Main

In the eukaryotic cell, N-glycosylation of secretory proteins is an essential process that involves the transfer of a high-mannose glycan (GlcNAc2Man9Glc3) from a dolichylpyrophosphate (Dol-PP) carrier, catalysed by oligosaccharyltransferase1. The biosynthesis of this donor substrate is a sequential process that is initiated in the cytoplasm, requires the flipping of the GlcNAc2Man5-containing intermediate, and is completed on the luminal side of the ER, where the transfer of the final seven hexoses occurs2 (Fig. 1a). For the cytoplasmic reactions, the donor substrates are soluble, nucleotide-activated sugars. By contrast, the ER-luminal reactions depend on the membrane-embedded donor substrates dolichylphosphate-mannose (Dol-P-Man) and dolichylphosphate-glucose (Dol-P-Glc)2. ALG6 transfers the first of three glucose moieties onto the pre-assembled GlcNAc2Man9 glycan. Following the identification5 and cloning6 of the ALG6 locus, it was shown that deficiencies in ALG6 are a frequent cause of congenital disorders of glycosylation (CDGs)7, in which patients generally have hypo-glycosylated serum glycoproteins8,9. To understand the mechanism of ALG6, we sought to obtain high-resolution structural insight.

Fig. 1: Functional analysis of ALG6.
figure 1

a, Schematic of eukaryotic LLO biosynthesis and glycan transfer by oligosaccharyltransferase (OST)2,36. Enzyme names are indicated above the reaction arrows. The curved arrows across the ER membrane indicate flipping reactions. b, Tricine gel-based analysis of chemo-enzymatic LLO extension and of ALG6-catalysed glucose transfer. Glycans were transferred onto a fluorescent peptide using purified T. brucei STT3A enzyme11. The glycan structures are indicated schematically above the lanes. Note that α-1,2-mannosidase is unable to remove mannoses from the A-branch of GlcNAc2Man9Glc, demonstrating the presence of a glucose moiety. c, Synthesis scheme of Dol-P-hexose analogues including Dol25-P-Man (5) and Dol25-P-Glc (6). Note that the citronellyl-farnesyl moiety of Dol25 faithfully reflects the stereochemistry of native dolichol. DMAP, 4-dimethylaminopyridine.

Full size image

In vitro activity of purified yeast ALG6

Neither the donor nor the acceptor substrate of ALG6 is commercially available. To generate the acceptor substrate, we started from our previously developed10 method for chemo-enzymatic synthesis and analysis of Dol25-PP-GlcNAc2Man5, where Dol25 refers to a citronellyl-farnesyl moiety. The lipid-linked glycans were analysed by tricine gel analysis following transfer onto a fluorescent acceptor peptide catalysed by the eukaryotic, single-subunit oligosaccharyltransferase STT3A from Trypanosoma brucei10 (Fig. 1b). To append mannose moieties to Dol25-PP-GlcNAc2Man5, we heterologously expressed and purified the membrane-integral enzymes ALG3, ALG9 and ALG12 (Supplementary Fig. 1a) and synthesized the required mannose donor analogue Dol25-P-Man (Fig. 1c). Using a stepwise procedure, we extended Dol25-PP-GlcNAc2Man5 to the ALG6 acceptor substrate Dol25-PP-GlcNAc2Man9 (Fig. 1b). We also synthesized the required donor substrate Dol25-P-Glc de novo (Fig. 1c). Incubation of ALG6 with Dol25-P-Glc and Dol25-PP-GlcNAc2Man9 led to complete conversion to Dol25-PP-GlcNAc2Man9Glc (Fig. 1b). To confirm the physiologically relevant glycan structure, the peptide-attached glycans were incubated with α-1,2-mannosidase, which showed that ALG6-catalysed addition of glucose protected the A-branch of the Man9-containing glycan from mannosidase activity (Fig. 1b). ALG6 was found to be active in different detergents as well as reconstituted in lipid nanodiscs (Extended Data Fig. 1a).

Notably, our chemo-enzymatic approach enabled us to synthesize every lipid-linked oligosaccharide (LLO) intermediate of the biosynthetic pathway (Extended Data Fig. 1b). Our LLO analogues are readily accepted by both the single-subunit OST enzyme from T. brucei11 and the octameric yeast OST enzyme12.

Determination of the structure of ALG6

We used phage display and a synthetic library of antigen-binding Fab fragments13 to identify a conformational binder (6AG9-Fab) that increased the size and thermostability of detergent-solubilized ALG6 (Fig. 2a, Supplementary Fig. 2a, b) but did not interfere with its catalytic activity (Fig. 2b). The increased particle mass and size (Extended Data Fig. 2) aided particle alignment, and the electron microscopy (EM) reconstruction of nanodisc-reconstituted apo-ALG6 bound to 6AG9-Fab had an overall resolution of 3.0 Å (Fig. 2c, Extended Data Fig. 2, Extended Data Table 1, Supplementary Fig. 3). This allowed us to build an atomic model of ALG6 that lacked only 37 residues at the presumably flexible N terminus and 28 residues in the cytoplasmic loop that connects transmembrane helix 12 (TM12) and TM13, a segment that contains several short linear motifs including a di-arginine ER-retention motif14. 6AG9-Fab bound to a conformational, ER-luminal epitope of ALG6 and all interacting complementarity-determining region loops of the Fab fragment were well resolved (Extended Data Fig. 3). The EM maps revealed bound lipids and cholesterol hemisuccinate molecules in contact with transmembrane helices of ALG6 (Extended Data Fig. 4a).

