CD8+ T cell landscape in Indigenous and non-Indigenous people restricted by influenza mortality-associated HLA-A*24:02 allomorph

Indigenous people worldwide are at high risk of developing severe influenza disease. HLA-A*24:02 allele, highly prevalent in Indigenous populations, is associated with influenza-induced mortality, although the basis for this association is unclear. Here, we define CD8+ T-cell immune landscapes against influenza A (IAV) and B (IBV) viruses in HLA-A*24:02-expressing Indigenous and non-Indigenous individuals, human tissues, influenza-infected patients and HLA-A*24:02-transgenic mice. We identify immunodominant protective CD8+ T-cell epitopes, one towards IAV and six towards IBV, with A24/PB2550–558-specific CD8+ T cells being cross-reactive between IAV and IBV. Memory CD8+ T cells towards these specificities are present in blood (CD27+CD45RA− phenotype) and tissues (CD103+CD69+ phenotype) of healthy individuals, and effector CD27−CD45RA−PD-1+CD38+CD8+ T cells in IAV/IBV patients. Our data show influenza-specific CD8+ T-cell responses in Indigenous Australians, and advocate for T-cell-mediated vaccines that target and boost the breadth of IAV/IBV-specific CD8+ T cells to protect high-risk HLA-A*24:02-expressing Indigenous and non-Indigenous populations from severe influenza disease.

N ewly emerging respiratory viruses pose a major global pandemic threat, leading to significant morbidity and mortality, as exemplified by 2019 SARS-CoV2, avian influenza H5N1 and H7N9 viruses, and the 1918-1919 H1N1 pandemic catastrophe. Influenza A viruses (IAV) can cause sporadic pandemics when a virus reassorts and rapidly spreads across continents, causing millions of infections and deaths 1 . Additionally, seasonal epidemics caused by co-circulating IAV and influenza B viruses (IBV) result in 3-5 million cases of severe disease and 290,000-650,000 deaths annually 2,3 . Severe illness and death from seasonal and pandemic influenza occur disproportionately in high-risk individuals, including Indigenous populations. This is most evident when unpredicted seasonal or pandemic viruses emerge in human circulation. During the 1918-1919 influenza pandemic, 100% of Alaskan adults died in some isolated villages, while only school-aged children survived 4 . Western Samoa was the hardest hit with a total population loss of 19-22% 5 . As many as 10-20% of Indigenous Australians died from pandemic influenza in 1919 6 in comparison to <1% of other Australians, with some reports showing up to 50% mortality in Indigenous Australian communities 7 .
During the 2009 A/H1N1 influenza pandemic, Indigenous populations worldwide were more susceptible to influenza-related morbidity and mortality. Hospitalization and morbidity rates were markedly increased in Indigenous Australians 8,9 , with 16% of hospitalized pandemic H1N1 (pH1N1) patients in Australia being Indigenous. The relative risk for Indigenous Australians compared to non-Indigenous Australians for hospitalization, ICU admission, or death was 6.6, 6.2, or 5.2 times higher, respectively 10 . This was mirrored in Indigenous populations globally, including American Indians and Alaskan Native people (4-fold higher mortality rate compared to non-Indigenous Americans) 11 , native Brazilians (2-fold higher hospitalization rate) 12 , New Zealand Maori (5-fold higher hospitalization rate), and Pacific Islanders (7-fold higher hospitalization rate) 13,14 .
Although the impact of influenza pandemics is more pronounced in Indigenous populations globally, these disproportionate hospitalization rates also occur during seasonal infections. During 2010-2013, Indigenous Australians had increased influenzarelated hospitalizations across all age groups (1.2-4.3-fold higher compared to non-Indigenous) 15 . Indigenous populations, especially Australians and Alaskans, are also predicted to be at greater risk from severe disease caused by the avian-derived H7N9 influenza virus, with mortality rates being >30% and hospitalization >99% in China 16 . While higher influenza infection rates could relate to overcrowded living conditions, increased severity and prolonged hospitalization most likely reflect differences in pre-existing immunity that facilitates recovery. However, the underlying immunological and host factors that account for severe influenza disease in Indigenous individuals are far from clear.
Antibody-based vaccines towards variable surface glycoproteins, hemagglutinin (HA), and neuraminidase (NA), are an effective way to combat seasonal infections, yet they fail to provide effective protection when a new, antigenically different IAV emerges 17 . In the absence of antibodies, recall of pre-existing cross-protective memory CD8 + T-cells minimizes the effects of a novel IAV, leading to a milder disease after infection with distinct strains [18][19][20][21][22][23] . Such pre-existing memory CD8 + T-cells provide broadly heterotypic or cross-reactive protection and can recognize numerous IAV, IBV, and influenza C viruses capable of infecting humans 24 , promoting rapid host recovery. During the 2013 H7N9 IAV outbreak in China, recovery from severe H7N9 disease was associated with early CD8 + T-cell responses 21,25 . Patients discharged early after hospitalization had early (day 10) robust H7N9-specific CD8 + T-cells responses, while those with prolonged hospital stays showed late (day 19) recruitment of CD8 + and CD4 + T-cells. Thus, with the continuing threat of unpredicted influenza strains, there is a need for targeting cellular immunity that provides effective, long-lasting, and cross-strain protective immunity, especially for high-risk groups such as Indigenous populations. However, despite the heavy burden of disease in Indigenous communities, there is scant data on immunity to influenza viruses in Indigenous populations from around the world.
As CD8 + T-cell recognition is determined by the spectrum of human leukocyte antigens (HLAs) expressed in any individual, and HLA profiles differ across ethnic groups, defining T-cell epitopes restricted by HLAs predominant in some Indigenous populations is necessary to understand pre-existing CD8 + T-cell immunity to influenza. We previously analyzed the HLA allele repertoire in Indigenous Australians 26 and found that HLA-A*24:02 (referred to as HLA-A24 hereafter), an HLA associated with influenza-induced mortality during the 2009-pH1N1 outbreak 27 , is the second most prominent HLA in Indigenous Australians 26,28 . HLA-A24 is also common to other Indigenous populations highly affected by influenza 16 . Thus, analysis of prominent influenza-specific CD8 + T-cell responses restricted by HLA-A24 is needed to understand the relationship between this allele and disease susceptibility. These specificities will also inform strategies to prime effective T-cell immunity in vulnerable communities.
Here, we define CD8 + T-cell immune landscapes against IAV and IBV, restricted by the mortality-associated HLA-A24 allomorph. We identify IAV-and IBV-specific HLA-A24 immunopeptidomes and screen immunogenicity of novel peptides in HLA-A24-expressing mice, peripheral blood of Indigenous and non-Indigenous HLA-A24 + healthy and influenza-infected individuals, and human tissues. Our studies provide evidence of the breadth of influenza-specific CD8 + T-cell specificities restricted by a mortality-associated risk allomorph HLA-A24. These findings have implications for the incorporation of key CD8 + T-cell targets in a T-cell-mediated vaccine to protect Indigenous people globally from unpredicted influenza viruses.
