Introduction of a Glycosylation Site into a Secreted Protein Provides Evidence for an Alternative Antigen Processing Pathway: Transport of Precursors of Major Histocompatability Complex Class I–Restricted Peptides from the Endoplasmic Reticulum to the Cytosol

Recombinant vaccinia viruses (rVVs) expressing K^d, TAP (1 and 2), and influenza virus nucleoprotein (NP) have been described (23, 24). A rVV expressing ICP47 (25) was provided by B. Rouse (University of Tennessee, Knoxville, TN). The full-length NP gene modified by standard molecular genetic methods to encode molecules listed in Table 1 was inserted into the thymidine kinase locus of vaccinia viruses (VVs) by homologous recombination as described using the pSC11 plasmid to express foreign proteins under the control of the VV p7.5 early/late promoter (26). NP was directed to the secretory pathway using the signal sequence from β interferon (27). The NP-coding sequences of all of the rVVs were verified by sequencing PCR-amplified copies of full-length NP genes isolated from the rVV.

6–8-wk-old BALB/cByJ (H-2^d) and CBA/J (H-2^k) mice were obtained from Jackson Labs (Bar Harbor, ME). Mice were immunized with PR8 by intraperitoneal injection, and with rVV by intravenous injection. Target cells were infected with viruses as described previously (28). To generate T_CD8+, splenocytes derived from animals immunized with viruses 2–6 wk previously were stimulated in vitro for 7 d with either antigenic peptides at 1 μg/ml or virus-infected autologous splenocytes as described (29). In some cases, T_CD8+ lines were used in place of secondary cultures. Lines were propagated by weekly stimulation with L929-K^d cells that had been incubated for 3 h at 37°C with IMDM supplemented with 20 μg/ml human β₂-microglobulin (Sigma Chemical Co., St. Louis, MO) and 1.2 μg/ml synthetic peptide, fixed for 10 min at room temperature with 0.5% paraformaldehyde in PBS, and washed by three consecutive 30-min incubations with IMDM at 37°C. 1–2 d after stimulation with cells, fresh IMDM supplemented with 10% (vol/vol) supernatant from rat splenocytes stimulated with Con A was added. This stimulation protocol was repeated weekly. Microcytotoxicity assays were performed as previously described (28). Where indicated, brefeldin A (BFA; Sigma Chemical Co.) was added to cells at a concentration of 5 μg/ml before assay and at a concentration of 2.5 μg/ml during the cytotoxicity assay to block further antigen presentation (29). Data are expressed as percent-specific release defined as:

Experimental release − medium release/ total release − medium release × 100.

Synthetic peptides were provided by the Biological Resources Branch (National Institute of Allergy and Infectious Diseases, Bethesda, MD). Glycopeptides were produced as described (18). All peptides were purified by HPLC and analyzed by mass spectroscopy to confirm purity and mass.

2 × 10 ⁷ L929 cells infected for 1 h with VV were incubated in 2 ml Met-free DMEM for 30 min at 37°C. After adding 200 μCi [³⁵S]methionine in 200 μl Met-free DMEM, cells were incubated for an additional 10 min at 37°C. After washing with PBS, cells were incubated for 20 min at 0°C with 0.14 M NaCl, 1 mM EDTA, 1% Triton X-100, 1 mM PMSF, and 10 mM Tris HCl, pH 7.4, and the nuclei were pelletted by centrifugation at 15,000 g. Supernatants were incubated with protein A–Sepharose preloaded with the anti-NP mAb H16-L10. After extensive washing, beads were boiled in sample buffer according to Laemmli (30), or in 0.1 M sodium acetate, 3 mM EDTA, and 0.25% SDS, pH 6. In the latter case, samples were split in half, incubated overnight at 37°C with or without 2 mU endo-β-N-acetylglucosaminidase (endo H). All samples were analyzed by SDS-PAGE according to Laemmli. After fixing, gels were dried under vacuum and exposed to Phosphorimager screens (Molecular Dynamics, Sunnyvale, CA), which were imaged using the associated hardware and software. For quantitation of viral gene expression in cells used for the microcytoxicity assay, an aliquot of 10⁶ cells was removed, incubated in methionine-deficient medium for 15 min, and labeled with [³⁵S]-methionine for 45 s. Labeling was terminated by the addition of Laemmli sample buffer heated to 100°C. Samples were analyzed by SDS-PAGE, and the amount of protein in fixed gels determined using a Phosphorimager (Molecular Dynamics). Images of gels were arranged and labeled with Adobe Photoshop (Adobe Systems, Mountain View, CA) and printed with a Fujix 3000 pictography digital printer (Fuji Photo Film USA, Elmsford, NY).

