Broad CD8+ T cell cross-recognition of distinct influenza A strains in humans

Newly-emerged and vaccine-mismatched influenza A viruses (IAVs) result in a rapid global spread of the virus due to minimal antibody-mediated immunity. In that case, established CD8+ T-cells can reduce disease severity. However, as mutations occur sporadically within immunogenic IAV-derived T-cell peptides, understanding of T-cell receptor (TCRαβ) cross-reactivity towards IAV variants is needed for a vaccine design. Here, we investigate TCRαβ cross-strain recognition across IAV variants within two immunodominant human IAV-specific CD8+ T-cell epitopes, HLA-B*37:01-restricted NP338-346 (B37-NP338) and HLA-A*01:01-restricted NP44-52 (A1-NP44). We find high abundance of cross-reactive TCRαβ clonotypes recognizing distinct IAV variants. Structures of the wild-type and variant peptides revealed preserved conformation of the bound peptides. Structures of a cross-reactive TCR-HLA-B37-NP338 complex suggest that the conserved conformation of the variants underpins TCR cross-reactivity. Overall, cross-reactive CD8+ T-cell responses, underpinned by conserved epitope structure, facilitates recognition of distinct IAV variants, thus CD8+ T-cell-targeted vaccines could provide protection across different IAV strains.

I nfluenza A viruses (IAVs) rapidly evolve and cause significant morbidity and mortality (reviewed in refs. 1,2 ). Annual epidemics are responsible for >500,000 deaths worldwide 3 , while pandemics can cause >50 million deaths (reviewed in ref. 4 ). Although vaccines are available, they primarily induce neutralizing antibodies directed towards the rapidly mutating surface glycoproteins, rather than cross-reactive CD8 + T cell immunity 1,5 , mandating that these vaccines are updated and administered annually (reviewed in ref. 6 ). Furthermore, these vaccines are fallible when the circulating strains do not match the predicted vaccine strains 7 or in a scenario when a novel viral subtype enters the population. Thus there is an urgent need to understand correlates of T cell protection towards IAV to provide effective influenza vaccine design.
It is well established that CD8 + T cells with diverse T cell receptor (TCR) repertoires are greatly beneficial for disease outcome, contributing to reduced disease severity 27 , enhanced CD8 + T cell function 28 , cross-reactivity across different peptide variants 29,30 , and preventing viral escape 31,32 . Importantly, although CD8 + TCRs are typically highly specific for their cognate peptide, they can also recognize a broad range of peptide variants, thus allowing CD8 + T cells to have a powerful capacity to recognize not only their cognate peptide but also a range of viral mutants 11,30,[33][34][35][36] . In case of highly mutating influenza viruses, such cross-reactive CD8 + T cells are highly desirable as they elicit immune responses towards multiple viral strains and hence provide cross-strain protection.
The precise mechanisms underlying cross-recognition by influenza-specific CD8 + TCRs in humans are unclear. To date, TCRαβ repertoires have only been dissected for two immunodominant influenza-specific human epitopes, HLA-A*02:01restricted M1 58 30 and HLA-B*35:01/*35:03/*07:02-restricted NP 418 30 , providing ∼50% of the cumulative population coverage. Thus it is important to understand cross-reactivity and diversity of CD8 + T cell TCRαβ repertoires directed against other prominent IAV-specific epitopes, if we are to rationally design a broadly protective CD8 + T cell-mediated influenza vaccine.

Results
Only HLA-B37 + donors elicit a NP 388 + CD8 + T cell response. Our previous work identified NP 338 as an immunodominant CD8 + T cell epitope in individuals expressing HLA-B37 39 . However, NP 338 was previously reported to be restricted by HLA-B*44:03 41 . Thus we first investigated whether the NP 338 peptide can be presented by other HLA allomorphs using the online SYFPETHI peptide-binding prediction tool 42 . A high prediction binding score for the NP 338 peptide was obtained for HLA-B37, HLA-B*44:02, and HLA-B*18:01 (HLA-B18) ( Table 1). The SYFPETHI website contains only the HLA-B*44:02 in its database but not the closely related allomorphs, such as HLA-B*44:03 and HLA-B*44:05 43 . The high prediction binding score for the NP 338 peptide obtained for HLA-B*44:02 is likely to be shared by HLA-B*44:03 and HLA-B*44:05 (HLA-B44), given their known overlapping peptide repertoire 44 . Both HLA-B44 [44][45][46][47] and HLA-B18 46,48 display a preference for P2-E and PΩ-F/Y in their bound peptides, and both residues are present in the IAV-derived NP 338 peptide.
