Emerging structural insights into glycosyltransferase-mediated synthesis of glycans

Abstract

Glycans linked to proteins and lipids play key roles in biology; thus, accurate replication of cellular glycans is crucial for maintaining function following cell division. The fact that glycans are not copied from genomic templates suggests that fidelity is provided by the catalytic templates of glycosyltransferases that accurately add sugars to specific locations on growing oligosaccharides. To form new glycosidic bonds, glycosyltransferases bind acceptor substrates and orient a specific hydroxyl group, frequently one of many, for attack of the donor sugar anomeric carbon. Several recent crystal structures of glycosyltransferases with bound acceptor substrates reveal that these enzymes have common core structures that function as scaffolds upon which variable loops are inserted to confer substrate specificity and correctly orient the nucleophilic hydroxyl group. The varied approaches for acceptor binding site assembly suggest an ongoing evolution of these loop regions provides templates for assembly of the diverse glycan structures observed in biology.

Animal cell surfaces are covered with thousands of distinct glycan structures attached to proteins and lipids¹. These glycans participate in numerous biological phenomena including modulating cellular communication, mediating immune functions, targeting proteins to specific subcellular organelles, assisting in protein folding, altering recycling of receptors, and protecting proteins from degradation^2,3. Cells in diverse tissues, or in different stages of development, have distinct glycan structures on their surfaces, leading to differential functional consequences in those contexts. When a cell divides, these glycan structures are replicated; however, because glycans are not directly synthesized from a genome-encoded template, as proteins and nucleic acids are, the fidelity is dependent on the specificity of the glycosyltransferases (GTs) that synthesize complex glycans by adding one sugar at a time. Most eukaryotic GTs are localized to the lumen of the secretory pathway and modify proteins and/or lipids as they transit to the cell surface or extracellular spaces⁴. Most GTs catalyze sugar transfer from a donor molecule (nucleotide sugar or lipid-linked sugar) to the hydroxyl group of an acceptor molecule⁵. For GTs that employ nucleoside diphosphate (NDP) sugar donors, the generalized reaction can be depicted as:

$$\text{NDP}-\text{sugar}+\text{acceptor}-\text{OH}\to \text{NDP}+\text{acceptor}-\text{O}-\text{sugar}$$

We have learned a great deal about how GTs bind donor nucleotide sugars, but specificities of GTs reside in their ability to selectively modify the correct hydroxyl group on an acceptor containing many equally reactive hydroxyls (for example, complex oligosaccharides). Here we review recent structures of ternary complexes of eukaryotic GTs including donors, donor analogs, and acceptors, providing details of how these enzymes modify the correct hydroxyl group and provide fidelity to accurately replicate the glycan structures on cell surfaces.

The Carbohydrate-Active Enzymes (CAZy) database (http://www.cazy.org) annotates nearly 500,000 GT sequences across all kingdoms of life and organizes them into 106 different families⁶, and the human genome encodes ~210 GTs⁴ in 45 CAZy families. The first three-dimensional structure of a GT was that of rabbit glycogen phosphorylase, deposited in 1990⁷. The fact that most eukaryotic GTs in the secretory pathway require disulfide bonding, N-glycosylation, and eukaryotic chaperones for proper folding makes production of sufficient amounts of these enzymes for structural analysis a challenge^4,8. This limitation has been a major barrier in the field, but recent availability of eukaryotic expression systems (insect cells; HEK293 cells⁴) has allowed for elucidation of more structures of eukaryotic enzymes in the past decade. As a result, since 2000 the number of independent GT structures has grown steadily to ~209 from 53 different GT families (56% bacterial, 38% eukaryotic, 4% archaeal, 1% viral; Supplementary Fig. 1; Supplementary Dataset 1).

Although there are many GT families, structural analysis has shown that their catalytic domains fall into a small number of protein folds: GT-A, GT-B, and GT-C (Table 1)^5,9. GT-A enzymes contain a single Rossmann-like fold, a domain found in enzymes that bind nucleotides. Most, but not all, GT-A fold enzymes require divalent cations for activity and contain a ‘DxD’ motif that is involved in coordinating the metal and nucleotide sugar. GT-B fold enzymes contain two Rossmann-like folded domains, but do not require divalent cations for activity or contain DxD motifs. GT-C fold enzymes, such as oligosaccharyltransferases, contain multiple transmembrane α-helices and employ lipid-linked sugar donors. For the sake of brevity, we will not discuss GT-C enzymes, CAZy GT1 family enzymes that act on small-molecule metabolites, or the nuclear/cytoplasmic O-GlcNAc transferase, which have all been recently reviewed^10–12, but will focus on recent eukaryotic GT-A and GT-B fold structures that synthesize cell surface and secreted glycoproteins.

Table 1 |.

Characteristics of GT-A, GT-B, and GT-C GTs

The GT structures in the PDB that are listed in Supplementary Dataset 1 were divided by fold type, catalytic mechanism, and metal dependency for sugar donor interaction and listed by CAZy family along with a cartoon protein structure display for a representative member of each fold family (GT-A, MGAT2 (PDB: 5VCM and 5VCS³¹); GT-B, POFUT1 (PBD: 5KY0²⁰); GT-C, STT3 (PDB: 6C26⁸⁶)). In one case (GT6, asterisk) there is one bacterial family member (BACOVA_01932⁸⁹) that is a metal-independent retaining enzyme, whereas other eukaryotic GT6 family members (GGTA1²⁵ and ABO⁴⁷) are metal dependent.

