key: cord-0326583-6bjjds0u
authors: Nguyen, T. H. O.; Koutsakos, M.; van de Sandt, C. E.; Crawford, J. C.; Loh, L.; Sant, S.; Grzelak, L.; Allen, E. K.; Brahm, T.; Clemens, E. B.; Auladell, M.; Hensen, L.; Wang, Z.; Nussing, S.; Jia, X.; Gunther, P.; Wheatley, A. K.; Kent, S. J.; Aban, M.; Deng, Y.-M.; Laurie, K. L.; Hurt, A. C.; Gras, S.; Rossjohn, J.; Crowe, J.; Xu, J.; Jackson, D.; Brown, L. E.; La Gruta, N.; Chen, W.; Doherty, P. C.; Turner, S. J.; Kotsimbos, T. C.; Thomas, P. G.; Cheng, A. C.; Kedzierska, K.
title: In-concert immune dynamics during natural influenza virus infection and recovery in acute hospitalized patients
date: 2020-09-18
journal: nan
DOI: 10.1101/2020.09.17.20197012
sha: b26e97d937e64d52f721b91cd11f9bc52bfe98b9
doc_id: 326583
cord_uid: 6bjjds0u

We report in-concert dynamics of 18 key immune parameters, related to clinical, genetic and virological factors, in patients hospitalized with influenza across different severity levels. Influenza disease was associated with correlated increases in IL6/IL-8/MIP-1; cytokines and lower antibody responses. Robust activation of circulating T follicular helper cells (cTfhs) correlated with peak antibody-secreting cells (ASC) and influenza heamaglutinin-specific memory B-cell numbers, which phenotypically differed from vaccination-induced B-cell responses. Influenza-specific CD8+/CD4+ T-cells increased early in disease and remained activated during patient recovery. Here, we describe the broadest to-date immune cellular networks underlying recovery from influenza infection, highly relevant to other infectious diseases.

Thus, there is a clear deficiency in our understanding of the specific interplay between genetics, 110 innate and adaptive immune responses driving recovery from acute viral pneumonia. 111

Here, we utilized longitudinal samples obtained from patients hospitalized with acute 112 influenza disease to elucidate how innate and adaptive immune responses work in concert to 113 resolve influenza disease. We report, at a very high level of resolution, the overall breadth and 114 kinetics of 18 key immune parameters: the antiviral/inflammatory cytokines and chemokines, 115 hemagglutinin (HA)-directed antibodies, HA-probe-specific B cells, antibody-secreting cells 116

(ASCs), circulatory CD4 + T follicular helper (cTfh) cells, influenza peptide/MHC-specific CD8 + 117 and CD4 + T cells, IFN-γ-producing CD8 + T cells, CD4 + T cells, Natural Killer (NK) cells, 118

Mucosal-associated invariant T (MAIT) and γδ T cells, and granzymes A, B, K, M and perforin 119 expression in CD8 + , CD4 + , NK and MAIT cells. This comprehensive panel of immune 120

parameters, combined with host genetic factors and patient clinical data, enabled a detailed 121 dissection of human factors driving susceptibility, severity and recovery in patients hospitalized 122

with seasonal influenza viruses, highly relevant to other infectious diseases, especially newly-123 emerging respiratory infections. 124 125 126

Results 127

Patient DISI cohort 128 We recruited 64 patients admitted to the Alfred Hospital (Prahran, Australia) between 2014-129 2017 into our "Dissection of Influenza-Specific Immunity" (DISI) cohort. The inclusion criteria 130

for the DISI study included hospital admission of consenting adult patients with influenza-like 131

illness (ILI). The longitudinal study involved serial blood and nasal swab samples collected 132 from influenza-PCR-positive patients (Flu+, n=44) within 1-2 days of hospital admission to 133 discharge and follow-up blood samples ~30 days later, allowing analyses of the recovery phase 134 (Fig. 1a) . Patients who were influenza-PCR-negative (Flu-, n=20) were included as ILI-matched 135 controls. The DISI cohort included one patient #11 who became infected with H3N2 while in 136 hospital with AML, and died 34 days later from febrile neutropenia. We also had one death in 137 the control group, patient #28 who was PCR-negative at admission, but subsequently became 138 IBV-PCR-positive while in hospital and died 61 days later in ICU. The remaining cohort was 139 admitted to the Respiratory/General Ward with one case patient #62 requiring 1 day in ICU. 53 140 patients presented with ILI, while 6 cases and 1 control patient presented with pneumonia 141

( Supplementary Tables 1 and 2) . 142 We successfully recalled 80% (35/44) of Flu+ patients at a median of 41 days after 143 disease onset and 75% of our Flu-patients (15/20, median 39 days). Overall, Flu+ patients were 144 predominantly (55%) infected with the H3N2 influenza A virus (IAV) subtype ( Fig. 1b and  145 Supplementary Table 1) , followed by the co-circulating IAV 2009 pandemic H1N1 (pH1N1)-146 like strain (16%) and two co-circulating influenza B virus (IBV) strains (Phuket/3073/2013 147 (9%) and Brisbane/60/2008 (7%)), representing the dominant strains for each year in Australia. 148 IAV-infected patients (median 58 yrs, n=34) were significantly older than IBV-infected (45 yrs, 149 n=10) and Flu-patients (47 yrs, n=20) (Fig. 1c) . Apart from age, patient demographics in Flu+ 150

and Flu-groups were well-matched, with 86% and 70%, respectively, having one or more high-151 risk conditions for severe influenza disease. Both groups had a median of 4 days in hospital 152

( Supplementary Tables 1-2) . Time in hospital was significantly lower than that for more 153 severely-ill H7N9 cohort 18 with a median 14 days for recovery (n=12, p<0.0001) and a median 154 of 33 days for those who died (n=6, p=0.0239) (Fig. 1d) . The lower hospital burden reflects the 155 lower severity of seasonal influenza compared to avian A/H7N9 disease. 156 157

Risk HLA class-I and IFITM3-SNPs do not contribute to disease severity in hospital settings 158 To determine whether the risk HLA class-I and IFITM3-SNPs genetic factors contributed to 159 influenza disease severity in hospitalized patients, we compared HLA type and IFITM3 SNP 160 alleles between Flu+ and Flu-patients. HLA types (Supplementary Table 2 ) were grouped 161 according to "universal" or "risk" alleles based on studies describing HLA-A*02:01/03:01 and 162

