Partial Enzymatic Deglycosylation Preserves the Structure of Cleaved Recombinant HIV-1 Envelope Glycoprotein Trimers*

Background: The heterogeneity and flexibility of HIV-1 envelope glycoprotein N-glycans interfere with structural and vaccine studies. Results: HIV-1 envelope trimers can be partially deglycosylated without affecting trimer integrity. Conclusion: HIV-1 envelope glycoprotein N-glycans do not contribute to trimer integrity once the protein is folded. Significance: Deglycosylated HIV-1 envelope trimers should be useful for structural and vaccine studies. The trimeric envelope glycoprotein complex (Env) is the focus of vaccine development programs aimed at generating protective humoral responses to human immunodeficiency virus type 1 (HIV-1). N-Linked glycans, which constitute almost half of the molecular mass of the external Env domains, produce considerable structural heterogeneity and are a major impediment to crystallization studies. Moreover, by shielding the peptide backbone, glycans can block attempts to generate neutralizing antibodies against a substantial subset of potential epitopes when Env proteins are used as immunogens. Here, we describe the partial deglycosylation of soluble, cleaved recombinant Env trimers by inhibition of the synthesis of complex N-glycans during Env production, followed by treatment with glycosidases under conditions that preserve Env trimer integrity. The partially deglycosylated trimers are stable, and neither abnormally sensitive to proteolytic digestion nor prone to aggregation. Moreover, the deglycosylated trimers retain or increase their ability to bind CD4 and antibodies that are directed to conformational epitopes, including the CD4-binding site and the V3 region. However, as expected, they do not react with glycan-dependent antibodies 2G12 and PGT123, or the C-type lectin receptor DC-SIGN. Electron microscopic analysis shows that partially deglycosylated trimers have a structure similar to fully glycosylated trimers, indicating that removal of glycans does not substantially perturb the structural integrity of the trimer. The glycan-depleted Env trimers should be useful for structural and immunogenicity studies.

The trimeric envelope glycoprotein complex (Env) is the focus of vaccine development programs aimed at generating protective humoral responses to human immunodeficiency virus type 1 (HIV-1). N-Linked glycans, which constitute almost half of the molecular mass of the external Env domains, produce considerable structural heterogeneity and are a major impediment to crystallization studies. Moreover, by shielding the peptide backbone, glycans can block attempts to generate neutralizing antibodies against a substantial subset of potential epitopes when Env proteins are used as immunogens. Here, we describe the partial deglycosylation of soluble, cleaved recombinant Env trimers by inhibition of the synthesis of complex N-glycans during Env production, followed by treatment with glycosidases under conditions that preserve Env trimer integrity. The partially deglycosylated trimers are stable, and neither abnormally sensitive to proteolytic digestion nor prone to aggregation. Moreover, the deglycosylated trimers retain or increase their ability to bind CD4 and antibodies that are directed to conformational epitopes, including the CD4-binding site and the V3 region. However, as expected, they do not react with glycan-dependent antibodies 2G12 and PGT123, or the C-type lectin receptor DC-SIGN. Electron microscopic analysis shows that partially deglycosylated trimers have a structure similar to fully glycosylated trimers, indicating that removal of glycans does not substantially perturb the structural integrity of the trimer. The gly-can-depleted Env trimers should be useful for structural and immunogenicity studies.
The Env complex of HIV-1 is a trimer of gp120/gp41 heterodimers. This structure, often referred to as a spike, attaches to receptors on target cells and mediates the fusion of the viral and cell membranes at the start of the infection process. Antibodies that bind to the native spike impede one or more stages in the virus-cell fusion process and thereby neutralize infectivity. The HIV-1 Env complex is, therefore, the focus of vaccine development strategies intended to prevent the establishment of infection via the induction of virus-neutralizing antibodies (NAbs). 4 The Env gp120 and gp41 subunits are extensively glycosylated, with N-linked glycans constituting approximately half the gp120 mass. There are between 18 and 33 (median, 25) canonical N-linked carbohydrate attachment sites (NX(S/T)) on gp120 proteins from different isolates (1,2). Although most of these sites are utilized, the glycosylation can be both isolatedependent and influenced by the producer cell type (3,4). The number of glycan sites present on Env, and the extent to which they are glycosylated, tend to increase during the course of HIV-1 infection, reflecting antibody-mediated selection pressures that drive resistance to autologous NAbs (5). Gp41 contains 4 or 5 canonical N-linked glycosylation sites, but there are conflicting reports about the extent to which they are actually occupied by a glycan (4, 6 -8). One or a few O-linked carbohydrates might also be present on gp120, but would contribute little to the glycan mass as the individual structures have been reported to be small (9,10).
Glycosylation is a post-translational modification that contributes to the proper folding of the Env protein in the endoplasmic reticulum. Thus, when N-glycosylation is blocked by enzyme inhibitors or by eliminating the canonical sites, gp120 does not fold properly (8,(11)(12)(13). The modification or removal of carbohydrates from virions can affect Env function and virus infectivity (14 -17). N-Glycans are also involved in the attachment of virions to certain cell types that express appropriate C-type lectin receptors, such as DC-SIGN or Langerin, events that may be important for the establishment and/or systemic spread of HIV-1 infection in a new host (18,19). Under some circumstances, the N-glycans can trigger immunosuppressive responses that may interfere with anti-Env antibody production (20 -25). The glycan component of Env helps counter host immunity by occluding underlying or topologically adjacent protein epitopes (23,26,27), and was long thought to be mostly immunosilent and only rarely recognized as foreign by the host. The glycan shielding can hinder the induction of NAbs against certain epitopes and prevent the binding of NAbs once they have been elicited. Thus, the N-linked glycans of Env constitute an important strategy by which HIV-1 evades the humoral immune response and complicates the design of HIV-1 Envbased vaccines.
Paradoxically, the N-glycans on gp120 are components of the epitopes for a subset of broadly active NAbs (bNAbs); such glycan-dependent bNAbs are much more common than previously appreciated (28 -30). The highly unusual domain-exchanged 2G12 antibody recognizes a discontinuous gp120 epitope made up of elements from 3 or 4 individual glycans (31)(32)(33)(34), whereas the discontinuous epitopes for bNAbs PG9 and PG16 are dependent on glycosylation at specific N-linked sites in the V1/V2 domain (28,35,36). bNAbs in the recently discovered PGT category also have glycan-dependent epitopes (29,37). Of note is that, whereas 2G12 recognizes only the tips of glycan structures, PG9 and PGT128 bind to composite epitopes containing elements of the glycan stems and the underlying peptide backbone (29,36,37).
The heterogeneity and flexibility of carbohydrate chains create substantial impediments to crystallizing gp120 monomers or soluble trimers for atomic structure determinations. The crystal structures of HIV-1 gp120 cores were obtained using enzymatically deglycosylated proteins (38 -41). To date, no crystal structure of an Env trimer has been obtained, in part because it has not been possible to produce stable and soluble deglycosylated Env trimers. Deglycosylating trimeric Env on virions has proven problematic, probably because steric factors restrict access of glycosidases to their substrates (17,42). Partially deglycosylated, uncleaved gp140 trimers have recently been described (43).
