Baseline innate and T cell populations are correlates of protection against symptomatic influenza virus infection independent of serology

Evidence suggests that innate and adaptive cellular responses mediate resistance to the influenza virus and confer protection after vaccination. However, few studies have resolved the contribution of cellular responses within the context of preexisting antibody titers. Here, we measured the peripheral immune profiles of 206 vaccinated or unvaccinated adults to determine how baseline variations in the cellular and humoral immune compartments contribute independently or synergistically to the risk of developing symptomatic influenza. Protection correlated with diverse and polyfunctional CD4+ and CD8+ T, circulating T follicular helper, T helper type 17, myeloid dendritic and CD16+ natural killer (NK) cell subsets. Conversely, increased susceptibility was predominantly attributed to nonspecific inflammatory populations, including γδ T cells and activated CD16− NK cells, as well as TNFα+ single-cytokine-producing CD8+ T cells. Multivariate and predictive modeling indicated that cellular subsets (1) work synergistically with humoral immunity to confer protection, (2) improve model performance over demographic and serologic factors alone and (3) comprise the most important predictive covariates. Together, these results demonstrate that preinfection peripheral cell composition improves the prediction of symptomatic influenza susceptibility over vaccination, demographics or serology alone. Thomas and colleagues examine preinfection baseline parameters of cellular and serologic immunity. Their findings collectively show that peripheral cell composition provides better correlates of immune protection from symptomatic influenza infection than vaccination, demographics or serology alone.

Article https://doi.org/10.1038/s41590-023-01590-2 and infection status: vaccinated-uninfected (n = 75), vaccinatedinfected (n = 33), unvaccinated-uninfected (n = 76) and unvaccinatedinfected (n = 22) (Fig. 1b). Participant age was comparable across sex and vaccination status, reducing the potential for confounding effects due to sample selection bias (Fig. 1c). Baseline serum and peripheral blood mononuclear cell (PBMC) samples were collected in the preseason period (March through May) for unvaccinated participants or 14 days after vaccination for vaccinees. Postseason serum samples were collected from September through January, with a majority collected in November. All participants exhibiting respiratory symptoms and meeting the World Health Organization (WHO)-defined criteria for influenza-like illness (ILI) were further tested by PCR. Participants with ILI and an associated influenza-positive PCR result were defined as having 'symptomatic' influenza. For the 2018 influenza season (May through September), 41 of 55 (74.5%) influenza virus infections were symptomatic, with a majority attributed to A(H1N1) viruses (n = 29, 70.7%), although infections with A(H3N2) (n = 5, 12.2%), B/Yamagata lineage (n = 1, 2.4%), B/Victoria lineage (n = 1, 2.4%) and untyped influenza titers and the T helper cell responses necessary to promote them, none are specifically designed to stimulate durable, virus-specific cellular immunity, thereby neglecting a potential ally in preventing and controlling influenza virus infection. In recent years, there have been several large-scale initiatives (for example, the Centers of Excellence for Influenza Research and Response (https://www.ceirr-network.org/) network and the Collaborative Influenza Vaccine Innovation Centers (https:// www.niaidcivics.org/) program, both funded by the National Institute of Allergy and Infectious Diseases (NIAID)) to improve influenza vaccine design by increasing surveillance, improving strain risk assessment and prediction, identifying universal humoral or T cell antigens, evaluating new delivery platforms and considering demographic-specific formulations 8,9 . Driving this push is the need to understand the complete range of protective immune responses limiting symptomatic influenza illness-termed correlates of protection (CoP)-which, to date, have been largely defined by serologic measurements alone. Few studies have examined how efficacious serologic data are as predictors of protection against influenza within the context of additional immune measures, such as cellular and innate immunity; how individual cell populations contribute to protection in the presence of existing antibody responses; or if and how these compartments coordinate.
Further complicating the study of CoP against influenza viruses is the substantial degree of baseline immune variation that exists across humans, now recognized as a key determinant in predicting the efficacy of some vaccines and therapeutics and in modeling disease outcomes [10][11][12][13][14][15] . Differences in host genetics, environmental and demographic factors, and infectious exposure histories influence the composition of the baseline innate, cell-mediated immune (CMI; adaptive T cells) and humoral compartments, and may augment varied responses to influenza virus infection or vaccination 13,[16][17][18][19][20][21][22][23][24][25][26][27][28][29] . Therefore, identifying baseline CoP necessitates the use of large, well-curated, influenza virus-seropositive human cohorts in which immune and demographic data are considered together. To address these challenges, the second iteration of the Southern Hemisphere Influenza Vaccine Effectiveness and Response Study (SHIVERS-II) was established to follow unvaccinated and vaccinated adults within a natural community setting in New Zealand. Using baseline serum and peripheral blood samples from 206 SHIVERS-II adult participants, we aimed to define baseline immune cell subsets correlated with protection against symptomatic influenza independently from or synergistically with humoral responses. Further, we investigated the quantitative and relative contributions of baseline cellular and humoral immune responses in mediating protective anti-influenza virus immunity. Our analytic approach incorporated high-dimensional flow cytometry data, serology measures, vaccination status and demographic data into statistical models, allowing the identification of individual baseline cell populations associated with increased risk of or protection against symptomatic influenza across vaccinated and unvaccinated adults while accounting for baseline immune and demographic variations. Using univariate analyses and multivariate partition and regression models, we demonstrate that the baseline composition of peripheral cells improves the prediction of influenza susceptibility over serology, vaccination or demographics alone. Our results underscore the complexity and variability of baseline cellular responses, support influenza vaccine design strategies targeting optimized cell subsets to induce long-lasting heterosubtypic immunity, and provide improved methods to compare vaccine efficacy.

Study population
A total of 206 adult participants (Table 1) were selected from the 2018 SHIVERS-II cohort for inclusion in this study (Fig. 1a) and analyzed according to a predefined analysis pipeline (Extended Data Fig. 1). The selected participants were divided into four roughly age-and sex-matched comparator groups based on influenza virus vaccination Article https://doi.org/10.1038/s41590-023-01590-2 A (n = 5, 12.2%) viruses were also recorded (Supplementary Table 1). Cryptic influenza cases, identified by seroconversion (fourfold or greater increase in hemagglutination inhibition (HAI) antibody titer ≥1: 40) in the absence of a PCR-confirmed symptomatic influenza episode, comprised the other 14 of 55 (25.5%) infections. Influenza virus strains associated with infection were proportional between male and female participants (Fig. 1d). Supplementary Tables 1 and 2 detail the vaccination status and demographics of the participants according to influenza virus strain.

