Lack of Heterologous Cross-reactivity toward HLA-A*02:01 Restricted Viral Epitopes Is Underpinned by Distinct αβT Cell Receptor Signatures*

αβT cell receptor (TCR) genetic diversity is outnumbered by the quantity of pathogenic epitopes to be recognized. To provide efficient protective anti-viral immunity, a single TCR ideally needs to cross-react with a multitude of pathogenic epitopes. However, the frequency, extent, and mechanisms of TCR cross-reactivity remain unclear, with conflicting results on anti-viral T cell cross-reactivity observed in humans. Namely, both the presence and lack of T cell cross-reactivity have been reported with HLA-A*02:01-restricted epitopes from the Epstein-Barr and influenza viruses (BMLF-1 and M158, respectively) or with the hepatitis C and influenza viruses (NS31073 and NA231, respectively). Given the high sequence similarity of these paired viral epitopes (56 and 88%, respectively), the ubiquitous nature of the three viruses, and the high frequency of the HLA-A*02:01 allele, we selected these epitopes to establish the extent of T cell cross-reactivity. We combined ex vivo and in vitro functional assays, single-cell αβTCR repertoire sequencing, and structural analysis of these four epitopes in complex with HLA-A*02:01 to determine whether they could lead to heterologous T cell cross-reactivity. Our data show that sequence similarity does not translate to structural mimicry of the paired epitopes in complexes with HLA-A*02:01, resulting in induction of distinct αβTCR repertoires. The differences in epitope architecture might be an obstacle for TCR recognition, explaining the lack of T cell cross-reactivity observed. In conclusion, sequence similarity does not necessarily result in structural mimicry, and despite the need for cross-reactivity, antigen-specific TCR repertoires can remain highly specific.

␣␤T cell receptor (TCR) genetic diversity is outnumbered by the quantity of pathogenic epitopes to be recognized. To provide efficient protective anti-viral immunity, a single TCR ideally needs to cross-react with a multitude of pathogenic epitopes. However, the frequency, extent, and mechanisms of TCR crossreactivity remain unclear, with conflicting results on anti-viral T cell cross-reactivity observed in humans. Namely, both the presence and lack of T cell cross-reactivity have been reported with HLA-A*02:01-restricted epitopes from the Epstein-Barr and influenza viruses (BMLF-1 and M1 58 , respectively) or with the hepatitis C and influenza viruses (NS3 1073 and NA 231 , respectively). Given the high sequence similarity of these paired viral epitopes (56 and 88%, respectively), the ubiquitous nature of the three viruses, and the high frequency of the HLA-A*02:01 allele, we selected these epitopes to establish the extent of T cell crossreactivity. We combined ex vivo and in vitro functional assays, single-cell ␣␤TCR repertoire sequencing, and structural analysis of these four epitopes in complex with HLA-A*02:01 to determine whether they could lead to heterologous T cell cross-reactivity. Our data show that sequence similarity does not translate to structural mimicry of the paired epitopes in complexes with HLA-A*02:01, resulting in induction of distinct ␣␤TCR repertoires. The differences in epitope architecture might be an obstacle for TCR recognition, explaining the lack of T cell crossreactivity observed. In conclusion, sequence similarity does not necessarily result in structural mimicry, and despite the need for cross-reactivity, antigen-specific TCR repertoires can remain highly specific.
␣␤T cells recognize peptides (p) bound to the MHC molecule, or human leukocyte antigen (HLA) 4 in humans, via their ␣␤T cell receptor (TCR). Each TCR comprises two chains, ␣ and ␤, that are composed of variable (V), joining (J), and constant (C) genes for the ␣-chain and an additional diversity (D) segment for the ␤-chain. The TCR has three complementarity determining regions (CDRs), which determine pHLA specificity (1). Although CDR1 and CDR2 are germ line encoded and found within the V gene segment, the CDR3 spans the junction between the V(D)J gene segments. During somatic recombination, TCR gene segments are rearranged, and the additional insertion/deletion of nucleotides at each gene junction creates a high level of diversity within the CDR3, termed a hypervariable loop (2). Thus, the CDR3 loops are often used to define epitopespecific ␣␤TCR clonotypes.
The clonal selection theory (3) suggests that protective immunity is a "one clonotype, one specificity" system, in which a single TCR recognizes a single pHLA complex. However, this theory was questioned by Mason (4), who hypothesized that 10 15 T cells would be required to recognize 10 15 non-self peptides. Because the weight of 10 15 T cells would be more than 500 kg in one individual, such single specificity of T cells is mathematically impossible. This suggests that each T cell must be able to recognize a vast number of peptides to ensure protective immunity in a process known as cross-reactivity. Thus, T cell cross-reactivity is critical for the efficacy of the immune system, whereby T cells exposed to one epitope can recognize viral variants (5)(6)(7) to provide immune protection upon subsequent challenge (5).
