|Home | About | Journals | Submit | Contact Us | Français|
Recombinant expression systems differ in the type of glycosylation they impart on expressed antigens such as the Human Immunodeficiency Virus Type-1 (HIV-1) envelope glycoproteins, potentially affecting their biological properties. We performed head-to-head antigenic, immunogenic and molecular profiling of two distantly-related Env surface (gp120) antigens produced in different systems: a) mammalian (293F) cells in the presence of kifunensine which impart only high mannose glycans; b) insect (Spodoptera frugiperda, Sf9) cells, which confer mainly paucimannosidic glycans; c) Sf9 cells recombinant for mammalian glycosylation enzymes (Sf9 Mimic™), which impart high mannose, hybrid and complex glycans without sialic acid; d) 293F cells, which impart high mannose, hybrid and complex glycans with sialic acid. Molecular models revealed a significant difference in gp120 glycan coverage between the Sf9- and wild-type mammalian cell-derived material that is predicted to impact upon ligand binding sites proximal to glycans. Modelling of solvent-exposed surface electrostatic potentials showed that sialic acid imparts a significant negative surface charge that may influence gp120 antigenicity and immunogenicity. Gp120 expressed in systems that do not incorporate sialic acid displayed increased ligand binding to the CD4-binding and CD4–induced sites compared to those expressed in the system that does, and imparted other more subtle differences in antigenicity in a gp120 subtype-specific manner. Non-sialic acid-containing gp120 was significantly more immunogenic than the sialyated version when administered in two different adjuvants, and induced higher titres of antibodies competing for CD4 binding site ligand-gp120 interaction. These findings suggest that non-sialic acid imparting systems yield gp120 immunogens with modified antigenic and immunogenic properties, considerations which should be considered when selecting expression systems for glycosylated antigens to be used for structure/function studies and for vaccine use.
The only viral target for neutralizing antibodies (NAb) against HIV-1 are the viral envelope glycoproteins (Env), comprised of a heterotrimer of three surface subunits (gp120) non-covalently linked to three transmembrane subunits (gp41). An ideal Env-based vaccine against HIV-1 would induce long-lived NAb responses recognizing a broad range of viral strains. To move towards this goal, considerable optimization of the antigen, adjuvant and delivery strategy will be required. This will include an understanding of the impact of Env glycosylation with regard to its effects upon antigenicity and immunogenicity. The choice of expression system might be especially relevant in the context of HIV-1 Env, since glycans contribute approximately 50% of the molecular mass of gp1201; 2; 3; 4; 5 and enzymatic glycosylation machinery differs substantially between, for example, insect- and mammalian-cell expression systems. When expressed in mammalian systems, between 13 and 16 of the N-linked glycans of gp120 are of the complex-type, while the remainder (of a total of 20–26) are of the high-mannose or hybrid-types2; 6; 7; 8. High-mannose glycans occur more frequently in the conserved regions, whereas those attached to variable loops, being more exposed to the glycosylation machinery of the Golgi, are more frequently processed into complex glycans6; 9. Conversely, in wild-type insect cell systems glycosylation is restricted to paucimannosidic structures10 and the degree of sequon occupancy may be lower11; 12.
Insect cell expression systems have the advantage of producing large quantities of proteins cheaply and without many of the potential biohazards associated with mammalian systems. In addition, insect cells carry out many post-translational modifications, including high-mannose type N- and O-linked glycosylation, resulting in glycoproteins generally considered to be of similar antigenicity and functionality to those prepared in mammalian cells13. When expressed in insect cell systems, Env gp160 undergoes appropriate cleavage to generate the gp41 and gp120 subunits, and the gp120 subunit retains CD4 binding14; 15; 16. For these reasons, insect cells have been used since the early years of HIV-1 vaccine research to produce Env antigens for immunogenicity and challenge studies in animals17 and immunogenicity in humans18. However, given the differences in glycan enzyme machinery between these different expression systems, the critera for selection of an optimal system might go beyond yield and ease of glycoprotein production, potentially affecting parameters such as antigenicity and immunogenicity.
Few studies have compared the impact of expression system on antigenicity in a head-to-head fashion, although one recent analysis compared gp120 expressed in different human cells lines for glycan content and relative binding to polyclonal sera8. By contrast, many have examined the effects of enzymatic removal of glycans or mutations deleting selected glycan sequons19; 20; 21; 22; 23; 24. While the outcomes of individual studies vary, together they suggest that partial or complete deglycosylation generally increases the binding of antisera and monoclonal antibodies (mAbs). However, the effects can be epitope-specific, with some regions being unaffected or even showing decreases in mAb binding25; 26. The site-specific deletion of glycan sequons from Env increased recognition by important neutralizing (N)mAbs such as b12, 447-52D and 2F5, especially when the mutants were expressed in insect cells20. Moreover, deletion of three-to-five Env glycan sequons from wild-type SIVmac239 virus resulted in dramatic attenuation in rhesus macaques and the induction of substantial protection against the wild-type virus with higher titers of NAbs than those obtained after wild-type virus challenge21; 22. Disappointingly however, subunit antigens based on such Env glycan mutants proved to be less good, or at best equivalent to, wild-type Env antigens at inducing NAb responses or controlling viremia in animal models19; 21; 23; 24. Moreover, since deletion of glycan sequons may alter glycoprotein processing and folding, this may have unpredictable effects on the presentation of conformational and discontinuous NAb epitopes such as those present on gp120. However one study showed that only a small number of protein-proximal glycan residues are important for folding while the rest may function to block antibody binding, suggesting that for immunogen design glycans may need to be re-engineered or otherwise modified rather than deleted27; 28. Such an approach has shown promise in a recent study in which complex glycans on gp120 were replaced with trimmed oligomannose structures by expression in a cell line lacking N-acetylglucosamine transferase I, resulting in increased binding of ligands to the CD4bs and the V3 loop27.