HLA-B*08:01/18:01/27:05/57:01 molecules binding universally conserved influenza peptides 27 , 163 . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) preprint

The copyright holder for this this version posted September 18, 2020. . https://doi.org/10.1101/2020.09.17.20197012 doi: medRxiv preprint versus HLA-A*24:02/68:01 types positively correlating with pH1N1 mortality 15 and 164 morbidity 16 . Given that our cohort was mainly Caucasian (82% Flu+, 95% Flu-, 165

Supplementary Table 1), the frequencies of universal and risk alleles were comparable 166 between two groups (Fig. 1e) . Similarly, for IFITM3 SNP analyses, the non-risk rs12252-T/T  167  homozygous allele was observed in 91% of Flu+ patients (remaining were C/T heterozygous)  168 and 100% of Flu-patients, whereas the risk C/C homozygous allele was absent (Fig. 1f) . The  169  rs34481144 SNP showed a higher frequency of the A/G-heterozygous allele for both groups  170  (41% and 50%, respectively), with equal or similar proportions of the risk A/A and non-risk  171 G/G homozygous alleles (Fig. 1g) . Further analyses of Flu+ patients with the risk A/A 172 rs34481144 SNP versus Flu+ patients with A/G-heterozygous allele and non-risk G/G 173 homozygous alleles showed no differences in disease severity in hospital settings, as measured 174 by sequential (sepsis-related) organ failure assessment (SOFA) scores 28 (Fig. 1h) .

Highly correlated cytokines cluster together during influenza virus infection 177

Hypercytokinemia of early inflammatory mediators (IL-6, IL-8, IL-10, MCP-1, MIP-1α and 178

MIP-1β) have been associated with an increase in disease severity causing death in a number of 179 hospitalized cohorts 18-21 . Pro-inflammatory cytokines and chemokines measured in the blood of 180 acute representative Flu+ patients showed increased cytokine levels, particularly IL-6 and IL-8, 181

between mild and moderate patients as defined by lower (0-1) or higher (2-6) SOFA scores, 182 respectively, and in the more severe patient requiring ICU support (Fig. 1i) . Heightened IL-6 183 levels were also associated with the more severe 2013 H7N9-infected patients (Fig. 1i) with less significant correlations with other cytokines/chemokines (Fig. 1j) seasonal vaccine strain (Fig. 2a) . Conversely, fewer H1-HA and B-HA sequence variations were 203 observed between isolates from the infected patients compared to the WHO reference strains 204 (Supplementary Fig. 1) . Interestingly, despite the H3N2 clade-mismatches, all H3N2-infected 205 patients (except one A/unsubtyped patient) generated HA inhibition (HAI) antibody titres of 40 206 and above (equivalent to log 2 (HAI/10)>2) against the H3 vaccine strain (same year as infection) 207

at follow-up, compared to 54% at acute timepoints (Fig. 2b) , suggesting a boost in strain-208 specific antibodies that may have contributed to the significant increase in vaccine-strain titres 209 (p<0.0001). However, only marginal increases in antibody titres were observed in H1-and B-210 infected patients at follow-up (Fig. 2b) . Geometric mean titres (GMT) were low for all three H1, 211

H3 and B subtypes during acute infection, but their ability to mount antibody responses at 212 follow-up was generally comparable to vaccine-induced responses in a healthy cohort but with 213 greater GMT fold-changes (Fig. 2c,d) . As expected, antibody titres were low and remained 214 unchanged in the Flu-group (Fig. 2c,d) . 215 Previous exposures to past influenza strains, influenced by age, can affect our immune 216 responses to current influenza virus infections 29 . To study the breadth of antibody responses, 217 . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) preprint

The copyright holder for this this version posted September 18, 2020. . https://doi.org/10.1101/2020.09.17.20197012 doi: medRxiv preprint HAI antibody landscapes were generated against current and older influenza strains from the last 218 century that the patients may have previously encountered (Supplementary Table 3 ). Antibody 219 responses to past strains were minimal during acute infection (V1 and V2) (Fig. 2e,f) . However, 220

at follow-up, Flu+ patients induced robust antibody responses to older strains from the same 221 subtype that they were infected with, a phenomenon known as back-boosting 29 , but not against 222 the other subtypes (Fig. 2e,f) . Therefore, compared to the Flu-and healthy cohorts, Flu+ 223 patients started with lower antibody titres at hospitalization, perhaps due to lower vaccine 224

coverage, but were able to mount robust and broad antibody responses at convalescence. 225 226

Activation of cTfh, ASCs and influenza-specific memory B cells prior to recovery 227

Tfh cells are essential for generating high-affinity memory B cells in germinal centers (GCs) 228

and are located in secondary lymphoid organs 30 . cTfh cells share phenotypic and functional 229

properties to GC Tfh cells, and we and others showed that inactivated influenza vaccination 230 (IIV) induces expansion and PD-1/ICOS-activation of CD4 + CXCR5 + CXCR3 + cTfh type-1 231 (cTfh1) cells, which correlated with antibody and CD19 + CD27 ++ CD38 ++ ASC responses 4,31 . 232

Here, we show for the first time that activated PD-1 + ICOS + cTfh1 cells emerged within 233

influenza-infected patients in parallel with ASC responses, both peaking between days 7-10 234 after disease onset in both Flu+ and Flu-groups, prior to patients' recovery. cTfh1 cells peaked 235 higher in Flu+ patients compared to Flu-groups, as shown by the separation of the 95% 236 confidence intervals ( Fig. 3a-c, Supplementary Fig. 2a ). The number of ASCs was 237 significantly higher (~2-8 fold) during acute infection compared to follow-up for both Flu+ and 238

