Characterizing human a -1,6-fucosyltransferase (FUT8) substrate specificity and structural similarities with related fucosyltransferases

Mammalian Asn-linked glycans are extensively processed as they transit the secretory pathway to generate diverse glycans on cell surface and secreted glycoproteins. Additional modification of the glycan core by a -1,6-fucose addition to the innermost GlcNAc residue (core fucosylation) is catalyzed by an a -1,6-fucosyltransferase (FUT8). The importance of core fucosylation can be seen in the complex pathological phenotypes of FUT8 null mice, which display defects in cellular signaling, development, and subsequent neonatal lethality. Elevated core fucosylation has also been identified in several human cancers. However, the structural basis for FUT8 substrate specificity remains unknown. Here, using various crystal structures of FUT8 in complex with a donor substrate analog, and with four distinct glycan acceptors, we identify

Mammalian Asn-linked glycans are extensively processed as they transit the secretory pathway to generate diverse glycans on cell surface and secreted glycoproteins. Additional modification of the glycan core by a-1,6-fucose addition to the innermost GlcNAc residue (core fucosylation) is catalyzed by an a-1,6-fucosyltransferase (FUT8). The importance of core fucosylation can be seen in the complex pathological phenotypes of FUT8 null mice, which display defects in cellular signaling, development, and subsequent neonatal lethality. Elevated core fucosylation has also been identified in several human cancers. However, the structural basis for FUT8 substrate specificity remains unknown.
Here, using various crystal structures of FUT8 in complex with a donor substrate analog, and with four distinct glycan acceptors, we identify the molecular basis for FUT8 specificity and activity. The ordering of three active site loops corresponds to an increased occupancy for bound GDP, suggesting an induced-fit folding of the donor-binding subsite. Structures of the various acceptor complexes were compared with kinetic data on FUT8 active site mutants and with specificity data from a library of glycan acceptors to reveal how binding site complementarity and steric hindrance can tune substrate affinity. The FUT8 structure was also compared with other known fucosyltransferases to identify conserved and divergent structural features for donor and acceptor recognition and catalysis. These data provide insights into the evolution of modular templates for donor and acceptor recognition among GT-B fold glycosyltransferases in the synthesis of diverse glycan structures in biological systems.
Initial structural studies on FUT8 over a decade ago determined the overall protein-fold (36), but the absence of bound ligands or substrates provided little insight into substrate recognition. Recently, structures were published for a FUT8-GDP-A2-Asn acceptor complex (41,42) that identified many aspects of substrate recognition. In the present paper we describe several FUT8 structures, including complexes showing that the occupancy of GDP, a sugar donor analog, correlates with the ordering of specific loop regions important for an induced-fit conformation change upon donor binding. We also describe four glycan acceptor complexes that illustrate the structural basis for FUT8 substrate specificity and the roles of acceptor-binding site complementarity and steric hindrance in acceptor substrate selection. Detailed kinetic analysis using a library of glycan acceptors and a collection of FUT8 active site mutants identify the key interactions required for high affinity substrate recognition and catalysis. We also compare the FUT8 structure with several GT-B fucosyltransferases in other CAZy families to identify far more structural similarities between these enzymes than previously appreciated. Despite the strong structural similarities, each enzyme employs a different strategy for acceptor substrate recognition and catalysis. Overall, the data provide a framework for understanding the evolution of donor-and acceptor-binding template structures used by related fucosyltransferases, and by extension other GT-B fold enzymes, in selective substrate recognition and synthesis of diverse glycan structures in biological systems.

Expression, purification, and enzymatic characterization of recombinant human FUT8
The soluble catalytic domain of human FUT8 was expressed as a secreted and N-terminally His-tagged GFP fusion protein in HEK293 cells as previously described for the production and biochemical characterization of numerous human glycosyltransferases ( Fig. S1A) (83). Purification of the enzyme from the conditioned media, fusion tag cleavage, and further purification led to an enzyme preparation amenable to enzymatic and structural studies (Fig. S1B). The oligomeric structure of FUT8 was examined by size exclusionmultiangle light scattering (SEC-MALS) revealing a single peak eluting with a predicted molecular mass of ;132 kDa consistent with a predominately dimeric form for the ;62.6 kDa expression product with tags removed (Fig. S1C).
Crystallization and structure of the FUT8:GDP complex Structural studies on FUT8 followed similar approaches that have been taken with other recent glycosyltransferase structures (32,83,85,86) and other recent structures of human FUT8:A2-Asn complexes (41,42). The soluble catalytic domain of FUT8 was initially crystallized in the presence of 5 mM GDP. X-ray diffraction data were collected to 2.25 Å resolution and the structure was solved by molecular replacement using the prior apo FUT8 structure (PDB 2DE0 (36)) as a search model. There were four chains in the asymmetric unit of the FUT8:GDP complex. The overall structure was comprised of a GT-B-fold but, as noted in the prior FUT8 structures (41,42,71), the enzyme has three large helical segments N-terminal to the GT-B-fold (two involved in homodimerization, see below) and an SH3 domain near the C terminus (Fig. 2).

Dimeric structure of FUT8
The SEC-MALS data for the purified enzyme indicated that FUT8 is a dimer in solution (Fig. S1). Similar to recent analysis of FUT8 structures (42,87), the asymmetric unit of the present FUT8:GDP complex shows that chains B and C form a dimer with an interface that buries 3192 Å 2 of surface area (Fig. 2). Chains A and D form analogous dimers with their crystallographically related symmetry mates. PISA analysis indicates that the interface is stable, with a favorable solvation freeenergy gain (D i G) of 229.8 kcal mol 21 and a p value of 0.057, which corresponds to a high likelihood that the surface is interaction-specific (88) The dimer interface is formed from the first two N-terminal a-helices of FUT8, which fold into a coiled-coil helix structure that associates with the corresponding coiledcoil pairs from adjacent subunits to form a 4-helix bundle, which buries heptad repeat residues in the hydrophobic core (Fig. 2, A and B) (89). Although this 4-helix bundle interaction was recognized in the recent FUT8:GDP:A2-Asn structures (41,42,87), it is notable that this dimer interface is also stabilized by extensive H-bonding and van der Waals interactions between the exposed face of the helical hairpin from one chain and the SH3 domain and residues from the GT-B-fold (residues 388-396) from the adjacent chain ( Fig. 2A). Interestingly, the interactions between the SH3 domain and the lateral face of the 4-helix do not involve prolines, despite the fact the interface is formed from same loop regions that canonical SH3 domains  Table S1 and the respective glycan acceptor structures shown in the lower panel. The nomenclature abbreviation is shown below each structure and the cartoon representations employ standard symbol nomenclature for glycans (116). Asterisk associated with the M5N2-Asn* structures indicates that this substrate contains an Fmoc modification (see "Experimental Procedures").