Mechanistic strategies

Nucleotide-sugar-dependent GTs use a restricted set of donor molecules to make the diverse collection of glycan structures in cellular systems and generate enzymatic products with distinct anomeric configurations. Each of the GT families can either invert or retain the stereochemistry of the anomeric linkage of the donor sugar during transfer, often to discrete hydroxyl groups on the acceptor structure^5,9. For example, β-1,4-galactosyltransferase 1 (B4GALT1) is an inverting enzyme. It transfers galactose (Gal) from UDP-α-Gal to the 4′-hydroxyl of N-acetylglucosamine (GlcNAc) to form a Gal-β-1,4-GlcNAc product (Fig. 1a)^13–15. Most inverting enzymes utilize an S_N2 single-displacement reaction mechanism (Fig. 1b) in which the acceptor nucleophilic hydroxyl attacks the donor sugar anomeric carbon and displaces the leaving group nucleotide moiety from the opposite face. The result is an inversion of the anomeric configuration of the product^5,9. In most instances the enzyme-bound sugar nucleotide donor assumes a ‘folded back’ conformation during sugar transfer, in contrast to the more extended conformation of the donor in solution^16–18. Significant conformational change in loop regions of the enzyme may also occur upon binding the sugar donor¹³. These enzymes employ a catalytic base (typically Glu, Asp, or His), which assists by deprotonating the nucleophile. A few inverting enzymes without an identifiable catalytic base have also recently been identified (POFUT1^19,20 and FUT1²¹). In both cases an S_N1 reaction mechanism was proposed in which the β-phosphate oxygen atom of the GDP-Fuc donor assists in the deprotonation of the acceptor hydroxyl group (Fig. 1c)^19–21. An active site water molecule may promote the reaction by providing a proton relay between the hydroxyl and the β-phosphate oxygen^20,21.

Fig. 1 |. GT reaction mechanisms.

a, B4GALT1 reaction. b, S_N2 inverting mechanism. c, S_N1 inverting mechanism. d, Double-displacement retaining mechanism. e, S_Ni-like same-side retaining mechanism. All mechanisms in b-e are drawn using a generic hexose in α-linkage to a nucleoside diphosphate. The β-phosphate is drawn out when it participates in the reaction (panels c and e). Enz-B and Enz-BH (purple) indicate the catalytic base that deprotonates the acceptor nucleophile. NMP and NDP are nucleoside monophosphate and nucleoside diphosphate, respectively. Red O indicates nucleophilic oxygen of acceptor substrate. Green C indicates anomeric carbon of donor sugar. R indicates acceptor substrate. The linkage of anomeric carbon in the hexose is indicated in both substrates and products.

The reaction mechanism for retaining GTs has been debated for many years, although several recent structures have provided significant insights^22–25. Similarly to retaining glycosidases that utilize a double-displacement mechanism with a transient covalent enzyme–substrate intermediate^5,9, retaining GTs were also originally thought to employ a covalent intermediate (Fig. 1d)⁵. However, no convincing evidence of a covalent intermediate for a native retaining GT has been obtained, and in most cases there is not an appropriate enzyme nucleophile in proximity to accept the intermediate or a catalytic base to deprotonate the acceptor nucleophile. An alternative mechanism for a same-side S_Ni-type reaction has recently been proposed in which the acceptor hydroxyl nucleophile is deprotonated by the donor β-phosphate oxygen and attacks the anomeric carbon atom of the donor sugar from the same side as the leaving nucleotide, resulting in retention of anomeric configuration (Fig. 1e). Several structural snapshots of enzyme–donor–acceptor Michaelis complexes for retaining GTs have recently appeared (XXYLT²², GALNT2²³, GGTA1²⁵, and GpgS²⁴). The position of the nucleophilic acceptor hydroxyl group relative to the anomeric carbon atom, together with snapshots showing progression of the reaction, strongly support the S_Ni-type mechanism.

The contrast of inverting and retaining catalytic mechanisms is illustrated by comparison of two GT-A fold enzymes, B4GALT7²⁶ and GGTA^24,25 (Fig. 2a,c). Both enzymes employ GT-A folds and metal-dependent interactions with their UDP-Gal donors. B4GALT7, an inverting GT, employs an Asp side chain on a conserved α-helix to deprotonate the Xyl C4 nucleophilic hydroxyl group of the acceptor glycan during its S_N2 in-line nucleophilic attack of the donor sugar (Figs. 1 and 2a). By comparison, GGTA1, a retaining GT, maintains a Glu residue at the position equivalent to that of the B4GALT7 catalytic base but shifts the position of the incoming acceptor Gal C3 hydroxyl nucleophile by >2Å away from the respective amino acid side chain, precluding its use as a catalytic base (Fig. 2c). This shift in the position for the acceptor nucleophile has two consequences for the catalytic mechanism. First, the catalytic base function is now provided by the β-phosphate oxygen of the sugar donor. Second, the angle of attack of the nucleophilic hydroxyl on the donor C1 atom is changed such that an in-line, direct displacement reaction is no longer possible (Fig. 2b). In its place, a same-side dissociative S_Ni-type mechanism has been proposed (Figs. 1 and 2c) to account for the retained anomeric configuration of the product, the altered geometry of nucleophilic attack, and the distinct mechanism for nucleophile deprotonation^24,25. Thus, positioning of the acceptor hydroxyl group relative to the donor anomeric carbon is critical in establishing the enzyme catalytic mechanism. These mechanisms have been reviewed in more detail elsewhere^25,27–30.

Fig. 2 |. Comparison of representative inverting and retaining catalytic mechanisms and enzyme topologies for substrate interactions.

