A Quantitative Analysis of the Variables Affecting the Repertoire of T Cell Specificities Recognized after Vaccinia Virus Infection¹

Vaccinia virus (VACV)³ is a member of the Poxviridae, a group of large DNA viruses that replicate in the cytoplasm of virus-infected cells (1). VACV is a close homolog of variola virus, the causative agent of smallpox, and has been commonly used as the vaccine to protect individuals against this disease. The double-stranded DNA genome is ∼195 kb long and is predicted to encode at least 218 open reading frames (ORF; GenBank accession number AY243312). The adaptive immune response to poxviruses involves B cells as well as both CD4⁺ and CD8⁺ T cells (2, 3). Adoptive transfer experiments have shown that CD8⁺ T cells can mediate control of VACV infection. Studies in mice and humans have to date identified ∼200 CD8⁺ T cell epitopes derived from 102 different proteins during VACV infection (Refs. 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17 and C. Oseroff and A. Sette, unpublished data). Of the >115,000 possible VACV-derived peptides 9 and 10 residues long, only 15 epitopes were recognized in HLA-A*0201-transgenic (Tg) mice following VACV infection (Ref. 12 and E. Assarsson and A. Sette, unpublished data). The question then is, what effectively determines the observed 8000-fold restriction of responses?

The term immunodominance is often utilized to describe the remarkable focus of the immune system on the recognition of only a relatively small number of potential determinants or epitopes (18, 19, 20, 21, 22). It is well appreciated that the ability of a given peptide to bind MHC (23, 24, 25), its capacity to be generated by Ag processing (26, 27, 28, 29), and the available TCR repertoire (30, 31, 32, 33) all contribute to immunodominance. At the quantitative level, the relative impact of each of these variables in generating the final repertoire of specificities is generally unclear and completely undetermined in the case of a complex pathogen such as VACV.

Immunoregulatory phenomena are also of importance. For instance, an ongoing immune response against a dominant epitope can diminish the response to a subdominant epitope (34). This cellular interference has been named immunodomination (35, 36) and is an additional, less well studied factor involved in the regulation of immunodominance. This observed phenomenon might be a result of competition for the APC surface, APC-derived factors, or costimulatory molecules (37, 38). Interestingly, the dominance/subdominance hierarchy can be somewhat fluid. Deletion or silencing of T cell responses against a dominant epitope can lead to the appearance of a previously undetectable response against subdominant epitopes (39, 40). Conversely, induction of T cell responses directed to a subdominant epitope through, for example, prepriming can eliminate responses directed against an otherwise dominant epitope (41). In addition, CD4⁺CD25⁺ regulatory T cells can influence immunodominance hierarchies by suppressing immunodominant T cell responses (42).

In the context of this study, we will be using the definitions established by Sercarz et al. (Ref. 20 and Table I). A dominant epitope is referred to as a peptide recognized in the context of a natural infection. A subdominant epitope is defined as an epitope not seen in the context of natural infection, but which nevertheless can induce a T cell response (usually by direct peptide immunization) that recognizes infected cells. A cryptic epitope, on the other hand, can induce a T cell response upon peptide immunization but is not generated by natural processing of the pathogen. These designations differ from the way they are utilized in other studies, where the term dominant epitope has been used to signify an epitope dominating the response in terms of magnitude or frequency of recognition (43), and a subdominant epitope represents a minor component of the response observed in natural infection. According to the definitions used in the present study, all epitopes recognized during infection would be designated as dominant, albeit some weaker and some stronger. We believe that this broader usage of dominance more accurately reflects the typically multispecific nature of an immune response. Truly subdominant epitopes that are generated by processing and recognizable by T cells but are not recognized during infection are seldom defined, and they are almost never assayed in studies analyzing mechanisms of immunodominance. Their identification is of importance, however, to enable the study of the mechanisms of immunoregulation.

The importance of immunoregulation in shaping the repertoire of T cell specificities, relative to other variables such as MHC binding, Ag processing, and TCR repertoire availability, is unclear. Knowledge about how immunodominance is regulated, and the ability to influence its effect on the immune response, is central for understanding the dynamics of pathogen escape and successful immune responses, and for the design of effective diagnostics and evaluation of new vaccine candidates.