Flu-groups (Fig. 3d) . Overall, activated cTfh1 cells were also trending higher at acute 239 compared to follow-up (~2 fold, p=0.0519, Mann-Whitney test, Fig. 3e ). Specifically, numbers 240 of activated cTfh1 cells in Flu+ patients were significantly higher than the cTfh2 and cTfh17 241 subsets, at both acute and follow-up timepoints (Fig. 3e) . Further, cTfh1 responses strongly 242 correlated with ASC responses during acute influenza virus infection (Fig. 3f) , but less so for 243 cTfh2 and cTfh17 subsets (Supplementary Fig. 3) . Acute ASC responses also correlated with 244 high antibody responses (log 2 (HAI/10)>2 or HAI>40, Fig. 3g) . However, acute cTfh responses 245 did not significantly correlate with antibody responses, perhaps due to the low titres observed in 246 some patients during acute infection, in contrast to a vaccination response where IIV-induced 247 cTfh1 responses correlated with the fold-change in antibody titres (d28 over d0 baseline) 4 . 248

To evaluate influenza-specific memory B cell responses during natural influenza virus 249 infection in comparison to vaccination, recombinant HA probes relevant to the 2014-2017 250 vaccine strains were used to detect class-switched IgD -HA-specific B cells in Flu+ patients 251 (Fig. 3h, Supplementary Fig. 2b) . These rHA probes have been extensively validated for their 252 specificity 4,32,33 . Numbers of rHA + IgD -B cells in total and per probe did not significantly 253 change between acute and follow-up timepoints (Fig. 3i) , and were comparable to the 254 magnitudes observed following IIV 4 . Phenotype and isotype distributions of the total B cells 255 remained stable between acute and follow-up timepoints (Fig. 3j,k) . Specifically, rHA + IgD -B 256 cells for H3 and B probes were more CD21 lo CD27 hi activated memory in proportion at the acute 257 timepoint compared to follow-up (H3: p=0.0007, B: p=0.0017), before becoming more 258 CD21 hi CD27 hi resting memory at follow-up compared to acute timepoints (H3: p=0.0004, B: 259 p<0.0001) (Fig. 3j,l top panels) . Conversely, in a healthy vaccinated cohort, the resting memory 260 phenotype was most prominent at baseline, before becoming more activated memory from days 261 7-28 following IIV (Fig. 3m, top panel) . The isotype distribution in Flu+ patients was enriched 262

for IgG -IgMcells (mostly IgA + ) for H3, and IgA + cells for the B-probe at acute timepoints, 263 before becoming largely IgG + at follow-up for each probe (Fig. 3j ,l bottom panels). In contrast, 264 the isotype distribution of rHA + IgD -B cells did not change in healthy donors following IIV 265 (Fig. 3m, bottom panel) . Thus, our analyses show striking differences after influenza virus 266 infection compared to vaccination of recruited rHA + IgD -B cells at both phenotypic and isotype 267 levels. 268

Although patients' HA-specific B cell responses did not positively correlate with their 269 acute antibody responses which were generally low, there was a significant positive correlation 270 between activated or resting memory HA-specific B cells with their acute ASC response, as well 271

as resting memory HA-specific B cells with the acute cTfh responses (Supplementary Fig. 4) . 272

. CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) preprint

The copyright holder for this this version posted September 18, 2020. . https://doi.org/10.1101/2020.09.17.20197012 doi: medRxiv preprint Taken together, we show prominent activation of cTfh1 cells during acute influenza virus 273 infection, at the time of the ASC and influenza-specific memory B cell responses. 274 275

Our previous work showed that rapid recovery from severe A/H7N9 was associated with early 277 IFN-γ-producing CD8 + T cell responses, while late recovery involved a network of humoral 278 (Abs) and cellular (CD8 + , CD4 + , NK) responses 12 . Here, Flu+ patients' PBMCs were incubated 279 with live seasonal H1N1, H3N2, B/YAM and B/VIC viruses to measure influenza-specific 280 innate (NK, γδ, CD161 + TRAV1-2 + or MAIT cells) and adaptive (CD4 + and CD8 + T cells) 281

immune responses elicited via IFN-γ production after 18 hrs ( Fig. 4a,b, Supplementary Fig.  282 5a). Kinetics of influenza-specific IFN-γ-producing cells across days after disease onset showed 283 low numbers of IFN-γ-producing populations across all cell subsets at patients' admission. 284

These however increased over the course of infection, before stabilizing at convalescence (Fig.  285  4c) . Despite lymphopenia, a fixed number of cells were added to the assay, and therefore the 286 ability of adaptive CD8 + and CD4 + T cells to generate IFN-γ-responses, as a frequency per 287

subset, continued to increase during acute infection until ~15 days after disease onset (Fig. 4d) . 288 Conversely, for innate cells, their ability to produce IFN-γ responses did not change, further 289

suggesting that adaptive T cells can increasingly respond to the virus during acute infection and 290

are the main drivers of early recovery. 291 292

To assess the cytolytic potential within the Flu+ patients' cell subsets, we measured cytotoxic 294 molecules: granzymes (A, B, K and M) and perforin ( Supplementary Fig. 6a) . Overall, MAIT 295 cells had the highest frequency of total cytotoxic molecules expressed, followed by NK cells and 296

CD8 + T cells, with minimal production in CD4 + T cells (Fig. 4e) 

To dissect epitope-specific CD8 + and CD4 + T cells during acute IAV infection, we performed 314 patient-specific tetramer staining, utilizing an extensive range of peptide/MHC class-I and class-315 II tetramers covering the most frequent HLA alleles, with an estimated population coverage of 316 63-100% across all ethnicities (Fig. 5a, Supplementary Table 5 ). Following tetramer-317 associated magnetic enrichment (TAME), robust tetramer + CD8 + T cell populations were 318 detected for all donors (n=19) and specificities tested directly ex vivo (Fig. 5b ,c, 319 Supplementary Fig. 6b,c) , even when using 10-20% less cell numbers generally required for 320

successful detection of antigen-specific CD8 + T cells in healthy donors [34] [35] [36] . In 17 patients, 321 tetramer + CD8 + T cell populations were detected at all the timepoints, while two patients had 322 tetramer + CD8 + T cells directed towards the subdominant B7-NP epitope detectable only at 323 follow-up. For comparison, A2-M1-tetramer + CD8 + T cells were also measured in Flu-patients 324 (n=4/4 detected) (Fig. 5d) . Apart from CD8 + T cells, we also identified influenza-specific CD4 + 325 T cells in 5/5 H3N2-infected patients with either HLA-DR*01:01-, DR*04:01-or DR*11:01-326 . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) preprint