Identification of novel HLA-A24-restricted IAV and IBV epitopes. To identify new A24/influenza-derived epitopes, we utilized an immunopeptidomics approach to sequence HLA-bound peptides on influenza-infected cells by liquid chromatography-tandem mass spectrometry (LC-MS/MS) 24 . Experiments were performed with the class I-reduced C1R B-lymphoblastoid cell line (minimal HLA-B*35:03 surface expression but normal levels of HLA-C*04:01 expression 33 ) and an HLA-A*24:02-transfected C1R cell line (C1R. A24) 34 . Initial experiments were performed in both cell lines at 16 h post-infection with and without the HKx31 IAV virus, isolating peptide-bound HLA-I molecules post-lysis utilizing the pan HLA-I antibody W6/32 (Supplementary Data 1). In subsequent experiments, to improve the confidence of assignment of binding to HLA-A24, HLA-C*04:01 depletion of the lysates was performed with the HLA-C-specific antibody DT9 prior to isolation of the remaining HLA-I (A*24:02»B*35:03) with W6/32 antibody. Utilizing this workflow 35 , we assessed peptide presentation in C1R.A24 cells at 2, 4, 8, and 12 h post-infection with either A/HKx31 or B/Malaysia/ 2506/2004 viruses. Analyses were restricted to up to 12 h postinfection due to observations of marked HLA-I downregulation at 16 h post-infection 36 .
In total, 12 immunopeptidome data sets containing HLA-A24restricted peptides were generated including 3 from uninfected C1R.A24 cells, 5 from HKx31-infected, and 4 from B/Malaysiainfected C1R.A24 cells ( Supplementary Fig. 1a and Supplementary Data 1). An additional 3 data sets for endogenous HLA-I of C1R cells (C1R W6/32 isolation of HLA-B*35:03 and HLA-C*04:01 after 16 h HKx31 infection; C1R.A24 -DT9 isolation of HLA-C*04:01 from uninfected cells and after 12 h HKx31 infection) and 2 data sets for endogenous HLA-II (C1R.A24: HLA-DR12, -DPB1*04:01,04:02 and -DQ7 from uninfected and 12 h HKx31 infection) were also generated as comparators to help establish true HLA-A24 binders ( Supplementary Fig. 1b, c). Comparisons to previously identified B/Malaysia-derived HLA ligands for C1R were also used 24 to distinguish HLA-B*35:03 and HLA-C*04:01 binding peptides from those binding to HLA-A*24:02. Across the 12 HLA-A24 data sets, a total of 9051 non-redundant peptide sequences were assigned to the human proteome at a 5% false discovery rate (FDR). As expected for HLA-I ligands, the majority of peptides were 9-11 amino acids in length but dominated by 9mers (Fig. 2a). Consistent with the HLA-A24 peptide-binding motif generated by NetMHC4.0 37,38 motif viewer, enrichment of Tyr/Phe at P2 and Phe/Leu/Ile at P9 were observed (Fig. 2b). Peptides binding the endogenous HLA-I of C1R were not removed in this analysis due to the similar preference of HLA-C*04:01 for 9mer peptides possessing Phe/Tyr at P2 and Phe/Leu at P9, which may result in shared ligands ( Supplementary Fig. 1d). To maximize the identification of potential virus-derived peptides, assignments to the viral proteome or 6-frame translation of the viral genome were considered without an FDR cut-off. Instead, lack of appearance in uninfected data sets and predicted binding affinity (NetMHCpan4.0 [39][40][41] for HLA-A24 were used to determine likely candidate epitopes. Thus, 52 HKx31-derived and 48 B/Malaysiaderived peptides were identified as potential HLA-A24-restricted epitopes (Fig. 2c), of which 26 IAV-derived and 29 IBV-derived peptides were identified at a 5% FDR (Supplementary Data 1). The identified peptides spanned the viral proteomes including frameshift proteins, representing 6 IAV proteins and 9 IBV proteins (Fig. 2d, e). Interestingly, most HKx31-derived ligands mapped to PB2 > PB1 > HA viral proteins and none were observed from NA or M1, while B/Malaysia-derived ligands predominantly mapped to NP/HA > NA. During the time course analyses, broadest peptide identification was achieved for both viruses between 8 and 12 h post-infection, while no influenza-derived peptides were identified at 2 h post-infection, and those identified at 4 h were of lower confidence (Fig. 2f, g and Supplementary Data 1). 10 potential HKx31-derived ligands were also identified for each of HLA-B*35:03 and HLA-C*04:01 based on predicted binding and/or appearance in control data sets (Supplementary Fig. 1e and Supplementary Data 1). Furthermore, analysis of the peptides presented by HLA-II molecules showed domination of the virusderived immunopeptidome by HA ( Supplementary Fig. 1f and Supplementary Data 1), as previously observed for IBV 24 . A  Several key determinants can orchestrate the main target and therefore the immunodominance of CD8 + T cells. These include the affinity of the peptide to the HLA I molecule, the affinity of the CD8 + T cells for an epitope, as well as the number of naive epitope-specific CD8 + T cells (as reviewed in ref. 43 ). During the acute phase (d8) of secondary challenge, CD8 + T-cell responses were similarly directed at 6 immunogenic peptides found in the primary infection (PB1 216-224 , PB1 498-505 , PB2 322-330 , PB2 463-471 , PB2 549-557 and PB2 549-559 ) (Fig. 3b). CD8 + T-cell responses towards the marginal epitopes observed in the primary infection were no longer detected. While the reason for the loss of HA 248-259 could be explained by sequence variation between HKx31 and PR8 (IYWTIVKPGDVL vs. YYWTLVKPGDTI) all internal proteins are shared between both viruses. Analysis of influenza-specific CD8 + T-cell numbers showed significant reductions in epitope-specific CD8 + T-cells for 9 out of 11 specificities (NP [39][40][41][42][43][44][45][46][47] , PA 130-138 , PB1 2-10 , PB1 216-224 , PB1 496-505 , PB1 498-505 , PB2 322-330 , PB2 703-710 and PB2 549-559 ) following secondary infection ( Supplementary Fig. 3). Changes in immunodominance after heterologous influenza virus infection have been studied previously, especially in B6 mice. Immunodominance hierarchy and crossprotective capacity of epitope-specific CD8 + T cells depend greatly on the first influenza encounter and presentation of viral epitopes by different cell types (reviewed in ref. 44 ).
Thus, our in vivo screening in HHD-A24 mice identified 6 IAV-derived immunogenic peptides during primary and secondary IAV infection, with prominent CD8 + T-cell responses being heavily biased towards PB1-and PB2-derived peptides (Supplementary Table 2). This is of key importance as the current T-cell vaccines in clinical trials focus mainly on internal proteins like NP and M1 [45][46][47][48][49] , which may be poorly immunogenic in HLA-A24-expressing individuals at risk of severe influenza disease.