The choice of expression system might also affect the immunogenicity of glycoproteins such as gp120 in diverse ways. Immunogenicity might be reduced in insect cell-expressed gp120 since sialic acid residues on complex glycans, not present on insect cell-expressed material, are important for extending glycoprotein half-life in vivo by shielding mannose from interaction with mannose receptors29. Conversely, it has recently been demonstrated that mannose receptors on professional antigen presenting cells mediate uptake into compartments involved in cross-presentation: this function might be enhanced in the presence of insect cell-produced glycoproteins expressing predominantly terminal mannose glycans30. The latter supposition is in agreement with the finding that gp120 expressed in insect cells was better able to induce CTL responses in Balb/c mice than that produced in mammalian Chinese hamster ovary (CHO) cells31. The reduced response to the CHO-cell material was overcome by enzymatic deglycosylation, suggesting that mammalian glycans were responsible for the poor immunogenicity31. An alternative view was recently proposed that terminal mannose groups might downmodulate antibody responses to gp120 via lectin interactions on antigen presenting cells leading to production of the immunosuppressive cytokine IL-1032; 33. In addition to regulating interaction with cell surface lectins including mannose receptors, the sialic acid residues of complex glycans also negatively regulate the interaction of gp120 with mannose-binding lectin in the serum34, which upon binding to an antigen can trigger the complement cascade resulting in complement-opsonization and improved antigen uptake. Finally, sialic acid has been demonstrated to suppress B cell responses via interaction with CD22, a potential mechanism to avoid self-recognition35; 36.
To investigate the effect of expression system on glycoprotein antigenicity and immunogenicity, we compared two insect systems: wild-type Sf9 (Sf9wt) and Sf9 Mimic™, with a mammalian system: 293 FreeStyle™ (293F) in the presence and absence of kifunensine, in a head-to-head fashion using gp120 from two distantly-related HIV-1 strains in order to describe both general and virus strain-specific effects. Sf9 Mimic™ cells are a recombinant Sf9 cell line that express five mammalian glycosylation enzymes and produce the majority of complex mammalian glycan modifications10 with the exception that they lack a donor for sialic acid and thus the complex glycans they produce have terminal galactose residues37. The inclusion of this additional cell line allows the contribution of complex glycans lacking sialic acid to antigenicity and immunogenicity to be assessed without the need for enzymatic desialyation.
HIV-1 is a highly diverse virus with strains differing by up to 20% within clades and 35% between clades in terms of the amino acid sequence, with Env being the most variable gene38. To study general effects on recombinant gp120 antigenicity and immunogenicity of the expression system used, we selected CCR5-tropic strains from two different clades: 97CN54, a CRF07_BC primary isolate in which the gp120 region, with the exception of part of the leader sequence, is entirely of clade C origin39; 40 (accession number: AF286226) and the clade B isolate Ba-L41 (accession number: AB221005). Alignment of the amino acid sequences of gp120 from these two strains using ClustalW (v1.83)42, showed that the strains are 26.3% divergent, with gp120CN54 having 14 additional amino acids: 10 extra residues in the V1 loop and 6 in the V2 loop, but 2 fewer in the V4 loop (data not shown). Analysis of potential sites of N- and O-linked glycosylation using N-Glycosite43 and Net-O-Glyc (v3.1)44 revealed that gp120CN54 has 23 sequons for N-linked glycosylation and no predicted sites for mucin-type O-linked glycosylation, whereas gp120Ba-L has 22 sequons for N-linked glycosylation and 1 potential site for mucin-type O-linked glycosylation. 15 of the N-linked glycosylation sites were conserved between the two strains with 12 occurring in conserved regions of the glycoprotein. Gp120CN54 has an additional N-linked glycosylation sequon in each of the V1- and V2-loops and the C4-region but one fewer in the V4-loop and C3-region when compared to gp120Ba-L (data not shown).
To inform our modeling analysis of glycan coverage we carried out mass spectrometric analysis of the glycan types present on gp120Ba-L produced in Sf9 cells, untreated 293F cells and 293F cells treated with 5 and 20 μm kifunensine, an inhibitor of an inhibitor of class I α -mannosidases. We confirm results from a previous study10, that Sf9-expressed gp120 contains mostly Man3GlcNAc2 but retains a minor population of untrimmed oligomannose structures that includes the 2G12 NmAb epitope (Fig.1). Mammalian cell-expressed gp120 contained the expected proportions of complex and high-mannose glycans, implying that the purification process did not impose any dramatic bias in the selection of glycan types on the different glycoprotein forms. It has been reported that insect cell-expressed material may have reduced numbers of glycans due to sequon skipping11; 12. Reduced numbers of glycans would be expected to modify the antigenicity of a highly-glycosylated protein, and might also influence immunogenicity. To limit differences between Sf9wt- and 293F-expressed gp120 to glycosylation type rather than number, we expressed gp120Ba-L in 293F cells in the presence or absence of 5 or 20μM kifunensine. To confirm that kifunensine treatment inhibited the processing of Man9GlcNAc2, we performed mass spectrometry on gp120 produced in the presence and absence of the inhibitor (Fig. 1). Cultivation of gp120Ba-L-expressing 293F cells in the presence of 5 μM kifunensine completely prevented complex glycosylation and resulted in the unprocessed Man9GlcNAc2 being the most prevalent glycan type (29.1%). However, the inhibition of class I α-mannosidases was incomplete as the remaining species in order of abundance were: Man8GlcNAc2, Man7GlcNAc2, Man5GlcNAc2 and Man6GlcNAc2. Increasing the kifunensine concentration to 20 μM better inhibited the mannosidases as only three glycan types could be detected: Man9GlcNAc2, Man8GlcNAc2 and Man7GlcNAc2.
