Disulfide Bond Formation and N-Glycosylation Modulate Protein-Protein Interactions in GPI-Transamidase (GPIT)

Glycosylphosphatidylinositol (GPI) transamidase (GPIT), the enzyme that attaches GPI anchors to proteins as they enter the lumen of the endoplasmic reticulum, is a membrane-bound hetero-pentameric complex consisting of Gpi8, Gpi16, Gaa1, Gpi17 and Gab1. Here, we expressed and purified the luminal domain of Saccharomyces cerevisiae (S. cerevisiae) Gpi8 using different expression systems, and examined its interaction with insect cell expressed luminal domain of S. cerevisiae Gpi16. We found that the N-terminal caspase-like domain of Gpi8 forms a disulfide-linked dimer, which is strengthened by N-glycosylation. The non-core domain of Gpi8 following the caspase-like domain inhibits this dimerization. In contrast to the previously reported disulfide linkage between Gpi8 and Gpi16 in human and trypanosome GPIT, our data show that the luminal domains of S. cerevisiae Gpi8 and S. cerevisiae Gpi16 do not interact directly, nor do they form a disulfide bond in the intact S. cerevisiae GPIT. Our data suggest that subunit interactions within the GPIT complex from different species may vary, a feature that should be taken into account in future structural and functional studies.

suggested to recognize long chain fatty acids of the GPI lipid mainly because its sequence contains a motif found in mammalian and yeast fatty acid elongases 5,12,18 .
Even though each of the GPIT subunits is postulated to have a unique function in the transamidation reaction, experimental tests of these assignments are hampered by the lack of structural information and a robust in vitro GPIT activity assay. Currently, only low-resolution structure mapping of partial luminal domains of PIG-K/ Gpi8, GAA1/Gaa1 and PIG-S/Gpi17 are available through the small-angle X-ray scattering (SAXS) method 19 . While working towards reconstituting the S. cerevisiae GPIT complex for structural studies we made unexpected findings on the properties of S. cerevisiae Gpi8 and S. cerevisiae Gpi16 with regards to disulfide pairings and N-glycosylation in the Gpi8-Gpi16 binary complex. Here we present these findings and discuss their implications for future structural studies on the complex.

Saccharomyces cerevisiae GPI-Transamidase Complex is Most Likely Dimeric. Towards recon-
stituting a GPI-transamidase (GPIT) complex for structural studies, we first investigated the interaction among the five GPIT subunits under physiological conditions. A triple tandem FLAG (3xFLAG) tag was individually engineered into the C-terminus of 4 subunits of the endogenous GPIT proteins to create 4 different strains of Saccharomyces cerevisiae (S. cerevisiae). In each case, expression of the tagged subunit was from the genomic locus of the native open reading frame and under natural regulation. As GPIT is essential for yeast viability, the ability to generate these strains indicated that C-terminal tagging did not affect function. We extracted the endogenous GPIT complex from each strain using 1% digitonin. Immunopurified S. cerevisiae GPIT from the four different 3xFLAG-tagged strains showed similar results on SDS-PAGE analysis, and the purified GPIT from the strain expressing Gaa1-3xFLAG on a 10% SDS-PAGE is shown in Fig. 1B. The full-length, membrane bound endogenous S. cerevisiae Gpi8 has been shown to possess multiple N-glycosylation sites 9 , resulting in the observed doublet. As shown for human PIG-U (equivalent to S. cerevisiae Gab1) 4 , the S. cerevisiae Gab1 migration position is much lower than its calcualted molecular weight (Fig. 1B). To characterize the endogenous GPIT complex, we first ran wild type (WT) and 3xFLAG-tagged S. cerevisiae lysates using blue native polyacrylamide gel electrophoresis (BN-PAGE) followed by either anti-Gpi8 or anti-FLAG Western blotting. Compared to regular native PAGE, BN-PAGE allows separation of the complexes based on the molecular mass while still preserving the intact protein complexes 20,21 . Given that GPIT likely consists of an equi-stoichiometric complex of Gpi8, Gpi16, Gaa1, Gpi17 and Gab1 15,18,22 , the predicted monomeric molecular mass of S. cerevisiae GPIT is ~290 kDa. S. cerevisiae GPIT migrated on BN-PAGE as high molecular mass bands between the molecular weight markers at 480 kDa and 720 kDa (Fig. 1C), suggesting that it is dimeric. The WT GPIT migrated somewhat faster than the 3xFLAG reconstituted counterpart, likely because of the highly charged nature of the FLAG tag sequence. Anti-FLAG purified GPIT ran homogenously on BN-PAGE at a position similar to the major band in the 3xFLAG reconstituted S. cerevisiae lysates (Fig. 1C). Human GPIT was also reported to be large from a velocity gradient with a sedimentation coefficient of ~17S 23 .
