Immunogenicity and Cross-Reactivity of a Representative Ancestral Sequence in Hepatitis C Virus Infection

Approximately 170 million people are infected with hepatitis C virus (HCV) worldwide, with the majority of infected individuals progressing to chronic infection (1). Within the United States, HCV infects 4 million individuals and remains the leading cause of liver transplantation and hepatocellular carcinoma (2, 3). The estimated morbidity associated with infection is anticipated to increase over the coming decade, and treatment is neither universally available nor completely effective. Thus, development of an HCV vaccine to prevent HCV infection remains a critical public health need.

Due, in part, to its highly error-prone NS5B polymerase, HCV circulates within and between individuals as a quasispecies (4). This swarm of viruses presents an immense challenge for vaccine development. Vaccine strategies meant to overcome this viral diversity must generate a broad immune response, capable of responding to a host of variations. Initial efforts at vaccine development focused on development of sterilizing immunity using an E1/E2 heterodimer (5). Although neutralizing Abs against envelope can play a role in clearance of infection and reinfection (6–8), specific envelope sequences that drive broadly neutralizing Ab production and sterilizing immunity remain unknown. Generation of robust CD4⁺ and CD8⁺ T cell responses is considered critical for long-term immunity (9–11). Development of a CTL response usually occurs early in infection and has a kinetic association with clearance of viremia (12). Failure of a CTL response to control viremia may be due to evasion of the T cell response through a number of mechanisms, including lack of CD4⁺ T cell help (13, 14), T cell exhaustion (15–18), or the emergence of viral escape mutations (19–25). Although the cellular immune response to HCV declines in chronic infection, progressive broadening of the T cell response to HCV is associated with enhanced control of HCV upon repeated reinfection (6). The generation of cellular immunity capable of controlling HCV infection analogous to that observed in people who successfully control repeated HCV infections has become a goal of HCV vaccine development (26).

The T cell responses generated by a vaccine must provide cross-reactivity against highly diverse circulating strains, making selection of HCV vaccine Ags a challenge. Sequence strain selection to induce a T cell response, either for peptide- or DNA-based vaccines, has borrowed heavily from HIV vaccine design (27, 28). Strategies that deliver only a handful of epitopes or provide cross-coverage of multiple epitopes of a single HCV protein may maximize recognition of that protein but at the expense of other regions that may be important in protective immunity. Selection of peptides with known high affinity for the MHC is another strategy that was shown to be effective for generating cellular immune responses using hepatitis B, lymphocytic choriomeningitis, and lassa virus vaccines in mice (29–31). High-affinity binders may serve as a stronger immunogenic agent, and peptides with high MHC affinity were shown to be recognized by hepatitis B virus- or HIV-infected humans and SIV-infected macaques (32–34). However, the frequency of recognition of epitopes selected on the basis of MHC affinity for HCV is unknown.

An alternative to the inclusion of specific epitopes that is commonly used in vaccine strain selection is the use of a single circulating strain. The choice of which viral strain to use is arbitrary, given that no circulating strain resembles the majority of other circulating strains of HCV. The extent to which any individual strain might have mutations that impair immune recognition and allow persistence is also unknown. Strain selection for HCV has focused on existing, well-characterized strains, such as the genotype 1a strain H77 (35). HCV strains between subtypes can differ in nucleotide composition by 20–25% (36). By comparison, a <2% amino acid difference can cause a failure in cross-reactivity of the polyclonal response to influenza vaccine (27).

An alternative to selection of specific epitopes or use of a single circulating variant is to use computational methods to minimize the degree of sequence dissimilarity between a vaccine strain and contemporary circulating viruses. One method is to create a consensus strain, whereby the most commonly predicted amino acid at each position is used at each position (27). A limitation of the consensus approach is that escape mutations can become the dominant sequence in regions where the restricting HLA allele is common, as recently demonstrated for HCV (37). This can happen because circulating viral sequences adapt to immune pressure. When an adaptive mutation at a viral residue escapes similar immune responses shared by multiple subjects (e.g., a T cell response restricted by a common allele), that mutation may arise many times in a population, such that a consensus of sequences obtained from those individuals may contain the common escape variant. Phylogenetic reconstruction places recent, host-specific changes near the tips of the tree, such that sequences found deeper in the tree may reflect an earlier, shared ancestor (38). Although not naturally occurring, ancestral strain sequences are similar to currently circulating HCV strains but may lack the divergent escape mutations carried by circulating strains.

