Crystal structures of β-1,4-N-acetylglucosaminyltransferase 2: structural basis for inherited muscular dystrophies

Crystal structures of human β-1,4-N-acetylglucosaminyltransferase 2, an enzyme essential for O-mannosylation, are reported, revealing a novel domain organization and providing the first rational basis for an explanation of the loss-of-function mutations observed in the clinic.

Abstract

The canonical O-mannosylation pathway in humans is essential for the functional glycosylation of α-dystroglycan. Disruption of this post-translational modification pathway leads to congenital muscular dystrophies. The first committed step in the construction of a functional matriglycan structure involves the post-translational modification of α-dystroglycan. This is essential for binding extracellular matrix proteins and arenaviruses, and is catalyzed by β-1,4-N-acetylglucosaminyltransferase 2 (POMGNT2). While another glycosyl transferase, β-1,4-N-acetylglucosaminyltransferase 1 (POMGNT1), has been shown to be promiscuous in extending O-mannosylated sites, POMGNT2 has been shown to display significant primary amino-acid selectivity near the site of O-mannosylation. Moreover, several single point mutations in POMGNT2 have been identified in patients with assorted dystroglycanopathies such as Walker–Warburg syndrome and limb girdle muscular dystrophy. To gain insight into POMGNT2 function in humans, the enzyme was expressed as a soluble, secreted fusion protein by transient infection of HEK293 suspension cultures. Here, crystal structures of POMGNT2 (amino-acid residues 25–580) with and without UDP bound are reported. Consistent with a novel fold and a unique domain organization, no molecular-replacement model was available and phases were obtained through crystallization of a selenomethionine variant of the enzyme in the same space group. Tetragonal (space group P42₁2; unit-cell parameters a = b = 129.8, c = 81.6 Å, α = γ = β = 90°) crystals with UDP bound diffracted to 1.98 Å resolution and contained a single monomer in the asymmetric unit. Orthorhombic (space group P2₁2₁2₁; unit-cell parameters a = 142.3, b = 153.9, c = 187.4 Å, α = γ = β = 90°) crystals were also obtained; they diffracted to 2.57 Å resolution and contained four monomers with differential glycosylation patterns and conformations. These structures provide the first rational basis for an explanation of the loss-of-function mutations and offer significant insights into the mechanics of this important human enzyme.

1. Introduction  

Congenital muscular dystrophy (CMD) encompasses a family of degenerative diseases that exhibit myopathy, contractures and in some cases abnormalities of the nervous system. Many of these patients do not survive long after birth, and many CMDs are due to a loss of functional glycosylation on α-dystroglycan (α-DG) and are referred to as secondary dystroglycanopathies. In higher organisms, an important cellular interaction is the dystrophin–glycoprotein (dystroglycan) complex that acts to bridge the actin cytoskeleton to laminin globular domain-containing proteins in the extracellular matrix (Ervasti & Campbell, 1993 ▸). Primary dystroglycanopathies generally arise from mutations in the gene encoding α-DG (DAG1) and can lead to CMD, but only a few such mutations have been identified to date (Hara et al., 2011 ▸; Geis et al., 2013 ▸; Dai et al., 2019 ▸). Instead, the vast majority of dystroglycanopathies result from mutations in genes encoding glycosyltransferases in the O-mannosylation pathway that specifically modify α-DG (Praissman & Wells, 2014 ▸). This is because the fully elaborated functional O-mannose structure terminating in a repeating disaccharide referred to as matriglycan serves as the receptor for laminin globular (LG) domain-containing proteins in the extracellular matrix (Yoshida-Moriguchi & Campbell, 2015 ▸). The enzymes responsible for glycosylation of α-DG in its mucin-like domain (residues 313–489) have been characterized and mutations in the genes encoding these enzymes have been identified as causal of secondary dystroglycanopathies (Sheikh et al., 2017 ▸; Brockington et al., 2001 ▸; Voglmeir et al., 2011 ▸).

