Hemagglutinin and neuraminidase antibodies are induced in an age- and subtype- dependent manner after influenza virus infection.

Data on the immunologic responses to neuraminidase (NA) is lacking compared to what is available on hemagglutinin (HA) responses, despite growing evidence that NA immunity can be protective and broadly cross-reactive. Understanding these NA responses during natural infection is key to exploiting these properties for improving influenza vaccines. Using two community-acquired influenza cohorts, we showed that the induction of both HA and NA antibodies after infection is influenced by age and subtypes. Such response dynamics suggest the influence of immunological memory, and understanding how this process is regulated will be critical to any vaccine effort targeting NA immunity.

the cornerstone of currently recommended influenza antivirals (https://www.cdc.gov/ mmwr/pdf/rr/rr6001.pdf). Although immunity to NA limits disease severity (1)(2)(3)(4), relatively little is known about population immunity to the protein or the robustness of the antibody response against it following infection. These facets of NA immunity have been identified as crucial knowledge gaps in the campaign to develop better seasonal influenza vaccines (5)(6)(7).
The paucity of data on NA immunity has historically been impeded by the lack of an appropriate and scalable assay. The traditional NA inhibition (NAI) assay that was based on the enzymatic cleavage of sialic acids on the substrate fetuin was laborious and required the use of hazardous chemicals such as arsenite and thiobarbituric acid (8). However, the development and validation of an enzyme-linked lectin assay (ELLA) has enabled rapid and high-throughput testing of NAI antibodies (9,10), leading to renewed interest in evaluating the role of NA antibodies during influenza virus infection and vaccination (1,2,4,11,12). The ELLA has shown high subtype specificity and reproducibility when tested with antigen-specific ferret antisera (9) and higher sensitivity compared to the traditional NAI assay (13).
In this study, we explored the antibody response to HA and NA following community-acquired influenza virus infection in two cohorts enrolled through the Southern Hemisphere Influenza Vaccine Effectiveness Research Surveillance (SHIVERS) study based in Auckland, New Zealand (14) (Fig. 1). The cohorts utilized distinct sampling protocols that were representative of epidemiological and clinical studies. We examined the seroconversion frequencies to HA using the standard hemagglutination inhibition (HAI) assay and NA using ELLA in individuals with PCR-confirmed influenza virus infections. Since recent studies suggest that the HAI assay may have limited sensitivity in detecting recent influenza virus infections (15)(16)(17), we further examined the added benefit of using NAI antibody responses as an alternative serologic marker of infection. Data from this study suggest that the dynamics of antibody responses to both HA and NA after infection are influenced by age and subtype and that the use of the ELLA to detect NAI antibodies after infection can circumvent some of the current limitations associated with HAI assays.

RESULTS
Cohort characteristics. Table 1 presents the characteristics of the study participants and sampling details in the two cohorts of this study. The 50 PCR-confirmed influenza positive serosurvey cohort participants (one participant was infected with both IAV and IBV) were relatively equally distributed across the three age groups (Ͻ 5 years old [yo], 5 to 19 yo, and Ն20 yo). Twenty-one (42%) participants identified as Maori (n ϭ 7), Pacific (n ϭ 5), or of Asian (n ϭ 9) descent, and the remaining participants were of European descent (classified as "others") ( Table 1). There were equal proportions of IAV (25/51, 49%) and IBV (26/51, 51%), and both B lineages were equally represented (B/Vic, 15/51, 29%; B/Yam, 11/51, 22%). The median time from preseason serum collection to symptom onset was 134 days (approximately 4 months; range, 22 to 193 days). The median interval between the preseason and postseason serum collections was approximately 7 months (218 days; range, 132 to 274 days). Median days (range) From first serum collection to symptom onset 134 , n ϭ 39 From symptom onset to first serum collection 12 , n ϭ 58 From first serum collection to swab specimen 141 (28-202) From swab specimen to first sera collection 6 (0-23), n ϭ 93 Between serum collections 218 (132-274) 16 , n ϭ 93 a n, Number of PCR-confirmed influenza cases. For the serosurvey cohort (n ϭ 50 individuals), one subject was positive for influenza A and B. For the immunology cohort, ethnicity data were only available for 93 participants.
