Kinetically distinct processing pathways diversify the CD8⁺ T cell response to a single viral epitope

Significance

CD8⁺ T cells eliminate infections and cancers through recognition of antigen-derived peptides displayed at the cell surface in combination with MHC class I molecules. We show that a single glycoprotein-derived epitope is generated from two sources: 1) the conventional cohort that is delivered to the endoplasmic reticulum, a fraction failing quality control and undergoing ERAD, and 2) an exceedingly minor fraction that is mislocalized to the cytosol during translation and immediately degraded. Notably, peptide derived from mislocalized antigen is delivered to the cell surface with faster kinetics and drives greater CD8⁺ T cell expansion and functionality. These findings provide key insights for development of vaccines intended to elicit CD8⁺ T cell-mediated protection.

Abstract

The source proteins from which CD8⁺ T cell-activating peptides are derived remain enigmatic. Glycoproteins are particularly challenging in this regard owing to several potential trafficking routes within the cell. By engineering a glycoprotein-derived epitope to contain an N-linked glycosylation site, we determined that optimal CD8⁺ T cell expansion and function were induced by the peptides that are rapidly produced from the exceedingly minor fraction of protein mislocalized to the cytosol. In contrast, peptides derived from the much larger fraction that undergoes translocation and quality control are produced with delayed kinetics and induce suboptimal CD8⁺ T cell responses. This dual system of peptide generation enhances CD8⁺ T cell participation in diversifying both antigenicity and the kinetics of peptide display.

CD8⁺ T lymphocytes (T_CD8+) limit viral replication by killing infected cells following activation by MHC class I (MHCI)-bound peptides generally 8 to 10 amino acids in length (1). Canonically, peptide production begins with degradation of viral proteins by the proteasome, the resultant fragments transiting to the endoplasmic reticulum (ER) via the transporter of antigenic peptides (TAP) for loading onto nascent MHCI. Nearly every nucleated cell of the body is capable of MHCI presentation processing, reflecting the critical nature of this system in host defense (2). Two aspects of the system are considered critical. First, peptide display at the cell surface must be rapid since many viruses replicate within a few hours of infection (3). Second, T_CD8+ must be highly sensitive, resulting in early detection of emerging peptide:MHCI complexes. While the mechanistic underpinnings of T_CD8+ sensitivity are understood in detail (4), the bases for rapid and efficient peptide generation remain poorly understood (5).

The prevalent paradigm intended to fill this gap is the defective ribosomal products (DRiP) hypothesis of peptide supply (6). In its original form the DRiP hypothesis proposed that a fraction of all nascent proteins are defective due to errors during production. These failures are targeted for degradation via quality-control mechanisms, one consequence being production of MHCI-binding peptides. A key refinement of the model stated that DRiPs fall into two categories (7). Rapidly degraded polypeptides (RDPs) are targeted for degradation immediately following or even during translation (half-life ≤10 min). This cohort is of particular interest as it enables rapid killing by T_CD8+ cells and is evidently the source of most MHCI-bound peptides (7). Slowly degraded polypeptides (SDPs) have a much wider half-life range (hours to days), consistent with the timing of conventional quality-control decisions (8) and the turnover of mature protein species (9), both also contributing to MHCI peptide supply (10, 11).

Much is understood about conventional quality-control decisions and the turnover of mature proteins (12, 13), but not the factors that result in RDPs and immediate peptide supply. For example, a landmark study found that redirecting the influenza hemagglutinin (HA) glycoprotein from the ER to the cytosol through signal sequence (SS) ablation, substantially increases antigen presentation (14). It was speculated that mislocalization results in misfolding and immediate targeting for degradation. However, we observed that intentional misfolding of influenza nucleoprotein did not detectably enhance antigen presentation (15). Subsequently, we reconciled these findings by demonstrating that exposure in the cytosol of hydrophobic regions that would normally be shielded (e.g., transmembrane [TM] domains), not misfolding, provides the signal for rapid degradation and enhanced peptide supply (11). This is consistent with the concept of “orphaned proteins” defined more recently by Juszkiewicz and Hegde as nascent proteins that are unsuccessful in partitioning to their intended organelle or multimeric protein complex (16). Such mislocalized proteins require swift recognition and degradation by cytosolic factors, since they can cause indiscriminate aggregation and, consequently, cellular toxicity, even at slightly elevated levels (16). Extending this model, we have proposed that cytosolic proteins will experience the same fate if not properly intercepted by the ribosome-associated chaperone system that shields hydrophobic segments emerging from the exit channel of the ribosome (8). Other mechanisms have been proposed to explain rapid protein degradation and immediate MHCI peptide supply, including protein degradation during the pioneer round of translation in conjunction with nonsense-mediated mRNA decay (NMD) (17), and a specialized ribosome that is assembled in response to infection, selectively delivering nascent polypeptides to the proteasome (7).

ER-targeted proteins, constituting ∼25 to 30% of the proteome, provide an intriguing conceptual challenge. Supply of peptide to MHCI would appear to depend upon the conventional quality-control mechanism of ERAD (ER-associated degradation). Given the kinetics of protein synthesis, translocation, quality-control decisions, and ERAD (8, 18, 19), ER-targeted processing substrates fall into the SDP category. Yet, it has been reported that epitopes can be produced from ER-targeted proteins with kinetics that imply the existence of an RDP cohort (20). Here, we illustrate the derivation of this RDP cohort from ER-targeted substrates and show its influence on the resulting T_CD8+ response_. These findings extend our understanding of rapid peptide presentation and with relevance to improved T_CD8+ vaccine design.