For use in both immunogenicity and structural studies, we have now identified nondenaturing conditions under which SOSIP gp140 trimers can be partially deglycosylated without compromising their structural integrity. The resulting trimers are stable, highly soluble, resistant to proteolysis, and appear to retain an appropriate conformation as judged by the binding of CD4 and monoclonal antibodies (mAbs) to conformation-sensitive epitopes. Furthermore, electron density maps at 14-Å resolution using negative stain electron microscopy show that, other than the expected reduction in their overall size, deglycosylation causes no detectable structural changes to the trimers. Hence, once the Env trimer has formed, most of its glycans are not involved in maintaining its integrity.

MATERIALS AND METHODS
Constructs-The pPPI4 vector system for expressing Env proteins has been described elsewhere (44). Single amino acid substitutions were made using the QuikChange mutagenesis kit (Stratagene, La Jolla, CA). In the current study, we used monomeric gp120s from the subtype B strain JR-FL. The KNH1144 SOSIP.R6 gp140 protein contains the trimer-stabilizing A501C, T605C, and I559P substitutions and a hexa-arginine (R6) motif at the C terminus of gp120 that increases the efficiency of gp120-gp41 proteolytic cleavage (44,45,53,54). The sequence was also modified to contain epitopes for mAbs 2F5, 4E10, and 2G12 (47). The resulting SOSIP variant, SOSIP.681, was used for most of the biochemical studies. A new variant, SOSIP.664, which lacks the C-terminal residues 665-681 of the gp41 ectodomain, was used for electron microscopy (EM) and SEC/MALS analyses.
Reagents-DC-SIGN-Fc was purchased from R&D Systems (Minneapolis, MN). HIVIg was obtained through the AIDS Research and Reference Reagent Program (ARRRP), Division of AIDS, NIAID, National Institutes of Health. mAbs 2G12, 4E10, 2F5, and D-50 were provided by Hermann Katinger through the ARRRP. 447-52D was obtained from Polymun Sciences, Klosterneuburg, Austria. mAb b12 was a gift from Dennis Burton (Scripps Research Institute, La Jolla, CA), mAbs 17b, 5.8c, and 1.4e were from James Robinson (Tulane University, New Orleans, LA). The CD4-IgG2 and sCD4 proteins from Progenics Pharmaceuticals have been described elsewhere (53). PGT123 was obtained through the IAVI NAC repository. Mannosidases, endoglycosidase F (Endo F), and endoglycosidase H (Endo H) were purchased from New England Biolabs (Ipswich, MA). Proteases trypsin, chymotrypsin, and proteinase K were obtained from Sigma.
Cell Culture and Transient Transfection-The human embryonic kidney (HEK) cell line 293T was grown under standard conditions (45). The 293S GnTI Ϫ/Ϫ line is a HEK 293 variant that does not express N-acetylglucosaminyltransferase I (GlcNAc transferase I; GnTI) (55). Proteins produced in these cells are designated, for example, as SOSIP.681 GnTlϪ/Ϫ . Both cell lines were transiently transfected with env genes using linear polyethylenimine (M r 25,000) as described previously (56). Briefly, DNA encoding an Env protein was diluted in Dulbecco's modified Eagle's medium (DMEM; Invitrogen), to 1/10 of the final culture volume and mixed with polyethylenimine (0.15 mg/ml final concentration). After incubation for 20 min, the DNA/polyethylenimine mixture was added to the cells for 4 h before replacement with culture medium containing 10% fetal bovine serum (FBS; Invitrogen), penicillin, streptomycin, and minimal essential medium nonessential amino acids (0.1 mM, Invitrogen). Supernatants were harvested 48 h after transfection. When appropriate, 100 M kifunensine, an inhibitor of the glycoprotein processing mannosidase I enzyme (TRC, Toronto, Canada), was added to the cultures 1 h before transfection, and the medium was replaced 4 h after transfection with a fresh batch containing 100 M kifunensine. All protein samples were kept on ice prior to subsequent analyses.
Endo-and Exoglycosidase Digestions-We tested many conditions for endo-and exoglycosidase treatment of gp120 monomers and gp140 trimers (data not shown, and see "Results"). Two final protocols were selected (unless indicated otherwise), depending on the analyses to be performed subsequently. Both methods were equally effective at deglycosylating trimers, while preserving their conformation. The SOSIP.681 GnTlϪ/Ϫ deglycosylation reactions were performed using 25 units of Endo H/g of purified Env trimer for 3 h at room temperature and neutral pH (PBS or culture media, pH 7.2). In control studies, the Env protein was first denatured by incubation for 10 min at 95°C with 0.02% SDS and 1 mM dithiothreitol (DTT), then treated with Endo H using conditions and a buffer specified by the manufacturer. Purified SOSIP.664 GnTlϪ/Ϫ proteins were deglycosylated using 0.3 units of Endo H/g of purified Env trimer for 3 h at room temperature in 100 mM sodium acetate, pH 6.0.
Limited Proteolysis-Purified trimers (17 g/ml) produced in 293S GnTI Ϫ/Ϫ cells and deglycosylated with Endo H (or mock treated) were incubated with serial 10-fold dilutions (1-0.001 g/ml) of various proteases for 1 h at 37°C. The proteins were analyzed by SDS-PAGE followed by silver staining or Western blotting.
SDS-PAGE, Blue Native PAGE, and Western Blotting-Env proteins were detected using SDS-PAGE and Western blotting procedures involving the anti-gp120 mAb ARP3119 (0.2 g/ml; Programme EVA Centre for AIDS Reagents) or the anti-gp41 mAb D50, HRP-labeled goat anti-mouse IgG (1:5,000 dilution) and the Western Lightning ECL system (PerkinElmer Life Sciences). Blue Native (BN)-PAGE was carried out using minor modifications to a published method (57). Purified proteins or cell culture supernatants were diluted with an equal volume of buffer (100 mM MOPS, 100 mM Tris-HCl, pH 7.7, 40% glycerol, and 0.1% Coomassie Blue), immediately prior to loading onto a 4 to 12% BisTris NuPAGE gel (Invitrogen). Usually, the gels were run for 2 h at 150 V (ϳ0.07 A) using 50 mM MOPS, 50 mM Tris, pH 7.7, as the buffer. Unless indicated otherwise, SDS-PAGE and BN-PAGE analyses were performed immediately after Endo H digestion.
