Protein Microarray Analysis of the Specificity and Cross-Reactivity of Influenza Virus Hemagglutinin-Specific Antibodies

Seasonal influenza is a serious public health problem because the viral infection spreads easily from person to person and because of antigenic drift in neutralizing epitopes. Influenza vaccination is the most effective way to prevent the disease, although challenging because of the constant evolution of influenza virus subtypes. Our high-throughput protein microarrays allow for interrogation of subunit-specific IgG and IgA responses to 283 different HA proteins comprised of HA1 and HA2 domains as well as full-length HA proteins. This provides a tool that allows for novel insights into the response to exposure to influenza virus antigens. Data generated with our technology will enhance our understanding of the factors that improve the strength, breadth, and durability of vaccine-mediated immune responses and develop more effective vaccines.

that improve the strength, breadth, and durability of vaccine-mediated immune responses and develop more effective vaccines. KEYWORDS hemagglutinin, influenza, protein microarrays D espite the availability of seasonal vaccination, influenza remains a serious cause of morbidity and mortality worldwide. Seasonal influenza virus epidemics result in 290,000 to 650,000 deaths per year (1). These annual outbreaks are sustained by the amino acid substitutions in the antigenic sites of the HA1 subunit, which enable viruses to escape recognition of protective antibodies generated previously and drive antigenic drift (2)(3)(4). In addition to seasonal influenza virus outbreaks, pandemics occur at irregular intervals. The recent 2009 influenza A H1N1 pandemic resulted in Ͼ18,449 reported deaths (5,6), although the actual number is thought to be considerably higher (7). Moreover, the potential for mortality rates in excess of a million, as seen sporadically over the past 100 years (8), remains a constant threat. Large pandemics are usually associated with the appearance of novel influenza virus subtypes in the human population. These novel subtypes originate from animal populations, and the pandemic viruses are often generated by reassortment of genetic segments between human and animal, typically avian or swine, influenza viruses (antigenic shift) (9). Highly pathogenic avian H5N1 and H7N9 viruses have caused zoonotic infections and have undergone genetic mutations and reassortment and, therefore, are considered high risk for public health. The potential for the emergence of pandemic H5N1 and H7N9 avian influenza has prompted the development of the U.S. National Prepandemic Influenza Vaccine Stockpile (10).
Influenza virus hemagglutinin (HA) proteins are divided into two phylogenetic groups: group 1 (encompassing the subtypes H1, H2, H5, H6, H8, H9, H11, H12, H13, H16, H17, and H18) and group 2 (encompassing the subtypes H3, H4, H7, H10, H14, and H15). The stalk domains of group 1 proteins share similar structures, as do the stalk domains of group 2 proteins (11). Protection against influenza virus infections is predominantly mediated by antibodies against the HA molecule on the virion surface. The HA binds to sialic acids in membrane glycoproteins and glycolipids on host cells, and antibodies are able to inhibit this interaction. Immunological and structural studies have revealed 4 or 5 important antigenic sites on the exposed head domain of the HA molecule, mutations within each of which are thought to promote antigenic drift and escape from a preexisting polyclonal response (2,3,(12)(13)(14).
The majority of current seasonal vaccines are produced from inactivated virions of the strain(s) predicted to be prevalent in advance of the influenza season. Antigenic drift causes seasonal protection to be short-lived, requiring frequent updating of vaccine antigens and vaccine readministration. There is thus considerable interest in the development of universal influenza virus vaccines (15)(16)(17), designed to elicit cross-reactive or broadly neutralizing antibodies (bnAbs), that would reduce the need for annual vaccination and also allow a broadly protective prepandemic vaccine to be stockpiled. The presence of bnAbs in humans against influenza virus glycoproteins was not fully appreciated until the 2009 H1N1 pandemic (18), thereby raising hopes that such a vaccine is possible. bnAbs target conserved structures on the virus, notably in the HA stalk (19)(20)(21), the receptor binding pocket in the HA head (19,22), and surface neuraminidase (NA) (23,24).
