Immunomic Analysis of the Repertoire of T-Cell Specificities for Influenza A Virus in Humans ^{▿ † ‡}

Influenza A virus strain selection and prediction of peptide/HLA binding.

We assembled a set of target strains representative of the H1N1 and H3N2 subtypes that were utilized for vaccine development in the periods 1968 to 1975, 1976 to 1986, 1987 to 1989, 1989 to 1998, and 1999 to 2004 (Table 1). Various potentially pandemic viruses such as the H2N2, H5N1, H7N7, H9N2, and H6N1 subtypes were also included. To obtain complete sequences from all strains, 14 (of the 23) selected virus strains with incomplete sequences were sequenced at Baylor College of Medicine. The sequencing results were submitted to GenBank. The protein sequences were scanned for the presence of main anchor motifs associated with the A1, A2, A3, A24, B7, B44 (23, 24, 26) and DR supertypes (28), and quantitative prediction matrix analyses were run as previously described (5). Using our in-house major histocompatibility complex (MHC) binding database, supertype matrices were generated based on the family of HLA alleles associated with the seven prevalent supertypes. A cutoff value of 100 nM was used for selection. Peptide conservancy was then calculated. By combining motif scanning and quantitative analyses with a preference for higher conservancy levels, a total of 4,080 peptides (8-, 9-, 10- and 11-mers for class I and 15-mers for class II) were predicted and synthesized (see Table S1 in the supplemental material).

Peptide synthesis.

Peptides utilized in the initial screening studies were purchased as crude material from Mimotopes (Minneapolis, MN). Peptides synthesized for use as radiolabeled ligands or utilized in the second round of screening were synthesized by A and A Labs (San Diego, CA) and purified to >95% purity by reverse-phase high-pressure liquid chromatography. Purity was determined using analytical reverse-phase high-pressure liquid chromatography and amino acid analysis, sequencing, and/or mass spectrometry. Peptides were radiolabeled by the chloramine T method (27).

Peptide/HLA binding assays.

Quantitative assays to measure the binding affinities of peptides to HLA molecules are based on the inhibition of binding of a radiolabeled standard peptide and were performed as described in detail previously (9, 27).

Characteristics of the study population.

Healthy males and females between 24 and 66 (average 35) years of age were used in this study. The ethnicities included were Caucasian, African-American, American Indian, Asian, and Hispanic. Exclusion criteria were a body weight of <45.4 kg and established pregnancy. Institutional review board approval and appropriate subject consent were obtained for this study.

PBMC isolation and HLA typing.

Peripheral blood mononuclear cells (PBMCs) were isolated from heparinized blood by gradient centrifugation with Histopaque-1077 (Sigma), and cells were resuspended in fetal bovine serum complemented with 10% dimethyl sulfoxide and cryopreserved in liquid nitrogen. PBMCs were typed for HLA-A, -B, and -DR by high-resolution PCR (Atria Genetics, San Francisco, CA).

CD8⁺ cell depletion.

To ensure class II restriction when measuring CD4⁺ T-cell responses, CD8⁺ cells were depleted from the PBMCs by negative selection using a Magnetic Cell Separation (MACS; Miltenyi Biotech, Bergisch Gladbach, Germany) purification system. When applicable, the CD8-depleted population was tested for the lack of recognition of a class I-restricted epitope previously identified in the same individual.

Ex vivo gamma interferon (IFN-γ) enzyme-linked immunospot (ELISPOT) assay.

A total of 4,080 peptides were synthesized and pooled (9 to 21 peptides per pool, with a majority of pools consisting of 20 peptides) by supertype, resulting in 208 pools (12 A1, 73 A2, 29 A3, 10 A24, 5 B7, 33 B44, and 46 DR supertypes). In initial experiments, PBMCs were incubated at 2 × 10⁵ cells per well in the presence of peptide pools corresponding to the donor's haplotype. Peptide pools yielding positive responses (≥20 net spot-forming cells [SFC]/10⁶, a stimulation index [SI] of ≥2, and a P value ≤0.05) were deconvoluted by subsequent testing of the PBMCs against each individual peptide at 10 μg/ml. The ELISPOT assays were performed as described previously (30). For negative control values, dimethyl sulfoxide alone was added at the same dilution as that present in the peptides/pools, and these values were subtracted from the experimental values. To assess statistical significance, a one-tailed Student t test was performed in which triplicate values of each condition were compared with those of the negative controls.

Definition of a target set of influenza virus sequences and epitope prediction.

