Antigenic characterization of highly pathogenic avian influenza A(H5N1) viruses with chicken and ferret antisera reveals clade-dependent variation in hemagglutination inhibition profiles

Highly pathogenic avian influenza (HPAI) A(H5N1) viruses pose a significant economic burden to the poultry industry worldwide and have pandemic potential. Poultry vaccination against HPAI A(H5N1) viruses has been an important component of HPAI control measures and has been performed in Vietnam since 2005. To systematically assess antigenic matching of current vaccines to circulating field variants, we produced a panel of chicken and ferret antisera raised against historical and contemporary Vietnamese reference viruses representing clade variants that were detected between 2001 and 2014. The antisera were used for hemagglutination inhibition (HI) assays to generate data sets for analysis by antigenic cartography, allowing for a direct comparison of results from chicken or ferret antisera. HI antigenic maps, developed with antisera from both hosts, revealed varying patterns of antigenic relationships and clustering of viruses that were dependent on the clade of viruses analyzed. Antigenic relationships between existing poultry vaccines and circulating field viruses were also aligned with in vivo protection profiles determined by previously reported vaccine challenge studies. Our results establish the feasibility and utility of HPAI A(H5N1) antigenic characterization using chicken antisera and support further experimental and modeling studies to investigate quantitative relationships between genetic variation, antigenic drift and correlates of poultry vaccine protection in vivo.


Introduction
Antigenic drift is the Achilles' heel of designing effective vaccines against rapidly evolving pathogens, such as influenza viruses. In poultry vaccination programs, the challenges are particularly acute when significant genetic variation exists among co-circulating strains, as is the case for highly pathogenic avian influenza (HPAI) A(H5N1) viruses. Antigenic matching, which refers to the antigenic similarity between a given vaccine strain and circulating field viruses, is typically measured by raising antisera in animal models and subsequently comparing antibody-toantigen reactivity titers of the vaccine strain (homologous titer) vs. circulating field viruses (heterologous titers). For many decades, human seasonal influenza virus antigenic matching has relied on hemagglutination inhibition (HI) assays using ferret antisera as the gold standard for measuring variation among viruses 1,2 . Twice each year, the World Health Organization (WHO) Global Influenza Surveillance and Response System (GISRS), in conjunction with Collaborating Centres for Influenza and Essential Regulatory Laboratories, produces genetic and antigenic analyses of viruses to systematically review diversity of influenza B and influenza A viruses (IAVs) from the Northern and Southern hemispheres, and recommends the antigen formulation for human seasonal influenza vaccines. The recommendations also include a selection of specific candidate vaccine viruses (CVVs) for non-seasonal influenza virus subtypes and genetic lineages as a precautionary measure to produce at-the-ready vaccine seed viruses for pandemic vaccine preparedness 3 .
Ferrets are the preferred animal model to study the susceptibility, virulence and transmission of IAVs because they are highly susceptible to both human and potentially zoonotic IAVs and experience similar clinical outcomes and immune responses to those of humans 4 . However, the production of ferret antisera is expensive and requires animal biocontainment laboratories, which is needed in part to prevent unwanted exposure of naive ferrets to circulating seasonal viruses. The importation of serologically influenza-naive (and specific pathogen free) ferrets into Vietnam is complicated due to customs procedures and the likelihood of infection during transport. Available animal containment facilities within Vietnam are also limited in space and resources, and there are currently no commercial entities that raise laboratory animals under containment for research purposes. Thus, the ferret model of infection is currently unavailable in Vietnam, and representative panels of ferret antisera are rarely available. Because chickens are the principle target for HPAI poultry vaccination and are readily available, they are an ideal animal model for producing reference antisera to evaluate antigenic variation among viruses and vaccines, host immune responses, and vaccine efficacy 5 . In this study, we aimed to produce antisera in chickens against representative HPAI A (H5N1) clade variants at the National Centre for Veterinary Diagnostics (Hanoi, Vietnam) and to use these antisera to characterize circulating HPAI A(H5N1) Vietnamese viruses by HI assay and antigenic cartography. We compared antigenic relationships among viruses using panels of ferret versus chicken antisera to explore the utility and feasibility of HI testing with chicken antisera to antigenically match currently available poultry vaccines used in Vietnam with circulating strains.

Collection and preliminary analysis of chicken antisera
The average volume of antiserum collected from individual chickens was 6.5 ml (ranging from 4 to 13 ml).