Bound ligands for B4GALT7 (a, PDB: 4M4K²⁶) and GGTA1 (c, PDB: 5NRD²⁵) are displayed along with putative catalytic bases. a, The inverting S_N2 catalytic mechanism for B4GALT7 is catalyzed by deprotonation of the Xyl acceptor hydroxyl (red sphere) by the catalytic base (Asp211 in the wild type enzyme mutated to Asn211 in PDB ID 4M4K) and attack of the anomeric carbon (green sphere) in line with the departing nucleotide leaving group. c, GGTA1 has a retaining S_Ni-like mechanism in which the nucleophile position (red sphere) is shifted relative to the inverting enzymes. The equivalent catalytic base residue (Glu317) is too far away for deprotonation, but the acceptor Gal hydroxyl (red sphere) is positioned adjacent to the β-phosphate oxygen for deprotonation (gray dotted line) and in proximity to attack the anomeric carbon (green sphere). b, Overlay of the two structures based on the Gal C1-phosphate oxygen bond illustrates the differences in geometry of nucleophilic attack on the anomeric carbons. d,e, Donor and acceptor binding sites for B4GALT7 (d) and GGTA1 (e) are displayed as topology diagrams illustrating the positions of residues that interact with donors and acceptors in loops extending from the core β-sheets of the Rossmann fold. β-Sheets are indicated as filled arrows (orange for conserved Rossmann-fold elements and yellow for nonconserved sheets among GT-A fold enzymes) and helices as red boxes connected by loop regions (green lines). Positions of interacting residues are indicated by colored lines with asterisks (gray, interacting with donors; blue, interacting with acceptors; magenta, DxD motif). Acceptors for each enzyme (B4GALT7: Xyl-β-1,4-Xyl and GGTA1: Gal-β-1,4-Glc) are indicated in abbreviated form, but with inverted orientation from standard nomenclature to indicate the proximity of the terminal sugar acting as nucleophile in the glycosyltransfer reaction.

Assembly of active site donor and acceptor modules

The recent structures of a growing number of eukaryotic GTs containing bound donors (or donor analogs) and/or acceptor structures (Supplementary Dataset 1) have also provided insights into how these enzymes can catalyze selective glycan synthesis. These structures demonstrated that GT active sites are assembled as modular components extending from the stable Rossmann-fold scaffolds that bind donor and acceptor molecules with precise geometry and reactivity (Fig. 2d,e). GT-A and GT-B fold enzymes both employ a parallel β-sheet Rossmann-fold structure to anchor loops that bind the nucleotide portion of the donor substrate (for example, the inverting enzyme B4GALT7 complex with UDP-Gal²⁶, Figs. 2d and 3; the retaining enzyme, GGTA1, Figs. 2e and 5). Direct interactions with the nucleotide are surprisingly variable among the collection of GTs³¹, yet each is capable of providing complementary specificity for the respective nucleotide base and donor sugar (Figs. 2–5). The GT-A fold enzymes that employ divalent cations use a carboxylate residue of the DxD motif (and sometimes His residues²⁶) to coordinate the metal, which interacts with the nucleotide phosphodiester of the donor (for example, MGAT1 interaction with UDP-GlcNAc³², Fig. 3b), whereas metal-independent GT-A fold enzymes generally employ Lys and Arg side chains for phosphodiester interactions (for example, GCNT1 interaction with UDP^33,34, Fig. 3b). GT-B fold enzymes employ one of their two Rossmann folds for metal-independent sugar nucleotide interactions using a similar set of Lys and Arg side chains (for example, POFUT1 interaction with GDP-Fuc²⁰), similarly to the metal-independent GT-A fold enzymes (Table 1).

Fig. 3 |. Structural gallery of eukaryotic GT-A fold inverting enzymes.

a, Eight eukaryotic GT-A fold inverting enzyme structures were aligned using Coot⁸⁸ (transparent surface) with bound donor analogs (white sticks), acceptor analogs (yellow sticks), divalent cations (spheres), and catalytic bases (sticks). Structures include B4GALT1 (PDB: 1TW5¹³), B4GALT7 (PDB: 4M4K²⁶), POMGNT1 (PDB: 5GGI⁴⁴), MGAT2 (overlay of PDB IDs 5VCM and 5VCS³¹), MGAT1 (PDB: 1FOA³²), B3GAT1 (PDB: 1V84⁴⁵), B3GAT3 (PDB: 1FGG⁴⁶), and GCNT1 (overlay of PDB IDs 2GAM and 3OTK^33,34). Boxes represent the regions of the respective structures shown in b. b, Enlarged views of the bound ligands are displayed in the same orientation and coloring as in a, with individual molecule components labeled. For GCNT1, the Lys and Arg residues that interact with the phosphodiester are indicated (sticks). c, Bound ligand components for enzymes that employ Asp as the catalytic base, shown in two different rotations. e, Bound ligand components for enzymes that employ Glu as the catalytic base, shown in two different rotations. d, Overlay of two representative enzymes (B4GALT7 employing an Asp catalytic base and B3GAT3 employing a Glu catalytic base) illustrating the differences in position for the nucleophiles and sugar donors as a consequence of the differences of length of the catalytic base side chains. Acceptors, donors (C1 as green sphere) metal ions and the nucleophilic hydroxyl (red sphere) are labeled, and the positions of nucleophilic attack (green dotted line) and catalytic base (black dotted line) action are indicated.

Fig. 5 |. Structural gallery of eukaryotic GT-A fold retaining enzymes.

a, Five eukaryotic GT-A fold retaining enzyme structures were aligned using Coot⁸⁸ (transparent surface) with bound donor analogs (white sticks), acceptor analogs (yellow sticks), and divalent cations (spheres). Structures are ABO (PDB: 2RJ7⁴⁷), GGTA1 (PDB: 5NRD²⁵), XXYLT1 (PDB: 4WNH²²), XXT1 (PDB: 6BSW⁴⁸) and EXTL2 (overlay of PDB IDs 1ON8 and 1OMZ⁴⁹). Boxes represent the regions of the respective structures shown in b. b, Enlarged views of the bound ligands are displayed in the same orientation and coloring as in a, with donors, acceptors, metal ions, catalytic bases labeled as in Fig. 3. c, Overlay of the aligned ligand components, shown in two different rotations.