Recent advances in producing bnAbs from individual human B cells are providing a clearer understanding of immunity to influenza virus (25) as well as other important viruses (26)(27)(28). However, the development of high-throughput tools to measure cross-reactivity of polyclonal serum lags behind, particularly to antigens that show high divergence, such as H1, H3, N1, and N2 of influenza virus. A number of protein microarrays have been produced previously to address this (29)(30)(31)(32)(33). However, these arrays had fewer numbers of HA proteins and coverage of fewer virus subtypes than our current high-density microarray, which comprises 283 purified HA variant proteins derived from 17 influenza A virus subtypes (H1 to H16 and H18) and influenza B virus strains. For H5, the array included both the American and the Eurasian lineages. Importantly, most H5 proteins on the array were derived from the Asian A/goose/ Guangdong/1996 lineage and included diverse clade 0, 1, 2.1.3, 2.2, 2.3.2.1, 2.3.4, 2.5, 3, 4, 5, and 7 isolates. Another focus of this study was analysis of sera from H3N2-infected individuals. For a good resolution of this response, we included H3 HAs from the 1968 pandemic, modern and historic vaccine strains, and currently circulating strains, as well as several swine and equine H3 HAs. HA molecules were expressed as HA1 only or as full-length HA1ϩHA2 molecules. These arrays were probed with available sera from naturally infected patients with laboratory-confirmed influenza virus infection during the 2010 to 2011 influenza season or from individuals boosted with an H5N1 vaccine. Distinct reactivity patterns were detected after H3N2 natural infection and after H5N1 booster vaccination. The data presented here demonstrate the potential of this assay to aid epidemiological surveys, guide vaccine development, and extend our understanding of the antibody response against influenza viruses.

RESULTS
The antibody response against HA1 proteins is subtype specific after boosting with an H5N1 vaccine and during H3N2 natural exposure. (i) Boosting with an H5N1 vaccine generates a robust subtype-specific IgG but no IgA response to HA1 proteins. In a previous clinical trial (clinicaltrials.gov ID NCT00680069), the safety and efficacy of administering different clades of H5N1 at prime and at boost were investigated (34). In a substudy of that trial, subjects were primed twice with either lowdosage or high-dosage nonadjuvanted H5N1 vaccine derived from A/Vietnam/1203/ 2004 (clade 1) and were boosted with a single intramuscular high dosage of a different H5N1 vaccine derived from the antigenically distinct A/Indonesia/05/05 (clade 2) virus more than 1 year later (see Fig. S1 in the supplemental material). The serum samples used in our microarray study were obtained from 25 subjects, 10 of whom were given the high-dosage primary series, and the remaining 15 the low-dosage primary series. All 25 subjects were eventually boosted with a high-dose boost. Data presented in Fig. 1 to 4 were obtained from the high-dosage primary series group only. Sera (n ϭ 10) from individuals who received the high-dosage primary series were first interrogated here for IgG and IgA responses at two time points-preboost (d0) and 28 days postboost (d28)-using the microarray of hemagglutinin proteins expressed as HA1 or HA1ϩHA2 molecules. Figure 1 shows data using HA1 proteins for antibody detection in the vaccine study. H5-boosted individuals exhibit a strong IgG response to nearly all H5 variants on the HA1 panel in addition to the administered strains (Fig. 1A). This response is subtype specific as demonstrated by a nearly 3-fold increase in average signal intensity on d28 compared to d0 in the H5 subtype (P Ͻ 0.0001) but not in other subtypes ( Fig. 1A and B; Fig. S2A). Analysis of the H5 response by clade (Fig. 1C) revealed broad reactivity across all the clades. This breadth is likely a consequence of the clade 1 prime followed by a clade 2 boost (35,36). In addition, there seems to be a durable IgG response to H5 at least 1 year after the primary vaccination series, as illustrated by the higher preboost H5 signal intensities in vaccinees (blue circles) than in those seen in individuals with no known history of H5 vaccination (orange area). This difference is not apparent in other subtypes.
In contrast to the IgG response, the overall serum IgA response to the HA1 panel is low at both time points (Fig. 1D). While the IgA response to H5 subtypes between the two time points is statistically significant ( Fig. 1E; P ϭ 0.0005), the magnitude of the difference is small (Fig. S2B). This is entirely consistent with an intramuscular route of entry for the vaccine compared to a mucosal route of entry seen in natural influenza virus exposure. Interestingly, the average d0 IgA response in vaccinees (blue circles) is higher than that seen in the reference group (orange area), which might suggest long-lived serum IgA against H5 HA1 proteins that persist at least 1 year after a two-prime H5N1 vaccination series. A larger study is needed to confirm these observed baseline differences.  Anti-IgG (A, B, and C) and anti-IgA (D and E) antibody responses after boosting with H5 vaccine are depicted as floating bar graphs or as Tukey box plots. Bar graphs are sorted by subtype and decreasing signal intensity (SI) at d28. Each bar represents the difference between the average signal intensities of two time points for a strain (blue dots, d0 SI; red dots, d28 SI; error bars, standard deviations [SD], n ϭ 10). Dark bars indicate a positive difference between the two time points (d28 Ͼ d0), while light bars indicate a negative difference (d0 Ͼ d28). A heat map of delta values (d28 Ϫ d0) is shown above each graph. The orange area represents the signal intensity distribution (average Ϯ 1 SD, n ϭ 11) of a reference group (GCRC) with no active influenza virus infection or history of H5 vaccination. The Tukey box plots in panels B and E show H5 alone and pooled group 1 or group 2 subtype signal intensities. Panel C shows average signal intensities at the two time points among the various H5 clades. Means are indicated by "ϩ." The two-tailed Mann-Whitney test for unpaired data was used to calculate statistical significance between the two time points, where P Ͻ 0.05 defines statistical significance. HA1, hemagglutinin head domain; d0, day 0; d28, day 28; GCRC, General Clinical Research Center (UC Irvine).