T cell cross-reactivity can also occur during viral infections, in which effector or memory CD8 ϩ T cells can recognize epitopes from an unrelated virus, a process known as heterologous T cell cross-reactivity. Heterologous T cell cross-reactivity has been extensively characterized in murine models, in which mice primed with one virus can be subsequently challenged with an unrelated virus during the effector phase (8 -12). Studies using similar viruses, such as lymphocytic choriomeningitis virus (LCMV) and Pichinde virus (9), or distinct viruses, such as LCMV and vaccinia virus (VV) (10), have shed some light onto the mechanisms of heterologous cross-reactivity. For example, the H-2K b -restricted NP 205 epitope from both LCMV (YTVKYPNL) and Pichinde virus (YTVKFPNM) are 75% identical, with differences only at position (P) 5 and P8 (9). Structures of the two H-2K b -NP 205 peptides revealed that P5 and P8, which share similar residues, were both buried inside the antigen-binding cleft, thereby enabling a high level of cross-reactivity toward these distinct viral peptides. Conversely, others have shown that heterologous cross-reactivity can occur toward epitopes with lower sequence identity. For example, the LCMV gp 34 epitope (AVYNFATM) and the VV A11R 198 epitope (AIVNYANL) share three identical residues (37% identity). However, previous exposure to LCMV provided a level of protection against VV (13). Despite sharing only three identical residues, the two pMHC complexes adopted similar structures, with the variable amino acids buried within the antigen-binding cleft, thus providing a molecular basis for T cell cross-reactivity (14).
A recent study utilizing a library of peptides containing systematic substitutions showed that TCRs are predominantly cross-reactive because they are tolerant of peptide residue substitutions rather than recognizing multiple distinct peptides (14). Moreover, the "tolerated" substitutions were either not in direct contact with the TCR or were conservative and thus permitted binding by the TCR, in line with the observations from Shen et al. (13). This suggests that T cells can cross-recognize distinct pMHC complexes because they are permissive of substitutions and recognize specific pMHC architectures rather than degeneracy in TCR binding. Accordingly, only a handful of studies have shown how a single ␣␤TCR could engage highly divergent pMHC complexes (15)(16)(17)(18).
Based on the previous reports of human heterologous crossreactivity (20,23,24,28), we focused our current study on the well described and highly prevalent HLA-A*02:01-restricted epitopes, namely M1 58 /BMLF-1 and NS3 1073 /NA 231 . These paired peptides share identical (3 and 6, respectively) as well as chemically conserved (2 each) residues, with a sequence homology of 56 and 88%, respectively. Therefore, they provide a good model to determine the molecular basis underlying heterologous T cell cross-reactivity in humans.
In this study, we combined single-cell ␣␤TCR repertoire sequencing with biophysical and structural analysis of the four epitopes in complex with the HLA-A*02:01 molecule. We also undertook functional studies (including ex vivo and in vivo T cell expansion in healthy individuals for both peptide pairs and in HCV-infected individuals for the NS3 1073 /NA 231 peptide pair) to determine the frequency and biological relevance of heterologous T cell cross-reactivity toward these HLA-A*02: 01-restricted epitopes. Our data show that the sequence similarity between the paired epitopes did not translate to structural mimicry. Namely, the paired epitopes exhibited distinct architectures and mobility within the HLA binding cleft and selected distinct ␣␤TCR repertoires. Together, these findings underlie a lack of heterologous cross-reactivity detected directly ex vivo by tetramer enrichment and in vitro via IFN-␥ and tetramer assays. Whereas T cell cross-reactivity is an intrinsic requirement for protective immunity, our data indicate that the sequence similarity of peptides alone is not a reliable indication of CD8 ϩ T cell cross-reactivity. In line with previous studies (13,14), our results highlight that pHLA architecture impacts CD8 ϩ T cell cross-reactivity.
The M1 58 and BMLF-1 epitopes share 56% sequence homology, with three identical residues at P1-Gly, P6-Val, and P9-Leu (similarly to the LCMV gp 34 and VV A11R 198 epitopes), as well as chemically conserved P2-I/L and P5-F/L residues. The NA 231 and NS3 1073 peptides share higher sequence homology (88%). To determine whether the sequence variation could impact the stability of each peptide within the HLA-A*02:01 antigen binding cleft, we performed a thermal stability assay. The thermal melting point (T m ) values were very similar for the paired pHLA complexes, with a T m of ϳ59°C for M1 58 and BMLF-1 and of ϳ56°C for NA 231 and NS3 1073 , suggesting that each pHLA complex displays similar stability.