We recently generated both a consensus sequence (cons1a) and a phylogenetic, reconstructed HCV sequence (bole1a) (39). However, the enhanced potential of any computer-generated sequence to elicit cross-reactive T cell responses has not been demonstrated for HCV. In this article, we assess the recognition frequency and cross-reactivity of HCV sequence peptides generated using the three vaccine-development methods described: peptide selection based on high MHC binding capacity, consensus, or the phylogenetically representative bole1a sequence. To our knowledge, we provide the first data in HCV to support the use of a synthetically generated sequence (bole1a) to elicit robust CD8⁺ T cell responses.

The Baltimore Before and After Acute Study of Hepatitis cohort is a prospective study of injection drug users at risk for hepatitis C infection. Eligible participants have a history of or ongoing i.v. drug use and are seronegative for anti-HCV Abs at enrollment. Written consent was obtained from each participant. Once enrolled, participants receive counseling to reduce i.v. drug use and its complications. Blood is drawn for isolation of serum, plasma, and PBMCs in a protocol designed for monthly follow-up, as previously described (40). HCV RNA measurements are used to identify the time of infection and to determine the outcome of infection. Participants with acute HCV infection were referred for evaluation of treatment. The study was approved by the Institutional Review Board at The Johns Hopkins School of Medicine.

A 5.2-kb region spanning the 5′-untranslated region to the NS3/NS4A junction was reverse transcribed, amplified by nested PCR, and cloned, as described previously (41). Briefly, total RNA was extracted from serum using the QIAamp viral RNA mini kit (QIAGEN), according to manufacturer’s instructions. The 5.2-kb PCR product was generated by a reverse-transcription nested-PCR strategy. Purified PCR products were cloned into a pCR-XL TOPO vector and transformed into One Shot TOP10 chemically competent cells. Forty clones were picked and amplified with Templiphi to screen for the presence of the insert. Positive clones were sequenced and aligned using CodonCode Aligner (Codon Code, Dedham, MA).

Quantitative assays to measure the binding of peptides to purified class I molecules were based on the inhibition of binding of a radiolabeled standard peptide (42). Briefly, 1–10 nM radiolabeled peptide was coincubated at room temperature with 1 μM to 1 nM purified MHC in the presence of 1 μM human β₂-microglobulin (Scripps Laboratories, San Diego, CA) and a mixture of protease inhibitors. After a 2-d 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, Longwood, FL), coated with the W6/32 Ab, and measuring bound cpm using the TopCount microscintillation counter (Packard Instrument). Alternatively, following the 2-d incubation, the percentage of MHC-bound radioactivity was determined by size-exclusion gel-filtration chromatography using a TSK 2000 column.

Full-length genotype 1a polypeptide sequences (n = 390) were downloaded from GenBank that were circulating in humans and were nonartificial sequences. Sequences were aligned to H77, and the consensus residues were calculated with the MargFreq program.

Construction of the bole1a sequence was described elsewhere (39). Briefly, the bole1a sequence was constructed using Bayesian phylogenetic and ancestral sequence-reconstruction methods, along with covariation analysis on the same set of 390 full-length sequences from which the consensus was derived.

PMBCs were thawed and stimulated with 10 μg/ml synthetic peptide and 0.5 μg/ml anti-CD28 and anti-CD49d Abs. Cells were maintained at a density of 2 × 10⁶ cells/ml in RPMI 1640 (Sigma-Aldrich), 20% human serum (Sigma-Aldrich), and 10 mM HEPES buffer (Sigma-Aldrich) with 2 mM glutamine and 50 U/ml penicillin-streptomycin. Cells were incubated at 37°C and 5% CO₂, with the addition of rIL-2 on days 3, 4, 7, and 10 and again following a second round of peptide stimulation on day 10.