FUT8 substrate recognition
use to recognize proline-rich ligand sequences (90) (Fig. 2A). The N termini of the two chains are adjacent at the base of the 4-helix bundle, but ;77 amino acids of N-terminal "stem region" (91) were not resolved in the present structure and would be linked to the type 2 transmembrane segment in the full-length FUT8 enzyme (Fig. 2C). The program DISOPRED3 (92) predicts that the linker region of human FUT8 will most likely be intrinsically disordered.
GDP occupancy in the FUT8:GDP complex and the role of flexible loop conformations Although the overall fold of the FUT8:GDP complex was similar to the prior apo structure (PDB 2DE0 (36)) and the recent FUT8:GDP:A2 complexes (41,42), there were notable differences in conformations of several loops between each of the four chains in the GDP complex and also with the previously published structures. The loop regions are comprised of Loop 1 (residues 245-273), Loop 2 (residues 365-378), and Loop 3 (residues 436-443) (Fig. 3B). These three flexible loops enclose the GDP-binding subsite. Chains A and B of the FUT8: GDP complex contained unambiguous electron density for a well-ordered GDP, chain C contains a weakly ordered GDP, and chain D contains no density for the nucleotide (Fig. 3, A  and B). The conformations and ordering of the three loops are strongly correlated with the occupancy of bound ligand. The fact that the crystals were grown in saturating GDP suggests that variable occupancy of the nucleotide is due to the crystal lattice selecting for loop conformations that weakened binding of the donor. Thus, in solution it is likely that these loops are flexible and undergo an induced fit folding upon binding of the donor substrate (Fig. 3D). The structures of each of the chains will be discussed separately below.
Chains A and B of the FUT8:GDP complex exhibit full occupancy for the bound GDP and Loops 1, 2, and 3 interact with the bound nucleotide (Fig. 3, A and B). The loop conformations of chain A are in analogous positions to the previously described FUT8:GDP:A2-Asn complexes (41,42). Loop 2 harbors Arg 365 and Lys 369 , which have direct ionic interactions with the b-phosphate oxygen of the nucleotide and help to enclose the bound nucleotide (Fig. 4, A and B). Minor differences in Loop 2 conformations are seen between chains A and B and are likely due to crystal packing (Fig. 3C). Loop 3 adopts a single conformation in chains A and B, and contributes significant packing interactions to the nucleotide and ribose through Ala 436 and Arg 441 , respectively. In addition, Arg 441 forms a complex ion pairing interaction with Asp 368 in Loop 2, which in turn bridges to Arg 365 in Loop 2 to completely enclose the bound nucleotide (the "Arg 365 -Asp 368 -Arg 441 cage" (Fig. 4C).  (3.6 mM K i ) than GMP (2.3 mM K i ) (71). The ribose O3 hydroxyl also H-bonds with the peptide nitrogen of Tyr 220 , and the purine ring of the nucleoside interacts with His 363 , Thr 408 , and Asp 453 (Fig. 4, A and B). There are additional minor differences in conformation for solvent-exposed side chains between chains A and B, but interactions with the bound GDP are conserved. . Chains A and B show full occupancy of the bound GDP, whereas chain C shows partial occupancy, and chain D shows no occupancy as indicated by the difference density map (F o 2 Fc) for the donor analog contoured at 3s. B, zoom-in representations of the GDP-binding sites. Upper panels, zoom-in of the difference density maps (F o 2 F c ) for the donor analog as indicated by the boxed regions in panel A with cartoon representations of the respective protein regions. Middle panels, chains A-D and the apo-FUT8 structure (PDB 2DE0 (36)) are shown in cartoon representation (chain A, tan; chain B, green; chain C, slate; chain D, yellow; Apo-FUT8 2DE0, orange). Significant differences in conformation were observed in three loops (Loop 1 (residues 245-273, green), Loop 2 (residues 365-378, red), and Loop 3 (residues 436-443, cyan)) with labels in the respective color. In chains C and D, disorder in Loops 2 and 3 were observed and the ends of the respective ordered regions are indicated by the red (Loop 2) or cyan (Loop 3) spheres, respectively. The 2DE0 structure also had a disordered region for Loop 2 indicated by the red spheres. GDP (yellow sticks) is modeled in the binding site of each chain for reference. Lower panels, surface representation of the four chains in FUT8:GDP and the apo-FUT8 (2DE0) structures with GDP (yellow sticks) modeled in the binding site of each chain for reference. Coloring is as shown in the middle panels. C, the progressive conformational changes in Loops 1-3 are illustrated by an overlay of the loops from chains A-D and the PDB 2DE0 structure. A fully extended "flipped-out" conformation is represented by the 2DE0 structure and progressive closure of Loop 2 and Loop 3 and repositioning of Loop 1 in chains C and D leads to partial occupancy in the GDP-binding site for chain C. Further closure of Loops 1-3 in chains A and B leads to full occupancy of the donor analog and completion of the induced fit donor interactions.
In contrast, chain C has significantly weaker electron density for the bound GDP ( Chain D has no visible density for GDP in the binding site ( Fig. 3, A and B). The disorder for Loop 2 was similar to chain C, but residues 430-435 in Loop 3 are positioned to indicate that this peptide segment is extended away from the GDP-binding site reminiscent of the position of the same loop in the PDB 2DE0 structure. A similar loss of binding residues is found for Loops 2 and 3 compared with chain C. In addition, Loop 1 in chain D is altered and does not position Tyr 250 within H-bonding distance of the ribose O 2 hydroxyl in the nucleotide. Thus, loss of interactions with all three loops explains the absence of a bound GDP in chain D.
Finally, a comparison with PDB 2DE0 shows that the loops in the unliganded structure are similar in flexibility to those in chain D (Fig. 3B). In PDB 2DE0, Loop 2 extends away from the active site with residues 373-376 being disordered. Although Loop 3 is ordered in PDB 2DE0, it has moved .25 Å away from the GDP-binding site. The conformation of Loop 1 is similar to that observed in the FUT8:GDP chain D. These altered conformations are additional evidence that all three loops are likely disordered in solution, and supports an induced-fit folding upon nucleotide binding (Fig. 3D).