We next compared the four peptide-HLA-A*02:01 structures (Fig. 1). The structures of HLA-A*02:01 in complex with NS3 1073 (PDB code 3MRG (27)), BMLF-1 (PDB code 3MRE (27)), and M1 58 (PDB code 2VLL (29)) were determined previously, and here we report the HLA-A*02:01-NA 231 structure to a resolution of 2.0 Å ( Table 1). The four pHLA complexes adopted similar overall structures, with an average root mean standard deviation of 0.40 Å on the C␣ atoms of the ␣1␣2 domains. All four peptides adopted an extended conformation within the HLA-A*02:01 cleft, whereby the P2 and P9 residues were used as anchors and were buried inside the cleft, whereas P3 acted as a partial anchor residue.
Structure of M1 58 versus BMLF-1 in Complex with HLA-A*02:01-The M1 58 peptide has been referred to as a "plain vanilla" peptide, because its structure is featureless in the cleft of HLA-A*02:01; the two large Phe residues at P5 and P7 are buried inside the cleft, and only the small P6-Val side chain is exposed for TCR contact (Fig. 1a, panel i) (29 -31). The P5-Leu and P7-Ala of the BMLF-1 peptide (27) adopted a buried conformation similar to the P5 and P7 of the M1 58 epitope (Fig. 1a, panel ii); however, the smaller P5 and P7 side chains of the BMLF-1 peptide were shifted toward the HLA ␣2-helix. As a result, the conserved P6-Val adopted a more central and fully solvent-exposed conformation in the BMLF-1 peptide than in the M1 58 peptide (C␣ displacement of 1.6 Å; Fig. 1a, panel iii). Moreover, as judged by temperature factor analyses (relative B factor of each peptide residue compared with the HLA antigen binding cleft B factor), the central part (P4 -P7) of the BMLF-1 peptide was more flexible than the M1 58 peptide, with a maximum mobility at the conserved P6-Val (Fig. 1b). Peptide mobility can influence peptide conformation, as well as CD8 ϩ T cell recognition (32,33). Therefore, despite a conserved P6-Val and 56% sequence homology, the M1 58 and BMLF-1 peptides were presented differently by the HLA-A*02:01 molecule.
Structure of NA 231 versus NS3 1073 in Complex with HLA-A*02:01-Despite the sequence similarity of the NA 231 and NS3 1073 epitopes (27), the structures of both peptides differed (Fig. 1c). For example, the conserved aromatic residue at P7 (NS3 1073 P7-Trp; NA 231 P7-Phe) adopted contrasting conformations. The NA 231 P7-Phe was buried inside the cleft and interacted with Trp 147 and His 114 of HLA-A*02:01 ( Fig. 1c, panel i) and was not available for TCR recognition. Conversely, the NS3 1073 P7-Trp was solvent-exposed, with its aromatic side chain potentially available for TCR interaction (Fig. 1c, panel ii). The different P7 conformations changed the central part of the peptide (Fig. 1c, panel iii), resulting in the conserved P6-Cys sitting higher in the HLA-A*02:01 cleft for the NA 231 peptide compared with NS3 1073 (C␣ deviation of 2.4 Å). This resulted in a 20% decrease in TCR accessible surface area for the NA 231 peptide (270 Å 2 ) compared with NS3 1073 (335 Å 2 ) (Fig. 1c, panel iii). Temperature factor analysis revealed that the NS3 1073 peptide was rigid, whereas the NA 231 peptide showed higher mobility at P4 to P6 (Fig. 1d). Taken together, the structural analysis revealed that HLA-A*02:01 presents the two peptide pairs with distinct epitope conformations and differing flexibility.
There were no evident TRAV or TRBV biases in the NA 231specific CD8 ϩ T cells (Table 3 and Fig. 2b, panel i). The most common CDR3␣ lengths for the distinct clonotypes were 8 or 10 aa (both 28.3 Ϯ 10.4%) (Fig. 3a, panel i), whereas CDR3␤ was most commonly 10 aa long (26.7 Ϯ 30.6%) (Fig. 3a, panel ii). No shared motifs were observed within either of the CDR3 loops ( Fig. 3b, panel iii). The number of NA 231 -specific TCRs available was low because of their low pHLA affinity and the predominant naïve phenotype of the NA 231 -specific CD8 ϩ T cells.