Fig. 2: ALG6–6AG9-Fab interactions.
figure 2

a, Size-exclusion chromatography of purified ALG6 in detergent (64 kDa) in the presence and absence of 6AG9-Fab (49 kDa). b, Analysis of ALG6 activity as described in Fig. 1b but in the presence of 6AG9-Fab. 6AG9-Fab was incubated with ALG6 at a 5-fold or 15-fold molar excess to test inhibition. The lane labelled Man9-LLO is a control sample for size comparison. c, Refined and B-factor-sharpened EM map of nanodisc-reconstituted ALG6 bound to 6AG9-Fab coloured by proximity to protein chains of ALG6 (light brown), 6AG9-Fab heavy chain (red) and 6AG9-Fab light chain (grey).

Full size image

ALG6 topology and GT-C architecture

ALG6 has 14 transmembrane helices and two long loops (EL1 and EL4) that form helices in the ER lumen (Fig. 3a–c). A disulfide bridge covalently links the ER-luminal end of TM14 with the external loop EL4 that connects TM7 and TM8 (Fig. 3b, c). While the overall fold and membrane topology of ALG6 are unlike those of previously described structures, we noticed that the transmembrane arrangement and topology of the first half of the protein resemble partial structures of previously reported GT-C enzymes. We superimposed the structure of ALG6, which belongs to the GT57 subfamily according to the CAZy database15, onto those of PglB16 (GT 66), AglB17 (GT66), STT312,18 (GT66), ArnT19 (GT83) and PMT1/PMT220 (GT39). We found that the first seven transmembrane helices (including the luminal helices EL1-h1 and EL1-h2) are structurally similar, whereas the remainder of the proteins differ in fold, topology and number of transmembrane helices (Fig. 3d, Extended Data Fig. 5). This suggests that GT-C enzymes contain a structurally conserved, N-terminal module consisting of TM1–TM7, and a structurally variable C-terminal module (Fig. 3b–d). The structural similarity of the conserved module is not reflected in sequence conservation, as the primary sequences of the compared proteins are dissimilar21. However, the first external loop contains the catalytically essential aspartate residue Asp69 (see below).

Fig. 3: ALG6 structure and GT-C architecture.
figure 3

a, Cartoon representation of ALG6 in rainbow colouring (blue at the N terminus and red at the C terminus). b, Schematic representation of ALG6 topology, with transmembrane and external loop helices numbered and coloured as in a. The pink dot indicates the proposed catalytic residue Asp69. The yellow bar depicts a disulfide bond. aa, amino acids. c, Top view of the schematic representation of ALG6 coloured as in b, with numbered circles depicting transmembrane helices and cylinders depicting helices in the ER-luminal loops. d, Comparison of the topologies of ALG6 and other selected GT-C enzymes of known structure. Ribbon diagrams of the top views of the structures are shown as in c. The constant GT-C modules are shaded grey, whereas the variable GT-C modules are shaded in different colours for each structure. Transmembrane helices are numbered. Note that GT39 and GT66 contain similar variable modules (shaded green). The pink asterisks indicate the locations of the catalytic sites. The red dashed arrows indicate the entry site for the dolichol-linked donor substrates, which are depicted using standard glycobiology symbols.

Full size image

Our interpretation of a modular architecture is supported by the observation that in all published structures of GT-C enzymes, the active site and the substrate-binding cavities are located at the interface of the conserved and variable modules (pink asterisks and red dashed arrows in Fig. 3d). The architectural similarity further extends to how GT-Cs bind to their dolichyl(pyro)phosphate-linked donor substrates: the dolichol moiety invariably interacts with TM6 of the conserved GT-C module, whereas the distinct attached carbohydrates, which include single sugars or complex glycans, interact with the variable module. The donor substrates reach their respective active sites by ‘diving under’ arch-like structures formed by the external (or ER-luminal) loops that either link the conserved to the variable modules, or are part of the variable modules (Fig. 3d). These loops contain, at the minimum, a helix running parallel to the membrane but may contain entire domains (Fig. 3d). We conclude that the conserved module of GT-Cs probably served as a platform that allowed distinct functionalities to evolve, and that the modularity has allowed GT-C enzymes to accommodate different donor and acceptor substrates and to generate diverse active sites for distinct catalytic mechanisms during glycan transfer reactions.

An analysis of residues that have been reported to be implicated in ALG6-associated human CDGs7,22,23,24,25 revealed that none of them point directly at the presumed active site. Rather, they are second shell residues with respect to the active site or are located even further away (Extended Data Fig. 6). Such mutations do not fully abolish ALG6 function, but rather reduce the enzymatic activity to a level that is not yet lethal22.