The O-Man post-translational modification pathway for α-DG begins in the endoplasmic reticulum. First, the enzymes POMT1 and POMT2 catalyze the transfer of mannose from dolichol monophosphate mannose (Dol-P-Man) to serine and threonine residues of α-DG in an α-linkage. Further bifur­cation of the pathway proceeds with the addition of an N-acetylglucosamine (GlcNAc) with either a β-1,2 or a β-1,4 linkage mediated by POMGNT1 and POMGNT2, respectively (Fig. 1 ▸). At most sites on α-DG, POMGNT1 catalyzes the addition of a β-1,2-linked GlcNAc to the initial mannose in the cis-Golgi (Takahashi et al., 2001 ▸; Stalnaker et al., 2010 ▸). Branching of this core M1 structure by another GlcNAc gives rise to the core M2 structure. At only a handful of sites on α-DG, POMGNT2 will extend the initial mannose with a β-1,4-linked GlcNAc in the ER (Manzini et al., 2012 ▸; Yoshida-Moriguchi et al., 2013 ▸; Yagi et al., 2013 ▸), leading to formation of the core M3 glycan structure (Fig. 1 ▸). In this sense, and in contrast to the enzyme nomenclature, α-DG encounters POMGNT2 first in the secretory pathway. Following β-1,4-linked GlcNAc addition by POMGNT2, the glycan is subjected to further modification with a β-1,3-linked N-acetylgalactosamine by B3GALNT2. Phosphorylation of the reducing-end mannose at the 6-position by POMK results in a phosphotrisaccharide core M3 glycan structure. The core M3 phosphotrisaccharide can be extended by the enzymes FKTN, FKRP, TMEM5 and B4GAT1. In respective order, the modifications include the addition of tandem ribitol-phosphate units in phosphodiester linkages by FKTN/FKRP (Gerin et al., 2016 ▸; Kanagawa et al., 2016 ▸), TMEM5-catalyzed xylose addition to the distal ribitol-phosphate (Praissman et al., 2016 ▸; Manya et al., 2016 ▸) and the addition of glucuronic acid in a β-1,4 linkage to the xylose by B4GAT1 (Praissman et al., 2014 ▸; Willer et al., 2014 ▸). At this point, the functional component of the glycan can be added by the enzyme LARGE1. This modification is termed matriglycan and consists of the addition of a repeating disaccharide composed of an α-1,3-linked xylose-β-1,3-linked glucuronic acid. Matriglycan is responsible for binding to the LG domains of ECM proteins (Hara et al., 2011 ▸; Inamori et al., 2012 ▸).

Figure 1.

Known O-Man structures on α-dystroglycan (figure adapted from Halmo et al., 2017). (a) POMGNT1 is responsible for generating the core M1 glycan structure that can be branched by MGAT5B to generate core M2, whereas POMGNT2 is responsible for generating the core M3 glycan structure. Both of the core M3 sites identified to date are shown; only two of more than a dozen core M1 sites are shown for comparison.

In addition to its essential role in humans, several aspects of POMGNT2 function are puzzling and have remained largely unresolved. A number of POMGNT1 and POMGNT2 mutations have been identified as causal of secondary dystroglycanopathies (Manzini et al., 2012 ▸; Yoshida-Moriguchi et al., 2013 ▸; Endo et al., 2015 ▸). Yet, despite the fact that both POMGNT1 and POMGNT2 catalyze the addition of GlcNAc to mannose using the same GlcNAc donor (UDP-GlcNAc), there is no similarity between the primary sequences of the two enzymes. In fact, these two enzymes are predicted to reside in separate and distinct CAZy families, with POMGNT1 in glycosyltransferase family 13 and POMGNT2 in glycosyltransferase family 61 (Lombard et al., 2014 ▸). Additionally, previous work has shown that POMGNT2 displays significant sequence selectivity near the site of O-mannosylation in vitro (Halmo et al., 2017 ▸). This is in contrast to the promiscuous nature reported for POMGNT1 activity in vitro (Halmo et al., 2017 ▸). As evidence has emerged that POMGNT2 has a more targeted selectivity than POMGNT1, understanding the molecular basis of the structure and function of POMGNT2 is imperative. In order to advance our understanding of POMGNT2 function, we have pursued a crystallographic investigation of this enzyme and reveal a unique fold and domain organization. This investigation provides a foundation for advancing our understanding of the mechanism of function of POMGNT2 and enables an explanation of the biochemical and clinical observations reported to date.

2. Methods  

2.1. POMGNT2 expression and selenomethionine labeling  

POMGNT2 has been referred to as AGO61, C3ORF39 and GTDC2 in previous literature. In this work, we exclusively use the nomenclature POMGNT2. A protein-expression construct encoding the catalytic domain of POMGNT2 (protein O-linked-mannose β-1,4-N-acetylglucosaminyltransferase 2; UniProt Q8NAT1; residues 25–580) was generated by PCR from a Mammalian Gene Collection clone followed by Gateway recombination into the pDONR221 vector (Halmo et al., 2017 ▸). The PCR amplification extended the truncated POMGNT2 coding region by inclusion of flanking Gateway attL recombination sites as well as an extension of the amino-terminus of the coding region with a TEV protease recog­nition site as described previously (Halmo et al., 2017 ▸). Gateway LR recombination of the TEV-POMGNT2-pDONR221 vector with the mammalian Gateway-adapted expression vector pGEn2-DEST generated the POMGNT2-pGEn2 expression construct. The fusion-protein construct encodes an N-terminal signal sequence, 8×His tag, AviTag, ‘superfolder’ GFP, the TEV protease recognition site and the truncated POMGNT2 coding region behind a CMV promoter (Moremen et al., 2018 ▸). This POMGNT2-pGEn2 expression vector was used for transient transfection of HEK293S (GnTI⁻) cells (ATCC) in suspension culture using polyethylenimine (linear 25 kDa PEI; Polysciences) as a transfection reagent as described previously (Meng et al., 2013 ▸). The cultures were diluted 1:1 with culture medium containing 4.4 mM valproic acid (2.2 mM final concentration) 24 h after transfection, and protein production was continued for a further five days at 37°C. For metabolic labeling of HEK293S (GnTI⁻) cells with selenomethionine (SeMet), cells were transfected as described above and 12 h after transfection the medium was exchanged to custom methionine-free Freestyle 293 expression medium (Invitrogen) for 6 h to deplete methionine pools; the cultures were subsequently resuspended in methionine-free Freestyle 293 expression medium containing 60 mg l⁻¹ SeMet and protein production was continued for a further 4–5 days at 37°C (Moremen et al., 2018 ▸).