As a model antigen we used influenza virus NP synthesized by rVV. NP is synthesized on free ribosomes in the cytosol and efficiently translocated to the nucleus due to a nuclear localization sequence. In H-2^d mice, the T_CD8+ response to NP focuses on an immunodominant K^d-restricted peptide located at residues 147–155 (TYQRTRALV). Conversion of the Gln₁₄₉ to Asn creates a N-linked glycosylation site. To study the effects of glycosylation on antigen processing, we produced rVVs expressing NP with Gln₁₄₉ or Asn₁₄₉ directed to the ER by appendage of a NH₂-terminal ER insertion sequence (termed secreted NP [SNP]; Table 1). We controlled for effects of the Gln₁₄₉→ Asn substitution on peptide liberation, class I binding, and TCR recognition by using rVVs expressing the same proteins in the cytosol. To examine whether any alterations in presentation of SNP^N are related to glycosylation, we produced rVVs expressing cytosolic or ER-targeted NPs in which the glycosylation site created by the Gln₁₄₉→ Asn substitution is destroyed by converting Thr₁₅₁ to Ala (Table 1).

Obviously, for this to be a useful strategy, the amino acid substitution alone cannot adversely affect the binding of the peptides to K^d. This was examined in two ways. First, RMA/S K^d cells incubated at 26°C to accumulate cell-surface peptide–receptive molecules were exposed to decreasing amounts of peptides for an hour at 26°C, and then for 2 h at 37°C to denature class I molecules lacking peptides. BFA was included with peptides to prevent the transport of additional K^d molecules to the cell surface. Conformed class I molecules were detected using the mAb SF1.1.1 conjugated to FITC. The peptide concentration required to preserve half of the cell-surface K^d molecules provides an approximation of peptide affinity for K^d. This revealed that the Asn₁₄₉- and Asn₁₄₉Ala₁₅₁-containing peptides have K_as approximately 2.5–3-fold higher than the wild-type peptide (K_a = 1/[peptide concentration], yielding values for variant and wild-type peptides, respectively, of 1–1.2 × 10⁷ M⁻¹ versus 4.2 × 10⁶ M⁻¹) (Fig. 1). By contrast, a peptide containing Asp₁₄₉ (which would result from PNGase-mediated removal of the N-linked oligosaccharide from Asn₁₄₉) was bound with an estimated K_a 230-fold lower than the wild-type peptide (1.8 × 10⁴ M⁻¹).

The relative affinities of the peptides were confirmed by examining the thermal stability of cell-surface K^d molecules loaded with a saturating concentration of peptide at 26°C and incubated at 37°C for up to 8 h in BFA in the absence of peptides. This provides a measure of the off-rate for peptide binding, which is the major factor in peptide affinity for class I molecules. K^d molecules were most efficiently stabilized by the N₁₄₉-containing peptides, and were only marginally stabilized by the Asp₁₄₉-containing peptide.

In the experiments that follow, we used T_CD8+ populations or lines raised to the wild-type sequence as a measure of presentation of the 147–155 determinant from the different cytosolic or ER-targeted NPs. The effects of the amino acid substitutions on peptide antigenicity varied somewhat with the T_CD8+ population or line used. In two of the experiments described below, the Asn₁₄₉- and Asn₁₄₉Ala₁₅₁-containing peptides sensitized target cells for lysis by NP-specific T_CD8+ with similar or slightly greater efficiencies than the wild-type peptide (see Figs. 3 and 5). The Asp₁₄₉-containing peptide routinely required 100–1,000 times the concentration of Gln₁₄₉- or Asn₁₄₉-containing peptides to achieve a similar level of lysis (not shown).

To confirm that attachment of an N-linked oligosaccharide to Asn₁₄₉ abolishes peptide antigenicity, we coupled Asn₁₄₉ in the TYNRTRALV synthetic peptide to the disaccharide maltobiose. This resulted in at least a 5 log₁₀-fold reduction in the capacity of the peptide to sensitize cells for lysis (not shown). Oligosaccharides added in the ER are much larger than this disaccharide, consisting of between 10 and 14 saccharide subunits, depending on the degree of modification by ER glucosidases and mannosidases (31), and would be expected to interfere to an even greater extent than maltobiose with peptide binding to class I molecules or interaction with the TCR. These findings demonstrate that (a) we can meaningfully compare presentation of variant and wild-type peptides using T_CD8+ effectors raised to the wild-type sequence, and, (b) glycosylation will destroy peptide antigenicity due either to the presence of the N-linked oligosaccharide, or after its removal by PNGase, the poor antigenicity of the Asp₁₄₉ containing peptide.