Utilizing the A*0201 Tg mice as a model system, this study aims at quantifying the relative importance of cellular processing, MHC class I binding, and TCR repertoire in shaping T cell responses to VACV. The present data suggest that roughly 1 of 40 peptides is a high-affinity binder. Of those binders, 1 of 2 is immunogenic and can elicit a T cell response. Furthermore, 1 of 7 immunogenic peptides is generated by natural processing in the infected cell. Finally, of all peptides that bind MHC, are immunogenic, and are generated by Ag processing, only 1 of 10 is actually recognized after VACV infection, thus underlining a potential role for additional factors, including immunoregulatory mechanisms, in shaping immunodominance.

Peptides used in initial screening experiments were synthesized as crude material by Mimotopes, Pepscan Systems, or A and A Labs, as described elsewhere (44). Peptides used as radiolabeled ligands for binding assays or used in immunizations were synthesized by A and A Labs and purified to >95% homogeneity by reverse-phase HPLC. Peptides were radiolabeled with the chloramines-T method, as described elsewhere (45).

Quantitative assays to measure the binding affinity of peptides to purified HLA-A*0201 (A*0201) molecules are based on the inhibition of binding of a radiolabeled standard peptide and were performed as previously described (44, 45). Briefly, 1–10 nM radiolabeled peptide was coincubated at room temperature with 1 nM–1 μM purified MHC in the presence of 1–3 μM human β₂-microglobulin (The Scripps Laboratory) and a mixture of protease inhibitors. After a 2-day incubation, binding of the radiolabeled peptide to the corresponding MHC class I molecule was determined by capturing MHC-peptide complexes on Greiner Lumitrac 600 microplates (Greiner Bio-One) coated with the W6/32 Ab, and measuring bound cpm using the TopCount microscintillation counter (Packard Instruments). For competition assays, the concentration of peptide yielding IC₅₀ of the binding of the radiolabeled peptide was calculated. Peptides were typically tested at six different concentrations covering a 100,000-fold range, and in 3 or more independent assays. Under the conditions used, where [label] < [MHC] and IC₅₀ ≥ [MHC], the measured IC₅₀ values are reasonable approximations of the true K_d values (46, 47).

For on/off rate measurements, representative dominant (n = 4; sequences ILDDNLYKV, VLYDEFVTI, RLYDYFTRV, and SLSAYIIRV), subdominant (n = 2; sequences MLNGIMYRL, FTSDYPFYV), cryptic (n = 3; sequences NIAEYIAGL, YQSFLFWFL, GLLDRLYDL), and negative (n = 3; sequences YIDAYVSRL, VLPFDIKYI, FTSSFYNYV) epitopes, as defined in the results, that contained a naturally occurring tyrosine residue necessary for radiolabeling, were selected for study. Each peptide was labeled using the chloramines-T method. MHC-peptide complexes were formed following the standard assay protocol described above, but with an excess of labeled peptide to maximize the yield of MHC-peptide complexes. After a 2-day incubation to achieve equilibrium, further reaction was quenched with a 10,000-fold excess of unlabeled hepatitis B virus core_18–27 (IC₅₀ 1 nM). MHC bound counts were determined at the 30-, 60-, 90-, 120-, 180-, 250-, 450-, 1440-, 1860-, 2940-, and 3300-min time points after 30 min capture on W6/32 mAb-coated plates. t_1/2 was calculated using the formula t_1/2 = ln2/K_off, where K_off is the slope of the regression line of ln (% bound) vs time (seconds).

A*0201/K^b Tg mice used in this study were the F₁ generation derived from crossing homozygous Tg mice (H-2^b haplotype) expressing a chimeric gene consisting of the α1 and α2 domains of HLA and the α3 domain of H-2K^b (48) with BALB/c mice (The Jackson Laboratory). (BALB/c × C57BL/6)F₁ (CB6F1) mice were purchased from The Jackson Laboratory. All animal procedures were performed according to National Institutes of Health guidelines and Institutional Animal Care and Use Committee-approved animal protocols.