Baseline serology measures are associated with protection
Anti-influenza virus antibodies targeting hemagglutinin (HA) and neuraminidase (NA) are known CoP [30][31][32][33][34][35] . To determine the degree of preexisting humoral immunity to dominant influenza viruses circulating in the region as well as to those present in the 2018 quadrivalent influenza vaccine [36][37][38] , we measured inhibiting antibody titers through inhibition assays (HAI or NA inhibition (NAI)) and total binding antibody titers through ELISA. The overall correlation between inhibiting and total anti-HA (Fig. 2a) or anti-NA (Fig. 2b) serology measures against homosubtypic influenza virus targets was positive, reflecting a large degree of concordance between the assays. Three correlation clusters were also identified across all serology measures (Extended Data Fig. 2a), suggesting an association between existing anti-influenza antibody responses to homo-and heterosubtypic targets. In 2018, the quadrivalent influenza vaccine was 38% effective at preventing influenza-associated hospitalizations 38 . In our study, this vaccine elicited HAI titers of ≥1:40, a purported cutoff for protection, in roughly half of the study participants sampled 14 days after immunization (Extended Data Fig. 2b). Vaccinated participants had significantly increased HAI and NAI titers against all targets compared to unvaccinated participants (Fig. 2c,d). The ELISA results showed that the total binding antibody titers were also significantly increased in vaccinees, except the titers of antibodies against B/Victoria lineage HA and A(N1), which were lower in vaccinees than in unvaccinated participants (Fig. 2e,f). Further, we compared the antibody levels in uninfected participants and participants with cryptic influenza to those in participants with symptomatic influenza. For these and other downstream analyses, uninfected and cryptic infection cases were grouped together (uninfected/cryptic group), as these cases did not meet the study criteria for symptomatic influenza and represent a more protected group. Nearly all NAI titers were significantly increased in the uninfected/cryptic group (Fig. 2h), whereas only the total anti-A(H1) and anti-B/Yamagata lineage NA binding antibody levels were significantly increased in uninfected/cryptic cases (Fig. 2i,j). Previous studies have suggested that demographic factors can influence humoral responses, illness susceptibility and influenza vaccine effectiveness 17,[39][40][41] . Therefore, we examined the correlations between age, sex, body mass index (BMI) and serology measures, using locally estimated scatterplot smoothing, to inform downstream statistical modeling. Age and BMI were not correlated with sex (Extended Data Fig. 2c); however, we observed significant relationships between age or BMI and several serology measures (Extended Data Fig. 2d-k). Although these results indicate that the effect of demographics on antibody levels is limited to specific targets, this variability may influence serology measures in downstream modeling and was therefore taken into account.
As higher antibody titers were observed in the uninfected/cryptic group, we used generalized logistic regression models (GLMs) to establish the risk of symptomatic influenza virus infection given individual serology measures while adjusting for demographics and influenza vaccination status. Elevated NAI titers against A(N1), A(N2), B/Victoria lineage NA and B/Yamagata lineage NA and total binding antibodies targeting A(H1) and B/Yamagata lineage HA were significantly associated with protection (odds ratio (OR) < 1) (Fig. 2k,l). In line with the elevated HAI titer observed in symptomatic influenza cases (Fig. 2g), increased risk (OR > 1) was associated with an elevated A(H3) HAI titer (Fig. 2k). This likely reflects the dominance of regional A(H3N2) virus infections in the previous year (2017) (ref. 42), skewing preexisting antibody levels away from an effective titer needed to neutralize the A(H1N1) viruses circulating during 2018 (ref. 38). Together, the serology results from SHIVERS-II are consistent with the results of numerous studies demonstrating protection mediated by anti-influenza antibodies. Further, these results suggest that the baseline levels of anti-NA inhibitory antibodies are particularly important in determining the risk of symptomatic influenza infection in this cohort.

Immune measures vary across vaccination and infection status
As our participants are relatively evenly divided by vaccination status, this allowed us to observe the statistical behavior of all variables separately across vaccination and infection groups. Statistical differences in individual baseline demographic, serologic and cellular covariates were evaluated between participants with and without symptomatic infection and are presented in Supplementary Table 3. Regardless of vaccination and infection status, no significant differences in demographic parameters (age, sex, BMI and ethnicity) were observed (Supplementary Table 3). Regarding serology measures, unvaccinated participants in the uninfected/cryptic infection group had higher median values of total binding antibodies to A(H1) and B/Yamagata lineage NA and NAI titers against B/Yamagata lineage, whereas the median A(H3) HAI titer was elevated in symptomatic influenza cases (Supplementary Table 3). Regardless of vaccination status, higher median NAI titers against A(N1), A(N2) and B/Victoria lineage were observed in the uninfected/cryptic group (Supplementary Table 3).
To evaluate variations in the baseline cell profile across participants, PBMCs were stained with fluorescent antibody panels distinguishing a wide array of phenotypic and functional myeloid (Extended Data Fig. 3) or lymphoid (Extended Data Fig. 4) cell subsets. To promote influenza virus-specific cytokine production for functional intracellular cytokine staining (ICS) in the lymphoid/functional panel, a portion of PBMCs from the participants were stimulated with conserved influenza virus peptide pools or live A(H1N1) or A(H3N2) viruses. Across vaccinated and unvaccinated participants, 11 cell populations had a  1 (2018). RT-PCR, reverse transcription followed by PCR. b, Following consented enrollment, demographic information and whole blood samples were collected from all vaccinated and unvaccinated participants in the preseason period (nonvaccinated baseline) and 14 days after vaccination (vaccinated baseline). Participants meeting the WHO-defined criteria for ILI were tested for influenza viruses by PCR, and confirmed cases were sampled further during acute infection. All enrolled participants were sampled after the season. Cryptic infections were adjudicated in the postseason period from ILI-and PCR-negative participants with a fourfold or greater increase in HAI antibody titers without postvaccination HAI seroconversion. Right, 206 enrolled participants were selected for study inclusion from four baseline comparator groups (unvaccinated-uninfected, unvaccinatedinfected, vaccinated-uninfected and vaccinated-infected) based on age-and sex-matching. c, Sex (assigned at birth) of n = 206 participants stratified by vaccination status and age (years) and compared by two-sided Wilcoxon rank-sum test (unvaccinated female (n = 58) versus unvaccinated male (n = 42), P = 0.07; vaccinated female (n = 66) versus vaccinated male (n = 42), P = 0.63). Boxes represent the median and 25th-75th percentiles; whiskers indicate the minimum (left) and maximum (right) values no further than 1.5 times the interquartile range (IQR); notches extend to 1.58 × IQR/sqrt(n), providing the 95% confidence interval (CI). P < 0.05 indicates significance. d, Participants' sex stratified by influenza virus infection status and strain. Influenza A (FluA) viruses include A(H1N1) and A(H3N2) strains; influenza B (FluB) viruses include the B/Victoria lineage and B/Yamagata lineage strains. Article https://doi.org/10.1038/s41590-023-01590-2 significantly increased frequency in uninfected/cryptic cases (Supplementary Table 3). Conversely, 20 cell populations had a significantly increased frequency in symptomatic infection cases. Although some of these cell populations had elevated frequencies regardless of vaccination status, the overall cell profiles were distinct, suggesting an effect of immunization on protective or susceptible cellular profiles.
We also identified seven populations that increased the risk of symptomatic influenza (OR > 1) among vaccinated (plasmacytoid dendritic cells (pDCs), granzyme B (GzmB) + IFNγ − NK cells, activated NK cells, γδ T cells, CC chemokine receptor 5 (CCR5) + CD8 + memory T cells) or unvaccinated (activated NK cells, γδ T cells, CCR5 + CD8 + memory T cells, total CD4 + T cells, TNF&aplha; + CD8 + T cells) participants (Fig. 3a). The overall cellular CoP profile defined by the univariate analyses indicates a diversity of polyfunctional responses to influenza viruses, engaging both the adaptive and innate compartments (Fig. 3b, left). In contrast, the susceptibility-associated profile skewed toward nonspecific inflammatory populations and single-cytokine producers (Fig. 3b, right). Comparison of cell profiles from unvaccinated participants across protective and nonprotective status also indicates that the baseline composition of innate populations is an important determinant of protection from symptomatic influenza in the absence of recent vaccine-induced immunity.