We then solved the binary structures of the NP 338 peptide presented by HLA-B37, HLA-B18, and HLA-B44 (Supplementary  Table 2, Fig. 1). The NP 338 peptide adopted an extended conformation in the cleft of HLA-B37, with the P2-Glu and P9-Phe acting as anchor residues in the B-and F-pocket of HLA-B37, respectively 49 . In addition to the primary anchor residues, the P5-Arg acted as a secondary anchor residue, fully buried into the Fpocket, and formed a salt bridge with Asp77 from the α1-helix in the HLA-B37 molecule (Fig. 1a). The remaining six residues were The http://www.syfpeithi.de/bin/MHCServer.dll/Info.htm prediction tool was used to assign a prediction score that the WT NP 338 peptide would be bound by the particular HLA allele. Prediction scores were grouped into a predicted score range for clarity. ND refers to HLA alleles, which have not been assigned a HLA superfamily. Alleles in bold were selected for further analysis all solvent exposed and represented potential contacts for the TCR. Comparison of the HLA-B37-NP 338 with HLA-B18-NP 338 and HLA-B44-NP 338 complexes show an overall similar HLAbinding cleft (Fig. 1d), with a root mean square deviation (r.m.s. d.) of 0.2 and 0.4 Å, respectively. Similar to the HLA-B37-NP 338 structure, the NP 338 peptide was anchored to HLA-B18 and HLA-B44 by P2-Glu and P9-Phe. However, the NP 338 peptide adopted a strikingly distinct conformation when bound to HLA-B37 as compared with HLA-B18 (r.m.s.d. of the peptide 1.7 Å) and HLA-B44 (r.m.s.d. of 1.8 Å for the peptide). In HLA-B18, P5-Arg is partially buried in the D-/E-pocket and only exposed its guanidinium group to the solvent, stabilized by a salt bridge with the peptide P3-Asp (Fig. 1b). Polymorphism at position 116 between the two HLA molecules change the architecture of the peptide. The HLA-B18 contains a small Ser116, while the HLA-B37 has a large Phe116. As a result, P9-Phe of the peptide is anchored deeper in the cleft in the HLA-B18 molecule, which would sterically clash with the P5-Arg conformation observed in HLA-B37. In addition, the salt bridge observed in HLA-B37 between Asp77 and P5-Arg would be lacking in HLA-B18 due to the presence of Ser77. In HLA-B44, P5-Arg was fully exposed to the solvent (Fig. 1c). Owing to the presence of four tyrosine residues (9,74,99,116) in the cleft of HLA-B44 located underneath P5-Arg, Arg cannot be accommodated inside the cleft and adopt a different conformation from the one observed in HLA-B37, that only shares Tyr74. The additional anchor residue observed in the B37-NP 338 complex could explain the higher stability of this pHLA complex (Supplementary Table 1). These distinct conformations adopted by the NP 338 peptide, due to HLA polymorphism, changed the surface presented to the TCR. On one hand, the NP 338 peptide revealed a hydrophobic surface when bound to HLA-B37, while on the other, it was an epitope with central positively charged residues in the cleft of HLA-B18 and HLA-B44.     (Table 2). Following 10-16 days of peptide stimulation, we detected robust CD8 + T cell responses by interferon-γ (IFNγ) and tumor necrosis factor-α (TNFα) production as well as CD107a expression in the majority (4/5) of HLA-B37 + donors (Fig. 1f, g), with a high level of polyfunctionality (Fig. 1g). However, NP 338 + CD8 + T cell responses were undetectable in all (5/5) HLA-B*44:02, HLA-B*44:03, or HLA-B*18:01 individuals (Fig. 1f, g), despite their CD8 + T cells having responded strongly to non-specific stimulation by phorbol 12-myristate 13-acetate or Concanavalin A (Fig. 1f).
Overall, these data suggest that HLA-B37, but not HLA-B44 or HLA-B18, represents the HLA restriction for the NP 338 peptide. The lack of NP 338 + CD8 + T cell responses in the context of HLA-B44 and HLA-B18 might be due to the distinct conformations, or lower stability, of NP 338 as compared to HLA-B37.
Cross-reactive NP 338 + CD8 + T cells towards NP 338 variants. Having shown robust NP 338 + CD8 + T cell responses in HLA-B37 donors, we next determined the level of cross-recognition towards the main NP 338 variants occurring in IAVs. Our conservation analysis of the NP 338 viral peptides across vaccine and circulating IAV strains, including pH1N1, H5N1, and H7N9, found 16 natural variants of NP 338 peptide (Table 2), with the most common mutations occurring at position (P) 5, 6, and 7 of the peptide. Two major variants; FEDLRVSSF (NP 338 -L7S) and FEDLRLLSF (NP 338 -V6L) were dominant in H1N1, pH1N1, H5N1, and H7N9 or H3N2 strains, respectively, representing 93-100% of the distinct strains and were thus selected in addition to the previously published (WT) peptide for further analysis.