2.2. Purification of POMGNT2  

Protein purification, deglycosylation and tag removal employed workflows similar to previous structural studies of rat ST6GAL1 (Meng et al., 2013 ▸) and human ST6GALNAC2 (Moremen et al., 2018 ▸). Briefly, the conditioned culture medium was loaded onto an Ni²⁺–NTA Superflow (Qiagen) column equilibrated with 20 mM HEPES, 300 mM NaCl, 20 mM imidazole pH 7.4, washed with column buffer and eluted successively with column buffers containing stepwise increasing imidazole concentrations (40–300 mM). The eluted fusion protein was pooled, concentrated and concurrently mixed with recombinant TEV protease and EndoF1 at ratios of 1:10 relative to GFP-POMGNT2 for each enzyme, respectively, and incubated at 4°C for 36 h to cleave the tag and glycans. In order to remove any uncleaved enzyme, the digested material was loaded onto an Ni–NTA column and the catalytic domain of POMGNT2 with cleaved glycans was flowed through a Bio-Rad gel-filtration system. The protein was further purified on a Superdex 75 gel-filtration column (GE Healthcare) and the peak fractions of POMGNT2 were collected and concentrated to 5 mg ml⁻¹ by ultrafiltration for crystallization.

2.3. Crystallization  

High-throughput hanging-drop crystallization screens were performed on a Mosquito system (TTP Labtech) using commercially available sparse-matrix screens. The diffraction quality of the crystals consistently improved when UDP was present at 2–10 mM in the protein solution, although UDP was not always observed in the electron density. Vapor-diffusion (hanging-drop) trays were used to obtain crystals using a 1:1 ratio of protein:precipitant with a precipitant solution consisting of 0.1 M potassium/sodium tartrate, 0.1 M bis-Tris pH 7.5, 10% PEG 10 000. Crystals grew within 24 h and were prepared for cooling by incremental addition (no more than a 2.5% increase) of glycerol, DMSO and ethylene glycol to a final concentration of 10% for each cryoprotectant.

2.4. Data collection, phasing and refinement  

All data collection was performed on beamline 22-ID at the Advanced Photon Source (APS) and all data reduction was performed with HKL-2000 using the ‘scale anomalous’ function in SCALEPACK. Eightfold redundant data were collected from selenomethionine-labeled and native POMGNT2 crystals at 12 782 eV. 12 selenium sites (out of 13 possible sites; see Table 1 ▸) were identified using Phaser. The 12 positions were further refined and solvent flattening was performed using REFMAC, leading to interpretable maps. AutoBuild was then used to build an initial model that included only elements of secondary structure. This model was used as a molecular-replacement model for phasing the higher resolution data sets. Iterative rounds of model building and refinement were carried out using Coot (Emsley & Cowtan, 2004 ▸; Emsley et al., 2010 ▸) and Phenix (Adams et al., 2010 ▸; Liebschner et al., 2019 ▸) for each of the two models. The peptide was placed in electron density first, followed by sugar molecules and cofactors, with water molecules being added last.

Table 1. Data-collection and refinement statistics for POMGNT crystals.

Values in parentheses are for the outermost shell.

	Native 1	Native 2	SeMet
Data collection
Beamline	22-ID, APS	22-ID, APS	22-ID, APS
Space group†	P4212	P212121	P4212
a, b, c (Å)	129.8, 129.8, 81.6	142.3, 153.9, 187.4	130.4, 130.4, 81.6
Wavelength (Å)	0.97	0.97	0.97
Resolution range (Å)	50.0–1.98 (2.05–1.98)	50.0–2.57 (2.66–2.57)	50.0–3.2 (3.31–3.20)
Unique observations	49420	131723	12199
Completeness (%)	99.7 (98.0)	99.1 (93.1)	100.0 (99.6)
R merge ‡ (%)	9 (35)	13 (65)	18 (54)
CC1/2	0.99 (0.81)	0.96 (0.86)	0.99 (0.94)
Multiplicity	8.3 (6.7)	14.6 (12.6)	15.1 (11.4)
〈I/σ(I)〉	27.0 (2.8)	20.3 (2.2)	18.1 (4.3)
Phasing (SeMet)
Sites			12
BAYES-CC			49.6 ± 11.3
Figure of merit			0.47
Refinement
Protein atoms	4237	17040
Solvent atoms	196	149
Resolution limits (Å)	50.0–2.00	50.0–2.57
R cryst (%)	19.8	21.3
R free (%)	22.7	24.7
Ramachandran outliers (%)	1.0	1.5
Side-chain outliers (%)	1.1	1.8
R.m.s.d., bond lengths (Å)	0.008	0.002
R.m.s.d., angles (°)	0.968	0.600
Average B factor (Å2)	30.2	37.8
Ramachandran plot
Most favored (%)	96	95
Allowed (%)	3	3.5
Outliers (%)	1	1.5
PDB code	6xfi	6xi2