Biophysical Analyses-Purified SOSIP.664 trimers (900 g) were treated with Endo H (Roche Applied Science) in 100 mM sodium acetate buffer, pH 6.0, at room temperature for 3 h. Endo H-treated or untreated SOSIP.664 trimers were analyzed on a Superose-6 10/30 column (GE Healthcare) using an AKTA Avant FPLC system (GE Healthcare). A size exclusion chromatography (SEC) system was coupled in-line with the following calibrated detection systems: 1) HP1 1050 Hewlett-Packard UV detector (Norwalk, CT); 2) MiniDawn Treos multi-angle light scattering (MALS) detector (Wyatt Corporation, CA); 3) quasielastic light scattering detector (Wyatt Corporation, CA); and 4) Optilab T-reX refractive index (RI) detector (Wyatt Corporation). These combined measurements allow both the absolute molar mass of an eluting glycoprotein, and the individual contributions made by protein and carbohydrate components, to be determined (58 -60). Briefly, the carbohydrate contribution to the molar mass can be assessed because glycans contribute directly to the RI signal but not the UV signal. Astra V software (Wyatt Corporation) was used to apply the protein conjugate template to the measured data. The extinction coefficient of the SOSIP.664 polypeptide, A 280 0.1% , was determined to be 1649 ml mg Ϫ1 cm Ϫ1 from its primary amino acid sequence (EXPASY). The specific refractive index increment values (dn/ dc) were estimated to be 0.168 and 0.179 for the mock-and Endo H-treated SOSIP.664 samples, respectively. The hydrodynamic radii, r H , of the mock-and Endo H-treated SOSIP.664 samples were determined from the quasi-elastic light scattering/RI signals of the samples eluting from the SEC column, using Astra V software.
MALDI-TOF MS Analysis-Peptides generated by trypsin digestion of SOSIP.681 GnTϪ/Ϫ trimers were detected by MALDI-TOF MS. Briefly, trimers were treated with Endo H under native or denaturing conditions, or with PNGase F under denaturing conditions, and diluted to 0.5 mg/ml before an overnight digestion with trypsin (0.3 g in 30 l of 25 mM ammonium bicarbonate) at 37°C. The reaction products were extracted and analyzed by MALDI-TOF MS in positive reflection mode (Autoflex; Billerica, MA). Glycopeptides and deglycosylated peptides were identified by comparing the MS data with the National Center for Biotechnology Information (NCBI) data base, and with the KNH1144 SOSIP.681 sequence, using Mascot (Matrix Science, London, UK, version 2.2.04). The experimentally identified peptides were matched to ones derived from an in silico (i.e. theoretical) trypsin digest of SOSIP.681, allowing for the following possible peptide modifi-cations: 1) oxidation of methionines; 2) deamidation of asparagines from which an N-glycan was removed by PNGase F; and 3) addition of a HexNac (GlcNAc) residue to asparagines from which an N-glycan was removed by Endo H. No fully glycosylated peptides could be detected in the MALDI-TOF analyses.
Immunoprecipitation Assays-Twenty-fold concentrated supernatants from transiently Env-transfected 293T or 293S GnTI Ϫ/Ϫ cells were incubated overnight at 4°C with mAbs or CD4-IgG2 in 500 l of RIPA buffer (150 mM NaCl, 1 mM EDTA, 50 mM Tris-HCl, pH 7.0). A 50-l volume of protein G-coated agarose beads (Pierce/Thermo Fisher Scientific) was then added, with rotation mixing, for 2 h at 4°C. The beads were washed extensively with ice-cold RIPA buffer containing 0.01% Tween 20 before proteins were eluted by heating for 5 min at 100°C in 50 l of SDS-PAGE loading buffer supplemented with 100 mM DTT. The immunoprecipitated proteins were analyzed on 8% SDS-PAGE gels (125 V, 2 h).
Electron Microscopy-Purified glycosylated or deglycosylated SOSIP.664 trimers (0.1 mg/ml) were applied to freshly glow-discharged, carbon-coated 400-mesh Cu grids (Ted Pella Inc., Redding, CA) and stained with 2% uranyl formate. The grids were viewed using an FEI Tecnai F20 electron microscope at a magnification of ϫ100,000 and an accelerating voltage tension of 120 kV. Images were acquired on a Gatan 4k ϫ 4k CCD camera at 0, 15, 30, 45, and 55°tilt angles using a defocus range of 600 to 720 nm. The tilt angles provided additional particle orientations to improve the image reconstructions. The pixel size of the CCD camera was calibrated at this magnification to be 1.09 Å using a two-dimensional catalase crystal with known cell parameters.
Image Processing-Particles were manually selected from micrographs using the DoGPicker package (61, 62). The contrast transfer function estimations for untilted micrographs were completed using the ctffind3 program and applied using the ACE2 program (61,63). Contrast transfer function estimations for the tilted micrographs and particles were calculated using ctftilt, and phases were flipped using applyctf in EMAN (61,64,65). Particles were extracted and binned by four into dimensions of 80 ϫ 80 pixels for calculating reference free class averages with the Sparx package (66). An ab initio three-dimensional image reconstruction was generated from each set of reference free class averages using the EMAN2 package (65,66). To test for model bias, each ab initio model was refined separately against the glycosylated and deglycosylated reference free class averages. In total, two models were produced for each set of reference free class averages. The two image reconstructions for each set were nearly indistinguishable, suggesting that refinement with the reference free class averages was able to overcome any possible model bias. This procedure was followed through 80 iterations of projection matching using the Sparx package (66). The final three-dimensional image reconstructions were calculated using images of single particles with a spatial resolution of 2.17-Å per pixel and refined using the EMAN package (64). C3 symmetry was imposed throughout the three-dimensional image reconstruction process. According to a Fourier shell correlation of 0.5 the resolution of the three-dimensional reconstruction was 14 Å for the untreated trimer and 20 Å for the Endo H-treated trimer.

Deglycosylation Strategy for Recombinant Env-
The considerable heterogeneity of the glycan structures present on HIV-1 Env compromises the efficiency of various deglycosylation strategies (17,42). In previous studies using monomeric gp120 or virions, many different glycosidases were tested under a range of experimental conditions, some of which are harsh enough to probably be incompatible with Env trimer stability (8). To avoid such problems, we adopted a two-stage strategy in which we first made trimers bearing a more homogeneous set of glycans that could then be digested by enzymes under mild conditions of temperature, salt, and pH (67). The specificities of the various enzymes that we evaluated are shown in Fig. 1A.
For optimizing deglycosylation conditions in pilot experiments, we used unpurified monomeric gp120 from the clade B strain JR-FL rather than purified, cleaved, disulfide-stabilized, soluble gp140 trimers. We then used the optimized protocol to deglycosylate the trimers. Because SOSIP gp140 from the clade A strain KNH1144 trimerizes more efficiently than the corresponding JR-FL gp140 (68), we used KNH1144 SOSIP.681 trimers for the deglycosylation studies (see "Materials and Methods" and Fig. 1B for more details). We also used a variant construct from which the MPER was deleted to prevent aggregation in the absence of detergent (SOSIP.664; Fig. 1B).
We previously described two methods for generating recombinant Env monomers and trimers that lack complex carbohydrates and, thus, contain only oligomannose sugars (69). First, we produced SOSIP.681 trimers in 293S GnTI Ϫ/Ϫ cells in which the attachment of GlcNAc to Man 5 GlcNAc 2 in the cis-Golgi is blocked; Man 5 GlcNAc 2 moieties are the most prevalent carbohydrate structure on the expressed glycoproteins. When we produced gp120 and trimeric SOSIP.681 in 293S GnTI Ϫ/Ϫ cells, they contained Man 5 GlcNAc 2 glycans, as expected (69). However, a considerable number of Man 6 -9 GlcNAc 2 moieties were also present because mannose residues were trimmed inefficiently at some sites (69). In our second approach, we treated 293T cells with the mannosidase inhibitor kifunensine and, thereby, produced Env proteins on which the predominant carbohydrate structure was Man 9 GlcNAc 2 (Fig. 1A) (69).