Overall, no common TCR␣␤ clonotypes were utilized for the recognition of M1 58 and BMLF-1 or the NA 231 and NS3 1073 peptides. On a broader level, only a single common TRAV8/ TRBV28 pair was identified in the NA 231 and NS3 1073 TCR␣␤ repertoires; however, they were completely distinct clonotypes. Furthermore, no common TRAV/TRBV pairs were detected in  (19 -21 days). CD8 ϩ T cell lines were stained with the cognate peptide, whereas PBMCs (from healthy donors) were stained with either NA 231 QE or NS3 1073 tetramers and were magnetically enriched. The cells were stained as lymphocytes, singlets, CD3 mid-high , dump Ϫ (PBMCs only), CD8 ϩ , tetramer ϩ cells, and single-cell sorted. Then the ␣␤TCR repertories were determined using a multiplex RT-PCR. the M1 58 and BMLF-1 ␣␤TCR repertoires. These findings further highlight the high level specificity of memory CD8 ϩ T cells for their pHLA. Thus, despite high sequence homology, the paired peptides were structurally different and mobilized distinct ␣␤TCR repertoires with no overlap.

Heterologous CD8 ؉ T Cell Cross-reactivity in Humans
Lack of Heterologous Cross-reactivity after in Vitro Amplification-To determine whether the reported heterologous cross-reactivity could be a consequence of an in vitro amplification, epitope-specific CD8 ϩ T cell lines from healthy (Fig. 6) or HCV-infected donors ( Fig. 7 and Table 4) were assessed for cross-reactivity by tetramer co-staining and an IFN␥ ϩ intracellular cytokine staining (ICS) assay. M1 58 ϩ and BMLF-1 ϩ CD8 ϩ T cell lines specifically responded to their cognate peptide, with 22.4 Ϯ 16.8 and 17.9 Ϯ 7.1% of the tetramer ϩ CD8 ϩ T cells (Fig.  6, a and b) and 25.0 Ϯ 12.3 and 9.7 Ϯ 7.5% being IFN␥ ϩ TNF␣ ϩ , respectively (Fig. 6, c and d). The frequency of CD8 ϩ T cells able to bind both tetramers and produce IFN␥ ϩ TNF␣ ϩ was just above background, showing the absence of CD8 ϩ T cells able to cross-react on both the M1 58 and BMLF-1 epitopes (Fig. 6).
No Cross-reactivity between Different NA 231 Variant Peptides-Based on the low number of NA 231 tetramer ϩ CD8 ϩ T cells detected in our study, and because the NA protein is highly variable, we investigated whether the donors may have been exposed to other NA variants. We first examined the conservation of the NA 231 peptide in the pH1N1, recent trivalent seasonal influenza vaccine strains, and Oceania-originated isolates ( Table 5). A total of 28 variant peptides were identified. Mutations were common in the anchor P2 and P9 residues (buried residues; Fig. 1c), which are unlikely to affect TCR binding ( Table 5). The P2 mutation, V2I, was common to the vaccine (71%) and Oceania (ϳ67%) isolates and thus further increased the sequence homology to the NS3 1073 peptide. Three major variants from the NA 231 WT CVNGSCFTV were identified and chosen for further analysis, namely CINGTCTVV (P2-Ile-P5-Thro-P7-Thr-P8-Val), CVNGSCFTI (P9-Ile), and CI NGSCFTI (P2-Ile-P9-Ile) ( Table 5).
To assess whether our HCV-infected donors had been exposed to a different NA 231 variant (Table 5), CD8 ϩ T cell lines were established against each or a pool of the peptides containing WT NA 231 and three NA 231 variants, as well as WT NS3 1073 and a single NS3 1073 variant. No NA 231 -specific CD8 ϩ T cells were expanded, despite extended in vitro amplification of up to 18 days (Fig. 8a). CD8 ϩ T cells were successfully expanded against the NS3 1073 peptides in both donors (Fig. 8b),

Heterologous CD8 ؉ T Cell Cross-reactivity in Humans
and CD8 ϩ T cells stimulated with the pool of peptides were able to respond only to both NS3 1073 peptides (Fig. 8c). Together, these data show that there is no cross-reactivity between WT and variants NA 231 with NS3 1073 in donors with chronic HCV infection, in line with previously published data (26).

Discussion
Heterologous T cell cross-reactivity is a phenomenon whereby an individual CD8 ϩ T cell can recognize and respond to more than one viral antigen derived from unrelated viruses (41). It has the potential to be either beneficial, by increasing the chances that any given viral peptide is recognized by a CD8 ϩ T cell (28), or detrimental, by contributing to the CD8 ϩ T cell-mediated immunopathology associated with diseases such as EBV and HCV infection (23,24). Because the few reports on T cell cross-reactivity in humans have shown conflicting results (20 -25, 41), we investigated four epitopes from ubiquitous viruses (influenza, EBV, and HCV) restricted by the highly prevalent HLA-A*02:01 molecule. We aimed to elucidate the extent (if any) of the proposed heterologous cross-reactivity toward two distinct sets of HLA-A*02:01 restricted peptides: M1 58 (29) with BMLF-1 (27) and NA 231 with NS3 1073 (27).