Potential differences in the antigenicity and immunogenicity of gp120 may result from differential glycan type, conformation, sequon usage, or all of these. To investigate the first possibility we constructed molecular models of gp120 based upon predicted N-linked glycosylation of the 293F, kifunensine-treated 293F and Sf9wt cell systems as described in materials and methods. A caveat for these models is that we were unable to include glycan mobility as a major factor in protein surface masking, since little is known of the flexibility of clusters of glycans in dense populations as found on gp120. In the absence of precise assignments for each glycan position, we modeled glycan heterogeneity by representing the proportion of each glycan species determined by mass spectrometry, and distributed them using a random assignment algorithm as detailed in methods. In this way, we generated three models for each kind of expression system. Molecular dynamics was used to optimize the stereochemistry for each of these models, allowing for a limited account of flexibility. The exact assignments of glycans to the models are shown in Tables S1–3. Glycans conferred by Sf9 Mimic™ cells are essentially the same as those conferred by mammalian cells except for sialic acid termini. Therefore, models of glycan coverage from 293F cells should be largely representative for glycans from Sf9 Mimic™ cells in terms of protein coverage, but not in terms of surface charge. Glycans conferred by kifunensine-treated 293F cells are retained as oligomannose structures whereas those conferred by Sf9 cells are principally processed to Man3F1-containing structures with a minor population of oligomannose structures.
A representative of the refined models for gp120HxBc2 expressed in 293F, kifunensine-treated 293F and Sf9 cells are shown in Figure 2a, b and c respectively. The extent of gp120 glycosylation is evident from the models, in which the protein core is labelled red, glycans containing terminal mannose are cyan, and complex glycans green. Using all these models, glycan coverage was quantified by calculating the amount of protein surface exposed to hypothetical spherical probes of different radii, approximating to the penetration of an amino acid side chain (1.4Å), beta turns (2.5Å and 5Å) and an antibody combining region (10Å) (Fig. 3). Although models for the same expression system vary greatly in glycan composition (Tables S1–3), they are quite similar to each other in terms of glycan coverage. The occlusion of the solvent-exposed protein surface by the glycan canopy was extensive in all expression systems, and the difference in surface coverage between gp120 from kifunensine treated cells and gp120 from 293F cells was not significant for small ligands of radii equivalent to amino-acid side-chains or polypeptide beta turns, components that can make up portions of an antigen-Ab combing region interface. Only when we modelled the interaction with a probe of radius 10Å, equivalent to an entire antibody-combining region, did a significant difference in total percentage occlusion of the solvent-exposed protein surface emerge between kifunensine-treated and untreated 293F-expressed material (P < 0.01). However, this difference amounted only to an extra 8% total coverage (a 10.5% increase in coverage) compared to gp120 produced in the kifunensine system. A more profound difference in coverage was observed for gp120 produced in the Sf9 system, which revealed a 10% reduction in total coverage (a 13.6% reduction in coverage) compared to the 293F cell-produced gp120. This decrease in protein surface coverage imparted by the Sf9 cell production system would probably exert substantial effects on the exposure of ligand binding sites proximal to glycans, such as the IgG1b12 antibody epitope and the CD4bs45.
Another property that sialic acid imparts onto a glycoprotein that might influence antibody binding and/or immunogenicity by altering interaction with antigen presenting cells, is increased negative charge, which would be lacking on insect cell-expressed material. We therefore modelled how the typical Sf9wt- and 293F-imparted glycan structures affect the protein surface electrostatic potential (Fig. 4 a and b). As predicted, gp120 expressed in untreated mammalian cells is modelled to have a net negative average electrostatic potential of −2.1 k/Te. However, when expressed in other systems not imparting sialic acid, the solvent-exposed protein surface had an average weak positive charge, a reversal of the electrostatic potential and a difference of 2.3kT/e (P< 0.001).