Antigenic analysis of HPAI A(H5N1) virus isolates Ferret-based map
The antigenic distances between viruses based on HI analyses using ferret antisera are displayed in Fig. 1 (Fig. 1). The antisera produced against clade 2.3.4 viruses also reacted with high specificity for related viruses, although some clade 1 viruses were an exception ( Fig. 1). Antisera against clade 7.1 viruses cross-reacted within this clade, but these viruses were outliers compared to all the other clades.
The use of ferret antisera for one-way analyses of additional test antigens yielded heterologous HI titers of sera against related and unrelated HPAI A(H5N1) within clade viruses, ranging from equivalent to 64-fold titer differences compared to homologous titers. From the ferret antisera-based map (Fig. 1

Chicken-based map
Two-way analyses of chicken antisera and the chicken antisera-based map ( Fig. 2; Supplemental Table 2) showed a similar trend of within-clade clustering as that observed in the ferret antisera-based map ( Fig. 1)  , little reactivity to Vietnamese clade viruses was observed. In general, the results of the two-way and one-way analyses generated from HI tests using chicken antiserum panels were compatible to those produced with ferret antisera.

Combined chicken and ferret antisera-based map
In general, the homologous HI titers indicated that four of fourteen pairs (28%) of chicken/ferret antisera had identical homologous HI titers (with titers ranging from 1:320 to 1:1280); eight pairs (57%) had higher HI titers in ferrets than in chickens, ranging from 2-to 8-fold higher (2 pairs had a 2-fold difference, 2 pairs had a 4-fold difference and 4 pairs had 8-fold difference); and 2 pairs (14%) had higher HI titers in chickens compared to ferrets (1 pair had a 2-fold difference and 1 pair had 4-fold difference). Side by side comparisons of chicken and ferret antisera run in the same HI test are shown in Fig. 3  locations of each other (71%, n = 10/14). Antiserum pairs for clades 7.1 (n = 1), 1 (n = 1) and 0 (n = 1) showed nearly identical positions, as did most pairs in clades 2.3.2.1 (n = 3/5) and 2.3.4 (n = 1/2). However, three antigens that had homologous HI titers within 4-to 8-fold between chicken and ferret antisera pairs displayed greater fold reductions in side by side comparisons of heterologous titers generated from one-way testing of viruses. For instance, the clade 2. 3 , also displayed greater fold reductions (between 8and 32-fold differences) when comparing the reactivity of the homologous chicken and ferret antisera. The HI table shows that this virus produced a very high homologous HI titer of 1:1280 in the ferret, whereas the immune response in the chicken produced a titer of 1:160 (8-fold difference). This could be an anomaly in the ferret antiserum produced against this virus, suggesting that careful evaluation of initial homologous titers is warranted prior to use in routine HI testing.