Structure of the FUT8:GDP:A2-Asn complex
To examine the acceptor substrate interactions, we crystallized a FUT8:GDP:A2-Asn complex. The structure was solved at 2.4 Å resolution in space group P6 5 22 with two chains in the asymmetric unit. Application of crystallographic symmetry reveals the same dimer that we previously observed. The two chains are essentially the same (rmsd 0.20 Å for 464 Ca atoms) with the exception of minor surface loops and solvent-exposed side chain differences. The structure is similar to chain A of the FUT8:GDP (rmsd 0.34 Å for 459 Ca atoms), with the exception that the conformation of Loop 2 is significantly different from the rest of the chains in the FUT8:GDP complex.
Both chains in the FUT8:GDP:A2-Asn complex display welldefined electron density for GDP and the entire A2-Asn molecule, which binds in a complementary surface groove on the face of the GT-B-fold and extends from a position adjacent to the GDP-binding site toward the SH3 domain (Fig. 5A). The FUT8:GDP:A2-Asn structure and its ligand interactions (Fig. 5, B and C) are essentially identical to those identified in the recently published FUT8:GDP:A2 complexes (41, 42) (the rmsd range from 0.45 to 0.54 Å for 450-467 corresponding Ca atoms in PDB 6TKV (41)). Briefly, GlcNAc residue N1 (Fig. 5) is positioned adjacent to the GDP-binding site with the O6 hydroxyl facing toward the GDP ligand through H-bonding interactions with Glu 373 . This residue is positioned in Loop 2 and is fully ordered in its interaction with the hydroxyl nucleophile of the acceptor. Glu 373 also interacts through a salt bridge to Lys 369 and subsequently to the b-phosphate oxygen of the GDP. These interactions suggest that Glu 373 acts as the catalytic base to deprotonate the N1 O6 nucleophilic hydroxyl and then in turn protonate the phosphate of the leaving GDP product (see "Catalytic Mechanism" below). Additional interactions include H-bonds from the N1 O3 hydroxyl to Asp 295 and Lys 216 , the Asn amide nitrogen with Gly 217 , and hydrogen bonds between Gln 470 and the N2 Nacetyl glycosidic oxygen. A further matrix of hydrogen bonding interactions extends throughout the length of the A2-Asn structure all the way to the nonreducing terminal GlcNAc residues (N3 and N4). Many of the interactions with the terminal A2-Asn residues are contributed by the SH3 domain (Fig. 5B, green residue labeling with green boxes). All of the monosaccharides in the acceptor have at least one H-bond to the enzyme surface with the exception of the internal Man residue in the GlcNAc-b1,2-Man-a1,3-Man arm (M2 residue). The details of the acceptor interactions are shown in Fig. 5, B and C. Most notable among these interactions is the position of the SH3 domain, which sterically blocks the linear extension of the A2 glycan beyond the b-linked Man residue (M1) and leads to a bifurcation of the two terminal glycan branches in opposite directions across the surface of the SH3 domain. Each branch has interactions with the SH3 domain including residues N3, M3, and N4. The close interaction of the M1 residue with the SH3 domain surface also precludes binding of glycans containing a bisecting GlcNAc residue.