Structure of Dol25-P-Glc-bound ALG6

ALG6 forms a large, hydrophilic cavity facing the ER lumen and a groove-shaped cavity facing the lipid bilayer. The residues that line these cavities are not only highly conserved among ALG6 homologues but are also in the functionally related glucosyltrasferase ALG82,26, suggesting that they are involved in substrate binding or form the active site of the enzyme (Extended Data Fig. 7). To investigate the roles of these cavities, we collected a second cryo-EM data set for the ALG6–6AG9-Fab complex (Extended Data Fig. 8, Extended Data Table 1). Here, ALG6 was solubilized in digitonin and pre-incubated with the synthetic donor substrate Dol25-P-Glc. Although the overall resolution of the structure is 3.9 Å and thus lower than that of nanodisc-reconstituted apo-ALG6, there is a well-resolved density for bound donor substrate in a lipid-exposed groove formed by TM6, TM7 and TM8 (Fig. 4a, b, Extended Data Fig. 4b). The shape of the density can clearly accommodate a Dol25-P-Glc molecule. The citronellyl-farnesyl moiety is located in the conserved, lipid-facing groove and interacts mostly with hydrophobic residues from TM6. The loop EL4 forms an arch above the groove on the ER-luminal side, resulting in a funnel-like entrance, and the Dol25 moiety bends under this arch at the level of the membrane boundary. This causes the phosphate group, which represents the leaving group of the glucose transfer reaction, to be lodged in a slightly positively charged surface region of ALG6 (Extended Data Fig. 4b).

Fig. 4: Donor substrate binding and catalytic site of ALG6.
figure 4

a, Surface representation of substrate-bound ALG6 in light brown with an EM density map for Dol25-P-Glc in the blue mesh carved to 3.0 Å, with bound Dol25-P-Glc in stick representation. The front of ALG6 was clipped for clarity. The inset shows a zoomed-in view. The red dashed arrow indicates the anomeric C1 of the glucose moiety. An ordered water molecule close to the C1 carbon is shown as a red sphere. b, Structure of the active site, with ALG6 shown in light brown ribbon representation with EL4 coloured in light blue. Acidic residues in the active site are shown in stick representation and labelled, and bound Dol25-P-Glc is depicted in stick representation, with carbon atoms coloured green. c, Analysis of substrate conversion by ALG6 variants, with mutations indicated above the lanes. WT, wild type.

Full size image

The EM density revealed a feature large enough to accommodate the glucose moiety of Dol25-P-Glc, but the quality of the map does not allow exact positioning. This is in part due to the fact that this density also covers the adjacent side chain of the conserved histidine residue His378 (Fig. 4a, b). We could nevertheless assign the orientation of the glucose moiety, with one face packing against the surface of ALG6 and the other pointing towards the solvent. We built the model such that the α-anomeric position of the C1 carbon is accessible for a nucleophilic attack by the mannose C3 hydroxyl group of the acceptor glycan. This binding mode can ensure the specificity of ALG6 for Dol-P-Glc over Dol-P-Man, as the axial C2 hydroxyl group of a mannose would cause a steric clash with the enzyme surface (Fig. 4a). The specificity of ALG6 can be demonstrated in vitro, where only glucose, but not mannose, is transferred from a Dol25-P carrier (Extended Data Fig. 9a).

We could not unambiguously identify a second dolichol-binding groove in the ALG6 structure, and multiple smaller grooves are potential candidates for binding the acceptor substrate Dol-PP-GlcNAc2Man9 (Extended Data Fig. 4c). Because several of the transmembrane helices of ALG6 are not perpendicular to the lipid bilayer, the protein appears to induce a distortion of the lipid bilayer of the nanodisc, with the lipid bilayer appearing to thin out at the side of the cavity opposite to the ‘arch’ generated by EL4 (Extended Data Fig. 4d). Such membrane thinning might facilitate the binding of the dolichol-linked donor substrate from the membrane. Deformation of a membrane caused by a protein has been reported previously in the TMEM16 scramblase, where it has been proposed to facilitate or assist in lipid flipping27.