The most likely mechanism for heterologous T cell crossreactivity would be that an individual CD8 ϩ T cell with a single TCR can recognize unrelated but similar pHLA complexes. This suggests that molecular mimicry would underpin heterologous T cell cross-reactivity, in a similar fashion to the interepitope cross-reactivity (5). Structures of the HLA-A*02:01 in complex with each paired peptide showed, however, that despite the sequence similarity of the paired peptides, the pHLA complexes display distinct architectures. The M1 58 and BMLF-1 peptides share 56% sequence homology, and the conserved P6-Val adopts a different conformation between the two peptides. In BMLF-1, the P6-Val is solventexposed and available for TCR contact, whereas it is buried within the HLA-binding cleft in the M1 58 structure (29). This solvent-exposed P6-Val is important in the recognition of the BMLF-1 pHLA complex, as demonstrated by the structure of a prototypical public TCR (AS01) in complex with HLA-A*02: 01-BMLF-1 (37). The highly conserved CDR3␤ loop positions itself atop the P6-Val, and the RXXXGN motif directly interacts with the P6-Val (36). The BMLF-1 peptide does not undergo structural change upon AS01 TCR binding, and moreover, alanine substitution of the P6-Val dramatically decreased the TCR affinity by 13-fold (37). Therefore, the buried conformation of P6-Val in the M1 58 complex would prevent binding by BMLF-1-specific CD8 ϩ T cells. From the structure of a conserved prototypical TRBV19 ϩ TCR (JM22) in complex with HLA-A*02: 01-M1 58 , it was observed that the M1 58 also does not undergo structural change upon TCR binding. Therefore, the solventexposed P6-Val of the BMLF-1 pHLA complex would cause steric clashes with the JM22 CDR3␤ loop, thus preventing binding (29,30). PBMCs from healthy HLA-A*02:01 ϩ donors were tetramerstained or co-stained with either M1 58 and BMLF-1 or NA 231 and NS3 1073 conjugated to PE or APC. The samples were magnetically enriched and antibodystained (as outlined under "Experimental Procedures") and gated on lymphocytes, singlets, CD3 ϩ , dump Ϫ CD8 ϩ , and tetramer ϩ . a, representative dot plots describing the gating strategy used for data analysis of tetramer magnetic enrichment experiments. b, representative dot plots of naïve and memory tetramer positive populations and their respective phenotypes using ␣CD27 and ␣CD45 RA phenotypic markers. c, representative dot plots of PBMCs from two donors stained with either WT or HLA-A*02:01-QE mutant NA 231-specific tetramers and their phenotype using ␣CCR7 and ␣CD45 RA phenotypic markers. d, frequency of naïve, low affinity, and memory epitope-specific CD8 ϩ T cell populations in healthy HLA-A*0201 ϩ donors after magnetic enrichment.
With regards to the NA 231 and NS3 1073 peptide pair, although they have higher sequence homology (88%), they show opposing conformations for their conserved P6-Cys and P7-Phe/Trp residues. The NA 231 P6-Cys is accessible for TCR binding, whereas the P7-Phe is buried. Conversely, the NS3 1073 P6-Cys is buried, and the P7-Trp is solvent-exposed. These changes completely alter the peptide conformation presented by HLA-A*02:01 to CD8 ϩ T cells for recognition.
Given the structural differences in the pHLA landscapes, the ␣␤TCR repertoire being utilized for the recognition of these peptides was also distinct. Paired analysis from three donors showed that no common TRAV/TRBV clonotype pairs were utilized to recognize either M1 58 or BMLF-1. The M1 58 ␣␤TCR repertoire has a distinct TRBV19 bias, as previously reported (20,30,31), whereas the BMLF-1 ␣␤TCR repertoire has a TRAV5 bias, which tends to pair with TRBV20 -1 (37). Furthermore, the ␣␤TCR repertoire for recognition of NA 231 and NS3 1073 was completely unrelated. Together, these data   NOVEMBER 18, 2016 • VOLUME 291 • NUMBER 47 highlight that the paired peptides are presented differently by HLA-A*02:01 and consequently elicit very different ␣␤TCR repertoires. Thus, it is unlikely that heterologous T cell crossreactivity could be occurring at a high frequency in HLA-A*02: 01-positive donors between those two peptide pairs.