Serum samples from infected patients display the greatest increase in both IgG and IgA responses to HA1 proteins of the H3 subtype compared to other subtypes (Fig. 2). It is noteworthy that all H3 strains on the HA1 panel are reactive, indicating there is cross-reactivity within a given subtype. This could be due to the redundancy of epitope coverage provided in the polyclonal anti-HA1 response and/or the extent of epitope conservation within an HA subtype. Interestingly, there appear to be some H3 strains that react more strongly, on average, to healthy reference serum IgG (orange area), but weakly to acute-phase serum IgG (t1, blue circles) ( Fig. 2A). The reference group, though healthy, is expected to have IgG antibodies to H3 proteins due to the very high likelihood that these individuals had past infections with H3; H3N2 is a commonly encountered circulating strain and a component of seasonal vaccines. In addition, there seem to be some cross-reactive IgG responses to other subtypes, including H1, H9, and H10, which are not observed in the vaccine group (Fig. 1A). Though statistically significant, the magnitude of the seroconversion among these non-H3 subtypes, indicated by the length of the bars, is far lower than that seen in H3 strains ( Fig. 2B;  Fig. S2C).
The robust subtype-specific IgA response to the HA1 panel seen in this cohort ( Fig. 2C and D; Fig. S2D) contrasts with the very low IgA response observed in the vaccine group. This is consistent with mucosal exposure, which is the presumed route through which these patients were infected. The average patient IgA response to some H3 strains, curiously, is higher at convalescence (t2, red circles) than seen in the healthy FIG 2 Antibody reactivity against HA1 proteins during natural exposure. Anti-IgG (A and B) and anti-IgA (C and D) antibody responses during H3-confirmed natural exposure are depicted as floating bar graphs or as Tukey box plots. Bar graphs are sorted by subtype and decreasing signal intensity (SI) at t2. Each bar represents the difference between the average signal intensities of two time points for a strain (blue dots, t1 SI; red dots, t2 SI; error bars ϭ SD, n ϭ 5). Dark bars indicate a positive difference between the two time points (t2 Ͼ t1), while light bars indicate a negative difference (t1 Ͼ t2). A heat map of delta values (t2 Ϫ t1) is shown above each graph. The orange area represents the signal intensity distribution (average Ϯ 1 SD, n ϭ 11) of a reference group (GCRC) with no active influenza virus infection or history of H5 vaccination. The Tukey box plots show H3 alone and pooled group 1 or group 2 subtype signal intensities. Means are indicated by "ϩ." The two-tailed Mann-Whitney test for unpaired data was used to calculate statistical significance between the two time points, where P Ͻ 0.05 defines statistical significance. HA1, hemagglutinin head domain; t1, time point 1; t2, time point 2; GCRC, General Clinical Research Center (UC Irvine).
Influenza Hemagglutinin Protein Microarrays reference group (orange area) (Fig. 2C). A study consisting of a much larger cohort of patients with acute-and convalescent-phase influenza virus infections may confirm these preliminary findings.