The copyright holder for this this version posted September 18, 2020. . https://doi.org/10.1101/2020.09.17.20197012 doi: medRxiv preprint HA-tetramers (Fig. 5e,f) , despite ~10-fold lower numbers of antigen-specific CD4 + T cells ex 327

vivo. DR*15:02-HA tetramers were also tested but yielded no specific cells in 2/2 patients. 328

Overall, tetramer precursor frequencies (where 10 -3 denotes 1 per 1,000 cells or 1e-3), 329

were stable across timepoints, with no significant differences observed between acute and 330 follow-up (Fig. 5f) . Pooled tetramer + CD8 + T cell precursor frequencies in Flu+ patients 331 (mean±SD, range: 2.62e-4±6.37e-4, 8.07e-7 to 3.56e-3) and Flu-patients (7.00e-5±8.11e-5, 332

8.38e-6 to 2.65e-4) were significantly higher 79-fold (p<0.0001) and 21-fold (p=0.0026), 333 respectively, than pooled tetramer + CD4 + T cells (3.30e-6±3.20e-6, 6.19e-7 to 1.17e-5) 334

( Supplementary Fig. 6d) . Frequencies of pooled tetramer + CD8 + T cells were similar between 335

Flu+ and Flu-patients and fell within the ranges we described previously for HLA-A2 + H7N9-336

Flu+ patients 37,38 . 337

Although the frequency of tetramer + CD8 + T cells were similar between acute and 338

follow-up timepoints in Flu+ patients, and Flu+ versus Flu-patients, the activation phenotypes 339 markedly differed. Based on four activation markers expressed (CD38, HLA-DR, PD-1 and 340 CD71), higher levels of activation were detected at the acute timepoint compared to follow-up 341 (Fig. 5g, Supplementary Fig. 6e) . Notably, in Flu+ patients, expression of two or more 342 activation markers was most evident between days 6-10 of disease onset for A2-M1 + CD8 + T 343 cells, which was in stark contrast to the mainly single-PD-1 expressing or non-activated 344 phenotypes exhibited among the Flu+ follow-up samples, Flu-patients (Fig. 5h) , and the parent 345 CD8 + T cell populations. High levels of activation at acute compared to follow-up samples were 346

clearly observed across all tetramer + CD8 + T cell specificities and, despite limited numbers, in 347 tetramer + CD4 + T cells when compared to parental CD8 + and CD4 + T cell populations (Fig. 5i , 348 Supplementary Fig. 7) . 349 Similarly, analysis of CD27/CD45RA/CD95 phenotypes showed significant differences 350 in the activation of tetramer + CD8 + and CD4 + T cells in Flu+ patients when compared to the 351 parental CD8 + and CD4 + T cell populations at acute and follow-up timepoints. In general, 352

among Flu+ patients, tetramer + CD8 + T cells consisted of central memory-like (Tcm,  353 CD27 + CD45RA -) and, to a lesser extent, effector memory-like (Tem, CD27 -CD45RA -) 354

phenotypes, while tetramer + CD4 + T cells were predominantly of Tcm-like cells. (Fig. 5j,  355 Supplementary Fig. 6f) . These proportions remained stable across timepoints but were 356 significantly enriched in Tcm compared to the overall CD4 + and CD8 + T cell populations, 357

respectively. 358

Overall, we show an extensive breadth of highly activated, non-cultured, ex vivo 359 influenza-specific memory CD4 + and CD8 + T cells detected during acute IAV infection, which 360

were still present in a less activated state following recovery. 361 362

Immune correlates of influenza severity 363 We also probed clinical, genetic and immune correlates of influenza severity and patients' 364 recovery. We found no significant associations between influenza disease severity (SOFA 365 scores 0-6, clinical presentation: ILI and/or pneumonia), clinical parameters (age, sex, influenza 366 strain, days of disease onset, days in hospital, risk factors and vaccination status) and genetic 367 host factors (IFITM3 rs34481144 SNP alleles). This might be as the majority of hospitalized 368 DISI patients were already in high-risk groups based on their co-morbidities/risk factors (86% 369

Flu+, 70% Flu-), including chronic respiratory disease (61% Flu+, 40% Flu-). Furthermore, 370

patients infected with seasonal influenza viruses experienced less severe disease, with short 371 hospital stays, no requirements for mechanical ventilation, and only one patient requiring ICU 372

for one day, in contrast to the more severe avian H7N9-infected patients 18 . 373

To investigate any potential links between the overall breadth of immune responses in 374 human influenza disease, including cytokines, antibodies, ASCs, activated cTfh1s, IFN-γ-375

producing innate (NK, γδT, MAIT cells) and adaptive (CD4 + and CD8 + T cells) immune cells, 376 and key clinical/genetic parameters and SOFA scores as a measure for disease severity, we 377 generated heat maps with all of the above mentioned parameters. Unsupervised heat maps of 378 acute and convalescent timepoints were generated for both Flu+ (Fig. 6a,b, Supplementary  379  Fig. 8a,b) and Flu-groups, as well as the entire cohort (Supplementary Fig. 8c-e) . For Flu+ at 380 . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) preprint

The copyright holder for this this version posted September 18, 2020. . https://doi.org/10.1101/2020.09.17.20197012 doi: medRxiv preprint the acute timepoints, regions of relatively lower cytokine levels of IL-6, MIP-1α, IL-8, MIP-1β, 381

MCP-1, were linked to higher levels of functional IFN-γ-producing immune cells, ASCs and 382 activated cTfh1s (Fig. 6a) . As shown in Fig. 1j , these cytokines were strongly positively 383 correlated to each other. At convalescence, lower cytokine regions were linked to higher IFN-γ-384 producing immune cells, but not ASCs or activated cTfh1 cells, which peaked during acute 385 infection (Fig. 6b) . In both acute and convalescent phases, reciprocal regions of higher 386

cytokines levels and lower IFN-γ-producing immune cells (and lower ASCs and activated cTfh1 387 cells for acute) were observed. However, these observed regions were not defined specifically 388 by SOFA scores or other clinical/genetic parameters. 389