Stimulator cells used for peptide-specific IFN-γ release were: Jurkat cells transfected with the same HLA construct expressed in the Tg mice used in this study (A*0201/K^b) (49); LPS-stimulated B lymphoblasts were obtained by cultivating splenocytes in RPMI complete culture medium (RPMI 1640, supplemented with 10% FBS, 4 mM l-glutamine, 5 × 10⁻⁵ M 2-ME, 0.5 mM sodium pyruvate, 0.1 mM MEM nonessential amino acids, 100 μg/ml streptomycin, and 100 IU/ml penicillin) in the presence of LPS (8.5 μg/ml) and dextran sulfate (7 μg/ml; Sigma-Aldrich, for 3 days at 37°C; and CD11c⁺ dendritic cells (DC) were purified from mice inoculated with 7 × 10⁶ B16 cells expressing fms-like tyrosine kinase 3 ligand s.c. 12 days previously. DC purification was performed using the MACS separation system (Miltenyi Biotec) according to the manufacturer’s recommendations. All cells were grown in RPMI complete culture medium. All supplements were purchased from Sigma-Aldrich. Cells at 2 × 10⁶/ml were either pulsed with peptides or infected with VACV Western Reserve (WR) strain at multiplicity of infection of 9 for 2 h, after which the cells were washed three times in 15 ml of medium.

The WR strain of vaccinia virus was obtained from Dr. Bernard Moss (National Institute of Allergy and Infectious Diseases).

HLA-A*0201/K^b Tg mice were immunized s.c. at the base of the tail with individual peptides (10 μg/mouse) and the IA^b-restricted helper epitope hepatitis B virus_128–140 (Ref. 50 ; 140 μg/mouse) in PBS emulsified in IFA. After 11–14 days, the mice were sacrificed (three per group pooled), and CD8⁺ T cells purified using the MACS separation system (Miltenyi Biotec) were used as effectors in mouse IFN-γ ELISPOT assays. For the in vivo infections in the tolerance experiments, 2 × 10⁶ PFU VACV or PBS were given i.p. to HLA-A*0201/K^b Tg mice 1 or 2 wk before peptide immunizations. Total spot-forming cells (SFC) was calculated as SFC per 10⁶ cells × total number of CD8⁺ T cells recovered for each group.

The IFN-γ ELISPOT assays were performed as previously described (12) with the following additions: 2 × 10⁵ splenic CD8⁺ cells were cultured with 10⁵ peptide-pulsed or VACV WR-infected stimulator cells per well in flat-bottom 96-well nitrocellulose plates. Responses against an irrelevant peptide (hepatitis C virus core_132–140, DLMGYIPLV) were measured to establish background values that were subtracted from the experimental values. The experimental values were expressed as the average net spots per 10⁶ CD8⁺ cells for each peptide. To determine the level of significance, a Student t test was performed using the mean of triplicate values of the responses against relevant peptides vs the responses against irrelevant control peptide. The net number of spots per 10⁶ effector cells was calculated as [(number of spots against relevant peptide) − (number of spots against irrelevant control peptide)] × [(10⁶)/(number of effector cells/well)]. The stimulation index was calculated as follows: (number of spots against relevant peptide)/(number of spots against irrelevant control peptide). Responses were considered positive when the three following criteria were met: net SFC/10⁶ ≥ 20, stimulation index ≥2.0, and p ≤ 0.05 (Student’s t test). Target cells were washed three times after peptide pulsing to reduce the levels of excess peptide in the culture. This was done to limit presentation of the peptides by endogenous mouse MHC class I molecules expressed on the T cells themselves. In addition, A*0201-deficient cells were used in parallel to ensure that the peptides were presented on A*0201 and not on the endogenous mouse class I expressed on the mouse DC used as APCs. Peptides were considered as immunogenic in the context of A*0201 when the response to peptide-pulsed JA-A*0201/K^b was equal to, or higher than, the response to peptide-pulsed A*0201⁺ DC or LPS, and when the response to peptide-pulsed A*0201⁺ LPS blasts or DCs was significantly higher than that to peptide-pulsed A*0201⁻ (derived from CB6F1 mice) LPS blasts or DC. A positive control peptide (ILDDNLYKV) was included in all assays to ensure efficient viral infection of the target cells and epitope presentation on the cell surface.

Regulation of immunodominance involves a chain of components spanning from viral protein expression and Ag presentation to regulation at the cellular level. Expression of a given epitope sequence during infection is a prerequisite for its recognition by cellular immunity. It is currently unknown whether all VACV-derived proteins are expressed and accessible for immune recognition during infection. However, when compiling results from epitope identification studies presently available in the literature we found that at least 102 VACV-derived ORFs contain epitopes that are recognized by CD8⁺ T cells during infection (4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17) (Oseroff et al., unpublished data) and therefore by definition produced in amounts sufficient for recognition. It must be emphasized that this number is likely an underestimation, as additional proteins targeted by CD8⁺ T cell responses may yet be identified (Supplemental Table IA). ⁴ Of the >115,000 possible 9- and 10-mer peptides encoded in the entire VACV genome, these 102 antigenic proteins encode a total of 67,150 peptides. The studies described below are limited to these 102 proteins.