As a first step in our study, we selected a panel of 23 representative influenza A virus strains to be used for epitope predictions. This approach avoided potential bias in the analysis, introduced by uneven representation of different subtypes in the sequence database available to us at the initiation of analysis (early 2005). To select virus strains that have been used for vaccination or are known to cause infection in the human population, representative strains for each of the H1N1 and H3N2 subtypes responsible for influenza virus infections from the periods 1968 to 1975, 1976 to 1986, 1987 to 1989, 1989 to 1998, and 1999 to 2004 were chosen (Table 1). In addition, we included representative strains from the 1918 (H1N1), 1957 (H2N2), and 1968 (H3N2) pandemics and from the avian subtypes H5N1, H7N7, H6N1, and H9N2.

To provide high coverage across different ethnicities, we considered six HLA class I supertypes (A1, A2, A3, A24, B7, and B44) (23, 24, 26) and a single HLA class II (DR) supertype encompassing many common alleles (DR1, DR2, DR4, DR7, DR8, DR9, DR11, DR13, DR51, and DR53) (10, 28, 32). Previous studies have shown that while the frequency of each allele in a supertype may vary dramatically among different populations, the frequency of each supertype is relatively constant (26). The combined coverage by the six HLA class I supertypes of the major ethnicities is >98% (24). Similarly, the DR supertype allows for 86.5 to 97.3% coverage (9, 28). For each of the HLA supertypes mentioned above, peptides were selected according to the presence of specific supertype motifs and average relative binding matrices generated by utilizing our in-house MHC binding database. A cutoff value of 100 nM for predicted affinity was used for selection. Approximately 17,000 peptides matching these characteristics were identified within the representative sequences. We next directed our selection procedure toward peptides with higher conservancy levels and chose 4,080 peptides for testing by immunogenicity assays (see Table S1 in the supplemental material). Reflecting our focus on conserved epitopes, almost all peptides with a conservancy level of ≥40% were synthesized and tested. A smaller sample of less-conserved peptides was also selected. The number of predicted peptides from the various supertypes varied according to the frequency with which the amino acids associated with each supermotif were found in proteins. For example, the A2 supertype peptides were most numerous, as the corresponding motif is composed of commonly found amino acids. Conversely, the B7 supertype peptides were the least numerous, as this supertype is characterized by a stringent requirement for the presence of a P residue in position 2 (26, 32). Likewise, the distribution of peptides predicted for each protein reflected the relative size of each protein, with the most peptides predicted for the PB1 and PB2 proteins (759 and 757 residues, respectively) and the least peptides predicted for the M2 and NS1 proteins (97 and 121 residues, respectively).

Human T-cell responses to influenza virus infection are highly diverse.

Human T-cell responses have been shown to be very diverse in the context of human immunodeficiency virus (HIV) (www.lanl.gov), cytomegalovirus (CMV) (29), and poxvirus (14, 18) infections. By contrast, based on a recent analysis of the published literature, a somewhat more limited breadth of responses has been characterized in the case of influenza virus (3). Therefore, we decided to investigate the influenza virus-induced responses in humans in more detail. Forty-four healthy volunteers between 24 and 66 years of age were recruited for leukopheresis or blood donations. HLA-typed, cryopreserved PBMCs were tested for recognition of supertype-matched pools comprising 10 to 20 peptides/pool in IFN-γ ELISPOT assays. When CD4⁺ responses were measured, PBMCs were depleted of CD8⁺ cells to ensure class II restriction. Positive pools (SFC/10⁶ ≥ 20, SI ≥ 2.0, and P ≤ 0.05) were deconvoluted to identify the individual peptides responsible for the responses. To ensure data quality and consistency, these peptides were resynthesized and purified and retested for all responding donors. To characterize these epitopes further, we measured their ability to bind to the most common HLA alleles representing the corresponding supertype (see Table S2 in the supplemental material). A peptide was considered a binder if it bound at least one allele with an affinity of ≤500 nM for class I and ≤1,000 nM for class II (25, 28). Altogether, these experiments identified 65 high-affinity binding epitopes that were found positive in at least two replicate assays using cells from the same individual. It is worth mentioning that additional epitopes were identified that generated reproducible T-cell responses but bound poorly to the HLA alleles included (data not shown). This might be explained by the fact that only the most common alleles within each supertype were tested for binding and that the binding of some peptides might have been initially mispredicted. However, as we do not know the HLA restriction of these peptides, we decided to exclude them from the remaining analyses.