Heterologous CD8 ؉ T Cell Cross-reactivity in Humans
Using a combination of ex vivo and in vitro techniques, including high sensitivity tetramer magnetic enrichment, we were unable to detect any heterologous memory CD8 ϩ T cell cross-reactivity in healthy (n ϭ 12) or HCV-infected (n ϭ 5) donors. The prominent M1 58 and BMLF-1 CD8 ϩ T cell subsets were identified in all donors directly ex vivo and expanded to high levels following in vitro amplification. Despite this, no heterologous T cell cross-reactivity was detected even following extensive in vitro amplification. In addition, because the NA protein is variable between influenza strains, we tested a panel of common NA 231 variants in HCV-infected donors for heterologous T cell cross-reactivity with the NS3 1073 peptide. However, no responses were detected against the WT and variant NA 231 peptides. Together, these data imply that heterologous T cell cross-reactivity is not occurring frequently toward these epitopes in both healthy and HCV-infected donors.
The precise reasons for the conflicting reports on human heterologous T cell cross-reactivity are unknown. However, differences in methodology of T cell culture, donor cohorts, directionality, and strain specificity of the viral infection might influence the results. In our study, we have chosen to avoid prolonged stimulation of CD8 ϩ T cells or exposure to high concentration of numerous cytokines, to reproduce as closely FIGURE 6. Lack of heterologous T cell cross-reactivity was observed following in vitro culture of PBMCs with either the M1 58 or BMLF-1 peptides. M1 58 and BMLF-1-specific CD8 ϩ T cell lines (stimulated for 9 -19 days) from healthy donors were assessed for heterologous cross-reactivity by tetramer co-staining (with PE and APC) and an IFN␥ ϩ TNF␣ ϩ ICS assay. All CD8 ϩ T cell lines were antibody-stained (as described under "Experimental Procedures"), and the samples were gated on lymphocytes, singlets, CD3 mid-high , and CD8 ϩ T cells. a, representative dot plot (donor H-13) (panel i) and summary (panel ii) of epitope-specific tetramer ϩ CD8 ϩ T cells in M1 58 -specific CD8 ϩ T cell lines (n ϭ 3). b, representative dot plot (donor H-13) (panel i) and summary (panel ii) of epitope-specific tetramer ϩ CD8 ϩ T cells found in BMLF-1 280 -specific CD8 ϩ T cell lines (n ϭ 3). c, representative dot plots (donor H-13) (panel i) and summary (panel ii) of IFN␥ ϩ TNF␣ ϩ production (no-peptide control subtracted) derived from M1 58 -specific CD8 ϩ T cell lines (APCs; n ϭ 6) in response to each peptide and controls. d, representative dot plots (donor H-13) (panel i) and summary (panel ii) of IFN␥ ϩ TNF␣ ϩ production (no-peptide control subtracted) made by BMLF-1-specific CD8 ϩ T cell lines (APCs; n ϭ 6) in response to each peptide and controls. The bar charts represent the means with the error bars representing standard deviation.
as possible a physiological CD8 ϩ T cell response. If heterologous T cell cross-reactivity occurs only in extreme circumstances of prolonged and potent T cell stimulation in vitro, it is therefore unlikely to happen frequently in vivo in humans. As shown in the present study, NS3 1073 -specific CD8 ϩ T cells were frequent in HCV ϩ donors but rare in healthy donors and with a naïve phenotype, and no cross-reactivity with NA 231 was observed. Conversely, Urbani et al. (22) were able to detect T cell cross-reactivity for the NA 231 /NS3 1073 pair, but only in a few HCV ϩ donors with severe liver pathology. The directionality of viral infection, viral strains, and acute infection could thus impact on the presence or absence of heterologous CD8 ϩ T cell cross-reactivity.