Enzymatic characterization of FUT8 active site mutants
To confirm the contributions of binding site amino acids to acceptor glycan interactions we mutated each of the respective amino acids to alanine, tested the activity, and performed kinetic analysis on the active mutants (Fig. 6C). A subset of the residues in the GDP-binding site had previously been tested for enzyme activity (36) and we confirmed that mutation of these and other residues that directly interact with the GDP lead to complete enzyme inactivation (Table S3 and Fig. 6, B and C). In addition, we probed residues that contribute to acceptor glycan interactions and found that mutation of three residues, Asp 295 that H-bonds with the O3 hydroxyl of N1, Tyr 498 that provides van der Waals packing interactions with the hydrophobic face of the b-Man residue (M1), and Asp 495 that H-bonds with M1 and the terminal N3 residue, all lead to complete loss of enzyme activity (Fig. 6A). Mutations of other interacting residues result in significantly decreased catalytic efficiency, but not complete inactivation (Fig. 6C). These reductions in k cat /K m include Lys 216 that H-bonds with N1 (160-fold reduction in k cat /K m ), Gln 470 that H-bonds with N2 (80-fold reduction), Asp 494 adjacent to N3 (800-fold reduction), His 535 that H-bonds with N3 (11-fold reduction), and Gln 502 that H-bonds with N4 (16-fold reduction). A similar reduction in activity was previously seen for mutants in the SH3 domain including H535A as well as H535A/Lys541A and T550A/L552A double mutants in prior studies (93) and a collection of residues conserved between Figure 5. FUT8 interactions with the A2 acceptor. A, the structure of the FUT8:A2-Asn acceptor complex is shown as a difference density map (F o -F c ) of the acceptor contoured at 3.5 s before modeling the GlcNAc 2 Man 3 Glc-NAc 2 -Asn structure (yellow sticks). The inset shows a cartoon representation (see Fig. 2 for cartoon nomenclature) of the A2-Asn glycan with labeling of each monosaccharide inside each respective symbol and the corresponding labeling is also employed in the density map. B, the FUT8:A2-Asn complex is shown in yellow stick and cartoon representation with interacting residues shown in cyan stick representation and hydrogen bonds as magenta dashed lines. Monosaccharide residues of the A2-Asn structure are labeled as in Panel A. GDP is shown in yellow stick representation. Residues arising from the SH3 domain are indicated by green residue labeling enclosed in green boxes. C, Ligplot (117) representation depicting packing interactions (red, feathered lines) and hydrogen bonds (green, dashed lines) of A2-Asn (highlighted in yellow) in the FUT8 active site (orange ball and stick with atomic colors).
human FUT8 and chicken c-Src (87) demonstrating the critical role of the SH3 domain in providing acceptor substrate specificity and affinity. Thus, whereas the donor-binding site is exceptionally sensitive to mutational perturbation, the acceptor-binding site has fixed points of essential interaction and other regions that contribute incremental binding energy to catalytic efficiency. The kinetic studies with varied acceptor substrate structures indicated a preference for glycans containing an unmodified GlcNAc-b1,2-Man-a1,3-Man arm (A1-Asn and A2-Asn structures), and reduced activities for substrates that harbor additional modifications to the core A2-Asn structure. The impacts of these modifications on k cat /K m are highlighted in Fig. 7A. To examine substrate interactions with alternative acceptor structures we performed additional co-crystallization studies with A3'-Asn (Fig. 7B), A3-Asn (Fig. 7C), and NM5N2-Asn (Fig. 7D) acceptors. In each case we obtained well-defined electron density for at least a portion of the corresponding acceptor glycan.
The structure of the FUT8:GDP:A3'-Asn complex was solved at 3.3 Å resolution, in the same space group as FUT8:GDP:A2-Asn. FUT8:GDP:A3'-Asn has two chains in the asymmetric unit and both chains display unambiguous electron density for the GDP and for all residues of the A3'-Asn acceptor. The overall enzyme structure was similar to the two chains of the FUT8: GDP:A2-Asn complex (rmsd 0.22-0.29 Å for 465 Ca atoms between the corresponding chains) and the donor and acceptor interactions within the respective binding subsites were identical to the FUT8:GDP:A2-Asn complex. The major difference was the presence of an extra b1,4-GlcNAc residue (N5) on the GlcNAc-b1,2-Man-a1,3-Man arm that distinguishes the A3'-Asn structure from A2-Asn ( Fig. 7B and inset). In both chains, the N5 residue faces into the solvent and does not contribute directly to acceptor-binding interactions.
The structure of the FUT8:GDP:A3-Asn complex was determined at 2.47 Å resolution in space group P6 5 with eight chains in the asymmetric unit. All eight chains display unambiguous electron density for the GDP, but density for the acceptor was evident only in chains A, D, E, and H. In each case where the acceptor density was present, only a single distal arm and the core of the glycan structure (GlcNAc-b1,2-Man-a1,3-Man-b1,4-GlcNAc-b1,4-GlcNAc-b-Asn) was resolved (Fig. 7C). The Man-a1,6-(M3) branching residue from the core b-Man (M1) was disordered along with the two distal GlcNAc residues (N4 and N6) extending from this arm (Fig. 7C and inset). The disorder for these three monosaccharides in the acceptor Figure 6. Structural representation of the FUT8 active site mutants depicting the effects on enzyme kinetics. Residues in the FUT8 acceptorbinding site (A) or donor site (B) were mutated to Ala, expressed, purified, and assayed for enzyme activity using the A2-Asn acceptor substrate (Table  S3). C, k cat /K m values for each of the respective mutants are shown. In A and B, residues that led to enzyme inactivation when mutated to Ala are indicated by cyan stick representations. Residues that reduced activity are indicated by tan stick representation. The K541A mutation had a minor impact on k cat /K m and is indicated in green stick representation. For each modification, the respective glycan structure is shown in red text and the effect on enzyme activity is listed based on enzyme assay data from Table S1. In cases where the respective acceptor was listed as "inactive" there was no activity detected in the in vitro enzyme assays. Values for k cat /K m are listed by comparison to activity using the A2-Asn substrate. The lack of activity toward the G2-Asn substrate or the glycans with bisected GlcNAcs were based on prior assays from other groups (73,75,82). The Polder maps (115) (magenta mesh and contoured at 2.8 s and 3.5 s, respectively) were obtained from the crystal structures of three additional acceptor complexes and are shown in B-D. The Polder maps (115) for A3'-Asn, A3-Asn, and NM5N2-Asn acceptors were calculated following a modified procedure to reduce model bias (see "Experimental procedures"). B, the structure of the FUT8:GDP:A3'-Asn complex was essentially the same as the FUT8:GDP:A2-Asn complex except for the additional b1,4-GlcNAc residue (N5) extending from the Man-a1,3-residue (M2). This residue extends into the solvent and has no additional interactions with the enzyme surface. The inset in B indicates the cartoon representation of the A3'-Asn glycan with residue labeling that is also used to label the residues in the difference density (Polder) map. C, the structure of the FUT8:GDP:A3-Asn complex retained the density for the GlcNAc-b1,2Man-a1,3Man-b1,4GlcNAc-b1,4GlcNAc-bAsn region of the A2-Asn complex. However, the entire extension from the Man-a1,6-arm (residues M3, N4, and N6) was disordered and not resolved in the structure (whited out region in the dotted box for the inset; structure and dotted oval in the Polder map (115)). D, the structure of the FUT8:GDP:NM5N2-Asn complex also retained the electron density for the GlcNAc-b1,2Man-a1,3Man-b1,4GlcNAc-b1,4GlcNAc-bAsn region similar to the A2-Asn complex along with density for the Man-a1,6-(M3) residue extending from the Man-b1,4-residue (M1). However, the additional terminal Man-a1,3-(M4) and Man-a1,6-(M5) residues in the NM5N2-Asn structure were disordered and not resolved in the structure (whited out region in dotted box for the inset structure and dotted oval in the Polder map (115).

FUT8 substrate recognition
complex indicates that the predicted steric hindrance (Fig. 7A) for the additional b1,6-GlcNAc residue (N6) altered the positioning of the entire branch.
The structure of the FUT8:GDP:NM5N2-Asn complex was determined at 3.2 Å resolution. Similar to FUT8:GDP:A3-Asn, this structure was also in space group P6 5 with eight chains in the asymmetric unit. For this complex, the electron density for both the GDP and the GlcNAc-b1,2-Man-a1,3-[Man-a1,6-]Man-b1,4-GlcNAc-b1,4-GlcNAc-Asn portion of the acceptor was visible in all eight chains, although the order of the GlcNAc-Asn region of the acceptor varied between chains (Fig. 7D and inset). The electron density for the M3 branch extending from the core b-Man (M1) is well-defined. However, density for the M4 and M5 residues was weak and they were left unmodeled.
The steric constraints of the acceptor-binding site clearly indicate that the 1,6-substituents on the M3 residue in the A3-Asn and NM5N2:Asn complexes are not compatible with the same binding geometry as the A2-Asn structure (GlcNAc-b1,6-(N6) for A3-Asn and Man-a1,6-(M5) for NM5N2-Asn). These modifications to the M3 residue lead to rotation or disorder of the more flexible Man-a1,6-linkage between M3 and M1 to accommodate the bulky branched substituents and likely account for the reduced catalytic efficiencies for these glycan acceptors.

