Single-particle electron microscopy structure of UDP-glucose:glycoprotein glucosyltransferase suggests a selectivity mechanism for misfolded proteins

The enzyme UDP-glucose:glycoprotein glucosyltransferase (UGGT) mediates quality control of glycoproteins in the endoplasmic reticulum by attaching glucose to N-linked glycan of misfolded proteins. As a sensor, UGGT ensures that misfolded proteins are recognized by the lectin chaperones and do not leave the secretory pathway. The structure of UGGT and the mechanism of its selectivity for misfolded proteins have been unknown for 25 years. Here, we used negative-stain electron microscopy and small-angle X-ray scattering to determine the structure of UGGT from Drosophila melanogaster at 18-Å resolution. Three-dimensional reconstructions revealed a cage-like structure with a large central cavity. Particle classification revealed flexibility that precluded determination of a high-resolution structure. Introduction of biotinylation sites into a fungal UGGT expressed in Escherichia coli allowed identification of the catalytic and first thioredoxin-like domains. We also used hydrogen-deuterium exchange mass spectrometry to map the binding site of an accessory protein, Sep15, to the first thioredoxin-like domain. The UGGT structural features identified suggest that the central cavity contains the catalytic site and is lined with hydrophobic surfaces. This enhances the binding of misfolded substrates with exposed hydrophobic residues and excludes folded proteins with hydrophilic surfaces. In conclusion, we have determined the UGGT structure, which enabled us to develop a plausible functional model of the mechanism for UGGT's selectivity for misfolded glycoproteins.

The enzyme UDP-glucose:glycoprotein glucosyltransferase (UGGT) mediates quality control of glycoproteins in the endoplasmic reticulum by attaching glucose to N-linked glycan of misfolded proteins. As a sensor, UGGT ensures that misfolded proteins are recognized by the lectin chaperones and do not leave the secretory pathway. The structure of UGGT and the mechanism of its selectivity for misfolded proteins have been unknown for 25 years. Here, we used negative-stain electron microscopy and small-angle X-ray scattering to determine the structure of UGGT from Drosophila melanogaster at 18-Å resolution. Three-dimensional reconstructions revealed a cagelike structure with a large central cavity. Particle classification revealed flexibility that precluded determination of a highresolution structure. Introduction of biotinylation sites into a fungal UGGT expressed in Escherichia coli allowed identification of the catalytic and first thioredoxin-like domains. We also used hydrogen-deuterium exchange mass spectrometry to map the binding site of an accessory protein, Sep15, to the first thioredoxin-like domain. The UGGT structural features identified suggest that the central cavity contains the catalytic site and is lined with hydrophobic surfaces. This enhances the binding of misfolded substrates with exposed hydrophobic residues and excludes folded proteins with hydrophilic surfaces. In conclusion, we have determined the UGGT structure, which enabled us to develop a plausible functional model of the mechanism for UGGT's selectivity for misfolded glycoproteins.
The folding of glycoproteins that transit through the endoplasmic reticulum (ER) 3 is reflected in the structure of their N-linked glycan. Newly synthesized proteins are modified with the intact glycan Glc 3 Man 9 GlcNAc 2 , which is then trimmed by glucosidases I and II to generate a monoglucosylated form. This trimmed form is recognized by the ER-specific lectin chaperones calnexin and calreticulin, which recruit additional chaperones to fold the nascent protein (1)(2)(3)(4)(5). Upon release, the terminal glucose is cleaved by glucosidase II, producing a shortened glycan, Man 9 GlcNAc 2 , which can no longer bind the lectin chaperones. Incompletely folded or misfolded proteins after the first cycle of folding are recognized by UDP-glucose:glycoprotein glucosyltransferase (UGGT), which adds back a glucose residue to regenerate the monoglucosylated form for additional rounds of lectin chaperone-assisted refolding (6,7). By glucosylating misfolded proteins, UGGT plays an important role in preventing misfolded proteins from exiting the ER.
In addition to its role as a sensor of protein folding, UGGT contributes to the maturation of the major histocompatibility complex of class 1 molecules (MHC1), ensuring that a highaffinity peptide has been loaded onto the MHC1 heavy chains (28,29). The MHC1 is composed of MHC-encoded heavy chains, which are glycosylated, and a ␤ 2 -microglobulin domain. The assembly of the complete MHC1 involves many components of the ER folding pathways. UGGT scans the peptide-MHC1 interaction through glucosylating the heavy chain and directs the MHC1 for reassociation with the calreticulin-ERp57-tapasin complex for a new peptide to be loaded onto the heavy chain-binding groove. UGGT ensures that only high-affinity peptides are loaded onto the MHC1.
A mechanistic understanding of UGGT activity has been limited as the protein has proven to be a difficult crystallization target. Sequence analysis and biochemical data have suggested that it has five domains. The N-terminal portion consists of three thioredoxin-like domains followed by a ␤-strand-rich domain and the catalytic glucosyltransferase domain. The catalytic domain is the most strongly conserved across species (9,30). The non-catalytic domains are less conserved with 20 -40% sequence identity among species. The only crystal structure is of the third thioredoxin-like domain from Chaetomium thermophilum (31).
In this report, we characterize the structure of UGGT from Drosophila melanogaster and Penicillium chrysogenum. Solution analysis by small-angle X-ray and light scattering confirmed that D. melanogaster UGGT (DmUGGT) is monomeric and forms a complex with Sep15 in a 1:1 ratio. We identified the Sep15-binding site on UGGT using hydrogen-deuterium exchange with mass spectrometry (HDX-MS) and mutagenesis. Small-angle X-ray scattering (SAXS) and single-particle electron microscopy (EM) revealed that UGGT adopts an open cage shape with a large central cavity. We identified the domains in an EM model of P. chrysogenum UGGT (PcUGGT) using biotinylation sites decorated with monovalent streptavidin. The three-dimensional model of UGGT allowed us to propose a mechanism for substrate recognition and selection.