The antibody response against whole HA proteins (HA1؉HA2) is subtype specific after boosting with an H5N1 vaccine but appears cross-reactive in H3N2 natural exposure. (i) Boosting with an H5N1 vaccine generates a subtype-specific IgG response to HA1؉HA2 proteins. The IgG and IgA response profiles of serum samples from H5-boosted vaccinees to the HA1ϩHA2 panel are similar to those found against the HA1 panel, where a subtype-specific IgG response is detected, but not IgA ( Fig. 3; Fig. S2A and B). The magnitude of seroconversion to IgG by d28, indicated by FIG 3 Antibody reactivity against whole HA (HA1ϩHA2) proteins after vaccination. Anti-IgG (A, B, and C) and anti-IgA (D and E) antibody responses after boosting with H5 vaccine are depicted as floating bar graphs or as Tukey box plots. Bar graphs are sorted by subtype and decreasing signal intensity (SI) at d28. Each bar represents the difference between the average signal intensities of two time points for a strain (blue dots, d0 SI; red dots, d28 SI; error bars, SD, n ϭ 10). Dark bars indicate a positive difference between the two time points (d28 Ͼ d0), while light bars indicate a negative difference (d0 Ͼ d28). A heat map of delta values (d28 Ϫ d0) is shown above each graph. The orange area represents the signal intensity distribution (average Ϯ 1 SD, n ϭ 11) of a reference group (GCRC) with no active influenza virus infection or history of H5 vaccination. The Tukey box plots in panels B and E show H5 alone and pooled group 1 or group 2 subtype signal intensities. Panel C shows average signal intensities at the two time points among the various H5 clades; clade 2.5 is not represented in this data set. Means are indicated by "ϩ." The two-tailed Mann-Whitney test for unpaired data was used to calculate statistical significance between the two time points, where P Ͻ 0.05 defines statistical significance. HA1, hemagglutinin head domain; d0, day 0; d28, day 28; GCRC, General Clinical Research Center (UC Irvine). the length of the bars, however, is lower than that seen with HA1 ( Fig. 3A and B;  Fig. S2A). Similar to the reactivity against the HA1 panel, the average baseline IgG response to H5 subtypes on the HA1ϩHA2 panel (blue circles) is higher than the average response of the reference group (orange area) (Fig. 3A). This may indicate long-lived anti-H5 antibodies induced by the priming vaccination. Of note, unlike signals against the HA1 panel, IgG signals against HA1ϩHA2 from all subtypes were already elevated at the first time point (Fig. 3B versus Fig. 1B). This was also observed to a lesser degree in the infected group. Overall, this is suggestive of improved detection of preexisting antibody by the HA1ϩHA2 antigen, presumably by detection of antibodies against HA2, which forms part of the conserved stalk region (see Discussion).
Similar to findings using HA1 for detection, the H5 response to HA1ϩHA2 revealed broad reactivity across the different clades (Fig. 3C), again presumably because of the clade 1 followed by clade 2 vaccination regimen. Although IgA signals against the H5 subtype variants are significantly elevated at 28 days postboost ( Fig. 3D and E), the magnitude of this increase is less than that seen for IgG. This observation that the vaccine boosts H5 IgG responses more than IgA against the whole molecule is consistent with similar differential isotype response against the HA1 panel (Fig. 1).
(ii) Natural H3N2 influenza virus infection generates subtype-specific and cross-reactive IgA and IgG antibody responses to HA1؉HA2 proteins. Patients naturally exposed to influenza show H3-specific IgG and IgA responses to the HA1ϩHA2 panel ( Fig. 4; Fig. S2C and D). Like those observed in H5 vaccinees, these Influenza Hemagglutinin Protein Microarrays responses are elevated at the first time point (t1, blue dots), again presumably because of detection of antibodies to HA2. Remarkably, unlike the H3-specific response seen when using HA1 alone, the response to HA1ϩHA2 was broadly reactive across other group 2 subtypes, as well as group 1 subtypes and influenza B virus, as demonstrated by the increase in signal intensity upon follow-up (t2, red dots). This is seen in both IgG ( Fig. 4A and B) and IgA (Fig. 4C and D). On average, IgG signal intensities of the reference group against this panel are higher than those seen in these patients even at follow-up (Fig. 4A, red dots versus orange area). In contrast, the IgA profiles of the two groups look similar. The reasons for the differences seen between reference and patient group profiles are unclear, though they may be attributable to an individual's age or health.