To define IBV-specific CD8 + T-cell responses following the secondary challenge, B/Malaysia/2506/2004-infected mice were challenged i.n. after 6-8 weeks with 400 pfu B/Phuket/3073/2013. At d8 after challenge, IBV-specific CD8 + T-cell responses resembled those after primary IBV infection (Fig. 3d), but were focussed more towards the immunodominant epitopes, similar to IAV-specific A24/CD8 + T-cell responses following secondary challenge. Although the B/Phuket virus differs in one amino acid in the NP 164-173 /NP 165-173 and NA 213-222 restimulating with the B/Malaysia variants showed cross-reactivity of CD8 + T-cells between both variants. All other immunogenic epitopes were conserved between both strains. In contrast to IAV infection, the total number of epitope-specific CD8 + T-cells for all immunogenic epitopes after secondary challenge remained comparable in the spleen ( Supplementary Fig. 3). Thus, our data identified prominent A24/CD8 + T-cell responses directed toward IBV during primary and secondary influenza virus infection in HHD-A24 mice (Supplementary Table 3).
IAV/CD8 + T-cells in HLA-A24 + Indigenous and non-Indigenous individuals. Having identified prominent IAVderived CD8 + T-cell epitopes towards the primary and secondary infection in HHD-A24 mice, it was of key importance to define immunodominant CD8 + T-cell sets in HLA-A24-expressing Indigenous and non-Indigenous individuals. For humans, different influenza strains of the same subtype are often co-circulating and mutating to create different variants of the same epitope region. In addition to our identified panel of HLA-A24-binding peptides, we searched the Influenza research database (https://www.ncbi.nlm. nih.gov/genomes/FLU/Database/nph-select.cgi?go=database) to a Length distribution of human proteome-derived HLA-I ligands of C1R.A24 isolated using the pan HLA-I antibody W6/32. Numbers of non-redundant sequences of ≤20 amino acids identified at a 5% FDR across the 12 experiments in which HLA class I was isolated from C1R.A24, filtered for peptides identified in HLA class II isolations at a 5% FDR. b Sequence logo derived from human 9mer peptides in a using Seq2logo2.0 75 . c Peptide length distribution of IAV (HKx31) and IBV (B/Malaysia) derived peptides identified as likely HLA-A*24:02 ligands in this study (no FDR cut-off applied) as shown in Supplementary Data 1. d, e Distribution of IAV-(d) and IBVderived (e) A*24:02 ligands from c across the viral proteomes, including potential identifications from alternate reading frames (PB1 + 2, PB1 + 3, PB1-1, HA + 2, HA + 3 and NA + 3). f, g Identification of specific ligands derived from IAV (f) and IBV (g) in isolations performed at 2, 4, 8, 12, and 16 h (16 h IAV only) post infection. Colored squares represent the identification of each peptide derived from viral proteins as indicated in legend from d, e. In all panels, n = number of peptides.
include the most frequent virus strains circulating in South-East Asia and Australia and identified naturally occurring variants of our HLA-A24 binding peptides utilizing the "Identify short peptides in proteins" analysis from the Influenza research database (fludb.org) (n = 61-3877 sequences dependent on protein) to include in epitope mapping (Supplementary Table 4). We probed memory CD8 + T-cell populations, by firstly stimulating PBMCs from healthy Indigenous HLA-A24-expressing donors with 5 IAV peptide pools for 13 days. Each pool contained 7-13 peptides (Supplementary Table 2), of which each variant (Supplementary Table 4) was always included in the same pool as the wildtype peptide identified in the immunopeptidome studies. We observed CD8 + T-cell responses towards pools 1 and 2 via an IFN-γ/TNF ICS assay ( Fig. 4a and Supplementary Fig. 4), and subsequently, cell cultures from those pools were restimulated with individual peptides (+variants) to map the immunogenic epitopes.  [39][40][41][42][43][44][45][46][47] were poorly immunogenic in non-Indigenous donors who instead responded well to the PB2 549-557 epitope, absent in 4/5 Indigenous donors. Such differential epitope preference and immunodominance hierarchies between Indigenous and non-Indigenous donors is perhaps influenced by different HLA co-expression backgrounds or infection history.
The 9mer A/PB2 549-557 peptide adopted a canonical extended conformation within the cleft of HLA-A24, with anchor residues at P2-Tyr and P9-Trp, and a secondary anchor residue at P5-Ile. Solvent exposed residues were at P4-Trp, P6-Ile, P7-Arg, and P8-Asn, representing a large surface accessible for TCR interaction. The P9-Trp of the peptide formed a network of interactions with HLA-A24 tyrosine residues at positions 116, 118, and 123 as well as the Leu95 (Supplementary Fig. 3a-c), likely assisting with stabilizing the complex reflected in the higher stability observed for the HLA-A24-A/PB2 549-557 complex than with other peptides (Supplementary Table 5).
The B/PB2 550-558 peptide differed from the A/PB2 549-557 peptide at positions 5 (Ile to Val), 6 (Ile to Leu), 7 (Arg to Lys), and 9 (Trp to Leu) (Fig. 4f). Both peptides shared the same anchor residue at P2-Tyr and similar solvent-exposed residues (except for P7-Lys) but differed at PΩ (P9). As Leu possessed a shorter side chain than Tyr at PΩ, the IBV peptide was not buried as deeply into the F pocket, which may explain the lower stability observed for the B/PB2 550-558 peptide (T m of 57°C) compared to A/PB2 549-557 (T m of 62°C) (Supplementary Table 5). Structural overlay of HLA-A24 presenting the A/PB2 549-557 and B/ PB2 550-558 peptides showed that the antigen-binding cleft and both peptides adopted a similar conformation with an average root mean square deviation (r.m.s.d.) of 0.31 and 0.37 Å, respectively (for Cα atoms) (Fig. 4f), consistent with T-cell cross-reactivity observed towards these two peptides.
Although the 11mer A/PB2 549-559 generated similar responses to the 9mer A/PB2 549-557 in HHD-A24 mice (Fig. 3a, b), it was not immunogenic in peptide-pool screening in humans (Supplementary Table 2 Pool 4) (Fig. 4a) as perhaps the minimal 9mer epitope was not exposed for T-cell recognition, due to the two additional C-terminal residues (P10-Glu and P11-Thr) (Supplementary Fig. 6a, b). Similar to the 9mer peptide conformation P2-Tyr and P9-Trp of the 11mer PB2 549-559 act as primary anchor residues buried in the HLA-A24 antigen-binding cleft with the structural overlay of the peptides showing a r.m.s.d. of 0.48 Å (Supplementary Fig. 6a-c). Strikingly, the extra P10-Glu and P11-Thr residues of the 11mer extended outside the antigen-binding cleft, creating an unusual conformation that disturbed the interaction between the peptide and the HLA-A24 Lys146 at the C-terminal of the cleft. The Lys146 residue is a conserved residue in HLA molecules that helps stabilize the pHLA complex 50 . In the 9mer PB2 549-557 peptide, Lys146 interacts with the carboxylic group of the PΩ residue ( Supplementary Fig. 6d), however, this interaction is lost in the 11mer due to the presence of the extra two residues P10-Glu and P11-Thr ( Supplementary  Fig. 6e), thereby likely decreasing the pHLA stability compared to the shorter A/PB2 549-557 peptide (Supplementary Table 5). Thus, the bulged conformation of the extra residues in the A/PB2 549-559 may represent a challenge for TCRs interacting with the Cterminal end of the peptide. In compliance with the in vitro data, the structural data support the potential cross-reactivity of CD8 + T-cells between the A/PB2 549-557 and the B/PB2 550-558 , verifying our previous findings of broad CD8 + T-cell immunity against influenza virus infections.