Catalytic mechanism

Given that ALG6 is an inverting glycosyltransferase, its reaction is thought to occur via an SN2 mechanism28. This implies that the C3 hydroxyl group of the terminal A-branch mannose of Dol-PP-GlcNAc2Man9 has to be activated for a nucleophilic attack on the anomeric C1 of the glucose moiety to form an α-1-3-glycosidic bond. This is generally achieved by deprotonation of the attacking hydroxyl group by an aspartate or glutamate side chain acting as a general base, either directly or by forming catalytic dyads or triads with suitable groups such as imidazole rings (histidine), hydroxyl groups of protein side chains or ordered water molecules29,30,31. In our Dol25-P-Glc-bound ALG6 structure, we identified five acidic residues in external loops that could act as a general base (Fig. 4b): Asp69 and Asp99 from EL1, Glu306 and Asp307 from EL4 and Glu379 from EL5. We mutated all five residues to alanine and, in the case of aspartates or glutamates, to asparagine or glutamine. We also mutated His378, which is in direct contact with the glucose moiety, to investigate its potential function in substrate binding. We expressed and purified the resulting ALG6 variants (Supplementary Fig. 1b, c) and tested their in vitro activity using our coupled assay (Fig. 4c). Activity was unaffected by mutation of Glu306, Asp307 and Glu379. Mutating Asp99 to an alanine abolished ALG6 function, whereas mutation to an asparagine did not reduce enzymatic activity. This suggests that Asp99 has a functional role, possibly in acceptor substrate binding, but is not the essential catalytic base. Finally, mutating Asp69 to an alanine abolished ALG6 function, and mutating it to an asparagine strongly reduced activity. We conclude that the structurally conserved Asp69 is the catalytic base in ALG6. The corresponding aspartates (Extended Data Fig. 5b) have been demonstrated to be catalytically essential in oligosaccharyltransferases12,16,17 and the arabinosyltransferase ArnT19 and proposed to be essential in the mannosyltransferase PMT1–PMT220. Hence, an aspartate or glutamate at the N-terminal end of the helix EL-h1 may be catalytically essential in all GT-C enzymes. Mutating His378, which is in the immediate vicinity of the glucose moiety of Dol25-P-Glc, to alanine, asparagine or glutamine, strongly reduced the activity of ALG6 but did not fully abolish it. Given the location of this residue, it is most likely to be involved in the binding and orientation of the donor substrate. Although it has been shown that at least one of the ER-luminal mannose transfer reactions of the LLO biosynthesis pathway is metal-dependent32, our EM map showed no evidence for a bound divalent metal ion (M2+), nor does ALG6 contain a DXD sequence motif similar to those shown to be essential for M2+ binding in oligosaccharyltransferase12,16,20. Purified ALG6 indeed retains activity in the presence of EDTA (Extended Data Fig. 9a), demonstrating that the reaction is metal ion-independent.

To probe the reaction mechanism further, we tested whether ALG6 could process synthetic Dol25-P-Glc analogues in which the glucose moiety carried distinct substituents that were originally designed for labelling studies (Extended Data Fig. 9a). We found that substituting the C4 hydroxyl group with a 2-azido-N-ethylacetamido group abolished glucose transfer, which can be rationalized by the structural data as this would lead to a clash with the side chain of His378 (Fig. 4b). Substituting the C6 hydroxyl group of glucose with an azido group did not impair glucosyl transfer, which is in line with the structural data given that this hydroxyl group is facing the solvent and can probably rotate freely (Fig. 4a). ALG6 activity was abolished when a larger substituent (tetraethylene glycol (PEG4)-linked rhodamine 110) replaced the C6 hydroxyl group (Extended Data Fig. 9a), where it probably hinders access of the acceptor substrate to the active site (Fig. 4a, Extended Data Fig. 4b).

On the basis of our structural and functional data, we propose a three-state mechanism for ALG6 function (Extended Data Fig. 9b). The structure of substrate-bound ALG6 suggests that Dol-P-Glc binds before the Man9-containing acceptor substrate, because the glucose moiety is at the bottom of the active site cavity. We therefore propose that donor and acceptor substrates bind sequentially and that Asp69 acts as a general base that abstracts the proton of the 3-hydroxyl group of the terminal A-branch mannose of the acceptor substrate to activate it for a nucleophilic attack. This may require a conformational change during which the glucose moiety of the donor substrate moves closer to the acceptor substrate and to the catalytic residue Asp69, akin to induced fit. An analogous conformational change has been described in another GT-C enzyme: the bacterial oligosaccharyltransferase PglB33. In ALG6, an induced fit mechanism would allow the enzyme to prevent futile donor substrate hydrolysis. Upon formation of a ternary complex, the glucose moiety may not only become better ordered but might also deviate from the chair conformation, as has been previously observed in other glycosyltransferases and glycosyl hydrolases34,35.

Conclusions

Our ALG6 structure not only defines the architecture of eukaryotic ALG enzymes that are involved in ER-luminal N-glycan synthesis but also suggests an unexpected structural modularity of GT-C enzymes. The conserved GT-C module acts as both a scaffold and a substrate activator by providing a catalytically essential general base residue, whereas the variable module allows different GT-Cs to catalyse diverse glycosyl transfer reactions and to bind to distinct substrates. We identified key residues involved in ALG6 catalysis and provide the basis for future studies of the catalytic mechanism of this and related GT-C enzymes. Our approach of combining synthetic chemistry with chemo-enzymatic glycan extension presents a highly efficient and controlled way to generate LLO analogues that are not only essential for studying ALG6–substrate interactions but may also allow all enzyme-catalysed reactions of the ER-based protein N-glycosylation pathway to be studied in molecular detail.