Recipients of an H5 vaccine booster dose, after either low-or high-dosage primary series H5 vaccine, show significantly higher specific IgG responses to H5 molecules. The paired serum samples analyzed in this substudy were derived from subjects who received two doses of either a high (90 g) or a low (15 g) dosage of the H5N1 A/Vietnam/1203/2004 (clade 1) influenza virus vaccine as the primary series. Subsequently, all subjects received a single high-dosage (90 g) H5N1 A/Indonesia/ 05/05 (clade 2) vaccine. The results in Fig. 5A and B show no significant difference in the IgA levels between individuals who received either the high-or low-dosage priming at baseline (d0) or at 28 days postboost (d28). The day 0 time point gives the relative antibody remaining greater than 1 year after the high-or low-dosage vaccine. In contrast, subjects who received high-or low-priming-dosage vaccine show a statistically significant difference in the baseline (d0) IgG antibody response to HA1 (P ϭ 0.033) and HA1ϩHA2 (P ϭ 0.033) (Fig. 5C and D). Similarly, the level of IgG is significantly higher after boost (d28) for subjects who received the high-dosage priming. The magnitude of increase in antibody response between d0 and d28 for HA1 and HA1ϩHA2 after boosting is not significantly affected by the vaccine dosage from the primary series (Fig. S3). These data indicate that the most efficient anamnestic responses were not necessarily associated with the higher-dosage heterologous primary series vaccination.

DISCUSSION
In the U.S., there are no killed whole-virion influenza virus vaccines. FluMist (Medimmune Vaccines, approved since 2003) is the only currently available live, attenuated whole-virion vaccine. Subunit vaccines in the U.S. are recombinantly expressed HA vaccine (Flublok, Protein Sciences, approved since 2013). All other vaccines are technically split vaccines (i.e., whole virus disrupted by detergents and then purified to a desired HA content). Recombinant influenza virus vaccines would be beneficial in the event of a pandemic or a shortage of vaccine supply because of the shorter processing time required for large-scale manufacture and independence from an egg supply. Moreover, virus propagation often results in HA mutations adapted for growth in embryonated chicken eggs, which can affect the antigenicity of the virus (38)(39)(40).
Determining which specific virus of a given hemagglutinin (HA) subtype is included in seasonal influenza virus vaccines is guided by predictions of the viruses that will dominate in a future outbreak. The strategy is suboptimal considering that the immunity engendered by current vaccination approaches is generally subtype specific or even strain specific, and efficacy against strains that emerge in subsequent epidemics and pandemics is unpredictable. The rules for predicting and enhancing strain coverage engendered by seasonal influenza virus vaccination are not well understood but would benefit from a clearer understanding of the extent of HA cross-reactivity engendered by vaccination and infection.
To develop a high-throughput way to measure cross-reactivity of polyclonal serum, particularly to antigens that show high divergence, we evaluated a high-density microarray that consisted of 283 purified HA variant proteins derived from 17 influenza A virus subtypes (H1 to H16 and H18), with HA molecules expressed as HA1 only or as HA1ϩHA2 full-length molecules. Using this large number of proteins, compared to smaller numbers used in previous studies (30)(31)(32)(33)41), allowed for the inclusion of both HA1 and full-length HA proteins (in contrast to some earlier studies that used only HA1 [30][31][32]). HA1 is more likely to detect strain-specific responses, while full-length HA contains the conserved stalk domain and will also detect cross-reactive antibodies that target this region. In addition, in some settings a response to an influenza virus vaccine can be very strain specific without boosting cross-reactive antibodies. In this case, the inclusion of a large variety of antigenically different HAs increases the chances of detecting this response.
Here, we first examined antibody profiles from an H5N1 vaccine study and demonstrated that a heterologous (clade 1-clade 2) prime-boost vaccination gave a broadly cross-reactive H5 IgG antibody response across all the clades represented on the array. This is consistent with a report using samples from the same clinical trial (34) in which higher hemagglutination inhibition and microneutralization geometric mean titers were seen to both clade 1 and 2 strains in subjects primed with clade 1 vaccine and later boosted with clade 2 vaccine, compared to naive subjects who received two doses of clade 2 vaccine 28 days apart. In another study, broad levels of H5 cross-reactivity FIG 5 IgA and IgG aggregated signal intensities in low-dose and high-dose H5 vaccine recipients. H5 vaccine recipients are stratified by the vaccine dose that they received 1 year prior to boost (low, 15 g; high, 90 g). Their antibody reactivities against H5 strains at two time points, d0 and d28, are depicted as Tukey box plots, with medians represented by horizontal bars. "d0" refers to 1 year after primary vaccination, and "d28" refers to 28 days after receiving a 90-g boost. (A and B) IgA antibody responses of low-and high-dose vaccinees, against HA1 only (A) and HA1ϩHA2 proteins (B) on d0 and d28. (C and D) IgG antibody responses of low-and high-dose vaccinees, against HA1 only (C) and HA1ϩHA2 proteins (D) on d0 and d28. Differences between high-and low-dose groups were analyzed using the Wilcoxon rank sum test, and those between time points were analyzed using the Wilcoxon signed-rank test where significance was set at P Ͻ 0.05.