Structural overlay of 9mer and 10mer B/NP peptides were different due to the 10mer's extra residue at the N-terminus (r.m. s.d. of 1.36 Å), which shifted the anchor residues ( Supplementary  Fig. 7b, c). The substitution of P2-Tyr (NP 164-173 ) for P2-Phe (NP 165-173 ) occurred without major structural rearrangement, as both residues were large aromatic residues filling the B pocket ( Supplementary Fig. 7f). However, the additional residue changed the secondary anchor residue at P3 from a small P3-Ser   Fig. 7g). The larger P3-Phe might stabilize the B pocket of the HLA-A24 better than the small P3-Ser, and therefore could explain the 7°C higher Tm observed for the NP 164-173 than the NP 165-173 in complex with HLA-A24 (Supplementary Table 5), which could also reflect the immunogenicity of this peptide. The largest structural difference between the two peptides was observed at the center of the peptide (P6/7-Arg) with a maximum displacement of 3.9 Å for the Cα atom ( Supplementary Fig. 7d). The P7-Arg of the NP 164-173 peptide sat higher than the backbone of the NP 165-173 peptide ( Supplementary Fig. 7c) and was a prominent feature for potential TCR interaction of the 10mer peptide, contrasting with the hydrophobic nature of the 9mer NP 165-173 peptide. Thus, the structures of HLA-A24 presenting the two NP peptides showed that, despite being overlapping peptides that differ only by one residue, the NP 164-173 and NP 165-173 peptides adopt different structural conformations. As a result, both peptides exposed different residues to the solvent, and hence would most likely be recognized by different TCRαβ repertoires.
Protective capacity of novel HLA-A24 IBV-epitope in HHD-A24 mice. To determine the protective capacity of the novel CD8 + T cell epitopes in HHD-A24 mice, we performed a proof of principle experiment and vaccinated mice with 3 immunogenic IBV peptides (NP 164 , NP 392 , NA 32 ) using a well-established prime/boost approach 24 , then infected mice i.n. with 1 × 10 3 pfu B/Malaysia (Fig. 5a). Vaccination with HLA-A24-restricted peptides resulted in significant protection against IBV. This was shown by decreased disease severity on d4, d5, and d6 after IBV infection as measured by the bodyweight loss ( Fig. 5b; p < 0.05, unpaired two-tailed t-test) as well as a significant~89% reduction in viral titers in the lung on d7 after IBV infection when compared to the mock-immunized group (p = 0.001, unpaired twotailed t-test) (Fig. 5c). Additionally, there was a significant decrease (p < 0.05, unpaired two-tailed t-test) in the levels of inflammatory cytokines (MIP-1β, MIP-1a, RANTES) in d7 BAL of peptide-vaccinated mice in comparison to the mockimmunized animals (Fig. 5d). Thus, CD8 + T cells directed at our novel HLA-A24-restricted IBV-specific epitopes provide a substantial level of protection against influenza disease, as they can markedly decrease body weight loss, accelerate viral clearance and reduce the cytokine storm at the site of infection.
Our findings demonstrate the presence of highly activated influenza-specific CD8 + T-cells against the published A/PB1 498 epitope and the IBV epitopes identified here in HLA-A24 + patients with acute influenza infection and memory pools across different human tissues, highly relevant to the Indigenous population.
Distinct HLA-A24 pMHC tetramer staining patterns reflect KIR3DL1 binding. It was apparent from the tetramerenrichment assays that some healthy donors contained large populations of HLA-A24-tetramer-binding CD8 + T-cells prior to enrichment (up to 6% of CD8 + T-cells) (Fig. 7a). This appeared to be donor-dependent but not entirely CD8 + T-cell specificitydependent. We found such oversized (0.32-6.73% in unenriched PBMCs) tetramer + CD8 + T-cell populations for A/PB1 498 in 10 out of 23 donors and in 14 out of 26 donors for B/NP 165 tetramers, but not for B/NA 32 (0/4 donors), which was further enriched with TAME (Fig. 7a, b). Ex vivo peptide stimulation and ICS performed with PBMCs from Non-LIFT 8 donor showed minimal IFNγ/TNF response ( Supplementary Fig. 10), despite the high frequency of tetramer-positive CD8 + T cells, indicating that these high frequency, low-affinity tetramer-binding ex vivo populations were not peptide-specific. Thus, it is important to note that our tetramer analyses in Fig. 5 excluded this oversized low-intensity staining tetramer-binding CD8 + T-cell population. Such oversized tetramer-binding CD8 + T-cell population could potentially be a unique HLA-A24-tetramer-binding phenomenon occurring in selected donors and hence potentially impair TCRspecific CD8 + T-cell binding. Therefore, we sought to better understand HLA-A24-tetramer-binding in donors with conventional and largely oversized HLA-A24-tetramer CD8 + T-cell populations.