Dimerization of the Caspase-like Domain of the Luminal Region of S. cerevisiae Gpi8. The luminal region of Gpi8 contains a previously identified caspase-like domain 24 followed by a non-core domain (Fig. 2). The caspase-like domain within the luminal region of Gpi8 has previously been expressed in the reducing environment of Escherichia coli (E. coli) cytoplasm 24 . Gpi8 contains four Cys residues in its luminal domain, with three in the caspase-like domain (Fig. 2), and localizes in the oxidizing environment of the endoplasmic reticulum (ER) lumen. Thus, we decided to express Gpi8 as a secreted protein in insect cells to promote proper folding and disulfide pairing in the oxidizing environment of the secretory pathway. To keep the oxidizing environment, we also purified the recombinant Gpi8 under a non-reducing condition in the absence of dithiothreitol (DTT).
Expressing the caspase-like domain of S. cerevisiae Gpi8 (23-306aa) as a secreted protein in insect cells resulted in an appreciable yield (Fig. 3A). In contrast to the mostly monomeric state of the same Gpi8 construct expressed previously in the E. coli BL21 cytoplasm 24 , the insect cell expressed Gpi8 appeared to form stable dimers in gel filtration (Fig. 3A). In addition, our SDS-PAGE analysis showed that insect cell expressed Gpi8 has two bands around 47 kDa implying that the protein was probably posttranslationally modified during insect cell expression (Fig. 3B). There are two predicted N-glycosylation sites in S. cerevisiae Gpi8 (23-306aa). We first confirmed that the insect cell expressed Gpi8 was N-glycosylated during the expression by treating the protein with endoglycosidases such as PNGase F (Fig. 3B). The treatment generated a triplet of bands with the two upper bands most likely corresponding to glycosylated forms and the lowest band being completely deglycosylated. We then expressed S. cerevisiae Gpi8 in insect cells in the presence of 2.5 mg/L tunicamycin, which serves to block addition of N-acetylglucosamine to dolicholphosphate in the first step of N-linked oligosaccharide formation. The purified S. cerevisiae Gpi8 expressed in the presence of tunicamycin showed a migration location on SDS-PAGE similar to the lowest band after deglycosylation (Fig. 3B). Interestingly, its size exclusion chromatography (SEC) profile indicated mostly dimers, but with significant amount of monomers (Fig. 3C), suggesting that glycosylation somehow enhanced dimerization.
Given that E. coli BL21 expressed S. cerevisiae Gpi8 (23-306aa) reported previously exhibited mostly monomeric species, we attempted bacterial expression using three different bacterial cell lines, E. coli BL21, E. coli Rosetta-gami 2, and E. coli SHuffle. Rosetta-gami 2 harbors mutations in both the thioredoxin reductase (trxB) and glutathione reductase (gor) genes, while SHuffle constitutively expresses disulfide isomerase DsbC. In theory, both may enhance disulfide bond formation in the E. coli cytoplasm. Expression of S. cerevisiae Gpi8 in E. coli BL21 and E. coli SHuffle both resulted in almost equal amounts of dimers and monomers while Rosetta-gami 2 showed very low expression (Fig. 3C). Because of the similar expression results from E. coli BL21 and E. coli SHuffle, we only used E. coli BL21 in further experiments. Re-run of the insect cell and E. coli expressed dimer peaks of Gpi8 on SEC showed that they stayed as dimers ( Supplementary Fig. S1), suggesting that the dimer peaks cannot easily dissociate into monomers. In contrast, the monomer peak of E. coli BL21 expressed Gpi8 eluted at both dimer and monomer positions ( Supplementary Fig. S1), suggesting that dimers could be formed from monomers. Collectively, these data suggest that Gpi8 can form dimers when expressed either in the more reducing environment in the E. coli cytoplasm or the more oxidized environment in the insect cell secretary pathway.