All env constructs were subcloned into the vector pTri-Ex1.1 allowing expression from the same plasmids in both insect and mammalian systems. Equal amounts of gp120CN54 expressed in 293F, Sf9wt and Sf9 Mimic™ and gp120Ba-L expressed in 293F and Sf9wt, as determined by BCA assay, were probed by Western blot for purity and assayed by ELISA with a panel of mAbs, the pAb D7324, sCD4, and polyclonal antisera, for expression system-dependent antigenic differences. The median 50% binding titers obtained from 3 independent ELISAs for each condition were established and are represented in Tables 1 and and22 as fold-changes in ligand binding to Sf9wt or Sf9 Mimic™-expressed material compared to 293F-expressed material. Antigenic differences were detected in most regions in both strains of gp120, although some differences in antigenicity were specific for the individual gp120s. For gp120Ba-L, the binding of the NmAbs IgG1b12, which binds an epitope overlapping the CD4bs45 and 2G12 which binds a conserved glycan epitope, were significantly increased by expression in Sf9wt cells (Table 1). For gp120CN54, expression in the Sf9-systems significantly decreased binding of a C1-region mAb and the NmAb 447-52D, which binds a V3-loop epitope. As is common in a number of C-clade isolates, 2G12 failed to bind to gp120CN54 expressed in any system46; 47. Both gp120s showed substantial and significantly increased binding of sCD4 (8.5- and 15.7-fold for Ba-L and CN54 respectively), suggesting a dramatic increase in exposure of this surface on the Sf9-expressed material, which is in accord with the enhanced binding of IgG1b12 to gp120Ba-L and computational modelling of glycan coverage (Fig 2). By contrast, recognition of both strains of Sf9-produced gp120 by polyclonal antisera raised against mammalian cell-expressed gp120 (ARP421 and ARP422) was significantly reduced. MAb binding to the CD4-induced (CD4i) surface was also increased for insect cell-derived compared to mammalian cell-derived antigen, and increases in CD4i-site mAb binding were generally greater when sCD4 was present (Table 2), probably reflecting the increased sCD4 binding. Similarly, binding of mAb A32 to its complex conformational epitope was also increased for both strains when expressed in the Sf9wt system, and this binding was further enhanced for gp120CN54 in the presence of sCD4. The Sf9 Mimic™ and Sf9wt systems imparted broadly similar antigenic profiles to gp120CN54: differences in the binding of the CD4i-site mAb E51 in the absence of sCD4 and the C1–C4 region-specific mAb A32 in the absence or presence of sCD4 were not significant (Table 2). These data suggest that the presence of more complex glycans is not sufficient, in the absence of sialic acid, to recapitulate the mammalian cell glycosylation phenotype.
Having established that there were gp120 strain-dependent and -independent antigenic differences between gp120s expressed in the various systems, we tested their immunogenicity in Balb/c mice (Fig 5). Since HIV-1 gp120 is a weakly immunogenic glycoprotein with potential immunosuppressive properties32, we carried out immunization of mice with antigen in extrinsic adjuvant, and compared two different adjuvants to control for the outcome of adjuvant-specific effects on gp120 immunogenicity observed by others33; 48. Groups of mice were primed subcutaneously with 10 μg of gp120CN54 produced in Sf9, Sf9 Mimic™ or 293F cells, and gp120Ba-L produced in Sf9 or 293F cells formulated in a CpG-containing commercial adjuvant that triggers innate immune activation via TLR-949, and boosted three weeks later with the same amount of antigen alone. Serum samples were assayed by ELISA against a CHO mammalian cell-expressed trimeric CN54 gp140 glycoprotein. We assayed against glycoprotein produced in a distinct system for two reasons: i) to avoid assay bias that might occur when using glycoproteins produced in the original expression systems used to generate the test antigens; ii) to reduce any potential cross-reactivity present in the sera against contaminants of the expression system unrelated to the recombinant antigen under analysis. We found that antigens expressed in Sf9 and Sf9 Mimic™ cells induced significantly greater antigen-specific serum IgG responses compared to 293F-expressed gp120 (Fig 5a). We subsequently immunized mice with a subset of the mammalian and insect cell-derived antigens in Freund's Complete Adjuvant (FCA), an potent adjuvant that acts largely independently of TLR receptor mediated mechanisms50, and hence has a different mode of action to the CpG adjuvant used previously. As before, the titres of antigen-specific IgG produced against CN54 and Ba-L gp120 produced in insect cells were significantly higher than those against the gp120s produced in mammalian cells (Fig 5b). We therefore conclude that insect cell-derived gp120 is more immunogenic than its mammalian cell-expressed counterpart irrespective of the type of adjuvant used or its mode of action.
Since the antigenicity analysis demonstrated that certain epitopes were differentially recognized on the insect compared to the mammalian cell-expressed gp120, we considered that this might translate into differential recognition of these epitopes by B cells in vivo. We therefore analyzed sera derived from mice immunized with gp120Ba-L or gp120CN54 produced in the two systems for their ability to compete with mAbs or soluble receptor probes of defined specificity. For the competition assay we used CHO cell-derived gp140CN54 detection as before. Initial titrations of the sera into this ELISA determined the optimal dilutions for use in the competition assay (data not shown). Because the volumes of mouse serum were limiting we were only able to analyze competition with a small number of probes, and so we chose ligands that bound gp140CN54 with high avidity that were specific for the CD4bs (mAb HJ16 and CD4IgG2) and mAbs specific for the gp120 V2 and V3 loops (HG68 and HR10 respectively). Figure 6 shows that sera from the mice immunized with Sf9-derived material competed significantly more with both CD4bs-specific reagents than sera from mice immunized with 293F-derived material. These results are in line with the antigenicity data presented in Table 1 that demonstrate increased binding of CD4bs-specific probes to insect cell-derived compared to mammalian cell-derived gp120. A similar relative increase in competition was seen with gp120CN54-derived serum for the V3 mAb but not for the V2 mAb, demonstrating that the increased competition observed with sera derived from insect cell-produced gp120 was not global in nature, but restricted to specific epitopes. Only modest competition was observed for the V2 and V3 loop mAbs with sera from gp120Ba-L-immunized mice: this result was to be expected since the gp140CN54 detection antigen would be unlikely to contain many appropriately cross-reactive epitopes within these hypervariable regions.