Our data fit within the current published literature (22,25,26) and the lack of heterologous CD8 ϩ T cell cross-reactivity observed between the HLA-A*02:01-restricted epitopes (M1 58 with BMLF-1 and NA 231 with NS3 1073 ) highlights that sequence similarity does not translate to structural mimicry and therefore does not lead to T cell cross-reactivity. In addition, limited sequence homology can lead to molecular mimicry, whereby different pMHC complexes can share structural features that will in turn be recognized by cross-reactive T cells (17,42). Our data also fit with a recent study showing that TCRs are cross-reactive because of their "tolerance" to peptide changes (substitution with a similar residue or at a position not involved with TCR contact), leading to an overall molecular conservation (14). In conclusion, CD8 ϩ T cells can display a high level of specificity for their pHLA complex because of their ␣␤TCR, ensuring that autoimmunity is still a rare event within the population. PBMCs from HCV-infected individuals were stimulated (for 11-27 days) with NA 231 or NS3 1073 peptides, and heterologous cross-reactivity was evaluated by tetramer co-staining and an IFN␥ ϩ TNF␣ ϩ ICS assay. The samples were stained (as described under "Experimental Procedures") and gated on lymphocytes, singlets, CD3 mid-high , and CD8 ϩ T cells. a, representative dot plot (donor HCV-2) (panel i) and summary (panel ii) of epitope-specific tetramer ϩ CD8 ϩ T cells found in NA 231 -stimulated CD8 ϩ T cell lines (n ϭ 3). b, representative dot plot (donor HCV-2) (panel i) and summary (panel ii) of epitope-specific tetramer ϩ CD8 ϩ T cells found in NS3 1073 -specific CD8 ϩ T cell lines (n ϭ 3). c, representative dot plots (donor HCV-2) (panel i) and summary (panel ii) of IFN␥ ϩ TNF␣ ϩ production (no-peptide control subtracted) made by NA 231 -stimulated CD8 ϩ T cell lines (APCs; n ϭ 3) in response to each peptide and controls. d, representative dot plots (donor HCV-2) (panel i) and summary (panel ii) of IFN␥ ϩ TNF␣ ϩ production (no-peptide control subtracted) derived from NS3 1073 -specific CD8 ϩ T cell lines (APCs; n ϭ 3) in response to each peptide and controls. The bar charts represent the means with the error bars representing standard deviation.

Experimental Procedures
Protein Purification, Crystallization, and Structure Determination-Soluble HLA-A*02:01 containing the M1 58 , BMLF-1, NA 231 , or NS3 1073 peptides were prepared as described previously (43,44). Crystals of the HLA-A*02:01 in complex with the NA 231 peptide were grown by the hanging drop, vapor diffusion method at 20°C with a protein/reservoir drop ratio of 1:1, at a concentration of 10 mg/ml in 10 mM Tris-HCl, pH 8, 150 mM NaCl using 20% PEG 6K, 0.2 M NaCl, 0.1 M sodium citrate, pH 6.5. Prior to being flash frozen in liquid nitrogen, the crystals were soaked in a cryoprotectant solution containing mother liquor solution with 30% (w/v) PEG. The data were collected at the Australian Synchrotron (Clayton, Australia) on the 3BM1 Beamline (ADSC-Quantum 210 CCD detector) (45), processed with XDS software (46), and scaled using XSCALE software (46). The molecular replacement was performed with the PHASER program (47) using a previously solved HLA-A*02:01 peptide structure as the search model (PDB code 3GSO (43)). Manual model building was conducted using the Coot software (48) followed by maximum-likelihood refinement with the PHENIX program (49). The final HLA-A*02:01-NA 231 model has been validated using the Protein Data Base validation web site (Table 1), and the coordinates have been submitted under the PDB code 5SWQ. All molecular graphics representations were created using PyMOL (50). Peptide sequence alignment was performed using the Clustalw server, and the sequence homology encompasses identical and chemically similar residues.
Thermal Stability Assay-To assess the stability of each peptide within the HLA-A*02:01 molecule, a thermal shift assay was performed as previously described (30).
Human Samples-Buffy packs from healthy blood donors (seronegative for HIV, HBV, and HCV) were obtained from the Australian Red Cross Blood Service (West Melbourne, Australia). Blood from HCV-infected subjects was collected as a part of social network study of intravenous drug users (Burnet Institute, Melbourne, Australia (51). HCV status was assessed via serology and the presence of viral RNA at the Victorian Infectious Diseases Reference Laboratory (Parkville, Australia). Samples were HLA-typed by the Victorian Transplant and Immunogenetics Service (West Melbourne, Australia) at the Australian Red Cross Blood Service tissue-typing laboratory. HLA-A*02:01-positive, HBV-negative, HIV-negative injecteddrug users were selected. Donors used in this study are listed in Table 4. The experiments conformed to the NHMRC Code of Practice and were approved by the University of Melbourne Research Human Ethics Committee.
Separation of PBMCs from Whole Blood-PBMCs were separated from peripheral blood by density gradient centrifugation over Ficoll-Paque Plus (Sigma-Aldrich). Diluted blood was layered onto 15 ml of Ficoll-Paque Plus and centrifuged. PBMCs were collected from the interface of the Ficoll, washed three times, resuspended in freezing medium (10% dimethyl sulfoxide; Sigma-Aldrich) in heat-inactivated FCS (Bovogen Biological, Keilor East, Australia), and stored in liquid nitrogen.