were also achieved in humans using recombinant vaccinia virus Ankara (MVA) to deliver a clade 1 H5 (42). Cross-reactivity was induced against both clade 1 and clade 2 H5 variants, although the titers were overall higher for the homologous, clade 1 H5s. It has also been reported that the priming dosage of a clade 1 H5 vaccine does not differentially impact the IgG response to a subsequent dose of the clade 2 H5-specific vaccine (43). However, we observed significantly higher IgG signals at the higher dose (90 g versus 15 g), not only 1 year after the priming regimen but also after the clade 2 booster immunization. These data would support the notion that a higher dose for priming might be preferable.
It has been described that vaccination with one dose of an H5N1 vaccine boosted cross-reactive antibody responses to the stalk domain while a booster dose of H5N1 vaccine induces mostly antibodies that target the head domain of HA (44,45). We found that booster vaccination with the clade 2 H5 vaccine induced antibodies reactive with other group 1 whole HA (HA1ϩHA2) inefficiently. This is expected since these individuals had already been primed with an H5 vaccine, and the booster dose, therefore, likely induced a recall response that was specific to the head domain of the H5. In addition, we studied subjects naturally infected with H3N2 influenza virus sampled during the 2010 to 2011 influenza season. Natural infection engendered a broadly cross-reactive response within the subtype experienced during infection, in this case H3. Highlighting this was the finding that natural H3N2 infection boosted IgA responses against all H3 variants displayed. Encouragingly, we also found that H3N2 infection boosted antibody responses to group 2 HAs. These finding are similar to a recent study that analyzed sera from H3N2-infected individuals for cross-reactivity by ELISA-a highly sensitive but low-throughput assay format (46).
We also observed that H3N2 infection elicited both IgA and IgG. IgA, particularly secretory IgA (S-IgA), is likely to be the primary means of protection acquired against influenza virus that infects via the respiratory mucosa, although other Ig isotypes and cellular mechanisms can also play a protective role in the absence of IgA (47)(48)(49)(50)(51). Live attenuated influenza virus vaccine (LAIV) is administered intranasally and induces both S-IgA and IgG in the upper respiratory tract, in part by mimicking the natural route of entry. LAIV also induces broader immunity against antigenically drifted strains than inactivated vaccines (52). The efficacy of inactivated seasonal vaccines against influenza may be improved if the induction of IgA could be enhanced, such as by intranasal delivery (53,54).
Interestingly, we consistently observed preexisting heterosubtypic antibody at day 0 in both seasonal cases and vaccinees when using full-length HA1ϩHA2 as detection antigen, compared to HA1 alone. This elevated baseline has the effect of reducing the fold increase at the second time point, thus making the effect of the clade 2 boosting appear more dramatic when using the HA1 alone. One explanation is that the conformation of HA1 is more authentic when assembled in the full-length molecule and therefore better suited for detection of preexisting antibody. Alternatively, and more likely, the preexisting antibodies are recognizing epitopes in the full-length molecule not present in HA1 alone, i.e., the HA2 polypeptide that, with part of HA1, forms the stalk domain. This is consistent with the notion that repeated influenza virus exposure(s) and/or vaccinations to different variants result each time in a primary, strainspecific response to the variable globular head domain but boosting and gradual accumulation of antibodies against the conserved stalk region (46,55).
Although the array can help characterize specificity and cross-reactivity, additional functional assays such as virus neutralization or ADCC assays are required to assess the immunological significance of the antibodies detected. Although the majority of neutralizing epitopes in HA map to the globular head, binding of antibodies to HA in the microarray may not necessarily correlate with protection. Moreover, head-reactive neutralizing antibodies typically show limited cross-reactivity. Indeed, cross-reactivity may correlate better with ADCC as these antibodies preferentially recognize epitopes of the stalk region of HA and are located mainly in HA2. Of note, recent studies have provided evidence that nonneutralizing, broadly binding antibodies (isolated from both mice and humans) can provide strong protection, at least in animal models (56)(57)(58).
It is also important to discuss the limitations of this technology. The data generated using the array are only as good as the recombinant HA probes used. Misfolded or denatured HA will detect different antibodies than correctly folded HA. The system used to express the antigens might also play a role since, e.g., bacterial systems typically do not attach glycans while glycan sizes vary between insect cell-and different mammalian cell-based expression systems. Differences in glycosylation might influence the detected antibody response as well. Also, while large amounts of data can be generated via influenza virus protein arrays, excellent quality control needs to be in place to account for batch-to-batch variations, trending, printing errors, and other technical issues. Finally, as mentioned above, only binding antibody can be measured, and functionality needs to be assessed using additional assays.