The function of the ER-targeting sequence was confirmed by SDS-PAGE analysis of total cell extracts of rVV-infected cells metabolically labeled with [³⁵S]methionine. As seen in Fig. 2,A, the secreted forms of NP migrated more slowly than the cytosolic/nuclear forms. SNP and SNP^NRA comigrated, whereas SNP^N migrated more slowly, which is consistent with the existence of one site for N-linked glycosylation in SNP and SNP^NRA (NP has a natural glycosylation site at position 21) and two sites in SNP^N. This was confirmed by digestion of immunocollected SNP and SNP^N with endo H, which eradicated the difference in mobility between the ER-targeted and nontargeted forms of NP (Fig. 2 B). It is important to note that all three forms of SNP were efficiently delivered to the ER and glycosylated at all potential sites, since we did not detect, by either immunoprecipitation or by Western blotting (not shown), full-length forms that were nonglycosylated or partially glycosylated. Upon prolonged chasing at 37°C, all three forms of SNP became endo H–resistant and were secreted into the medium with a t_1/2 of ∼3 h (not shown).

These biochemical findings were confirmed using anti-NP mAbs in indirect immunofluorescence of fixed and permeabilized rVV-infected cells (not shown). Non-ER–targeted forms were largely localized to the nucleus as expected, whereas secreted forms were located in the ER and Golgi complex. Upon prolonged infection (16 h), weak NP-specific staining was detected in the nuclei of cells expressing SNP, SNP^N, or SNP^NRA (“SNPs”).

Together, these findings demonstrate that (a) SNPs are efficiently targeted to the ER, (b) all potential glycosylation sites are used in SNPs to within the limits of detection, and (c) a minor portion of SNPs make their way to the nucleus.

The efficiency of generating the NP_147-155 peptide from the six different forms of NP was determined using a NP-specific T_CD8+ line that recognizes the wild type, Asn₁₄₉- and Asn₁₄₉Ala₁₅₁-containing peptides with similar efficiency (Fig. 3, inset). The efficiency of peptide generation was estimated by adding BFA to L-K^d cells at increasing intervals after infection with rVVs (Fig. 3; 32). As with other proteins, targeting NP to the ER reduced the efficiency of peptide generation. Modification of amino acids 149 or 151 had little effect on presentation of cytosolic NP. The same alterations, however, had clear effects on presentation of SNP. The Gln₁₄₉→ Asn substitution significantly reduced presentation of SNP. The similar levels of presentation of SNP^NRA and SNP indicates that glycosylation of Asn₁₄₉ is responsible for decreased presentation of SNP^N relative to SNP.

To eliminate the possibility that the differences in antigen presentation reflected differences in levels of expression of the various antigens (and not real differences in antigen processing), we quantitated the amounts of NP produced by the target cells used in this experiment. Cells were radiolabeled with [³⁵S]methionine for 45 s and immediately immersed in boiling SDS-PAGE sample buffer. This method allows recovery of virtually all of the biosynthesized proteins, and not just those extracted into relatively mild detergents that are capable of binding antibody (the technique most commonly used to quantitate antigen expression). SDS-PAGE of extracts revealed that newly synthesized NP could be easily detected due to the strength of the VV p7.5 promoter combined with the ability of VV to shut down host cell protein synthesis. As seen in Fig. 2,A, the difference in presentation between the secreted forms of NP cannot be attributed to differences in levels of expression. Indeed, increasing the amount of SNP^N synthesized by increasing the virus dose by threefold (SNP^N_High) did not increase presentation to T_CD8+ (Fig. 3). NP-specific T_CD8+ from H-2^k mice recognize residues 50–57 in association with K^k (33). Unlike the K^d-restricted determinant, the presentation of NP_50-57 was not adversely affected by the Gln₁₄₉→ Asn substitution in SNP (Fig. 4). In fact, VV-SNP^N–infected cells were recognized at slightly higher levels by K^k-restricted T_CD8+ than cells expressing SNP, whereas the very same target cells gave the reverse hierarchy of lysis using K^d-restricted T_CD8+. This provides functional evidence that the diminished recognition of VV-SNP^N–infected cells relative to VV-SNP–infected cells is not related to diminished levels of SNP^N expression. Moreover, this experiment indicates that the inhibitory effect of glycosylation on SNP antigen processing is limited to the determinant containing the glycosylation site.