Based on the functional implications of IFN-γ-producing cells detected here and our 390 previous H7N9 study 12 , we analyzed specifically the numbers of IFN-γ-producing cells across 391 acute (V1) and convalescent timepoints (F up ) as a function of patients' disease severity via 392 binned SOFA scores due to unequal numbers of SOFA scores across the cohort, skewing 393 towards low (0-1) SOFA scores, classifying the least severe. At the earliest acute timepoint 394 (V1), decreases in innate IFN-γ-producing γδT and MAIT cells were significantly associated 395

with more severe patients with higher SOFA scores (2-6) (Fig. 6c) , but not for NK cells, CD4 + 396 and CD8 + T cells. In contrast, at convalescence, lower adaptive IFN-γ-producing CD4 + and 397

CD8 + T cell responses were associated with higher severity (Fig. 6c) . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this this version posted September 18, 2020. . https://doi.org/10.1101/2020.09.17.20197012 doi: medRxiv preprint

Our cytokine analyses showed that inflammatory cytokines IL-6, IL-8, MIP-1α and 434 MIP-1β were strongly positively correlated with each other, after adjusting for age and the 435 sampling days from disease onset, and supports previous studies on their collective role as 436 biomarkers for influenza severity 18-21 . These cytokine correlations were still robust following 437 recovery at the convalescenct phase compared to other cytokines that were only correlated with 438 each other during acute infection. Furthermore, lower expression of IL-6, IL-8, MIP-1α and 439

MIP-1β were associated with higher IFN-γ-producing immune cells following influenza virus 440 infection assay, as well as ASCs and activated cTfh1 cells, and vice versa. 441

For antibody responses, HAI antibody titres were lower in Flu+ patients during acute 442 infection compared to Flu-patients and healthy controls, but increased significantly following 443 recovery against the infected strain and previous strains from the same subtype. It is unclear 444

whether the SHIVERS study compared antibody responses between acute versus convalescent 445 phase within mild non-hospitalized and severe hospitalized patient groups, although they did not 446 observe any differences when comparing antibody responses of mild versus severe groups at 447 both acute and convalescent timepoints 26 . Given that we observed back-boosting of cross-448 reactive antibody responses post-infection at the follow-up timepoint, we postulate that having a 449 natural influenza virus infection, compared to a vaccine-induced response, elicits more effective 450 cross-reactive antibody responses post viral clearance warranting further investigations. 451

Furthermore, acute GMT antibody titres were all below 40 and lower than the Flu-group 452

suggesting that these antibody levels were not protective prior to infection. These findings 453 provide a strong argument for influenza vaccination since the Flu+ group had lower vaccine 454 coverage (48%) than the Flu-group (65%) and supports WHO's guideline of 40 being the 50% 455 protective cut-off which has been controversial. Influenza vaccination is of utmost importance in 456 this cohort given that 86% of Flu+ patients had a significant risk factor for severe influenza 457 disease and 71% of those having a chronic respiratory disease. 458 We have recently performed a comprehensive study to measure influenza-specific B and 459

T influenza-specific B cells were of an activated-memory phenotype comprising IgG and IgA 473 isotypes but were predominantly a resting-memory IgG + population at convalescence. 474

Conversely, IIV induced an expansion of activated-memory influenza-specific B cells but there 475 were no changes in the isotype distribution which were predominantly IgG + B cells, given that 476 the influenza vaccine is not delivered to a mucosal site. IgA + memory B cells are able to localize 477

in the blood and mucosal sites of inflammation such as the respiratory tract during influenza 478 virus infection to provide a rapid immune response, while IgG + memory B cells generally 479 circulate throughout the body 45 . Therefore, in accordance with our earlier antibody findings, we 480

suggest that future B cell-based influenza vaccines promoting cross-reactive B cell responses 481 encompassing both activated-memory IgA + and IgG + subclasses may provide further protection 482 at the site of infection. 483

Our study utilized a whole live virus assay to measure the kinetics of influenza-specific 484 responses in NK cells, γδ T cells and CD161 + TRAV1-2 + MAIT cells (innate-like), and CD8 + 485 and CD4 + T cells (adaptive). We observed robust and increasing proportions of adaptive CD8 + 486

and CD4 + T cell responses within the first 2 weeks of infection, despite the patients' overall 487 . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this this version posted September 18, 2020. . https://doi.org/10.1101/2020.09.17.20197012 doi: medRxiv preprint lymphphopenic state, illustrating the ability of these cell types to drive patient recovery. In 488 contrast, innate lymphocytes did not increase in their ability to respond to influenza virus 489 infection by cytokine production, and lower IFN-γ-producing γδ T cells and MAIT cells were 490 significantly associated with higher disease severity based on SOFA scores at the earliest V1 491 timepoint. Similarly, we previously showed the robustness of the adaptive CD8 + and CD4 + T 492 cell immune response in driving recovery from severe H7N9 disease 12 . These findings have  493  important implications for future influenza vaccines because, apart from the CD4 + T cell help  494  compartment, IIV fails to induce CD8 + T cell responses nor any innate responses from NK cells,  495 γδ T cells and MAIT cells, as we previously reported 4 . 496 We probed influenza-specific T cells ex vivo using a range of peptide/MHC class I and 497

class II tetramers covering the most frequent HLA alleles with an estimated population coverage 498 of 63-100% across all ethnicities. Tetramer + CD8 + and CD4 + T cells were highly activated at 499 acute timepoints, compared to a less activated profile at follow-up, and mainly expressed 500 combinations of PD-1 with CD38 and CD71, and HLA-DR to a lesser extent. In contrast to 501 severe H7N9 influenza 13 , Ebola 46 and COVID-19 disease 39 , with prominent CD38 + HLA-DR + 502 populations in the total CD8/CD4 compartments, we did not observe marked populations of 503 CD38 + HLA-DR + T cells in either tetramer + or parent populations, perhaps reflecting a less 504

severe cohort of diseased patients in our DISI study. Nonetheless, we provide acute and 505 convalescent populations of rare HA 306-318 -specific CD4 + T cells in H3N2-infected patients that 506

were HLA-DR*01:01 + (n=3), HLA-DR*04:01 + (n=1) or HLA-DR*11:01 + (n=1), which 507 complements other studies assessing cells in HLA-DR*04:01 + healthy donors and rheumatoid 508 arthritis patients following influenza vaccination 47,48 . Therefore, our results represent the 509 broadest array of influenza-specific tetramer + T cell responses during acute influenza infection, 510

providing essential tools for future T cell-immune monitoring of newly-emerging influenza 511

viruses. 512

Tracking antibody, B and T cell responses (and other immune mediators) may be 513 predictive of patients' severity or recovery from influenza virus infections that require 514 hospitalization. Our current analyses of immunological and virological parameters within a 515 clinical hospital framework provide a unique and key dataset on the mechanisms underpinning 516 influenza severity and susceptibility in the human population. Furthermore, these efforts 517 represent a substantial progression to elucidate the host immune responses underlying the 518 recovery from this acute disease as well as other infectious diseases such as COVID-19 39 . 519 520 521