Deglycosylation of Oligomannose-enriched gp120-We first studied the effect of various enzymes on 293T cell-expressed JR-FL gp120 that contains a heterogeneous mixture of complex and oligomannose carbohydrates (designated gp120 WT ). The efficiency of deglycosylation was evaluated by SDS-PAGE and Western blot analysis, using size reduction as an indicator of glycan removal ( Fig. 2A). We note that SDS-PAGE analysis does not provide absolute values for glycoprotein sizes, because glycoproteins do not migrate on these gels in a way that is strictly proportional to their molecular weight (M r ) and, hence, to their glycan content. The glycosidase-mediated M r reductions, as measured on SDS-PAGE, are therefore merely apparent values that generally overestimate the extent of glycan loss.
For each enzyme, the buffer conditions recommended by the manufacturers were used, i.e. 50 mM sodium citrate, pH 4.5-5.5 for ␣1-2,3,6-mannosidase, Endo H and Endo F1. There was no detectable size reduction when gp120 WT was treated with ␣1-2-, ␣1-6-, or ␣1-2,3-mannosidase, implying either that the enzymes were inactive or that they removed too little carbohydrate for any size reduction to be visible using SDS-PAGE. In contrast, a small, but clear size reduction (ϳ5 kDa) was observed when gp120 WT was exposed to ␣1-2,3,6-mannosidase, which is consistent with our earlier findings (21,22) (Fig.  2A). Endo H or Endo F1 digestion lead to a slightly greater apparent molecular mass decrease of ϳ10 -15 kDa (Fig. 2A).
These enzymes had only limited effects because most sugars on gp120 WT are of the complex-type and, hence, not substrates; in addition, steric constraints may impede enzyme access to some N-glycans that would otherwise be targets.
The same set of enzymes were then used to treat JR-FL gp120 proteins that had been expressed in 293T cells in the presence of kifunensine (designated gp120 kif ); most N-glycans should, therefore, be of the Man 9 GlcNAc 2 form. The gp120 kif proteins were of a similar size to gp120 WT but migrated as a more compact band, which is consistent with their decreased glycan heterogeneity (compare the first lane of Fig. 2, A and B). There was, again, no detectable size difference when gp120 kif was exposed to either ␣1-2or ␣1-6-mannosidase (Fig. 2B). This outcome was expected with ␣1-6-mannosidase because Man 9 GlcNAc 2 glycans do not contain terminal ␣1-6-mannose residues. However, we had anticipated that ␣1-2-mannosidase would be active; each Man 9 GlcNAc 2 glycan theoretically contains four ␣1-2-linked mannoses, so this enzyme should reduce the size of gp120 kif by ϳ17 kDa (i.e. four mannoses, each of 0.18 kDa, removed from ϳ24 glycans). A reduction of this magnitude would be easily visible on an SDS-PAGE gel, and yet it was not observed. Although ␣1-2-mannosidase has been effective on other glycoprotein substrates (32), steric hindrances may limit its activity on gp120. We, therefore, abandoned the use of this enzyme. The size of gp120 kif was slightly reduced (ϳ5 kDa) by ␣1-2,3-mannosidase, and more so (an apparent shift of ϳ15 kDa) by ␣1-2,3,6-mannosidase. Treatment of gp120 kif with FIGURE 1. Schematics of Env constructs and glycosidase specificities. A, two different oligomannose N-glycans (Man 5 GlcNAc 2 and Man 9 GlcNAc 2 ) that are present on Env proteins produced in GnTl Ϫ/Ϫ cells are depicted, along with the cleavage sites for the enzymes used in this study. Env produced in the presence of kifunensine would only contain Man 9 GlcNAc 2 glycans. B, schematics of the gp120 and gp140 proteins. JR-FL gp120, conserved regions in blue, variable regions in green; KNH1144 gp140, conserved regions in red, variable regions in yellow. The gp140 was modified to reconstitute the 2G12 epitope (S295N substitution), and the 2F5 and 4E10 epitopes in the MPER (A662E, G664D, S668N, and N671T substitutions) (47). The gp140 also contains the A501C and T605C substitutions to create the SOS disulfide bond between gp120 and gp41 (44), the I559P mutation to promote trimerization (45), and a hexa-arginine cleavage site (R6) to enhance cleavage efficiency (54). The type of N-glycan indicated to be present at each glycosylation site (oligomannose, complex or undefined) is based on glycan characterization of IIIB gp120 (83). However, which type of glycan is actually present at some of these sites may vary between strains (83), and can also depend on the oligomerization status of Env proteins (69,84). The gp41 glycans have not been characterized.
Endo H or Endo F1 had a much more marked effect; the apparent reductions in the size of gp120 kif to ϳ60 kDa imply that these enzymes remove a considerable amount of its carbohydrate content.
Broadly comparable results were obtained using gp120 GnTϪ/Ϫ proteins produced in 293S GnTI Ϫ/Ϫ cells (Fig.  2C). ␣1-2and ␣1-6-Mannosidase had no detectable effect on the size of these proteins. This outcome was again unexpected because the gp120 GnTϪ/Ϫ glycans should have both ␣1-6mannose and ␣1-2-mannose moieties at their termini (albeit the latter on only a minority of glycans). Again, perhaps steric factors limit access of these enzymes to their substrates on gp120. As seen with gp120 kif , there was a small decrease in the size of gp120 GnTϪ/Ϫ upon exposure to ␣1-2,3-mannosidase or ␣1-2,3,6-mannosidase. The apparent decrease of ϳ10 kDa in the size of ␣1-2,3,6-mannosidase-treated gp120 GnTϪ/Ϫ was slightly less than the ϳ15 kDa shift seen with gp120 kif . The quantitative difference arises because there are fewer susceptible mannose residues on gp120 GnTϪ/Ϫ (which contains a mixture of Man 5-9 GlcNAc 2 glycans) than on gp120 kif (which contains only Man 9 GlcNAc 2 ). Finally, Endo H and Endo F1 reduced the size of gp120 GnTϪ/Ϫ to ϳ60 kDa, an outcome similar to when gp120 kif was treated with the same enzymes (Fig. 2,  B and C).
Deglycosylation of Oligomannose-enriched gp120 under Mild Conditions-The recommended buffers for optimal function of Endo H, Endo F1, and mannosidases use 50 mM sodium citrate at pH values in the range 4.5-5.5. Although the antigenic structure of gp120 is not compromised under such conditions (34), we considered it possible that gp140 trimers would be more pH-sensitive. We note that KNH1144 SOSIP.664 trimers do indeed dissociate below pH 5.0 (see below). To avoid such problems, we investigated whether endo-and exoglycosidases could still digest carbohydrates at a more physiological pH, by treating gp120 GnTϪ/Ϫ proteins with these enzymes for 2 h at room temperature in PBS, pH 7.2 (Fig. 2D).