Although GT-A and GT-B fold enzymes both use analogous approaches for nucleotide sugar donor interactions, the enzymes differ in their interactions with extended acceptor structures. GT-A fold enzymes employ extensions and insertions into their single Rossmann core fold to assemble acceptor binding site modules³¹. The diversity in positions and structures of the loop insertions among the GT-A fold enzymes is quite striking (Fig. 2d,e), even among enzymes in the same CAZy family (for example, GT29 NeuAc transferases^4,35–38 and GT14 xylosyl and GlcNAc transferases^33,39), and unique complementarities for each of the respective acceptors are achieved using these loop extensions (see below).

For GT-B fold enzymes, the two Rossmann folds are linked in tandem with donor (A domain) and acceptor (B domain) binding subsites facing into the cleft between the two domains (Fig. 7)⁵. The acceptor binding subsites are assembled largely from loops extending from the B domain into the cleft (for example, MGAT5⁴⁰ and FUT1^21,41), but there are several examples of GTs that employ residues from both domains for acceptor interactions (POFUT1²⁰, POFUT2⁴², and POGLUT1⁴³). In addition, the geometry of the cleft can vary significantly among the GT-B fold enzymes to accommodate extended, branched glycan structures (for example, MGAT5⁴⁰) and FUT1^21,41)) or globular protein domain structures (POFUT1²⁰, POFUT2⁴², and POGLUT1⁴³, Fig. 7).

Fig. 7 |. Domain specific GTs (POFUT1, POFUT2, and POGLUT1) with substrates.

a, POFUT1 (overlay of PDB IDs 5KXH and 5KY3²⁰). b, POGLUT1 (overlay of PDB IDs 5L0R and 5L0U⁴³). c, POFUT2 (PDB: 5FOE⁴²). Upper panels: all three GT-B fold enzymes are displayed in transparent surface representations with A domains in light green and B domains in darker green. Folded domains (EGF repeats for POFUT1 and POGLUT2; TSR for POFUT2) are in cyan cartoon with respective consensus sequences in yellow. Acceptor amino acid is shown in red sticks, with the acceptor oxygen as a red sphere. Donor nucleotides, white sticks. Note that the POGLUT1 structure has the donor analog UDP-CH₂-Glc. Lower panels: enlarged views of the bound ligands are displayed in the same orientation and coloring as the upper panels, with individual molecule components labeled. Boxes in the upper panels represent the regions of the respective structures shown in the lower panels.

The overall conclusions from these diverse substrate-bound structures are that GTs have a conserved and restricted positioning for the nucleotide sugar in the donor binding subsite based on the constraints of interacting loops that extend from the Rossmann fold (Fig. 2d,e). By contrast, the striking diversity in binding geometries for glycan, peptide, lipid or metabolite acceptors shows a common theme for precise positioning of the nucleophilic hydroxyl group appropriately for catalysis leading to either inversion (S_N2- or S_N1-type mechanisms) or retention (S_Ni-type mechanism) in anomeric configuration upon sugar transfer (Figs. 1 and 2a–c)³¹.

Structures of eukaryotic secretory-pathway GTs containing bound acceptors now total 18 enzymes for glycan-directed GTs and 7 structures for peptide-directed GTs (Supplementary Dataset 1). Glycan-directed eukaryotic enzymes include the metal-dependent inverting GT-A fold enzymes B4GALT1¹³, B4GALT7²⁶ (GT7), POMGNT1⁴⁴ (GT13), MGAT2³¹ (GT16), B3GAT1⁴⁵, and B3GAT3⁴⁶ (GT43) (Fig. 3). Metal-independent GT-A fold inverting enzymes include GCNT1³⁴ (GT14) (Fig. 3), ST3GAL1³⁷, ST6GAL1³⁵, and ST8SIA3³⁸ (GT29) (Fig. 4). Enzyme–acceptor complexes for metal-dependent retaining GT-A fold enzymes include ABO⁴⁷ and GGTA1²⁵ (GT6), XXYLT1²² (GT8), XXT1⁴⁸ (GT34), and EXTL2⁴⁹ (GT64) (Fig. 5). Glycan-directed inverting GT-B enzymes include MGAT5⁴⁰ (GT18) and FUT1^21,41 (GT37). Polypeptide-directed GT-A fold enzymes include the inverting enzyme XYLT1 (GT14) and the retaining enzymes GALNT2 and GALNT4 (GT27) (Fig. 6). GT-B fold inverting enzymes include POFUT1²⁰ (GT65), POFUT2⁴² (GT68), and POGLUT1⁴³ (GT90) (Fig. 7), as well as OGT⁵⁰ (GT41). These structures are described in more detail in the following sections.

Fig. 4 |. Structural gallery of eukaryotic GT29 sialyltransferases as acceptor-bound complexes.

a, Three eukaryotic GT29 sialyltransferase structures (GT-A fold (variant 2)) were aligned using Coot⁸⁸ (transparent surface) with bound donor analogs (white sticks), acceptor analogs (yellow sticks), and catalytic bases (sticks). Structures include ST6GAL1 (PDB: 4JS2³⁵), ST3GAL1 (PDB: 2WNB³⁷), and ST8SIA3 (PDB: 5BO9³⁸). Boxes represent the regions of the respective structures shown in b. b, Enlarged views of the bound ligands are displayed in the same orientation and coloring as in a, with individual donors, acceptors, catalytic bases labeled as in Fig. 3. For ST6GAL1, the large Gal₂GlcNAc₂Man₃GlcNAc₂-Asn ligand has been simplified to the terminal Gal-β-1,4-GlcNAc-β-1,2-Man acceptor. c, Overlay of the three bound ligand complexes in two different orientations.