Adequate binding of a candidate epitope to MHC molecules is a prerequisite for T cell recognition. Thus, to estimate the universe of potential CD8⁺ epitopes in a given pathogen, it is necessary to determine the fraction of peptides that can bind MHC class I molecules. In this context, it is important to define an affinity threshold to differentiate peptides that can be recognized by T cells from those that bind with affinities too low to be biologically relevant. To derive such a binding affinity threshold, the capacity of a set of known HLA-restricted CD8⁺ T cell epitopes to bind their corresponding restricting alleles was measured. This set comprises 125 epitopes restricted by at least 1 of 22 different common HLA alleles (Supplemental Table IB). As shown in Fig. 1, and consistent with previously reported data (25), the vast majority of epitopes (85%) bound their restricting allele with an affinity of 500 nM or better, and most (75%) bound with an affinity of 100 nM or better. This lower affinity threshold also captured 100% of the known VACV-derived A*0201 epitopes (4, 5, 6, 12, 13). Thus, for the analyses described in the following sections, we utilized the 100 nM binding affinity threshold.

To quantify the role of different variables in determining which peptides are recognized during infection, the fraction of VACV-derived peptides that bind A*0201 was first determined. To do this, a combined approach relying on sequence motifs and in vitro biochemical measurements of peptide affinity was utilized. Previous work demonstrated that canonical motifs could efficiently identify peptides with the capacity to bind MHC class I with biologically relevant affinity (51). However, it was also noted that a significant proportion of binders do not strictly conform to canonical motifs, and were therefore missed. To overcome this limitation, the present analysis uses a very broad A*0201 motif previously characterized by our laboratory (44, 52), which identifies peptides with L, I, V, M, A, T, or Q in position 2 and L, I, V, M, A, and T at the C terminus. This motif is present in 21 of 22 (95%) well-characterized EBV-, HIV-, influenza-, and tumor Ag-derived A*0201-restricted epitopes independently reported in the literature (see Supplemental Table IB). Based on our database of A*0201 binding affinities, it is estimated that ∼95% of the peptides with the capacity to bind A*0201 with affinities <100 nM are motif positive. Fewer than 0.2% of peptides without this motif are A*0201 binders (J. Sidney and A. Sette, unpublished data, www.immuneepitope.org, and Ref. 53). Thus, for the binding analysis we have focused on peptides containing the extended A*0201 motif that allows us to capture the majority of potential epitopes.

To exclude protein expression as a possible factor in our analysis of epitope recognition, we have restricted our analysis to VACV proteins that are known to contain epitopes that induce CD8⁺ T cell responses after infection. By this strategy, we can be certain that the lack of recognition of a potential epitope is not due to inadequate expression of the VACV protein. Thus, we selected 18 VACV WR-derived antigenic proteins; 13 of these contain CD8⁺ T cell epitopes recognized after i.p. infection of HLA-A*0201 Tg mice (Ref. 12 and E. Assarsson and A. Sette, unpublished observations) and 5 contain epitopes restricted by other MHC class I molecules. The complete set of 1657 A*0201 motif-containing peptides (Table II) contained within these proteins were synthesized and tested for A*0201 binding. Overall, 15.9% (263 of 1657) of the motif-positive peptides tested bound A*0201 with an affinity of 100 nM, or better (Table II). These A*0201 motif-positive binders represent 2.5% of all possible 9-mer and 10-mer peptides in this set of proteins. The total number of possible peptides contained in the 102 antigenic proteins thus far identified in the VACV proteome is 67,150. Because no significant difference in the frequency of A*0201 binders was found between proteins containing epitopes restricted by A*0201 and those restricted by other class I molecules, this suggests that VACV-derived proteins targeted by CD8⁺ T cell responses contain ∼1700 high-affinity A*0201 binders (67,150 × 0.025 = 1,679).