Online Methods 522

Study participants and design 523

The Dissection of Influenza-Specific Immunity (DISI) cohort enrolled consenting adult patients 524 admitted to The Alfred hospital during the 2014-2017 peak influenza seasons with influenza-525 like illness. Influenza-positive patients were PCR-confirmed (n=44), while ILI influenza-526 negative patients with other respiratory diseases were treated as negative controls (n=20). 527

Bloods were collected within 24-72 hours of hospital admission (Visit 1, V1), every 2-5 days 528 until discharge (V2, V3, V4 etc), then followed up approximately 30 days later. Nasal swabs 529

were collected at V1 and V2 timepoints while in hospital. Clinical data collection included 530 vaccination status, sequential (sepsis-related) organ failure assessment (SOFA) score 28 as a 531 measure of disease severity, and any significant risk factors (Supplementary Tables 1 and 2) . 532

The H7N9-infected hospitalized patient cohort (n=18) has previously been described 18 . 533

Healthy adults vaccinated in 2015 (TIV, n=16) and 2016 (QIV, n=26) have previously been 534 described in detail 4 , where blood samples were collected prior to vaccination (day -1 or 0) and 535 on days 7, 14 and 28 following vaccination. Healthy blood donors were recruited from The 536

University of Melbourne and Deepdene Surgery. Buffy packs were sourced from Australian Red 537

Cross Lifeblood, respectively (Supplementary Table 4 ). 538

Human experimental work was conducted according to the Declaration of Helsinki 539

Principles and according to the Australian National Health and Medical Research Council Code 540 of Practice. All participants provided written informed consent prior to the study. The study was 541 . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) preprint

The copyright holder for this this version posted September 18, 2020. . https://doi.org/10.1101/2020.09.17.20197012 doi: medRxiv preprint approved by the Alfred Hospital (ID #280/14) and University of Melbourne (ID #1442952.1 and 542 #1443389.4) Human Research Ethics Committees. Up to 40 mls of DISI patient blood was 543 collected in sodium heparin tubes (including 1 serum tube) and processed within 24 hours. Up to 544 60 mls was collected from healthy donors and ~50-60 ml for buffy packs. PBMCs were isolated 545 by Ficoll-Paque (GE Healthcare, Uppsala, Sweden) density-gradient centrifugation and 546 cryopreserved. DNA was extracted from the granulocyte layer using a QIAamp DNA Mini Kit 547 (Qiagen, Hilden, Germany) and sent to the Victorian Transplantation and Immunogenetics 548

Service (Australian Red Cross Lifeblood, West Melbourne, Victoria, Australia) for HLA class I 549 and class II molecular typing using the Luminex platform and microsphere technology (One 550

Lambda, Canoga Park, CA, USA), with LABType SSO HLA typing kits (One Lambda). Nasal 551 swab samples were kept cold during transport and stored at -80°C within 4 hours. 552

In cases where fewer than the total number of donors were analyzed (Supplementary  553  Table 6 ), DISI patients were selected for influenza-specific cellular assays based on their 554

influenza status and remaining sample availability, such as the class I and II TAME experiments 555 which were carried out on IAV + patients and relied on HLA type and tetramer availability. HIV-556 positive patients (n=5, Supplementary Table 2) were not included in such experiments to fulfil 557 biocontainment regulations. No other blinding or randomization protocols were applied and no 558 outliers (i.e. the two death patients) were excluded. 559 560 IFITM3 SNP analysis 561

Amplification of exon 1 rs12252 region was performed by PCR on genomic DNA using forward 562 (5′-GGAAACTGTTGAGAAACCGAA-3′) and reverse (5′-CATACGCACCTTCACGGAGT-563 3′) primers, as previously described 49 . Amplification of the rs34481144 promotor region was 564 performed using forward (5′-GGAAACTGTTGAGAAACCGAA-3′) and reverse (5′-565 CATACGCACCTTCACGGAGT-3′) primers, as described 17 . 566 567

Cytokine analysis 568 Patient's sera or plasma was diluted 1:4 for performing cytokine bead assay (CBA) using the 569

Human Tetramer-associated magnetic enrichment (TAME) 618 TAME was performed on thawed PBMCs (3-38e6) for the detection of influenza-specific CD4 + 619 and CD8 + T cells, as described 36,37 . Peptide/MHC class I and class II monomers (Fig. 5a) were 620 generated in-house 50,51 , before 8:1 molar ratio conjugation with either PE-streptavidin (SA) or Mann-Whitney (unpaired) and Wilcoxin (paired) tests were two-tailed. Friedman (matched) and 643

Kruskal-Wallis (unmatched) tests were used to compare more than two groups. Tukey's 644 multiple comparison test compared row means between more than two groups. Partial 645 correlation plots (Fig. 1j) showing significant (FDR-adjusted p-value <0.05) coloured squares 646

were partialed to account for the variance caused by log 10 (age) and the sampling days after 647 disease onset. To build antibody landscapes (Fig. 2f) , local regression (LOESS) was applied in 648 R 3.5.3 52 . LOESS with 95% confidence intervals was also used to plot Fig. 3c, 4c and 4d . Linear 649 . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) preprint

The copyright holder for this this version posted September 18, 2020. . https://doi.org/10.1101/2020.09.17.20197012 doi: medRxiv preprint relationships between day of sample and percent of IFN-γ producing cells were assessed with 650 ordinary least squares (lm function) in R with 95% confidence intervals shaded in grey and 651 correlation coefficients (r 2 ) reported in Fig 4d. Correlations (Fig. 3f,g, Supplementary Fig. 4 ) 652

were assessed using Spearman's correlation coefficient (r s ). Polyfunctionality pie charts (Fig.  653 4g) were generated using Pestle v1.8 and Spice v5 software 53 and p-values were calculated using 654

Permutation Test. P-values lower than 0.05 were considered statistically significant and exact p-655 values are shown in the figure. All statistical tests are indicated in the figure legends. 656 657

Data availability 658

The data that support the findings of this study are available from the corresponding author upon 659 request. 660 661 662 663

. CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) preprint

The copyright holder for this this version posted September 18, 2020. Fig. 5b) .