Results

Mislocalization of ER-Targeted Proteins to the Cytosol Dramatically Increases Epitope Presentation.

In exploring the basis for MHCI direct presentation of peptides derived from ER-targeted antigens, we considered two distinct and potentially coexistent cohorts: 1) copies that have failed quality control within the ER and are returned to the cytosol for proteasome-mediated ERAD; the relative delay in peptide production defines these copies as SDPs and 2) copies mislocalized to the cytosol through failure to translocate to the ER, leading to rapid targeting for degradation, fitting the definition of RDP. Indeed, an estimated 1 to 10% of nascent ER-targeted proteins are mislocalized, the fraction depending on natural variation in SS efficiency or metabolic conditions such as ER stress (16, 21). Mislocalization results in unshielded hydrophobicity from the retained SS, and, in many cases, an exposed TM domain.

We created a panel of constructs where the same antigen is targeted to ER with differing levels of efficiency, as conferred by divergent SS, or to the cytosol by default. As the parent antigen we used the murine guanylate cyclase-C (GCC) glycoprotein, which forms stable, homodimer complexes at the plasma membrane (22). Many transmembrane proteins are heterooligomeric and undergo rapid degradation without coexpression of their binding partners (23). GCC was modified by addition of a triple HA tag for tracking the parent protein, and the MHC H2K-b-restricted OVA_257–264 (SIINFEKL) (Fig. 1A).

Fig. 1.

Mislocalization of conventional ER-targeted proteins to the cytosol dramatically increases epitope density at the cell surface. (A) Schematic representation of GCC constructs expressed in recombinant VACV. Signal sequence (SS), triple HA tag (HA), Ova_257−264 SIINFEKL epitope (S), transmembrane domain (TM). (B) SDS/PAGE of l-Kb cells infected with equivalent doses of GCC-expressing VACV. Samples were collected 6 h postinfection. Protein synthesis inhibitors (PSI) used were cyclohexamide and ementine. PNGase, untreated samples digested with PNGase. All GCC constructs were detected by an anti-HA antibody; anti-GAPDH immunoblotting was used as a loading control. (C) Time course of SIINFEKL:Kb complex expression by WT or GCC-expressing VACV-infected l-Kb cells, measured by staining of surface complexes with 25‐D1.16 antibody as detected by flow cytometry. Statistical differences emerge for cyto-GCC at 3 h p.i. (P < 0.001) and at 5 h p.i. for ER-targeted constructs (P < 0.001 vs. cyto-GCC and between each other). Error bars represent SD of triplicate samples; data are representative of at least three independent experiments.

Hegde and colleagues have previously shown SS translocation efficiencies vary under conditions of acute ER stress (21). Thus, we replaced the native SS of GCC with that of binding Ig protein (BiP), predicted to have high translocation efficiency, and the less efficient SS of the prion protein (PRP) (21), which we therefore expected to be mislocalized to a greater extent. Rerouting to the cytosol was achieved by removal of the SS altogether (“cyto-GCC”). Since different amounts of protein will be delivered to the cytosol in each case, we anticipated that these manipulations would distinctly impact presentation of the SIINFEKL epitope.

To test this prediction, we introduced the constructs into recombinant vaccinia viruses (rVACVs) under the control of a synthetic early/late promoter (24). VACV transcription occurs in the cytosol (25), eliminating the potential complication of NMD-associated peptide supply. We infected L929 cells expressing MHCI Kb (l-Kb) with our rVACV panel and assessed GCC expression at 6 h postinfection (p.i.) (Fig. 1B). BiP- and PRP-GCC, were detected at an expected molecular weight of ∼80 kDa while cyto-GCC migrated to ∼65 kDa due to the lack of glycosylation. Treatment of BiP- and PRP-GCC lysates with peptide-N-glycosidase F (PNGase F) converted BiP- and PRP-GCC to the same molecular weight as cyto-GCC. Cyto-GCC expression was much lower than BiP- and PRP-GCC due to rapid degradation; following addition of protein synthesis inhibitors at 2.5 h postinfection, BiP- and PRP-GCC remained detectable, indicating their relative stabilities, whereas the cyto-GCC construct was untraceable. Furthermore, addition of the proteasome inhibitor epoxomicin produced no measurable changes in the levels of ER-targeted constructs but substantially increased the levels of cyto-GCC. Despite epoxomicin treatment we were unable to detect a mislocalized cohort (∼65 kDa) for either PRP- or BiP-GCC, indicating a small fraction of total protein in either case. Given the exquisite sensitivity of T cells (1) and our previous experience with biochemically undetectable antigens (26, 27), this potential cohort could nevertheless be highly immunogenic.

Next, infected cells were stained for Kb:SIINFEKL complexes on the cell surface (Fig. 1C). Both BiP-GCC and PRP-GCC supplied epitopes above background levels at 9 h p.i. However, the difference between the two SS-driven constructs was marginal and both were overshadowed by epitope generation from cyto-GCC. Furthermore, when examining the kinetics of presentation, for both ER-targeted constructs, Kb/SIINFEKL complexes were detected between 4 and 5 h p.i., lagging well behind cyto-GCC (∼3 h p.i.). These results underscore the earlier finding (14) that mislocalization of ER-targeted antigen to the cytosol profoundly enhances MHCI epitope presentation.