^†All angles are 90° in both crystal forms.

^‡ R _merge = \textstyle \sum_{hkl}\sum_{i}|I_{i}(hkl)- \langle I(hkl)\rangle|/\textstyle \sum_{hkl}\sum_{i}I_{i}(hkl), where I(hkl) is the intensity of an individual measurement of the reflection with indices hkl and 〈I(hkl)〉 is the mean intensity of that reflection.

3. Results  

3.1. Overall structure and domain organization  

The overall model of POMGNT2, determined to 1.98 Å resolution in the presence of UDP (Table 1 ▸), reveals a single peptide packed in the asymmetric unit that consists of three distinct domains when analyzed by the Domain Identification Algorithm (DIAL) server (http://caps.ncbs.res.in/DIAL/) (Fig. 2 ▸). Domains 1 and 2 of POMGNT2 are unique and constitute the catalytic portion of POMGNT2. UDP is bound within an active-site cleft at the interface of domains 1 and 2 (Fig. 2 ▸) and no metal ions are observed within the active site, consistent with the lack of a DxD metal-binding motif. This is addressed further below, but it is important to point out that several loops, as well as the entire peptide backbone between domains 2 and 3 (Fig. 2 ▸, dashed line), were not modeled owing to a lack of electron density. This is indicative of considerable movement in several regions of the protein within the crystal lattice, and is reflected in the B factors as well as the refinement statistics of our models (Table 1 ▸). The lack of electron density for the peptide backbone between domains 2 and 3 is consistent with domain 3, a lectin-binding domain, having a high degree of mobility relative to the rest of the enzyme. Unlike domain 3, the catalytic portion (domains 1 and 2; Fig. 2 ▸) of POMGNT2 is structurally unique. In fact, the top two hits from a structural alignment of the catalytic portion of POMGNT2 using DALI were a family 9 glucosyltransferase from Veillonella parvula (PDB entry 3tov; Midwest Center for Structural Genomics, unpublished work) and an iron-containing glycosyltransferase (TibC) from enterotoxigenic Escherichia coli H10407 (PDB entry 4rap; Yao et al., 2014 ▸). However, while these structural alignments confirm that POMGNT2 has some similar functional folds (i.e. a partial Rossmann-like fold in domain 2), the alignments are poor and limited in their scope. Specifically, only 209 amino acids (out of 346) in the A chain of the family 9 glucosyltransferase structure aligned, with an r.m.s.d. of 4.0 Å and a Z-score of 9.8. For TibC, 215 amino acids (out of 390) aligned, with a Z-score of 9.3 and an r.m.s.d. of 4.7 Å. The Z-score for the alignment of protein structures with similar folds, even within a large enzyme superfamily, is typically much higher. Taken together, these data indicate that the structure of the catalytic portion of POMGNT2 is unique and, relative to enzymes that catalyze similar reactions, there are significant structural differences in the active site. For example, there are no acidic amino acids that might function as a catalytic base.

Figure 2.

Cartoon representation of the POMGNT2 model seen in the asymmetric unit with the three distinct domains identified (numbered 1–3) and two functional domains. No electron density was observed for the linker region (residues 167–177; dashed line) between the catalytic domain (domains 1 and 2) and the lectin domain (domain 3). The atoms of amino acids for which disease-causing mutations have been identified are shown as spheres and labeled with the wild-type residue. Inset: the UDP molecule is shown in stick representation with C, O, N and P atoms colored tan, red, blue and orange, respectively.

Both POMGNT2 and POMGNT1 are essential to catalyze the first steps of essential post-translational modifications involving complex glycan structures. Given the physiological significance of both enzymes, the cellular proximity of their function, their notable functional differences and the fact that the structure of POMGNT1 has been reported, it is logical to compare the structural properties of POMGNT2 in light of what is known about the structure and function of POMGNT1. The lectin-binding domain of POMGNT2 (domain 3; Fig. 2 ▸) may play a role in substrate recognition, as has been proposed for a similar domain that is also present in POMGNT1. This was referred to as a ‘stem’ domain (Kuwabara et al., 2016 ▸) in POMGNT1 and contained both α and β structural elements. In POMGNT2 this domain is dominated by β structure, with the exception of a single short α-helix. Another striking observation is that this domain is found at the C-terminus of POMGNT2 and at the N-terminus of POMGNT1. In fact, all three domains in POMGNT2 are oriented in reverse order in the peptide chain when compared with POMGNT1. It is important to note that Kuwabara and coworkers described the catalytic portion of POMGNT1 as a single domain containing two functional motifs or ‘lobes’.