Phenotypic analyses comparing tetramer-enriched fractions revealed that tetramer-binding CD8 + T-cells of donors with oversized populations were predominantly of the CD45RA + CD27 − effector (T EMRA ) phenotype (mean 73.3 and 71.7% for PB1 498 and NP 165 , respectively), while those from donors with conventional tetramer + CD8 + T-cells were predominantly T CM (mean 31.6 and 45.5%), T EM (10.3 and 8.3%) and T Naive (12.5 and 22.6%) in phenotype (Fig. 7c). To determine factors underlying this phenomenon, we performed scRNA-seq on single-cell-sorted TAME-enriched A/PB1 498-505 + CD8 + T-cell populations from two donors with oversized populations (non-LIFT 8 and 12) and two donors with conventional-size populations (non-LIFT 14 and 10) (Fig. 7d). Unsupervised hierarchical clustering analysis revealed three gene clusters (Fig. 7e). Highly expressed genes from Cluster 1 were associated with A/PB1 498 + CD8 + T Naive cells predominantly from donor 14. Most notably, A/PB1 498 + CD8 + T EMRA cells from donors 8 and 12 (oversized population) were grouped together and highly expressed genes from Cluster 2, characterized by high levels of T-cell effector genes NKG7, GNLY, CCL5, and granzymes B and H (GZMB and GZMH) but not K (GZMK), found in Cluster 3. In contrast, A/ PB1 498-505 + CD8 + T EMRA cells from donors 14 and 8 were grouped with highly expressed genes from Cluster 1 and 3, but not Cluster 2 except for CCL5 and IFITM1 genes, revealing  [32][33][34][35][36][37][38][39][40] were enriched in human blood and tissues. Representative FACS plots of enriched fractions are shown for healthy donors, acute influenza-infected patients, and tissues from deceased organ donors. b Tetramer + precursor frequencies in the blood of healthy LIFT and non-LIFT, and influenza-infected donors where acute and convalescent time-points are shown together (IBV-inf). Samples from donor IBV2-inf were collected at acute square and convalescent timepoint triangle. c Activation status of epitope-specific CD8 + T-cells in healthy, acute and convalescent donors. Statistical significance was calculated using a two-tailed Mann-Whitney test. d T-cell differentiation phenotype of epitope-specific cells in healthy LIFT and non-LIFT donors, and influenza-infected patients. e Tetramer + precursor frequencies of epitope-specific cells in human tissues. f CD103 and CD69 expression on epitope-specific cells in the lung compared to secondary lymphoid organs (SLO; CD8 + T cells 2 spleens, 1 lymph node, 3 tonsils; A/ PB1 498 -specific cells 2 spleens, 2 tonsils; B/NP 165 -specific cells 2 spleens, 1 lymph node, 2 tonsils). b, e Lines represent the median. c, d, f Bars represent mean + SD.
distinct characteristics of A/PB1 498 + CD8 + T EMRA cells within the two tetramer-binding populations.
NKG7 (natural killer cell granule protein 7) and GNLY (granulysin) are key CD8 + T-cell effector genes 53 located on the same immunoregion locus containing all the natural killer-receptor genes including the killer cell immunoglobulin-like receptors (KIR), within the leukocyte receptor complex (1 Mb,chromosome 19q13.4) 54 . Since NKG7 and GNLY were the top-hit genes associated with A/PB1 498 + CD8 + T EMRA cells from donors 8 and 12, we hypothesized that a KIR was interacting with the peptide/ HLA-A24 complex. KIR are expressed by a proportion of CD8 + Tcells 55 and KIR3DL1, in particular, has been previously shown to bind some but not all A24 pMHC tetramers 56 implying a degree of selectivity in the interaction. Staining for KIR3DL1 revealed its expression on CD27 − CD8 + T-cells, with the highest frequency of KIR3DL1 + cells detected in the T EMRA population in a donor that exhibited strong Pre-TAME tetramer-binding (Fig. 7f). Co-staining with the A/PB1 498-505 tetramer showed that all tetramer-binding CD8 + T-cells were positive for KIR3DL1, indicating that KIR3DL1 could potentially be binding to the tetramers (Fig. 7f). Blocking of KIR3DL1 prior to tetramer-staining markedly reduced the oversized population after TAME enrichment, to the levels of conventional tetramer + CD8 + T cell pools, revealing the true A24/PB1 498 -specific CD8 + T-cell population (Fig. 7g). Thus, much of the oversized population comprises tetramer-binding KIR3DL1 + CD8 + T-cells with other TCR specificities. Future studies are needed to understand whether KIR3DL1 binding of peptide-HLA-A24 complexes are competing with TCR interactions to mount robust peptide-HLA-A24-specific CD8 + T-cell responses, thus impacting influenza-specific immunity in Indigenous and non-Indigenous HLA-A24-expressing people at risk of severe influenza disease.

Discussion
Indigenous populations worldwide are highly affected by pandemic and, to a lesser degree, seasonal influenza disease. In line with previous studies 57-60 , our results show a high frequency of HLA-A24 allele expression in Indigenous populations in the Pacific region, an allele identified as an influenza mortalityassociated allomorph 27 . In our cohort of Indigenous Australians, 36% of individuals expressed at least one HLA-A24 allele. With little understood about the nature of the HLA-A24-restricted influenza-specific CD8 + T-cell response, there was a need to identify the breadth of influenza CD8 + T-cell epitopes for this atrisk population. Our analysis of previously published epitopes revealed a small number of HLA-A24-restricted IAV epitopes reported as immunogenic targets, while no IBV targets for HLA-A24 were known. Our in-depth mass-spectrometry-based immunopeptidomics approach defined the breadth of peptides presented by HLA-A*24:02 during IAV and IBV infection (Supplementary Tables 2 and 3) and provided important insights into the characteristics of the associated CD8 + T-cell responses that could predispose to more severe influenza disease. It should be noted that the peptidomes identified by our approach are characteristic of C1R cells (derived from EBV-transformed B cells 33 ), which constitutively express immunoproteasomes. Although the presence of reactive T cells confirms the relevance of identified peptides, there is a possibility that some immunogenic peptides may be missing from our analysis due to tissuespecific processing. Of the 52 peptides presented by HLA-A*24:02 during IAV infection, most mapped to PB2 (18 peptides, 35%) and PB1 (14 peptides, 27%) viral proteins, with no peptides originating from NA or M1. Consistent with this preference for PB2 and PB1 peptides, the CD8 + T-cell response in IAV-infected HHD-A24 mice focused mostly on four epitopes from PB1 (PB1 216-222 and PB1 498-505 ; 36% primary splenic response) and PB2 (PB2 549-557 and PB2 549-559 ; 31% primary splenic response). In HLA-A*24:02 + donors, memory responses to overlapping peptides PB1 496/498-505 were consistently observed, while interesting differences were seen in the hierarchy of other epitopespecific responses, with Indigenous donors responding to PA 649-658 and NP [39][40][41][42][43][44][45][46][47] , and non-Indigenous donors instead responding to the PB2 549-557 peptide. Such differential response characteristics, possibly related to HLA co-expression or infection history, are important considerations for the design of T-cell vaccines for high-risk Indigenous populations. Importantly, CD8 + T-cells specific for the dominant A/PB1 498-505 peptide were identified with an activated phenotype in the blood of patients with acute IAV infection and across different human tissues, including a population of T RM cells in the lung, providing evidence of their involvement in the influenza-specific response.
Hertz et al. previously showed that HLA-A24 has a low targeting efficiency for conserved regions of the pH1N1 virus, which was indicative of low cross-reactive memory responses that may have contributed to the impaired pH1N1 CD8 + T-cell immunity observed in HLA-A24 + individuals during the 2009 pandemic 27 . Our data reveal that HLA-A24 presents a breadth of peptide ligands (52 peptides) derived from 6/8 IAV proteins. Notably, the variable HA and NA viral glycoproteins play a minimal role in HLA-A24-restricted CD8 + T-cell immunity to IAV. Instead, the focus on epitopes from PB1 and PB2 that are well conserved across virus strains circulating in South-East Asia and Australia suggests that the prominent HLA-A24-restricted CD8 + T-cell responses are likely to confer broad cross-reactive immunity to IAV. This is of key importance as the current T-cell vaccines in clinical trials focus mainly on structural proteins like NP, M1, and M2, and would therefore not elicit cross-protective CD8 + T-cell responses in HLA-A24 + individuals at risk of severe influenza disease.