Nevertheless, this study supports the utility of the influenza virus purified HA protein microarray as a rapid and high-throughput tool to survey seroreactivity against hundreds of HA variants and inform vaccine and adjuvant development. Here, we investigated antibody profiles induced by H5N1 vaccination and H3N2 natural infection. While we observed interesting differences, these two types of exposures cannot be directly compared since humans are naive to H5N1 but not to H3N2 and since the HAs of the two viruses belong to different HA groups. In future studies, we will therefore investigate differences between exposure to the same virus subtype via natural infection and via vaccines administered by different routes, different dosing regimens, adjuvants, and excipients, to advance our understanding of factors that improve the strength, breadth, and durability of vaccine-mediated immune responses and discover and develop more effective vaccines. Sera were provided to the University of California Irvine (UCI) for assay without patient identifiers and were classified as exempt by the UCI IRB. Sera were collected at two time points: at time of presentation ("t1" ϭ acute-phase infection), and at follow-up ("t2" ϭ convalescent-phase infection) 7 to 31 days later.

Methods
(ii) H5N1 vaccine study. Sera from 25 subjects (age 18 to 64 years) were obtained from future-use consented specimens archived from a clinical trial during which a single intramuscular low-dosage (15 g) or high-dosage (90 g) booster dose of an inactivated subvirion A/Indonesia/05/05 (clade 2) H5N1 vaccine was administered more than 1 year (1.4 to 3.7 years) after receipt of various dosages of a two-dose primary series with a clade 1 (A/Vietnam/1203/2004) H5N1 vaccine (identifier NCT00680069) (34) (see Fig. S1 in the supplemental material). Samples were collected from boosted individuals at baseline (d0) and 28 days after the boost (d28) and used to probe protein arrays. All 25 paired serum specimens were from individuals who received the high-dosage (90 g) clade 2 booster vaccine; 10 paired serum specimens were from high-dosage (30 to 90 g) clade 1 primary series recipients and 15 paired serum specimens were from low-dosage (3.75 to 15 g) clade 1 primary series recipients. Sera were provided to the University of California Irvine (UCI) for assay without patient identifiers and were classified as exempt by the UCI IRB and University of Maryland, Baltimore (UMB) IRB.
(iii) Control sera. As a reference, serum samples collected in early 2008 from healthy blood donors at UC Irvine's General Clinical Research Center (GCRC) who were not known to have received H5N1 vaccine were probed on the arrays. Throughout that season, H1N1, H3N2, and influenza B viruses cocirculated, with H3N2 being the most commonly reported strain (60).
Protein microarray manufacture and probing. (i) Protein microarrays. HA protein microarrays were fabricated as previously described (61) with modifications. Briefly, purified influenza virus antigens, representing 17 influenza A virus subtypes and influenza B virus, were obtained from Sino Biological Inc. (Beijing, China) ( Fig. S4; Table S1). The protein set comprised 283 lyophilized influenza virus hemagglutinin (HA), 4 neuraminidase (NA), and 1 nucleoprotein (NP) variants, with HA molecules expressed either as separate subunits (HA1 or HA2) or as a whole molecule (HA1ϩHA2). The neuraminidase, nucleoprotein, HA2, and NS1 features were not used for the analysis in this study. Proteins were derived from roughly equal numbers of human and nonhuman (avian, swine, etc.) isolates and were expressed in baculovirus or human cell expression systems. Approximately half of the total HA1ϩHA2 molecules printed on the array were expressed in a baculovirus system, and 2% of total HA1 molecules were similarly expressed (Fig. S4C). H5 strains are highly represented on the array, making up almost a third of the HA protein set, and included HAs from the American and Eurasian lineage. The majority of the Eurasian lineage HAs were derived from the A/goose/Guangdong/1996 virus, which gave rise to the highly pathogenic Asian H5N1 viruses. HAs from clades 0, 1, 2.1.3, 2.2, 2.3.2.1, 2.3.4, 2.5, 3, 4, 5, and 7 were included. This was followed by H1 (17%), H3 (16%, including H3 HAs from the 1968 pandemic, modern and historic vaccine strains, and currently circulating strains as well as several swine and equine H3 virus HAs), and H7 (12%) subtypes, which are relevant for being commonly encountered during seasonal epidemics (Fig. S4A). Proteins expressed in insect cells presented signal intensities that were overall higher than corresponding ones expressed in human cells as demonstrated by the low transformed human-cell-to-baculovirus ratios (Fig. S5). The same conclusions were drawn whether the whole data set was pooled or whether data from the two expression systems were analyzed independently in this pilot study. Therefore, analyses from the pooled data set are presented throughout. Prior to printing, each lyophilized antigen was reconstituted to a concentration of 0.1 mg/ml in phosphate-buffered saline (PBS) with 0.001% Tween 20 (T-PBS). In addition to these purified proteins, 14 nonstructural (NS1) antigens were expressed by an Escherichia coli-based in vitro transcription/translation (IVTT) system (RTS 100 E. coli HY kit; Biotechrabbit GmbH, Hennigsdorf, Germany) from purified DNA. Negative control ("No DNA" control) was included by performing IVTT reactions in the absence of DNA template. Purified proteins and crude IVTT reaction mixtures were printed onto nitrocellulose-coated glass AVID slides (Grace Bio-Labs, Inc., Bend, OR, USA) using an Omni Grid 100 microarray printer (Genomic Solutions).