ROC thresholds define protection and susceptibility cutoffs
From the univariate analyses, we identified individual immune measures associated with protection from or increased risk of developing symptomatic influenza. To provide a quantitative value denoting   Tables 3 and 4. c, Representative thresholds are depicted as cell frequency (% parent) with their associated ROC curves. A threshold defines the cell frequency at which the ROC curve sensitivity (true positive rate) equals 0.5 and represents the cutoff above which an individual factor correctly associates 50% of cases as protected or susceptible. AUC values indicate the overall quality (true versus false positives) of the individual measure in discerning protection or susceptibility. CD4 + , CD4 + T cell; CD8 + , CD8 + T cell; CK, cytokine; single CK, single-cytokine-producing; dual CK, dual-cytokine-producing; TEMRA, terminally differentiated effector memory. protection or susceptibility, we derived 'threshold' values for each cell population (Supplementary Table 5) or serology measure (Supplementary Table 6) through receiver-operating characteristic (ROC) analysis. Thresholds define the value at which an individual parameter accurately identifies ~50% of true positive cases (sensitivity) while minimizing false positives (specificity). Therefore, a threshold is a quantifiable measure at and above which a factor accurately associates with protection or susceptibility. The area under the ROC curve (AUC) defines the quality of the immune measure as a classifier. For example, a baseline frequency of CD4 + dual-cytokine producers above a threshold of 0.022% identifies at least 50% of vaccinated individuals as protected from symptomatic influenza (Fig. 3c). Conversely, a threshold frequency of ≥12.7% activated NK cells correctly identifies at least 50% of individuals as susceptible to symptomatic influenza (Fig. 3c). Comparing the protection-associated factors with the highest AUC values (Supplementary Tables 5 and 6), we observed that serology measures were more accurate in classifying cases in unvaccinated participants (for example, NAI B/Yamagata lineage AUC = 0.9, NAI B/ Victoria lineage AUC = 0.86 and NAI A(N1) AUC = 0.85), whereas cellular measures were more accurate in vaccinated participants (for example, dual-cytokine-producing CD4 + T cell AUC = 0.78, IL-2 + CD4 + T cell AUC = 0.77 and ICOS + cT FH cell AUC = 0.77). This is likely because the overall antibody levels are elevated in vaccinated individuals, so differences in cell populations provide improved resolution; in unvaccinated individuals, increased baseline antibody levels are more variable and therefore more accurate classifiers likely reflecting a more recent exposure to influenza antigens. However, as these are single-parameter measures, we cannot determine from the AUC values alone whether other confounding factors affect these classifications.

Cryptic influenza cases are associated with unique cell profiles
Susceptibility to symptomatic influenza encompasses risks associated with (1) virus exposure and (2) symptom development following exposure. Outside of regional prevalence estimates, it was difficult to determine whether uninfected study participants were indeed exposed to influenza viruses during the 2018 season. Therefore, we analyzed a subset of participants with confirmed exposures: symptomatic influenza cases (confirmed by PCR, n = 41) and cryptic infection cases (confirmed by seroconversion, n = 14). Univariate regression analyses demonstrated that elevated baseline levels of several anti-HA or anti-NA inhibitory or total binding antibody measures were significantly associated with reduced risk of symptomatic influenza, whereas demographic factors and recent influenza vaccination history were not (Fig. 4a). We then compared cell frequencies between symptomatic and cryptic cases. In concordance with our univariate results, the frequencies of both activated and GzmB − IFNγ + NK cells were increased at baseline in symptomatic cases, further suggesting a role in susceptibility (Fig. 4b).
Conversely, we identified 15 cell populations with increased baseline frequencies in cryptic infection cases, comprising a unique set of protective responses, including several cell populations (conventional DC type 2 (cDC2) and IL-21 + cT FH cells) not identified in the prior univariate analysis (Fig. 4c). Similar to the univariate protective profile, the baseline cell populations with increased frequencies in cryptic cases were diverse and polyfunctional, reflecting adaptive T cell activation and cytokine production (CD4 + effector T cells, IL-17 + CD4 + T cells, IL-2 + CD4 + T cells, IL-2 + CD8 + T cells, IFNγ + CD8 + T cells, dual-cytokine-producing CD8 + T cells), humoral engagement (cT FH subsets) and innate immune activities (cytotoxic and cytokine-producing NK cells, mDCs).

Immune cell populations cluster into co-regulated modules
The above univariate analyses consider cell populations independently and do not account for coordinated responses often observed across the adaptive and innate immune compartments. We sought to understand the cell-to-cell relationships within our dataset and determine which populations represent co-regulated immune cell modules, defined here as cell populations with a strong positive correlation in cell frequencies suggestive of parallel responses. Using Pearson's bivariate correlation, we identified 9 myeloid (Fig. 5a) and 13 lymphoid/functional (Fig. 5b) cell modules across all study participants. We also evaluated immune cell correlations separately in vaccinated and unvaccinated participants (Extended Data Fig. 5) and found them to largely reflect the same correlation groups identified from the complete set of participants. Correlation results from the complete set of participants were used for downstream analyses. The cell frequencies from individual populations within a given module were averaged to provide the final 'cluster' frequency values, which were compared according to influenza virus infection status. Clusters with increased average cell frequencies in both influenza-negative and cryptic infection cases included myeloid clusters 1 (mDCs, CD14 + CD16 + intermediate monocytes), 3 (cytotoxic and cytokine-producing NK cells, neutrophils) and 8 (cDC2) (Fig. 5c), as well as lymphoid/functional clusters 4 (TNF&aplha; + , IFNγ + , IL-17 + and dual-cytokine (IL-2/ TNF&aplha;/IFNγ)-producing CD4 + T cells; IFNγ + and dual-cytokine (IL-2/TNF&aplha;/IFNγ)-producing CD8 + T cells), 8 (CD4 + and CD8 + effector T cells), 9 (PD-1 + CD4 + and PD-1 + CD8 + T cells, PD-1 + and CXCR3 + cT FH cells) and 13 (total CD8 + T cells, GzmB + CD4 + and GzmB + CD8 + T cells, γδ T cells) (Fig. 5d). Conversely, the cell frequency of myeloid cluster 7 (activated NK cells) was found to be increased in symptomatic cases ( Fig. 5c), supporting the previous observations made in the univariate GLM analysis. Together, these results reflect the involvement of coordinated cellular responses in determining influenza virus susceptibility in cases of confirmed exposure.   (4) combined model comprising all available variables. Model performance was measured by sensitivity and specificity metrics and by scoring data from the testing set in ROC curves (Fig. 6a). We observed that both the base and myeloid models categorized participants with 79% accuracy (Fig. 6a, bottom right), whereas the lymphoid model increased sensitivity and specificity and substantially improved categorization accuracy to 86%, demonstrating the influence of lymphoid populations in categorizing influenza cases. As the combined model (84% accuracy) did not improve accuracy over the lymphoid model, we examined the relative contributions of lymphoid and myeloid populations in model performance. Three additional random forest models-myeloid-only, lymphoid-only and lymphoid + myeloid-were also built, trained and compared (Extended Data Fig. 6). The lymphoid-only model (79%) performed better than the myeloid-only model (72%) and was as accurate as the base model (79%) in classifying symptomatic influenza cases. The lymphoid + myeloid model also exhibited higher accuracy (81%) than the base model. These data indicate that lymphoid populations improve overall accuracy in classifying symptomatic influenza cases compared to myeloid populations, serology, vaccination status and demographics alone. Using variable importance (VIP) analysis of the combined model, we then derived 'importance' values representing the individual effect strength on the dependent variable. Strikingly, the top 4 (and 25 of the top 30) most important covariates used in categorizing symptomatic and uninfected/cryptic influenza cases were of cellular origin, with ICOS + cT FH cells having the highest importance (Fig. 6b). Indeed, the frequency of ICOS + cT FH cells was significantly higher in uninfected/cryptic influenza cases across both unvaccinated       and vaccinated participants (Fig. 6c) and had a predictive accuracy of 81.6% following cross-validation and adjustment for age, sex and vaccination status (Fig. 6d). Together, the results from the partition analyses emphasize the strong contributions of both baseline innate and adaptive cellular immunity in regulating responses to influenza virus and mitigating symptomatic disease.