In contrast to IAV, the protein origins of IBV peptides presented by HLA-A24 differed greatly. From 48 IBV-derived presented peptides, the majority originated from NP (10, 21% of total), as well as HA (11, 23% of total) and NA (9, 19% of total with a total of 42% for surface glycoproteins). In terms of immunogenicity, our data from transgenic mice showed that the immunogenic HLA-A24-binding peptides were predominantly derived from NP (40% of response) and NA (40% of the response). More importantly, numbers of CD8 + T-cells directed towards our novel epitopes were preserved during the secondary IBV challenge, indicating optimal memory establishment and recall, which contrasted with the situation for the secondary IAV challenge. As in mice, HLA-A24-restricted influenza-specific CD8 + T-cell responses in Indigenous and non-Indigenous human donors were also targeted towards NP, with NP 165-173 Fig. 7 Comparison of conventional and KIR-binding CD8 + T-cells. a Staining pattern of donors with conventional-size tetramer staining and donors with oversized tetramer + CD8 + T cell populations for the A/PB1 498-505 and B/NP 165-173 tetramers pre-and post-enrichment. b Presence (+) or absence (−) of conventional (blue) and oversized tetramer + CD8 + T cell populations (red) populations across all samples tested. c Phenotypic analysis of tetramerenriched CD8 + T-cells from conventional-size and oversized populations for both A/PB1 498-505 and B/NP 165-173 tetramers (conventional A/PB1 498 -binding cells n = 7 biologically independent samples; pre-TAME A/PB1 498 -binding cells n = 6 biologically independent samples, conventional B/NP 165 -binding cells n = 9 biologically independent samples, pre-Tame B/NP 165 -binding cells n = 7 biologically independent samples). Bars indicate mean + SD. Statistical significance was calculated using a two-tailed Mann-Whitney test. d Tetramer-staining pattern of conventional and oversized-binding staining for scRNAseq pre-and post-enrichment. e Unsupervised clustering of mRNA expression of enriched tetramer-binding CD8 + T-cells (n = 30 cells per donor). All listed genes in the SC3 plot have a p-value < 0.05 and AUC > 0.65. p-values have been calculated using the two-tailed Wilcoxon signed-rank test without adjustments. f KIR3DL1 expression on different CD8 + T-cell phenotypes in non-LIFT 8 (top panel) and co-staining of A/PB1 498-505 tetramer and KIR3DL1 without enrichment. g Effects of KIR3DL1 blocking on conventional and pre-TAME-binding tetramer + CD8 + T-cells pre-and post-enrichment. and NP 164-173 being prominent CD8 + T-cell specificities alongside CD8 + T-cell epitopes derived from NA, HA, PB2, and PA ( Table 3). The breadth of the HLA-A24-restricted IBV response highlights the power of identifying epitopes with our massspectrometry-based immunopeptidomics approach and might explain, at least partially, why Indigenous populations have not been reported to be at risk from severe IBV disease. As for IAV, IBV epitope-specific CD8 + T-cells were activated during acute IBV infection in HLA-A24 + individuals and were found distributed across tissues including the lung in non-infected individuals.
Broadly cross-reactive CD8 + T-cell responses that provide universal immunity across multiple strains or subtypes of influenza viruses have a crucial role in protection from severe influenza disease 24 . Here we demonstrate cross-reactive responses between IBV lineages for the B/NP 165-173 peptide, as well as cross-reactive IAV/IBV responses between the A/PB2 549-557 peptide and IBV PB2 550-558 variants in HHD-A24 transgenic mice ( Supplementary Fig. 11) and humans. The A/PB2 549-557 peptide is conserved between H3N2 and H1N1 IAVs 61 , and shares 55% amino acid identity with the cross-reactive IBV PB2 550-558 variants. Structures of HLA-A24 with A/PB2 549-557 and B/PB2 550-558 showed that the antigen-binding cleft and both peptides adopted a similar conformation, providing a structural basis for T-cell cross-reactivity between these epitopes. Interestingly, IBV was more effective than IAV at expanding crossreactive CD8 + T-cells, suggesting that infection history may play a role in determining patterns of cross-reactivity and that selection of peptide sequences that promote the greatest crossreactivity is a consideration for universal influenza T-cell vaccines.
Structural analysis of the overlapping peptides A/PB2 549-557/559 and B/NP 164/165-173 showed that despite the difference of only two or one amino acids in length respectively, these peptides each adopted different conformations with HLA-A24 and are likely to induce distinct TCR repertoires. In the case of B/NP 164/165-173 , responses to both peptides are equivalently immunodominant in HLA-A24 + individuals, providing breadth to the overall CD8 + Tcell response. However, only the A/PB2 549-557 epitope showed immunogenicity in HLA-A24 + individuals, with the instability and bulged conformation of the longer A/PB2 549-559 epitope potentially proving challenging for TCR recognition. Such intricacies in epitope presentation and CD8 + T-cell recognition offer opportunities to either maximize or tailor responses through vaccination.
Our present study not only provides comprehensive data on generating CD8 + T-cell immunity against severe influenza disease in HLA-A24-expressing Indigenous and non-Indigenous people worldwide but also unravels three potential reasons why HLA-A24 has a role in disease susceptibility: (i) The antigenic origin of HLA-A24 IAV epitopes. The majority (62%) of IAV peptides presented by HLA-A24 are derived from PB1 and PB2, which is in stark contrast to previous studies in humans 52,[62][63][64] and mice 65 showing that immunodominant peptides for other HLAs are derived predominantly from NP, PA, or M1. This is problematic for the current vaccine candidates in clinical trials which do not have a PB1 or PB2 component 48,66,67 , and also raises the question of whether PB1 or PB2-specific CD8 + T-cell responses are equivalently robust and protective compared to immunodominant responses derived from NP, PA or M1 presented by other HLAs. (ii) Qualitative deficiencies in the HLA-A24 IAV-specific CD8 + T-cell response. In HHD-A24 mice, the magnitude and breadth of IAV-specific CD8 + T-cell responses were greatly reduced during secondary IAV (but not IBV) challenge compared to primary infection, implicating possible functional defects at memory establishment or recall levels. (iii) Non-epitope-specific binding of peptide/HLA-A24 complexes to KIR. This can possibly limit TCR recognition and thus TCR-specific activation of influenza-specific CD8 + T-cells and could also provide an explanation for increased susceptibility to other infectious diseases such as sepsis and tuberculosis observed in Indigenous populations 14,15 . The above observations provide a platform for further investigations to understand the mechanisms driving a greater risk of severe influenza disease in HLA-A24 + individuals.