Fig. 6 |. Glycosyltransferases recognizing linear peptides.

a, GALNT2 (PDB: 5AJP⁶⁹). Enzyme represented in transparent surface representation, with the catalytic domain in green and lectin domain in light green. Donor nucleotide, white sticks; acceptor peptide, cyan sticks; GalNAc on acceptor peptide, red sticks; manganese, magenta sphere. Lower panel: enlarged view of the bound ligands are displayed in the same orientation and coloring as that shown on top with individual molecule components labeled. Acceptor amino acid is shown in red sticks, with the acceptor oxygen as a red sphere. b, XYLT1 (overlay of PDB IDs 6EJ7 and 6EJ8³⁹). The catalytic domain of the enzyme is represented in transparent surface representation in green. Donor nucleotide, white sticks; acceptor peptide, cyan sticks. Lower panel: enlarged cut-through view of the active site in tan with donor UDP-Xyl at the bottom, showing superposition of eight acceptor peptides in sticks (reproduced with permission from ref. ³⁹). Boxes in the upper panels represent the regions of the respective structures shown in the lower panels.

Glycan-modifying glycosyltransferases

Metal-dependent, inverting GT-A fold enzymes.

Several recent glycan-directed enzyme–substrate complexes have been solved for inverting GT-A fold enzymes that provide notable insights into acceptor interactions (Fig. 3). Alignment of six metal-dependent, inverting GT-A fold enzymes containing bound acceptors (B4GALT1¹³, B4GALT7²⁶, POMGNT1⁴⁴, MGAT2³¹, B3GAT1⁴⁵ and B3GAT3⁴⁶), as well as MGAT1 as an intact UDP-GlcNAc donor complex and GCNT1 as a representative metal-independent inverting GT-A fold enzyme, revealed conserved catalytic elements (Fig. 3). Each enzyme acts on a distinct glycan acceptor structure and employs a unique collection of acceptor binding interactions assembled from loops and extensions emerging from different positions in the respective core Rossmann folds (Fig. 2d,e). Despite the diversity of interactions, the positioning of the catalytic hydroxyl nucleophile is similar among all of the inverting enzymes in its proximity to the catalytic base and the C1 of the donor sugar residue (Fig. 3). However, two distinct patterns are revealed in the structural alignments. For five of the enzymes, the catalytic base is an Asp residue, and the corresponding position of the nucleophile and C1 of the donor sugar are almost identical (Fig. 3c). By contrast, for B3GAT1, B3GAT3, and GCNT1 the catalytic base is a Glu residue, and the positions for the catalytic base, nucleophile, and C1 of the respective sugar donor are also essentially identical, but in a position that is shifted relative to the enzymes that employ Asp as the catalytic base (Fig. 3e). All eight enzymes position the catalytic bases at the end of a highly conserved α-helix. As a result, the differences in the lengths of the side chains for the catalytic bases result in a coordinated shift in position for both the corresponding nucleophilic hydroxyls and C1 of the sugar donors to accomplish their in-line S_N2 catalytic mechanisms (Fig. 3d). Surprisingly, metal-independent GCNT1 and metal-dependent B3GAT1 and B3GAT3 share identical positioning of the catalytic base, nucleophile, and C1 of the sugar donor (or presumed C1 position for GCNT1 and B3GAT1) (Fig. 3e). These data suggest that the identity of the catalytic base, not the metal-dependent or -independent donor sugar interactions, determines the positioning of the donor and nucleophile for the GT-A fold inverting enzymes.

A highlight of these differences in acceptor interactions is the comparison of enzyme–acceptor complexes for MGAT2 (GT16) and POMGNT1 (GT13)³¹ (Fig. 3). These enzymes are structurally similar in their Rossmann-fold cores: both employ UDP-GlcNAc as donors, and both generate β-1,2-GlcNAc linkages on terminal Man residues using a metal-dependent inverting catalytic mechanism. MGAT2 initiates N-glycan branching in the Golgi during complex type N-glycan synthesis. The structure of its GlcNAcMan₃GlcNAc₂-Asn acceptor complex reveals substrate interactions through an extended binding site that anchors a terminal GlcNAc-β-1,2-Man-α-1,3-Man ‘recognition arm’ at a distal exosite and positions the Man-α-1,6-Man acceptor arm in the catalytic site³¹ (Fig. 3). By contrast, POMGNT1 modifies peptide-linked O-Man residues found predominately on α-dystroglycan, a component of the multiprotein dystrophin glycoprotein complex at the cell surface that interacts with the extracellular matrix in many tissues, including muscle^51,52. Defects in α-dystroglycan glycosylation can lead to congenital dystroglycanopathies with muscle, eye, and brain pathologies^51,53. POMGNT1 recognizes terminal Man residues on dystroglycan, but additional interactions with the peptide backbone also contribute to substrate specificity. Although acceptor substrates for MGAT2 and POMGNT1 are structurally distinct, and the acceptor binding sites are completely different, these two enzymes present their Man acceptor residues and their reactive C2 hydroxyl nucleophiles in essentially identical positions relative to the catalytic base and donor sugars (Fig. 3 and ref. ³¹). Thus, the extended acceptor binding sites act as templates complementary to their respective acceptor structures to achieve common positioning of the acceptor residue and identical catalytic mechanisms.