MAIT cells were defined as CD161 + TRAV1-2 + and were validated by the MR1-5'OP-RU-899 tetramer in 51% of samples (Supplementary Fig. 5c,d) . c,d, Numbers (n=22) and frequencies 900 (n=30) of influenza-specific IFN-γ-production in patients' immune cell subsets as days of 901 disease onset where 95% confidence intervals are shaded in grey. e-g, Patient data from PBMCs 902 left over from flow through fraction following TAME. *NK cells were defined by live/CD14 -903 /CD19 -/CD3cells. . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) preprint

The copyright holder for this this version posted September 18, 2020. . https://doi.org/10.1101/2020.09.17.20197012 doi: medRxiv preprint Concatenated FACS plots of TAME-enriched class II-tetramer + cells gated on CD4 + T cells 919 from H3N2-infected patients. f, Precursor frequencies of tetramer + cells from Flu+ and Flu-920 patients at acute (V1, V2, V3 or V4) and follow-up timepoints. g, Representative overlay FACS 921 plots of activation markers expressed on TAME-enriched tetramer + cells compared to their 922 unenriched parent population. h, Frequency of A2-M1 + CD8 + T cells from individual Flu+ and  923  Flu-patients expressing different combinations of activation markers PD-1, CD38, HLA-DR  924 and CD71, where CD71 was replaced by Ki-67 in the staining panel for Flu-patients. i, Overall 925 activation status of TAME-enriched tetramer + cells compared to their unenriched parent 926 population of CD4 + or CD8 + T cells in Flu+ patients. j, T cell differentiation phenotype of 927 TAME-enriched tetramer + cells in relation to the unenriched parent population of CD4 + or CD8 + 928 T cells. i,j, Mean and SD are shown for all acute and follow-up timepoints, except for the acute 929 tetramer + CD4 + group (n=2), which were plotted individually. Statistical significance (p<0.05) 930

was determined using Tukey's multiple comparison test for (i) number of activation markers 931 present (0, 1 or 2+) and (j) T cell differentiation subsets. 932 933 Fig. 6 . Analyses of immune responses and clinical and genetic host factors. a,b, 934

Unsupervised heatmaps of immune, clinical and genetic parameters in Flu+ patients at a, acute 935 and b, convalescent timepoints. Interactive heatmaps (.html) are shown in Supplementary Fig.  936 8a-e for Flu+, Flu-and combined datasets. Regions of low (maroon) and high (pink) cytokine 937 clusters are boxed. c, Box plots of IFN-γ-producing cells following influenza virus infection 938 assay, at the earliest acute (V1) and convalescent (F up ) timepoints as a function of patients' 939 disease severity via binned SOFA scores of 0-1 versus 2-6. Median, IQR and whiskers 940 extending to the largest or smallest values no further than 1.5 times the IQR are shown. 941

Nonparametric Wilcoxon rank sum test with continuity correction was used for comparisons 942 between SOFA categories. Wilcoxon signed rank test (a paired test) was used to compare 943 differences between V1 and F up among the same individuals. Tests were carried out on the 944 actual data, although plots were on a log10+1 scale for ease of visualization. 945 . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) preprint

The copyright holder for this this version posted September 18, 2020. . https://doi.org/10.1101/2020.09.17.20197012 doi: medRxiv preprint

Influenza Collaborators. Mortality, morbidity, and hospitalisations due to 666 influenza lower respiratory tract infections, 2017: an analysis for the Global Burden of 667 Disease Study

Estimates of global seasonal influenza-associated respiratory 669 mortality: a modelling study

Circulating TFH cells, serological memory, and tissue 672 compartmentalization shape human influenza-specific B cell immunity

Cytotoxic T-cell immunity to 675 influenza

Prior H1N1 influenza infection and susceptibility of Cleveland Family 677 Study participants during the H2N2 pandemic of 1957: an experiment of nature

Cross-recognition of avian H5N1 influenza virus by human cytotoxic 680 T-lymphocyte populations directed to human influenza A virus

Cellular immune correlates of protection against symptomatic 683 pandemic influenza

Preexisting influenza-specific CD4+ T cells correlate with 685 disease protection against influenza challenge in humans

High levels of virus-specific CD4+ T cells predict severe pandemic 687 influenza A virus infection

Natural T Cell-mediated Protection against Seasonal and Pandemic 689 Influenza. Results of the Flu Watch Cohort Study

Recovery from severe H7N9 disease is associated with diverse response 692 mechanisms dominated by CD8(+) T cells

Human cytotoxic T lymphocytes directed to seasonal influenza 696 A viruses cross-react with the newly emerging H7N9 virus

HLA targeting efficiency correlates with human T-cell response 699 magnitude and with mortality from influenza A infection

An Increased Frequency in HLA Class I Alleles and 702

Haplotypes Suggests Genetic Susceptibility to Influenza A (H1N1) 2009 Pandemic: A 703 Case-Control Study

SNP-mediated disruption of CTCF binding at the IFITM3 promoter is 705 associated with risk of severe influenza in humans

Early hypercytokinemia is associated with interferon-induced 707 transmembrane protein-3 dysfunction and predictive of fatal H7N9 infection

Cytokine response patterns in severe pandemic 2009 H1N1 and seasonal 710 influenza among hospitalized adults

Mucosal immune responses predict clinical outcomes during 712 influenza infection independently of age and viral load

Innate immune function and mortality in critically ill children with 715 influenza: a multicenter study

The Hurdles From Bench to Bedside in the Realization and 717 Implementation of a Universal Influenza Vaccine

Progression of whole-blood transcriptional signatures from interferon-719 induced to neutrophil-associated patterns in severe influenza

Severe pandemic H1N1 2009 infection is associated with transient NK 722 and T deficiency and aberrant CD8 responses

Severe H7N9 infection is associated with decreased antigen-presenting 724 capacity of CD14+ cells

Severe Influenza Is Characterized by Prolonged Immune Activation: 726 Results From the SHIVERS Cohort Study