To assess the cross-reactive potential of HLA-B37-NP 338 + CD8 + T cells, pooled or variant-specific NP 338 T cell lines were generated using PBMCs from HLA-B37 + donors. We also generated HLA-B37-NP 338 and mutant HLA-B37-NP 338 -V6L and HLA-B37-NP 338 -L7S tetramers. The ability to recognize, and functionally respond, to each of the NP 338 variant was then assessed using tetramers and an intracellular cytokine staining assay, respectively (Fig. 2). Staining with all three pHLA tetramers and functional assays showed similar proportions of CD8 + T cell staining/functional levels in all WT or variant-specific NP 338 -T cell lines (Fig. 2a, c), suggesting that the majority of NP 338specific CD8 + T cell lines were able to cross-recognize (Fig. 2a, c) and respond to (Fig. 2b, c) each of the three distinct peptides, indicating broad cross-reactivity towards these epitopes (Fig. 2b, c). Interestingly, the NP 338 -V6L tetramer showed a slightly stronger staining than the WT and NP 338 -L7S tetramers (Fig. 2a).
CD8 + T cell lines typically recognized and responded most prominently to the peptide they were generated against, with the lowest level of cross-reactivity being observed by the NP 338 -V6Lspecific CD8 + T cell lines, when stimulated with the NP 338 -WT (21.45 ± 18.86% by tetramer, 20.98 ± 19.80% by IFNγ, and 18.88±18.90% by TNFα) and NP 338 -L7S variant (16.31 ± 14.47% by tetramer, 13.63 ± 12.04% by IFNγ, and 13.04 ± 11.65% by TNFα). These data suggest that a high level of cross-reactivity across different variants exists, although some variant-specific reactivity can be also detected. These data highlight that such cross-reactive CD8 + T cells may offer protection against H1N1, H5N1, H7N9, and possibly other novel IAV strains.
To further assess whether the WT NP 338 peptide elicited CD8 + T cell responses of high functional avidity, CD8 + T cell lines were generated against the WT NP 338 peptide (n = 3), and their functional response was assessed towards decreasing concentrations of the WT NP 338 peptide (Fig. 2d). All data were normalized to the maximal response (when stimulated with maximum peptide). High and stable functional avidity towards NP 338 (IFNγ, TNFα, and CD107a) was observed in cell lines derived from all three donors (Fig. 2d).
Our data show that HLA-B37 + individuals display polyfunctional and highly cross-reactive CD8 + T cell response to the NP 338 epitope and its major variants.   s.d. of 0.12 and 0.09 Å, respectively (Fig. 3e, f). The hydrophobic P6-Val residue sits near the α2-helix and interacts with Val152, Gln155, and Asp156 of the HLA (Fig. 3g). The substitution of P6-Val for P6-Leu in the H3N2 was readily accommodated in the HLA-B37 cleft and did not change either the peptide or HLA conformation (Fig. 3h). The P7-Leu residue sits above Trp147, with its side chain interacting with Ala150, Val152, and Lys146 (Fig. 3i). The substitution from P7-Leu to P7-Ser did not impact the conformation of either the peptide or HLA molecule (Fig. 3j).
We also showed that the two NP 338 variants could stabilize HLA-B37 as efficiently as the NP 338 WT peptide (Supplementary  Table 1). Overall, the two major variants of the NP 338 epitope adopted the same conformation as the NP 338 epitope and represent molecular mimics of the WT peptide, which might favor T cell cross-reactivity.
Overall, despite entirely private TCRαβ repertoires being detected within each individual, a high level of inter-variant cross-reactivity was achieved in each donor, facilitated by key TCRs.
Biased TRBV19 + TCR chains mainly contacts the HLA-B37. To further understand the molecular mechanism underlying T cell cross-reactivity, we then investigated the cross-reactive clonotype described above (EM2 TCR; TRAV30/TRBV19, Supplementary Table 4) that was able to recognize NP 338 and its two most common variants. We solved the structure of the EM2 TCR in complex with HLA-B37-NP 338 (Supplementary Table 2, Fig. 6). The EM2 TCR docked centrally onto HLA-B37-NP 338 with a 66°angle (Fig. 6b), falling in the range previously observed for TCR-pHLA-I (average of 63°) 50 . The buried surface area (BSA) of the complex was of~1800 Å 2 , also falling in the previously observed range for TCR-pHLA-I (average of 1900 Å 2 ) 50 . The EM2 TCR α-chain contributes to 40% of the contact surface ( Fig. 6c), while the biased TRBV19 + β-chain contributes to 60% of the BSA, providing a direct basis for the TRBV19 bias observed in HLA-B37 + donors in response to NP 338 and its variants. The peptide contributed to 25% of the pHLA BSA, with all six solventexposed residues contacted by the EM2 TCR. CDR2β (27%), CDR3α (23%), and CDR3β (20%) are the highest contributors to the BSA, followed by CDR1α, CDR2α, and CDR1β (6-8% each), as well as the framework (FW) residues of the β-chain (9%) (Fig. 6c).