S. cerevisiae Gpi8 (23-306aa) Can Form Homo-dimers Through an Interchain Disulfide Bond. S.
cerevisiae Gpi8 (23-306aa) contains three Cys residues at sites 85, 199 and 246, respectively (Fig. 2). C199 is the predicted catalytic residue. Because of the different degrees of dimerization in the different redox environments, we asked whether the dimers could be linked via disulfide bonds. To elucidate this, we run non-reducing SDS-PAGE in the absence of DTT in the SDS sample buffer. Strikingly, we observed a major band at around 70 kDa representing the dimeric state of the insect cell expressed Gpi8 (Fig. 3D), indicating linkage by interchain disulfide bond in the dimer. Insect cell expressed Gpi8 in the presence of the glycosylation inhibitor tunicamycin showed a mixture of dimers and monomers on non-reducing SDS-PAGE (Fig. 3D). Similarly, E. coli expressed Gpi8 also showed a mixture of monomers and dimers, consistent with the SEC profile (Fig. 3C,D). In comparison, when DTT was added to the SDS sample buffer, only monomers were observed on SDS-PAGE for all Gpi8 samples (Fig. 3D).
To further elucidate the relationship between disulfide bonds and dimer formation, we tested the effect of adding DTT to the protein purification buffers. For both insect cell expressed Gpi8 in the presence of tunicamycin and E. coli expressed Gpi8, the SEC profiles shifted so that a majority of the proteins came out in monomeric peaks (Fig. 4A). These data suggest that disulfide bond formation is critical for the stability of the observed dimers. Consistently, previously reported recombinant S. cerevisiae Gpi8 caspase-like domain expressed in E. coli and purified in the presence of DTT is also mostly monomeric 24 . However, insect cell expressed, fully glycosylated Gpi8 did not dissociate into monomers in the presence of DTT (Fig. 4A). We suspect that the N-glycans may have shielded the access of DTT to the disulfide bond at the dimer interface. This is supported by the complete dissociation into monomers on reducing SDS-PAGE under the denaturing condition (Fig. 3D).

Cys85 of S. cerevisiae Gpi8 is Responsible for Disulfide Bond Formation.
Having confirmed the hypothesis that S. cerevisiae Gpi8 (23-306aa) can form disulfide-linked homo-dimers, we individually mutated the three Cys residues within the protein construct ( Fig. 2A) and investigated its SEC profiles under both reducing and non-reducing purification processes. While the C199A and C246A mutations did not change the dimer to monomer ratios in comparison with the WT (Supplementary Fig. S2), the C85A mutation caused a shift to mostly monomeric positions on the SEC profile even in the absence of DTT (Fig. 4B). Non-reducing SDS-PAGE showed that the dimer peak did not contain disulfide-linked species (Fig. 4B). Furthermore, multi-angle light scattering (MALS) measurement confirmed the dimeric and monomer molecular masses of the peaks (Fig. 4B). Re-run of the monomer peak on SEC in the absence of DTT showed only the monomer peak ( Supplementary Fig. S2). Unlike the dimer peak from either insect cell or E. coli expressed WT Gpi8 which stayed as dimers, the dimer peak of C85A exhibited a mixture of dimers and monomers in the SEC re-run, suggesting equilibrium between dimers and monomers in the absence of disulfide linkage ( Supplementary Fig. S2).
Because E. coli expressed C85A mutant and insect cell expressed WT Gpi8 gave high quantities of stable monomer and dimer respectively, we used small-angle X-ray scattering (SAXS) to gain further structural insights on Residues identical among the different species are shown in white on a red background. Conserved, but not identical, residues are shown as red letters. Secondary structures predicted in the S. cerevisiae Gpi8 model from Robetta are shown above the protein sequences in blue. Because Gpi8 has a caspase-like domain, we aligned S. cerevisiae Gpi8 with caspase-7 (PDB code 1F1J) 29 and the corresponding secondary structure from caspase-7 are shown in pale blue. Identical residues between S. cerevisiae Gpi8 and caspase-7 are shown in white with a pink background. Dashed lines represent deletions in caspase-7 sequence relative to S. cerevisiae Gpi8.