Methods

Overexpression and purification of ALG3

A synthetic gene (GenScript) of full-length ALG3-PreScission-EYFP-1D4 from Saccharomyces cerevisiae (UniProtKB: P38179) was optimized for expression in Spodoptera frugiperda (GeneArt, Thermo Fisher). The gene was cloned into a pOET1 vector (Oxford Expression Technologies) and was expressed in S. frugiperda (Sf9) cells transfected with baculovirus that was generated with FlashBAC DNA (Oxford Expression Technologies). Cells were cultured in serum-free SF-4 Baculo Express ICM medium (AMIMED) at 27 °C. Cells were transfected at a density of 1 × 106 cells per ml and were harvested after 3 days. For purification, the cells were resuspended in 150 mM NaCl, 50 mM HEPES pH 7.4 with 0.1 mg/ml DNase I, 1:100 protease inhibitor cocktail (Sigma), 0.1 mg/ml PMSF and were lysed by dounce homogenization before solubilization by addition of 1% n-dodecyl-β-d-maltopyranosinde (DDM; Antrace), 0.2% cholesteryl hemisuccinate (CHS; Anatrace) and 10% glycerol. After 1 h of solubilization, cell debris was pelleted by centrifugation at 30,000 rpm in a type T45 Ti rotor (Beckmann). The supernatant was added to anti-1D4 antibody-coupled beads and was incubated for 1 h. The beads then were washed with 2 × 20 column volumes (cv.) of washing buffer (150 mM NaCl, 40 mM HEPES pH 7.4, 0.03% DDM and 0.006% CHS). Then the protein was eluted by incubation of washing buffer, supplemented with 0.5 mg/ml 1D4 peptide for 1 h.

Overexpression and purification of ALG6, ALG9 and ALG12

Synthetic genes (GenScript) of Flag-EYFP-PreScission-full length ALG6 (UniProtKB: Q12001), ALG9 (UniProtKB: P53868) or ALG12 (UniProtKB: P53730) from S. cerevisiae were optimized for expression in Homo sapiens (GeneArt, Thermo Fisher). Proteins were expressed, and cells were lysed, solubilized and pelleted as described above for ALG3. The supernatant was added to anti-Flag antibody-coupled resin and was incubated for 1 h. The beads then were washed with 2 × 10 cv. washing buffer (150 mM NaCl, 40 mM HEPES pH 7.4, 0.03% DDM and 0.006% CHS). Then the protein was eluted by incubation of washing buffer, supplemented with 0.2 mg/ml Flag peptide for 1 h. For functional assays, ALG6 was eluted with washing buffer supplemented with 3C-protease. For the substrate-bound structure, the immobilized ALG6 was solubilized, immobilized to Flag resin and washed as described above. Subsequently the immobilized protein was washed with 10 cv. washing buffer supplemented with 0.1% digitonin (Huberlab) and with 2 × 10 cv. digitonin washing buffer (150 mM NaCl, 40 mM HEPES pH 7.4 and 0.1% digitonin). The sample was eluted with 3C-protease and was subsequently purified by size-exclusion chromatography (SEC) in the same buffer.

ALG6-mutant generation

ALG6 mutants were generated by site-directed mutagenesis PCR and were expressed and purified by immobilization to Flag-resin and elution with peptide as describe above for wild-type ALG6. The proteins were further purified by SEC.

Nanodisc reconstitution of ALG6

A mixture of yeast polar lipids (Yeast Extract Polar, Avanti Polar Lipids) and CHS (80:20, w/w) was diluted to a concentration of 10 mM in buffer (150 mM NaCl and 40 mM HEPES pH 7.4). Lipids were solubilized by addition of 10 mM DDM and sonicated for 20 min. Purified EYFP-PreScission-ALG6 in DDM:CHS-supplemented buffer, purified MSP1D1 protein37 and solubilized lipids were mixed in a ratio of 1:6:390. After incubation for 10 min at 4 °C, the mixture was incubated for 20 min at room temperature. Detergent was removed by addition of 0.8 mg activated Biobeads (Bio-Rad) per ml of reconstitution mixture and subsequent incubation at 4 °C overnight. Biobeads were removed and the reconstitution mixture was 10× diluted in washing buffer (150 mM NaCl and 40 mM HEPES pH 7.4). To remove empty nanodiscs, the reconstitution mixture was incubated with anti-Flag antibody-coupled resin for 1 h. After washing the resin with 2 × 10 cv. washing buffer, the sample was eluted by incubation with 3C-protease.

Expression and purification of EYFP

His6-EYFP-1D4 was cloned into a modified pET19b vector (Novagen) and overexpressed in BL21 gold cells (DE3) (Stratagene). Cells were grown in Terrific Broth medium at 37 °C to an OD600 of 3.0 before induction with 1 mM isopropyl β-d-1-thiogalactopyranoside (IPTG). For protein expression, cells were then grown overnight at 25 °C and harvested by centrifugation at 10,000g.