All of the CDR loops of the EM2 TCR contacted the HLA-B37 molecule (Fig. 6d), with the addition of two FW residues from the β-chain (Gln67β and Lys83β). From the EM2 α-chain, the CDR1/ 2 loops contributed to~10% of the TCR-HLA BSA and 16% for the CDR3α (Supplementary Table 5). The CDR1/2α loops made a contact with the region before the hinge of the HLA α2-helix (residue 151-158, Fig. 6f), while the CDR3α loop contacted the N-terminal part of the HLA α1-helix (residues 58 and 62, Fig. 6g). Within the β-chain, the CDR1 made only few contacts with the HLA (~2% TCR-HLA BSA), and the contacts were driven by CDR2β (34%), CDR3β (14%), as well as FWβ (11%) (Fig. 6d). The CDR3β loop contact with the HLA focused mostly on Gln155, which has its side chain sandwiched by the Tyr30α (CDR1α) and the 110 SAMGT 114 motif from the CDR3β loop (Supplementary Table 5, Fig. 6h). The FW residue Lys83β formed Fig. 2 NP 338 -specific CD8 + T cell lines cross-react with variant NP 338 peptides. PBMCs from HLA-B*37:01 + donors (n = 4) were stimulated with a pool (10 μM each, 30 μM total) or individual (10 μM) NP 338 peptides for 10-11 days and cross-reactivity was assessed by tetramer staining and ICS assays. Gating strategy: lymphocytes, singlets, live + , CD3 mid-high , CD8 + T cells; as per Supplementary Figure 2a, b. a Representative dot plots of a pooled NP 338specific CD8 + T cell line stained with each NP 338 variant tetramers. b Representative dot plots of IFNγ + TNFα + production by pooled NP 338 -specific CD8 + T cells. c Summary of tetramer staining, IFNγ (minus no peptide control), and TNFα (minus no peptide control) production by pooled or variant-specific NP 338 -specific CD8 + T cell lines in Donors 1-3. d Summary of IFNγ, TNFα, and CD107a production, minus no peptide control, of WT-specific CD8 + T cell lines from Donors 1, 2, and 4 in response to concentrations ranging between 10 μM and 0.125 μM of the WT-NP 338 peptide. Values are represented as a proportion of maximum tetramer staining or cytokine production against the line-specific peptide a network of salt bridge with the HLA-B37 Glu76 and CDR1β Asp30, hovering above P8-Ser (Fig. 6i). The CDR2β loop form a hairpin loop that sat above the HLA α1-helix covering a large stretch of the HLA helix spanning from residue 65 to 79 (Fig. 6b, d, j and Supplementary Table 5). The residue Gln67β from the FW, located downstream of the CDR2β loop was also contacting the α1-helix of the HLA-B37 molecule. Gln57β, Asn60β, Asp61β, and Gln67β were forming a large hydrogen network with the HLA-B37 molecule.
Overall, CDR2β and FWβ make extensive interaction with the HLA-B37 molecule, providing a basis for the strong bias TRBV19 + usage observed in the donors.
EM2 TCR contacts the NP 338 peptide via a hydrophobic network. The peptide is predominantly contacted by CDR3α (47%) and CDR3β (26%), as well as by CDR1β (17%) and CDR2β (10%) (Fig. 6b, e, Supplementary Table 5). The motif SGG within the CDR3α loop, with small or no side chain, allow the loop to be in close proximity and lodge itself between the side chains of P1-Phe and P3-Asp (Fig. 7a). The proximity of the loop allows the Gln113α to form hydrogen bonds with P3-Asp of the peptide and hydrophobic interaction with P4-Leu. Therefore, CDR3α loop forms a lid-like structure that covers the N-terminal part of the peptide (Supplementary Table 5). Within the EM2 TCR β-chain, all the CDR loops were contacting the peptide C-terminal section from residue P4-Leu to P8-Ser. P4-Leu was contacted by Gln57β from the CDR2β loop, while the CDR1β loop was interacting with both P7-Leu and P8-Ser that form a hydrogen bond with Asp30β (Supplementary Table 5). P6-Val and P7-Leu were both contacted by the CDR3β loop that sat on top of the central region of the peptide and form Van der Waals interaction. The 109 MSAM 112 motif from the CDR3β loop could hover on the central part of the peptide due to the small side chains of the 110 SA 111 residues. The CDR3β loop was thus making contacts with P6-P7 residues of the peptide mainly via Van der Waals bonds (Supplementary Table 5).