To generalize these conclusions, we extended the study to HLA-A*1101, another common HLA class I molecule. The extended motif for A*1101 specifies the presence of A, S, T, L, I, V, M, F, or Q residues in position 2 and R, K, or Y residues at the C terminus. A set of 4 proteins known to be recognized by A*1101-restricted T cell responses (12) and a set of 18 proteins recognized in the context of other class I specificities were selected for analysis. All 1268 motif-containing 9- and 10-mer peptides in the 23 proteins studied were synthesized and tested for their capacity to bind A*1101. In total, 442 (34.9%) of the peptides bound A*1101 with an IC₅₀ of 100 nM or better. Considering all possible 9- and 10-mer peptides in these Ags, 3.0% bind A*1101 with an affinity of 100 nM or better (Table III). Thus, the detailed HLA binding data for A*0201 and A*1101 are consistent and indicate that ∼2.5 to 3.0% of all 9-mer and 10-mer peptides in a given protein can be expected to bind a certain HLA molecule with high affinity.

To analyze the role of T cell repertoire in limiting the immune response to potential epitopes, we evaluated the fraction of high-affinity binding peptides that had the capacity to induce a CD8⁺ T cell response upon peptide immunization. A set of 90 A*0201 high-affinity binding (IC₅₀ < 100 nM) peptides were randomly chosen from 35 proteins found to be antigenic in the context of HLA-A1, -A2, -A3, -B7, or -B44 in human vaccinees (13), A*0201 in VACV-infected Tg mice, or H-2D^b or H-2K^b in VACV-infected C57BL/6 mice (11). From 12 to 14 days after peptide immunization of A*0201 Tg mice, CD8⁺ splenocytes were cultured with peptide-pulsed target cells for 20 h after which IFN-γ-producing cells were enumerated in an ELISPOT assay.

First, to make certain that the immunization protocol was adequate, the 15 A*0201-restricted epitopes previously identified in these mice during VACV infection were assayed (Table IV, group 1; see 12). As expected, 15 of 15 of these known epitopes were found to be immunogenic (Table IV, Immunogenic column, and Fig. 2). Next, the set of 90 peptides sampled from A*0201 high-affinity binders not recognized in the context of a natural infection was evaluated (Table IV, Group 2, and Fig. 2). Of these A*0201 binding peptides, 50 (56%) were found to be immunogenic. Given that we estimated that there were 1679 A*0201 binding peptides in the 102 VACV proteins included in this analysis, these data led to the estimate that in addition to the 15 dominant epitopes, around 930 [0.56 × (1679 − 15)] peptides exist that are A*0201 high-affinity binders and immunogenic, but that are not normally recognized following VACV infection. We emphasize that this is a conservative estimate, as VACV-derive proteins in addition to the 102 considered in this analysis might be identified in the future as being targets for immune recognition. Nevertheless, based on the available data, the overall immunogenicity rate of A*0201 binders is estimated as 56% [(932 + 15)/1679]. These data indicate that the availability of a suitable TCR repertoire is a significant, albeit relatively minor, limiting factor in determining immunodominance.

In the same series of experiments, we evaluated whether the T cells elicited by peptide immunizations could recognize VACV-infected cells in vitro, as an indicator of the peptide being generated in the course of cellular processing of VACV-derived proteins. Preliminary flow cytometry experiments were performed utilizing various cell types and VACV-GFP or anti-VACV antisera to measure efficiency of infectivity. On the basis of these results, VACV-infected splenic CD11C⁺ DCs were selected as APCs for the IFN-γ induction assays. In the series of experiments summarized in Table IV (“Processed” column), we found that T cells elicited by 14 of 15 of the epitopes recognized in the course of natural infection recognized VACV-infected DC. These data demonstrate that the recognition of infected DC in vitro is an acceptable indicator of natural processing in vivo after VACV infection. In contrast, only 7 of the 51 (14%) other immunogenic peptides elicited T cells capable of recognizing VACV-infected cells (Table IV). Following the classic definition of dominance and subdominance (6, 23, 24), these 7 peptides represent subdominant epitopes, and the remaining 44 immunogenic peptides represent cryptic epitopes.

These results led to the estimation that an additional 130 (14% of estimated 930 immunogenic peptides) subdominant epitopes might exist in the set of 102 VACV-derived antigenic proteins. Thus, the overall frequency of dominant (n = 15) and subdominant (n = 130) peptides generated by natural processing is estimated as ∼15% (145/945).

Although the peptides above had been previously tested when measuring responses to pools of peptides after VACV infection of A*0201/K^b Tg mice (12), all seven peptides categorized as subdominant epitopes were retested as single peptides to confirm the lack of recognition after VACV infection. As expected, it was found that none of the subdominant epitopes were recognized during infection, whereas all dominant epitopes included as controls in the same experiment were recognized (data not shown).