The MGAT2 exosite interactions with the GlcNAc-β-1,2-Man-α-1,3-Man ‘recognition arm’ were also found to resemble the bound substrate geometry and interactions for an unrelated enzyme in the N-glycan biosynthetic pathway, Golgi α-mannosidase II (MAN2A1)⁵⁴. MAN2A1 acts one step before MGAT2 in N-glycan biosynthesis and cleaves two Man residues from its GlcNAcMan₅GlcNAc₂-Asn substrate to produce the GlcNAcMan₃GlcNAc₂-Asn acceptor for MGAT2. The two enzymes have completely different protein folds and catalytic mechanisms, yet they both employ convergent exosite architectures for interaction with the same ‘recognition arm’ (GlcNAc-β-1,2-Man-α-1,3-Man) substructure common to both substrates³¹. This use of convergent exosite interactions by MGAT2 and MAN2A1 for recognition of a common glycan substructure is in contrast to the larger collection of GT-A fold enzymes that use similar core fold elements for sugar donor interactions and highly divergent loop regions to recognize and position a diverse number of acceptor structures for similar mechanisms of sugar transfer (Fig. 3).

Metal-independent, inverting GT-A fold enzymes.

Analogous use of conserved sugar donor interactions and divergent acceptor recognition is also seen for the metal-independent GT29 sialyltrans-ferases^4,35–38 (Fig. 4a,b). These enzymes generate distinct NeuAc capping structures on glycoproteins (N- and O-glycans) and glycolipids based upon unique acceptor specificities among the 20 mammalian GT29 isoforms. Terminal sialylated glycan epitopes provide a broad collection of ligand structures involved in developmental, immunological and pathological interactions in mammalian systems^55–58. The GT29 sialyltransferases employ a modified GT-A fold structure (GT-A variant 2) comprised of conserved ‘sialylmotif’ element sequences in the core of the Rossmann-fold scaffold to assemble a metal-independent cytidine monophospho-N-acetylneuraminic acid (CMP-NeuAc) binding site using adjoining proximal loop regions (Fig. 4)^36,59,60.

GT29 sialyltransferases fall into six sequence subclades that differ in the acceptor monosaccharides they bind and the linkage positions they produce: NeuAc-α-2,3-Gal (two subclades), NeuAc-α-2,6-Gal, NeuAc-α-2,6-GalNAc (two subclades), and NeuAc-α-2,8-NeuAc linkages³⁶. Representative acceptor-bound complexes have been determined for three of the subclades and revealed surprising differences in acceptor interactions among members of this family. Each of the sialyltransferases position the nucleotide or nucleotide sugar donors similarly to the conserved sialylmotif element sequences and employ a conserved His residue as catalytic base. The structure of ST6GAL1, the major N-glycan NeuAc-α-2,6-Gal-synthesizing enzyme^60,61, revealed an enzyme-bound Gal₂GlcNAc₂Man₃GlcNAc₂-Asn acceptor complex demonstrating interaction of the terminal Gal-β;-1,4-GlcNAc unit in the enzyme active site with the terminal Gal C6 hydroxyl positioned adjacent to the His catalytic base^35,36 (Fig. 4a,b). By contrast, ST3GAL1, the major enzyme synthesizing sialylated core 1 O-glycan structures (NeuAc-α-2,3-Gal-β-1,3-GalNAc-Ser/Thr)^60,61, was crystallized with a bound Gal-β;-1,3-GalNAc-oNP acceptor³⁷ and revealed a completely different ‘flipped’ geometry for the terminal Gal acceptor residue that positions the C3 hydroxyl adjacent to the His catalytic base as the nucleophile for glycan transfer (Fig. 4a–c)³⁶. By comparison, ST8SIA3, an enzyme involved in generating NeuAc-α-2,8-NeuAc-glycan structures on glycoproteins and glycolipids^60,61, was solved as a ternary complex containing a CMP-3F-NeuAc donor analog and a NeuAc-α-2,3-Gal-β;-1,4-GlcNAc-6-SO₄ acceptor³⁸. In contrast to the two Gal-specific sialyltransferases, the latter enzyme employed completely different acceptor interactions to position the NeuAc C8 hydroxyl adjacent to the His catalytic base as the nucleophile for transfer (Fig. 4a–c). The recent structure of ST6GALNAC2, an enzyme that synthesizes an alternative sialylated core 1 isomer structure relative to ST3GAL1^60,61, was determined as a CMP complex, but without a bound acceptor analog⁴. The positioning of potential loops for acceptor interactions were quite distinct from those of ST6GAL1, ST3GAL1, or ST8SIA3, suggesting that even within this single CAZy family there is considerable diversity in acceptor substrate recognition through distinct binding-site architectures.

Metal-dependent, retaining GT-A fold enzymes.

Substrate interactions for retaining GT-A fold enzymes also share many features with those of GT-A fold inverting enzymes (Fig. 5). Five retaining GT-A fold enzyme–acceptor complexes (ABO⁴⁷, GGTA1²⁵, XXYLT1²², XXT1⁴⁸, and EXTL2⁴⁹) demonstrate analogous metal-dependent interactions with the phosphodiester of the nucleotide, mediated by an enzyme-associated DxD motif and a similar overall position for the nucleotide sugar (Fig. 5). Retaining GT-A fold enzymes also assemble their glycan acceptor sites from unique loops emerging from the core Rossmann fold to provide complementary templates for glycan interactions. The enzyme–acceptor complexes displayed in Fig. 5 indicate that the approach of each acceptor nucleophile to the catalytic site is unique, with the exception of the two GT6 enzymes ABO and GGTA1, which employ an analogous approach for their nucleophilic Gal residues. As indicated in the mechanisms summary above (Figs. 1 and 2a–c) these latter enzymes presumably do not use active site amino acids for acceptor deprotonation, but position their respective acceptor nucleophiles adjacent to the donor β-phosphate oxygen for deprotonation and the proposed same-side dissociative S_Ni-type mechanism.

Inverting GT-B fold enzymes.