Sequence and structural alignment of FUT8 with other related GTs
Previous sequence (94), HCA (95), and structural (42) alignments of fucosyltransferases have compared a-1,2and a-1,6fucosyltransferases with mammalian protein-O-fucosyltransferases (POFUTs) and identified three conserved sequence motifs that are common among the sequence families (Fig. S2, red boxes), including residues involved in binding the nucleotide portion of the sugar donor (42). Structures are now available for mouse (38), human (37), and Caenorhabditis elegans (33) POFUT1 (GT), C. elegans (39), and human (40) POFUT2 (GT68), Arabidopsis FUT1 (GT37 (29, 30)), NodZ (GT23, (35,96)), and human FUT8 (GT28) and structural alignment using the PROMALS3D server (97) allowed us to make a more comprehensive comparison between these enzymes to identify additional conserved structural elements. The resulting alignment revealed extensive structural similarity far beyond the three previously identified motifs (Fig. S2, blue boxes). Visualization of these conserved elements (Fig. S3A, Conserved) indicated the core of both Rossmann-folds were highly conserved among these enzymes including more than eight helical segments and four of the b-strands within the Rossmann-fold associated with the donor-binding site. Six b-strands of the second Rossmann-fold are less conserved in position, but still maintain a high degree of structural similarity among the collection of enzymes (Fig. S3, A and B). The degree of structural similarity is striking considering pairwise primary sequence similarities were ,15%. The PROMALS3D alignment also identified 13 amino acids that are most conserved among the structures, some of which were noted in a more focused comparative analysis of the sugar donor-binding site (42). Displaying these amino acids within the framework of the FUT8 structure shows that 9 of the 13 amino acids are clustered within the base of the GDP-binding site proximal to the Rossmann-fold (Fig. S3C) and either directly interact with the donor or contribute to the architecture of the donor site. The other four residues are distributed more widely within the FUT8 structure and do not interact with substrates or contribute to catalysis. For FUT8, the conserved binding site residues are not a part of Loops 1-3 that undergo conformational changes upon donor binding and, surprisingly, these three dynamic loops are not conserved in position or sequence among the other fucosyltransferase structures.
In contrast to the similarity in the underlying Rossmann-fold and donor-binding subsite, each of the enzymes are entirely different in their modes of interactions with their respective acceptor molecules (Fig. S3). Although the positions of acceptor interactions are approximately similar in the clefts between the two Rossmann-folds, each enzyme employs unique loop insertions into the Rossmann-fold scaffold to assemble the acceptor binding subsites (Fig. S3). For the POFUTs that recognize larger protein domain structures, the interactions are assembled through the generation of a deep cleft between the two subdomains. For the glycandirected fucosyltransferases (FUT8 and AtFUT1), the interactions are through a more flattened face of the cleft and, for FUT8, the recruitment of an additional SH3 domain insertion near the C terminus. This clearly indicates the presence of evolutionary divergence in the respective loop regions to provide unique acceptor binding specificities, whereas the core scaffold of the GT-B structure and donor interactions are more conserved.
An even more striking observation from the aligned structures is the similarities in positions for the acceptor nucleophiles in the respective structures (Fig. 8B). FUT8 and AtFUT1 (29,30) structures have been solved as extended glycan acceptor complexes, whereas POFUT1 (38) and POFUT2 (39) have been solved in complex with EGF and thrombospondin repeat domain acceptors, respectively. In each instance, the corresponding acceptor hydroxyl has been identified in a position for nucleophilic attack. Surprisingly, despite the differences in the respective acceptor structures, the nucleophilic hydroxyls are positioned similarly. The nucleophilic hydroxyls for the two glycan-directed enzymes (FUT8 and AtFUT1) superimpose almost exactly. Similarly, the nucleophilic Ser/ Thr hydroxyls for the POFUTs also superimpose with each other at a position ,2 Å from the nucleophiles for the glycandirected enzymes.

FUT8 catalytic mechanism
Among the six aligned fucosyltransferases in Fig. S3, only two have a clearly identified catalytic base (FUT8 and POFUT2 (39)). POFUT1 and AtFUT1 have no apparent ionizable side chain in proximity to the acceptor hydroxyl and are proposed to use a S N 1 mechanism employing a proton shuttle (20,30,33,38). NodZ was not solved as an acceptor complex and its catalytic base was not identified (35,96).
To identify the catalytic mechanism of FUT8 we superimposed the aligned structure of the intact GDP-Fuc from the POFUT2:GDP-Fuc complex (40) in the donor-binding site (Fig.  8A). This resulted in the placement of the Fuc residue in the FUT8 active site without steric hindrance and predicted Hbonds for the fucose O2 with the backbone nitrogen of Gly 219 and for O4 with Asp 368 (from Loop 2). This latter interaction is consistent with the loss of FUT8 activity for the D368A mutant (Table S3). In addition, C1 of the Fuc residue is directly in line for nucleophilic attack by the O6 hydroxyl of N1 (Fig. 8A). Thus, the structure of the FUT8:GDP:A2-Asn complex and the modeled donor confirm the inverting catalytic mechanism of the base-catalyzed S N 2 nucleophilic attack of the sugar nucleotide donor by the N1 O6 hydroxyl.
Although both FUT8 and POFUT2 perform similar S N 2 inline nucleophilic attacks of their respective GDP-Fuc donors, the catalytic base for POFUT2 (E52 (39)) originates from a completely different region of the protein structure compared with FUT8 (Fig. 8B). The catalytic base (Glu 373 ) for FUT8 is on the opposite side of its acceptor nucleophile relative to Glu 52 in POFUT2, whereas POFUT1 and AtFUT1 do not have ionizable side chains at either position. Clearly, whereas the enzymes conserve residues for donor and nucleophile binding, the constraints for positioning the catalytic base and enzymatic mechanism are not conserved among these enzymes.