In the immunology cohort, 82 of 94 (87%) paired sera were collected from adults (Ն20 yo). Forty-six (49%) participants identified as Pacific (n ϭ 23), Maori (n ϭ 15), or of Asian (n ϭ 8) descent, and the remaining 47 were of European descent. Most infections were due to influenza A viruses (IAVs; 74%) or influenza B viruses (IBVs) of Yamagata lineage (20%). Because PCR testing and serum collection was conducted only following presentation with symptoms, the median time from symptom onset to first serum collection was 12 days (range, 1 to 28).
HAI and NAI-antibody responses in the serosurvey cohort. The serosurvey cohort allowed us to collect true baseline samples prior to substantial community influenza activity. HAI and NAI antibody titers were determined on pre-and postseason serum samples from individuals with laboratory-confirmed influenza during the season. For participants with IAV infections (Table 2), the baseline HAI geometric mean titers (GMTs) were low in this cohort across all age groups, whereas there were relatively small increases in baseline NAI GMTs with increasing age (the GMTs and 95% confidence intervals [95% CI] for the 0 to 4, 5 to 19, and Ն20 yo groups were 5 [5 to 5], 24 [15 to 40], and 13 [7 to 24], respectively; P ϭ 0.0005). The geometric mean fold change (GMFC) between the pre-and postseason sera decreased with increasing age for HAI (GMFC [95% CI] for the 0 to 4, 5 to 19, and Ն20 yo groups were 22 [14 to 35], 10 [8 to 12], and 4 [3 to 6], respectively; P ϭ 0.0005) but not for NAI. There were no age-dependent differences on the NAI GMFC responses. Baseline titers did not appear to influence the magnitude of fold change for the HAI or NAI antibody titers ( Fig. 2A and B).
We next looked at seroconversion events detected by each assay, as determined by a Ն4-fold increase in titer and meeting the minimum threshold titer of 40. The frequency of HAI seroconversion in IAV was lowest in adults (44% [95% CI ϭ 4 to 85%]) compared to children (86% [95% CI ϭ 51 to 121%]) and young children (100%) (P ϭ 0.0133). In contrast, the frequencies of NAI seroconversion ranged from 67 to 100% and were not statistically different by age.
HAI and NAI antibody responses in the immunology cohort. The challenge of conducting immunologic studies in medical care facilities is that infections, as well as the corresponding immune responses, are often well under way by the time patients present. Therefore, true baseline blood draws are difficult to obtain. This was likely the case in our immunology cohort, since the baseline antibody GMTs to HA and NA were higher compared to the serosurvey cohort across all age groups (Table 3). Within this cohort there were no statistical differences in the IAV baseline GMT for HAI or NAI among the different age groups, but there were age-specific differences in the GMFC responses (HAI GMFC, P ϭ 0.0014; NAI GMFC, P ϭ 0.067). Subjects Ͻ5 yo showed the highest GMFC in both assays. The frequency of HAI seroconversions was highest among young children (0 to 4 yo; 83% [95% CI ϭ 41 to 126%]) compared to adults (Ն 20 yo; 34% [95% CI ϭ 22 to 46%]), while NAI seroconversions did not show any agedependent trends. Unlike the serosurvey cohort, a large percentage of IAV cases across a broad range of baseline HAI and NAI titers failed to mount at least a 4-fold increase in antibody titers ( Fig. 4A and B). An exception was the NAI response against IBV, where nonresponders (FC Յ 2) had high baseline (Ͼ320) NAI titers ( Fig. 4C and D).

NAI antibodies as serologic markers of influenza virus infections.
Given the poor detection of HAI seroconversion in certain age groups, we sought to determine whether the NAI antibody responses could be used as an alternative serological marker to detect recent infections. We first determined the overall seroconversion frequencies in both cohorts by either assay or both assays, and we then evaluated the performance of these assays using a statistical metric.