Based on biochemical assays that have been performed in vivo, the primary functional difference appears to be that POMGNT1 is promiscuous for O-mannosylated peptides, whereas POMGNT2 displays significant primary amino-acid selectivity near the site of O-mannosylation (Halmo et al., 2017 ▸). We were initially optimistic that the crystal structure of POMGNT2, when combined with data on acceptor binding to POMGNT1 and POMGNT2 (Akasaka-Manya et al., 2011 ▸; Halmo et al., 2017 ▸), could explain the functional differences. However, as the model of POMGNT2 now reveals, there are substantial differences in the organization of the domains as well as in the domain structure itself.

3.2. Binding mode of UDP and the mechanism of POMGNT2  

At the present time, we have been unable to obtain crystals of POMGNT2 with a bound acceptor molecule. Numerous attempts were made using several acceptors in pre- and post-turnover states. In all cases, UDP was the only molecule that was ever seen in the active-site cleft. Fig. 3 ▸ shows a LigPlot diagram detailing the protein–UDP interactions that we observe in the POMGNT2 model. The active site is located within a cleft that is found between domains 1 and 2, similar to that reported for POMGNT1 (Kuwabara et al., 2016 ▸). Interestingly, there is a loop containing a disulfide bond between Cys436 and Cys437 at the opposite end of the active site to the bound UDP molecule. This disulfide is observed in both models (Supplementary Fig. S1), and reduction of this disulfide could certainly influence the active-site structure and the binding of the acceptor. Unlike POMGNT1, POMGNT2 does not contain a DxD metal-binding motif; instead, we observe two strictly conserved arginine residues (Arg294 and Arg298) that form hydrogen bonds, and most likely engage in ionic interactions, with the diphosphate portion of UDP (Fig. 3 ▸). In contrast, POMGNT1 exhibits cation-dependent activity, with the diphosphates of UDP interacting with a manganese ion (PDB entry 5ggi; Kuwabara et al., 2016 ▸) within the active site. A reasonable conclusion is that the UDP-binding mode we observe for POMGNT2 represents a post-turnover, thermodynamically stable, conformational state. Further interpretation is complicated due to the low structural homology with any other glycosyl- or acetylglucosaminyltransferase. However, structures are available of the closest functional homolog, POMGNT1, with substrates/products bound. Through a comparative structural analysis, some functional conclusions can be drawn. At the bottom of the active-site cleft, within 5 Å of the UDP β-phosphate and Arg298, POMGNT2 contains a histidine residue (His345). Relative to the UDP-binding mode, this side chain is oriented similarly to Asp476, the catalytic base in POMGNT1. More noteworthy, based on the best alignment produced by a DALI search, His345 is similarly positioned close to the sugar donor in the TibC model (Fig. 4 ▸). In addition, His345 is hydrogen-bonded to Glu363, a buried residue that is also hydrogen-bonded (2.5 Å) to Tyr390. Given these interactions, a reasonable prediction is that the pK _a of the His345 side chain is substantially more basic, providing additional support for this amino acid acting as a catalytic base in the mechanism of POMGNT2. Experimentally, a recombinant variant enzyme (H345D) could be isolated; however, the purified H345D mutant did not display any measurable activity against a synthetic O-Man peptide derived from α-DG (Fig. 4 ▸). The detection limit of these assays suggests that this corresponds to at least a 100-fold reduction in k _cat compared with the wild-type enzyme. The amino acids on either side of His345 within the active site are predominantly hydrophobic and may play a role in the acceptor specificity observed for POMGNT2. These include several aromatic amino acids (Tyr374, Tyr377 and Trp442) and hydrophobic amino acids (Ile446 and Met165). Aromatic residues have an important role in acceptor recognition through carbohydrate–π inter­actions, and clinical mutations at Met165 have already been identified. Both of these points are discussed further in the following sections.

Figure 3.

LigPlot diagram detailing the interaction of UDP (purple ball-and-stick representation) with specific atoms of the POMGNT2 model. Hydrogen bonds are shown as green dashed lines and the sunburst icons represent hydrophobic interactions.

Figure 4.