In subjects with detectable CD8⁺ T cell ELISPOT responses against HCV, we examined the sequences of the epitope regions over time. When changes away from the consensus sequence occurred in the region of a CD8 T cell epitope, synthetic peptides corresponding to that sequence, as well as the consensus sequence, were synthesized commercially by Genemed Synthesis (San Antonio, TX).

Ex vivo HCV CD8⁺ T cell responses were quantified by Human IFN γ ELISPOT Ready-SET-Go! assays (eBiosciences), according to the manufacturer’s instructions. Briefly, PBMCs were screened for recognition of HCV-specific Ags using pools of overlapping peptides covering the entire HCV polyprotein and previously defined optimal epitopes. Once responses were determined, additional analyses were performed using decreasing concentrations of synthesized peptides to determine the effects of amino acid substitutions on recognition and the degree to which there was cross-reactivity with circulating variants of the epitope and the bole1a or consensus sequence peptides.

To compare the magnitude of T cell responses generated in response to optimal and variant peptides, the same ELISPOT assay was performed, using decreasing concentrations of synthesized peptides as Ag. Briefly, polyvinylidene difluoride plates were coated with 2.5 μg/ml recombinant human anti–IFN-γ Ab (Endogen M-700A) in 100 μl PBS/well at 4°C overnight. Plates were washed with sterile PBS eight times before blocking with RPMI 1640 + 10% FCS for 30 min. Either 20,000 or 30,000 cells in R10 media were added to the wells. Decreasing peptide concentrations (10 μl/well, 10–0.001 μg/ml) were added to the well in duplicate. PHA served as a positive control. Plates were incubated for 20 h at 37°C with 5% CO₂. Following incubation, plates were washed with 200 μl sterile PBS eight times and blotted dry. Biotin-labeled anti–IFN-γ (0.25 μg/ml, 100 μl; M-701B; Endogen) was added to each well and incubated for 90 min at room temperature. Plates were washed and incubated with streptavidin-alkaline phosphatase (100 μl; 170-3554; Bio-Rad) for 45 min at room temperature. Following additional washes, the plates were developed with BCIP/NBT Tris-buffer (pH 9.5) solution (170-6532 and 170-6539; Bio-Rad), according to instructions in the manual. Plates were dried overnight and read on a Zeiss ELISPOT plate reader.

Statistical analysis was done using SigmaPlot software version 12.0 (Systat Software). Log odds of frequency of recognition based on HLA-matched subjects was calculated. Linear regression was performed on the log odds versus log of the IC₅₀ from the in vitro binding assay for HLA-matched sequences for positive responses in all subjects and among those that cleared acute infection.

Peptides that bind strongly to MHC can induce strong pathogen-specific immune responses in other chronic viral infections (32–34). As a result, selection of pathogen peptides with high MHC-binding capacity has been used as a strategy in vaccine strain selection. To test the hypothesis that high-affinity binding to Class I MHC is associated with increased frequency of HCV peptide recognition, the likelihood of recognition of previously identified class I-restricted epitopes was correlated with IC₅₀ binding to its corresponding HLA allele and subtype. Our cohort is routinely screened during acute infection for responses against overlapping genotype 1a peptides, including both long regions of the HCV-H77 polypeptide and known class I epitopes (19). These epitopes have been confirmed as class I-restricted epitopes in the literature using cytolytic assays, IFN-γ ELISPOT, intracellular cytokine staining, and/or tetramer staining (43). Assessment during the acute phase of HCV infection is critical, because the breadth of HCV epitope recognition declines with progression to chronic infection (10, 44).

A total of 60 acutely HCV-infected subjects had HLA genotype data available and was screened for HCV-specific T cell responses to 36 epitopes with known HLA restriction and binding affinity (Table I). To avoid spuriously negative results, the calculated frequency of responses against each known HLA-restricted epitope in the ELISPOT assay includes data for only those subjects with matching HLA genotype and subtype. A positive response for a subject to any epitope was defined as recognition of that epitope in ELISPOT testing at any time during infection. The frequency of recognition varied widely between epitopes. For example, we found that nearly half of HLA-A*0201 individuals (14/33, 42%) recognized the HLA-A*02–restricted epitope NS3_1406–1415 (KLVALGINAV) (Table I). In contrast, 15/36 (42%) of the tested epitopes were never recognized. To determine whether recognition frequency was correlated with HLA-binding affinity, affinity and recognition frequency were compared.