Discussion
Prior studies on FUT8 examined the range of glycan substrates that can be recognized (69)(70)(71)(72)(73)(74)(75)(76)(77)(78) and how these modifications modulate a diverse array of biological functions (6, 8-10, 12, 14, 15, 17, 44-68, 98, 99). Our detailed kinetic analysis demonstrates the importance of the N-glycan GlcNAc-b1,2-Man-a1,3-Man arm as a key determinant for high affinity substrate recognition. Glycans that have not been modified by MGAT1  Fig. S3. The acceptor monosaccharide residues for the respective glycans (FUT8 (green sticks) and AtFUT1 (light blue sticks)) or protein domains (POFUT1 (purple sticks) and POFUT2 (gray sticks)) are shown as single monosaccharides or amino acids (full Ser side chain modeled for mouse Factor VII EGF1 domain from an mutant Ala residue in the mouse POFUT1:GDP-Fuc:EGF domain structure (38)). The respective hydroxyl nucleophiles are shown as magenta spheres. The donor analogs (GDP for all structures except POFUT1) are shown in white stick representation. The GDP-Fuc structure from the POFUT1:GDP-Fuc complex is shown in white stick representation except for the Fuc residue, which is shown as cyan sticks with C1 as a yellow sphere. The proposed catalytic base for FUT8 (Glu 373 in Loop 2) is shown in pink stick representation. Yellow dotted lines represent the deprotonation trajectory of the nucleophilic hydroxyl by Glu 373 and the subsequent S N 2 nucleophilic attack of the GDP-Fuc sugar donor as proposed in the FUT8 catalytic mechanism. The proposed catalytic base for POFUT2 (E52) is shown in white stick representation. Yellow dotted lines represent the deprotonation trajectory of the nucleophilic Ser hydroxyl by Glu 52 and the subsequent S N 2 nucleophilic attack of the GDP-Fuc sugar donor in the POFUT2 catalytic mechanism. The two catalytic bases interact from the opposite faces of their respective nucleophilic hydroxyls and emerge from completely different regions of the conserved core structures of their respective enzymes.
FUT8 substrate recognition (e.g. Man 5-3 N 2 -Asn) were ineffective substrates for human FUT8 even though recombinant expression data previously indicated these structures can be modified when mannose trimming is inhibited or FUT8 has been overexpressed (77,78,100). The A1-Asn and A2-Asn processing intermediates exhibited the highest catalytic efficiencies, while A3'-Asn, A3-Asn, G1, or incomplete processing to NM5N2-Asn significantly reduced catalytic efficiency, and extension to A4-Asn abolished activity (Fig. 7A). Thus, FUT8 exhibits a highly constrained active site that restricts steric access to all but a few structures despite the fact that all N-glycan processing intermediates contain the same core Man 3-GlcNAc 2 -Asn acceptor structure.

Structures of FUT8 donor complexes
To address the restricted substrate recognition, we first examined structures of FUT8 in complex with the donor analog, GDP. Loop regions were identified that were extended away from the donor-binding site when no GDP was bound and a progressive increase in GDP density was found as the loops flipped in to enclose the donor analog. Thus, catalysis proceeds by initial GDP-Fuc interaction with the donor-binding site predominately through hydrogen bonding between the nucleotide and the base of the donor-binding pocket. The highly flexible Loops 1-3 progressively become ordered while enclosing the sugar nucleotide and locking the donor in place through the Arg 365 -Asp 368 -Arg 441 cage. Binding of the glycan acceptor with its GlcNAc O6 hydroxyl positioned as nucleophile leads to deprotonation by Glu 373 , in-line S N 2 attack of the donor C1, and shuttling of the proton from Glu 373 through Lys 369 to the b-phosphate of the leaving group (Fig. 8A). The importance of each of these donor interactions was demonstrated through binding site mutants that all individually led to enzyme inactivation.

Structures of FUT8 acceptor complexes
The structures of several high and low affinity FUT8 acceptor complexes also allowed mapping of steric contributions to substrate binding affinity and specificity. Initial structural studies on the FUT8:GDP:A2-Asn complex identified analogous interactions found in prior structural studies on equivalent complexes (41,42). The Man-b1,4GlcNAc-b1,4-bAsn acceptor core bound within the cleft between the two Rossmann-folds, whereas the C-terminal SH3 domain presented a flattened surface and steric barrier for branched substrate interactions that established acceptor glycan specificity. Our kinetic analysis of site directed mutants also probed the roles of individual amino acids in substrate binding and catalysis to validate the structural model.
The studies were then extended by examining additional acceptors complexes with reduced binding affinities. The A3'-Asn structure extended the additional b1,4-GlcNAc into solvent relative to the A2-Asn complex (Fig. 7B). A similar solvent-exposed b1,4-Gal extension was predicted for the G1 structure based on comparison with the A2-Asn complex. Both acceptor modifications reduce catalytic efficiencies by 3-4-fold likely as a result of entropic penalties from the solvent-exposed sugar rather than steric restrictions.
In contrast, modifications to the M3 residue, especially at the O6 position, led to a steric clash with the enzyme surface (Fig.  7A). As a result, both the NM5N2-Asn (a1,6-Man on M3) and A3-Asn (b1,6-GlcNAc on M3) had significant disorder for residues on the M3 branch, whereas the remainder of the acceptor structure was well resolved. This disorder is consistent with reduced catalytic efficiencies for these acceptors (8.4-and 20fold reduction in k cat /K m , respectively) and a displacement of the M3 residue as a consequence of the O6 modification. NM5N2-Asn had slightly better catalytic efficiency than A3-Asn, consistent with the well-defined electron density for the M3 residue in the crystal structure and the reduced size of the Man-a1,6substituent versus GlcNAc-b1,6-for A3-Asn. These comparisons of catalytic efficiencies and acceptor complex structures highlight the role of the SH3 domain as a critical binding interface for tuning branch-specific acceptor affinity through both complementarity interactions and steric restrictions.

Conserved and distinctive structural elements for substrate recognition among FUTs
The FUT8 structure also allowed a comparison with other known GT-B fold fucosyltransferases to identify conserved and divergent structural features. These alignments revealed a far more extensive similarity within the core Rossmann-folds and previously noted donor-binding site residues (42) and other regions identified in sequence alignments (95,101). Surprisingly, no conservation in position or sequence for the three dynamic loops that enclose the FUT8 donor-binding site were identified indicating that the mechanism of induced-fit donor binding was not preserved among the other fucosyltransferases.
Despite the similarities in GT-B fold structure and sugar donor-binding interactions, no structural similarity was observed for the loops involved in acceptor recognition (Fig. S3). POFUT1 and POFUT2 employ deep clefts between the two Rossmann-folds for recognition of their respective protein substrate domains, but use completely different sets of loop interactions (38,39). AtFUT1 (30) and FUT8 employ a more flattened face of the enzyme surface for acceptor interactions, but also use completely different loop structures and an additional SH3 domain for FUT8.