Baseline immune measures predict influenza susceptibility
Although partition analysis helps determine which combined or individual baseline variables best categorize symptomatic and uninfected/ cryptic cases, these models are limited in risk assessment. Indeed, the random forest and VIP analyses here do not ascribe effect directionality nor associate any factor with increased or decreased risk of symptomatic influenza. Therefore, we used GLMs to assess influenza risk given all measures of immune and demographic variation present at baseline. A multivariate GLM was constructed to identify the differential effects of cell populations, anti-influenza virus antibodies, vaccination status and demographics on determining the overall risk of symptomatic influenza (Fig. 7). GLMs with highly correlated factors (collinearity) often produce unreliable coefficients with high standard error (SE) values 43 . Owing to the high dimensionality and observed correlation between the cellular and serology immune measures in our data, we took several steps to reduce variable collinearity (Pearson's correlation and variance inflation factor (VIF)) and optimized selected covariates by comparing stepwise models through Akaike information criterion (AIC) and Bayesian model averaging (BMA) (Extended Data Fig. 1). The final multivariate GLM was constructed using the set of selected covariates against a response variable denoting symptomatic infections (Fig. 7a). The model comprises 19 covariates, including 4 (CD107a + CD4 + T cells, intermediate monocytes, A(N1) NAI and B/Victoria lineage NAI) that are significantly associated with protection and 2 (TNF&aplha; + CD8 + T cells and γδ T cell receptor (TCR) + T cells) that significantly predict increased symptomatic influenza risk. Using CD107a + CD4 + T cells as an example, the GLM readout shows that for each 4.2% increase in CD107a + CD4 + T cell frequency, the odds associated with developing symptomatic infection decrease by 11.1-fold (1/0.09).
Although neither age (P = 0.051) nor 2018 vaccination status (P = 0.859) had significant effects on risk, these variables were included in the GLM to account for variability across participants. Owing to the strong correlations observed among cell populations, we can further infer protection or susceptibility associations by identifying the myeloid or lymphoid modules within which these populations reside (Fig. 7b) and by referencing the univariate associations for a given cell population (Fig. 3b) and cellular factors, when considered together, synergistically predict influenza virus infection outcomes and contribute to the relative risk of developing influenza disease. Results from this study support recent calls to target adaptive T cell populations in next-generation vaccine designs 9 and strongly suggest that both cellular and serology measures are necessary to fully assess the efficacy of influenza vaccines.