Cartoon representation (left) of the DALI-generated structural overlay between TibC (yellow cartoon and C atoms) and POMGNT2 (tan cartoon and C atoms) as well as assay data (right) for the H345D variant of POMGNT2. The substrate-bound form of TibC (PDB entry 4rb4) was used for structural alignment. The peptide substrate 21man317 was prepared as described by Halmo et al. (2017 ▸) and three 18 h assays were performed using purified POMGNT2 (3 µg) proteins. The UDP-Glo assays (20 µl total volume) were performed according to the manufacturer’s instructions (Promega, Madison, Wisconsin, USA) as described in Section 2.

3.3. Evidence for a functional dimer  

Although a single monomer of the POMGNT2 peptide was observed in the asymmetric unit of the tetragonal lattice (space group P42₁2), there is compelling evidence, based on the packing of the unit cell in both data sets, that the functional form of POMGNT2 is a dimer. In particular, no direct interactions between the lectin-binding domain and the catalytic domain are seen within the asymmetric unit. In fact, the shortest distance between an atom from the lectin-binding domain and an atom from the catalytic domain is 7.4 Å. The closest contact distance between the UDP molecule and the lectin-binding domain is 33 Å. To provide a possible explanation, we generated all of the symmetry mates and looked for another monomer that satisfied two criteria. Firstly, the monomer must bury a significant amount of surface area when forming a dimer with the monomer in the asymmetric unit. The second requirement was that the lectin-binding domain of the symmetry molecule must be closer to the UDP molecule of the original monomer. In POMGNT1, the lectin-binding domain has been proposed to facilitate the GlcNAcylation of neighboring (or nearby) O-mannose moieties (Kuwabara et al., 2016 ▸). An acceptable solution that satisfied both criteria was identified and is shown in Fig. 5 ▸. According to a PISA analysis, the dimer interface shown in Fig. 4 ▸ has a buried surface area of 2904 Ų. Moreover, the closest contact distance between the bound UDP molecule of one monomer and the lectin-binding domain of the other monomer is reduced to less than 25 Å. Given the length of the linker domain and observations for other enzymes containing such recognition domains, a reasonable hypothesis is that the lectin-binding domain is quite dynamic and multiple orientations of this domain relative to the active site of the catalytic domain are also observed. These observations are most likely to be the result of enzyme-preparation protocols. Specifically, deglycosylation of the enzyme does not result in a homogenous sample (different sites on the enzyme may still be glycosylated), leading to multiple conformations of the monomer and to the orthorhombic space group (P2₁2₁2₁; Table 1 ▸). Interestingly, the asymmetric unit of the ortho­rhombic crystals, as discussed further below, contains four monomers and two copies of the proposed dimer originally identified in the packing analysis of the tetragonal data set. The shift in space group is clearly due to each of the lectin-binding domains adopting a slightly different orientation relative to domains 1 and 2 that contain the active site. This is discussed further in Section 3.5. Analytical ultracentrifugation experiments will be useful in clarifying the equilibrium between oligomeric states in future experiments.

Figure 5.

Cartoon representation and transparent surface representation showing the POMGNT2 monomer found in the asymmetric unit (yellow) and a symmetry-related monomer (purple) that forms a dimer interface with 2904 Ų buried surface area. The loop region between the domains is implied with a dashed line. No electron density was observed for these amino acids (residues 167–177).

3.4. Disease-causing variants of POMGNT2  

Several homozygous single point mutations and one heterozygous double mutation in the gene encoding POMGNT2 have been identified in patient populations (Manzini et al., 2012 ▸; Yoshida-Moriguchi et al., 2013 ▸; Endo et al., 2015 ▸). The relative positions of all of these mutations are shown in Fig. 2 ▸. Surprisingly, all of the disease-related mutations appear to be confined to domain 2 or the cleft between domains 1 and 2 (Fig. 2 ▸). The fact that the disease mutations all cluster within one domain of the structure confirms that there must be some selective pressure to maintain this fold. Although, all things being equal, the first principles of protein folding/evolution dictate that protein-misfolding variants could arise from any part of the enzyme structure, this is clearly not the case. Therefore, we must conclude that there must be some, albeit very low, level of enzyme activity for these variants.

The R158H mutation is not too distant from the active-site cleft where UDP is bound; it was identified by homozygosity mapping combined with whole-exome sequencing and was confirmed by Sanger sequencing (Manzini et al., 2012 ▸). In the wild-type enzyme, this amino acid is involved in a salt bridge at the interface of motifs 1 and 2 within the catalytic domain (Fig. 6 ▸ a). The salt bridge is buried and any mutation here would have a high probability of significantly destabilizing the structure. The patient with the R158H mutation displayed classical features of the most severe form of CMD: Walker–Warburg syndrome (WWS). The W197* mutation is caused by a 590 G–A transition in the POMGNT2 gene (Manzini et al., 2012 ▸). This mutation results in premature protein truncation in the glycosyltransferase domain. Interestingly, the side chain of Trp197 is not buried, but rather is exposed to solvent and is only two amino acids away from Glu195, which forms the salt bridge with Arg158 (Fig. 6 ▸ b). The N atom of the tryptophan ring is close enough (within 3.4 Å) to form a hydrogen bond to the backbone carbonyl O atom of Phe159 (Fig. 6 ▸ b). Autopsies of patients with this mutation show classical features of WWS (Manzini et al., 2012 ▸). Given these physiological observations and the location of the Arg158–Glu195 salt bridge relative to Trp197, it seems clear that this loop region is of functional significance. Moreover, this region of POMGNT2 is also at the interface of domains 1 and 2. The acceptor/substrate molecules are large and a certain amount of flexibility between the functional domains may be required. We certainly observe this when we align the catalytic portion of POMGNT2 from all of the monomers that we have modeled in this study (Fig. 7 ▸). The rest of the disease mutations are found in domain 2, where UDP is bound. In fact, domain 2 of the catalytic portion of POMGNT2 is also where we observe the greatest degree of structural homology with other glycosyl- or acetylglucos­aminyltransferases, albeit with very poor alignment scores.