SAXS analysis
We chose to study UGGT from two sources. DmUGGT had been previously characterized biochemically and is easily purified from Sf9 insect cells due to incorporation of an N-terminal melittin signal peptide that secretes the protein into the culture medium (2,32,33). We also studied the fungal PcUGGT, which could be expressed in Escherichia coli and purified in high yield (supplemental Fig. S1). The two proteins are 44 and 34% identical to human UGGT with the strongest similarity in the C-terminal catalytic domain.
To characterize the proteins, we turned to SAXS and EM, two structural techniques that have been used successfully to study proteins that are difficult to crystallize. SAXS analysis of DmUGGT and PcUGGT confirmed that the proteins were monomeric with no aggregation detected in Guinier plots (Fig.  1A). Kratky plots of the scattering data showed an elongated bell curve that returned to zero, which is characteristic of a folded protein (supplemental Fig. S2). The radius of gyration for Dm-and PcUGGT, calculated from the slope of the Guinier plots at low angle, was estimated to 4.9 nm and independent of protein concentration. The distance distribution curves for both species showed similar bell-shaped curves consistent with D max values in the range of 100 -140 Å (Fig. 1B).
We used the SAXS data to generate ab initio bead models for DmUGGT and PcUGGT using the ATSAS suite of programs (34). The two models were very similar and displayed a C-shaped structure with two distinct regions or lobes (Fig. 1, C and D). The larger lobe was 80 ϫ 70 ϫ 50 Å with a depression at its center and was connected to an elongated lobe of 70 ϫ 40 ϫ 40 Å. The distance separating the lobes was strongly dependent on the D max used for the structure calculations. The structures calculated with larger D max values showed a T-shaped conformation, whereas in the smaller D max structures the elongated lobe was curved with a distance of roughly 35 Å between the ends of the lobes (supplemental Fig. S3).