To eliminate potential redundancies, we applied a cluster algorithm (http://tools.immuneepitope.org) to the set of 65 epitopes to identify nested or overlapping peptides, as well as homologous variants (≥80% identical), from different strains of influenza virus. Within a cluster, a representative optimal peptide was selected based on conservancy level, number of responding individuals, and magnitude of responses. It should be noted that if nested peptides were differentially recognized by the responding donors, each was maintained as a separate epitope. A total of 54 distinct epitopes (38 class I- and 16 class II-restricted epitopes) were identified accordingly (Table 2). Thus, this analysis reveals a heretofore unappreciated extreme breadth in the T-cell responses to influenza virus in humans. The epitopes varied in the frequency of their recognition by the panel of HLA-matched donors, from 3 to 47%, with an average of 2.5 class I- and 3.2 class II-restricted epitopes recognized by a given donor. In general, each individual donor recognized a different and unique set of epitopes. Thus, this analysis underlines the large donor-to-donor variation in the epitopes recognized in the influenza virus system.

We also wanted to compare this set of epitopes with previously published data. Of 56 unique HLA epitopes reported in the literature and restricted by the same alleles considered in our analysis, 16 were independently predicted and tested in our analysis (data not shown). The remaining 40 were mostly less conserved (29 peptides had a conservancy of <40%) and were therefore not included in our study. Of the 15 epitopes previously identified and tested in our study (6, 7, 11, 16), only 4 (VSDGGPNLY, GILGFVFTL, RMVLASTTAK, and TTYQRTRAL) were identified as epitopes in our donor population (Table 2).

PB1 and M1 are major targets of both CD4⁺ and CD8⁺ T-cell responses.

A protein might be effectively targeted only at a certain stage of infection, when its expression is at a suitable level. It is therefore likely that immune responses targeting a broad range of viral proteins would be more efficient in combating a virus than those recognizing a single protein. However, the majority of the influenza A virus epitopes described in the literature are derived from the HA and NP proteins (3). The 4,080 peptides considered in this study were derived from 10 different influenza virus proteins, allowing examination of the extent to which additional antigens might also be targets of cellular immunity. Indeed, it was found that most proteins were recognized.

A somewhat uneven representation of peptides selected for testing was observed, with the PA, PB1, and PB2 antigens contributing the most and M2, NS1 and NS2 contributing the least. This uneven representation is due to differences in protein sequence variability (HA and neuraminidase are more variable than PA, PB1, and PB2) and size (M2, NS1, and NS2 are relatively small proteins). Epitopes were derived from all antigens except M2 and NS1, with PB1 contributing the highest number of epitopes for both CD4⁺ and CD8⁺ T-cell responses (Fig. 1A; Table 3). This is in contrast to the reported literature, where only 3% of the epitopes described were derived from PB1. The highly skewed distribution of previously reported class I- and class II-restricted epitopes in favor of HA and NP (3) likely reflects the recognition of more variable epitopes, as well as a bias in the antigens investigated.

To normalize the data for the different numbers of peptides tested from the different proteins, a recognition frequency (hit rate) was determined as the number of peptides yielding positive responses divided by the number of peptides tested for each protein. The highest recognition frequency for both CD4⁺ and CD8⁺ T-cell responses was observed for M1. This also showed that even when compensating for the differences in the number of peptides tested, PB1 was second to M1 (Fig. 1B) in contributing the recognized epitopes. Interestingly, although PB2 is a subunit of the same RNA polymerase complex as PB1, the recognition rate for PB2 was eight times lower than that for PB1 (Table 3).

When comparing the antigen distribution of the epitopes identified herein with those reported in the literature, we found that while HA contributed to only a single class II-restricted epitope in our experiments (Table 3), about one-third of all published epitopes were derived from HA (Table 3; Fig. 1C). This discrepancy can be explained by the fact that HA is a protein commonly studied in influenza research, and therefore there is a higher probability that epitopes would be sought and identified from this protein. In addition, fewer HA-derived peptides were tested in this study as a consequence of our selection of peptides that were more than 30% conserved. However, despite testing 161 HA-derived predicted class I-restricted peptides, not a single epitope was identified for class I, suggesting that HA is not a common target for CD8⁺ T-cell responses (Table 3).

In conclusion, combining the current epitopes with those in the literature data set demonstrates that immunity to influenza virus infection in humans is very diverse and comprises responses to many different virus antigens.

Identified T-cell epitopes are highly conserved among various influenza virus strains and subtypes.