Interestingly, the comparison of HLA-B37-NP 338 in its free and liganded state with the EM2 TCR revealed that the peptide (r.m.s. d. of 0.2 Å) and HLA-B37 (r.m.s.d. of 0.3 Å) did not change conformation upon TCR binding (Fig. 7c). The hydrophobic nature of the interaction between the EM2 TCR and the NP 338 peptide is complementary with the conformation of the NP 338 peptide bound to HLA-B37 that presents only the hydrophobic residue to the TCR. This might explain the lack of response for NP 338 when presented by both HLA-B18 and HLA-B44 molecules, in which the peptide presents its charged P5-Arg to the TCR (Fig. 1b, c).
Altogether, the structure of the EM2 TCR-HLA-B37-NP 338 complex revealed that the TCR is using hydrophobic interaction from its CDR3 loops to engage with the all stretch of the NP 338 peptide.
EM2 TCR affinity for NP 388 -V6L is higher than for NP 388 -WT. We next evaluated the ability of the EM2 TCR to recognize the NP 338 and its two major variants (NP 338 -V6L and NP 338 -L7S)  Table 6). The EM2 TCR bound to the HLA-B37-NP 338 complex with a K d of~130 µM (Fig. 7d), displaying a low affinity for an antiviral CD8 + T cell, the average being 35 µM 50 . The EM2 TCR bound to NP 338 -L7S with a similar affinity to the one observed for the NP 338 bound to HLA-B37 (K d~1 50 µM, Fig. 7e). Interestingly, the EM2 TCR affinity was increased by four-fold for the NP 338 -V6L (K d~3 0 µM, Fig. 7f) compared to NP 338 when bound to HLA-B37. The structures of each NP 338 and variants in complex with the HLA-B37 molecule (Fig. 3, Supplementary Table 2) allowed us to understand the impact of the NP 338 mutation onto the EM2 TCR recognition. Overlay of the HLA-B37-NP 338 -L7S with the EM2 TCR-HLA-B37-NP 338 structures showed that the smaller P7-Ser would be readily accommodated by the EM2 TCR without structural changes (Fig. 7g). This might explain the similar affinity observed for the EM2 TCR towards both peptides (K d1 30-150 µM). In a similar fashion, an overlay of the HLA-B37-NP 338 -V6L with the EM2 TCR-HLA-B37-NP 338 structures (Fig. 7h) shows that the larger side chain of the P7-Leu might create additional contact with the CDR3β loop and might in turn provide a basis for the improved affinity observed for the EM2 TCR towards the HLA-B37-NP 338 -V6L (K d~3 0 µM).
Thus our data demonstrate that CD8 + T cell cross-reactivity in the context of NP 338 peptide and its variants is favored by similar conformations adopted by the epitopes.
Key TCRs are cross-reactive towards A1-NP 44 variants. To investigate whether cross-reactivity and thus protection against distinct IAV strains is NP 338 -specific or occurs towards other IAV-specific epitopes, we assessed CD8 + T cell responses towards     Table 7). To investigate cross-reactivity towards these variant NP 44 peptides, PBMCs from healthy HLA-A1 + donors (n = 3) were tetramer stained with each of the NP 44 variants directly ex vivo and magnetically enriched and single-cell sorted (Fig. 8a). The TCRαβ repertoire was then determined using multiplex RT-PCR (Fig. 8b). Two of the three donors (Donors 1 and 20) had both NP 44 -and NP 44 -S7N-specific CD8 + T cell populations. Interestingly, Donor 21 had only a NP 44 -specific CD8 + T cell population, with no detectable S7N-specific CD8 + T cell population (Fig. 8b) Fig. 6 Structure of EM2 TCR in complex with the HLA-B37-NP 338 . a Overview of the EM2 TCR (α-chain in pink and β-chain in blue) bound to the HLA-B37-NP 338 complex (HLA in white and peptide in black). b The EM2 TCR-binding footprint on the HLA-B37 surface (white) and NP 338 peptide (gray surface) is represented by the atomic contact made by each TCR residues colored in teal, green, and purple for the CDR1, 2, and 3 of the α-chain; red, orange, and yellow for the CDR1, 2, and 3 of the β-chain; and pink and blue when contacted by the framework residue of the αand β-chain, respectively. The pie charts represent the buried surface area at the interface between c the TCR with the HLA-B37-NP 338 or with d HLA-B37 alone or e with the peptide alone. The different colors represent the TCR gene segment as per b. Interaction of the EM2 TCR with the HLA-B37 molecule (white) are represented for f CDR1α and CDR2α; g CDR3α; h CDR3β; i CDR1β and FWβ; and j for CDR2β. The same color scheme is used as per b for the different segments of the TCR. The sphere represents the Cα atom of glycine residue, the blue dashed lines represent hydrophobic interaction, while the red dashed lines indicated hydrogen bond or salt bridges ARTICLE NATURE COMMUNICATIONS | https://doi.org/10.1038/s41467-018-07815-5 isolated from Donor 1 could recognize both NP 44 variants, while the remainder of the repertoire was variant specific (Fig. 8b). Interestingly, the TCRαβ repertoire of Donor 20 was entirely NP 44 variant specific. We previously solved the structures of both NP 44 and NP 44 -S7N peptides in complex with the HLA-A1 molecule 23 . In similar fashion as HLA-B37 and the NP 338 variants, the NP 44 variants adopted similar conformation in the cleft of HLA-A1 molecule. This suggests that here too molecular similarity might underpin CD8 + T cell cross-reactivity and favor the recognition of multiple IAV strains.