Scientific RepoRts | 7:45912 | DOI: 10.1038/srep45912 factors affecting dimerization. Both Guinier plots of the proteins show absences of aggregation, so we processed the data and calculated the pair-distance distribution function P(r) (Fig. 4C). We generated ab initio molecular envelopes for both the C85A monomer and the WT dimer. The SAXS envelope of the C85A mutant appears to be of a tubular shape with one end larger than the other. The dimer envelope matched well with two monomer envelopes (Fig. 4D), suggesting the validity of the calculations. Previously, Toh and colleagues 19 performed SAXS analysis on S. cerevisiae Gpi8 (24-337aa) expressed from E. coli, showing a similar monomeric envelope for Gpi8, whereas we report here the envelope and model for the dimeric species as well.
Because Gpi8 has been predicted to have a caspase-like fold 24 , we used the protein structure prediction server Robetta 25 to obtain structural information on the luminal region of S. cerevisiae Gpi8. Five independent structural models were generated. The caspase-like domain (23-306aa) showed remarkable similarity among the five models, with pairwise RMSDs between 0.7 Å to 1.2 Å (Supplementary Fig. S3). Within the caspase-like domain, the conserved Cys85 and the active site Cys199 are exposed in the model while Cys246 is completely buried. Because insect cell expressed Gpi8 (23-306aa) is bonded via a disulfide linkage to form homo-dimers, we placed two Gpi8 (23-306aa) Robetta models into the Gpi8 (23-306aa) SAXS envelope with Cys85 from each molecule facing and interacting with each other (Fig. 4E). Such a placement also positioned the two glycosylation sites adjacent to the dimerization interface (Fig. 4E), which could have played a role in dimer stabilization and in protecting the disulfide bond from DTT. The model fitting in the dimer envelope was then transferred to the C85A monomer envelope using the superposition between the dimer and monomer envelopes, with reasonable agreement (Fig. 4D,F).
The Effects of the Non-core Domain on Cys85 Disulfide Bond Formation. The luminal domain of S. cerevisiae Gpi8 (23-379) contains a non-core domain of around 80 residues after the conserved caspase-like domain that has no recognizable sequence similarity to known domains (Fig. 2). We expressed full-length luminal domain of Gpi8 (23-379aa) in both E. coli cytoplasm and as a secretary protein in insect cells. Surprisingly, the SEC profiles even in the absence of DTT showed that both full-length luminal Gpi8 proteins are mostly monomers (Fig. 5A), unlike the caspase-like domain alone. We tested if the presence of the non-core domain somehow inhibited dimerization using deletion studies. Indeed, progressive deletion of parts of the non-core domain seemed to correlate with re-appearance of the dimer peak (Fig. 5B), supporting the hypothesis.
Because Cys85 is a main factor in dimer formation for the caspase-like domain, we wondered if Cys85 is paired internally in the full-length luminal Gpi8 construct. Only one Cys residue is present in the non-core domain of S. cerevisiae Gpi8, Cys373 (Fig. 2). When we mutated Cys373 to Ala within the full-length luminal construct, the SEC profile under the non-reducing purification condition appeared as a mixture of dimers and monomers (Fig. 5C). Therefore, the conserved Cys85 must have once again involved in the disulfide-linked dimer formation in the absence of intramolecular pairing with Cys373. In the Robetta models that we generated, the non-core region preferentially forms a structure with a central three-stranded β -sheet flanked by α -helices  (23-306aa) is within the S. cerevisiae Gpi8 (23-306aa) C85A monomer SAXS envelope. The residues in magenta represent insertions in S. cerevisiae Gpi8 relative to the caspase-7 structure. Active site Cys199 and His157 are colored in Green. Cys85 is in yellow. (Fig. 5D). The beginning of the non-core domain is apparently loopy and flexible, which created different relative orientations between the caspase-like and non-core domains in the different models. In one of the models, the locations of Cys85 and Cys373 are juxtaposed for potential disulfide bond formation (Fig. 5D).