Discussion
Vietnam is one of four countries (along with China, Egypt and Indonesia) where poultry vaccination has become a routine practice to control HPAI A(H5) viruses. To date, most government-supported poultry vaccination programs in Vietnam have employed inactivated, oil-inwater emulsion vaccines imported from the Harbin Veterinary Research Institute in China, which have been periodically updated and reformulated in an attempt to optimize antigenic matching following shifts in predominant H5 clade circulation in China. Phylogenetic analyses of codon-complete genome sequences of A(H5) viruses isolated from poultry outbreaks in Vietnam have identified a total of 56 genotypes and a circulation of at least 16 distinct clades [10][11][12][13][14][15][16] 13 . Given the rapid evolution and frequent introduction of new A (H5) clades, maintaining a comprehensive panel of reference ferret antisera for antigenic characterization of viruses is crucial to assess antigenic matching of existing poultry vaccines. However, despite the importance of these reagents for successful vaccine strain selection and implementation policies, production of these panels remains resource-intensive and expensive to generate. To explore the utility of chicken antisera as a cost-effective strategy to meet these research goals, we ran HI assays using both chicken and ferret antisera and compared their ability to discriminate between antigenically distinct viruses using antigenic cartography.
Our results indicated that antigenic distances calculated using multidimensional scaling or 'mapping' approaches generated similar results for both the chicken and ferret HI data sets. In both instances, antigenic clustering corresponded to genetic grouping as determined by HA phylogenies. Although analyses of the chicken and ferret antisera showed consistent reactivity patterns among tested the viruses and yielded similar comparisons of antigenic groupings, differences in reactivity and relative map positions were observed for 4 of the 14 chicken and ferret antisera pairs. For example, ferret antisera were unable to resolve antigenic differences between clade 1 and 2.3.4 viruses, which had overlapping antigenic space compared to the clearly defined delineation between these two clades using chicken antisera (Fig. 1,  Fig. 2). On the other hand, clade 2.3.2.1 viruses clustered well in antigenic maps from ferret HI tests (within 1 antigenic unit) but were more widely dispersed in tests performed with chicken antisera. Variation in reactivity profiles of chicken vs. ferret antisera may be attributable to differences in antisera production. First, the production of ferret antisera, involved the inoculation of ferrets with live virus, followed by an adjuvanted boost 14 days post-infection. In contrast, the production of antisera in chickens required that inactivated virus be used for inoculation, as HPAI viruses cause fatal infections in chickens, which would have abrogated the possibility of collecting influenza-specific serum antibodies. Ferrets, on the other hand, experienced infections with the wild-type viruses that likely triggered both innate and cell-mediated immunity and stimulated pattern recognition receptors in a manner similar to natural infection. The production of reference anti-sera in chickens requires that they be immunized with an inactivated antigen due to the extreme lethality of HPAI A(H5) infections in chickens. Thus, the immune response measured by our chickenbased maps more closely mimicked vaccine-induced immunity rather than natural infection. These differences may be expected to impact the overall antibody titer but not the specificity of the serum antibodies. Although live virus infections may stimulate a broader, more robust immune response in chickens and reflect a natural infection, its practical application for generating antisera may be less advantageous. Utilization of a live virus requires that attenuated, reassortant/reverse genetics viruses are available because of the lethality of HPAI A(H5) viruses in chickens. In resource-limited situations, genetically modified, attenuated strains are expensive and time-consuming to produce, whereas inactivated virus is easily obtained from field strains. Furthermore, because poultry vaccines used in Vietnam rely on inactivated vaccine formulations, the generation of antisera with inactivated virus may be a better model of the post-vaccination immune response and better reflect the immune response to vaccination in the field. A direct comparison of antisera raised in ferrets using inactivated virus as an immunogen was not possible due, in part, to existing protocols approved by the Institutional Animal Care and Use Committee of the CDC that require the use of infectious virus but, more importantly, because previous studies showed that inactivated virus did not consistently elicit sufficient HI antibody titers when given intranasally (data not shown). A second important methodological difference was the use of different adjuvants for the ferrets and chickens. Adjuvants are compounds that have been shown to non-specifically augment host responses 17 . The precise mechanisms and differences between adjuvanted and non-adjuvanted responses have rarely been defined, and the impact they may have on the breadth of antibody coverage or specificity of the response is rarely measured with precision. For the experiments reported in this study using ferret antisera, Titermax Gold adjuvant was used in the boost inoculum, while the adjuvant used to produce reference chicken antisera was Montanide TM ISA 70 VG. Both TiterMax Gold (http://www.titermax.com/technicalinformation.html) and Montanide TM ISA 70 VG have been developed to create stable water-in-oil (W/O) emulsions. Although both adjuvants are metabolized in a similar fashion, it has not been assessed whether their use results in differences in antigenic cross-reactivity of serum antibodies. Finally, besides the methodological differences noted, there is likely to be some variation in how the ferret immune system targets viral epitopes compared to that of chickens. Host-specific variation in antigen processing, antibody production, and ultimately, the specificity of an antibody for an antigen may lead to variation in HI reactivity patterns between antisera raised in each host.
Plotting of the antigenic distance relative to percent survival of chickens post-vaccine challenge demonstrated an inverse relationship between survival rate and antigenic distance. In general, the comparison of antigenic distances and percent survival of chickens following vaccine challenge suggested a trend toward lower survival rates as the antigenic distance between a challenge virus and vaccine virus increased. Indeed, aggregate results of in vivo challenge studies showed that Re-1 (clade 0) reliably protected chickens against clinical disease when challenged with either clade 1 or clade 2. What remains to be seen is whether the precise quantitative measures of antigenic distances, as afforded by the cartographic analyses presented here, could be used to model and predict outcomes from experimental challenge studies. For example, it should be feasible to examine, compare, and visualize how the breadth of immunity induced by bivalent vaccine formulations (i.e., comprising two A(H5) clade variants) differs from immune responses to monovalent formulations through an antibody landscaping approach that builds upon the A(H5) maps presented here.
In summary, our data demonstrate the utility of using panels of reference chicken antisera for the systematic analysis of antigenic drift variants of A(H5) viruses and indicates that the chicken-and ferret-based antiserum panels used in HI assays yielded comparable antigenic profiles, which were relevant for evaluating A(H5) vaccine antigenic matches. The great advantage of using antisera raised in chickens for these analyses is the feasibility of generating the antigenic data quickly in-country and within the existing constraints and resources of national animal health laboratories in Vietnam. We strongly advocate for more integrated analyses of antigenic distance measures based on cartography, together with output from in vivo transmission studies, to develop reliable surrogate markers of protective immunity and standardized measures of vaccine performance. Strategic use of new computational tools is needed to meet the long term goal of achieving robust vaccine-induced heterosubtypic immunity and to overcome the Achilles' heel of antigenic drift.