Discussion
Immune responses to influenza viruses are complex and include coordinated innate, CMI and serologic responses to clear active infections; build durable immune memory; and rapidly neutralize subsequent challenges. Equally nuanced are the determinants of risk and susceptibility to symptomatic influenza, which comprise individual comorbidities and demographic risk factors, vaccination and infection histories, and variations in baseline immune profiles. In our study, we investigated which baseline immune cell populations affect the risk of developing symptomatic influenza and examined how differences in preexisting anti-influenza virus antibody titers, influenza vaccination status and participant-level demographics influence these associations. Our univariate analyses demonstrated that protection from symptomatic influenza correlates with increased frequencies of diverse and polyfunctional influenza virus-specific CD4 + and CD8 + T cells; cells associated with the engagement of humoral responses, including cT FH cells, mDCs and T H 17 cells; and innate immune effector CD16 + cytotoxic and cytokine-producing NK cells, many of which have been observed in other studies. As expected, protection profiles were distinct between vaccinated and unvaccinated adults. The vaccine-associated protective profile favored effector and polyfunctional adaptive populations, pointing toward influenza virus-specific T cell activation. Increased frequencies of antigen-presenting mDCs and cT FH populations in vaccinees are also suggestive of humoral crosstalk, discussed in detail below. In unvaccinated participants, protection was associated with increased baseline frequencies of populations exhibiting analogous, albeit antigen-agnostic, functions to those observed following vaccination. For example, increased frequencies of T H 17 cells, which were protective in our study, have been shown to be involved in anti-influenza virus responses either directly 44 , through B cell engagement promoting immunoglobulin G (IgG), IgA and IgM 45,46 , or by recruiting preexisting B cells into the lung without vaccine priming [46][47][48] . Therefore, T H 17 cells may compensate for the absence of other vaccine-induced CD8 + or CD4 + T helper subsets. Cytokine-producing and cytotoxic NK cells exhibiting virus-limiting potential similar to CD8 + T cells were also strongly protective in unvaccinated participants, supporting studies demonstrating the active roles of NK cells in anti-influenza virus responses and in limiting disease severity [49][50][51][52][53][54][55][56] . NK cells, however, may also be a double-edged sword. One group reported a link between IFNγ + NK cells and extrapulmonary inflammation leading to poor outcomes in pregnant women with influenza virus infection 57 . Coinciding with these findings, we also observed increased risk of symptomatic influenza in participants with increased baseline frequencies of activated NK cells (CD56 low HLA-DR hi CD16 − ), suggesting that NK cells have pleiotropic roles during influenza virus infection.
A main takeaway from this study is the importance of cell-humoral crosstalk promoting increased baseline levels of anti-influenza virus antibodies. We report that elevated baseline NAI titers are the best serology predictors of symptomatic influenza and are associated with protection in this cohort, supporting other recent findings [32][33][34][58][59][60] . However, we also found that the addition of lymphoid cell population composition improves the prediction of symptomatic influenza cases, arguing that baseline antibody levels cannot be considered alone. Indeed, sustained B-T cell crosstalk within germinal centers, as well as cT FH cells in peripheral blood, is associated with productive anti-influenza virus humoral response [61][62][63][64][65] . In our study, evidence of this communication is supported by the association of CD4 + populations, including cT FH subsets, with protection. In fact, ICOS + cT FH cells represent the most important variable in determining the accuracy of the combined random forest model. From our analysis of co-regulated cell modules, we found that ICOS + cT FH cells correlate with CD4 + and CD8 + memory T cell populations (lymphoid/functional cluster 11), linking increased ICOS + cT FH cell frequency to long-term memory. We also identified correlations across T H 17, IFNγ + , and polyfunctional CD4 + and CD8 + T cells (lymphoid/functional cluster 4), which are known to improve outcomes during influenza virus infection 66 . Further, our multivariate regression modeling identified that reduced risk of symptomatic influenza is independently associated with CD107a + CD4 + T cells, CD14 + CD16 + intermediate monocytes, and anti-influenza virus NAI titers targeting A(H1N1) and B/Victoria lineage viruses. CD107a is a marker of degranulation 67 and implies that these CD4 + T cells were specifically activated by influenza virus antigens during ICS and flow cytometry. That these populations are observed at baseline indicates that cT FH and other influenza-responsive T cells persist following influenza antigen exposure in prior seasons and may confer 'carryover' protection in a subset of participants 68 . However, as our study was limited to immune measures present in circulation, we did not evaluate contributions from mucosa-associated cells, secretory IgA, or IgG produced by respiratory-resident memory B cells, which are likely CoP (reviewed in refs. 22,69). As ICOS + cT FH and other functional lymphoid cells rank highly in our study, we hypothesize that cellular correlates might provide better insights toward mucosal immune responses (including antibody responses) compared to serum antibody measures.
A strength of our statistical modeling approach is the ability to define not only CoP but also baseline factors that increase the risk of symptomatic influenza. Among the strongest risk-associated cell populations identified across univariate and multivariate models were γδ T cells. A prior study from our group demonstrated that γδ T cells are integral for lung repair in neonatal mice 70 ; however, in this adult SHIVERS-II cohort, the opposite effect was observed. This incongruence could be attributed to differences in γδ TCR repertoires, which change substantially with age, tissue localization and infection history 71,72 . Pediatric γδ T cells exhibit nonoverlapping TCR repertoires, whereas adult γδ T cells heavily favor Vγ9 and Vδ2 chain usage associated with polycytotoxic cytokine profiles 72 . Therefore, it is possible that, within the context of adult influenza virus infection, γδ T cells could contribute to inflammation-mediated pathology rather than tissue repair.
The results presented in this study emphasize the need to evaluate anti-influenza virus responses and vaccine efficacy from both the serologic and cellular standpoints, as well as argue that the composition of baseline cell populations is a better overall predictor of influenza susceptibility than serology and vaccination status alone. As baseline cellular and humoral immune landscapes significantly vary within human populations and across study cohorts, we propose that the protection-associated cell profiles defined in this study are a better metric to evaluate influenza susceptibility than any one cell population or serology measure. Although individual cellular CoP against symptomatic influenza likely differ between cohorts, their functions may contribute to a converging protective immune profile, which can be more easily compared across studies. Further, the statistical and predictive models described here will also serve as useful tools to evaluate vaccine efficacy, identify new targets for next-generation vaccines promoting both cellular and humoral responses, and identify additional at-risk human populations for vaccination. However, further questions remain. It will be important for follow-up studies to validate these risk and protection models across more diverse cohorts to determine whether there are population-specific baseline cell profiles correlated with vaccine failure or disease severity. Assessment of CMI and innate responses is especially important during pregnancy 73 and in adults over the age of 65 years, who are known to generate poor protective humoral responses following immunization 74,75 . Article https://doi.org/10.1038/s41590-023-01590-2 The continuing aim of SHIVERS-II is to assess the immunological parameters of influenza virus infection and vaccination over several years. As the data presented in this study are derived from participants enrolled in the inaugural year of the SHIVERS-II study, there is great potential for future studies using this cohort to validate and expand on these results for years to come.

Study limitations
We acknowledge several limitations of our study. The first is the relatively small sample size across comparator groups and the increased circulation of the A(H1N1) virus during the study year. An over-representation of infections caused by the A(H1N1) virus means that no strong strain-specific conclusions regarding influenza B or A(H3N2) influenza viruses can be drawn. Future studies using samples collected in subsequent years of the SHIVERS-II cohort will focus on providing further validation to the models described here and on expanding the analyses to include a wider range of influenza virus types and subtypes. Second, participants enrolled in the SHIVERS-II cohort represent a limited regional, age and genetic scope; therefore, specific interpretations of these data should be made solely within this context. Further validation of our risk modeling and associated cell profiles should be compared across a more diverse ethnogeographic space to account for the large degree of baseline variability across participants. Finally, limited considerations were made regarding genetic contributions to the risk of influenza as, although ethnicity was reported, this is not a robust evaluation of genetic background 76 . Despite these limitations, data presented here are from a subset of year 1 participants, and it is our directive to continue to investigate this important population across multiple years of the SHIVERS-II study.

Online content
Any methods, additional references, Nature Portfolio reporting summaries, source data, extended data, supplementary information, acknowledgements, peer review information; details of author contributions and competing interests; and statements of data and code availability are available at https://doi.org/10.1038/s41590-023-01590-2.

SHIVERS-II study design and participant definitions
The SHIVERS-II study is a population-based, longitudinal (2018-2022) prospective cohort study in Wellington, New Zealand, initiated to evaluate cellular and serologic immune responses at baseline and during recall among adults with current influenza virus vaccination and/or infection. During year 1 (2018), more than 22,000 individuals aged 20-69 years were randomly selected for recruitment from three participating primary health organizations' healthy patient networks. Selected individuals received a study invitation and information packet by mail, and those interested provided consent through an online electronic consent form and questionnaire in which demographic and contact information, vaccination status, and health and ILI status were provided. A total of 2,195 individuals were ultimately enrolled following written informed consent (Fig. 1a). Each participant had one baseline blood draw (preseason, usually during March through May) and a postseason blood draw (usually during October through December). Consented participants received a New Zealand $30 gift card after each blood or swab sample collection to recognize their time and effort. Whole blood samples were collected in vacuum tubes containing heparin. Additional samples, including blood draw samples or nasal swabs, were obtained from participants with PCR-confirmed influenza virus infection or who received an influenza virus vaccination during the influenza season (May through September). Briefly, participants who received an influenza virus vaccination during this time provided a blood draw sample 14 days after vaccination (postvaccination baseline). All participants were monitored, through weekly surveys, for ILI, defined by the WHO as "acute respiratory illness with cough and a history of fever/ measured fever of ≥38 °C and illness onset within the past 10 days" (ref. 77). If ILI presented, a respiratory specimen (nasopharyngeal or throat swab) was collected and subjected to PCR molecular testing for respiratory viruses (see methods below). Participants with both ILI and an associated influenza-positive PCR result were categorized as having symptomatic influenza virus infection, and additional acute (1-2 weeks after ILI onset) and convalescent (4-7 weeks after ILI onset) blood draws were conducted. Respiratory samples from these PCR-positive influenza cases were further processed to determine the influenza virus subtype (see methods below). Inhibiting antibody titers against HA and NA were determined from baseline and postseason serum samples by HAI assay or NAI enzyme-linked lectin assay (NAI-ELLA), respectively (see methods below). Total binding antibody titers against purified, full-length HA and NA were measured by ELISA (see methods below). HAI titers were used to detect cryptic influenza virus infection cases not identified by PCR. Briefly, cryptic influenza virus infections were defined by (1) the absence of an ILI-associated influenza-positive PCR result and (2) one of the following seroconversion criteria: unvaccinated participants with a fourfold or greater increase in the preseason to postseason HAI titer, vaccinated participants with a fourfold or greater increase in the postvaccination to postseason HAI titer, or vaccinated participants with a fourfold or greater increase in the preseason to postseason HAI titer without seroconversion following vaccination (that is, seroconversion cannot be accounted for by vaccination), with the second titer being at least 1:40 in all cases.