S. cerevisiae Gpi8 Does Not Form a Disulfide-linked Hetero-dimer with S. cerevisiae
Gpi16. Our data indicated that Cys85 has a high propensity for forming inter-or intra-molecular disulfide bond in S. cerevisiae Gpi8. In the intact human GPIT complex, previous studies have also implicated that PIG-K and PIG-T are linked together through a disulfide bond in their luminal domains, Cys92 of PIG-K and Cys182 of PIG-T 15 , equivalent to Cys85 of S. cerevisiae Gpi8 and Cys202 of S. cerevisiae Gpi16. Collectively, these data all suggest that the conserved Cys85 of S. cerevisiae Gpi8, or its equivalent in human PIG-K, is a critical residue for the structure of Gpi8 itself or of the GPIT complex. However, the possibility of alternative disulfide pairings relating to Cys85 also causes confusion in its physiological role within the GPIT complex.
We wondered if S. cerevisiae Gpi8 and S. cerevisiae Gpi16 interact differently from the human counterpart. To address this question, we first expressed S. cerevisiae Gpi16 luminal domain (20-550aa) as a secreted protein in insect cells. Since S. cerevisiae Gpi16 contains 8 Cys residues likely with complicated disulfide pairings, we did not attempt expression in E. coli. Insect cell expression of Gpi16 resulted in a good yield of the secreted protein. The elution position in the SEC profile corresponded to the molecular mass of monomeric Gpi16 (Fig. 6A).
Having obtained purified proteins of both S. cerevisiae Gpi8 and S. cerevisiae Gpi16, we tested whether they form a complex through co-expression in insect cells. We reasoned that the proper redox environment in the secretory pathway could be essential for formation of the disulfide-linked Gpi8-Gpi16 dimer, if they behave similarly to the case for human GPIT. We co-expressed Gpi16 with either Gpi8 caspase-like domain (23-306aa) or the Gpi8 full-length luminal domain (23-379aa). Both Gpi16 and Gpi8 constructs are His-tagged and both proteins were purified together from Ni-affinity chromatography. However, the SEC profiles and native PAGE of the co-expressed proteins showed that neither Gpi8-Gpi16 pairs co-migrated (Fig. 6B-D). Because dimeric Gpi8 (23-306aa) has a similar molecular mass as monomeric Gpi16, their elution positions in the co-purified mixture are similar (Fig. 6B). Lack of interaction was nonetheless clearly seen on native PAGE of the gel filtration peak, in  which Gpi16 and Gpi8 migrated separately as individual bands without generation of any band shift due to complex formation (Fig. 6C). Similarly, individually purified Gpi8 (23-306) and Gpi16 also did not form a complex on native PAGE (Fig. 6C).
There may be at least two reasons why S. cerevisiae Gpi8 and S. cerevisiae Gpi16 failed to form a complex despite the predicted disulfide bond linkage. One possibility is that Gpi8 and Gpi16 only form a complex in the context of the entire five-subunit GPIT complex. Another possibility is that the S. cerevisiae GPIT is somewhat different from the human GPIT in disulfide pairing, as not all Cys residues are conserved ( Fig. 2A). To elucidate whether Gpi8 and Gpi16 are disulfide linked in the S. cerevisiae GPIT complex, we purified endogenous GPIT complexes using the engineered strains containing 3xFLAG tag on either Gpi8 or Gpi16, similarly as what we did in Fig. 1. To preserve potential inter-chain disulfide pairings, we used non-reducing SDS-PAGE followed by anti-FLAG Western blot (Fig. 6E,F). Anti-FLAG blots on the Gpi8-3xFLAG GPIT complex and the Gpi16-3xFLAG GPIT complex showed major bands at close to the expected molecular weights of Gpi8 and Gpi16, respectively. Both Gpi8 and Gpi16 run somewhat slower than the expected monomeric molecular weights likely due to the addition of the extended 3xFLAG tag. Unlike the case for human GPIT 15 , there are no major higher molecular weight bands that could correspond to the disulfide linked S. cerevisiae Gpi8-Gpi16 subunits. The minor bands in both anti-FLAG blots for Gpi8 and Gpi16 are above the 260 kDa marker and therefore unlikely correspond to the Gpi8-Gpi16 complex.