Antigen preparation
Viruses were grown in 9-or 10-day-old influenza-free, clean-embryonated chicken eggs (C-ECE). Ten C-ECEs per isolate were used, and the allantoic fluid was harvested from each C-ECE after incubation at 37°C for 48-72 h. After harvesting, viruses were inactivated with betapropiolactone (BPL) at a final concentration of 0.05% for 4 h at 37°C, after which the virus samples were stored at 4°C overnight and then adjusted to pH 7.2 using a sterile 7.5% NaHCO 3 solution. Inactivation was confirmed through three serial egg passages without any evidence of viral growth as detected by HA. Inocula for chicken antisera production were prepared by mixing inactivated antigens [at 1280 hemagglutination units (HAU)] with an adjuvant (Montanide ISA 70) in a 3:7 v:v ratio, which were emulsified by sonication for 15-20 s as per manufacturer's instructions (http://www.seppic.com/animal-health/ vaccine-adjuvant/montanide-isa/-montanide-isa-70-vg). The emulsified antigen-adjuvant mixtures were stored at 4°C and used within 24-48 h of preparation.

Chicken antisera production
Chickens were obtained as day-old chicks from a commercial hatchery that was certified as being free of avian influenza viruses by the Department of Animal Health (DAH) in Hanoi, Vietnam. The antibody-negative status of chickens was reconfirmed at 5 weeks of age, prior to being used for antisera production, by testing with a commercial NP ELISA Flock Check AI (Idexx Laboratories, Westbrook, ME, USA). Five chickens were used per viral antigen for antisera production. Briefly, chickens were immunized subcutaneously in the nape of the neck with 0.5 ml of immunogen/adjuvant using a 5-ml syringe and a 22 gauge needle. To produce antisera against commercial poultry vaccines, vaccines were reconstituted as per manufacturer's instructions (Harbin Veterinary Research Institute, Harbin, China). At 3 weeks post-immunization, chickens were sedated by intravenous administration of ketamine/xylazine based on body weight, and blood was collected by cardiac puncture. Whole blood tubes were incubated for 3-5 h at 37°C on their sides, after which they were placed at 4°C overnight to maximize serum volume. Sera were harvested by centrifuging the samples to pellet the clots, after which the supernatants were decanted and stored at −20°C in 1ml aliquots until further use. Procedures for animal work at the NCVD were adapted from the OIE (available at: http:// web.oie.int/eng/normes/mcode/en_chapitre_1.7.8.htm).

Ferret antisera production
Ferret antisera against the selected A(H5) viruses were produced as previously described at the Centers for Disease Control and Prevention (CDC) Influenza Division in Atlanta, GA, USA. Briefly, outbred male ferrets with body weights ranging from 900 to 1800 g and less than one year of age were obtained from Triple F Farms (Sayre, PA, USA). Ferrets were confirmed as being antibody-negative for influenza A and B viruses and were inoculated intranasally (in) with doses ranging between 10 4 and 10 6 EID 50 of live virus diluted in PBS. At 14 days post-inoculation, each ferret was boosted by subcutaneous injection in both hind legs with at least 1024 HAU of virus mixed with Titermax Gold Adjuvant (Sigma-Aldrich, MO, USA). Antisera were collected 14 days post-boost. All work with animals was carried out in biosafety level 3 laboratories with enhancements to meet USDA/APHIS guidelines and performed under protocols approved by the Institutional Animal Care and Use Committee of the CDC.

Comparison of chicken and ferret antisera via HI assays
Antigenic characterization was performed using the post-infection ferret and chicken antisera in parallel. Sera were heat inactivated at 56°C for 30 min, absorbed with packed chicken red blood cells (CRBCs) to remove nonspecific agglutinators, and then tested through an HI assay with 0.5% CRBCs following standard procedures 19 . Starting dilutions of 1:10 were used, and endpoint titers were calculated as the reciprocal of the last HI positive serum dilution. Naive, negative control sera from ferrets and chickens were used in all HI tests to control for potential background reactivity (data not shown).