An in vitro competitive-binding assay of peptide to the restricting HLA allele was used to assess the strength of binding to the MHC (Table I), with some of these results reported previously (45, 46). This assay measures HLA-binding affinity of test peptides by determining the concentration of test peptide required to inhibit, by 50%, binding of radiolabeled peptides to purified HLA molecules of known subtypes. Requiring high concentrations of test peptide to compete with binding of the radiolabeled peptide to the restricting MHC (high IC₅₀) suggests low affinity for that HLA molecule. We compared the frequency of epitope recognition among HLA-matched individuals with the inhibitory concentration in the in vitro HLA-binding assay and found no significant relationship (Fig. 1A). For all subjects and responses, HCV peptides with higher HLA-binding affinity were no more likely to be recognized than were those with low binding affinity (y = −0.1254x − 0.4992, r² = +0.0492, odds ratio = 0.882, 95% confidence interval: 0.69–1.12). Moreover, 12/19 (63%) high-affinity HCV peptides (IC₅₀ < 50 nM) were never recognized at any time point examined in our large cohort of acutely HCV-infected subjects. This lack of relationship between frequency of recognition and HLA-binding affinity held, even when we examined the subset of subjects who cleared acute infection and, therefore, had successful immune responses (Fig. 1B, y = −0.1464x − 0.2524, r² = −0.0925, odds ratio = 0.866, 95% confidence interval: = −13.0–12.7). Having found no association between HLA-binding affinity and T cell recognition, we evaluated immunogenicity of vaccine strains generated in alternative ways.

The use of consensus or representative sequences was proposed in vaccine design to minimize the genetic differences between vaccine strains and contemporary isolates. A consensus sequence includes the most common amino acid at each position. However, such a sequence is subject to the frequencies of common HLA alleles in the population and the forces by which these alleles shape circulating viral sequence. Evolutionary forces on the population level due to common HLA alleles play a contributory role in the frequency of circulating escape variants (37, 38). The methods used to generate bole1a minimized the genetic distance from circulating sequences while maximizing the likelihood that selected residues represented universally shared (i.e., rather than individual) evolutionary forces. For comparison, we used the same 390 full-length genotype 1a HCV polypeptide sequences that were used to construct bole1a to generate a consensus sequence (cons1a). Bole1a sequence contains a larger number of known T cell epitopes than do the H77 and HCV-1 strains (39), despite H77 and HCV-1 having been used widely to identify HCV epitopes. Bole1a is less likely to contain escape mutations that would impair T cell recognition (27). The cons1a and bole1a sequences were compared for homology across 15 epitopes located between Core and NS3. For 13 of 15 epitopes, the sequences were identical, with the two exceptions noted in Table II. In general, the consensus amino acid residue was clearly defined, with ≥90% frequency of a single residue. Where the consensus amino acid differed from bole1a, the frequency at the differing position was between 0.405 and 0.523.

T cell recognition of artificial HCV sequences has not been demonstrated. Therefore, we initially sought to determine whether the rationally generated HCV vaccine sequences cons1a and bole1a could induce robust T cell responses and whether the sequences differed in their capacity to do so. The cons1a and bole1a sequences were tested for their capacity to expand T cells already primed in vivo against circulating HCV. Lines were generated against peptides encoding the two cons1a and bole1a sequences that differed and then tested for recognition of both the cons1a and bole1a sequences by IFN-γ ELISPOT (Fig. 2A, 2B). Supplemental Table II indicates the circulating virus present at the time that the T cells were obtained from the individuals tested. In both cases, the bole1a sequence peptides were able to expand T cell responses that recognized bole1a and consensus sequences for that epitope. In contrast, incubation with the consensus sequence expanded T cells well for only one of the two epitopes. Expansion with consensus NS3_1436–1444 (ATDALMTGF) resulted in minimal recognition of either variant of that epitope (Fig. 2A). Expansion with consensus E2_610–618 (HYPYRLWHY) resulted in recognition of both variants but of much lower magnitude than that achieved with bole1a sequence expansion (Fig. 2B). Overall, T cells expanded with the bole1a sequence-encoded peptides had greater-magnitude responses against both bole1a and consensus sequences than did those sequences expanded with cons1a (Fig. 2C). Given the improved expansion using bole1a compared with cons1a, future comparisons of the immunogenicity of rationally designed strains with circulating HCV strains were completed using bole1a sequence for these epitopes.

