School-aged children (aged 5–17 years) were enrolled from September to November 2015 in a serologic study in Marshfield, Wisconsin (Figure 1; Supplementary Table 1) [9]. Children included in the antibody landscape analysis (n = 72; age 7–17 years [median age, 13 years]) received 2015–2016 trivalent inactivated influenza vaccine (IIV3, Fluzone, Sanofi Pasteur). Of these, 41 were previously enrolled in 2014–2015 and received either 2014–2015 IIV3 (Sanofi Pasteur) or quadrivalent live attenuated influenza vaccine (LAIV4, FluMist, MedImmune). Children were grouped based on their 2014–2015 A(H3N2) infection and vaccination status: 12 were vaccinated (9 received LAIV and 3 received IIV) and had laboratory-confirmed A(H3N2) infection during the 2014–2015 season (VI group; age 7–13 years [median age, 10.5 years]), 29 were vaccinated and uninfected with A(H3N2) in 2014–2015 (VU group; age 7–17 years [median age, 12 years]), and 31 children were newly enrolled in 2015–2016 from the Marshfield Epidemiologic Study Area who were unvaccinated and uninfected with no documented influenza infection nor medically attended acute respiratory illness in the prior 5 seasons (UU group; age 9–17 years [median age, 13 years]) (Supplementary Table 1) [9]. In 2015–2016, none of the participants had A(H3N2) infection identified during active surveillance. Sera collected prevaccination and 21–28 days postvaccination in 2015–2016 from all children and in 2014–2015 from those enrolled in both seasons were tested.

Written informed consent was obtained from parents/guardians of the children, and assent, when applicable. This study was approved by the institutional review boards of the Centers for Disease Control and Prevention and the Marshfield Research Clinic Institute.

Vaccine responses were measured by hemagglutination inhibition (HI) assays using 0.5% turkey erythrocytes as previously described [12]. Serum samples were treated with receptor-destroying enzyme and preabsorbed with packed turkey erythrocytes to remove nonspecific agglutinins as needed. Serial 2-fold dilutions of sera were made from an initial 1:20 dilution. The HI titer was defined as the reciprocal of the last dilution of serum that completely inhibited hemagglutination. At least 2 independent experiments were performed for all time points from the same individual with each sample run in duplicates.

All viruses were propagated in 9- to 11-day-old embryonic chicken eggs and sequenced (Supplementary Table 2).

For each individual, antibody landscapes were constructed for each time point using antibody responses against 16 A(H3N2) viruses representative of antigenic clusters that circulated between 1968 and 2016: A/Aichi/2/1968 (Aichi/1968), A/Victoria/03/1975 (Victoria/1975), A/Bangkok/1/1979 (Bangkok/1979), A/Shanghai/11/1987(Shanghai/1987), A/Beijing/353/1989 (Beijing/1989), A/Beijing/32/1992 (Beijing/1992), A/Wuhan/359/1995 (Wuhan/1995), A/Sydney/05/1997 (Sydney/1997), A/Panama/2007/1999 (Panama/1999), A/Fujian/411/2002 (Fujian/2002), A/California/07/2004 (California/2004), A/Wisconsin/67/2005 (Wisconsin/2005), A/Brisbane/10/2007 (Brisbane/2007), A/Perth/16/2009 (Perth/2009), A/Texas /50/2012 (Texas/2012), and A/Switzerland/9715293/2013 (Switzerland/2013).

To define likely A(H3N2) immune priming, children were grouped into cohorts using 2 approaches. In the first approach, we assumed that the year of birth directly implied priming with a virus that circulated during that year; thus, participants were grouped by birth years as “birth cohorts.” To achieve sufficient numbers per birth cohort, 2 birth years were grouped as 1 “birth cohort,” except in 2006–2008, where 3 birth years were grouped as 1 birth cohort. To construct antibody landscapes for each birth cohort, we plotted geometric mean titers (GMTs) against viruses arranged chronologically by the year of isolation.

In the second approach, we determined likely priming virus based on an individual’s postvaccination antibody landscape without assuming that initial influenza exposure occurred in the first years of life [13–15], and grouped these children into “priming cohorts.” Likely A(H3N2) priming virus was assumed to be antigenically similar to the earliest reference strain that circulated after a child’s birth with titer ≥40 post 2015–2016 vaccination (Supplementary Tables 3 and 4). Circulation periods for historical A(H3N2) viruses were estimated based on data published for influenza surveillance (Supplementary Table 3). Following the method of Fonville et al [8], we generated an antigenic cartographic map using ferret antisera titers against the 16 A(H3N2) viruses and computed antigenic distance along a “summary path” using R version 3.6.1 software (Supplementary Methods and Supplementary Figure 1). Log-transformed postvaccination titers (log₂ [titer / 5]) were plotted against 16 A(H3N2) reference viruses arranged on the horizontal axis based on the antigenic distance measured from the antigenic map.