In conclusion, by combining the estimates of motif-containing peptides with the percentage of MHC class I binding within the set of proteins known to be targets of CD8⁺ T cell responses, we have shown that ∼1/40 (2.5%) of all 9- and 10-mer peptides bind A*0201 with high affinity. Of the high-affinity binders, ∼1 of 2 (56%) elicited CD8⁺ T cell responses upon peptide immunization. Furthermore, of those immunogenic peptides, ∼1 of 7 (15%) are generated by natural processing in VACV-infected DC (summarized in Fig. 3). We estimate that out of the ∼70,000 potential 9- and 10-mers in the 102 proteins recognized by CD8⁺ T cell responses, ∼130 additional subdominant epitopes might exist along with the 15 dominant epitopes that have been identified (Fig. 3). Thus, besides processing, binding, and TCR repertoire, other mechanisms appear to limit the repertoire of epitopes eliciting responses after infection by an additional factor of 10, as only 15 of 141 possible epitopes are recognized during infection.

In an attempt to explain this phenomenon, dominant epitopes (recognized during infection), subdominant epitopes (immunogenic and processed), cryptic epitopes (immunogenic, but not processed), or nonimmunogenic peptides were compared with regard to several parameters related to MHC class I binding and T cell recognition. First the average binding affinity levels were calculated for each epitope category. As shown in Fig. 4,A, differences in MHC class I binding affinity do not seem to explain immunodominance, because epitopes from all categories bound MHC within overlapping affinity ranges, with the binding affinity being highest for the subdominant epitopes. Under the conditions utilized, the equilibrium IC₅₀ values measured are reasonable approximations of the true K_d values. Because K_d represents the ratio between off- and on-rates, two peptides can have the same equilibrium affinity but different off- and on-rates. Because it has been suggested in the literature that off-rates correlate most strongly with immunodominance (54, 55), we also wanted to assess whether off-rates of the different categories of peptides might explain their immunodominance status. When the stability of a sample of four dominant, two subdominant, three cryptic, and three nonimmunogenic peptides was tested, a suggestive, but not statistically significant, trend was noted for off-rates and dominance (Fig. 4 B).

Next, the influence of TCR avidity was studied. A*0201/K^b Tg mice were immunized as above with three dominant, four subdominant, and three cryptic epitopes. After 2 wk, the CD8⁺ splenic T cells were cultured with autologous DCs and corresponding peptides at increasing concentrations between 10 pg/ml and 10 μg/ml, and T cell responses were determined by an IFN-γ ELISPOT assay. Avidity was determined as the EC₅₀. As shown in Fig. 4 C, no consistent difference in avidity between the different categories was seen. Furthermore, there was no clear correlation (or inverse correlation) between the magnitude of particular epitope-specific CD8⁺ T cells and avidity.

In terms of magnitude of T cell responses, the average values for immunogenicity (see Table IV) were slightly lower for the subdominant epitopes than for the dominant epitopes, but ranges were largely overlapping (Fig. 4,D). Finally, it has been reported that the number of peptide-MHC class I complexes on the cell surface of infected APCs might determine whether a peptide will be recognized during infection or not (56). The magnitude of the T cell response to VACV-infected cells can be an indication of whether a peptide is efficiently generated by natural processing in the infected cell. Indeed, a somewhat higher magnitude was noted in the dominant compared with the subdominant group (from Table IV), but again ranges were largely overlapping (Fig. 4 E). Taken together, the data in this section did not identify clearly responsible factors to account for the apparent 10-fold reduction in the immunodominant epitope repertoire.

Based on the results above, it appears that immunoregulatory phenomena might play additional significant roles in limiting the repertoire of epitopes recognized in VACV infection. Indeed, it has been shown that ongoing immune responses against a given epitope can inhibit the response to a different epitope (34). Therefore, it is possible that subdominant T cells are somehow depleted, tolerized, or impaired during the antiviral immune response to dominant epitopes. This possibility was investigated by immunizing naive mice, or mice infected 7 days previously with VACV, with either dominant or subdominant epitopes and comparing the peptide-specific T cell responses in the naive and infected animals. As expected, previous VACV infection was associated with an increased T cell response (2- to 7-fold) to the dominant epitopes after peptide immunization. In contrast to this, the response to the subdominant epitopes was clearly decreased (5- to 6-fold) in mice that were infected with VACV before immunization (Fig. 5,A). Because VACV infection leads to an increase in the number of splenocytes, it was possible that an increased number of CD8⁺ T cell would explain the decreased response to subdominant peptides observed. Therefore, the same data were also plotted as total SFCs by multiplying with the total number of CD8⁺ T cells for each group (Fig. 5 B). This demonstrated that part of the observed decrease in subdominant T cell responses observed in VACV-infected mice was due to an increase in splenic CD8⁺ T cells. However, there still seems to be a decrease in the subdominant T cell responses upon previous VACV infection. These observations suggest that recognition of subdominant epitopes might be impaired during VACV infection by some yet undefined mechanisms.