Overall, these data show that TCRαβ cross-reactivity is not limited to the recognition of HLA-B37-NP 338 and highlight the importance of cross-reactive CD8 + T cells in the recognition of virus-infected cells and thus their potential to protect against distinct IAV strains.

Discussion
There is great interest in the development of a universal CD8 + T cell-mediated vaccine towards IAV. However, to realize this, first a thorough understanding of prominent human influenza-specific CD8 + T cell epitopes is needed. We have previously shown that NP is the most immunogenic antigen in both HLA-A*02:01-positive 51 and -negative individuals 39 . However, as the conservation of the known influenza epitopes varies between 16% and 56%, depending on ethnicity 23 , understanding inter-epitope cross-recognition across diverse IAV strains is needed. In this study, we have used the variable NP 338 and NP 44 peptides to dissect CD8 + T cell recognition of different peptide variants and thus distinct IAV strains.
Clone  NP 338 has two major variant peptides, covering~93% (NP 338 -L7S) and~96% (NP 338 -V6L) of H1N1 and H3N2 viruses, respectively. Although CD8 + T cell responses are highly peptide specific 34 , they can tolerate variations within CD8 + T cell peptides 22,29,30,34 . We identified functional cross-reactivity towards the variant NP 338 peptides, albeit to different levels. Using a multiplex RT-PCR for the detection of the TCRαβ repertoire directly ex vivo, we confirmed that CD8 + T cells were cross-reactive towards these distinct NP 338 variants, at a clonotypic level, with ∼55% TCRαβ repertoire being capable of recognizing multiple NP 338 variants. Interestingly, unique clonotypes with distinct TRAV-TRBV pairings in each individual recognized all three NP 338 variants. Furthermore, this crossreactivity was increased following in vitro amplification, potentially suggesting that activation increases expansion of crossreactive clones, allowing certain TCRαβ clonotypes to recognize distinct viral variants and potentially prevent viral escape. Additionally, we showed that the importance of cross-reactive TCRαβ repertoires is not limited to the recognition of NP 338 and that cross-reactive CD8 + TCRαβ can recognize HLA-A1restricted NP 44 variant peptides.
To understand the molecular basis of cross-reactive TCRαβ recognition, we solved the structure of NP 338 and its variants bound to the HLA-B37 molecule, as well as the structure of a cross-reactive TCR in complex with HLA-B37-NP 338 . First, our data show that the two NP 338 variants were structural mimics of the WT peptide. Second, the NP 338 peptide conformation was not altered by TCR engagement, and therefore the conserved structure of the NP 338 variants favored TCR cross-reactivity, a mechanism that is thought to underpin cross-reactivity and alloreactivity more broadly 45,52,53 . The structures also demonstrated that only HLA-B37, and not HLA-B18 nor HLA-B44, binds the NP 338 with a secondary anchor residue (P5-Arg), revealing the hydrophobic surface of the NP 338 peptide, which might help cross-reactivity 36,54 . This hydrophobic surface of the NP 338 was disrupted when the peptide was presented by HLA-B44 or HLA-B18 molecules, which exposed the charged P5-Arg, and for which no T cell recognition was observed.
Together, these data imply that HLA-B37 + donors may have a level of protection against H1N1, pH1N1, H3N2, H5N1, H7N9, and possibly novel IAV strains. Furthermore, the findings highlight the potential of inducing inter-strain cross-protection by the inclusion of a single peptide variant in a CD8 + T cell-mediated vaccine, suggesting that vaccine epitope identification will no longer need to focus solely on conserved epitopes for universal inter-strain cross-protection.