Figure 6.

Stick representations and 2F _o −F _c composite OMIT map contoured at 1.2σ (green cage) showing the interactions of amino acids in the 1.98 Å resolution POMGNT2 model that have been implicated in disease. Sites of mutation that have been identified in patients include Arg158 (a), Trp197 (b), Met165 (c), Pro253 (d), Gly413 (e) and Arg445 (f).

Figure 7.

Cartoon representation showing an alignment of the N-terminal fold of the catalytic subunit for all of the monomers of POMGNT2 built in this work. The locations of N-acetylglucosamine molecules within different monomers are represented by van der Waals spheres.

Met165 is located within 4.0 Å of one of the UDP phosphates (Fig. 6 ▸ c) at the edge of the same cleft where UDP is bound. However, the thioether side chain is swung away from the UDP molecule and tucked into a hydrophobic pocket. A mutation at Pro253 (P253L), in combination with a mutation at Met165 (M165T), was identified in patients with limb girdle muscular dystrophy (LGMD) and intellectual disability without brain malformation (Endo et al., 2015 ▸). In our model, the electron density for Pro253 is well defined (Fig. 6 ▸ d) and the B factors are lower than average, suggesting that this region of the peptide is less flexible. The preceding amino acid is a glycine and therefore a reasonable prediction is that mutations at Pro253 significantly influence the flexibility of this region. Thermodynamically speaking, the peptide bond preceding a proline residue may adopt either the cis or trans configuration with equal favorability. Pro253 is also close to the proposed dimer interface and therefore mutations here could potentially influence the stability of this dimer.

In contrast to Pro253, Gly413 is found in a loop region with slightly higher than average B factors (Fig. 6 ▸ e), consistent with the presence of two glycine residues (Gly412 and Gly413). A homozygous missense variant p.G413V (c.1238G>T) was identified in the POMGNT2 gene in fibroblast cells that were taken from a patient with WWS (Yoshida-Moriguchi et al., 2013 ▸). This glycine residue is at the beginning of an interesting α-turn near a region of the protein model. Considering the higher than average B factors, this amino acid may be important for enzyme dynamics.

A homozygous 1333 C–T transition in the POMGNT2 gene results in an Arg445-to-Ter (R445*) substitution and premature protein truncation (Manzini et al., 2012 ▸). The patient with this mutation had classic features of WWS. While Arg445 is not involved in a salt bridge, this residue is near the C-terminal end of an α-helix and forms a hydrogen bond to the C-terminal cap, Asn448 (Fig. 6 ▸ f). Mutations here will influence the stability of this helix.

To further investigate the R158H, G413V and R445* variants of POMGNT2, constructs expressing GFP-tagged, secreted, soluble versions of these variants were made and transfected into HEK293F cells, a system that we have previously used to express and characterize multiple glycosyltransferases, including POMGNT2 (Halmo et al., 2017 ▸). While all of the variants showed transcriptional expression at the mRNA level, no enzyme or detectable level of activity was observed for any of the variants, unlike the wild-type enzyme (data not shown). However, if the loss of function is in fact simply due to protein misfolding, then why are the variants clustered in and around a single domain? In light of the structural information presented here, the genotype–phenotype relationship may need to be re-evaluated. Regardless, patients carrying mutations that encode the R158H, G413V and R445* POMGNT2 variants all have the most severe WWS muscular dystrophy, while patients with the M165T and P253L variants, which have previously been shown to have partial activity (Endo et al., 2015 ▸), display the milder LGMD. Taken together, these data suggest that the POMGNT2 activity threshold for a severe phenotype may simply reside well below our detection limits for functional enzymes.