The antigenicity of the 5 and 20 μM kifunensin-treated gp120Ba-L was probed with mAbs, sCD4 and a recombinant immunoglobulin-CD4 chimeric protein expressing 4 gp120 binding sites, CD4-IgG251: very similar results were obtained for both and so only the 5 μM data are shown here. Few significant differences were observed for the wt compared to kifunensin-treated gp120 (Table 3). An exception was the CD4bs ligands IgG1b12 and CD4-IgG2 that showed significantly increased binding to the kifunensine-treated material. This is consistent with the increased binding of ligands to the CD4bs observed with insect cell-derived compared to mammalian cell-derived material. In addition, the binding of the V3-loop mAb 19b and the glycan-specific broadly neutralising mAb 2G12 were significantly increased by the treatment.
The immunogenicity of gp120 from untreated wt and kifunensine-treated 293F cells was analysed by immunization of Balb/c mice in the CpG-based adjuvant, and sera were titrated on both untreated and kifunensine-treated immunogens to take into account assay bias introduced by the modification. Sera were tested for antigen-specific IgG 3 weeks after the antigen-alone boost at week 6. Antigen-specific IgG responses were detected against gp120 from both wt and kifunensine-treated 293F cells (Fig. 7). The responses against gp120 from the kifunensine-treated cells were greater than those against wt gp120, but the differences did not reach significance when titrated on wt gp120. By contrast, mice immunized with gp120 from the kifunensine-treated cells achieved a 3.8-fold higher IgG titre against the kifunensine-treated gp120 compared to wt gp120, and this difference was significant (P = 0.0496). This result implies that eliminating sialic acid by kifunensine treatment increased antibody responses to gp120, but this was only detectable using kifunensine-treated material in the ELISA, presumably reflecting recognition of epitopes better exposed on the modified compared to unmodified gp120.
In the present study we have examined the influence of expression system on the antigenic and immunogenic characteristics of HIV-1 gp120, and constructed 3D models to facilitate our understanding of mechanisms underlying these differences. Production of gp120 in insect cells resulted in substantially different antigenicity from the equivalent material produced in mammalian cells, and immunogenicity that was increased compared to mammaliam cell-expressed gp120 (Fig. 8). These results have obvious implications for the design of vaccine antigens based upon expression of recombinant HIV-1 Env. Firstly, alterations in antigenicity will influence the exposure and/or conformation of particular epitopes, and therefore their presentation to B cells, which will influence the specificity of the antibody response. This might be beneficial or detrimental for induction of NAb, since alterations in antigenicity will modify both neutralization-relevant and neutralization-irrelevant epitopes. Gp120 expression in insect cell systems preferentially increased mAb and soluble receptor binding to the CD4bs and CD4i surfaces, probably as a result of decreased glycan bulk increasing exposure of the underlying protein surface. Since IgG1b12 is a broadly neutralizing mAb recognising a conserved surface on gp12045, the increase in the binding of this NmAb to gp120Ba-L expressed in insect cells might favour the induction of similar specificities by immunization. In this respect our competition ELISA data support the idea that the CD4bs is not only more exposed on insect cell-produced gp120, but is also more immunogenic, a finding of potential importance for antibody-based vaccine design. Although thus far it has not proved possible to recapitulate IgG1b12-like responses in vivo by immunization, such responses have been detected in HIV-1-infected individuals, suggesting that this region is immunogenic52. Since increased IgG1b12 binding to insect cell-expressed gp120 was viral isolate-specific, occurring in gp120Ba-L but not gp120CN54, this may be epitope context-dependent and not a broadly applicable observation. Indeed, virus strain selectivity of IgG1b12 binding to glycan variants may explain why a previous study found that gp120 derived from the IIIB isolate produced under kifunensine treatment showed increased binding of the 2G12 antibody compared to wt gp120, but failed to show substantially increased IgG1b12 binding53. However, the improved recognition of the CD4bs by sCD4 on both Ba-L and CN54 insect-derived gp120 may stimulate increased presentation of other neutralizing epitopes on this relatively conserved surface to B cells in a more generalized manner, such as that recognized by the novel CD4bs-specific mAbs HJ1654 and VRC01, VRC02 and VRC0355. Kifunensine treatment of gp120Ba-L increased binding of IgG1b12 and CD4-IgG2, consistent with the reduction in glycan bulk rendering the CD4bs more accessible to ligands. The increased binding of NmAb 2G12 to gp120Ba-L is also of interest for vaccine development, and may reflect a higher avidity of 2G12 for its epitope when the gp120 expresses a higher proportion of terminal mannose glycans. Consistent with this interpretation, kifunensine treatment also increased 2G12 binding to gp120Ba-L, in a manner analogous to that previously described for enhanced 2G12 recognition of kifunensine-treated mammalian cells and mammalian cell-produced gp12053. The increase in CD4i epitope exposure was, with two exceptions (X5 on gp120CN54 and 17b on gp120Ba-L), generally applicable to both gp120s. It may be that reducing the bulk of the glycans allows greater exposure of CD4i surfaces that are otherwise masked in mammalian cell-expressed material. A second explanation may be that increased CD4i mAb binding reflects increased numbers of gp120 with bound sCD4. Alternatively, reducing glycan complexity and charge may influence folding of gp120 molecules. Regardless of the mechanism, it seems unlikely that the CD4i surface is a useful target for induction of neutralizing antibodies as it is largely inaccessible to IgG on the native viral spike either prior to, or during viral entry56. Our analysis shows that glycan structure will impact upon analyses of glycoprotein structure and function, and should be taken into account. One obvious example from our work is that the affinity of sCD4 for insect cell-derived material is substantially higher than for mammalian cell-derived material, and may bias interpretations of HIV-1 Env-receptor interaction energetics. Caution should therefore be exercised when extrapolating from data acquired using expression systems that differ from those in which virus would normally be produced in vivo.