Vaccination strategies based on conserved T-cell epitopes would be of great value to overcome seasonal variations in influenza virus antigens. This prompted us to focus our studies on the identification of highly conserved epitopes. Accordingly, the present study identified a number of well-conserved epitopes. Specifically, 18 of the 54 (33%) identified epitopes were conserved across the entire panel of 23 influenza virus strains considered. An additional eight (15%) epitopes were conserved in 80 to 96% of all strains (Table 2). A range of 36 to 44 different epitopes were totally conserved (100% identity) in a given influenza virus strain (data not shown). A similar result was observed when we focused on the avian subtypes H5N1, H6N1, H7N7, and H9N2, in which a range of 37 to 40 epitopes were found to be conserved in all strains within at least one of the avian subtypes, and 28 epitopes were conserved across all avian strains analyzed (Fig. 2A; data not shown). This implies that T-cell memory in influenza virus-infected and/or vaccinated individuals includes some reactivity against sequences also encoded in avian viruses, which further highlights the potential advantages of utilizing T-cell epitopes for vaccine development.

Compared to the epitopes already described in the literature, the epitopes identified in this study were significantly more conserved among the 23 strains included in the present study (Fig. 2A). However, in the context of vaccination strategies, it is also important that the epitopes remain conserved in newly emerging strains. To address this point, we determined epitope conservancy levels among a total of 57 H1N1, H3N2, and H5N1 strains that emerged between 2005 and 2007 and therefore were not initially included in this study. The conservancy level of the epitopes remained relatively constant; 25 out of the 54 epitopes were conserved in 80% of the strains (Fig. 2B). The oldest strain included in this study is the A/Brevig Mission/1/18 H1N1 strain, which was the causative agent of the Spanish influenza pandemic that occurred 90 years ago. Thus, it appears that some of the epitopes are derived from genomic regions that are completely conserved over long time periods and may therefore be of interest in the development of a universal vaccine.

We have reasoned that since humans are more likely to have been exposed to highly conserved peptides than to less well conserved peptides, especially if repeated infections throughout the years are taken into account, then this would mean that the hit rate for the recognition of a particular peptide would increase with its conservancy level. Accordingly, the fraction of epitopes identified (number of epitopes recognized/number of peptides tested) within each conservancy category was analyzed. The hit rate within the set of 100% conserved epitopes was higher (2.7%) than for peptides with a lower degree of conservation (1.0 to 1.1%). Similarly, the fraction of donors responding to a specific epitope also increased with the conservancy level of the epitope, with 8% responding to peptides <40% conserved and 14% responding to 100% conserved peptides. While epitope-specific T cells must have been primed by influenza virus infection of these subjects, we have not determined whether the epitope-specific T cells can also recognize target cells infected with the virus. As each individual generally recognized multiple epitopes, it is difficult to attribute the recognition of infected cells to any given epitope specificity. This could, however, be addressed by the derivation of epitope-specific T-cell lines by repeated in vitro stimulation. Such experiments are considered for future studies.

The set of identified epitopes provides broad population coverage.

With regard to HLA supertype restriction, we have identified epitopes for all seven supertypes considered, including, to our knowledge, the first demonstration of A24-restricted influenza virus-derived epitopes. The distribution of the epitopes identified was uneven, with only one epitope restricted by B7 and 17 epitopes restricted by A2, reflecting the relatively uneven distribution of motifs. In fact, when the number of peptides yielding positive responses was divided by the number of peptides tested for each supertype, the hit rate was found to be relatively constant across the class I supertypes (0.9 to 1.6%) (Table 3).

Next, we addressed the extent to which the epitope sets identified would allow coverage of individuals in major ethnic groups. Based on the peptide binding data (see Table S2 in the supplemental material) for each of the different HLA supertype molecules and the reported frequencies of each HLA allele in different ethnic populations, we calculated theoretical population coverage for both the class I and class II epitope sets, using the population coverage calculation tool available through the Immune Epitope Database (4). For class I, the coverage was high throughout the major different populations, spanning from 90.5% of Australians to 99.9% of North Americans, with an average of 98.5% (Fig. 3A). On average, each individual was calculated to be capable of binding 6.5 epitopes. Similarly for class II, the average population coverage was 89.5%, spanning from 67.1% of South Americans to 96.0% of Europeans, and each individual had the capacity to bind 6.8 epitopes (Fig. 3B).

To further experimentally assess the issue of population coverage, we obtained PBMCs from an additional set of 20 untyped random human blood donors and tested them for their reactivity against peptide pools comprising the set of either class I or class II epitopes. A total of 15/20 (75%) individuals responded to the class I peptide pool (Fig. 4A) and 16/20 (80%) subjects responded to the class II-restricted epitope pool (Fig. 4B). Altogether, 17/20 (85%) subjects responded to either the class I and/or class II pools. The high population coverage observed is more impressive when considering the diverse infection histories of the individuals and would be expected to increase in a vaccination setting. In summary, these data suggest that the set of conserved epitopes identified herein provide a broad population coverage.