3.5. ‘Scars’ left behind from standard N-glycosylation sites  

POMGNT2 is produced in human cells and is subject to physiologically relevant post-translational modifications, including extensive glycosylation. The enzyme was therefore treated with endoglycosidase F1 and subjected to further purification prior to crystallization. This process is imperfect in that not all of the post-translationally modified sites may be completely cleaved, thus presenting an additional challenge for crystallization trials. Upon careful model building and refinement, an explanation for the orthorhombic crystals became clear. Specifically, at least three different asparagine residues were differentially glycosylated. The location of these sites is highlighted by the N-acetylglucosamine residues represented as spheres in Fig. 6 ▸ and includes asparagine residues Asn99, Asn337 and Asn543 (see also Supplementary Figs. S2–S4). Interestingly, the N-acetylglucosamine moiety on Asn543 is involved in a crystallization contact with Trp409 (Supplementary Fig. S4). Specifically, the hydrophobic face of the glucose ring, consisting of the aliphatic protons, is stacked against the planar indole ring of the tryptophan side chain. This type of interaction has been predicted to be favorable and weak, corresponding to approximately −0.5 to −1.0 kcal mol⁻¹, based on computer modeling (Laughrey et al., 2008 ▸). Moreover, these carbohydrate–π interactions have been investigated using a variety of analytical techniques and have been proposed to play a key role in substrate recognition (Laughrey et al., 2008 ▸; Chen et al., 2013 ▸).

4. Discussion  

This work presents the first series of structures of POMGNT2, the human enzyme responsible for the first committed step towards functional core M3 glycosylation of α-dystroglycan. Even though both enzymes catalyze the addition of N-acetylglucosamine (a β-1,2 versus a β-1,4 linkage), POMGNT2 has been reported to be significantly more sequence-specific when catalyzing O-mannosylation in vitro (Halmo et al., 2017 ▸). However, mutations in both enzymes have been identified as causal of secondary dystroglycanopathies. In both cases, the majority of the disease-related mutations cluster around the active-site cleft or domain 2 of the catalytic portion of POMGNT2. These data are inconsistent with the prevailing wisdom that the disease-causing variants are simply misfolded, inactive, variants. In addition, and unlike POMGNT1, POMGNT2 does not contain the DxD metal-binding motif; instead, the active site has a pair of arginine side chains that are most likely to have a role in stabilizing the phosphate charges of the UDP-GlcNAc substrate. Ultimately, further structural analysis of POMGNT2 complexed with a non­hydrolyzable donor and/or various acceptor molecules will facilitate a more detailed discussion of enzyme mechanism and substrate selectivity. However, the data reported here, combined with variants of the enzyme identified in the clinic, provide significant new insights into the enzyme mechanism. This includes the identification of a catalytic base (His345) and residues in the immediate region that could play a role in acceptor recognition (Tyr374, Tyr377, Trp442, Ile446 and Met165).

Another unique aspect of the POMGNT2 structure is the lack of structural homology with any known glycosyltransferases, an observation that is consistent with a unique role for POMGNT2 as a ‘gatekeeper’ for O-mannosylated sites in human proteins. In fact, the closest structural alignments, as determined by DALI, consistently give high r.m.s.d. values for alignments involving less than 200 amino acids. Essentially, the best alignments cover at most one or two helices near or within the active site and a few strands that are similarly located. Moreover, the r.m.s.d. values for these few elements of secondary structure are still greater than 3 Å. In addition to the comparisons with POMGNT1 discussed above, additional support for the assignment of His345 as the catalytic base in POMGNT2 comes from the loss of function associated with the H345D variant. The rationale for this mutation was based on the highest DALI hit for the catalytic domain of POMGT2 being a heptosyltransferase named TibC from enterotoxigenic E. coli. Structures of TibC with substrates bound are available and a DALI alignment placed His345 of POMGNT2 within 4 Å of the sugar moiety in the TibC model (Fig. 4 ▸; PDB entry 4rb4; Yao et al., 2014 ▸).

In addition to the structural features discussed above, POMGNT2 also contains a large loop that partially covers the active-site cleft opposite the dimer interface. The fact that this loop contains a disulfide bond between two adjacent amino acids, Cys436 and Cys437 (Supplementary Fig. S1), may indicate a mechanism for redox sensing. Consistent with this observation, POMGNT2 is expressed and functional in the endoplasmic reticulum, which is a slightly more oxidizing environment. However, this is a striking observation considering the proximity to the UDP-binding site. Undoubtedly, reduction of this disulfide will influence the local structure and accessibility to the active site through an increase in the flexibility of this loop. Whether or not this could play an additional role in substrate specificity and the curious selectivity that is observed for POMGNT2 remains to be established.

In this work, we have determined the first series of structures of POMGNT2, a human glycosyltransferase with critical physiological functions. Our analysis provides significant new insights into the mechanism and also provides a rational basis for mutations that have been observed in the clinic. Future work is ongoing that will test the mechanistic hypotheses presented here.

Supplementary Material

PDB reference: UDP-bound POMGNT2, 6xfi

PDB reference: UDP-free POMGNT2, 6ix2

Funding Statement

This work was funded by National Institutes of Health, National Institute of General Medical Sciences grants GM111939 and S10 OD021762.