We enriched the insect cell and mammalian cell-derived gp120s using a Galanthus Nivalis lectin column that has high affinity for mannose groups. We cannot exclude that this enrichment selected for specific glycoforms of gp120 that could influence the antigenicity of the material. However, since both insect and mammalian cell-derived gp120 naturally present clusters of terminal mannose groups, it seems unlikely that this would substantially bias selection of particular variants. This conclusion is consistent with the MS analysis which suggests that the glycan composition of the differentially-produced gp120s reflects that described by others using different purification strategies10; 57; 58; 59. It is possible that a proportion of the gp120 populations that we have probed antigenically may consist of mixtures of conformers or potentially even some misfolded material, leading at least in part, to the differences in antigenicity observed. However, one of the principal tools for probing the conformational integrity of an antigen is by antibody binding to conformational and discontinuous epitopes. Since all forms of gp120 tested here reacted with all conformation-dependent mAbs and with sCD4, we assume that a major fraction of the glycoproteins is most likely to be properly folded. Some differences observed in antigenicity were env isolate-dependent rather than expression system-dependent, adding weight to the idea that the differences observed were due to intrinsic structural features of the glycoproteins rather than protein misfolding.
Our finding that the insect cell-expressed gp120 antigens are more immunogenic in the context of adjuvant than the same antigens expressed in mammalian cells is of interest for vaccine development. To our knowledge only one other study has compared immunogenicity of insect cell (Drosophila S2) and mammalian cell-derived (293F) gp12048. Grunder and colleagues administered mammalian cell-expressed gp120 in Ribi adjuvant with insect cell-expressed wt or gp120 containing additional T-helper epitopes, administered in Ribi and FCA/FIA adjuvants. In this study, immunogenicity appeared to be dependent both upon the adjuvant used, since no immunogenicity for the S2-derived gp120 was observed in Ribi whereas robust immunogenicity was observed for the same glycoprotein in FCA/FIA, and also on the presence of the T helper epitope48. However, further comparison between the studies is difficult since Grundner et al assayed sera pooled from each group and hence were unable to carry out quantitative statistical analysis. Another study compared de-mannosylated mammalian cell-derived gp120 with its untreated counterpart, which differs from our analysis of insect compared to mammalian cell-produced gp120, and concluded that de-mannosylation increased immunogenicity in an adjuvant-dependent context33. Thus when administered in Alum there was a significant increase in immunogenicity of the demannosylated material compared to the untreated gp120, but this difference disappeared when administered in QuilA adjuvant33. Our conclusion is therefore that gp120 immunogenicity is dependent upon both the presence and proportion of terminal mannose-containing glycans, and on the adjuvant used to elicit the immune response. Our findings of increased humoral immunogenicity with insect cell-expressed gp120 concur with prior findings of increased cellular immunogenicity31 and support the idea that the insect cell-expressed gp120 may be processed and presented more efficiently by antigen presenting cells in vivo.
Full consideration of the effects of expression system on the antigenicity and immunogenicity of a candidate vaccine antigen should be taken into account when assessing which system should be employed. This is particularly true for antigens based upon HIV-1 Env, which is a highly conformational, highly glycosylated antigen in development as a recombinant subunit immunogen for eliciting neutralizing antibodies. In this respect the choice of adjuvant will be of central importance, since this will influence the folding and presentation of the antigen to the immune system. Based upon our data, an insect cell system or kifunensine treatment of mammalian-cell-derived material, which increases immunogenicity while simultaneously increasing the presentation of epitopes including those of IgG1b12 and 2G12 might be beneficial. However, the final choice must be further informed by the relative ability to induce durable, high-titer and broadly active NAb responses and cellular immune responses, a question best assessed empirically in vivo.