Our findings provide important insights into the design of new T-cell-targeted vaccines and immunotherapy protocols to reduce influenza disease mortality and morbidity in Indigenous people globally. Current seasonal influenza vaccines are effective at inducing antibody responses to the currently circulating strains, however, they do not protect against newly emerging pandemic viruses 68,69 . It was shown previously that the inactivated influenza vaccines do not induce CD8 + T-cell immunity 70 . Given the potential that CD8 + T cells have to protect against pandemic influenza viruses 18,20 , vaccinations are clearly not harnessing the full power of the immune system, especially against pandemic viruses to which Indigenous populations are highly susceptible. Development of a vaccine that induces long-lasting broadly crossreactive CD8 + T-cell immunity would provide at least some level of protection against distinct influenza variants, even strains with pandemic potential. Such a vaccine would minimize influenzarelated deaths in global populations, especially high-risk groups, which includes HLA-A24-expressing Indigenous and non-Indigenous people. HHD-A24 transgenic mice are a valuable tool for identifying immunogenic epitopes following virus infection in an animal system in vivo. However, as there are limitations to using mono-chain HLA-transgenic mice, our study infers the immunogenic epitopes in the context of HLA-A24 allele in mice. These CD8 + T cells were, however, further screened and validated in human PBMCs across several peptides presented by all HLA molecules from a number of different HLA-A24 + donors. Thus, our comprehensive analysis of peptide presentation and immunogenicity defines the candidate IBV and IAV peptides needed for a CD8 + T-cell-targeting vaccine that is effective in HLA-A24 + individuals. Understanding how best to augment these key responses to confer stronger protective immunity is the next step.

Methods
Human blood and tissue samples. Human experimental work was conducted according to the Declaration of Helsinki Principles and according to the Australian National Health and Medical Research Council Code of Practice. All blood and tonsil donors provided written consent prior to study participation. Lung tissues, lymph node, and spleen samples were obtained from deceased organ donors after receiving written informed consent from next-of-kin. Lungs were sourced from the Alfred Hospital's Lung Tissue Biobank. Lymph node and spleen were provided by DonateLife Victoria. Buffy packs were sourced from the Australian Red Cross Lifeblood (West Melbourne, Australia  (ID 2012(ID -1928 and Tasmanian Health and Medical HREC (ID H0017479). Human PBMCs and cells from tissues were isolated and cryopreserved as previously described 70 . Indigenous donors (LIFT) were recruited as described before 26 .
HLA typing and analysis of human PBMCs. NGS HLA typing for HLA class I and class II on genomic DNA isolated from granulocytes were performed by the Victorian Transplant and Immunogenetics Service (West Melbourne, VIC, Australia). Co-expression Circos plots were generated using R V.4.0 (R Core Team, Vienna, Austria), RStudio: Integrated Development for R (RStudio, Inc., Boston, USA), and the cyclize package 71 .
Expansion of antigen-specific memory CD8 + T-cells from human PBMC. Cryopreserved PBMCs (3.3-5 × 10 6 ) from healthy non-LIFT and LIFT donors were used to expand antigen-specific CD8 + T-cells modified from Koutsakos et al. 24 . In brief, one-third of PBMCs were pulsed with a pool of up to 31 peptides (including circulating variants, Supplementary Table 4) at a total concentration of 10 µM at 37°C in RPMI. After 1 h, cells were washed twice with RPMI and mixed with the remaining autologous PBMCs. To expand virus-specific CD8 + T-cells, infected C1R.A24 cells were washed twice with serum-free RPMI to remove excess FCS and infected with A/HKx31 or B/Malaysia/2560/2004 at a MOI of 5 and incubated at 37°C. After 1 h, RF10 was added and cells were incubated for further 11 h at 37°C before cells were placed at 4°C for 14 h. Infected C1R.A24 cells were washed twice to remove any residual virus and added at a 1:10 ratio to PBMCs. For additional stimulation, virus-expanded PBMCs were restimulated by the addition of virusinfected C1R.A24 cells on day 8 at a 1:10 ratio. Cells were then incubated for a total of 10-15 days in RF10 media with 10 U mL −1 of recombinant human IL-2 (Roche Diagnostics, Mannheim, Germany) being added on day 4 and half-media changes every 1-2 days onwards.
Large-scale infection for immunopeptidome analysis. For large-scale infections, C1R or C1R.A24 were cultured to high density in RF10 media slowly rotating in 17 dm 2 filter-capped roller bottles (Corning) at 37°C, 5% CO 2 . Cells were harvested and infected with influenza A or B virus at a MOI of 5 in RPMI at a density of 1 × 10 7 cells mL −1 in 50 mL tubes for 1 h at 37°C with slow rotation. Infected cells were returned to roller bottles with the addition of 1:1 conditioned media:RF10 to a final density of 1.4 × 10 6 cells mL −1 and incubated a further 1-15 h (37°C, 5% CO 2 , slow rotation). HLA expression and infection efficacy were validated by surface staining LC-MS/MS analysis of HLA-bound peptides. Cell pellets of 0.7-1.3 × 10 9 C1R or C1R.A24 were lysed via cryogenic milling (Retsch Mixer Mill MM 400), resuspension in 0.5% IGEPAL CA-630, 50 mM Tris-HCl pH 8.0, 150 mM NaCl and protease inhibitors (cOmplete Protease Inhibitor Cocktail Tablet; Roche Molecular Biochemicals) and incubation at 4°C for 1 h with slow rotation. Lysates were cleared by ultracentrifugation and HLA isolated by immunoaffinity purification using protein-A-agarose-bound antibodies as described 35,73 . Antibodies were either w6/32 (pan class I) alone or sequential DT9 (HLA-C specific), w6/32 (pan class I), and mixed class II (equal amounts LB3.1, SPV-L3 and B721, capturing HLA-DR, -DQ and -DP, respectively).
Peptide/MHC complexes were dissociated and fractionated by reversed-phase high-performance liquid chromatography (RP-HPLC) as described 24,35,74 . 500 µL fractions were collected throughout the gradient, and the peptide containing fractions combined into 9 pools, vacuum-concentrated, and reconstituted in 15 µL 0.1% formic acid (Honeywell) in Optima™ LC-MS water. Reconstituted fraction pools were analyzed by LC-MS/MS using a SCIEX 5600+ TripleTOF mass spectrometer equipped with a Nanospray III ion source as previously described 74 .
LC-MS/MS data analysis. Spectra were searched against a proteome database consisting of the human proteome (UniProt/Swiss-Prot v2016_04), and either the A/X31 or the B/Malaysia/2506/2004 proteome plus a 6-reading frame translation of the viral genome, using ProteinPilot software (version 5.0, SCIEX), considering biological modifications and employing a decoy database to calculate the false discovery rate (FDR). Subsequent analyses were based on the best hypothesis for distinct peptides. Sequence motifs were generated utilizing peptides assigned at confidences greater than that required for a 5% FDR using Seq2logo2.0 75 (default settings). Likely HLA-A*24:02 binders were determined based on appearance across the experiments/antibodies and predicted binding (netMHCpan4.0 [39][40][41]. For peptides identified in their native form (and lacking Cys residues) that were synthesized for functional analysis, fragmentation patterns and retention times of representative spectra were compared to the synthetic and the quality of the match described (Supplementary Data 1).