Discussion
GPIT is an important enzyme complex that has been associated with human genetic diseases such as paroxysmal nocturnal haemoglobinuria (PNH) 16 and other intellectual disability syndromes 17 . In plant, the catalytic subunit Gpi8 of GPIT plays an essential role in stomata formation and fertility 26 . Here, we report an initial attempt to characterize S. cerevisiae GPIT by purification of the endogenous complex and by recombinant expression. Our findings suggest a dimeric GPIT complex, likely mediated by the Gpi8 subunit, which shows similarities to caspases. This Gpi8-mediated dimerization appears to be stabilized by disulfide bond formation and by glycosylation. The dimeric GPIT supposedly provides a large surface for all the transamidation reaction steps. This scenario is different from the previously reported data on human GPIT in which Gpi16 and Gpi8 stabilize each other by a disulfide bond 15 and exist in the ternary complex of Gaa1, Gpi8, and Gpi16 6 , which may be important for future structural studies on this complex.

Materials and Methods
Purification of Endogenous Yeast GPIT and Blue Native PAGE Assay. GPIT was separately purified from each of four different S. cerevisiae (S. cerevisiae) strains containing 3xFLAG-Gpi8, 3xFLAG-Gaa1, 3xFLAG-Gpi16, or 3xFLAG-Gpi17 using the following steps. The yeast cells were cultured in YPD medium at 37 °C until ~9,000 OD 600 units of yeast (~2.4 liter culture at a cell density of OD 600 ~3.7). The yeast cells then were lysed using glass beads (VWR cat # 101454-156) in disruption buffer containing 5% glycerol, 20 mM HEPES at pH 7.5, 10 mM MgCl 2 and protease inhibitors (Sigma cat # S8830). After separating glass beads and cell debris at 1,000 × g, microsomal membranes were collected in high speed at 4 °C for 20 min. GPIT extraction was done in 1% digitonin (Sigma cat # D141-500 MG), 20 mM HEPES pH = 7.5, 100 mM NaCl and protease inhibitors for 60 min on ice. Collected digitonin extract of microsomes were pre-cleared for non-specific binding using IgG beads in buffer containing 0.1% digitonin, 20 mM HEPES at pH 7.5, 100 mM NaCl and protease inhibitors before being treated with anti-FLAG agarose. The bound yeast GPIT complexes were eluted with 5 mg/mL 3xFLAG at 4 °C with shaking for 2 hour, and the samples were loaded either on a 10% SDS-PAGE gel followed by colloidal Coomassie Blue staining or on a freshly made 6% Blue native PAGE to determine native protein masses and oligomeric states. We casted the Blue native PAGE following the published protocol 21 . Cloning and Protein Expression in E. coli. Bacterial expression of S. cerevisiae Gpi8 (23-306aa) BL21 , S. cerevisiae Gpi8 (23-339aa) BL21 , S. cerevisiae Gpi8 (23-359aa) BL21 , S. cerevisiae Gpi8 (23-379aa) BL21 , S. cerevisiae Gpi8 (23-306aa) Cys85Ala , S. cerevisiae Gpi8 (23-306aa) Cys199Ala and S. cerevisiae Gpi8 (23-306aa) Cys244Ala used the following cloning and expression methods. The luminal portion of the Saccharomyces cerevisiae Gpi8 sequence (S. cerevisiae Gpi8; NCBI RefSeq: NM_001180639.1) was used as a template for PCR reactions, and each PCR product was sub-cloned into NdeI-XhoI sites of pET26a (Novagen). The S. cerevisiae Gpi8 constructs were transformed into E. coli BL21-CodonPlus ® (DE3)-RIPL competent cells (Agilent Technologies) and E. coli Rosetta-gami 2 (DE3) pLysS (Novagen) for cytosolic protein expression. The E. coli containing the expression vectors were inoculated in Luria-Bertoni (LB) broth containing 100 μ g/ml kanamycin and 100 μ g/ml chloramphenicol until an OD 600 of 0.6 was reached, and isopropyl β -D-thiogalactoside (IPTG) was added to a final concentration of 1 mM. The culture was allowed to grow over night at 20 °C before harvesting at 4,000 × g for 15 min.