For purification, cells were resuspended in PBS (Gibco) supplemented with 1 mM PMSF and 0.025 mg/ml DNase I. Cells were lysed by three cycles of sonication (50% amplitude and 50% duty cycle). Cell debris was pelleted by ultracentrifugation at 35,000 rpm, in a type T45 Ti rotor, for 30 min. The supernatant was incubated with Ni-NTA resin for 1 h and was subsequently washed with 10 cv. 150 mM NaCl, 50 mM Tris-HCl pH 7.5 and 10 cv. 300 mM NaCl, 40 mM Tris-HCl pH 8.0 and 50 mM imidazole, before elution with 5 cv. 300 mM NaCl, 40 mM Tris-HCl pH 8.0 and 400 mM imidazole. Immediately afterwards, the protein buffer was exchanged into 150 mM NaCl and 50 mM Tris-HCl pH 7.5.

Enzymatic biotinylation of MSP1D1

MSP1D1 fused to a C-terminal Avitag was biotinylated in biotinylation buffer (75 mM NaCl, 50 mM bicine pH 8.3, 10 mM Mg acetate, 10 mM ATP and 0.25 mM biotin) with 2 µM BirA protein, overnight at 4 °C. 3C protease (2.5 µM) was added for 6 h at 4 °C. 3C-protease and BirA were removed by reverse binding to Ni-NTA. Subsequently, the biotinylated MSP1D1 protein was desalted into 150 mM NaCl, 50 mM HEPES pH 7.4 and subsequently used for nanodisc reconstitution as described above.

Synthetic antibody generation

For phage display experiments, enzymatically biotinylated MSP1D1 was used for the nanodisc reconstitution and no reverse binding or cleavage after nanodisc reconstitution was performed. Aggregates were removed by SEC after the nanodisc reconstitution.

To validate the immobilization efficiency of the final sample, pull-down assays of the target proteins immobilized on streptavidin MagneSphere paramagnetic particles (Promega) were performed, followed by biopanning using Fab Library E13,38; DNA was kindly provided by S. Koide (University of Rochester Medical Center, Rochester, NY, USA) and the phage library was prepared by S. Mukherjee (University of Chicago, Chicago, IL, USA). Five subsequent rounds of selection were performed in 25 mM HEPES pH 7.4 and 150 mM NaCl (selection buffer) supplemented with 1% bovine serum albumin (BSA)39,40.

The phage display selection was performed according to published protocols39,41. In brief, in the first round, 400 nM Flag-EYFP-Prescission-ALG6 in nanodiscs was immobilized onto paramagnetic particles for biopanning, followed by three washes and resuspension of the particles directly in log-phase Escherichia coli XL-1 blue cells for infection. Phages were amplified in 2×YT media supplemented with ampicillin (100 µg/ml) and M13-KO7 helper phage (109 p.f.u./ml). From the second to the fifth round, sorting was performed using a Kingfisher magnetic beads handler (Thermo Fisher) with decreasing concentration of target protein: 200 nM for the second round, 100 nM for the third round, 60 nM for the fourth round and 30 nM for the last round. Phage elution was done by disrupting immobilized nanodiscs with 1% Fos-choline 12 in selection buffer. In addition, for rounds 2–5, each phage pool was negatively selected against streptavidin-coated particles and then used as an input for the next round. Throughout the entire selection process, soluble competitors were included in excess: 2 µM non-biotinytaled empty nanodiscs (same reconstitution procedure as for ALG6, without the addition of ALG6 and using non-biotinylated MSP1D1) and 2 µM purified EYFP protein. Phage pools from the fourth and fifth round were used for binder screening.

Synthetic antibody screening

Single-point phage ELISA was performed to screen and validate the binding affinities of selected binders39. Briefly, E. coli XL-1 blue were transfected with selected phage pools and plated on LB/agar plates with ampicillin. Colonies were used to inoculate 2×YT media with 100 µg/ml ampicillin and 109 p.f.u./ml M13-KO7 helper phage. Overnight amplified cultures were pelleted by centrifugation and tenfold diluted supernatants were used for ELISA. For this purpose, 96-well plates coated with 2 µg/ml NeutrAvidin (Thermo Fisher) were blocked with selection buffer with 1% BSA, and Flag-EYFP-Prescission-ALG6 protein reconstituted in biotinylated nanodiscs (50 nM) was immobilized in each well. Subsequent ELISA assays were performed in selection buffer with 2% BSA. Diluted phage particles were assayed against the target protein by using an HRP-conjugated anti-M13 monoclonal antibody (GE Healthcare). Detection was done with TMB substrate (Thermo Fisher) quenched with 1.0 M HCl after 3 min and subsequent absorbance measurement at 450 nm. Biotinylated empty nanodiscs were used as a control and only target-specific binders were selected.

Selected binders were sequenced at the University of Chicago Comprehensive Cancer Center DNA Sequencing facility and cloned into a pRH2.2 vector (kind gift of S. Sidhu (University of Toronto, Toronto, ON, Canada)) using the In-Fusion Cloning kit (Takara). The antigen-binding fragments were expressed in E. coli BL21 gold cells and purified by Protein A chromatography and ion-exchange chromatography as described previously42.