There was no detectable effect of ␣1-2,3,6-mannosidase on the size of gp120 GnTϪ/Ϫ proteins under these reaction conditions (Fig. 2D), which contrasts with the size decrease seen using the optimal enzyme digestion buffer (Fig. 2, B and C). The differential outcome is consistent with reports that the optimal pH for this enzyme is in the range 3.5-4.5, with no activity at pH 7.2 (71). In contrast, Endo H was still able to reduce the size of gp120 GnTϪ/Ϫ from ϳ110 to ϳ60 kDa when the pH was increased to 7.2 (Fig. 2D). Hence, Endo H can remove a considerable amount of the carbohydrates from the glycan-modified gp120 GnTϪ/Ϫ proteins even under apparently suboptimal digestion conditions that better preserve the structure of gp140 trimers. Similar results were obtained when 293T cell culture medium, pH 7.6, was used instead of PBS (data not shown).
Deglycosylation of Oligomannose-enriched gp140 Trimers-Env trimer glycans on SIV virions are more resistant to glycosidase digestion than those on gp120 monomers, probably because steric factors impede the access of the enzymes to their substrates on the trimers (17). We therefore tested whether the native reaction conditions identified above would also work with soluble gp140 trimers. Cleaved KNH1144 SOSIP.681 gp140 trimers were produced in 293T cells in the absence (SOSIP.681 WT ) or presence of kifunensine (SOSIP.681 kif ), or in 293S GnTI Ϫ/Ϫ cells (SOSIP.681 GnTϪ/Ϫ ), as previously described (69). The various (unpurified) trimers were then treated with Endo H or Endo F1, fractionated by SDS-PAGE, and probed with the anti-gp120 mAb ARP3119 after Western blotting (Fig. 3A). As observed when monomeric gp120 was similarly treated, the gp120 subunits of both SOSIP.681 kif and SOSIP.681 GnTϪ/Ϫ (i.e. gp120 in the context of a trimer) were reduced in size to, apparently, ϳ60 kDa. For comparison, we denatured the same trimers using SDS plus DTT, and then treated them with Endo H using conditions optimal for enzyme activity (sodium citrate, pH 5.5). The resulting trimers were indistinguishable in size, i.e. ϳ60 kDa, from ones that had been exposed to the enzyme under nondenaturing conditions (Fig.   FIGURE 2. Deglycosylation of gp120 with endo-and exoglycosidases. The indicated glycosidases were used to treat the following gp120s under mild conditions. A, 293T cell-produced gp120 that contains both complex and oligomannose (i.e. heterogeneous) N-glycans; B, oligomannose-enriched gp120 kif from kifunensine-treated 293T cells (i.e. Man 9 GlcNAc 2 N-glycans); C, oligomannose-enriched gp120 GnTIϪ/Ϫ from 293S GnTI Ϫ/Ϫ cells (i.e. Man 6 Man 5-9 GlcNAc 2 N-glycans). The enzyme-treated gp120 variants were analyzed by reducing SDS-PAGE and Western blot. The positions of the molecular mass marker proteins and the approximate apparent molecular mass of the enzyme-treated gp120s are indicated; D, glycosidase treatment of gp120 GnTlϪ/Ϫ under mild conditions. The gp120s were analyzed by SDS-PAGE analysis and Western blotting. 3A). Hence, using nondenaturing conditions at neutral pH appears to be adequate for removing all the glycans from SOSIP.681 GnTϪ/Ϫ trimers that are actually Endo H-sensitive under the test conditions. However, we discuss below how biophysical techniques yielded additional information that caused us to refine the above conclusion.
To monitor the deglycosylation of gp41, which contains potential N-linked glycosylation sites at positions 611, 616, 625, and 637, we used the gp41-specific mAb D50 or HIVIg to probe Western blots after SDS-PAGE analysis. Endo H treatment of SOSIP.681 GnTϪ/Ϫ decreased the apparent size of the gp41 component by ϳ10 kDa, as detected by Asp-50 (Fig. 3B). However, a ladder of three additional gp41-related bands was detected when HIVIg was used for blotting (Fig. 3B). The different outcome probably reflects the more sensitive detection of trace amounts of partial digestion products when HIVIg is used. Furthermore, the detection of three additional bands implies that only three of the four potential glycosylation sites in gp41 are actually used. This supposition was confirmed by mass spectrometry studies (see below).
To further facilitate crystallization studies, we determined whether Endo H (in PBS at pH 7.2) was active on a complex of purified trimeric KNH1144 SOSIP.681 gp140 GnTϪ/Ϫ protein with sCD4 and mAb 17b. The trimers were deglycosylated to similar extents whether sCD4 was present or not (Fig. 3C). A residual, slower migrating band was apparent in the presence of 17b, suggesting that this mAb hinders Endo H access to one or more glycans. Overall, however, the sCD4-gp140 -17b ternary complex was efficiently deglycosylated by Endo H.
We used BN-PAGE to study the integrity and stability of mock-or Endo H-treated SOSIP.681 GnTϪ/Ϫ trimers, because deglycosylation can increase the tendency of proteins to aggregate, and/or cause oligomeric complexes to dissociate (72). The Endo H-treated, deglycosylated trimers migrated on BN-PAGE gels as a single species, with no indication of aggregate formation or dissociation into monomers (Fig. 3D).
Deglycosylation of Purified Oligomannose-enriched gp140 Trimers-The above studies were carried out using unpurified Env proteins in the supernatants of transiently transfected cells. To further refine the deglycosylation protocol, we next assessed the effectiveness of Endo H treatment of purified trimers. During the course of this study, we found that KNH1144 SOSIP gp140 trimers from which the MPER was deleted (SOSIP.664) adopt a monodisperse distribution in the absence of detergent. 5 This property makes them more suitable than SOSIP.681 trimers for BN-PAGE, biophysical analyses, and EM studies. We confirmed that Endo H removed similar amounts of glycans from KNH1144 SOSIP.664 and SOSIP.681 trimers, as judged by SDS-PAGE analysis (not shown). Hence we used SOSIP.664 trimers for most of the following analyses.
Using purified SOSIP.664 GnTϪ/Ϫ trimers as the substrate and Endo H as the enzyme, we evaluated how varying the incubation time, and using buffers of different pH, affected deglycosylation efficiency (Fig. 4). BN-PAGE analysis revealed that purified KNH1144 SOSIP.664 GnTϪ/Ϫ trimers were stable in the pH range 5.0 -9.5, but dissociated to monomeric gp140 when the pH was below 5.0. The incubation time and the pH both influenced the deglycosylation process, which was faster at pH 5.0 than 7.0. However, deglycosylation proceeded to the same final extent at the higher pH, provided the incubation time was increased (Fig. 4). Accordingly, we chose to perform the subsequent enzyme digestions for 3 h at pH 7.0, unless indicated otherwise.