We next compared the presentation of the various rVVs to primary T_CD8+ in vivo by determining the amount of virus required to prime for secondary in vitro responses to synthetic peptides (Fig. 3 B). Targeting NP to the secretory pathway did not significantly alter its immunogenicity. Remarkably, introducing the Gln₁₄₉→ Asn substitution into SNP decreased the immunogenicity of the rVV 100-fold relative to the other viruses (in additional experiments we detected priming after inoculation with 10⁶ plaque-forming units/ mouse [not shown]). This effect was specific for the secreted form, since the same substitution actually increased the immunogenicity of cytosolic NP 10-fold. The low immunogenicity of VV-SNP^N cannot be attributed to poor infectivity in vivo since it primed for VV-specific responses with similar efficiency as the other rVVs (not shown).

These findings demonstrate that the presence of an N-linked glycosylation site in a secreted protein specifically affects its antigenicity and immunogenicity, presumably due to the attachment of an N-linked oligosaccharide in the ER. Creation of the same site in a cytosolic version of the protein has no significant effect on antigenicity.

The involvement of TAP in the presentation of NP, NP^N, SNP, and SNP^N was examined using T2-K^d cells, a TAP-deficient human lymphoblastoid cell line expressing K^d from a transfected gene (34; Fig. 5,A). None of the corresponding rVVs were able to sensitize T2-K^d cells for lysis by NP-specific T_CD8+. The precise hierarchy of presentation of the various gene products observed in L-K^d cells was faithfully restored, however, by coinfection with VV-TAP (1 and 2), which encodes both human TAP subunits (35). Lysis of SNP^N-expressing cells was at levels less than half that observed with SNP. This cannot be attributed to less efficient recognition of the Asn₁₄₉-containing peptide by the T_CD8+ line used since NP^N was presented slightly more efficiently than NP and the synthetic peptides were recognized with similar efficiency. Nor can it be attributed to decreased expression of SNP^N relative to SNP. As in the previous experiment, cells were briefly radiolabeled and the expression of NPs quantitated after SDS-PAGE of whole cell lysates. This demonstrated that SNP^N was expressed in similar amounts to SNP (Fig. 5 legend). To eliminate the possibility that a minor population of cells synthesized most of the antigen, NP expression was examined by immunoperoxidase staining of fixed and permeabilized target cells. As seen in Fig. 5, a similar percentage of cells expressed the various forms of NP (SNP^N was actually detected in a greater percentage of cells than SNP).

These findings demonstrate that the TAP-dependent presentation of the 147-155 peptide derived from SNP is blocked by glycosylation. TAP has recently been shown to directly bind to class I molecules (3, 4), and its potential as a molecular chaperone in class I folding has been suggested (36). Thus, it is possible that the TAP dependence of presentation of the NP_147-155 determinant reflects a role for TAP in facilitating the association of K^d with the free peptide in the ER lumen, and not the transport of 147-155 peptide from the cytosol. To examine this possibility, we used an rVV expressing the herpes simplex virus protein ICP47 to specifically inhibit the transport function of TAP (23, 37– 40). Coexpression of TAP with ICP47 blocked the TAP-mediated rescue of both NP and SNP (Fig. 5 B).

These findings demonstrate that the 147-155 determinant produced from SNP is delivered to class I molecules by TAP from the cytosol, and that this presentation is partially blocked by glycosylation.

Based on its dependence on TAP-mediated peptide transport, we conclude that antigenic peptides from SNP are produced in the cytosol. Substituting Asn for Gln₁₄₉ in SNP specifically decreases the immunogenicity and antigenicity of the 147-155 determinant. This cannot be attributed to negative effects of the substitution on peptide affinity for K^d or TCR recognition. Nor can it be attributed to diminished TAP-mediated transport or liberation by cellular proteases of the Asn₁₄₉-containing peptide, since NP^N is presented with similar or even greater efficiency than wild-type NP. Rather, these findings implicate glycosylation as the cause of reduced antigenicity, which is confirmed by the enhanced presentation observed when the glycosylation site is destroyed by substituting Ala for Thr₁₅₁. Since the Asn₁₄₉-containing peptide in cytosolic NP is presented normally, the effect of glycosylation on SNP^N must occur before TAP-mediated transport. This indicates that most antigenic peptides from SNP are derived from protein that is delivered to the ER and then routed back to the cytosol where it is processed by cytosolic proteases into peptides that are delivered to K^d molecules via TAP.