Profound Differences in Epitope Display between ER-Targeted and Cytosolic Parent Antigens In Vitro Translate into Equivalent T_CD8+ Responses In Vivo.

We then investigated the impact of the observed antigen presentation differences on T cell activation. The Kb:SIINFEKL-specific B3Z T cell hybridoma (28) was cultured overnight with l-Kb cells infected with increasing doses of our rVACV panel. T cell activation reached maximum levels in cyto-GCC at infectivity doses above 0.1 multiplicity of infection (MOI) (Fig. 2A). In this range, responses to BiP- and PRP-GCC were significantly lower than cyto-GCC (P < 0.001) and indistinguishable from one another. The differences, however, were not nearly as great as the cell surface Kb/SIINFEKL complexes (Fig. 1C), reflecting the extreme sensitivity of T_CD8+.

Fig. 2.

Profound differences in epitope display between ER-targeted and cytosolic parent antigens in vitro, translate into equivalent T_CD8+ responses in vivo. (A) l-Kb cells infected with equivalent doses of recombinant VACV were cocultured overnight with SIINFEKL-specific hybridomas that express β-galactosidase upon activation. Cells were lysed and conversion of the β-galactosidase substrate methyl-umbelliferyl-β-D-galactoside was measured in arbitrary “MUG” units. Flow cytometric detection of SIINFEKL-specific T_CD8+ responses to (B) direct infection by i.p. priming of B6 mice with 1 × 10⁴ PFU VACV or (C) cross-presentation in B6D2F1/J F1 mice primed intraperitoneally with 1 × 10⁶ P815 mastocytoma cells, previously infected with the indicated rVACVs. The TSYKFESV immunodominant VACV epitope was used as a control for priming. Error bars represent SD of triplicate samples; data are representative of at least three independent experiments. P values: *P < 0.05, ***P < 0.001. ns, not significant.

Extending this line of investigation, we examined effector responses in vivo. C57BL/6 mice were immunized with wild-type (WT) VACV vs. the rVACV panel, and epitope-specific splenic T_CD8+ were identified by IFNγ-based flow cytometry 7 d postimmunization, at the peak of the acute response. Under these conditions all three constructs elicited comparable responses, negating the differences in epitope expression and hybridoma recognition (Fig. 2B). Responses to the immunodominant VACV epitope TSYKFESV were uniform, confirming comparable priming.

In seeking an explanation for the equivalent responses to the three constructs in vivo, we first addressed the confounding variable of cross-presentation. Stable, mature proteins like BiP- and PRP-GCC, are cross-presented much more efficiently than unstable substrates like cyto-GCC (29). Thus, we considered the possibility that cross-presentation of ER-targeted constructs counterbalances their relatively weak direct presentation. Following a previously published approach (29), MHC mismatched P815 mastocytoma cells (H2d) were infected with our panel of rVACV for 6 h, a sufficient time for antigen expression, and then exposed to ultraviolet (UV) light in the presence of psoralen to induce DNA damage prior to injection into mice. This treatment prevented direct infection of host cells by infectious virus released by the injected cells and consequent direct presentation. Recipient mice were H2bxd B6D2F1/J, preventing an allogeneic response to the H-2Kd-expressing P815 cells and allowing for assessment of cross-presentation to H-2Kb-restricted T_CD8+. Seven days after injection, spleens were collected and T_CD8+ responses measured by flow cytometry. In agreement with previous studies (29, 30), the unstable cyto-GCC elicited nearly undetectable responses (0.05%) while the ER-targeted constructs stimulated low but detectable T_CD8+ numbers (0.2%) (Fig. 2C). Nevertheless, the scale of cross-presentation responses was ∼25-fold lower than the direct infection setting where nearly 5% of the responding CD8⁺ T cells were specific for the SIINFEKL epitope.

Collectively, these data suggest that although ER-targeted constructs generate very low levels of epitopes in comparison with the unstable cyto-GCC construct, the priming potential in vivo is equivalent. Furthermore, cross-presentation does not explain the in vivo equivalence, as responses induced by this mechanism are considerably lower than those elicited by direct presentation. Indeed, our findings agree with previous studies describing the direct presentation pathway as the major contributor to T_CD8+ priming in poxvirus infections (31). Consequently, focus moving forward was on direct presentation.

Establishing a Model System to Track Peptide Production from the ER-Targeted and Mislocalized Cohorts.

In order to obtain mechanistic insight into the causes of equivalent in vivo responses, we devised a system that allowed us to assess the relative contributions of the ER-targeted and mislocalized cohorts to epitope production. Since the BiP- and PRP-based constructs were only marginally different in in vitro experiments and not detectably different in in vivo experiments, we moved forward with only BiP-GCC in subsequent experiments.