Subsequently, multi-point protein ELISA was performed to estimate the binding affinity. ELISA plate preparation and target immobilization were done similar to single-point ELISA, followed by purified Fabs in threefold dilutions (starting from 3 µM) assayed against target protein by HRP-conjugated mouse anti-human IgG F(ab’)2 monoclonal antibody (Jackson). Detection was done as described above. Data points were plotted assuming sigmoidal dose response and EC50 values were calculated for affinity estimation.

Fab-binding characterization using SEC

For SEC binding analysis, 3C-protease-cleaved ALG6 in DDM:CHS-supplemented buffer was incubated with fourfold molar excess of Fab for 30 min on ice and subsequently analysed by SEC. Peak fractions were subsequently analysed by SDS–PAGE.

Thermostability experiments were performed as described previously43, with 3C-protease-cleaved ALG6 in DDM:CHS-supplemented buffer pre-incubated with threefold molar excess of Fab for 30 min on ice and by measuring A280 instead of fluorescence during SEC.

EM sample preparation

For the apo structure, ALG6 in MSP1D1 lipid nanodiscs was incubated with 6AG9-Fab in 1:2 molar ratio for 1 h at 4 °C and subsequently purified by SEC. Pooled peak fractions were concentrated to 0.33 mg/ml. Quantifoil holey carbon grids, Cu, R 1.2/1.3, 300 mesh, were glow discharged for 45 s, 25 mA using a PELCO easiGLOW glow discharger. Sample (4 µl) was applied to the cryo-EM grids and blotted for 1–2 s before plunge freezing in a liquid ethane–propane mixture with a Vitrobot Mark IV (Thermo Fisher Scientific) operated at 4 °C and 100% humidity.

For the substrate-bound structure, ALG6 in digitonin-supplemented buffer at 7.2 mg/ml was incubated with 1.5-fold excess of 6AG9-Fab for 1 h at 4 °C. Subsequently, the sample was purified by SEC and peak fractions were pooled and were concentrated to 6.3 mg/ml. The sample was supplemented with a final concentration of 150 µM Dol25-P-Glc and incubated for 1.5 h before grid preparation. Grids were prepared the same way as for the nanodisc sample, but with a 3.5-s blotting time.

EM data collection

Data were recorded on a Titan Krios electron microscope (Thermo Fisher Scientific) operated at 300 kV, equipped with a Gatan BioQuantum 1967 filter with a slit width of 20 eV and a Gatan K3 camera. Movies were collected semi-automatically using SerialEM44 at a nominal magnification of 105,000 and a pixel size of 0.42 Å per pixel in super-resolution mode. The defocus range was −0.6 to −2.8 µm. Each movie contained 40 images per stack with a dose per frame of 2.3 electrons/Å2.

EM data processing, model building and refinement

For the apo structure of ALG6 in nanodiscs, 6,801 movies were corrected for beam-induced motion using MotionCor245. The micrographs were visually inspected and 5,143 were used for further processing in RELION 3.046. Contrast transfer function (CTF) was estimated using gCTF47. Particles (n = 3,437,475) were auto-picked and extracted with threefold binning to a pixel size of 2.52 Å/pixel. After two subsequent rounds of 2D classification, 445,702 particles were used to generate an initial model. The particles were subjected to two subsequent rounds of 3D classification and the remaining 171,764 particles were subjected to 3D refinement. Subsequently, the particles were re-extracted at 0.84 Å/pixel and subjected to 3D refinement yielding a map at 3.2 Å resolution. Subsequent particle polishing and 3D refinement lead to a map at 3.1 Å resolution. Application of a tight mask that masked out most of the flexible constant domain yielded a map at 3.0 Å resolution.

For the substrate-bound structure in digitonin, 6,187 movies were corrected for beam-induced motion using MotionCor245. The micrographs were visually inspected and 4,000 were used for further processing in RELION 3.046. CTF was estimated using gCTF47. Particles (n = 995,878) were auto-picked and extracted with threefold binning to a pixel size of 2.52 Å/pixel. After two subsequent rounds of 2D classification, 287,977 particles were selected and were used to calculate an initial model. After one round of 3D classification, 115,590 particles were selected for one round of 3D refinement and subsequent re-extraction at 0.84 Å/pixel. Subsequent 3D refinement yielded a final map at 3.9 Å.

Model building was performed in Coot48. Most of the Fab, except for the binding interface, was built based on a pre-existing model at high resolution (Protein Data Bank (PDB) ID 5UCB). The structures were refined in PHENIX49 and ligand files were generated in PHENIX eLBOW50. The correlation between model and map was validated using PHENIX mtriage51.

Figure preparation and data analysis

Local resolution estimations were calculated in ResMap52. Figures were prepared in PyMol53 and UCSF Chimera54 and UCSF ChimeraX55. Graphs were generated in GraphPad Prism 8. Protein sequence alignments were performed using CLUSTAL W56, Clustal Omega57 and EMBOSS Needle58. Figure 3d was generated using the following structures: S. cerevisiae ALG6 (this study), Campylobacter lari PglB (PDB 3RCE), Cupriavidus metallidurans ArnT (PDB 5F15) and S. cerevisiae PMT1 (PDB 6P25).