EM analysis
We next carried out negative-stain EM studies of DmUGGT and PcUGGT (Table 1). Single-particle analysis using uranyl formate stain showed that the protein particles had no preferred orientation on the grid, and the size distribution of the particles was very homogeneous, key characteristics to facilitate three-dimensional reconstructions. Initial analysis was done in EMAN2.0 (35,36) to create two-dimensional reference-free class averages. The resulting models of DmUGGT and PcUGGT after initial model constructions and three-dimensional refinements showed high consistencies and similarities. Back-projections of the three-dimensional maps agreed with the two-dimensional class averages observed experimentally. The refined models, determined to a resolution of ϳ25 Å using EMAN2.0, exhibited a distinctive C-shape when viewed from the side and a ring shape from the top ( Fig. 1 and supplemental Figs. S4 and S5).
To confirm our EM model, we processed the data using different software. Briefly, the boxed particles of DmUGGT were re-extracted and further selected through two-dimensional alignments in RELION2.0 (37). The new particle subset was imported into Scipion (38) for further two-dimensional classification using RELION and CL2D (39). Initial volumes were estimated using a different method (40) followed by RELION three-dimensional classifications and RELION autorefinement (Fig. 2, A and B). No significant differences were observed in the classes obtained using the different processing software, but the DmUGGT structure determined using Scipion provided slightly improved resolution (Fig. 2C). The higher resolution structure confirmed the claw-like architecture composed of a ring-shaped lobe connected to an arm-like lobe (Fig. 2, B and D).
As an additional check, we calculated the theoretical X-ray scattering curve from the Scipion DmUGGT EM model and compared it with the experimental data (supplemental Fig. S6). A good fit was observed at small angles (q Ͻ 0.06 Å Ϫ1 ), but the curves diverged at larger angles, which could result from distortion during dehydration of the EM samples or conformational heterogeneity. To address the latter, we re-examined the Scipion three-dimensional classifications and detected the presence of multiple particle classes (Fig. 2D). The classes were very similar but showed distinct and reproducible position and density differences. We detected multiple classes irrespective of the classification parameters used, which suggests that flexibility is an intrinsic property of the protein. To visualize the differences in the structures, we generated morphings between the classes (supplemental Movies 1 and 2). The DmUGGT classes were evenly populated with small incremental changes, suggesting continuous movement or flexibility in the structure. Despite these differences, the organization of UGGT as two lobes with a central cavity was consistent across all structural models (Fig. 2D).

Domain identification
To identify the features in the EM and SAXS maps, we used the DOLORS strategy of labeling with monovalent streptavidin (mSA) using inserted biotinylation sites (41). We chose to apply the strategy to PcUGGT because its facile expression allowed screening of a large number of insertions. Building on the bioinformatics analysis of UGGT from C. thermophilum (31), we examined the primary sequence of PcUGGT and selected 16 putative loop regions for insertion of biotinylation sites (Fig.  3A). After expression, purification, and mSA labeling, we obtained five PcUGGT-biotin-mSA samples for EM analysis (supplemental Fig. S7).
Two of the streptavidin-labeled PcUGGT samples gave clearly defined additional density that allowed us to assign parts of the EM map to specific domains. PcUGGT-biotin-mSA-m3 (mutant 3) contained a biotinylation site inserted after lysine 261 in the first thioredoxin domain, and PcUGGT-biotin-mSA-m15 (mutant 15) contained a biotinylation site inserted after aspartic acid 1377 in the catalytic domain (supplemental Fig.  S7). The reference-free classifications of the boxed particles of the streptavidin-labeled mutants were very similar to the classes observed for PcUGGT with additional density consistent in size with a monovalent streptavidin label (Fig. 3, B and C). The structures were determined to resolutions of 20 Å (Table 1 and supplemental Fig. S8).
The EM map of PcUGGT-biotin-mSA-m3 showed additional density on the ring lobe and identifies this as the N-terminal portion of UGGT (Fig. 3D). We posit that the ring is

Electron microscopy of UGGT
formed by the three thioredoxin-like domains and that the armlike lobe is formed by the ␤-strand-rich and catalytic domains.
In agreement with this, the EM map of PcUGGT-biotin-mSA-m15 showed distinct additional density at the end of the armlike lobe. Localization of the catalytic domain at the end of the arm allows us to determine that the ␤-strand-rich domain is in  the density connecting the catalytic domain and N-terminal lobe.