Our system leveraged the previously described strategy of introducing an N-linked glycosylation site into a model epitope (32, 33). During ERAD, N-linked sugar groups are removed by PNGase in the cytosol prior to proteasomal degradation, converting the glycosylated asparagine (N) to aspartic acid (D), altering the primary amino acid sequence of the epitope. In contrast, substrates delivered to the cytosol retain the original N residue. This system was applied to the classical SIINFEKL epitope such that the SIINFSKL and SIIDFSKL variants serve as markers of cytosolic delivery and ERAD, respectively. Specifically, we replaced the original SIINFEKL epitope with SIINFSKL (pretranslocation) and SIIDFSKL (post-ERAD) epitope variants in both BiP- and cyto-GCC constructs, resulting in four new recombinant viruses as denoted in Fig. 3A. If the system performed as expected, and a fraction of the BiP-GCC construct is mislocalized to the cytosol, then the “BiP-GCC-N” construct should produce a mixture of both SIINFSKL and SIIDFSKL peptides. In contrast, the “cyto-GCC-N” construct should produce only the SIINFSKL epitope. As controls, the “BiP-GCC-D” and “cyto-GCC-D” constructs express only the final deamidated form of the epitope, SIIDFSKL (Fig. 3A).

Fig. 3.

N-linked glycosylation variants discern cytosolic and ER antigen processing pathways. (A) Schematic representation of the glycosylation variants used in these studies and the changes in primary amino acid sequence depending upon processing pathway. (B, Left) Peptide off-rate determined by flow cytometric analysis of MHC levels on RMAS cells. MFI, mean fluorescence intensity. (Right) Test of cross-reactivity between N- and D-specific T cell populations. B6 mice were primed intraperitoneally with 1 × 10⁸ hemagglutinating units (HAU) of recombinant influenza viruses. Splenocytes were collected at day 7 postinoculation and restimulated overnight with 100 ng/mL peptide directly ex vivo and evaluated for IFNγ production by flow cytometry. The ASNENMETM influenza epitope was used as a control for priming. (C) SDS/PAGE of l-Kb cells infected with equivalent doses of GCC-expressing VACV. Samples were collected 6 h postinfection. Protein synthesis inhibitors (PSI) used were cyclohexamide and ementine. PNGase, untreated samples digested with PNGase. All GCC constructs were detected by an anti-HA antibody; anti-GAPDH immunoblotting was used as a loading control. (D) IFNγ ELISpot of T_CD8+ cells in a dose–response curve for the ERAD inhibitor CB-5084. Target l-Kb cells were pretreated with the indicated concentration of CB-5083 and infected with VACV recombinants, then incubated overnight with N (D)- or D (E)-specific T_CD8+ cells obtained from recombinant influenza primed mice. Best fit curve statistical analysis indicates a higher probability that BiP-N and cyto-N (D) have the same ED₅₀ values while BiP-N and cyto-D (E) are more likely to have different ED₅₀ values (SI Appendix and Fig. 1C). Error bars represent SD of triplicate samples; data are representative of at least three independent experiments. P values: **P < 0.01, ***P < 0.001. ns, not significant.

The system required validation and further development at several levels. First, we determined if any of the epitope permutations impacted association with the Kb-presenting molecule. Since neither the E→S nor the N→D amino acid changes affect anchor residues (34) and both SIINFSKL and SIIDFSKL Kb-binding rates are comparable to SIINFEKL according to a peptide-binding algorithm (35), we expected minimal perturbation. To confirm this, we tested the binding kinetics of the original SIINFEKL epitope along with SIINFSKL and SIIDFSKL (Fig. 3 B, Left). RMA/S (Tap^−/−) cells loaded with saturating peptide concentrations in the presence of brefeldin A (BFA), were stained with anti-Kb MHC antibody at various time points in order to determine dissociation rates. As expected, all three epitopes demonstrated comparable peptide off-rates over a 5-h time course. Next, sodium dodecyl sulfate–polyacrylamide gel electrophoresis (SDS/PAGE) analysis of infected l-Kb cells verified that the engineered changes do not affect steady-state levels of the parent antigens (Fig. 3C). Moreover, protein synthesis inhibitors, epoxomicin, and PNGase treatments produced similar results as with our original SIINFEKL constructs. We also observed a slight increase in the mass of BiP-GCC-N due to the addition of a new glycosylation site following a prolonged gel run (SI Appendix, Fig. S1A).

We next developed tools for detection of N- and D-specific epitope presentation. Initially, we attempted to use the antibody or specific for the WT Kb/SIINFEKL complex, with the hope that this reagent would cross-react selectively with the N or D epitope but this was not the case (SI Appendix, Fig. S1B). Instead, we generated recombinant influenza viruses in which the SIINFSKL and SIIDFSKL epitopes were embedded in the nonstructural protein 1 (NS1). Since NS1 is cytosolic, the recombinant flu viruses were expected to elicit N- or D-specific populations exclusively. Indeed, priming of B6 mice with either NS1-N or NS1-D flu elicited highly specific, noncross-reactive T_CD8+ cells (Fig. 3 B, Right). The D-specific population was induced at a higher magnitude than the N-specific population (P < 0.001), a trend we observed with our VACV viruses as well (P < 0.001) (Fig. 4A). This was not a confounding factor for our experiments since the cytosolic counterparts (cyto-GCC-N/D) could be used for normalization. Thus, N- and D-specific T_CD8+ elicited by the engineered flu viruses provide highly specific readouts of SIINFSKL and SIIDFSKL presentation.

Fig. 4.