Chemical synthesis of substrate analogues

β-Mannosyl (3) and β-glucosyl (4) triesters were synthesized starting from 2,3,4,6-tetracetylated glycosides with an optimized procedure based on a previously described method59. Diphenyl-protected glycosyl phosphates were readily hydrogenated and coupled to farnesylcitronellol in presence of trichloroacetonitrile. Acetyl deprotection gave β-glycosyl phosphoisoprenoids Dol25-P-Man (5) and Dol25-P-Glc (6). To generate potentially clickable analogues of 6, an azide was introduced in the 6 position following known procedure60, and in the 4 position introducing a short linker via O-alkylation of orthogonal-protected glucopyranoside61 with bromoacetonitrile followed by reduction of the nitrile with BH3•SMe2 (ref. 62) and acylation of the primary amine. β-Phosphorylation of the azide-bearing lactols was performed using diallylphosphoryl chloride63 to avoid hydrogenation steps. The desired configuration was confirmed by ROESY (Supplementary Methods). Diallyl phosphates were deprotected, converted to the phosphoric acid and coupled with farnesylcitronellol11, as described above. More detailed procedures are described in Supplementary Methods.

Synthesis and purification of Dol25-PP-GlcNAc2Man9

Dol25-PP-GlcNAc2Man9 was produced as previously described10. Dol25-PP-GlcNAc2Man5 (20 µM) and Dol25-P-Man (120 µM) were mixed in 150 mM NaCl, 50 mM HEPES pH 7.4, 0.03% DDM and 0.006% CHS. Purified ALG3 (1.5 µM) was added and the mixture was incubated overnight at 4 °C. The reaction was inactivated by heating for 10 min at 95 °C. After cooling down to 4 °C, the reaction was supplemented with 10 mM MgCl2, and 1.5 µM ALG9 and 1.5 µM ALG12 were added and incubated overnight at 4 °C. For purification of Dol25-PP-GlcNAc2Man9, the reaction was lyophilized followed by a first extraction with CHCl3:MeOH (2:1, v/v) to remove contaminants of smaller LLO species. The remaining pellet was then subjected to extraction with CHCl3:MeOH:H2O (10:10:3, v/v/v). Subsequently, the extracted substrate was dried under N2 and was resuspended in 150 mM NaCl and 40 mM HEPES pH 7.4. The yield was quantified as described previously10.

Note that a similar synthetic route for in vitro synthesis of an LLO analogue was previously reported64. However, the lipid carrier in that study was a phytanol moiety instead of dolichol.

In vitro glycosyl transfer assays

Glycosyl transfer reactions for ALG1, ALG2 and ALG11 were performed as previously described10. Glycosyl transfer reactions for ALG3, ALG9 and ALG12 were performed in reaction buffer (150 mM NaCl, 50 mM HEPES pH 7.4, 0.03% DDM and 0.006% CHS) supplemented with 10–20 µM acceptor substrate (starting from purified Dol25-PP-GlcNAc2Man5 as the acceptor substrate with sequential addition and heat inactivation of ALG enzymes) and 50–120 µM donor substrate (Dol25-P-Man). Reactions for ALG9 and ALG12 were supplemented with 10 mM MgCl2. Protein concentrations were ranging from 0.5 to 2 µM. Reactions were carried out overnight at 4 °C. Reactions were stopped by heating at 98 °C for 10 min. This ensured that, during the ALG-mediated glycosyl transfer reactions, no other active ALG enzymes from previous reactions were present and eliminated fluorescence from EYFP fusion proteins, which otherwise would affect subsequent tricine gel analysis.

Reactions for ALG6 were carried out in reaction buffer (150 mM NaCl, 50 mM HEPES pH 7.4, 0.03% DDM and 0.006% CHS) supplemented with 10 µM purified Dol25-PP-GlcNAc2Man9 acceptor substrate, 50 µM of the respective donor substrate and 0.2 µM tag-cleaved ALG6. For reactions in the presence of EDTA or Fab, the entire reaction mixture was preincubated for 1 h on ice, before adding donor and acceptor substrates. The reactions were incubated for 6 h at 10 °C. Subsequently, the reactions were stopped by heating at 98 °C for 10 min. Reactions of ALG6 mutants were carried out as described above, at 10 °C for 24 h. All ALG reaction products were analysed by TbSTT3A-mediated oligosaccharyl-transfer and subsequent tricine gel analysis as previously described10. Owing to lower substrate specificity of TbSTT3A for Dol25-PP-GlcNAc2Man9Glc1, oligosaccharyl-transfer reactions for ALG6 reactions were performed for 12 h at 30 °C. For the oligosaccharyl-transfer reactions for ALG6 reactions that were carried out in the presence of EDTA, 30 µM MnCl2 was added, to avoid inhibition of STT3.

Statistics and reproducibility

Unless otherwise stated, ALG6 assays and tricine gel-based analyses were conducted once as shown in the figures.

Reporting summary

Further information on research design is available in the Nature Research Reporting Summary linked to this paper.