Identification of Sep15-binding site
In cells, UGGT is tightly associated with Sep15, a small 15-kDa thioredoxin-like protein. Isothermal titration calorimetry experiments measured the affinity to be 20 nM (27). To characterize the site of its interaction on UGGT, we expressed and purified Sep15 from E. coli cells. The complex of DmUGGT and Sep15 was prepared by mixing an excess of Sep15 with purified DmUGGT followed by gel filtration chromatography to separate the complex from excess Sep15. Formation of the complex could be observed by a shift in the elution time, an increase in the molecular weight measured by multiangle light scattering, and SDS-PAGE analysis (supplemental Fig. S9).
We utilized HDX-MS to identify the Sep15-binding site on DmUGGT. The technique is based on measuring differences in the exchange of amide hydrogens in complex and free protein.
The peptide coverage in MS analysis of DmUGGT was excellent, allowing measurement of the solvent accessibility of 95% of the protein (supplemental Fig. S10). Comparison of the exchange rates for DmUGGT and DmUGGT-Sep15 complex showed a marked difference in a roughly 40-residue region predicted to be in the first thioredoxin-like domain (Fig. 4A). Residues in the region displayed a significantly reduced rate of hydrogen-deuterium exchange in the presence of Sep15 ( Fig.  4B and supplemental Fig. S11). Outside of this region, there were no significant differences (supplemental Figs. S12 and S13).

Discussion
UGGT has been the target of numerous structural efforts over the last 25 years. Here, we used EM and SAXS to characterize its overall shape. The main features of the structure are very robust. We observed highly similar conformation for UGGT from two evolutionarily distant organisms with only 33% amino acid sequence identity. It is also notable that the structures calculated from the SAXS and EM data are very similar. The back-projections of the SAXS models in Fig. 1 agree well with the EM two-dimensional class averages in Fig. 2, and the overall three-dimensional shapes are similar. Ab initio SAXS modeling is not well suited for the analysis of hollow structures such as UGGT, but it nonetheless provided a strong independent confirmation of the EM maps.
The identification of the Sep15-binding site is similarly robust. The HDX-MS data are remarkably clear and identify a highly charged segment in the first thioredoxin-like domain as the binding site. The region is loosely structured in the absence of Sep15 as determined by its complete deuteration at 15 s. Homology modeling suggests the segment could adopt an ␣-helix upon binding of the Sep15. Sep15 is composed of a cysteine-rich domain that binds UGGT and a selenocysteinecontaining thioredoxin-like domain (25)(26)(27). The protein is highly conserved across eukaryotes and thought to act as a protein-disulfide reductase due to the oxidation potential of its selenocysteine (26).
Although limited to 20-Å resolution, the UGGT structure is surprisingly informative and suggests a mechanism for substrate recognition. A large number of studies have characterized the specificity of UGGT for misfolded proteins (7, 10 -22). Full activity requires both exposed hydrophobic residues and the Man 9 GlcNAc 2 glycan, which is the site of monoglucosylation and release. The two elements have to be in the same molecule, and in fact addition of deglycosylated (endoglycosidase H-treated) misfolded proteins inhibits UGGT activity (42). There is a bias toward glycans adjacent to hydrophobic residues and against bulky folded proteins. Studies with synthetic substrates composed of the acceptor glycan linked to methotrexate have shown that the addition of dihydrofolate reductase reduces activity (17). Allosteric coupling between a domain that acts a sensor of misfolded protein and the glucosyltransferase domain has been proposed, although it fails to explain the requirement for both recognition elements to be on the same molecule.
The UGGT structure suggests a simpler mechanism for the selection of misfolded proteins that is consistent with the literature and involves minimal complexity. We hypothesize that the catalytic site is located inside the UGGT cavity to sequester the glucosyltransferase activity away from glycans on folded, hydrophilic substrates (Fig. 5). We expect that the cavity will be lined with hydrophobic surfaces to favor the recruitment of misfolded proteins that have exposed hydrophobic residues. In a manner similar to the chaperonins GroEL/GroES, UGGT recruits substrates into a hydrophobic cavity. Misfolded proteins are kinetically favored through their higher rate of occupancy in the cavity. The size of the pocket favors the binding of intact, partially folded proteins over small hydrophobic glycopeptides (43), and flexibility allows UGGT to act on different misfolded proteins. This simple mechanism explains why the acceptor glycan and exposed hydrophobic segment need to be in the same molecule and why misfolded, deglycosylated proteins inhibit UGGT activity. Testing of this model will require a higher resolution structure of UGGT to identify the nature of internal surfaces of the cavity and the orientation of catalytic domain. It is interesting that the recently determined structure of the third thioredoxinlike domain from C. thermophilum UGGT revealed the existence of the extensive detergent-bound hydrophobic patch (31). Future studies are necessary to determine whether the other thioredoxin-like domains in the N-terminal lobe similarly contain hydrophobic surfaces.