Epitopes arising via mislocalization, but not ERAD, induce optimal CD8⁺ T cell responses. (A) Flow cytometric detection of IFNγ T_CD8+ cell responses to direct infection by i.p. priming of B6 mice with 1 × 10⁴ plaque forming units (PFU) of recombinant VACV. (B) The TSYKFESV immunodominant VACV epitope was used as a control for priming. (C) IFNγ ELISpot of N- or (D) D-specific CD8⁺ T cell kinetics of epitope recognition. l-Kb cells were infected with the indicated rVACVs, fixed with paraformaldehyde (PFA) at the indicated time points postinfection and incubated overnight with T_CD8+ cells obtained from recombinant influenza-primed mice. For C, time points between 0 and 2.5 h (ns), for subsequent time points (P < 0.01 or P < 0.001). For D, Bip-N vs. BiP-D and cyto-D values were lower for all time points (P < 0.001) except 1 h (ns vs. BiP-D). BiP-D vs. cyto-D for 0 to 2.5 h (P < 0.01) were not significantly different after 3 h. Error bars represent SD of triplicate samples; data are representative of at least three independent experiments. P values: **P < 0.01, ***P < 0.001. ns, not significant.

Finally, the utility of this system for tracking translocation vs. mislocalization depends upon complete glycosylation within the SIINFSKL epitope. Otherwise, it would not be possible to distinguish between the cytosol-targeted vs. ER-targeted cohorts. To this end, we assessed the impact of the ERAD inhibitor CB-5083 on the presentation of N and D epitopes arising from different constructs (Fig. 3 D and E). CB-5083 inhibits only ATPase p97, which is responsible for removing substrates from the ER membrane prior to proteasomal degradation, while other well-known ERAD inhibitors such as eeyarestatins inhibit both ATPase p97 and the sec61 translocon, bidirectionally inhibiting flow across the ER membrane (36–38). If N epitope presentation from BiP-GCC-N is dependent upon only mislocalization, T_CD8+ cell activation should not be impacted by CB-5083. In contrast, reduction of the N fraction by CB-5083 would indicate translocation and partial glycosylation at the epitope. l-Kb cells pretreated with CB-5083 were infected with indicated viruses, fixed in order to “freeze” the peptide display at the cell surface also preventing assembly and release of live virus, then incubated with N- or D-specific T_CD8+ cells in an IFNγ ELISpot assay. At higher concentrations, CB-5083 causes cytotoxicity due to the induction of the unfolded protein response (39). Thus, target cell recognition declined at higher CB-5083 concentrations for all constructs; however, BiP-GCC-N and cyto-GCC-N slopes remained the same. In contrast, only the BiP-GCC-N construct in the context of the D epitope recognition demonstrated a steep decline starting at the 0.1 µM CB-5083 concentration while BiP-GCC-D and cyto-GCC-D displayed similar curves (SI Appendix, Fig. S1C). Thus, the N peptide from BiP-GCC-N tracked with the cyto-GCC-N construct, leading to the conclusion that all immunologically detectable N peptide from BiP-N is derived from parent protein that is mislocalized to the cytosol.

In sum, through use of noncross-reactive N- and D-specific CD8⁺ T cells, we ascertained that the engineered SIINFSKL and SIIDFSKL epitopes are effectively presented by the Kb class I molecule and report out targeting of parent protein to the cytosol or ER. This positioned us to assess the relative contributions of ER-targeted vs. mislocalized glycoprotein in the T_CD8+ response.

Epitopes Arising via Mislocalization, but Not ERAD, Induce Optimal CD8⁺ T Cell Responses.

As before, we primed B6 mice i.p. with our rVACV panel, and IFNγ⁺ T_CD8+ cells were quantified 7 d p.i. Both BiP- and cyto-GCC-N and, BiP- and cyto-GCC-D construct pairs generated similar levels of N- or D-specific responses within each set (Fig. 4A). Thus, the biochemically undetectable mislocalized fraction arising from the BiP-GCC-N construct could stimulate a T cell response on par with the cyto-GCC-N control. Strikingly, BiP-GCC-N induced a much smaller number of D-specific T_CD8+ cells compared to the BiP-GCC-D and cyto-GCC-D controls (Fig. 4 A, Right). Taken together, these results indicate that: 1) the fraction of epitopes arising from mislocalization of ER targets is necessary and sufficient for inducing an optimal T cell response; and 2) under the described conditions, accounting for the inherently higher response by D-specific T_CD8+ (Figs. 3B and 4A), epitopes generated via ERAD induce a relatively minor fraction of the total potential T_CD8+ response.

Finally, employing the same N/D system, we examined the kinetics of epitope expression as an orthogonal approach to assess the relative contributions of mislocalized vs. translocated parent protein. Accordingly, l-Kb cells fixed at various times p.i. were coincubated with N- or D-specific T_CD8+ cells in an ELISpot assay. As expected, the N epitope was produced from BiP-GCC-N and cyto-GCC-N with similar kinetics; T cell activating levels were reached between 1 and 2 h p.i. in both cases with a steady rise until the final 4-h mark (Fig. 4C). Responses to the D epitope from Bip-GCC-N were substantially delayed compared to those elicited by cyto-GCC-D (Fig. 4D) and the maximal stimulation capacity was considerably lower, also consistent with results in Fig. 4A. In further agreement with our findings so far, BiP-GCC-D displayed an intermediate phenotype, starting out at the same level as BiP-N at 1 h (not significant [ns]), but reaching the same response levels as cyto-GCC-N (ns) by 3 h postinfection (Fig. 4D). In sum, these results underscore the importance of the cytosolic antigen processing pathway, accessed either by default or by mislocalization, in generating epitopes that are rapidly recognized by T_CD8+. At the same time, the cytosolic pathway appears to be necessary and sufficient to drive optimal T_CD8+ effector numbers.