Antibodies (Abs), mAbs and other reagents were obtained from the same sources as previously described60, with the exception of: CD4-IgG251 manufactured by Progenics Inc. and obtained from The International AIDS Vaccine Inititative Neutralizing Antibody Consortium Repository; CD4-binding site (CD4bs)-specific mAbs 15e and 21h obtained from the National Institute of Health (NIH) AIDS Research and Reference Reagent Program; the rabbit antiserum to gp120, ARP421 obtained from the Centralised Facility for AIDS Reagents (CFAR, NIBSC, Potters Bar, UK. Donated by S, S. Ranjbar); D7324, the goat polyclonal Ab (pAb) to the gp120-C-terminal region, from Aalto Bio Reagents Ltd (Dublin, Ireland); human mAbs HJ16 (CD4bs), HR10 (V3) and HG68 (V2) from D. Corti and A. Lanzavecchia54. Anti-species immunoglobulin-HRP conjugates were obtained from Jackson ImmunoResearch Europe Ltd (Soham, UK). HIV-1 env coding for soluble gp140CN54, gp120CN54 and gp120Ba-L were subcloned into the vector pTri-Ex1.1 (Invitrogen) with a C-terminal 6 × His tag. Vectors were expressed in 293 FreeStyle™ (293F) cells following mass transfection and gp120CN54 and gp120Ba-L in Sf9 and Sf9 Mimic™ cells (Sf9wt and Sf9 Mimic™ respectively) following infection with recombinant baculoviruses. Gp120Ba-L was expressed in 293F cells in the presence or absence of 5 or 20 µM kifunensine (TRC, Canada). Glycoproteins were purified by a combination of lectin followed by nickel (reactive with 6 × His-tag) affinity chromatography. Briefly, supernatants were initially enriched on the Galanthus nivalis sepharose column (Vector Laboratories), followed by buffer exchange with ProBond native binding buffer (Invitrogen) supplemented with 10 mM imidazole and purification using the Invitrogen ProBond purification system. Eluted glycoproteins were buffer exchanged with 20 mM Tris pH 7.4, aliquotted and frozen at −70°C until use. Glycoprotein purity was confirmed by SDS-PAGE and Coomassie blue staining, and all preparations were estimated to be >90% pure.
8–12 week old Balb/c mice were obtained from the specified pathogen-free animal breeding facility of the Sir William Dunn School of Pathology (University of Oxford, UK). All experiments were performed under appropriate licence in accordance with the UK Animals (Scientific Procedures) Act 1986. Antigens were screened for endotoxin content and contained ≤ 1 EU.mL−1. Antigens at doses of 2, 5 or 10 μg as described in figure legends were formulated in PBS alone, emulsified with Freund's Complete Adjuvant (FCA) at a 1:1 volume ratio, or with the CpG-based adjuvants ImmunEasy (Qiagen) (10 μL/vaccine dose) or CpG 1018 (MWG Biotech) (50 μg/vaccine dose) formulated in 25 kDa, branched PEI (Sigma-Aldrich) at an N:P ratio of 15. The immunogen preparations were made up to 100μL using sterile, endotoxin free 5% (w/v) D-glucose (Sigma-Aldrich) and injected subcutaneously into the right flank. Typically, immunizations were given at weeks 0, 3 and 6, with test tail bleeds (50 μL) collected immediately prior to each immunization.
ELISAs were performed as described previously60 except the plates were washed six times following the addition of antisera or mAbs and after the addition of the anti-species IgG HRP conjugates. The absorbance of pre-immune serum was subtracted from that of the post vaccine bleeds before calculation of the endpoint titre by interpolation of the point of intersect between the assay cut-off and a sigmoid dose-response curve fitted to the dilution series in GraphPad Prism v5.061.
Antigens produced in the various expression systems were coated on to ELISA plates and probed with the anti-Env Abs or CD4-IgG2 as described. For the sCD4 binding study, sCD4 was coated onto the ELISA plate at concentrations ranging from 0.31 to 20 nM. The gp120 antigens were then added at a concentration of 1 μg/mL−1 and coincubated for 1 h at room temperature (RT) before the addition of 10 μL/well of sample buffer containing six-fold the median binding titer of ARP422, such that the final concentration of ARP422 equated its 50% binding titer. After a further 1 h at RT the plates were washed and the ARP422 detected using anti-rabbit-IgG-HRP followed by the standard ELISA development protocol. For the assays in which sCD4 was added before the ligand, gp120-coated wells were incubated with 50 μL of a 2 μg/mL solution of sCD4 for 30 min at RT before the addition of 50 μL of a 2-fold concentrated dilution series of the ligand without washing out the sCD4. Titrations were performed in 3–5 replicates per plate and on at least two independent occasions, and comparisons were made between the median relative binding avidity in nM−1 or the 50% binding titer in the case of antisera. Ligands that did not reach saturating levels were compared by analysis of the absorbance obtained at the highest concentration of ligand used. The mAb competition ELISA to assay for specificity in immune mouse serum was carried out as follows. CHO-derived CN54 gp140 was coated onto the ELISA plate and the plate blocked and washed as described previously60. Test or control (immune mouse serum from mice immunized with hen egg lysozyme formulated in FCA) serum diluted to 1/50 in 50 μL PBS/1% BSA was added to the plate for 1 h at RT, followed by one of the competition mAbs to yield a final concentration previously determined to give 50% binding in this assay format. Plates were washed 3 times and bound mAb detected with anti-human IgG-HRP at 1/1000 in PBS/1%BSA for 1 h at RT followed by washing, addition of substrate, stopping and reading as previously described60. The % inhibition was calculated using 100-[(test value-control value)/(maximum value-control value) × 100].