Mutation within CD8⁺ T cell epitopes occurs frequently, with 69% of epitopes undergoing mutation in the first 6 mo of infection (19). We and other investigators previously demonstrated selection of, as well as subsequent dominance of, HCV variants that evade neutralizing Ab and CD8⁺ T cell responses (7, 19). Thus, any given circulating HCV strain from an individual is likely to contain a large number of escape mutations that allow evasion of T cell recognition (21, 47). The capacity of a vaccine strain to induce CD8⁺ T cell responses to naturally occurring sequence variants will likely be reduced if the vaccine strain used contains escape mutations.

Naturally occurring HCV variants within epitope regions were identified through longitudinal hemigenomic sequencing of HCV from subjects (Supplemental Table I) who recognized CD8⁺ T cell epitopes of known HLA restriction. Identified amino acid substitutions within these epitopes and their frequency in the larger GenBank dataset are listed in Table III. Bole1a sequences were identified in circulating sequences for all epitopes investigated, confirming that sequences algorithmically chosen are also found in natural infection. However, circulating variant sequences identified in our cohort were not always present in the larger GenBank dataset, suggesting that the relative frequency of individual variant sequences will vary.

Given that representative HCV strains should contain fewer escape mutations than do naturally circulating strains, we hypothesized that CD8⁺ T cells already primed in vivo against HCV would better recognize bole1a sequence than any individual circulating strain. Expansion capacity of the bole1a sequence relative to naturally occurring sequence variants was tested. Lines were generated against peptides encoding the bole1a sequences or naturally circulating sequence variants. These lines were tested for recognition of the bole1a sequence and the corresponding naturally occurring sequence variants using peptide dilutions in an IFN-γ ELISPOT assay (Fig. 3A–G). Lines were successfully generated against 8 of 8 (100%) peptides encoding bole1a sequences and for 12 of 18 (67%) peptides derived from naturally occurring sequence variants that did not match bole1a (Figs. 2A, 2B, 3A–3G). Incubation of T cells with every naturally occurring sequence variant of the Core_41–49 epitope (Fig. 3A), as well as three of the four NS3_1073–1081 (Fig. 3F) epitopes, failed to generate any ELISPOT responses above background to bole1a or sequence-variant Ag peptides. This suggests that the naturally occurring variants of these epitopes are less able than bole1a to expand cross-reactive T cell responses. In addition, when expansion of T cells with naturally occurring sequence variants did occur, the responses were of diminished magnitude relative to those obtained using the bole1a sequence to expand. The single exception occurred for p7_790–790, for which expansion with either the bole1a sequence or a naturally occurring sequence variant produced comparable (within 2-fold) recognition of both sequences across four of five peptide dilutions (Fig. 3E). Fig. 3H shows the summation of ELISPOT responses against 10, 1, and 0.1 μM of the bole1a and every variant peptide for each epitope when bole1a is used to expand (left column) or a naturally occurring variant is used to expand (right column). For six of seven (86%) epitopes, expansion with bole1a sequence resulted in stronger summed responses against itself and sequence variants, as measured by total spot-forming cells (SFC)/million cells, than did expansion with circulating variants (Fig. 3H). The single exception, the p7_790–790 sequence, still resulted in an overall strong magnitude of response against both the bole1a sequence and the naturally occurring sequence variant. Thus, bole1a best or comparably expands T cells specific for HCV epitopes, again supporting the superiority of the bole1a sequence over circulating variants in expanding T cell responses broadly specific for HCV.

