAc in wild-type GTA, the AAGlyB-3 structure provides valuable information on the function and importance of the two key residues, 266 and 268 in determining the specificity of the blood group GTs. Importantly, this mechanism of donor recognition and discrimination may also be used by other GTs. For example, the equivalent residues 266 and 268 in the Forssman synthetase, a close homologue of blood group synthases in GT6, are glycine and alanine as in AAGlyB (44). The Forssman synthetase is an ␣3-GalNAc transferase that transfers a GalNAc residue onto the glycolipid called the Forssman antigen. A hypothesis could be that, for the enzyme to allow the donor to adopt a productive configuration and act as an ␣3-GalNAc transferase, the active site of the enzyme must comprise a glycine residue either in position 266 (as in AAGlyB and FS) or 268 (as in GTA).
Finally, it is notable that the kinetic parameters for 3, including k cat and acceptor K m , are comparable with those of its parent molecule UDP-GalNAc (Table 2), despite the clashing position of the GalNAc with HAA. In contrast, the kinetic profile of 1 differs markedly from that of its parent UDP-Gal by having a lower k cat and a higher acceptor K m . Furthermore, because both AAGlyB-1-HAA and AAGlyB-3-HAA are forming the alternative conformation of the C terminus the lack of hydrogen bonding between HAA and the C-terminal His 348 residue cannot be the only factor contributing to the elevated acceptor K m in the case of 1. Instead, it is more likely that acceptor binding is preceded by additional conformational changes and/or donor hydrolysis.
In summary, we show for the first time, the structure of a human blood group GT, AAGlyB, in complex with a UDP-Gal-NAc derivative. Intriguingly, this structure adopts a conformation different from the canonical tucked under conformation. The new donor conformation of the GalNAc in 3 is induced by the presence of residues, Gly 266 and Ala 268 , important determinants for donor substrate specificity in blood group GTs. The structure of the AAGlyB-3 complex therefore provides, for the first time, a detailed structural explanation for the dual specificity of AAGlyB. These insights may prove useful for ongoing efforts to expand the donor substrate specificity of GTs for synthetic applications. Finally, the new structural data shows the AAGlyB-3-HAA complex with a well defined internal loop and C terminus in the "pseudo-closed" conformation as also seen for the AAGlyB-1-HAA complex. This confirms that the nature of the substituent is important for the alternative closure of the binding pocket and will facilitate the design of new substituents with higher binding affinity of the donor.
Author Contributions-R. J. collected the x-ray diffraction data, solved the AAGlyB-3 and AAGlyB-3-HAA crystals structures, as well as carried out the crystallographic analysis and the enzymological studies. G. K. W. designed the inhibitors and T. P. carried out the synthetic work. R. J. and M. M. P. designed the study and analyzed the data and discussed the results. R. J. wrote the paper and prepared the crystallographic figures and tables, with contributions from the other authors. All authors commented on the manuscript.