Diverse and cross-reactive TCR repertoires are important for viral control and preventing viral escape 27,29 , and we thus dissected the importance of TCRαβ diversity in the recognition of the variant NP 338 peptides. Our data show that cross-recognition of distinct IAV strains is underpinned by a diverse TCR repertoire and surprisingly that no shared clonotypes could be detected between individuals. This highlights that cross-reactivity and TCRαβ repertoire diversity may provide further protection against distinct IAV strains. Furthermore, TCRαβ diversity is likely to aid in limiting viral escape, offering protection against any quasispecies that arise during infection. Additionally, we have shown that cross-reactivity towards distinct IAV strains is not limited to HLA-B37-NP 338 , and cross-reactive TCRs towards the HLA-A1-restricted NP 44 were also observed. In both contexts, we have isolated cross-reactive clonotypes that were not shared between individuals, showing the plasticity of the TCR repertoire whereby multiple TCRs are able to cross-react and provide protection against multiple IAV strains.
Overall, our study provides evidence that inter-epitope crossreactivity is common for the human influenza epitopes and occurs for prominent HLA-B37-restricted NP 338 and HLA-A1restricted NP 44 epitopes, in addition to previously published HLA-A*02:01-restricted M1 58 30 . CD8 + T cells can recognize variations within CD8 + T cell epitopes and cross-reactive CD8 + TCRαβ repertoires can recognize variant HLA-B37-restricted NP 338 and HLA-A1-restricted NP 44 peptides across donors. Together, these data highlight the importance of CD8 + T cell cross-reactivity and diversity in the protection against distinct IAV strains and the induction of such diverse and cross-reactive CD8 + T cells by a CD8 + T cell-mediated vaccine could provide universal and superior protection against distinct IAV strains. Isolation of PBMCs. Buffy coats were obtained from the Australian Red Cross Blood Service (ARCBS) and whole blood donations from healthy volunteers. PBMCs were separated using density gradient separation and were cryopreserved until use. All samples were HLA-typed by the Victorian Transplant and Immunogenetics Service (VTIS, West Melbourne, VIC, Australia) at the ARCBS tissuetyping Laboratory.
Single-cell multiplex RT-PCR for TCRαβ analysis. Epitope-specific CD8 + T cells from PBMCs or CD8 + T cell lines were tetramer and antibody stained, as described above. Following surface staining, cells were re-suspended in sort buffer (PBS +0.1% BSA). Single epitope-specific CD8 + T cells were sorted on the BD FACS Aria III (BD Biosciences) directly into PCR plates (Eppendorf, Hamburg, Germany). cDNA was synthesized using the VILO RT Kit at 1/20 of the manufacturer's recommendation, supplemented with 0.1% Triton X (Sigma, MO, USA) 30,37,56 . The resulting cDNA was subject to a nested PCR containing 40-Vα and 27-Vβ primers and Cα and Cβ primers (Supplementary Table 9). PCR products were purified using Exo-Sap or Exo-Star (GE Healthcare, Buckinghamshire, UK) and were sequenced using BigDyeV3.1 (Applied Biosystems, CA, USA). Sequencing products were cleaned using DyeEx sequencing plates (Qiagen, Limburg, Netherlands) and sequencing was performed by the Department of Pathology at the University of Melbourne (Melbourne, VIC, Australia). Sequences were analyzed using Finch TV (Geospiza, WA, USA) and the IMGT query software 57,58 (www. imgt.org/IMGT_vquest). CDR3 amino acid sequences described within the text are productive (no stop codons and an in-frame junction) TCRαβ pairs and start from CDR3-position 1, as determined by the IMGT query software. CDR3 length is calculated from position 4 and excludes the final xF motifs. TRAV and TRBV nomenclature was derived from IMGT www.imgt.org/IMGTrepertoire/ LocusGenes/#J. NP 338 + CD8 + T cells with two alpha or beta chains were dissected by performing the internal round of PCR with the specific individual primers and subsequently sequenced using the relevant constant reverse primer or variablespecific primers (Supplementary Table 9). Where possible, both chains were resolved; however, in a few instances only the dominant chain could be resolved. TRBV19 + TCRβs were sequenced with the TRBV19 forward primer (and are thus referred to as TRBV19) and CDR3 regions were determined using the EXPASY translate tool.
In cases when NP 338 + CD8 + TCRα or TCRβ chains could not be resolved, sequences (two cells per clonotype) were resolved by cloning and colony PCR. PCR products were ligated into a pGEM-T Easy Vector (Promega, WI, USA) overnight at 16°C. DH5α cells were then transformed with this vector and were grown, plated on Luria agar plates (containing 100 μg/mL ampicillin, Media Preparation Unit, Department of Microbiology and Immunology, University of Melbourne, VIC, Australia) for 16-18 h, and were selected using X-gal (800 μg). Individual transformed colonies were amplified by PCR with 20 pmol sense and anti-sense vector primers (Supplementary Table 9). PCR were amplified as follows: 1 cycle of 95°C for 5 min, 30 cycles of 95°C for 30 s, 57°C for 30 s, and 72°C for 1 min, followed by 1 cycle of 95°C for 1 min, 57°C for 1 min, and 72°C for 7 min. Products were visualized on a 2% agarose gel and were sequenced as above with 5 pmol of sense vector primer (Supplementary Table 9).