The level of cross-reactivity for seven epitopes was compared across individuals to determine whether recognition patterns were consistent across subjects and whether bole1a expanded T cells specific for variants not seen in that subject. We tested our CD8⁺ T cell lines against the indicated bole1a sequence for recognition of all of the epitope variant peptides identified in our sequencing of circulating HCV strains, as well as bole1a and consensus sequence epitopes (Fig. 4A–G, Table III, Supplemental Table II). The bole1a sequence was recognized by all subjects at the two highest concentrations of peptide for every epitope. However, the recognition patterns for specific variants did vary between subjects and by epitope. For example, expansion of T cells with either the peptide encoding bole1a sequence for the epitope NS3_1436–1444 or its variant resulted in similar recognition patterns of bole1a and the variant sequence (Fig. 4A) between subjects. In other cases, expansion with bole1a-encoded sequence resulted in better recognition of the bole1a sequence than any variant but with differing recognition profiles of variant sequences (Fig. 4B–D). Finally, expansion with bole1a-encoded sequence resulted in equal recognition of sequence variants and itself for three epitopes (Fig. 4E–G), although the extent of variant recognition differed across subjects. T cells expanded from both subjects recognized the D610H and D610N variants well, with diminished recognition of the other three sequence variants (Fig. 4E). These results suggest that expansion of T cells with bole1a is likely to generate T cell responses that recognize many circulating HCV variants in different HCV-exposed subjects. In contrast, expansion with any given circulating variant results in widely different levels of recognition, depending on the subject and HCV exposure.

Following vaccination against any virus, individuals subsequently exposed to that virus will have to respond to a multitude of sequences similar, but not identical, to those that they have already seen. Optimal sequences, or those sequences that stand the greatest likelihood of being cross-reactive in a vaccine, were defined as sequences that induce a maximal response when used to generate a line. For a given epitope position, the optimal sequence was defined experimentally as the sequence that induced the greatest recognition on IFN-γ ELISPOT at a peptide concentration of 1 μM. Lines generated that only produced responses <100 SFC/1E6 cells were excluded from consideration. Of the 10 epitopes assessed, 6 had a single optimal sequence. For four epitopes, two sequences were equally well recognized with responses of a magnitude within 2-fold of the strongest response. When two sequences produced comparable responses, both were deemed optimal. For example, the WPAPQGARSL and WPAPQGSRSL variants of the NS3_1111–1120 epitope were recognized comparably in multiple ELISPOT assays across multiple individuals, so both were accepted as optimal sequences.

The presence of the optimal sequence or sequences for the 10 epitopes tested was assessed across the 390 full-length genotype 1a sequence culled from GenBank, as well as the bole1a and cons1a sequences. The tabulated data are presented in Fig. 5. Bole1a was found in the group of sequences containing the highest number of optimal epitopes and contains optimal sequence for every one of the 10 epitopes. Twenty-nine other viral strains also contained all of the optimal sequences. In comparison, the cons1a and the H77 sequence were found to contain only 9 of 10 optimal epitopes, despite the H77 strain being used to define most of the known HCV epitopes.

Successful vaccination strategies against extremely genetically diverse viruses like HCV will require generation of an immune response that is both broadly cross-reactive and robust against the Ags to which vaccine recipients are subsequently exposed. Significant controversy exists about the best method for selection of a vaccine strain capable of inducing strong, cross-reactive T cell responses against divergent circulating strains. Clinical trials in humans comparing the capacity of different vaccines to induce protective responses are not possible because of the expense and limited access to those at risk for infection. Thus, surrogate measures for vaccine strain efficacy are needed. We elected to focus on HCV genotype 1a infectious strains as the most common infectious subtype in the United States, as well as one of the most difficult to eradicate by therapy. This study presents a comprehensive analysis of the efficacy of a computationally designed genotype 1a HCV sequence (bole1a) to expand HCV-specific T cell responses relative to a consensus sequence or individual circulating strains, as well as an assessment of the relationship between HLA-binding affinity and recognition.