A synthetic version of the Asn₁₄₉-substituted 147-155 peptide is glycosylated in a TAP-dependent manner when added to cells permeabilized by streptolysin O (our unpublished results). Thus, we interpret the efficient presentation NP^N to indicate that the Asn₁₄₉-containing peptide associates with K^d before the peptide can be glycosylated. This is consistent with the idea that TAP delivers peptides directly to the class I molecules they bind (3, 4). The practical importance of this is that the antigenicity and immunogenicity of peptides in cytosolic/nuclear proteins are probably not affected by the presence of a consensus site for N-linked glycosylation.

By contrast, the antigenicity, and particularly the immunogenicity of SNP, were greatly affected by the presence of a glycosylation site, indicating that glycosylation can have a major negative effect on the antigenicity of peptides in ER-targeted proteins. The disproportionate effect of the Gln₁₄₉⇒ Asn substitution on SNP immunogenicity relative to its effect on antigenicity indicates that presentation of antigen to primary T cells in vivo is fundamentally different from in vitro presentation. This may reflect differences in antigen processing in the APCs used in vivo versus the various tumor cells we have used. An intriguing possibility that we favor is that in vivo presentation of SNP relies largely on processing of exogenous antigen scavenged from infected cells by professional APCs (41). Exogenous antigens would consist of the metabolically stable proteins produced by cells, and in this case, the very large difference between presentation of SNP and SNP^N would be expected based on the high efficiency of glycosylation of Asn₁₄₉.

The present results do not distinguish in which secretory organelle egress of SNP to the cytosol occurs (anywhere from ER to lysosomes is possible). Cells are able to present exogenous antigens in a TAP-dependent manner, due to release of pinocytosed material to the cytosol. This pathway is largely limited to phagocytosing cells, however. Moreover, in additional experiments (not shown), we found that presentation of SNP could not be attributed to uptake of secreted material, since coincubation of VV-SNP–infected cells with ⁵¹Cr-labeled uninfected cells during the course of a 6-h cytotoxicity assay did not result in the lysis of uninfected cells (not shown). It was recently shown that two cytomegalovirus proteins, US2 and US11, have the remarkable ability to transport newly synthesized class I molecules from the ER to the cytosol, probably by inducing reverse translocation through the translocon (10, 11). The authors proposed that cytomegalovirus takes advantage of a constitutive operating cellular pathway that underlies the phenomenon of ER-associated degradation of defective or unassembled proteins. We propose that SNP is reimported from the ER to the cytosol via the same pathway.

The negative effect of glycosylation on peptide antigenicity could have several causes. First, if the oligosaccharide is not removed, peptide antigenicity would be eradicated by interference with TAP-mediated transport, K^d binding, or TCR recognition. Second, removal of the oligosaccharide by cytosolic PNGase would create an Asp₁₄₉-substituted peptide with low affinity for K^d. In additional experiments, we found that this peptide sensitizes cells 1,000-fold less efficiently than the wild-type peptide for lysis mediated by T_CD8+ raised to NP, and that these T_CD8+ do not recognize cells expressing rVV encoded cytosolic or ER-targeted Asp₁₄₉-containing peptides (our unpublished results). Thus, removal of the glycan would destroy peptide antigenicity as surely as its nonremoval.

The inhibitory effect of glycosylation on in vitro or in vivo presentation of SNP was incomplete. Based on the argument presented in the preceding paragraph, the residual presentation of SNP^N is almost certainly due to incomplete glycosylation of Asn₁₄₉. Although we cannot eliminate the possibility that a small amount of SNP^N escapes glycosylation at Asn₁₄₉, the degree of residual presentation of SNP^N relative to SNP seems to be inconsistent with our failure to detect SNP^N lacking one or both N-linked oligosaccharides by SDS-PAGE analysis of mAb-collected material. Instead, we believe that the residual presentation of SNP^N reflects the generation of peptides from defective ribosomal products (DRiPs; 42), misfolded, possibly truncated forms of SNP^N that are not inserted into the ER and are processed by the standard cytosolic route.

In conclusion, the present findings expand the possible mechanisms underlying the presentation of secreted and membrane-bound proteins. It will be important to test in future experiments whether the results we describe are limited to cytosolic proteins artificially routed to the secretory pathway, or also apply to naturally exported proteins that have evolved to be resistant to secretory proteases. The major mode of peptide generation from any given exported protein must be determined on a case by case basis until general principles have been more firmly established.

This work was supported in part by a National Institutes of Health grant (AI-36331) to L. Eisenlohr.