Peptides and Tetramers-The following HLA-A*02:01-restricted peptides were purchased from Genscript: HCV- CD8 ϩ T Cell Lines-PBMCs were thawed, washed, and rested for 1 h in RF10 (RPMI (Gibco) supplemented with 10 -20% FCS, 5 mM HEPES (MP Biomedicals), 100 g/ml streptomycin (Gibco), 100 units/ml benzylpenicillin (CSL, Parkville, Australia), 2 mM L-glutamine (MP Biomedicals), 55 M ␤-mercaptoethanol (Sigma-Aldrich), and 100 mM non-essential amino acids (Gibco)). Samples were split into stimulators and responders at a ratio of 1:2 for healthy donors or 1:1 for HCVinfected individuals. Stimulators were pulsed with 10 g/ml peptide for 90 min, washed, and added to the responders. Media half changes were done twice weekly with RF10 supplemented with 10 -20 units/ml of IL-2 with or without 50% T cell growth factor (supernatant from MLA44 cell). Cultures derived from HCV-infected individuals were supplemented with 25 units/ml IL-7 on day 3. The cultures were restimulated on day 7 and then weekly for CD8 ϩ T cell lines at a 1:10 -15 stimulators to responders ratio with C1R-HLA-A*02:01-expressing cells pulsed with 10 g/ml peptide for 60 min and gamma-irradiated at 8,000 rad. Sequences for conservation analysis were obtained from the NCBI Influenza Research Database. Full-length sequences were obtained and were aligned using the IEDB Analysis Resource. All vaccine strains, sequences from viruses isolated within Oceania, and all pH1N1 sequences were chosen for analysis. The frequency of each variant is shown in the table; the underlined residues are those that vary from the WT NA 231 epitope. Intracellular Cytokine Staining-CD8 ϩ T cell lines were stimulated with C1R-HLA-A*02:01-expressing cells pulsed with 10 g/ml of peptide for 1 h at a ratio of 1:2 stimulators to responders in the presence of IL-2. The cells were incubated for 1 h and then supplemented with 10 g/ml brefeldin A (Sigma-Aldrich) and incubated for a further 5 h. Cells were then surface-stained with ␣CD3-PeCy7 and ␣CD8-PerCP (BD Biosciences) for 30 min and fixed in 1% paraformaldehyde (PFA; Electron Microscopy Sciences, Hatfield, PA) for 15 min. The cells were permeabilized with 0.3% saponin (Sigma) diluted in PBS and stained for internal cytokines ␣IFN␥ (PE or FITC) and ␣TNF␣ (FITC or PE) overnight (BD Biosciences). The samples were acquired on the BD FACS Canto II (BD Biosciences) and analyzed using Flowjo software (Treestar).
Single-cell Sorting and Multiplex PCR-Epitope-specific CD8 ϩ T cells were antibody-stained, and CD3 ϩ CD8 ϩtetramer ϩ cells were single-cell sorted directly into 96-well PCR plates (Eppendorf, Hamburg, Germany) using the BD FACS Aria II (BD Biosciences). cDNA was synthesized using the VILO RT kit (Invitrogen). Multiplex nested PCR was performed as previously described (34,52). Briefly, the external round was performed with 40 V␣-external and 27 V␤-external primers and the C␣-external and C␤ external primers. The internal PCR rounds were performed individually with the 40 V␣-internal and C␣-internal or the 27 V␤-internal and the C␤-internal primers (34,52). PCR products were detected on a 2% agarose gel and purified using Exosap (Affymetrix) or Exostar (GE Healthcare). The sequencing reaction was performed using the C-internal primer and BigDyeV3.1 (Applied Biosystems) and cleaned on DyeEx sequencing plates (Qiagen). Sequencing was performed by the Pathology Department at the University of Melbourne (Melbourne, Australia). The sequences were analyzed using FinchTV, and V and J region usage was identified by IMGT query (53). CDR3 amino acid sequences described within the manuscript are productive (without stop codons and with an in-frame junction) ␣␤TCR pairs and start from CDR3 position 1, as determined by the IMGT software. CDR3 length and motifs are from position 4. Samples with two ␣ or ␤ chains were dissected by performing an internal round of PCR with the specific individual primers and then sequenced as above. Where possible, both chains were resolved; however, in a few instances only the dominant chain could be resolved. Samples with TRBV19 (M1 58 ϩ CD8 ϩ T cell repertoire) had the internal round of PCR repeated with TRBV19 and were subsequently sequenced with the TRBV19 forward primer. IMGT nomenclature was used (54). CDR3 motif conservation analysis (Fig. 8) was generated using the Seq2Logo 2.0 server (55) using the Shannon logo type, no clustering, 0 weight on prior, and seq2logo default amino acid color settings.