DmUGGT expression and purification
DmUGGT cloned into the pFastBac1 plasmid with a N-terminal melittin secretion sequence and a C-terminal hexahistidine tag was obtained from Daniel Tessier (2). The reconstructed plasmid was transformed into DH10Bac E. coli cells, and the baculovirus DNA was produced according to the Sf9 baculovirus expression system manual using serum-free media. Sf9 cells were amplified to 2 ϫ 10 6 /ml and infected by P3 virus with a ratio of 0.2% (v/v). The Sf9 cells continued to grow for 72 h at 27°C at 85 rpm. The supernatant containing the secreted DmUGGT was separated with the cells by centrifugation at 1000 rpm for 15 min. The supernatant was loaded into the pre-equilibrated Ni-NTA gravity column. After all the supernatant flowed through, the column was washed using Buffer A (30 mM Tris-HCl, pH 7.5, 300 mM NaCl, 20 mM imidazole). The protein was eluted from the column using Buffer B (30 mM Tris-HCl, pH 7.5, 300 mM NaCl, 300 mM imidazole). DmUGGT was concentrated to less than 5 ml using a 50-kDamolecular-mass-cutoff Amicon filter by centrifugation at 4000 rpm at 4°C before being injected into a Superdex 200 prep grade HiLoad 16/600 column, which was pre-equilibrated with Buffer C (30 mM Tris-HCl, pH 7.5, 100 mM NaCl). The fractions were analyzed by SDS-PAGE.

Expression and purification of PcUGGT
P. chrysogenum UGGT was codon-optimized and cloned into pCold I vector for bacterial expression using E. coli Rosetta Gami 2. Cells were grown until A 600 nm reaches 0.6 and then induced with 0.5 mM isopropyl 1-thio-␤-D-galactopyranoside at 37°C for 4 h. Cells were harvested by centrifugation, suspended in lysis buffer (Buffer A), and lysed by sonication. The cell lysate was centrifuged at 18,000 rpm at 4°C for 60 min, and the supernatant was subjected to Ni-NTA purification. The eluted sample was further purified by anion-exchange chromatography and size-exclusion chromatography. The fractions were analyzed by SDS-PAGE.

SAXS data collection and analysis
DmUGGT at 0.5-20 mg/ml was used for data collection on an in-house SAXSess system (Anton Paar) with exposures of 6 h per sample on a CCD chip for both protein and buffer samples. The SAXS data were processed and analyzed using the ATSAS software package (34). PRIMUS was used to subtract buffer scattering curves and normalize the data by protein concentration. GNOM was used to calculate P(r) curves with different D max values. For each D max value, 20 pseudoatom (bead) mod-high affinity substrate low affinity substrate els were generated by DAMMIF and then aligned and compared in DAMSEL to determine the most probable model. DAMSUP was used to align the models with the most probable model, and the aligned models were averaged and filtered in DAMAVER to generate the final SAXS structure. For comparison of the SAXS and EM data, EM2DAM was used to calculate a theoretical X-ray scattering curve from an input EM model.

EM sample preparation and image acquisition
DmUGGT (12 ng/l) and PcUGGT (7.5 ng/l) proteins were applied onto negatively glow-discharged carbon-coated grids (400-mesh copper grids) for 60 s, and excess protein was removed by blotting using filter papers. Freshly prepared 0.75% uranyl formate, pH 5, was applied on the grids after the protein was removed for 60 s and then blotted. The grids were dried for at least 30 min before electron microscopy. Images were collected using an FEI Tecnai G2 TF20 transmission electron microscope at 200 kV equipped with a Gatan Ultrascan 4000 CCD camera (model 895) at 62,000ϫ magnification and a 1.8-Å/pixel sampling rate. Micrographs were recorded at varying defocus values (Table 1).