Participant categorization and selection
Participants were categorized based on 2018 influenza virus vaccination (vaccinated or unvaccinated) or infection (symptomatic, cryptic or uninfected) status. The symptomatic influenza virus infection group met the WHO-defined criteria for ILI and had a positive influenza virus PCR test. Cryptic influenza virus infections were defined as cases with zero to one mild symptom that did not meet the WHO-defined ILI criteria (and therefore do not have an associated PCR test) but were confirmed to have had a cryptic influenza virus infection by seroconversion. All cryptic infections in this study are, by definition, subclinical as no influenza PCR-positive-associated ILI was reported. The uninfected group did not meet the WHO-defined ILI criteria, did not undergo an associated PCR test and were negative for cryptic infection. A total of 206 participants were selected from year 1 of the SHIVERS-II study and comprised 82 men (39.8%) and 124 women (60.2%) residing in the Wellington, New Zealand, catchment area who had an average age of 43.8 ± 12.6 years (range 20-68 years), had a BMI of 27.5 ± 5.6 kg m −2 (range 18.3-48.5 kg m −2 ) and predominantly self-identified as being of New Zealand-European descent (83.0%) ( Table 1). Participants were selected into four comparator groups: vaccinated-uninfected (n = 75), vaccinated-infected (n = 33), unvaccinated-uninfected (n = 76) and unvaccinated-infected (n = 22). Participants' age (in years) and sex (assigned at birth) were roughly matched when selecting across comparator groups. Sample size requirements for statistical testing of the covariates were calculated using the R package WMWssp (v.0.4.0) (refs. [78][79][80]. At a defined power of 0.8, our sample was determined to be sufficient in size for the planned comparisons.

Molecular testing for respiratory viruses
For the detection of respiratory viruses including influenza viruses, the CDC standard RT-PCR assay was performed on RNA extracted from the participants' nasopharyngeal or throat swabs, as previously described [81][82][83][84][85] . All samples were tested for a standard panel including influenza A (A(H1N1)pdm09, A(H3N2)) viruses, influenza B (B/Yamagata and B/Victoria lineages) viruses, respiratory syncytial virus, rhinovirus, human metapneumovirus, parainfluenza types 1-3, adenovirus and enterovirus.

Influenza virus antigenic typing
Antigenic typing of the infecting influenza virus strain was performed on all samples identified by qPCR as influenza virus positive with low cycle threshold values. Following virus isolation by sample inoculation into Madin-Darby canine kidney cells stably expressing human α-2,6-sialyltransferase 86 , antigenic typing of influenza viruses was performed using an HAI assay with standard antisera supplied by the WHO Collaborating Center in Melbourne, Australia. Any untypeable influenza A viruses were forwarded to the WHO Collaborating Center in Melbourne or the CDC in Atlanta, GA, for further characterization.

HAI assay
Anti-HA antibody titers were determined by HAI assay as previously overnight, heat-inactivated at 56 °C for 30 min and tested by HAI assay with 0.5% turkey red blood cells. Serum samples were serially diluted twofold in 96-well U-bottom plates beginning at a 1:10 dilution. Four agglutinating doses were added to each well, and the samples were incubated. Titers were read after 30 min of incubation with 50 μl of 0.5% turkey red blood cells.

NAI-ELLA assay
NAI antibody titers were determined by ELLA for influenza A and B viruses 37,87,88 . Briefly, recombinant viruses, composed of the NA from each of the influenza A viruses used for HAI and a mismatched H6 on a PR8 backbone, were generated by reverse genetics and used as antigens. Serum samples were tested at a starting dilution of 1:10. Because no recombinant antigen was available for influenza B viruses, ELLA was performed against the whole B/Brisbane/60/2008 virus (B/Victoria lineage) as previously described 89 .  (Gibco, cat. no. 15260037) and 0.01% Tween-20 (ThermoFisher, cat. no. 85113) in 1× PBS) for 6 h at 25 °C. Serum samples treated with receptor-destroying enzyme were titrated at 1:160-1:20,480 in blocking buffer, and 50 μl of each serum dilution or blocking buffer (negative control) was added to replicate wells following the removal of blocking buffer. The plates were then sealed and incubated at 4 °C overnight. Plates were washed three times with wash buffer (0.05% Tween-20 in 1× PBS), after which 100 μl of anti-human IgG (Fab-specific)-horseradish peroxidase secondary antibody (Sigma, cat. no. A0293, polyclonal, lot no. 0000201676, dilution 1:3,000) was added per well and plates were incubated at 4 °C overnight. The plates were then washed three times in wash buffer, and 50 μl of tetramethylbenzidine (Sigma, cat. no. T0440) was added to each well. After 8 min, 50 μl of stop solution (1 N H 2 SO 4 ; Sigma, cat. no. T0440) was added per well and the plates were read at 450 nm on the BioTek Synergy H1 microplate reader. To normalize the optical density (OD), average OD values from negative control wells were subtracted from the average OD of replicate sample wells per plate. AUC values were calculated from the xy plots of dilution by the normalized OD values in GraphPad Prism 9 (v.9.5.1 (528)), using the following parameters: baseline y = 0; minimum peak height, <10% of the distance from minimum to maximum y; all peaks above baseline; five significant digits.

Flow cytometry: surface and intracellular staining
Cryopreserved aliquots of PBMCs were thawed at 37 °C and suspended in RPMI 1640 supplemented with 10% heat-inactivated FBS (Gibco, cat.  TreeStar). Spectral cytometry panels conform to best-practice principles in spectral analysis 90,91 .

Flow cytometry: selecting cell subsets for primary analysis
Cell frequencies were resolved for more than 89 CMI and innate subtypes, including populations previously shown to correlate with anti-influenza virus response. For unbiased selections, several blinded quality control measures were used to identify cell populations for inclusion in the primary analysis. These included filtering out individual samples with low cell viability (≤5%) and/or high cellular debris (>25% of events) and eliminating cell populations with limited dynamic range in cell frequency. Using these measures, we reduced the number of individual myeloid populations from 23 to 16 and the number of lymphoid/functional cell populations from 66 to 41 for the primary analysis. Refer to Supplementary Table 7 for cell population naming conventions used in this study.