Catalytic mechanism and evolution of modular substrate recognition by glycosyltransferases
The use of unique acceptor-binding loop insertions in GT-B fold fucosyltransferases is highly reminiscent of hypervariable substrate-binding loops in the evolution of GT-A fold glycosyltransferases (32,102,103). In contrast to the two Rossmann domains in GT-B fold enzymes, GT-A fold glycosyltransferases have a single, conserved Rossmann-fold core and employ proximal residues for donor interactions, whereas more divergent extended loop insertions are free to rapidly evolve and diversify novel acceptor specificities (32,102,103). This rapid evolution of the loop insertions is constrained by the necessity of positioning the acceptor nucleophile for attack and access to a catalytic base for nucleophile activation relative to the C1 of the FUT8 substrate recognition sugar donor. GT-A fold inverting glycosyltransferases have highly conserved positions for the catalytic base, hydroxyl nucleophile, and generally use a conserved DXD motif for metal binding and sugar nucleotide positioning (32). For GT-A fold retaining enzymes the position for the hydroxyl nucleophile is shifted to no longer use of the enzyme-associated catalytic base, and instead a donor substrate-assisted S N 1 catalytic mechanism is employed (32).
The present studies on the GT-B fold fucosyltransferases indicate an analogous use of Rossmann-fold core structures, conserved proximal residues for donor interactions, and highly divergent loop regions for distinctive acceptor recognition. Surprisingly, whereas the fucosyltransferases all have inverting catalytic mechanisms, their strategies for use of a catalytic base are quite distinct. AtFUT1 has no discernable catalytic base and water-mediated proton shuttle mechanism has been invoked (30). For POFUT1 the donor b-phosphate oxygen has been proposed as the catalytic base with the possible involvement of a water-mediated proton shuttle (20,33,38). FUT8 and POFUT2 (40) both have clearly identified catalytic bases, but they originate from different regions of their protein structures and act from opposite faces in deprotonation of their respective nucleophilic hydroxyls. Thus, despite the highly similar core structures, positions of donor sugars and acceptor nucleophiles among the fucosyltransferases, their respective catalytic base mechanisms are entirely different.
Although a general understanding of GT-A fold glycosyltransferase evolution is emerging (32,102,103), a parallel understanding of GT-B fold enzymes has lagged behind largely because of their more complex domain architectures and fewer determined protein structures in complex with substrate analogs. However, evolutionary parallels between the two proteinfold classes are now evident and further insights into selective substrate recognition among the broader collection of GT-B glycosyltransferases will expand our understanding of how diverse glycan structures are elaborated in biological systems.

Experimental procedures
Cloning, expression, fusion tag cleavage, and purification of FUT8 in HEK293F cells The expression construct encoding the catalytic domain of human FUT8 (a1,6-fucosyltransferase, Uniprot Q9BYC5, residues 41-575) was generated by amplification of a Mammalian Gene Collection clone and transferring it into a pDONR221 vector through Gateway recombination (83). The PCR product, containing the Gateway att1 recombination sites, was subcloned into a pGEN2-DEST mammalian expression vector that employs a cytomegalovirus promoter and encodes an NH 2 -terminal sequence followed by an His 8 tag, an AviTag recognition site, "superfolder" GFP, the 7-amino acid recognition sequence of TEV protease (83), and the coding region of human FUT8. The protein was expressed through transient transfection in either WT HEK293F (Freestyle 293F, ThermoFisher Scientific) cells or mutant HEK293S (GnTl 2 ) cells (ATCC, catalog number CRL-3022). Briefly, cells were maintained in suspension culture at 1-3 3 10 6 cells/ml in a humidified CO 2 shaker incubator (37°C, 150 rpm). Transient transfection was performed at a cell density of 3-3.5 3 10 6 cells/ml in media containing 9 parts of Freestyle TM 293 Expression Medium (Life Technologies) and 1 part of Ex-Cell 293 serum-free medium (Sigma) using polyethyleneimine (linear 25 kDa PEI, Polysciences, Inc., PA) at a concentration of 5 mg/ml and a plasmid DNA concentration of 4 mg/ml. At 24 h post-transfection the suspension cultures were diluted 1:1 with culture media containing valproic acid (2.2 mM final concentration, Sigma). Cultures containing the secreted GFP-tagged protein were harvested 6 days post-transfection. For evaluation of enzyme kinetics, alanine mutants were generated using Q5 site-directed mutagenesis kit (New England Biolabs) and sequences were confirmed for the desired mutations. Both WT and mutant forms were expressed in the pGEN2-DEST vector and expressed as a soluble secreted form.
The WT and mutant forms of the protein were purified as described previously (Table S4) (83,86). The harvested supernatant was filtered through 0.8-mm filter membrane to remove particulate matter and then loaded onto nickel-nitrilotriacetic acid resin (Qiagen) pre-equilibrated with 20 mM HEPES, pH 7.4, 300 mM NaCl, 20 mM imidazole, and 5% glycerol. After washing steps, the GFP-FUT8 fusion protein was eluted using 300 mM imidazole and further pooled and concentrated for determination of enzyme kinetics. Removal of the GFP tag was performed by treating the concentrated protein with TEV protease at a ratio of 1:5 relative to the fusion protein and incubating the sample for 16 h at 4°C. The sample was then loaded to a nickel-nitrilotriacetic acid resin to remove the cleaved GFP and His-tagged TEV protease leading to a purified untagged protein in the flow-through fraction. For structural studies, the protein was further purified using a Superdex 75 column (GE Healthcare) pre-equilibrated with 20 mM HEPES, pH 7.4, 150 mM NaCl, and 60 mM imidazole and the peak fractions were collected.

SEC-MALS
The oligomerization of FUT8 was analyzed by injecting the enzyme (20 ml, 1 mg/ml) on an analytical grade Superdex 75 column pre-equilibrated with 25 mM HEPES, pH 7.4, 150 mM NaCl, 0.02% NaN 3 . Light scattering and deferential refractive index were measured using a MiniDAWN TREOS detector (Wyatt Tech.) and Optilab rEx detector (Wyatt Tech.) respectively. Data were analyzed using the ASTRA 6.0 software (Wyatt Tech.)

Alkaline phosphatase treatment of GDP-fucose
For enzyme kinetics, GDP-fucose (Carbosynth) was used as donor at a final concentration of 0.2 mM. Prior to performing the assay, GDP-fucose was treated with calf intestinal alkaline phosphatase (CIAP, Promega) to remove free GDP. Briefly, a reaction mix of 3 units/ml of CIAP/50 ml of 50 mM GDP-fucose in 13 CIAP buffer (5 ml) was prepared in a microcentrifuge tube. Multiple doses of CIAP (3 units/ml) were added every 2 h and incubated for a total of 6 h at 37°C with gentle shaking (;100-150 rpm). After 6 h, the concentration of GDP-fucose in the reaction volume was adjusted to 20 mM. The entire reaction mix was filtered through a Microcon 10-kDa centrifugal filter unit (Millipore) with Ultracel-10 membrane by centrifuging at 14,000 3 g for 30 min to remove CIAP. The filtrate was recovered and incubated with another 3 units of CIAP for 16 h at 37°C with gentle shaking (;100-150 rpm). Finally, the reaction mix was further filtered through a Microcon 10-kDa centrifugal filter unit as mentioned above. The filtrate was stored at 220°C and used as the stock solution of GDP-fucose. The CIAP-treated GDP-fucose gave lower background and improved signal to noise ratio in enzyme kinetic assays compared with untreated GDP-fucose.