Preexisting CD8+ T-cell immunity to the H7N9 influenza A 728 virus varies across ethnicities

The Third International Consensus Definitions for Sepsis and Septic 730 Shock (Sepsis-3)

Antibody landscapes after influenza virus infection or vaccination

Follicular helper CD4 T cells (TFH)

Induction of ICOS+CXCR3+CXCR5+ TH cells correlates with 735 antibody responses to influenza vaccination

Cross-lineage protection by human antibodies binding the influenza B 737 hemagglutinin

HIV-dependent depletion of 739 influenza-specific memory B cells impacts B cell responsiveness to seasonal influenza 740 immunisation

Maintenance of the EBV-specific CD8(+) TCRalphabeta repertoire 742 in immunosuppressed lung transplant recipients

Understanding CD8(+) T-cell responses toward the native and 744 alternate HLA-A*02:01-restricted WT1 epitope

Perturbed CD8(+) T cell immunity across universal influenza 747 epitopes in the elderly

Human CD8(+) T cell cross-reactivity across influenza A, B and C 749 viruses

Single-Cell Approach to Influenza-Specific CD8(+) T Cell Receptor 751 Repertoires Across Different Age Groups, Tissues, and Following Influenza Virus 752 Infection

Breadth of concomitant immune responses prior to patient recovery: 754 a case report of non-severe COVID-19

Activation of Peripheral T Follicular Helper Cells During 756

Acute Dengue Virus Infection

CXCR5(+) follicular cytotoxic T cells control viral infection in B cell 758 follicles

Follicular CXCR5-expressing CD8(+) T cells curtail chronic viral 760 infection

Follicular CD8 T cells accumulate in HIV infection and can kill 762 infected cells in vitro via bispecific antibodies

Dynamics of SIV-specific CXCR5+ CD8 T cells during 764 chronic SIV infection

Recalling the Future: Immunological Memory Toward Unpredictable 766 Influenza Viruses

Human Ebola virus infection results in substantial immune 768 activation

Efficient ex vivo analysis of CD4+ T-cell responses using 770 combinatorial HLA class II tetramer staining

Successive annual influenza vaccination induces a recurrent 772 oligoclonotypic memory response in circulating T follicular helper cells

Towards identification of immune and genetic correlates of severe 775 influenza disease in Indigenous Australians

The shaping of T cell receptor recognition by self-tolerance

Public T cell receptors confer high-avidity CD4 responses to HIV 779 controllers

R: A language and environment for statistical computing. R Foundation 781 for Statistical Computing

SPICE: exploration and analysis of post-783 cytometric complex multivariate datasets

We thank Jill Garlick, Janine Roney, and the research nurses at the Alfred Hospital

Thank 789 you to Professors David Fairlie and Jim McCluskey for the MR1-5'OP-RU tetramer. We also 790 thank the study participants for providing blood for research purposes

The Australian National Health and Medical Research Council (NHMRC) NHMRC Program 795

WC supported this work. MK and SN were 796 recipients of Melbourne International Research Scholarship and Melbourne International Fee 797 Remission Scholarship. CES has received funding from the European Union's Horizon 2020 798 research and innovation program under the Marie Skłodowska-Curie grant agreement No. 799 792532. EBC is a NHMRC Peter Doherty Fellow. MA and LH are recipients of Melbourne 800 International Research Scholarship and Melbourne International Fee Remission Scholarship

University of Melbourne. The Melbourne WHO Collaborating Centre for 803

Reference Research on Influenza is supported by the Australian Government Department of 804

SG is a NHMRC SRF-Level A Fellow. JR is supported by an ARC Laureate fellowship. 805 NLG is supported by a NHMRC Ideas grant

PGT is supported by the St

ACC is a NHMRC Career Development (level 2) Fellow. KK was supported by a 809 . CC-BY-NC-ND 4.0 International license It is made available under a NHMRC Senior Research Fellowship Level B (#1102792)

KK supervised and lead the study. KK, THON, MK, CES and LL designed the experiments

820 PGT and KK analyzed data. AKW and SJK provided invaluable rHA probes. SG and JR 821 provided invaluable pMHC-I

WC provided intellectual input into the study 823 design and data interpretation

Virus strain colours match 835 those in (b). d, Days in hospital for DISI cohort and H7N9 cohort. c,d, Bars indicate the median 836 and IQR, statistical significance (p<0.05) was determined using the Kruskal-Wallis test. e, 837 Frequency of universal versus risk HLA alleles between influenza-positive (Flu+) and Flu-838 patients. f,g, Pie charts of IFITM3 SNP allele genotyping at positions rs12252 and rs34481144, 839 respectively (risk allele in black). h, Box plots of rs34481144 alleles against SOFA scores 840 showing median, IQR and whiskers extending to the largest or smallest values no further than 841 1.5 times the IQR. i, Representative serum levels and distribution of pro-inflammatory cytokines 842 in patients, measured within the first 2-3 days of hospital admission (Visit 1, V1), with varying 843 disease severity. j, Partial correlation plots showing the degree of correlation between every 844 chemokine/cytokine pair in Flu+ patients at V1 and F up

influenza vaccine strains in red and 850 sequences isolated from the nasal swab of 12 H3N2-infected patients in blue. Patient number is 851 followed by the year of recruitment, yes (Y) or no (N) for prior vaccination in the year of 852 infection, and "mm*" indicates whether the vaccine was a clade mismatch in that year. Scale bar 853 represents the number of substitutions per site. b, Antibody HAI titers of Flu+ patients at acute 854 (V1 or V2) and follow-up timepoints from the relevant infected strain (mean and SD are 855 shown). Statistical significance (p<0.05) was determined using the Mann-Whitney test between 856 acute and follow-up per strain. c,d, Geometric mean titers (GMT) per strain in Flu+ and Flu-857 patients at acute and follow-up, and from a healthy vaccinated cohort at days 0 and 28 post-858 vaccination. b-d, Both H1N1 and H3N2 titers are shown for three A

Representative antibody landscapes from a patient 861 infected with H1N1, H3N2, B/YAM or B/VIC virus. Blue shading indicate period of potential 862 exposure based on the year born. f, Antibody landscapes of H1N1-and H3N2-infected (n=7 and 863 23, respectively) and H1N1-and H3N2-non