Beginning with the method of selecting peptides with high HLA-binding affinity as vaccine Ags, we assessed the frequency of peptide recognition relative to HLA binding capacity. If peptides bound more strongly to the MHC induce recognition more frequently, those peptides might be better candidates for a vaccine sequence. For hepatitis B virus, lymphocytic choriomeningitis virus, and HIV, high-affinity peptides have been proposed as vaccine Ags based on their enhanced immunogenicity (29, 31, 33). However, we found no association between HCV peptide binding to the MHC and frequency of recognition in an acutely infected HCV cohort. The lack of association also held among those who clear infection, suggesting that the strength of binding to the MHC does not predict the likelihood of developing an effective HCV response. High HLA-binding affinity, as a method for producing Ag, was proposed in previous studies (45, 46). To evaluate this strategy, a previous study assessed the capacity of peptide with high HLA-A*02–binding affinity to prime naive T cell responses in vitro using human cells, as well as in vivo in mice. Although some peptides were capable of inducing responses in both models, it remained unclear from that study whether in vivo priming of HCV-specific T cells favors recognition of high-affinity HLA binders (45). Our study compares in vivo human recognition of peptide epitopes restricted to multiple HLA alleles with widely diverse binding capacity. Another study proposed the use of high HLA-binding affinity with ability to bind multiple alleles to select HCV vaccine Ag; it found that peptides with comparable HLA-binding capacity could have markedly different immunogenicity (46). This is supportive of our study’s finding that HLA affinity beyond that needed to bind was not the major determinant of recognition. Once MHC binding beyond a relevant threshold is achieved, other factors, such as location of the epitope in the polypeptide, its abundance, thymic education, and T cell precursor frequency, appear to play a dominant role in selecting the actual epitope recognized (48, 49).

The most commonly proposed method in vaccine design for HCV is selection of an individual circulating sequence. However, any individual circulating sequence is likely to contain escape mutations that result in diminished T cell recognition relative to those generated against the initial infecting sequence (19) or a prototype sequence (50, 51). We compared two computationally derived sequences, consensus (cons1a) and a phylogenetically representative sequence (bole1a). Bole1a is designed to decrease the genetic distance between vaccine and circulating strains and to minimize the inclusion of escape mutations. Although the consensus sequence had shared identity for 13 of 15 epitopes, our head-to-head comparison of cons1a versus bole1a sequence at the two epitopes that differed demonstrated improved capacity to expand cross-reactive HCV-specific T cells responses using bole1a. One of the differences present in the consensus sequence represented a known escape mutation in an HLA-A*01–restricted epitope. The HCV NS3_1436–1444 ATDALMGY epitope was shown to predominate in HLA-A*01 populations, despite a known fitness cost (37).

Following vaccination against any virus, individuals subsequently exposed to that virus will have to respond to sequences that are similar, but not identical, to those that they have already seen. The ability to expand T cell populations that recognize common circulating sequence variants will be critical if any individual sequence is used as a vaccine strain. Bole1a reliably expanded T cell responses against a broad array of circulating variants, with greater magnitude of responses compared with naturally occurring sequence variants (Fig. 3H). The bole1a sequence can induce cross-reactive responses in different subjects against many variants, including those not circulating in that host.

Although bole1a expanded T cells capable of recognizing many circulating epitopes, some circulating variants were not recognized well by T cells expanded with any peptide sequence. For example, for the CINGVCWTV bole1a HCV NS3_1073–1081 peptide epitope, three of the four circulating variants of that sequence (CINGACWTV, CINGECWTI, and CINGECWTV) are very poorly recognized, regardless of the peptide variant used to expand T cells (Fig. 3F). Vaccination with any sequence is unlikely to induce a response to those sequences. Variant sequences were shown to escape the T cell response through a number of mechanisms, including decreased binding to the MHC (21) or exploitation of a hole in the repertoire (52). It is unknown how many T cell epitopes must be recognized to generate an effective T cell response. Thus, the impact of failure to generate a response to any one viral epitope is unclear. However, failure to generate a single T cell response is unlikely to render a vaccine ineffective.

Compared with T cells expanded with bole1a, T cell expansion with naturally occurring HCV variants generally resulted in diminished recognition of circulating sequence variants. Although some variant sequences produced cross-reactive responses, others induced no cross-recognition. In addition, some variant sequences failed to induce any T cell response, making selection of a naturally circulating sequence problematic.