Although far fewer structures of eukaryotic glycan-directed GT-B fold enzymes have been determined with bound acceptors, two recent structures have provided insights into glycan acceptor recognition by these enzymes. The first, FUT1, is an inverting enzyme that generates α-1,2-Fuc linkages on the termini of plant cell wall xyloglucan side branches. Xyloglucans are highly abundant cell wall polymers comprised of a β-1,4-glucan backbone decorated with a repeating branching pattern of α-1,6-Xyl, β-1,2-Gal and α-1,2-Fuc substituents. FUT1 structures were determined with bound GDP and octasaccharide acceptor ligands comprised of a Gal-β-1,2-Xyl-α-1,6-modified cellopentaose structure^21,41. Unlike the protein-domain-specific fucosyltransferases described below, the octasaccharide acceptor structure assumes a flattened planar arrangement that interacts across the face of the enzyme with only the Gal acceptor residue extending perpendicularly into the active site. The inverting catalytic mechanism for this enzyme is more complex, as there is no apparent catalytic base available for nucleophile deprotonation. As described earlier, an S_N1 mechanism (Fig. 1c) has been invoked with a water-mediated proton shuttle for nucleophile deprotonation²¹.

The structure of a second GT-B inverting enzyme, MGAT5, involved in the branching of complex type mammalian N-glycans, was also recently described. The enzyme extends a β-1,6-GlcNAc branch on GlcNAc₂Man₃GlcNAc₂-Asn glycan biosynthetic intermediates that can then be further extended to form poly-N-acetyllactosamine-containing structures⁶². MGAT5 expression is upregulated in response to Ras-Raf-Ets activation^63,64 and was identified as a cancer-associated GT^65,66, contributing to alterations in cell adhesion during metastasis in transformed tissues⁶⁷. Binding affinity for the UDP-GlcNAc sugar donor is unusually poor for a GT, but the enzyme has higher affinity for glycan acceptor interactions^65,66. The MGAT5 structure was solved in complex with a chimeric bisubstrate-type inhibitor (GlcNAc-β-1,2-Man-α-1,6-Man tethered to UDP-GlcNAc), but only the glycan acceptor component could be resolved bound exclusively by the B domain of the GT-B fold⁴⁰. Further structure determination in the presence of donor analogs will be required before the catalytic mechanism of this latter enzyme can be clarified.

Protein-modifying glycosyltransferases

Glycosyltransferases that directly modify hydroxyls of Ser or Thr in a protein must bind their acceptor protein substrates to orient the respective Ser/Thr hydroxyl nucleophile correctly for sugar transfer (Fig. 1). Two classes of protein-modifying GTs include those that modify linear peptides and those that modify folded, cysteine-rich domains.

GTs that modify linear peptides.

Structures of several eukaryotic enzymes that add sugars to unstructured peptides have been determined, including polypeptide GalNAc-transferases (GALNTs), protein O-xylosyltransferase 1 (XYLT1), O-GlcNAc transferase (OGt), and oligosaccharyltransferase (OST) (Supplementary Dataset 1). Here we will limit our discussion to GALNTs and XYLT1. Twenty GALNTs exist in the human genome, and structures of two of them (GALNT2 and GALNT4) have been solved in complex with acceptor peptides^23,68–70. GALNTs are Golgi-localized, membrane bound, retaining enzymes with two lumenally oriented domains: an N-terminal glycosyltransferase domain (metal-dependent GT27 GT-A fold; Supplementary Dataset 1) and a C-terminal lectin domain with a β-trefoil fold held together by a flexible linker²⁷. They transfer GalNAc from UDP-GalNAc to peptides rich in Ser, Thr and Pro, often leading to heavily glycosylated mucin-type regions. Some of the GALNTs prefer unmodified peptides as acceptor substrates, whereas others prefer partially GalNAc-modified peptides. The peptides bind in a hydrophobic groove on the surface of the enzymes, which can accommodate a variety of sequences. For GalNAc-modified peptides, the lectin domain binds to the GalNAc and positions the peptide so that a specific Ser/Thr residue at a particular distance from the GalNAc can be glycosylated (Fig. 6a). For instance, GALNT2 has a preference for adding GalNAc to a Ser/Thr residue about 10 residues N-terminal to the GalNAc-modified residue^27,69. Other GALNTs have different specificities, allowing heavy modification of Ser/Thr-rich regions. For instance, GALNT4 preferentially modifies a Ser/Thr residue C-terminal to a previously modified residue⁷⁰. Interestingly, the flexible linker between the catalytic and lectin domains has been shown to fine-tune acceptor substrate recognition and glycosylation⁷⁰. The GALNTs are the only known mammalian GTs with a lectin domain that contributes to the specificity of the enzyme. The differing specificities of GALNTs likely explain why a strict consensus sequence for predicting sites of GalNAc modification does not exist, although algorithms for prediction of sites are improving (see http://www.cbs.dtu.dk/services/NetOGlyc/ and http://isoglyp.utep.edu/index.php). In all cases, the enzymes orient the acceptor hydroxyl group on the same side of the anomeric carbon of the sugar as the nucleotide to perform an S_Ni-like reaction resulting in retention of configuration (Fig. 1e).