(ii) Microarray probing and development. Serum samples were diluted 1:100 in protein array blocking buffer (GVS, Sanford, ME, USA) supplemented with E. coli lysate (GenScript, Piscataway, NJ, USA) to a final concentration of 10 mg/ml and preincubated at room temperature (RT) for 30 min. These conditions were established previously to give robust antigen signals while also giving low (Ͻ5,000) background signals to E. coli for the majority of seropositive samples. Concurrently, arrays were rehydrated in blocking buffer (without lysate) for 30 min. Blocking buffer was removed, and arrays were probed with preincubated serum samples using sealed chambers to ascertain that there was no cross-contamination of samples between the pads. Arrays were incubated overnight at 4°C with gentle agitation. They were then washed at RT three times with Tris-buffered saline (TBS) containing 0.05% Tween 20 (T-TBS), followed by incubation for 2 h at RT with a mixture of Qdot800-conjugated goat anti-human IgG (Grace Bio-Labs, Inc.) and biotin-conjugated goat anti-human IgA (Jackson ImmunoResearch Laboratories, Inc., West Grove, PA, USA), diluted 1:250 and 1:200, respectively, in blocking buffer. Arrays were washed three times with T-TBS, followed by incubation with streptavidin-conjugated Qdot655 (Thermo Fisher Scientific, Waltham, MA, USA) diluted 1:250 in blocking buffer for 1 h at RT (62). Arrays were washed three times with T-TBS and once with water. Arrays were air dried by centrifugation at 500 ϫ g for 10 min. Images were acquired using the ArrayCAM imaging system from Grace Bio-Labs (Bend, OR). Spot and background intensities were measured using an annotated grid (.gal) file. The imager settings were set at gain 50 and 500-ms exposure time for both 655-and 800-nm channels.
Data analysis. Purified protein signal intensities used for calculations were first background corrected by subtracting sample-specific T-PBS buffer signals from purified protein spot signals. When the H5 vaccine study is discussed, only data collected from the high-dosage group were used for analysis of antibody responses, unless otherwise indicated. The two-tailed Mann-Whitney test for unpaired data and the two-tailed Kruskal-Wallis test with Dunn's multiple-comparison test were performed in GraphPad Prism 6 (GraphPad, La Jolla, CA, USA). A P value of Ͻ0.05 was considered statistically significant. The antigen identifiers and raw data are provided in Table S2 in the supplemental material.
For comparisons between low-and high-dosage vaccines, mean fluorescence intensities (MFIs) for all antigens per Ag group (HA1 and HA1ϩHA2) at t1 and t2 were added per individual sample to obtain an individual's aggregate reactivity for that time point (for that group of Ags). These were then plotted to compare the dosage groups (high and low). Differences between high-and low-dosage groups were analyzed using the Wilcoxon rank sum test in R, and those between time points were analyzed using the Wilcoxon signed-rank test, where significance was set at P Ͻ 0.05. Delta increases were calculated by subtracting the MFIs for each antigen t2 Ϫ t1 for each volunteer. Delta MFIs for all antigens per antigen group (HA1 and HA1ϩHA2) were then summed for each individual's aggregate reactivity for that group of antigens. These were then plotted to compare high-and low-dosage recipients. Differences between high-and low-dosage groups were analyzed using the Wilcoxon rank sum test in R, where significance was set at P Ͻ 0.05.