Bole1a contains a larger number of epitopes in the Immune Epitopes Database than did any circulating strain of HCV assessed (39). Experimentally, bole1a and 29 other sequences contain 10 of the 10 epitope sequences tested that were deemed most optimal for expansion and recognition, outperforming both cons1a and H77. The difference might have been greater had a larger number of epitopes been assessed. The H77 sequence contained nearly as many: 9 of the 10 optimal epitopes. This is not surprising, given that H77 has been the most commonly used sequence for epitope discovery. Many of the tested epitopes were derived from this sequence in the generation of peptides to screen for T cell responses. Given the historical reliance on H77 for epitope identification, that bole1a contained even more optimal epitopes supports its use. In addition to the experimentally observed reliable expansion of cross-reactive, robust T cell responses, this theoretical assessment of immunogenicity argues for the use of a representative sequence like bole1a.

The debate in the literature over HIV strain selection has suggested the use of multiple reference sequences, consensus sequences (27), most recent common ancestor sequence, or mosaic sequences. The STEP trial used gag, pol, and nef sequences from three reference strains in an Ad5-based vaccine and failed to reduce the incidence of infection, to decrease HIV RNA set points (53), or to induce a difference in CD8⁺ T cell responses (54). Part of this failure has been attributed to the immunogenicity of the Ad5 vector in prime and boost, but selection of reference strains may also play a role. Vaccination of macaques has not found a consensus sequence to be particularly immunogenic compared with alternative methods (55). However, immune pressure on HIV is intense, and consensus sequence contains escape mutations as a result. For example, the frequency of HIV variants in HLA-B*57–, HLA-B*27–, and HLA-B*51–restricted sequences has been correlated with the prevalence of the restricting allele in populations, suggesting an evolving relationship between HIV sequence and HLA frequency (56). Finally, the use of computer-generated sequences in vaccine design has been proposed for HCV (57), with coverage compared with reference strain H77; however, direct testing of immunogenicity and cross-recognition had not been demonstrated previously. To our knowledge, our study provides the first evidence that a computer-generated HCV sequence can both expand cross-reactive T cell responses and be recognized as Ag.

There are some limitations to our study. Lines generated against a sequence offer insight into recognition of sequence variants but may artificially expand one subset of the polyclonal population in a way that would not occur in vivo. We measured only IFN-γ production as a marker for T cell reactivity. Although poly-functional T cell responses are associated with control in some vaccine studies (58), there is an excellent correlation between IFN-γ production and control of HCV (12, 59). In addition, we did not examine sequence cross-reactivity in multiple genotypes as the result of a relative infrequency of nongenotype 1 infections in our cohort. However, certain epitopes and escape variants are conserved across genotypes. For example, the escape variant Y1444F is the same variant in genotypes 1a, 1b, and 3a infection (37). Thus, our results may be relevant in nongenotype 1 infections. We are also limited to assessing responses that appear in the peripheral blood. Other studies implicated an increased breadth and frequency in the CD8⁺ T cell response in the intrahepatic compartment versus the peripheral response (50, 60). Thus, we may have underestimated the frequency of responses detected. Furthermore, we were unable to assess HLA-binding affinity for all known epitopes, including those of HLA-C, as well as the B*27 and B*57 alleles. The presence of either the B*27 or the B*57 allele in HCV-infected subjects is associated with increased frequency of HCV clearance (61). Finally, we did not assess recognition or cross-reactivity of helper responses, which are known to play a role in controlling infection (13, 62).

We also demonstrate capacity to expand from previously HCV-exposed individuals rather than immunogenicity in subjects not infected with HCV. Ideally, vaccination of naive hosts with every variant and artificial sequence would be performed. The assessment of the subsequent immune response would allow a more accurate comparison of immunogenicity. Given the lack of feasibility of such experiments, we assessed capacity to expand and cross-reactivity as surrogates.

In conclusion, generation of a successful T cell vaccine will require a strain that is immunogenic as well as cross-reactive within genotype or with multiple genotypes. To our knowledge, we have demonstrated for the first time the efficacy of a synthetic representative sequence, bole1a, in expanding a robust and cross-reactive CD8⁺ T cell response and its superiority over other strategies. We anticipate that this strategy will help to direct vaccine sequence selection for hepatitis C infection.

We thank Drew Pardoll for helpful comments on the manuscript. We also thank David Hudson for assistance with subject recruitment.

The authors have no financial conflicts of interest.