Protein expression, purification, and crystallization. The EM2 TCR contained an engineered disulfide linkage in the constant domains between the TRAC and TRBC. The αand β-chains were expressed separately as inclusion bodies in a BL21 Escherichia coli strain. The inclusion bodies were washed and suspended in 6 M guanidine, then mixed into a cold refolding solution containing 5 M urea, 100 mM Tris-HCL (pH 8), 2 mM EDTA, 400 mM L-arginine-HCl, 0.5 mM oxidized glutathione, and 5 mM reduced glutathione. The refolding solution was dialyzed after 3 days against 10 mM Tris-HCl (pH 8) and 150 mM NaCl and then purified using anion exchange and size exclusion columns 59  containing the NP 338 and variant NP 338 -L7S and NP 338 -V6L peptides were prepared similarly to the TCR without urea in the refolding solution 59 . Crystals of the EM2 TCR-HLA-B*37:01-NP 338 complex or the pHLA individually in 10 mM Tris-HCl (pH 8.0) and 150 mM NaCl were grown by the hanging-drop, vapor-diffusion method at 20°C with a protein to reservoir drop ratio of 1:1 and a protein concentration of 6-10 mg/mL. The pHLA crystals formed in 20% PEG 6000, 0.2 M NaCl, 0.1 M Na citrate pH 6.5 while the EM2 TCR-HLA-B*37:01-NP 338 complex crystals formed in 19% PEG 3350 and 0.2 M di-ammonium tartrate. Monomers of pHLA with the BirA tag construct were biotinylated and tetramerized in a 1:4 ratio with PE-streptavidin (Life Technologies, CA, USA) in 6 additions over the course of 1 h.
Data collection and structure determination. Crystals were soaked in a cryoprotectant solution containing mother liquor solution with the PEG concentration increased to 30% (weight per volume) and then flash frozen in liquid nitrogen. Data were collected on the MX1 beamline at the Australian Synchrotron, Clayton using the ADSC-Quantum 210 CCD detector (at 100 K) 60 . Data were processed using the XDS software 61 and scaled using the XSCALE software 61 . The pHLA and EM2 TCR-HLA-B*37:01-NP 338 complex structures were determined by molecular replacement using the PHASER 62 program with the HLA-A*02:01-NLV complex as the search model for the HLA without the peptide (Protein Data Bank accession number, 3GSO 59 ) and the LC13 TCR for the TCR (Protein Data Bank accession number, 1KGC 63 ). Manual model building was conducted using the Coot software 64 followed by maximum-likelihood refinement with PHENIX 65 and Buster programs 66 . The TCR was numbered using the IMGT nomenclature 57 . The final models have been validated using the Protein Data Base validation website and the final refinement statistics are summarized in Supplementary table 2. Coordinates were validated by the PDB database. All molecular graphic representations were created using PyMol 67 .
Thermal stability assay. The thermal stability assay was performed in the Real Time Detection system (Corbett RotorGene 3000) using the fluorescent dye Sypro orange to monitor protein unfolding. The pHLA complexes (5 and 10 μM) in 10 mM Tris-HCl pH 8 and 150 mM NaCl were heated from 30 to 95°C with a heating rate of 1°C/min. The fluorescence intensity was measured with excitation at 530 nm and emission at 555 nm. The thermal melt point (Tm) represents the temperature for which 50% of the protein is unfolded.
Surface plasmon resonance measurement and analysis. Surface plasmon resonance experiments were conducted at 25°C on the BIAcore 3000 instrument with TBS buffer (10 mM Tris-HCl, pH 8, 150 mM NaCl, and 0.005% surfactant P20). TBS buffer was supplemented with 1% BSA to prevent non-specific binding. The human TCR-specific monoclonal antibody, 12H8 antibody 68 , was coupled to research-grade CM5 chips with standard amine coupling. The experiment was conducted with one empty flow cell (use for subtraction as blank) and one flow cell with the EM2 TCR captured to approximately 300 response unit per flow cell onto the 12H8 antibody 68 . Then a concentration range of 200 µM maximum of the pHLA complexes were passed over all the flow cells. The final response was calculated by subtracting the response of the antibody alone and that of EM2 TCR. BIAevaluation Version 3.1 was used for data analysis using the 1:1 Langmuir binding model. Experiments were carried out in duplicate (n = 2).
Statistical analyses. All statistical analyses were undertaken using Prism 6 (GraphPad, CA, USA). A Dunnett's two-way analysis of variance was used for all non-parametric non-paired analysis where multiple comparisons were made. Statistical significance was defined as *p ≤ 0.05, **p ≤ 0.01, and ***p ≤ 0.001.