Several different processes contribute to determining the final specificity of T cell responses during viral infection. These comprise expression of viral proteins, Ag processing and presentation, and TCR response, as well as regulatory mechanisms. It is not clear how much each of these steps influences the final number of recognized epitopes. Yewdell and Bennink presented a comprehensive review of the data in the late 1990s (57), and also very recently (58), and derived tentative estimates for some of the different steps that restrict immunodominance. These earlier estimates were generated by compilation of available disparate data from different antigenic systems (including work from our own laboratory). The present study, however, addresses this question in a single antigenic model. Most importantly, our work provides the first estimation of the relative importance of regulatory mechanisms in limiting the repertoire of responses to a complex virus such as VACV.

In the current study, we have focused on the A*0201 system. This system has been utilized to map responses restricted by human MHC class I molecules in various model systems (12, 59). The system has limitations, as pointed out in the literature (60), in terms of how accurately it identifies the same epitopes recognized in humans. However, we have utilized the HLA Tg model system because it allowed us to utilize accurate and reliable MHC binding motifs and assays, and to study the MHC class I molecule most commonly expressed in human populations. At the same time, the use of the Tg mouse model allowed us to draw on existing data defining the epitopes recognized in the course of natural infection (12). In addition, peptide immunizations were key to allowing quantification of the role of TCR availability and the identification of subdominant epitopes.

We would like to point out that the results obtained herein are likely to be generalized to other MHC alleles. In terms of binding analysis, similar frequencies of MHC-binding peptides were observed for another common HLA molecule (HLA A*1101). Similar ranges of binding affinities were detected for VACV epitopes restricted by other human (12, 13) and murine (7, 8, 11) class I MHC molecules. Also, a recent exhaustive study in the mouse H-2^b system revealed that in the course of VACV infection D^b and K^b molecules present 22 and 27 epitopes, respectively (11). Thus, the number of different VACV epitopes presented by a given MHC molecule seems to be in the same range in either the murine or human system.

The frequency by which peptides bind a given MHC molecule has been estimated by several studies, but accurate quantitative estimates have generally been difficult to obtain. This is because in most previous studies, narrow canonical motifs were used to preselect peptides and as a result the true frequency of MHC binding has been underestimated. In the present study, this was addressed using a comprehensive approach based on calibration of the relevant binding threshold using binding affinities of known epitopes, the use of expanded motifs, and the analysis of almost 3000 different measured binding constants. The binding data presented here focus on HLA-A*0201 and -A*1101 and suggest that ∼2.5–3% of the relevant sized peptides in a given protein can be expected to bind a certain HLA molecule. This rate is about 5 times higher than earlier estimates suggested (57), reflecting the more comprehensive motif and binding analysis of the present study.

MHC binding affinity did not strictly correlate with immunodominance. Strikingly, all seven subdominant epitopes identified were found to be very strong A*0201 binders. Each was in the top 25% rank for binding affinity among the peptides tested, with affinities <1.2 nM. It is possible to speculate that T cells specific for epitopes with very high MHC class I affinity become overstimulated and exhausted during infection and therefore fail to differentiate into effector T cells. A similar phenomenon has previously been noted in the lymphocytic choriomeningitis virus system (23). In contrast, nonimmunogenic peptides had a somewhat lower binding affinity. These observations suggest that, in general, a binding affinity that is neither too high nor too low may be optimal for epitope recognition during infection. It has also been suggested that peptide stability correlates best with immunodominance (54, 55, 61). However, in our hands, peptide stability did not correlate significantly better with immunodominance than did equilibrium binding measurements.