EM data processing and image analysis
For all the data sets, protein particles were picked using e2boxer from EMAN2.0 with a 140 ϫ 140-pixel box size. The particles were either extracted in EMAN2.0 (35,36) or extracted in RELION2.0 (37). Contrast transfer function corrections were done using CTFFind3. For EMAN2.0 processing, particles were classified using reference-free algorithm K-means, and only particles from well defined class averages were kept for further processing. The high-quality particles were used to generate initial models by the common line algorithm within EMAN2.0 and by the SIMPLE PRIME (44,45) method. We refined these models within EMAN2.0 using the gold-standard Fourier shell correlation method. UGGT models were compared using Chimera at thresholds levels giving equivalent voxel volumes. Back-projections of the models were calculated using EMAN2.0 e2project3d.py for comparison with the class averages (supplemental Figs. S4 and S5). For Scipion processing, two-dimensional classification from RELION2.0 was used to remove particles that were either false positives or belonged to classes featuring touching particles, resulting in a data set of 26,624 particles. The selected particles were imported into Scipion (38). Initial models were generated using RANSAC using the obtained two-dimensional averages (40). Three-dimensional classification used the generated initial models after low-pass filtering to 60 Å with C1 symmetry. The selected model was further processed with RELION 3D Auto-Refine in Scipion. For resolution determination, gold-standard Fourier shell correlation (FSC) was used. Back-projections were calculated and compared with the two-dimensional class averages from Xmipp CL2D. For comparison of the SAXS and EM data, EM2DAM (part of the ATSAS software suite) was used to calculate a theoretical X-ray scattering curve from the EM model.

Generation of PcUGGT-biotin mutant expression constructs
The biotin insertion sites were identified following the procedure reported previously (41). We carried out a secondary structure prediction of PcUGGT to identify as many sites as possible to introduce the 15-amino-acid AviTag insertion (LNDILEAQKIEWHEQ) as a biotinylation target across the various domains ( Fig. 3A and supplemental Fig. S7). A total of 16 sites were identified, presumably loops or unstructured linkers, distributed across the entire sequence of PcUGGT. Primers with AviTag cDNA were synthesized (Biocorp Inc.) and inserted into the pCold I vector bearing the wild-type PcUGGT sequence using a QuikChange lightning site-directed mutagenesis Kit (Agilent technologies). The sequences of PcUGGT-biotin mutants were confirmed through sequencing (Genome Quebec).

Preparation of PcUGGT-biotin monovalent streptavidin (mSA)-labeled samples
All the PcUGGT-biotin mutants were expressed in E. coli Rosetta Gami 2 as described above for wild-type PcUGGT production. Purification was also done using a similar strategy as the wild-type PcUGGT. Briefly, after the Ni-NTA affinity purification, the protein samples were subjected to in vitro biotinylation using the biotin ligase enzyme BirA (41). The successfully biotinylated mutants were then incubated with an excess of monovalent streptavidin (46) and further purified by gel filtration using a Superdex 200 16/600 column.

Negative stain electron microscopy, image processing, and model reconstructions of PcUGGT-biotin-mSA mutants
Preparation of samples for negative-stain EM and data collection was carried out using the same procedures described above for DmUGGT. Briefly, proteins at concentrations of 10 ng/l were stained with a solution of 0.75% (w/v) uranyl formate and imaged at 62,000ϫ magnification at 200 kV on the FEI Tecnai G2 TF20 transmission electron microscope. We collected roughly 100 micrographs for PcUGGT-biotin-mSA-m3, PcUGGT-biotin-mSA-m7, and PcUGGT-biotin-mSA-m15 separately. Similarly to DmUGGT processing, reference-free classification was applied to sort particles and to select the highquality particles. The negative-stain PcUGGT model was used as the initial model for refinement of the mSA-labeled mutants. The extra density observed in the mutant maps was attributed to the 50-kDa mSA label. Three-dimensional visualizations of the map were conducted in UCSF Chimera.