FUT8 kinetic analysis
Enzyme kinetics for the WT and mutant form of the enzyme were determined using the GDP-Glo TM Glycosyltransferase assay (Promega) that detects GDP as a released product of the glycosyltransferase reaction. Assays were performed in a 10-ml reaction volume consisting of a universal buffer (200 mM each of Tris, MES, MOPS, pH 7.5), 0.2 mM GDP-fucose (CIAPtreated, see above) in reactions containing varied concentrations of the respective glycan substrates: A1-Asn, A2-Asn, A3-Asn, A3'-Asn, A4-Asn, G1-Asn, NM5N2-Asn, M3N2-Asn, and M5N2-Asn-Fmoc. Reactions were carried out using purified GFP-fusion enzyme (WT or alanine mutants) at 37°C for 30 min. For acceptor titration, assays were performed using the desired substrate at a concentration range of 0-1 mM with 0.2 mM GDP-fucose as final concentration of donor. Similarly, for donor titration, reactions were carried out using GDP-fucose (0-0.2 mM) with 0.5-1 mM A2-Asn as final concentration. Reactions were stopped using 5 ml of GDP-detection reagent at a 1:1 ratio in a white, opaque 384-well-plate and incubated in dark for 1 h at room temperature. The released GDP product was detected based on luminescence using a GloMax Multidetection plate reader (Promega). The steady state parameters of K m , k cat , and V max values were calculated using a GDP standard curve and nonlinear curve fitting in GraphPad Prism 6 software.

Crystallization and data collection
The catalytic domain of FUT8 (10 mg/ml in a storage buffer containing 20 mM HEPES, pH 7.5, 50 mM NaCl, 300 mM betaine) was mixed with GDP or with GDP plus acceptor and screened for crystallization conditions using a TTP Labtech Mosquito Crystal robot and optimized using hanging drop vapor diffusion with 2-ml drops (1:1 protein:reservoir ratio). For the FUT8:GDP complex, crystals containing 5 mM GDP grew in less than 1 week from a reservoir of 1.4 M ammonium sulfate, 0.6 M L-proline, 0.1 M HEPES, pH 7.5. Crystals were transferred to the reservoir solution supplemented with 5 mM GDP, and 20% cryoprotectant (1:1:1 ethylene glycol, DMSO, and glycerol). FUT8:GDP crystallized in space group P6 5 and diffracted to 2.25 Å (Table S2).
For the FUT8:GDP:A3'-Asn complex, 10 mg/ml of protein was prepared in storage buffer containing 10 mM GDP and 10 mM A3'-Asn and crystals grew from a reservoir solution of 1.2 M lithium sulfate, 12 mM nickel chloride, and 0.1 M Tris, pH 8.5. Crystals were transferred to the reservoir solution supplemented with 10 mM GDP, 10 mM A3'-Asn. A 40:60 mixture of paratone and paraffin was used as the cryoprotectant. FUT8: GDP:A3'-Asn crystallized in space group P6 5 22 and diffracted to 3.3 Å (Table S2).
For the FUT8:GDP:A3-Asn complex, 10 mg/ml of protein was prepared in storage buffer containing 10 mM GDP and 10 mM A3-Asn and crystals grew overnight from a reservoir solution of 0.2 M L-proline, 10% PEG3350, and 0.1 M HEPES, pH 7.2.
All cryo-protected crystals were flash cooled in liquid nitrogen and X-ray data were collected at the SER-CAT 22-BM beamline using a MAR-CCD detector (FUT8:GDP and FUT8: GDP:A2-Asn) or at the 22-ID beamline using an Eiger-16M detector (FUT8:GDP:A3-Asn, FUT8:GDP:A3'-Asn, and FUT8: GDP:NM5N2-Asn) at the Argonne National Laboratory. The data were processed using XDS (109). Five percent of the data were set aside for cross-validation.

Phasing and refinement
The crystal structures of FUT8:GDP and FUT8:GDP:A2-Asn were solved using molecular replacement with the apo-FUT8 structure (PDB 2DE0 (36)) as the search model. The NH 2 -terminus (residues 41-107) and the C-terminal residue Lys 575 were disordered and left unmodeled. Successive rounds of automated refinement in Phenix (110) and iterative manual fitting using Coot (111) produced the final models (Table S2). The B-factors were refined using TLS (112).
During structural refinement, geometrical restraints were used to restrain monosaccharides with weak electron density in the lowest energy chair conformation ( 4 C 1 ) (110,113). The linkage torsion angles for the glycosidic bonds (f, c) in FUT8: A2-Asn, FUT8:A3'-Asn, FUT8:A3-Asn, and FUT8:NM5N2-Asn complexes were determined using the Carbohydrate Ramachandran Plot (CaRP) and fall into the energetically preferred areas of the GlyTorsion plot (114).

Structural analysis
Structure based sequence alignment was performed using default parameters on the PROMALS3D server (97) using structures of the human FUT8:GDP:A2-Asn complex, mouse POFUT1:GDP-Fuc:EGF domain complex (PDB 5KY3 (38)), the C. elegans POFUT2:GDP:thrombospondin repeat complex (PDB 5FOE (39)), the human POFUT2:GDP-Fuc complex (PDB 4AP6 (40)), the NodZ:GDP complex (PDB 3SIX (35)), and the AtFUT1:GDP:xylo-oligosaccahride complex (PDB 5KOR (29)). Alignment of the structural models was performed in Coot (111) followed by manual adjustment in PyMOL and displayed using PyMOL (Schrödinger). Structural elements identified in the PROMALS3D analysis were extracted from the full protein alignment for display as an overlay of the respective structures. The difference density maps (F o 2 F c ) shown in the figures for the bound GDP and the A2 acceptor were calculated subsequent to structure solution and initial restrained refinement of the polypeptide, but prior to the modeling of the respective ligands. In all cases, the final refined coordinates were used to depict the ligand model. Polder maps (115) for the A3'-Asn, A3-Asn, and NM5N2-Asn acceptors were calculated following a modified procedure to reduce model bias (103). Briefly, prior to calculating the Polder map, the acceptor was omitted and the remaining model subjected to Cartesian simulated annealing at 5000 K. The acceptor coordinates were then used to calculate a reduced-bias Polder map.