Cloning and Expression of Secreted Proteins in Insect Cells.
Expressing gpi8 and gpi16 genes as secreted proteins in insect cells may be necessary to promote proper folding and disulfide bond formation. The luminal portions of S. cerevisiae Gpi8 (UniParc: P49018) and S. cerevisiae Gpi16 (UniParc: P38875) were inserted into the baculovirus pFastBac-dual vector (Invitrogen) with an N-terminal gp67 signal peptide, which is an effective signal sequence for protein secretion, and a C-terminal His-tag. Each pFastBac-dual construct was first transformed into DH10Bac E. coli for transposition into the bacmid, and the isolated bacmid DNA was transfected in Spodoptera frugiperda Sf9 insect cells using Cellfectin II Reagent. Once virus plaques developed, the plaques with recombinant viruses were picked and expanded. The virus stock was then used to infect a fresh culture of insect cells, resulting in expression of Gpi8 or Gpi16 proteins. To inhibit glycosylation of insect cell-expressed secreted proteins, 2.5 mg/L or 5 mg/L tunicamycin (Sigma-Aldrich) was added into the medium during insect cells infection.
Scientific RepoRts | 7:45912 | DOI: 10.1038/srep45912 Protein Purification. Both bacterial expressed and insect cell-expressed proteins were purified using Ni-affinity chromatography and size exclusion chromatography (Superdex 200 10/300 GL). The bacteria cells were lysed on ice by sonication in Buffer A (50 mM Tris-HCl pH 8, 300 mM NaCl and 15 mM imidazole), added with 5 mM β -Mercaptoethanol if purifying proteins under a reducing condition. The supernatant that was separated from the E. coli lysate through centrifugation was incubated with Ni-NTA beads (GE Healthcare) for one hour at 4 °C with rotation. The insect cell medium that contained the expressed secreted protein was loaded onto a pre-packed Ni column directly after pH adjustment. The Ni-NTA beads bound with the protein were washed 3 times with Buffer B (50 mM Tris-HCl at pH 8, 300 mM NaCl, 35 mM imidazole, with or without 5 mM β -mercaptoethanol). The proteins were eluted with Buffer C (50 mM Tris-HCl at pH 8, 300 mM NaCl, 150 mM imidazole, with or without 5 mM β -mercaptoethanol). Subsequently, the concentrated protein eluent was subjected to size exclusion chromatography using Buffer D (20 mM Tris-HCl at pH 8 and 150 mM NaCl, with or without 2.5 mM DTT).

Non-reducing SDS-PAGE.
To capture disulfide-linked species, we used an SDS sample buffer without any reducing agent. The proteins were mixed with the sample buffer and heated at 95 °C for 5 minute before being loaded onto SDS-PAGE gels.
Structural Modeling. Because Gpi8 has been predicted to have a caspase-like fold, we used the full-chain protein structure prediction server Robetta 25 to model the structure of its luminal domain. We generated five independent structural models, and the caspase-like core domain models are highly similar to each other. The non-core region C-terminal to the caspase-like domain preferentially forms a structure with a central three-stranded β -sheet flanked by α -helices.

Small-angle X-ray Scattering (SAXS) Experiments and Data
Analysis. The small-angle X-ray scattering data of the various Gpi8 recombinant proteins were collected using the X9 beamline at the National Synchrotron Light Source of the Brookhaven National Laboratory. The purified proteins were stored in Buffer D containing 20 mM Tris-HCl at pH 8.0 and 150 mM NaCl before SAXS experiments. Each blank or sample was measured in triplicate, and the sample measurements were adjusted by subtracting the scattering from buffer-alone blank. Several protein concentrations (2.5 to 7.0 mg/ml) were used for each recombinant protein, in order to assess and remove any concentration-dependent inter-particle effects. All data were analyzed and processed using program packages PRIMUS 27 , and GNOM 28 . The SAXS envelopes were built using the program RasWin.