FIG 5
Percentages of individuals that seroconverted as determined by HAI assay, NAI assay, or both HAI and NAI assays in PCR-confirmed influenza cases in the immunology cohort. Error bars represent 95% confidence intervals.
Interassay diagnostic agreement with the PCR positivity and sensitivity and the specificity of serologic assays. We used statistics to evaluate the level of agreement between PCR subtype positivity and serologic seroconversions by each assay when used alone and in combination (Table 4). Higher scores (scaled from 0 to 1) indicate a better agreement between the two methods (18 . All assays had high specificity scores (range, 0.81 to 1), although, unsurprisingly, the combination of HAI and NAI assays offered the highest specificity (0.91 to 1).
The HAI and NAI assays also detected 4 and 20% seroconversions against H1N1 strain, respectively, in the serosurvey cohort and 1 and 13% seroconversions, respectively, in the immunology cohort ( Table 5).
Specificity of assay in the control (summer) cohort. To further examine the specificity of the assays, we tested paired sera collected for the diagnosis of nonrespiratory illnesses during the months when influenza activity was not epidemic. Our rationale for this was that any seroconversions detected were more likely due to assay complications rather than true infections. The percentages reported here represent seroconversion events against either IAVs (H1 and H3 subtypes) or IBVs (Yamagata and Victoria lineages) ( Table 6). As a group, no significant rise in GMFC was detected for any virus. The frequency of a HAI-seroconversion event against IAVs, was 3% (95% CI ϭ -1 to 6%) (3/94). All of these seroconversion events were directed against H3. The frequency of NAI seroconversion against IAVs was 4% (95% CI ϭ 0 to 8%, 4/94). The frequency of NAI seroconversion against IBV was 2% (95% CI ϭ 0 to 5%, 2/94), and no HAI seroconversion events were detected. Notably, three of the five individuals seroconverted to multiple antigens. Two individuals seroconverted to both N1 and N2, and one individual seroconverted to three antigens; H3 and NA to both B lineages (Table 7).
No individual seroconverted to both HA and NA simultaneously.

DISCUSSION
Our study evaluated the induction of HA and NA antibodies in individuals with community-acquired influenza through two cohorts with a distinct sampling design. We found that the dynamics of HA and NA antibody responses were age and virus dependent after influenza virus infection. HAI seroconversion events in IAV cases were highest in young children (Յ5 yo) but then decreased with increasing age. These young children, who are likely experiencing their first influenza virus infection, were also more likely to seroconvert to both HA and NA or show an HA-dominant antibody response. a GMT, geometric mean titer; GMFC, geometric mean fold change. The percent seroconversion (%SC) was determined as the percentage of individuals that seroconverted, as defined by a 4-fold rise in antibody titer by the specified assay between the first and second sera. b NA, not applicable. The latter finding is consistent with the expectation of HA being immunodominant compared to NA in HA-naive individuals (19,20). The age-dependent increase in baseline NAI, but not HAI, titers in our cohorts suggests either an age-associated shift in antibody immunodominance or (perhaps more likely) a closer match between the NA antigen used in our assay to the prior circulating strain (i.e., a less-drifted NA antigen) (12). This could explain the NA-only seroconversions seen in IAV-infected adults. The presence of NA-specific memory B cells could favor the NA-specific antibody responses over HA in these adults, a model previously proposed to explain the "damping" of NA antibody responses in HA-primed individuals (21)(22)(23). Several studies have reported that the elderly subjects may have a more NA-biased antibody response compared to the younger subjects and that evidence of original antigenic sin can be observed for NA antibody responses as well (12,24,25). Based on these observations, it appears that NA antibodies to strains encountered early in life may persist and, like HA antibodies, can be "back-boosted" (26). It is therefore likely that the relative conservation of this antigen can further contribute to boost the preexisting NA antibody response in the adults. Compared to the NAI assay, the HAI assay was surprisingly insensitive in detecting seroconversions against IBVs. This was also observed in our larger serosurvey cohort comprising of the PCR-negative cases (n ϭ 701), where the proportion of IBV NAI seroconverters was higher than the HAI seroconverters, particularly in young children (27). This is in contrast to the findings by Rajendran et al., who reported lower postvaccination NAI titers to IBV in children compared to adults (12), suggesting that, like IAV, there are differences in the NA antibody response after IBV infection versus vaccination (28). Whether the reduced IBV HAI sensitivity seen in our study was due to technical reasons (i.e., a mismatched antigen or the influence of egg-adapted antigens) or antigenic competition, we found that the NAI assay compensated for these limitations, since 100% of the IBV cases were captured.