Mislocalized and ERAD Processed Epitopes Induce T_CD8+ Cell Populations with Differing Functionalities.

Lastly, in light of the submaximal frequency of IFNγ⁺ D-specific T_CD8+ in response to the BiP-GCC-N construct (Fig. 4A), we set out to assess functionality more broadly. First, we evaluated the sensitivity of the responding T_CD8+ which can be impacted by the surface density of epitope:MHCI complexes (40). We performed a T cell functional avidity assay using IFNγ ELISpot as a readout (Fig. 5 A and B). Briefly, splenocytes from mice immunized with the indicated viruses were collected at day 7 and restimulated directly ex vivo with a range of peptide concentrations. The overlapping response curves indicate that D-specific T_CD8+ cells produced via priming with BiP-GCC-N, although differing in magnitude, are just as sensitive to peptide concentration as those induced by BiP- or cyto-GCC-D.

Fig. 5.

Mislocalized and ERAD-processed epitopes induce T_CD8+ cell populations with differing functionalities. (A and B) Day 7 splenocytes from B6 mice primed with the indicated rVACVs were cocultured with l-Kb target cells pulsed with titrated amounts of SIINFSKL or SIIDFSKL peptides and analyzed by IFNγ ELISpot. Spot numbers were normalized to the averaged maximum response recorded. (C and D) Total cytokine-producing T_CD8+ at 7 d postpriming with rVACVs in response to synthetic SIINFSKL (C) or SIIDFSKL (D) peptide. (E and F) Polyfunctionality analysis of the T_CD8+ population specific for the SIINFSKL (E) or SIIDFSKL (F) variants. Error bars represent SD of triplicate samples; data are representative of at least three independent experiments. P values: *P < 0.05, ***P < 0.001.

Next, we performed a polyfunctionality analysis of N- and D-specific T cell populations. B6 mice were immunized as described above and T_CD8+ responses were interrogated at day 7 p.i., this time assessing the expression of IFNγ, TNFα, and the degranulation marker CD107. N-specific responses to BiP- and cyto-GCC-N populations were similar (Fig. 5 C and E), as were D-specific responses to BiP- and cyto-GCC-D (Fig. 5 D and F). In particular, both pairs showed comparable total IFNγ expression (Fig. 5 C and D), as expected from our previous experiments. However, the BiP-GCC-N and -D constructs induced significantly higher numbers of total CD107⁺ expressing T cells compared to the cyto-GCC-N/D counterparts. There were several minor differences between the BiP- and cyto-GCC-N pair, including total TNFα expression (Fig. 5C) and the numbers of single-function IFNγ producers (Fig. 5E). Also, BiP-GCC-D displayed a higher number of single-function CD107⁺ T_CD8+ responses compared to cyto-GCC-D (Fig. 5F).

Finally, comparison of D-specific responses from BiP-GCC-N (ERAD fraction) with cyto-GCC-D, and by default BiP-GCC-D, revealed a significant deficit in the magnitude of total numbers of IFNγ⁺ and CD107⁺ responding cells (Fig. 5D). This difference is driven by a striking absence of CD107⁺IFNγ⁺ double and CD107⁺ single producers (Fig. 5F). Taken together, these results indicate that T cells elicited by both cytosolic- and ERAD-processed peptides develop similar sensitivities to antigen concentration; however, their functionalities are markedly different.

Discussion

Here we identify two distinct sources of the same epitope arising from a single ER-targeted protein (Fig. 6). The first is derived from the fraction of nascent protein translocated into the ER and targeted for ERAD. This cohort, therefore, fulfills the original definition of a DRiP (6). The clear delay in presentation places it into the SDP category (7). The second is the exceedingly minor fraction of protein that is not translocated into the ER but delivered by default to the cytosol, resulting in immediate degradation and presentation, unambiguously an RDP. Despite being such a minor component, this cohort contributes substantially to the epitope pool.

Fig. 6.

Modeling the relative contributions of each processing pathway to MHCI peptide supply. The immediately degraded (RDP) cohort mislocalized to the cytosol and the more slowly degraded (SDP) cohort targeted to the ER, failing quality control and undergoing ERAD constitute the DRiPs fraction, contributing to differing degrees to direct presentation as described above. The mature protein fraction that successfully exits the ER contributes minimally to direct presentation but is the primary contributor to cross-presentation, which is not a major driver of T_CD8+ activation in the VACV-based system used here (Fig. 2).