Defining co-regulated immune cell modules
Co-regulated immune cell modules were independently determined for myeloid or lymphoid/functional cell populations across all vaccinated and unvaccinated participants. For the myeloid cell populations, modules ('clusters') were identified by a significant positive correlation between cell frequencies (% parent), using Pearson's bivariate correlation and an FDR adjustment cutoff of q ≤ 0.05. To define modules within the lymphoid/functional compartment, cell population frequencies (% parent) were first averaged across virus (MOI = 4, A/Michigan/45/2015 (H1N1)pdm09 or A/Singapore/INFIMH-16-019/2016 (H3N2)) and peptide (peptide pools containing M1, NP and PB1, 1-5 μM per peptide) stimulation conditions to obtain the average stimulated cell frequency ('Stim Average'). Lymphoid/functional clusters were then identified by a significant positive correlation between Stim Average frequencies, using Pearson's bivariate correlation with an FDR adjustment cutoff of q ≤ 0.05. Analyses were performed using the base library in R (v.3.6.0) (ref. 92).

Statistical modeling
Univariate statistical comparisons (Supplementary Table 3) were performed independently on vaccinated and unvaccinated participants and are presented as differences between cases (symptomatic influenza) and controls (uninfected and cryptic influenza) within each group. Categorical variables and associated levels are displayed as counts and percentages across the target variable, and Fisher's exact test was used to compute P values to test the significance of the association of the variable across cases and controls. Continuous variables are represented by their median and IQR across the symptomatic and cryptic/uninfected groups, and the significance of the difference between the two groups was computed using a single-sided Kruskal-Wallis test. Statistical analyses were performed using the base library in R (v.3.6.0) (ref. 92) and sjTabone (v.0.1.0) (ref. 93). Six participants with consistent cell frequency outliers across multiple cell populations determined by Grubbs' test (P ≤ 0.05) were excluded, and a total of 200 participants were analyzed.
Univariate regression modeling was performed individually on all cellular covariates (separately for vaccinated and unvaccinated participants), on selected co-regulated cell modules, and on a subset of participants with cryptic and symptomatic infection (for serology measures). Binomial regression models were built from each cellular or serology measure or from the cluster frequency (independent variable) to evaluate the effects on symptomatic infection risk (dependent variable). Individual multivariate regression models describing the risk of symptomatic influenza, given individual serology measures adjusted for age (years), sex, BMI (kg m −2 ) and 2018 vaccination status, were also generated. Logit estimates from each GLM were transformed exponentially to obtain OR values.
For the multivariate logistic regression considering demographic, vaccination status, serology and cell immune measures, a stepwise approach was used to identify appropriate variables to include in the GLM. First, we identified highly correlated cell population clusters for both lymphoid/functional and myeloid compartments (Fig. 7b) or across serology measures (Extended Data Fig. 2a). The resulting Pearson's correlation coefficients were informative in selecting representative immune measures from each cluster to be used in statistical models. Inhibiting titers for all HAI and NAI were log 2 transformed. All serology measures and cell populations found to be influenced by vaccination status in the univariate analyses were considered as both individual covariates and interacting terms with the 2018 influenza vaccination status. As collinearity is a common problem in studies in which the underlying phenomenon that generates the observations is not fully specified, we also assessed multicollinearity across all covariates through VIF analysis. The results from the correlation and VIF analyses identified a 'selected' set of covariates that could be used in traditional regression methods without overfitting. A GLM was constructed with the final set of selected covariates against a response variable denoting symptomatic infection status. Stepwise regression models (both forward-selection and backward-elimination) were also constructed using AIC as the metric of model effectiveness. The final stepwise model had a lower AIC than the select GLM but resulted in higher residual deviance. BMA was also used to test the strength of the selected covariates when averaged over a wider range of models. Continuous dependent covariates were normalized using scaled mean and s.d. values (age, ELISA AUC, % parent cell frequency) or by log 2 transformation (HAI, NAI). Multivariate analyses were performed using the base library in R (v.3.

Univariate protection and susceptibility thresholds
ROC curves were used to test the effectiveness of each univariate immune measure-from the complete set of participants, the vaccinated subset and the unvaccinated subset-as a diagnostic indicator 95,96 . From the ROC analysis, we (1) measured trade-offs between the sensitivity and specificity of a given immune measure and (2) determined the optimal threshold achieving 50% sensitivity to generalize the behavior of the measure. The AUC was also determined for each Article https://doi.org/10.1038/s41590-023-01590-2 Extended Data Fig. 1 | Data analysis pipeline for predictive and statistical modeling. The analysis pipeline was designed to integrate participant-level demographic, serology, vaccine histories, and cellular flow cytometry data into statistical and predictive models. Univariate analyses, including Fisher's exact test, Kruskal-Wallis, logistic regression, and ROC thresholds were performed first on single, independent variables. These analyses help determine if an individual immune measure is statistically different between influenza virus infection and vaccination comparator groups (Fisher's exact test; Kruskal-Wallis), the risk of symptomatic influenza associated with an individual measure (logistic regression), and the threshold at which an individual immune measure can accurately describe 50% of symptomatic cases (ROC threshold). As univariate comparisons do not account for confounding factors, multivariate analyses were performed on combined variables including decision tree analysis (random forest) and logistic regression. The random forest allows comparison of performance (that is categorization accuracy) across models (ROC; Sensitivity & Specificity) as well as the relative importance of individual covariates within a model (VIP analyses). While random forest considers which models or individual covariates best categorize cases (symptomatic influenza) and controls (uninfected/cryptic), they do not provide information on association or risk. Multivariate generalized linear modeling (GLM) was used to determine the risk of symptomatic influenza associated with individual immune measures while accounting for the effects of others. The GLM was built on a select set of variables determined following reduction of dimensionality (correlation-based clustering) and multicollinearity (VIF) using stepwise regression (Akaike Information Criteria; AIC) and evaluated using Bayesian Model Averaging (BMA).

Reporting Summary
Nature Portfolio wishes to improve the reproducibility of the work that we publish. This form provides structure for consistency and transparency in reporting. For further information on Nature Portfolio policies, see our Editorial Policies and the Editorial Policy Checklist.

Statistics
For all statistical analyses, confirm that the following items are present in the figure legend, table legend, main text, or Methods section.

n/a Confirmed
The exact sample size (n) for each experimental group/condition, given as a discrete number and unit of measurement A statement on whether measurements were taken from distinct samples or whether the same sample was measured repeatedly The statistical test(s) used AND whether they are one-or two-sided Only common tests should be described solely by name; describe more complex techniques in the Methods section.
A description of all covariates tested A description of any assumptions or corrections, such as tests of normality and adjustment for multiple comparisons A full description of the statistical parameters including central tendency (e.g. means) or other basic estimates (e.g. regression coefficient) AND variation (e.g. standard deviation) or associated estimates of uncertainty (e.g. confidence intervals) For null hypothesis testing, the test statistic (e.g. F, t, r) with confidence intervals, effect sizes, degrees of freedom and P value noted

Software and code
Policy information about availability of computer code Data collection Flow cytometery data were collected using SpectroFlow v2.2 software (Cytek). All other biologic data were collected without software.