HLA-A*24:02 HHD mouse studies. All mouse studies were overseen by the University of Melbourne Ethics Committee (#171408). HHD-A24 mice were generated by François Lemonnier as described previously 42 . These mice express a chimeric MHC-I that consists of the murine α3 and transmembrane domain and the human α1 and α2 domain covalently linked the human β2m. For mouse infections, 6-12 week-old mice were infected intranasally using a positive displacement pipette with 30 µL of either 100 pfu of A/X-31 or 200 pfu of B/Malaysia/ 2506/2004 in PBS under isoflurane anesthesia. For the secondary challenge, mice were infected 6-8 weeks after primary infection with 200 pfu of A/PR8 or 400 pfu B/Phuket/3073/2013. To identify immunogenic peptides, spleen and bronchioalveolar lavage (BAL) were isolated on day 10 or day 8 for secondary infection, respectively. Spleen single-cell suspensions were prepared and incubated for 1 h at 37°C in Affinipure Goat anti-mouse IgG + IgM (Jackson Immunoresearch)-coated panning plates to deplete B cell populations. BAL was combined from 3 to 5 mice to achieve sufficient T-cell numbers. Single-cell suspensions were then stimulated with peptide pools or single peptides at 1 µM in the presence of Brefeldin A (BD Golgi plug) for 5 h at 37°C in RF10 with 10 U mL −1 IL-2 followed by staining with panel 1 (Supplementary Table 6). Cells were analyzed using BD Fortessa and FlowJo v10 (BD). For immunization studies, mice were vaccinated with 30 nmol of NA [32][33][34][35][36][37][38][39][40] , NP 392-400 and NP 165-173 emulsified in complete (Prime) or incomplete (Boost) Freund's adjuvant. 50 μL vaccine was injected on both sides at the base of the tail. Control mice were injected emulsified adjuvant without peptides. Two weeks after prime, mice were boosted and challenged with 1 × 10 3 pfu B/ Malaysia/2506/2004 intranasally 7 days after boost. On days 6 and 7, after infection lungs were isolated to determine viral load with a plaque assay as described before 76 .
Detection of cytokines in BAL using a cytometric bead array. Cytokine concentrations in the BAL were detected using the BD Cytometric Bead Array Mouse enhanced sensitivity master buffer kit (Cat#: 562248) as per the manufacturer's instruction. Data were acquired using BD FACS Canto II and analyzed by FCAP Array software (Soft Flow Inc. Pecs, Hungary).
Protein expression, purification, and crystallization. Soluble HLA-A*24:02 heterodimers containing either A/PB2 549-557 , A/PB2 549-559 , B/PB2 549-557 , B/ NP 165-173 , or B/NP 164-173 peptide were prepared as follows. A truncated HLA-A*24:02 construct encompassing the extracellular part of the HLA molecule (residues 1-276), and human beta-microglobulin (β2 m) were expressed separately in a BL21-pLyS Escherichia coli strain as inclusion bodies. The inclusion bodies were subsequently extracted, washed, and resuspended into a solution containing 6 M guanidine. Each pHLA complex was then refolded into a cold refolding solution (100 mM Tris-HCl pH 8, 2 mM Na-EDTA, 400 mM L-arginine-HCl, 0.5 mM oxidized glutathione, 5 mM reduced glutathione) by adding 30 mg of HLA heavy chain, 20 mg of β2 m and 4 mg of peptide. The refolding solution was then dialyzed in 10 mM Tris-HCl pH 8, and the protein was purified by a succession of affinity column chromatography.
Thermal stability assay. Thermal shift assays were performed to determine the stability of each pHLA-A*24:02 complex using fluorescent dye Sypro orange to monitor protein unfolding. The thermal stability assay was performed in the Real Time Detection system (Corbett RotorGene 3000), originally designed for PCR. Each pHLA complex was in 10 mM Tris-HCl pH 8, 150 mM NaCl, at two concentrations (5 and 10 mM) in duplicate, was heated from 25 to 95°C with a heating rate of 1°C min −1 . The fluorescence intensity was measured with excitation at 530 nm and emission at 555 nm. The Tm, or thermal melt point, represents the temperature for which 50% of the protein is unfolded.
Tetramer-associated magnetic enrichment in humans. TAME was performed on PBMCs (7.5 × 10 6 -2.7 × 10 8 ) of healthy, IAV-or IBV-infected donors, as well as lymphocytes isolated from tonsils, lung, and pancreatic lymph nodes (panLN) to detect CD8 + T-cells specific for IAV and IBV. pMHC-I monomers were made inhouse 81 and conjugated at an 8:1 molar ratio to PE-or APC-labeled streptavidin (SA) to generate tetramers. Cells were FcR-blocked for 15 min on ice and stained with APC or PE-conjugated tetramers at a 1:100 dilution for 1 h at RT, washed twice then incubated with anti-PE and anti-APC MicroBeads (Milenty Biotec). Unenriched, flow-through, and enriched fractions were surface stained with panel 3 (PBMC) or 4 (SLO and lung) (Supplementary Table 6). After 30 min staining on ice, cells were fixed for 20 min in 1% PFA and acquired by flow cytometry. In some experiments, KIR3DL1 blocking was achieved by the addition of anti-human NKB1 antibody (clone DX9, Cat 555964, BD Pharmingen) at 1:50 during the FcRblocking step.
Single-cell mRNAseq. A/PB1 498-505 + CD8 + T-cells were single-cell sorted into 96 well plates containing lysis buffer (1 µL RNase inhibitor and 19 µL Triton X-100) after TAME on a BD Aria III sorter. Libraries were generated as described previously 24 . A Nextera XT DNA Library Prep Kit was used for the generation of sequencing libraries and sequencing performed on a NextSeq500 platform with 150-base par high-output paired-end chemistry for 30 tetramer + cells/donor (120 cells total). Bioinformatical analysis. The quality of scRNA-seq was assessed with FastQC. TopHat2 with default parameters was used to align sequences to the Ensembl GRCh38 reference genome. A total of 112 out of 120 analyzed cells passed quality control and was used for further analysis. Gene expression was quantified utilizing Cufflinks suit (v 2.2.1) where FPKM was assessed using CuffQuant and values normalized based on total mRNA content with CuffNorm. Clustering was performed utilizing SC3 82 . Each gene has been assessed for the accuracy of predicting the 3 clusters using its expression level. The accuracy of the prediction has been measured using the area under the receiver operating characteristic (ROC). Furthermore, a p-value has been calculated using the two-tailed Wilcoxon signed-rank test. Pheatmap in R was used to visualize Heatmaps.
Reporting summary. Further information on experimental design is available in the Nature Research Reporting Summary linked to this paper.