That the ELLA was particularly sensitive for IBV suggests that NA antibodies against IBVs could be more broadly cross-reactive than HA antibodies, as had been reported recently for IAV (28). However, we cannot exclude the possibility of nonspecific inhibition of NA activity by HA antibodies, particularly those that bind outside of the globular head which were not detected by HAI assays (since most of our participants had very low preexisting HAI titers to IBVs) (29,30). How many of these antibodies are present and thus the extent of interference in polyclonal serum are unknown. The two sampling designs used in the serosurvey and immunology cohort are typical of most seroepidemiology and clinical studies. The serosurvey cohort utilized a pre-and post-influenza season sampling design. This allowed for baseline sera to be collected prior to the actual infection, although with such a design, a large initial cohort is required in order to capture sufficient infected cases during the influenza season. In contrast, the immunology cohort sampled individuals that actively sought health care. Correspondingly, there was a large variability in the time of the first sera sampled after symptom onset in this cohort, attributed in part to the logistical delay in sampling the influenza-like illness (ILI) cases that present to the general practitioners across Auckland. This likely contributed to the high baseline titers, thus decreasing the likelihood of meeting the seroconversion criteria (31). Indeed, when we examined in detail the sampling times broken down by age (Table 8), we found that the older-children group (5 to 19 yo), which had the highest baseline HAI titer in the immunology cohort, was also the group that had the largest time interval between symptom onset and first serum collection. This is likely associated with health-seeking behavior since older children are less likely to present with severe influenza. We attempted to correct for the variable sampling time in our analysis by excluding the outliers (i.e., those sampled Ͻ14 days or Ͼ20 days), but this resulted in the loss of statistical power without affecting the population average (data not shown). Hence, we kept the original analysis for this study. Nevertheless, despite the difference detected in the frequency of seroconversions, data from both cohorts, at least for IAV cases, suggested that adults were more likely than children to show discordant HA and NA antibody induction, confirming previous observations (32,33).
Despite H3N2 being the dominant circulating strain during our study, some seroconversions to H1N1 were also detected in our two cohorts. Seroconversions to NA were more frequent compared to HA, potentially due to the higher cross-reactivity of NAI antibodies or to coinfection events during the influenza season. For this reason, we used the summer cohort (where sera were collected for nonrespiratory disease testing when community influenza activity is low) as a proxy for an "influenza-negative" population to evaluate the specificity of these assays. Since three of the five paired sera showed seroconversion to multiple antigens (Table 7), it is likely that these are false-positive events. With the caveat that we could not discount the possibility of any "out-of-season" influenza cases, the false-positive frequencies were between 0 and 4% of the total reactions tested and appeared to be comparable between HAI and NAI assays. Principally, we found that including the NAI assay improved the serological detection of influenza cases in both our cohorts, although we acknowledge that the detection sensitivities of the HAI and NAI assays reported here may vary during different influenza seasons with different antigens. With further validation across different influenza seasons and antigens, including the NAI assay as a hierarchical or targeted testing approach can increase the power and accuracy of a seroepidemiological study and will circumvent some of the limitations and variable robustness associated with HAI assay.