GMTs of replicates were reported as final titers. Titers <20 (predilution of sera) were assigned as 10. GMT fold-rise was calculated as the ratio of postvaccination to prevaccination titers. Seroconversion was defined as ≥4-fold rise in antibody titers with postvaccination titers ≥40. Comparison between study groups were analyzed by 2-tailed t test or 1-way analysis of variance, and difference in proportions was compared using Fisher exact test. Multivariate linear regression with log₂-transformed titers or fold-rise in titers was used to examine associations between pre- and postvaccination or fold-rise in titers with age and probable priming viruses in both seasons. Statistical analyses were conducted using GraphPad Prism 8.1.1 and SAS 9.4 software.

Antibody landscapes in children against 16 A(H3N2) viruses representative of major antigenic clusters that circulated from 1968 to the study period 2016 were compared before and after 2014–2015 and 2015–2016 influenza vaccinations (Figure 2). Vaccination against either Texas/2012 or Switzerland/2013 not only induced antibody rises to the vaccine viruses, it also boosted responses to older viruses, including some that circulated prior to participants’ birth (Figure 2 and Supplementary Figures 2 and 3). Antibody responses in children were broader than those detected by primary infection ferret antisera against the vaccine viruses, which only recognized viruses that circulated in adjacent seasons (Supplementary Table 5).

To investigate the effect of immune priming on vaccine responses, we first grouped the children into 5 birth cohorts that were from birth years 1998–1999, 2000–2001, 2002–2003, 2004–2005, and 2006–2008. At prevaccination, the breadth of antibody landscapes, or the number of viruses with HI titers ≥40, increased with the age of the birth cohorts (Figure 3 and Supplementary Figure 4A). Children in the same birth cohort, regardless of their prior season vaccination status, had similar shapes of antibody landscapes before vaccination. At baseline, children in the UU group without documented vaccination and infection in the past 5 seasons (n = 31) had lower titers but similar shapes of landscapes to older viruses compared with children who received vaccine in the prior season (VI and VU groups, n = 41), indicating that these children were also exposed to A(H3N2) viruses early in life (Figure 3). Following 2015–2016 vaccination, children in older birth cohorts induced broader antibody responses (back-boost) to earlier viruses; the number of postvaccination titers ≥80 increased with birth cohort years (Supplementary Figure 4B). Of note, young children in the 2006–2008 birth cohort, all of whom received LAIV in 2014–2015, had no detectable HI antibody responses in 2014–2015 (Figure 3).

Next, we used each child’s antibody landscape to historic viruses and defined their likely immune priming virus [13]. Antigenic characterization (Supplementary Table 5) and antigenic cartographic map using ferret antisera (Figure 4A) illustrated the gradual and frequent antigenic drift of A(H3N2) viruses over the past 5 decades. Among the children received vaccination in both seasons, we identified 5 priming cohorts. These children were likely primed with Sydney/1997-like, Panama/1999-like, Fujian/2002-like, California/2004-like, and Wisconsin/2005-like A(H3N2) viruses earlier in their life. We constructed antibody cartographic landscapes for these 5 priming cohorts in both seasons (Figure 4B). Similar to the birth cohorts, children primed with older viruses had broader breadth of antibody landscapes at baseline. After vaccination, landscapes based on antigenic distance retained their shapes over 2 seasons within each priming cohort (Figure 4B).

The 2 approaches to identify likely immune priming yielded similar results. In each “birth cohort” or “priming cohort” (Figures 3 and 4), antibody responses were higher to the current vaccine virus and closely related contemporary viruses, rather than to the priming virus or other older viruses they may have been exposed to in the past. In 2015–2016, fold-rise in titers to the likely priming virus was significantly associated with the fold-rise to the vaccine virus Switzerland/2013 (P < .001) (Table 1, model 5). However, in the multivariate model, prevaccination titers to the likely priming virus were not significant predictors of responses to vaccine viruses (Table 1, models 1–4). Pairwise comparison of antibody responses to the vaccine viruses in each season among priming cohorts revealed no significant difference (P > .05) except that, in the 2014–2015 season, children likely primed with Wisconsin/2005-like viruses (n = 4) had lower responses to Texas/2012 compared to other priming cohorts (Supplementary Tables 6 and 7). Three of the 4 children in this priming cohort received LAIV; lower antibody responses were likely due to the vaccine type received [16], rather than immune priming.