Stability and Sensitivity of Deglycosylated gp140 Trimers to Proteolysis-Glycans can shield the underlying protein domains from proteases, so a deglycosylated protein might be atypically vulnerable to proteolytic digestion (23,42,72). We therefore compared the sensitivities of purified, deglycosylated, and mock-treated SOSIP.681 GnTϪ/Ϫ trimers to different concentrations (0.001-1 g/ml) of three common serine proteases: proteinase K, trypsin, and chymotrypsin. SDS-PAGE analysis followed by silver staining showed that the deglycosylated trimers were more sensitive to proteinase K (Fig. 5A). Thus, whereas the mock-treated trimers were resistant to digestion by 0.1 g/ml of proteinase K, and were only degraded at 1.0 g/ml (Fig. 5A, top panel), the deglycosylated trimers were completely degraded at 0.1 g/ml (Fig. 5A, bottom panel). In contrast, deglycosylation had no effect on trypsin and chymotrypsin sensitivity (Fig. 5, B and C). Overall, whereas glycan removal does increase the vulnerability of SOSIP.681 GnTϪ/Ϫ trimers to proteinase K degradation, the effect is not dramatic. Hence, the deglycosylated trimers should be stable enough to be used in immunization and crystallization studies.
Biophysical Characterization of Deglycosylated gp140 Trimers-To accurately determine the extent of deglycosylation achieved in our experiments, we separated purified KNH1144 SOSIP.664 GntlϪ/Ϫ gp140 trimers from gp140 monomers by size exclusion chromatography (SEC) on a Superose 6 column (Fig. 6A). A BN-PAGE analysis of the SEC fractions revealed that fractions D2 and D3 contain predominantly trimeric gp140, whereas mostly monomeric gp140 was present in fractions D4 and D5 (Fig. 6B). Fractions D5 and D6 also contained a contaminant, which was identified as BSA by mass spectrometry (not shown).
Determining how well gp140 trimers and gp140 monomers are deglycosylated under comparable conditions could reveal whether the trimerization interface influences the accessibility of some glycans to the enzyme. For this purpose, we analyzed fractions D2 and D3 (trimer) and D4 and D5 (monomer). When each fraction was treated with Endo H, BN-PAGE and SDS-PAGE analyses showed that the gp140 trimers and gp140 monomers were both deglycosylated to similar extents (Fig.  6B). The gp140 trimerization interface does not, therefore, shield a substantial subset of glycans from Endo H. A similar conclusion can be drawn from the experiments showing that the efficiency of glycan removal is similar whether the trimers are treated under native conditions or after prior denaturation (Figs. 3A and 6B). Untreated and deglycosylated trimers had similar SEC elution profiles, with no detectable differences in the distribution of the oligomeric species. However, the deglycosylated trimers eluted slightly later (elution volume 14.8 ml, compared with 14.2 ml), which is consistent with a decrease in their overall mass (Fig. 6C). In a quasi-elastic light scattering analysis, the hydrodynamic radii (r H ) of the untreated and Endo H-treated SOSIP.664 GnTϪ/Ϫ trimers were determined to be 12.3 and 9.9 nm, respectively, further evidence that an overall decrease in size occurs upon deglycosylation (Table 1).
In-line, MALS and RI detectors were used to make additional measurements during the SEC fractionations. Taken together  with the A 280 nm signal, these biophysical measurements allow the absolute molar mass of the eluting glycoprotein to be determined, as well as the contributions made by the individual protein and carbohydrate components. Together, the SEC-UV/ MALS/RI measurements show that the protein components of the untreated and deglycosylated SOSIP.664 GnTϪ/Ϫ trimers have the same molar mass of ϳ220 kDa (3 ϫ ϳ74 kDa for each monomer; Fig. 6C and Table 1). The overall reduction in the molar mass of the trimers (from ϳ390 to ϳ260 kDa) upon Endo H treatment is, therefore, due to a loss of carbohydrate mass (from ϳ175 to ϳ35 kDa, which is equivalent to ϳ12 kDa remaining carbohydrate per gp140 monomer).
The imprecision of the SEC-UV/MALS/RI technique is such that the ϳ35 kDa median estimate of the glycan molar mass remaining on the trimer could actually range from 18 to 52 kDa (corresponding to 0 -1 to 6 -7 Man 9 GlcNAc 2 glycans per gp140 monomer). This outcome results from intrinsic measurement errors (ϳ0.2-50%, as detailed in Table 2), and constrains the following discussions about the nature of the ϳ35 kDa of glycans that remain on each Endo H-treated gp140 trimer. First, if any O-linked sites are present, they would only contribute perhaps ϳ1 kDa of carbohydrate. Second, one GlcNAc moiety (203 Da) remains on every initially glycosylated asparagine residue; assuming every canonical glycosylation site is actually used (see below), the ϳ84 GlcNAc moieties would contribute ϳ17 kDa to the glycan mass of a SOSIP.664 trimer. Third, some trimer glycans may be inefficiently digested by, and/or inaccessible to, Endo H. Taking into account the ϳ17 kDa contributed by the nonremovable GlcNAc moieties, undigested carbohydrates could contribute ϳ18 kDa to the Endo H-treated trimer, or ϳ6 kDa per monomer. If the remaining glycans are of the Man 9 GlcNAc 2 type, which each have a mass of 1823 Da (excluding the protein-proximal GlcNAc already accounted for above), then Endo H fails to remove ϳ0 -7 of the 28 glycans on FIGURE 6. Biophysical analysis of SOSIP.664 GnTl؊/؊ trimers incubated with or without Endo H. A, untreated SOSIP.664 GnTlϪ/Ϫ trimers were fractionated using a Superose-6 column. The fractions were treated with or without Endo H and analyzed by reducing SDS-PAGE (B, left panels) or BN-PAGE (B, right panels). B, top panels, untreated trimers; middle panels, Endo H-treated trimers (3 h at 37°C, pH 5.5); bottom panels, denatured and Endo H-treated trimers. The species associated with each band are indicated. C, overlay of the A 280 UV absorbance traces from Superose-6 column fractionations of untreated (blue) and Endo H-treated (3 h at 37°C, pH 5.5; red) SOSIP.664 GnTlϪ/Ϫ trimers. UV/MALS/RI signals were recorded as the samples eluted from the column. Absolute molar masses calculated from these measurements are shown as thick blue and green connected dots under the main peaks for the samples obtained with and without Endo H treatment (see also Table 2). d, deglycosylated. each monomer. If some of the glycans contain fewer mannose moieties, that estimate would increase slightly. Again, we emphasize that this discussion is constrained by the considerable experimental error inherent in the measurement of what glycans remain undigested.
Mass Spectrometric Analysis of Deglycosylated Gp140 Trimers-Mass spectrometry analysis of tryptic peptides is a useful tool for assessing the utilization of glycosylation sites and the efficiency of deglycosylation protocols (73). We used MALDI-TOF to identify what N-glycans were removed by Endo H treatment. Because N-glycans are degraded during the course of MALDI-TOF spectrometry, glycan-containing peptides cannot be identified directly. However, the efficiency of deglycosylation can still be investigated by studying recovered peptides that contain modifications consistent with deglycosylation-related events. Thus, peptides from which PNGase F has removed an N-glycan can be distinguished from nonglycosylated peptides because they are deamidated and so have an aspartic acid in place of the original asparagine. Similarly, peptides from which Endo H has removed an N-glycan can be identified because one GlcNAc moiety remains attached to the asparagine residue.