Testing whether a peptide could elicit T cell responses upon direct peptide immunization was used to measure the impact of TCR repertoire availability on immunodominance. In agreement with what was previously estimated (57, 58), the current study suggests that ∼50% of all high-affinity binders are immunogenic. In terms of TCR repertoire, several factors are known to influence the specificities available. First, a number of studies have highlighted the potential role of previous exposure to unrelated Ags as a key factor in shaping the T cell response. Secondly, homology of the potential epitope with self-protein sequences might narrow the repertoire based on self-tolerance mechanisms either in the thymus or in the periphery. Thymic education has in fact been shown to influence the pattern of dominant and subdominant epitopes (62). We are currently investigating whether the nonimmunogenicity of roughly 50% of the high-affinity binders is due to homology with host proteins (which might have led to negative selection) by comparing the peptide sequences against the mouse proteome. A high functional TCR avidity has also been shown to be associated with high-magnitude responses (63). However, with the data generated in the avidity experiments, there was no clear correlation (or inverse correlation) between the magnitude of particular epitope-specific CD8⁺ T cells and avidity. In addition, TCR avidity does not seem to limit immunodominance, at least not if a certain required threshold level is reached.

Processing also plays an apparent role in selecting peptides recognized in the course of VACV infection. Earlier estimates indicated that as many as one of five of all high-affinity binding peptides are generated by natural processing (64). Our data on A*0201 antigenic proteins suggest a similar ratio of 1:7. Although Ag processing might be somewhat different in the in vitro and in vivo settings, the 93% processing rate obtained for the immunodominant epitopes suggest that the model system used herein allows a fair estimate. The extent to which the actual amount of peptide-MHC class I complexes present on the surface of infected APC dictates dominance vs subdominance has been the subject of debate. Several studies showed that suboptimal production, or increased yield of certain epitopes (41, 65, 66) does not necessarily result in corresponding lower or higher immunogenicity or immunodominance. When comparing dominant and subdominant epitopes, there was no significant difference in the relative amount of epitope presented on the infected cell as judged by the relative magnitude of recognition of infected target cells.

By combining the estimates of motif-containing peptides and MHC class I binding for all proteins targeted by CD8⁺ T cell responses (Table I), it seems that 1 of 40 nonomer and decamer peptides bind A*0201 with high affinity (Fig. 3). Of the high-affinity binders, one-half could elicit CD8⁺ T cell responses upon peptide immunization, one of seven were naturally processed in VACV-infected DC. These data suggest that efficiency of Ag processing, affinity for MHC class I molecules, and the nature of the TCR repertoire all play a role in establishing the potential of peptide-MHC complexes for immunogenicity in absolute terms. However, the influence of these variables alone appears to be insufficient to fully account for the paucity of distinct epitopes recognized during infection. It rather suggests that additional factors, including immunoregulatory mechanisms, reduce the number of VACV epitopes by at least a factor of 10.

The 10-fold reduction is in fact likely to be a conservative estimate, as only 102 VACV Ags known to be expressed were considered, and for this reason the actual number of potential peptides to be considered could be up to 2-fold higher. In addition, the conservative 100 nM threshold used might also lead to an underestimation given that it is likely that several peptides binding in the 100–500 nM range have the potential of being immunogenic/naturally processed. Similarly, our focus on motif-containing peptides might have led to the exclusion of small number of potential epitopes. Additional factors limiting responses may include regulatory mechanisms, such as immunodomination, where dominant T cell responses can suppress subdominant responses, and the induction of immunoregulatory T cells that might suppress certain T cell responses. Our preliminary results suggest that T cells specific for subdominant epitopes might become impaired during VACV infection in vivo. We are currently investigating the mechanisms involved in this effect, by analyzing whether cotransfer of T cells specific for dominant epitopes can directly inhibit T cell responses to subdominant epitopes, as well as studying the specificity of this effect using irrelevant epitopes as immunogens. Furthermore, because it has been shown that both the site and route of administration as well as the coexpression of other MHC molecules within the infected host can influence the actual repertoire of epitopes recognized (4, 8, 12, 13, 58), these variables need to be fully considered. As mentioned above, it is also possible that an inordinately high binding affinity might lead to exhaustion or tolerization of the T cells recognizing subdominant epitopes.

Whatever mechanism(s) underlies this phenomenon, our results emphasize that in the case of a complex pathogen such as VACV, immunoregulatory mechanisms might be even more important than Ag processing or TCR availability, in shaping and limiting the repertoire of T cell specificities in the course of natural infection.

We thank Timothy J. Pencille and Carrie Peltz for performing the MHC binding assays, as well as David Seiber, Dr. Jeff Alexander, and Dr. Mark Newman for providing HLA-A2.1/K^b-Tg mice.

The authors have no financial conflict of interest.