ER-targeted proteins have been pivotal in the evolving picture of T_CD8+ activation. Initially, T_CD8+ cells were presumed to recognize fully folded viral proteins associated at the cell surface with MHCI (41). Accordingly, target antigens were assumed to be membrane glycoproteins. The findings that internal proteins can be recognized by T_CD8+ were reconciled with the original model by findings that they can be detected in low numbers at the plasma membrane (42–44). These native antigen-based models were dramatically upended by the findings that T_CD8+, like T_CD4+, recognize fragments of antigen (45), that the pathway toward epitope generation originates in the cytosol (46), and that proteasome-generated fragments are transported to the ER by TAP for loading onto nascent MHCI (47, 48). Ironically, processing of cytosolic proteins fit easily into this new paradigm, glycoproteins less so. In light of the finding that retargeting of HA to the cytosol substantially enhances T_CD8+ recognition, it was proposed that presentation of wild-type glycoproteins is contingent upon a fraction of the nascent pool being mislocalized to the cytosol (49, 50). The “N-to-D” system was used to test this notion for an HIV gp120-based epitope (51). However, specificity of the T_CD8+ clone that was used limited focus on only the N form, and the possibility of incomplete glycosylation was not addressed. In subsequent studies of other epitopes available specificities did not allow the authors to determine the relative use of each pathway and whether a single epitope could be produced through both pathways (32, 52). At the same time, increased understanding of quality-control systems for glycoproteins shifted focus to ERAD-based mechanisms. However, the kinetics of ERAD were difficult to reconcile with the rapid kinetics with which glycoproteins can be presented (8, 20). For this reason, and informed by building biochemical evidence on the imperfection of the translocation machinery (53), we revisited this question.

At the outset of this investigation we anticipated the key tools to be SS with different translocation efficiencies and variants of the well-studied SIINFEKL (“SL8”) epitope (54). For the first tool, contrary to expectation, we found comparable protein expression and only slightly enhanced epitope liberation from GCC targeted to the ER with an inefficient SS, even with the considerable ER stress of a VACV infection (55, 56). These observations were at odds with the previously reported large divergence between our chosen SS (21). Kang et al. induced ER stress chemically (1,4-Dithiothreitol and thapsigargin) under cell-free conditions. Possibly, translocation is regulated differently under our experimental conditions and VACV-associated ER stress is handled by other mechanisms such as reduced speed of translocation and increased release of unfolded proteins to the post-ER secretory pathway (57). The actual protein systems used could also be a factor. Kang et al. (21) used PRP as the parent antigen. PRP, mislocalized to the cytosol, forms proteasome-resistant aggregates (58) that would accentuate this cohort relative to mislocalized GCC.

For our second tool, we introduced the E → S mutation at the P4 position creating an internal glycosylation site according to the N-X-S/T motif, allowing us to distinguish between untranslocated (SIINFSKL) and translocated (SIIDFSKL) processing substrate. Three attributes of this system were critical for its application: 1) Binding to the Kb restriction element was not impacted by the E → S and N → D changes. This is consistent with the crystal structure of the SIINFEKL:Kb complex, which revealed the Asn (P4) and Glu (P6) residues to be the most solvent, accessible within the Kb-associated peptide (34). Such positioning implies a minimal role in binding to Kb and a major role in T cell receptor contact. 2) Noncross-reactivity between the T_CD8+ responses to the N and D forms, also likely attributable to the relative positioning of the P4 side chain in the Kb-binding groove. As mentioned, in other uses of the N-to-D system, scope of analysis was narrowed by limitations in or lack of MHC binding by one of the two species (32, 33). 3) Finally, complete glycosylation at the Asn within the epitope following translocation was confirmed through use of a retrotranslocation inhibitor assessed by T_CD8+ activation. This readout was particularly important considering the high sensitivity of T_CD8+, which allows detection of species that are biochemically undetectable (59–61), as was the case here.

Results shown here align with our previous finding that unresolvable hydrophobicity in the cytosol leads to rapid degradation and increased presentation (11) and with more recent studies that provide a mechanistic basis. Specifically, a battery of cytosolic proteins sense exposed hydrophobicity, targeting the parent proteins for degradation (53). In the case of our GCC-based constructs, two different systems may be activated: ubiquilin family members which may favor the generally shorter (7 to 12 aa) SS and the BAG6 complex, which appears to favor the longer (15 to 25 aa) TM domains. Importantly, these systems can target proteins during translation (62, 63), explaining the rapid presentation of cytosolic and mislocalized proteins, as we have proposed (8). Of note, BAG6 knockdown has been demonstrated to reduce expression of surface MHCI (64), which is controlled by peptide supply (65, 66). Delayed presentation of GCC translocated into the ER may in part be explained by the lack of a SS (and BAG6 engagement), removed by signal peptidase during initial translocation. In contrast, mislocalized GCC contains both a SS and TM domain. In contrast, results do not align with our previous report in which an epitope derived from influenza nucleoprotein (NP) was presented with similar efficiencies when NP was delivered to the cytosol or the ER (67). However, that system was fundamentally different in that nascent NP targeted to either location was long lived.

Both the ER-targeted and mislocalized GCC fit the definition of a DRiP but, as we have argued previously (8), there are key distinctions. DRiP is the acronym for defective ribosomal product, implying that the error is intrinsic to the antigen itself. This is true for the bona fide translocated version, which can fail quality control for various reasons. However, the error associated with the mislocalized species is extrinsic to the antigen and may lie with the chaperoning or translocation machinery. In this light, referring to these cohorts as SDPs and RDPs, according to the refined DRiP model (7) seems more appropriate.