Overall, our findings highlight the age and subtype-dependent nature of HA and NA antibody responses after PCR-confirmed influenza virus infections. The differences in the antibody response to HA and NA that we observed between adults and children suggests a possible influence of immunological memory on the recall response. This is an important consideration for any vaccination strategy aimed at eliciting NA immunity since it suggests that the antibody response may differ with different age groups. The mechanism underlying such response dynamics should be further investigated.

MATERIALS AND METHODS
Ethics approval. This study received ethics approval from the New Zealand Northern (A) Health and Disability Ethics Committee under references NTX/11/11/102/AM02, AM05, AM06, AM13, and AM14.
Study design. The sampling designs for this study are summarized in Fig. 1. For the serosurvey cohort, randomly selected community participants enrolled at SHIVERS participating general practices provided pre-influenza season sera in March to June and post-season sera in October to November of 2015. During the influenza season (May to September), nasopharyngeal swabs from 209 unvaccinated participants with influenza-like illness (ILI; defined as cough and fever with onset Յ10 days) were tested by PCR for evidence of influenza virus infection. Fifty-one infections (24% positivity) were identified, and the sera from these individuals were analyzed for this study. The immunology cohort was recruited in the 2013 and 2015 seasons and consisted of 94 patients presenting to general practice clinics (n ϭ 29) with acute respiratory illness or to hospitals with severe acute respiratory illness (n ϭ 65) based on the World Health Organization disease classification (34). For this cohort, the first blood draw occurred following a positive influenza PCR test, and the second occurred at least 14 days later. Study data were collected and managed using Research Electronic Data Capture (REDCap) tools (35) hosted at the Institute for Environmental Science and Research (ESR). To examine the specificity of the assays, a panel comprising of 47 paired sera, collected approximately14 days apart during the summer of 2016 for arbovirus/ legionella testing at the ESR, was also evaluated.
Serologic testing. Serologic testing was performed at the National Influenza Center at ESR, New Zealand. Receptor-destroying enzyme (RDE)-treated sera were tested against the strains included in the Southern hemisphere vaccine for the relevant enrollment year (A/H3, A/H1, and the two B lineages; Table  9) according to standard practice (40). The ELLA was used to test for NAI-specific antibodies, as previously described (9,11,41). Recombinant influenza A viruses (IAVs) composed of NA from the viruses to be tested and a mismatched HA from A/Teal/Hong Kong/W312/1997 (H6N1) were generated using the reverse-genetics method and used as antigens. The NAs used in recombinant viruses were N1 from  influenza viruses, and the percentages of seroconversion events against IAV (A/H3 and A/H1) and IBV (B/Yam and B/Vic) were reported. Statistical analysis. Seroconversion was defined as a 4-fold increase between the first and second HAI or NAI titer against the subtype of the infecting virus. In addition, if the first titer was below the detection threshold (Ͻ10), the second titer had to be Ն40. IAV cases that were not subtyped were assumed to be A/H3N2, since it was the predominant subtype circulating in both seasons of the study. Non-lineage-typed IBVs were assumed to be the B antigens with the highest increase in titer. Geometric mean titers (GMTs) were reported as the back-transformed average of the log 2 antibody titers, and the geometric mean fold change (GMFC) was reported as the back-transformed average of the differences between the first and second log 2 antibody titers (41). Age-specific effects on the antibody responses were analyzed by using analysis of variance (ANOVA) tests.
Kappa () statistics were used to describe the level of agreement between PCR diagnosis and seroconversion status by HAI only, by NAI only, by both HAI and NAI, and by either HAI or NAI. The strength of agreement between the two assays was categorized based on the Landis and Koch kappa benchmark scale (23). Test sensitivity and specificity scores were also calculated for each assay or combination of assays. For the serosurvey cohort, the negative cases were the ILI individuals that were PCR negative during the season (n ϭ 209). For the immunology cohort, the negative cases were derived from the summer cohort. For IAV, we considered PCR subtype positivity as the gold standard outcome, and we used the homologous subtype seroconversion event as the test to ascertain agreement. For IBV, we only considered PCR type and any flu B serotype seroconversion for agreement. All calculations were generated by using the R package epiR (42,43).