Protein O-xylosyltransferases 1 and 2 (XYLT1 and XYLT2, respectively) catalyze transfer of xylose from UDP-Xyl to Ser residues with inversion of configuration, initiating the biosynthesis of glycosaminoglycan chain addition on proteogycans (Heparin Sulfate, Chondroitin Sulfate, Dermatan Sulfate)³⁹. XYLT1 and 2 are anchored to the luminal face of the Golgi complex with an N-terminal glycosyltransferase domain (metal-independent GT14 GT-A fold enzyme similar to GCNT1, Supplementary Dataset 1) linked to a C-terminal Xylo_C domain of unknown function. Transfer is specific for Ser residues generally surrounded by glycines, frequently with acidic residues positioned at −2 and −4 relative to the serine. A structure of ternary complex of XYLT1 with various acceptor peptides and UDP-Xyl revealed the molecular basis for this specificity (Fig. 6b)³⁹. The acceptor peptides bind a groove in the enzyme, placing the hydroxyl of the acceptor Ser residue precisely in position to perform a nucleophilic attack on the anomeric carbon of xylose (S_N2 reaction, Fig. 1b). The binding cleft narrows C-terminal to the Ser, explaining the preference for residues with small side chains in the +1 and +2 positions. The binding cleft is larger on the N-terminal side allowing more flexibility, although there is a general positive charge explaining the preference for acidic residues in the −3 and −4 positions.

GTs modifying folded cysteine-rich domains.

Co-crystal structures of domain-specific GTs with bound acceptor domains include protein O-fucosyltransferases 1 and 2 (POFUT1 (GT65) and POFUT2 (GT68)) and protein O-glucosyltransferase 1 (POGLUT1 (GT90)), all of which are inverting GTs with GT-B folds (Supplementary Dataset 1). All three are soluble enzymes localized in the ER that have the unique ability to recognize folded protein domains^71–73. POGLUT1 and POFUT1 add O-glucose or O-fucose, respectively, to properly folded epidermal growth factor-like (EGF) repeats containing the appropriate consensus sequences: C¹XSX(P/A)C² for POGLUT1 and C²XXXX(S/T)C³ for POFUT1 (in which C^1-3 are conserved cysteines in EGF repeats, and underlined residues are modified)^74–77. Although multiple proteins contain EGF repeats with these consensus sequences, the Notch family of receptors appears to be the major biological target for these modifications⁷⁸. POGLUT1 and POFUT1 are required for Notch receptor function in vivo^74–77.

POFUT2 adds O-fucose to properly folded thrombospondin type 1 repeats (TSRs) containing the consensus CXX(S/T)C (in which the C’s are either conserved cysteines 1 and 2, or cysteines 2 and 3, depending on whether the TSR is group 1 or group 2, respectively)⁷⁶. Nearly 50 proteins contain TSRs with this consensus sequence, including all members of the A disintegrin and metallo-protease with thrombospondin repeats (ADAMTS) family of extracellular proteases⁷⁸. Deletion of Pofut2 in mice results in an identical embryonic lethal phenotype to that with deletion of Adamts9, suggesting that ADAMTS9 requires addition of O-fucose glycans for proper function^79,80.

POFUT1, POFUT2, and POGLUT1 all have a large cleft between the two Rossman-like folds where the folded protein domains bind, covering approximately one third of the surface area of the domains (Fig. 7)^20,42,43,81. The cleft in each enzyme shows remarkable surface complementarity for the face of the domain bound in the cleft with multiple hydrogen bonds, especially within the loop containing the consensus sequences (see refs. ^20,42,43,81 for details). This orients the domain to place the hydroxyl group of the Ser/Thr acceptor precisely in position to perform nucleophilic attack on the anomeric carbon of the fucose or glucose (S_N2 reaction for POGLUT1^43,81 and POFUT2⁴², S_N1 for POFUT1^19,20, Fig. 1). Interestingly, POFUT2 also has multiple water molecule-mediated hydrogen bonds at the enzyme-substrate interface, presumably allowing flexibility of sequences for different TSRs to bind to the enzyme⁴². These structures clearly demonstrate why both a consensus sequence and a properly folded domain are needed for modification by these enzymes. All three enzymes have been implicated in quality control pathways for the proper folding of EGF repeats (POFUT1 and POGlUT1⁸²) and TSRs (POFUT2⁸³) since the glycosylated product stabilizes the folded structural domains^76,77,84.

Future directions

The GT structures presently in PDB represent only 53 of the 106 total CAZy families and only 15 of 45 CAZy families with vertebrate members (Supplementary Dataset 1). Since many of these enzyme families have multiple members with distinct acceptor and donor specificities, the present structures clearly represent only a minor fraction of the glycan synthetic capability in biological systems. The rate of new GT structures deposited in PDB has only minimally increased in recent years, but a greater number of recent structures for eukaryotic enzymes have resulted from significant improvements in eukaryotic expression strategies. Hopefully these efforts will expand more effectively in the coming decade and include a greater number of the more challenging GT-C fold enzymes. An increase in more complex enzyme structures determined by cryo-electron microscopy will also be expected in the coming years^11,85–87.

The goals of these studies will be to understand not only the structural basis of linkage-specific glycan synthesis, but also broader themes of how the GTs provide catalytic templates for glycan assembly and how faithful replication of glycan structures is accomplished with high fidelity among cells of common lineages. The restricted set of protein folds and the modular nature of the GT active site architectures have allowed independent evolution of specificities within donor, acceptor, and catalytic modules to expand the repertoire of glycan structures in biological systems. A major challenge for the future will be to improve our understanding of how selective pressures of evolution have led to the present collection of enzyme specificities and how novel specificities could evolve or be engineered. These insights will only come from a more comprehensive comparative analysis of GT sequences from varied organism sources paired with enzyme specificity studies and structures determined for GT enzyme–substrate complexes. Challenges for the latter goals will be large-scale production of the full complement of eukaryotic glycoenzymes and the development of glycan, glycopep-tide, and glycolipid libraries for use as substrates and ligands. These efforts will reveal mechanisms that drive evolution of new catalytic templates and enzyme specificities that lead to glycan diversification for cell surface and secreted glycoconjugates. The studies would also provide insights that allow engineering of novel enzyme capabilities for biochemical and biotherapeutic applications that would catalyze glycoscience advances in the future.