Our studies support an important observation made by Yewdell and Nicchitta, that RDPs are a considerably more efficient source of MHCI-binding peptides than SDPs (7), as schematized in Fig. 6. As mentioned, we have proposed that this is attributable to the robust degradation signal provided by unshielded hydrophobicity. Yewdell et al. (5) and Yewdell and Nicchitta (7) have proposed the possibility of immunoribosomes, a specialized subset of translating ribosomes which, through recruitment of specialized factors, are able to produce DRiPs with high efficiency and deliver them to the proteasome. Perhaps unshielded hydrophobicity triggers immunoribosome participation. Rock and colleagues have demonstrated that the turnover of mature, fully folded proteins is yet another source of MHCI-binding peptides and that they can be the major source of epitope from a cytosolic protein expressed via transduction (68). We did not examine epitope production from the mature protein cohort in our studies. However, considering the relative inefficiency of epitope production from the ERAD fraction compared to the mislocalized fraction as measured by T cell recognition, the transiting of mature GCC to the cell surface, and our use of a viral infection system that induces considerable stress and cell death within 24 h, we expect contribution of epitope from mature protein in our experimental system to be minimal, as depicted in Fig. 6. Relative contributions from the three protein pools are likely to be very different under conditions of steady-state expression in a cell not undergoing acute stress, conditions more relevant to the presentation of autoantigens and tumor antigens.

Despite large differences in SIINFEKL production from the cyto- and ER-targeted constructs, T_CD8+ activation in vitro and in vivo were virtually comparable, reflecting the high efficiency with which cytosolic antigen is processed and the exquisite sensitivity of T cells to very low amounts of antigen. The latter is attributable to mechanisms such as serial engagement of a single peptide:MHC complex by multiple TCRs, TCR rebinding, immunological synapse formation and a highly coordinated signal transduction cascade (1). Responses are not binary; however, when we previously titrated translation product through use of thermostable duplexes (“hairpins”) of different length in a VACV expression system, a graded impact on T_CD8+ responses was observed both in vitro and in vivo (59). We conclude that even low levels of presentation generated by naturally orphaned ER-targeted proteins, are sufficient to prime an optimal T_CD8+ effector response in vivo under conditions of conventional transcription and translation. Thus, proteins targeted to the ER and other organelles, unless expressed at very low levels, should not be substantially disadvantaged in eliciting T_CD8+ responses compared to cytosolic proteins.

An important remaining question is whether one processing pathway is more important from the standpoint of protection. The functionality profile of the T_CD8+ population stimulated by mislocalized GCC (BiP-GCC-N responses to the N epitope) seems to be geared toward cytotoxicity with a higher number of T cells expressing IFNγ and the degranulation marker CD107. In contrast, the more slowly presented ERAD processed fraction (BiP-GCC-N responses to the D epitope) induced fewer IFNγ⁺ producers. On closer inspection, this discrepancy is driven by the absence of CD107⁺IFNγ⁺ double positive cells, and we observed that the CD107⁺ single function population is also markedly reduced. While cytotoxic potential remains the hallmark of T_CD8+ cells, unconventional subsets have been reported to carry out noncytotoxic antiviral and immunoregulatory functions (69–71). Should tetramer tools become available, it would be interesting to determine whether T cells responding to ERAD-processed epitopes might have a different cytokine profile or whether simply they do not expand to the same extent. From the standpoint of rational T_CD8+ vaccine design it may be beneficial to preserve both processing pathways in order to diversify the kinetics of epitope display and to broaden T_CD8+ repertoire and functional diversity.

Materials and Methods

Antigen Presentation.

Cells were infected with rVACV and stained for OVA257-264 (SIINFEKL)-Kb surface complexes. Samples were fixed in 2% paraformaldehyde in phosphate-buffered saline and collected on an LSR II cytometer.

Flow Cytometry.

Spleens were harvested at day 7 postinfection and processed into red blood cell-free single-cell suspension. Splenocytes were stimulated overnight (18 h) with the indicated peptides in the presence of protein transport inhibitors. The samples were then stained with antibodies for viability, CD4, CD8, CD3, IFNγ, TNFα, CD107, and analyzed on an LSR II flow cytometer..

Mouse Inoculations.

C57BL/6 mice were inoculated with recombinant VACV or influenza viruses intraperitoneally for direct presentation experiments. For cross-presentation experiments, B6D2F1/J mice were immunized intraperitoneally with P815 cells previously infected with rVACV then inactivated. All mice experiments were performed in accordance with the Thomas Jefferson University and Children's Hospital of Philadelphia institutional animal care and use committees and under the NIH guidelines for proper animal welfare.

ELISpot Assays.

Effector T_CD8+ cells were isolated by negative selection from the spleens from mice inoculated as described above. Target cells consisted of peptide-pulsed or virus-infected l-Kb cells. Effector and targets were cocultured overnight in triplicates in ELISpot plates coated with an anti-IFNγ capture antibody at an effector:target ratio of 2:1.

Statistical Analysis.

Statistical analysis was performed using GraphPad Prism software. Unless indicated, graphs depict a representative individual experiment of at least three independent experiments with three to five mice per group. Data are represented as mean ± SD, and an unpaired two-tailed Student’s t test with 95% confidence interval was used to determine P values and significant differences between datasets.

Study Approval.

Animal procedures were reviewed and approved by the institutional animal care and use committees of Thomas Jefferson University and the Children’s Hospital of Philadelphia.

A full overview of the methods can be found in SI Appendix.

Data Availability.

All data are available within this manuscript and SI Appendix.