A(H3N2) vaccine antigen was updated from Texas/2012 (3C.1) in 2014–2015 to Switzerland/2013 (3C.3a) in 2015–2016 due to antigenic drift of the circulating viruses. Interestingly, among children enrolled in both seasons (VI and VU groups), following consecutive vaccination with these 2 antigenically distinct antigens, antibody landscape patterns generally remained similar between seasons (Figure 2A and 2B); similar landscape patterns were also retained within each priming cohort (Figure 3A and 4B). Antibody waning was evident between the postvaccination landscape in 2014–2015 and prevaccination in 2015–2016 (Figures 2–4). In children vaccinated in both seasons, there was a slight increase from 2014–2015 to 2015–2016 prevaccination landscape titers (Figure 5A).

When stratified by prior season vaccination status, at baseline, children unvaccinated and uninfected in the past 5 seasons (UU group, n = 31) had lower 2015–2016 prevaccination landscape titers than those who received 2014–2015 IIV3 (IIV/IIV, n = 29), including titers to the vaccine virus Switzerland/2013 and several viruses that circulated in the earlier influenza seasons (Texas/2012, Perth/2009, Wisconsin/2005, and California/2004; P < .05, Figure 5B). However following vaccination with Switzerland/2013, their landscape titers were largely similar (P > .05) for all viruses between the 2 groups except for 1 virus (Brisbane/2007) (Figure 5B) where the UU group showed significantly higher titers (P < .05). The magnitude of the titers increased but the patterns of the landscapes were retained between pre- and postvaccination in each group. Children in the UU group had significantly higher antibody fold-rise to the vaccine virus Switzerland/2013 (Supplementary Figures 3 and 5) and to the other earlier A(H3N2) viruses compared to children vaccinated in 2014–2015 (IIV/IIV group in Figure 5C and Supplementary Figure 3). Furthermore, in each group, the highest fold-rise was to the vaccine virus Switzerland/2013, rather than to any of the historic A(H3N2) viruses (Figure 5C). Following vaccination in 2015–2016, a greater proportion (94%) of children in the UU group seroconverted to Switzerland/2013 compared with children who received IIV3 in the prior season (63%; P = .0052).

In multivariate linear regression models to analyze the predictors for vaccine responses in each season, in 2015–2016, prevaccination titers against vaccine virus Switzerland/2013 were the strongest predictors of postvaccination titers to Switzerland/2013 (P < .001; Table 1, model 4). In 2014–2015, prevaccination titers to Texas/2012 were positively correlated with postvaccination titers to Texas/2012 and negatively correlated with fold-rise to Texas/2012 (P < .05; Table 1, models 1 and 2).

Among the 12 children who had reverse-transcription polymerase chain reaction–confirmed infection with A(H3N2) viruses in 2014–2015 (VI group), 9 had received LAIV and 3 received IIV in 2014–2015 (Supplementary Table 1). Circulating A(H3N2) viruses in 2014–2015 were antigenically drifted from the vaccine virus Texas/2012 but were antigenically similar to the updated vaccine strain Switzerland/2013 (3C.3a) in 2015–2016 [16]. Infection with A(H3N2) in 2014–2015 resulted in elevated titers across the landscape in 2015–2016 prevaccination sera, consistent with convalescent response following natural infection. Subsequent vaccination in 2015–2016 further boosted landscape titers (Figure 6A). Conversely, in vaccinated but uninfected children (VU group, n = 29), antibody waned across the landscape from postvaccination in 2014–2015 to prevaccination in 2015–2016, followed by a substantial increase of landscape titers postvaccination with Switzerland/2013 in 2015–2016 (Figure 6B).

Children infected with A(H3N2) after receiving Texas/2012 in 2014–2015 had lower postvaccination titers cross-reactive to Switzerland/2013 compared with those who were vaccinated but not infected (P = .004; Supplementary Figure 6A). Postvaccination (preinfection) antibody landscapes among vaccine failure cases (VI) were narrower in breadth and lower in magnitude than those from the vaccinated but uninfected children (VU). Here, postvaccination titers to all 9 recent viruses on the landscape from Sydney/1997 to Switzerland/2013 were significantly lower (P < .05) in those who became infected with A(H3N2) than those who were uninfected in 2014–2015 (Figure 5C).

In the following 2015–2016 season, postvaccination antibody landscapes did not differ significantly among the 3 groups of children (Figure 6D; VI, VU, and UU groups; P > .05). Children in the UU group had higher fold-rise in titers to most viruses than other groups, likely due to lower prevaccination titers (Supplementary Figure 6B).