Data analysis
All statistical analyses were performed in R v3.6.0, including sample size assessments (WMWssp v0.4.0), univariate statistical modeling (sjTabone 0.1.0), multivariate predictive modeling (caret v6.0-92), and Bayesian model averaging (BMA v3.18.17) except calculation of area under the curve (AUC) values for ELISA, which were determined with GraphPad Prism 9 (v9.5.1 (528)) using the following parameters: baseline y=0; minimum peak height <10% of the distance from min to max Y; all peaks above baseline; 5 significant digits. Visualization in R was performed using ggplot2 v3.4.0 and ggpubr v0.5.0. Flow cytometry data were analyzed using FlowJo v10.7.1 (TreeStar). Figures were assembled using Adobe Illustrator 2023 or created with BioRender.com (exported under a paid subscription with an associated publication license). A detailed description of each analysis can be found in the "Methods" section of the manuscript. A minimum dataset containing deidentified study participant information and biological assay results along with custom study-generated R code for analysis was uploaded to GitHub (https://github.com/kvegesan-stjude/SHIVERS2) per the data sharing agreement stipulated under the NRSA-NIAID Individual Postdoctoral Fellowship award number F32AI157296 (R.C.M). Additional basic R code can be made available upon reasonable request. The published article includes all datasets generated or analyzed as a part of this study. Individual Source Data are provided with associated figures (where appropriate) per the data sharing agreement. Raw flow cytometry source files can be made available upon reasonable request.
For manuscripts utilizing custom algorithms or software that are central to the research but not yet described in published literature, software must be made available to editors and reviewers. We strongly encourage code deposition in a community repository (e.g. GitHub). See the Nature Portfolio guidelines for submitting code & software for further information.

March 2021
Data Policy information about availability of data All manuscripts must include a data availability statement. This statement should provide the following information, where applicable: -Accession codes, unique identifiers, or web links for publicly available datasets -A description of any restrictions on data availability -For clinical datasets or third party data, please ensure that the statement adheres to our policy The published article includes all datasets generated or analyzed as a part of this study. Individual Source Data are provided with associated figures (where appropriate) per the data sharing agreement stipulated under the Ruth L. Kirschstein National Research Service Award (NRSA) Individual Postdoctoral Fellowship (NRSA-NIAID) award number F32AI157296 (R.C.M). Raw flow cytometry source files can be made available upon reasonable request.

Human research participants
Policy information about studies involving human research participants and Sex and Gender in Research.
Reporting on sex and gender Participant sex (assigned at birth) is reported as a biologic attribute. Attention was paid to participant sex during the design of the study and sample selection so rough equivalent Male and Female participants were represented. Sex was considered in the correlation and univariate analyses, and as a covariate in the multivariate modeling. Results reported apply to all sexes, unless specifically stated. Information on comparator groups (vaccine and influenza infection status) by sex can be found in Figure 1c-1d,

Recruitment
During year 1 (2018) of the SHIVERS-II study, more than 22,000 individuals aged 20-69 years were randomly selected for recruitment from 3 participating primary health organizations' healthy patient networks in Wellington, New Zealand, with 2,195 ultimately enrolled following written informed consent. The study staff identified prospective adults aged 20-69 years through random selection from those healthy individuals listed in the management systems of selected primary care general practices. The study staff mailed the invitation and study information packet to those selected individuals. Those interested individuals provided their consent through online electronic consent form and also filled in an online questionnaire by providing demographic and contact information, vaccination and health and influenza-like illness status. For those consented participants, study activities included blood/swab collections. Each participant received a NZ$30 gift card after each blood or swab sample collection to recognize their time and effort. Note that full information on the approval of the study protocol must also be provided in the manuscript.

Field-specific reporting
Please select the one below that is the best fit for your research. If you are not sure, read the appropriate sections before making your selection.

Life sciences Behavioural & social sciences Ecological, evolutionary & environmental sciences
For a reference copy of the document with all sections, see nature.com/documents/nr-reporting-summary-flat.pdf

Life sciences study design
All studies must disclose on these points even when the disclosure is negative.

Sample size
A total of 206 participants were selected from year 1 of the SHIVERS-II study comprising vaccinated-uninfected (n=75), vaccinated-infected (n=33), unvaccinated-uninfected (n=76), and unvaccinated-infected (n=22). Subjects' age (in years) and sex (assigned at birth) were roughly matched when selecting across comparator groups. Sample size requirements for statistical testing of the covariates were calculated using the R package WMWssp v0.4.0 with a defined power of 0.8 and were determined to be sufficient in size for the planned comparisons.
Data exclusions Following collection of flow cytometry data we applied several blinded quality control measures for unbiased selections to identify cell populations to include in our primary analysis. These included filtering out individual samples with low cell viability (5% and below) and/or high cellular debris (>25% of events), and eliminating cell populations with a limited dynamic range in cell frequency. In the downstream statistical and modeling analyses, subjects with significant cell frequency outliers consistent across multiple cell population types were determined by Grubbs' test and excluded where indicated (n=6 removed).

March 2021
(BioLegend cat# 502936, clone Mab11, lot# B312493, dilution 1:25) ELISA secondary antibody used: anti-Human IgG (Fab specific)-HRP secondary antibody (Sigma cat #A0293, polyclonal, lot# 0000201676, dilution 1:3000) Validation All antibodies were purchased from commercial suppliers including BD, BioLegend, Tonbo, ThermoFisher, Sigma, and eBiosciences with validation data and applicable citations available on product listings for all antibodies (see individual catalog numbers). Antibodies that have previously been validated in the literature were preferred and used at specified dilutions or according to the manufacturer's specifications.

Flow Cytometry
Plots Confirm that: The axis labels state the marker and fluorochrome used (e.g. CD4-FITC).
The axis scales are clearly visible. Include numbers along axes only for bottom left plot of group (a 'group' is an analysis of identical markers).
All plots are contour plots with outliers or pseudocolor plots.
A numerical value for number of cells or percentage (with statistics) is provided.

Instrument
Cytek 3-laser Aurora spectral flow cytometer Software SpectroFlow v2.2 software (Cytek) was used to collect data, which were analyzed using FlowJo v10.7.1 (TreeStar) Cell population abundance Following collection of flow cytometry data we applied several blinded quality control measures for unbiased selections to identify cell populations to include in our primary analysis. These included filtering out individual samples with low cell viability (5% and below) and/or high cellular debris (>25% of events), and eliminating cell populations with a limited dynamic range in cell frequency. Where possible, individual cell population frequencies were compared to known frequencies in published human cohorts.

Gating strategy
The flow cytometry gating strategy to resolve cell populations within the myeloid compartment is presented in Extended data figure 3 and Supplementary Table 7. All gating applied to leukocyte-sized, single, live cells and depict frequency as % of parent gate. The flow cytometry gating strategy to resolve cell populations within the lymphoid compartment by ICS is presented in Extended data figure 4 and Supplementary Table 7. All gates applied to lymphocyte-sized, single, live cells. Gates depict frequency as % of parent gate.
Tick this box to confirm that a figure exemplifying the gating strategy is provided in the Supplementary Information.