We studied tryptic peptides derived from purified SOSIP.681 GnTϪ/Ϫ trimers that had been Endo H-treated either under mild conditions (see above) or after denaturation by heating for 10 min at 95°C in SDS plus DTT. As a further comparison, we also generated a third set of tryptic peptides from denatured SOSIP.681 GnTϪ/Ϫ trimers that had been treated with PNGase F. In each case, approximately half of the SOSIP.681 GnTϪ/Ϫ tryptic peptides that contain a glycosylation site could be identified (Table 2).
A notable finding from the tryptic peptide comparison is that all the glycans removed from denatured trimers by PNGase F or Endo H were also lost when Endo H was used under native conditions (Table 2). Thus, Endo H was able to deglycosylate trimers with similar efficiencies under either nondenaturing or denaturing conditions, and to an extent comparable with what was achieved using PNGase F under denaturing conditions (Fig. 7). In other words, every glycan moiety on SOSIP.681 GnTϪ/Ϫ trimers that is identifiable in the MALDI-TOF analyses and that is vulnerable to Endo H, is removed under mild conditions compatible with maintaining trimer integrity; prior denaturation of the trimers makes no discernible difference.
The tryptic peptides from the untreated SOSIP.681 GnTϪ/Ϫ sample suggested that the Asn-197 canonical site was not glycosylated. However, the presence of modified variants of the corresponding peptide in the PNGase F-and Endo H-treated samples implies that this site is, in fact, sometimes glycosylated. Moreover, when a glycan is present at the Asn-197 site, it is sensitive to these enzymes. The peptide containing the two potential gp41 glycosylation sites at positions Asn-611 and Asn-616 was found to contain only one new aspartic acid in the PNGase F-treated sample, and only a single added GlcNAc moiety in the Endo H-treated samples. Hence only one of the two sites at Asn-611 and Asn-616 is actually glycan-occupied, but we cannot identify which one. The glycan-free status of one For SEC-UV/MALS/RI analysis of glycoproteins, a typical error is Ϯ 10%. The source of uncertainty predominantly arises from the amount of sample analyzed and the stability of the system (59).
f Errors propagated in the various calculations performed by Astra V range between 0.2 and 50%.

Deglycosylated HIV-1 Env Trimers
of the four canonical sites on gp41 is consistent with the outcome of the SDS-PAGE analyses (Fig. 3B).
The MALDI-TOF analyses show that, of the 28 canonical glycosylation sites on each protomer of a SOSIP.681 GnTϪ/Ϫ trimer, one is not used at all (either Asn-611 or Asn-616 in gp41), and one only sometimes (Asn-197 in gp120). Endo H can remove glycans from 17 sites, those at gp120 positions 88, 130,136,146,156,188,193,197,234,241,276,301,448,463, and gp41 positions 611 or 616, 625 and 637 (Table 2). We can draw no definite conclusions about the remaining 10 sites, i.e. those at positions 262, 295, 332, 339, 355, 362, 386, 392, 396, and 411, which form a glycan cluster on the outer domain of gp120. The SEC-UV/MALS/RI analyses implied that between 0 and 7 sites are insensitive to Endo H. These two independent estimates are, therefore, in reasonable agreement.
Antigenic Structure of Deglycosylated gp140 Trimers-To study whether deglycosylation affects the antigenic structure of trimers, we used a panel of mAbs against gp120 and gp41 neutralization epitopes, and also reagents that bind glycans. Purified deglycosylated and mock-treated SOSIP.664 GnTϪ/Ϫ trimers were immunoprecipitated with the various reagents, and then analyzed by SDS-PAGE and Western blot. Most mAbs and the CD4 receptor analog CD4-IgG2 reacted equally efficiently with the untreated and deglycosylated trimers (Fig. 8A). There was an ϳ50% increase in b12 binding, in agreement with previous studies showing that glycans, in particular the one at position 386, can restrict access to the b12 epitope (74,75). Deglycosylation had no effect on basal or sCD4-induced interactions between the trimers and the coreceptor binding site-directed antibody 17b. There were also no apparent differences in the binding of the gp41-MPER directed mAbs 2F5 and 4E10 (Fig.  8A, lower panel). However, as expected, the glycan-dependent ligands 2G12 and DC-SIGN did not bind to the deglycosylated trimers (Fig. 8A, upper panel).
The V3 region of gp120 and its base are important determinants of coreceptor usage that also contain neutralization epitopes (76). Three glycosylation sites are located in or close to V3 (Fig. 1A). Eliminating the glycan from residue 301 at the base of V3 via mutagenesis (77)(78)(79), or generating smaller glycans by producing gp120 in GnT Ϫ/Ϫ cells (14), increases the accessibility of V3 neutralization epitopes to antibodies. Consistent with these observations, mAb 447-52D and 19b both bound more strongly to deglycosylated trimers than untreated ones ( Fig. 8B and data not shown). The epitope for 19b includes residues in both flanks of V3, whereas 447-52D recognizes the relatively conserved tip (80,81). There was also a modest increase in the binding of two other V3-directed mAbs, 5.8c and 1.4e (Fig. 8B). Finally, we tested the recently described glycan-dependent mAb PGT123 (29). This mAb interacted efficiently with untreated SOSIP.664 GnTϪ/Ϫ trimers, but not Endo H-treated Ϫ" indicates that the peptide was not found, which suggests the peptide is glycosylated and hence not detectable in the MALDI-TOF analyses; "ϩ" indicates that an (unglycosylated) peptide variant was detected. d PNGaseF-treated SOSIP.681 GnTlϪ/Ϫ ; "N3 D" indicates that one asparagine in the peptide was deamidated, indicative of prior glycosylation and subsequent removal by PNGaseF; "2N3 D" indicates that two asparagines in the peptide were deamidated. e EndoH-treated SOSIP.681 GnTlϪ/Ϫ ; "HexNAc" indicates that the peptide was found with an additional HexNAc (i.e. GlcNAc) attached, indicative of glycosylation and subsequent glycan removal by EndoH; "2HexNAc" indicates that two HexNAc moieties were present on the peptide.

Deglycosylated HIV-1 Env Trimers
ones, consistent with the loss of the glycan from site 332 that is a component of the PGT123 epitope (Fig. 8C).
Negative Stain Electron Microscopy Analysis of SOSIP. 664 GnTϪ/Ϫ Trimer Morphology-We used negative stain electron microscopy (Fig. 9A) and image reconstruction to visualize untreated and Endo H-treated SOSIP.664 GnTϪ/Ϫ trimers (Fig.  9B). Both trimers exhibit a similar morphology, with the principal differences arising at the periphery of the head domains. Thus, additional density, likely corresponding to glycans, protrudes from the head of the glycosylated untreated SOSIP.664 GnTϪ/Ϫ trimer, where the glycan-rich gp120 outer domains are expected to be located. Overall, the electron microscopy images demonstrate that the trimers are morphologically similar to one another, but with a clear size reduction between the fully glycosylated trimer and its deglycosylated counterpart (Fig. 9C).