Sep15 expression and purification
Human Sep15 with deletion of the first 28 residues (Sep15-⌬28) was cloned into the pET29a vector with a C-terminal hexahistidine tag. The reconstructed plasmid was transformed into lobster E. coli strain. Cells were grown until A 600 nm reached 0.6 and then induced with 1 mM isopropyl 1-thio-␤-Dgalactopyranoside at 30°C for 4 h. The E. coli cells were harvested by centrifugation at 4500 rpm at 4°C for 15 min and resuspended in Buffer A supplemented with 1 mM DTT. Resuspended cells were lysed using a French pressure cell followed by centrifugation at 12,000 rpm at 4°C for 1 h. The supernatant after centrifugation was loaded into the pre-equilibrated Ni-NTA gravity column. After all the supernatant flowed through, the column was washed using Buffer A supplemented with 1 mM DTT. Sep15-⌬28 was eluted from the column using Buffer B supplemented with 1 mM DTT. The elution fraction was concentrated to less than 5 ml using a 10-kDa-molecular-masscutoff Amicon filter by centrifugation at 4000 rpm at 4°C and injected into a Superdex 75 HiLoad 16/600 column, which was pre-equilibrated with Buffer C supplemented with 1 mM DTT. The fractions were analyzed by SDS-PAGE.

Multiangle light scattering analysis
Analytical size-exclusion chromatography (SEC) of DmUGGT and the DmUGGT-Sep15-⌬28 complex was performed using a GE Healthcare 200 10/300GL column with a flow rate of 0.4 ml/min at 22°C. The SEC was coupled to a mini Dawn TREOS multiangle light scattering analysis instrument (Wyatt Technology) and an Optilab rEX (Wyatt Technology). The measurement of the Rayleigh scattering intensity as a function of the angle as well as the differential refractive index of the eluting peak in SEC can be used to determine the weight average molar mass (M w ) of eluted peaks, which can be used to estimate the oligomer states and protein complexes using the ASTRA (Wyatt Technologies) (47). The number average molar mass (M n ) was also determined to estimate the monodispersity of the peaks. BSA (4 mg/ml) was used to calibrate the system before running DmUGGT.

Hydrogen-deuterium exchange mass spectrometry
HDX-MS experiments were carried out as described (48). Briefly, stock solutions of 5 mg/ml DmUGGT or the DmUGGT-Sep15 complex were prepared in Buffer C. Similar buffer conditions were applied for the preparation of the corresponding D 2 O buffers. For the blank control, the initial dilution was made in H 2 O buffer. HDX was initiated by diluting stock DmUGGT solution 1.5:8.5 into the D 2 O-based buffer. HDX incubation periods were 15, 300, 900, and 3600 s, and the temperature was set at 25°C. HDX was quenched with chilled quenching buffer (300 mM glycine, 6 M guanidine hydrochloride, 400 mM tris(2-carboxyethyl)phosphine in H 2 O, pH 2.5) using a 1:1 dilution ratio. Quenched samples were flash frozen in methanol containing dry ice, and frozen samples were stored at Ϫ80°C until used.
Prior to UHPLC analysis, deuterated DmUGGT was digested in an on-line immobilized pepsin column prepared in house. Resulting peptides were loaded onto a C 18 analytical column (1-mm inner diameter, 50 mm; Thermo Fisher Scientific) equipped to an Agilent 1290 UHPLC system. Peptides for each sample were separated using a 5-40% liner gradient of acetonitrile containing 0.1% formic acid for 10 min at a 65 l/min flow rate. To minimize back-exchange, the column, solvent delivery lines, injector, and other accessories were placed in an ice bath. The C 18 column was directly connected to the electrospray ionization source of the LTQ Orbitrap XL (Thermo Fisher Scientific), and mass spectra of peptides were acquired in positive-ion mode for m/z 200 -2000. The deuteration (%) as a function of incubation time was determined using HDExam-iner 2.1 (Sierra Analytics, Modesto, CA). The first two amino acid residues in peptides were excluded from the analysis (49).

Cyanogen bromide-activated Sepharose for affinity binding
Sep15 was coupled to cyanogen bromide-activated Sepharose (GE Healthcare) according to the manufacturer's instructions. The coupling reaction was quenched, and unreacted sites were blocked with 100 mM Tris-HCl buffer, pH 8.0. For pulldown assays, the Sep15-coupled beads were incubated with DmUGGT at room temperature for 1 h in binding buffer (25 mM Tris-HCl, pH 8.0, 300 mM NaCl) and washed three times, and the bound proteins were eluted with SDS-PAGE loading buffer for analysis.