Antigenic cartography
Antigenic cartography was performed as previously described 20 using the open access software available through https://acmacs-web.antigenic-cartography.org/ . HI tables containing the full panel of HI titers (homologous and heterologous) were first normalized by calculating the difference between the log2 (HI titer) of a given virus-serum pair and the maximum log2 (HI titer) of that serum against a given virus. Antigenic distances were calculated using multidimensional scaling to minimize the differences between the target distances and the distances in the antigenic map. Fivehundred dimensional annealing runs were performed for optimization, with random restarts implemented to avoid being trapped in local optima. Metadata associated with all reference and test antigens were prepared in Microsoft Office Excel 2010 spreadsheets with an antigen table (test virus, antigen ID, strain name, clade, and abbreviation) and antisera table (test serum, serum ID, serum strain, serum clade, abbreviation, and serum type). Antigenic maps were visualized using Tableau Reader v9.1 by importing x/y coordinate output files and linking each coordinate with reference or test antigen metadata.

HI data matrices
A total of 18 ferret antisera were used for cross-HI tests with 15 homologous antigens and 25 test antigens.

Rationale of the HI test set-up
Following the conventions and terminology of antigenic characterizations of serological data sets, we refer to 'two-way' vs. 'one-way' analyses of HI titer matrices. Two-way analyses refer to interpretations of antigenic relationships based on complete data sets, where each virus is represented by both an antigen and its homologous antiserum. For two-way analyses, the spatial coordinates for a given antigen or antiserum reflect the differential reactivity to both homologous and heterologous viruses. For the purposes of comparing chicken vs. ferret models, we first performed two-way analyses to assess whether the chicken and ferret models generated antisera with comparable specificities. In contrast to 'two-way' analyses, a 'one-way' analysis involves interpretation of antigenic relationships when a given virus is represented solely by the viral antigen because no homologous antiserum has been produced (in this case the antigens are called 'test' antigens). For one-way analyses, the spatial coordinates for a given antigen are derived from the combined reactivity of a given antigen to a panel of heterologous antisera. The most common scenario for cartography is to have a small number of reference antigens and a much larger number of test antigens because antisera are expensive and timeconsuming to generate.
Plotting of antigenic distances to in vivo vaccine efficacy Antigenic distances, calculated from chicken antiserabased HI tests, between viruses previously tested in vaccine challenge studies and each of three poultry vaccines (i.e., Re-1, Re-5, and Re-6) were calculated as above. The distances were plotted against the percent survival of chickens that were vaccinated and challenged with the corresponding vaccine and virus. The methodology and results from the vaccine challenge studies were previously described [6][7][8][9] .

Chicken challenge studies
The commercial inactivated H5N1 Re-1, Re-5 and Re-6 poultry vaccines, which were previously and currently used in Vietnam and produced by the Harbin Veterinary Research Institute (People's Republic of China), were used to vaccinate chickens. Chickens were obtained at one day of age from a commercial hatchery. Serum samples were collected from 20 chickens in each experiment to ascertain that the birds were serologically negative for antibodies to the NP protein of influenza A viruses as determined by the commercial ELISA test Flock Check AI (Idexx Laboratories, Westbrook, ME). Two-week-old chickens (20 chickens/group) were vaccinated subcutaneously in the nape of the neck with 0.5 ml of the Re-1, Re-5 or Re-6 vaccines. This vaccination schedule was based on the manufacturer's recommendations. In each experiment, an additional 10 chickens were not vaccinated and served as the challenge controls.
Three weeks after vaccination, blood samples were collected and tested by hemagglutinin tests (HI) with inactivated antigens of A/VN/1203/2004 (for the Re-1 vaccinated chicken group), A/Anhui/1/2005 (for the Re-5 vaccinated chicken group), and A/Hubei/1/2010 (for the Re-6 vaccinated chicken group). Ten chickens with HI titers ≥ 3 log2 from each vaccinated group were selected for challenge experiments, which were inoculated intranasally with 10 6 TCID 50 of the selected H5 HPAI challenge virus. Chickens were observed daily for clinical signs and mortality. Oropharyngeal and cloacal swabs were collected at different days post challenge (dpc) to assess viral shedding. At the end of each experiment (10 days), blood was collected from all surviving chickens for antibody assays. All experiments were performed in biosafety level two plus enhanced facilities at the National Centre for Veterinary Diagnostics.