DISCUSSION
Our goal was to devise a way to produce conformationally intact, deglycosylated soluble HIV-1 Env trimers to facilitate crystallography and immunogenicity studies. By using a combination of a cell line (293S GnTI Ϫ/Ϫ ) in which the addition of complex sugars does not occur, followed by enzymatic deglycosylation with Endo H under mild conditions, we could make gp140 trimers that lack most of their original N-linked glycan content. The remaining glycan structures are positioned on the trimer in orientations that appear to make them inaccessible to Endo H and other glycosidases.
The intrinsic flexibility and heterogeneity associated with carbohydrate moieties on glycoproteins make it challenging to crystallize such material, but we believe not impossible. In general, any reduction in the glycan content should, to some extent, increase the likelihood of a glycoprotein crystallizing, provided its structural integrity and stability are still maintained. The partially deglycosylated Env trimers are, therefore, more suitable than their fully glycosylated counterparts for crystallization studies. It is relevant that the Endo H-resistant glycans remaining on the partially deglycosylated trimers are relatively homogeneous, as they are all inaccessible to various carbohydrate-processing enzymes of the glycosylation pathway (and also produced in a cell Contour levels corresponding to experimentally determined masses by SEC-UV/MALS/RI (see "Materials and Methods") were calculated using the "volume" algorithm in EMAN. Untreated trimers were contoured using a volume of ϳ390 kDa, whereas the Endo H-treated trimers were contoured with a volume of ϳ260 kDa. The overall shapes of the trimers are conserved. C, overlay of untreated (blue mesh) and Endo H-treated (cyan surface) trimers, highlighting the conserved morphology and the size difference between the trimers.
line with a block in the glycosylation pathway). Overall, we believe that the partial deglycosylation of the trimers represents progress toward a crystallization substrate. Whether the procedures we have described here are sufficient remains to be determined, but we think we have characterized a necessary part of the process. For successful crystallization of the partially deglycosylated trimer, we are now testing whether it is necessary to also use ligands, such as antibodies, to provide additional crystal packing contacts in regions where glycans are still present. Glycoprotein deglycosylation can cause several adverse effects, including protein aggregation, the dissociation of oligomeric complexes, an increased susceptibility to proteolytic degradation, and a loss of native conformation (72). However, we found that the partially deglycosylated SOSIP.664 trimers neither dissociated nor aggregated. Hence the Endo H-accessible glycans are not major contributors to intersubunit interactions within the trimer. Moreover, their loss does not expose hydrophobic surfaces that can cause proteins to aggregate, something that has been suggested to occur with Env trimers (32,82). Glycan removal did increase protease sensitivity to a modest extent, implying that some scissile sites in the peptide backbone become more accessible. However, the partially deglycosylated trimers are robust enough to be used for various purposes without taking extraordinary precautions to prevent degradation.
The electron microscopy studies gave no indication that glycan removal adversely affected the overall conformation of the trimers. This is further evidence that the Endo H-accessible N-glycans do not contribute to the structural integrity of trimeric Env once the structure is assembled into its native conformation. The efficient binding of mAbs to conformation-sensitive, protein-only epitopes also suggests that the structural integrity of the partially deglycosylated trimers is preserved. Indeed, epitopes associated with the CD4 binding site (e.g. bNAb b12) and the V3 region (e.g. MAbs 19b and 447-52D) were more accessible on the deglycosylated trimers. The elimination of topologically proximal glycans by mutagenesis is known to increase the accessibility of the b12 and V3 epitopes on monomeric gp120. More distant glycans may also be involved in shielding these epitopes on the trimer, however. As expected, the binding sites for the glycan-dependent bNAbs 2G12 and PGT123 (as well as the C-type lectin DC-SIGN) were lost upon glycan removal. Clearly, NAbs of this particular specificity would not be induced by immunization with glycan-depleted trimers. Whether the favorable effects on the exposure of protein-only bNAb epitopes outweigh the loss of glycan-dependent bNAb epitopes remains to be determined.
Although there is likely to be some heterogeneity between individual trimers, the mass spectrometry data show that glycans could be removed from positions 88,130,136,146,156,188,193,197,234,241,276,301,448, and 463 on gp120, and from positions 611 or 616, 625, and 637 on gp41. The loss of glycan-301 would account for the increase in V3 mAb binding (14), whereas the loss of glycan-448 may explain why 2G12 no longer binds (34).
By implication, the remaining 10 sites on each protomer of the trimer may still contain glycans after Endo H treatment, provided of course that they were actually glycosylated in the first place. These sites are 262, 295, 332, 339, 355, 362, 386, 392, 396, and 411, most of which are part of an oligomannose patch on the outer domain of gp120. It is possible that some of the corresponding peptides went undetected in the MALDI-TOF analyses for technical reasons, and not because they were glycosylated. SEC-UV/MALS/RI assays independently imply that ϳ0 -7 (or slightly more if the glycans in question are smaller than Man 9 GlcNAc 2 ) sites remain glycosylated after Endo H digestion. The estimates from these two techniques seem reasonably concordant, with any discrepancies being attributable to the lack of detection by MALDI-TOF and/or measurement imprecisions in the SEC-UV/MALS/RI analyses.
One specific discrepancy relates to the glycan at site Asn-332, which is critical for binding of the PGT123 mAb. Indeed, alanine scanning mutagenesis suggests that no other glycan contributes to the PGT123 epitope (29). No peptide corresponding to the Asn-332 site was detected in the MALDI-TOF analyses, so this technique did not provide confirmation that Endo H removes the glycan from residue 332. Nonetheless, the lack of binding of PGT123 to the deglycosylated trimers implies that the Asn-332 glycan is, in fact, Endo H-sensitive.
We found that only three of the four canonical glycosylation sites in gp41 are utilized. In the MALDI-TOF analyses, the peptide containing the two canonical sites at residues 611 and 616 was occupied by only one glycan (which was removed by Endo H). Thus, the gp41 ectodomain on SOSIP.681 GnTϪ/Ϫ trimers contains glycans at positions 625 and 637 and a third that is located at either position 611 or 616. All three are removed by Endo H treatment under mild conditions, although some minor, partially deglycosylated forms of gp41 were detected (Fig. 4B).
In general, our findings are consistent with a recent report on PNGase F-mediated deglycosylation of uncleaved JR-FL and CON-S gp140 trimers (43). Thus, in both studies, a subset of the total glycans could not be removed. However, there are also some differences. For example, the asparagine at gp41 position 637 was not occupied on the uncleaved JR-FL and CON-S gp140 trimers, but is used on the cleaved KNH1144 trimers. In addition, we found that the glycans at positions 301 and 448 were removed, whereas they were not efficiently digested away from the uncleaved trimers (43). Whether differences in the design of the trimers (cleaved versus uncleaved, with or without the gp41 fusion peptide), the HIV-1 sequences on which they are based (KNH1144 versus JR-FL/CON-S), the deglycosylation protocol (Endo H versus PNGase F), or the analytical method used to assess deglycosylation, underlie these differences remains unknown.
In summary, we report partial deglycosylation of cleaved HIV-1 Env gp140 trimers under conditions that conserve their most fundamental features. The glycan-depleted trimers may be useful for both structural and immunogenicity studies.