We produced models of gp120 as it would be produced by expression in Sf9, untreated 293F or kifunensine-treated 293F cells. To mimic gp120 produced in 293F cells under kifunensine treatment, high mannose glycans, which have been characterized on gp120 of the HIV-1BH8 isolate expressed in insect cells9, were modeled on the structure of gp120HxBc2 (PDB ID = 1GC161). N-linked glycan sites at amino acids 289, 397, 406, 448 and 463 of gp120HxBc2 do not have conserved counterparts in gp120BH8, so glycans on gp120BH8 closest to those amino acids were incorporated. Yeh et al. found that each site contains a distribution of glycans ranging in size from five to nine mannose residues9 that closely matched the distribution of glycans on gp120 produced from kifunensine-treated cells as determined by mass spectrometry (Figure 1). A method was incorporated to account for this micro-heterogeneity. Each N-linked glycan was assigned a numeric range equal to the percent that this glycan is found at a site. This was done for all glycans at each site. Next, each site was given a random number between 0 and 100. A particular glycan was modeled on a site if the random number fell within that glycan's numeric range. This method was used to generate three variants of gp120 from kifunensine-treated cells (Table S1). To mimic wild-type 293F glycosylation, complex and high mannose glycans, which were characterized on gp120HxBc2 expressed in CHO cells6 and chronically-infected lymphoblastoid (H9) cells7, were modeled on the structure of gp120HxBc2. Leonard et al. determined whether each N-linked glycan site contains a high mannose or a complex glycan6. To assign glycans on gp120 and to account for microheterogeneity, a method similar to the one described above for Sf9 glycan modeling was developed. Each N-linked glycan was assigned a numeric range equal to the molar percentage of the chromatography fraction, in which the glycan was characterized. The numeric ranges for complex glycans were separated from those for high mannose glycans, forming two groups. The ranges in each group were normalized to 100. Because several chromatography fractions from the attempt to characterize the glycan types remained unknown, their molar percentages were evenly distributed among the other, known fractions. Again, each N-linked glycan site was given a random number between 0 and 100. Each site was also labeled high mannose or complex depending on the type of glycan determined by Leonard et al. A particular glycan was modeled on a site if its random number fell within that glycan's numeric range and if the site's label corresponded to the glycan's type. Three variants of gp120 from 293F cells were generated with this method (Table S2). To mimic Sf9 glycosylation, all complex sugars were stripped from the models for 293F glycosylation and replaced by Man3F1. Furthermore, all high mannose sugars containing nine mannose residues were trimmed down to six mannose residues. This was done so that the model would better match the mass spectrometry data on gp120 expressed from Sf9 cells (Table S3). Glycans were modeled on the structure of gp120HxBc2 using Glyprot (http://www.glycosciences.de/modeling/glyprot/php/main.php). These models were refined in two steps using CNS62; 63. In the first step, they underwent simulated annealing, in which they were heated up to 2500–3000 K, freeing them from the local energy minimum. Then, they were slow cooled in steps of 25 – 30 K so that they approached a global energy minimum. Three trials were done for each condition and the condition that yielded models that differed most from the starting model was selected. In the second step, a model from the selected condition was subject to further molecular dynamics at 310 K for up to 106 steps at 10 fs steps until its rmsd from the initial model approached a plateau. This effectively placed the model in an energy-minimized region at physiological temperature.
To quantify the extent of glycan coverage of protein surface, the refined models were analyzed with spherical probes of different radii as previously described64. A 1.4 probe approximates the penetration of a side chain and a 2.5 or 5.0 probe approximates that of a beta hairpin turn. A 10 probe approximates to the reach of an entire antibody combining region. These probes were used to calculate the accessible surface of the models. Using GRASP, the accessible surface area excluded by the glycans was determined by finding the surface area of the polypeptide within carbon's van der Waal's radius, or 1.7 , of all glycan surfaces. These values were normalized over the accessible surface area of the entire polypeptide to obtain the percent glycan coverage.
Analyses were performed using GraphPad Prism v 5.0. Raw or log10-transformed data were only treated as normally distributed if this was first demonstrated by the Kolmogorov-Smirnov normality test with a Dallal-Wilkinson-Lilliefor P-value. For normally-distributed data, comparisons of more than two groups were carried out using the one-way ANOVA test. Comparison of two data sets used unpaired t-tests. For data that were not normally-distributed, Kruskal-Wallis analysis was performed to compare more than two data sets. The Mann-Whitney test was used to compare two data sets. Bonferroni's correction was applied when multiple two-group comparisons were made. Error bars represent the range unless stated otherwise.
We thank JE. Robinson, DR. Burton, D. Katinger, S. Zolla-Pazner, D. Corti and A Lanzavecchia for their generous gifts of mAbs, and the NIBSC CFAR, NIH AIDS Reagent Program and the International AIDS Vaccine Initiative (IAVI) Neutralizing Antibody Consortium (NAC) Repository for reagent supply. We acknowledge the support of the UK Medical Research Council (grant G0000635), the IAVI NAC and Fondation Dormeur. QJS is a Jenner Institute Fellow.
Publisher's Disclaimer: This is a PDF file of an unedited manuscript that has been accepted for publication. As a service to our customers we are providing this early version of the manuscript. The manuscript will undergo copyediting, typesetting, and review of the resulting proof before it is published in its final citable form. Please note that during the production process errors may be discovered which could affect the content, and all legal disclaimers that apply to the journal pertain.
The authors have no conflicting commercial interests relevant to the work published herein.
The nucleotide sequences for HIV-1 97CN001 (CN54), Ba-L and HxBc2 have been deposited in the GenBank database under the GenBankAccession Numbers AF286226, AB221005 and K03455 respectively. The atomic coordinates for the crystal structure of gp120HxBc2 are available in the Molecular Modeling Database (http://www.ncbi.nlm.nih.gov/structure/MMDB/mmdb.shtml) under MMDB # 8099.