ABSTRACT
The highly pathogenic avian influenza virus (HPAIV) H5N1 A/goose/Guangdong/1996 lineage (Gs/GD) is endemic in poultry across several countries in the world and has caused sporadic lethal infections in humans. Vaccines are important in HPAIV control both for poultry and in prepandemic preparedness for humans. This study assessed inactivated prepandemic vaccine strains in a One Health framework across human and agricultural and wildlife animal health, focusing on the genetic and antigenic diversity of field H5N1 Gs/GD viruses from the agricultural sector and assessing cross-protection in a chicken challenge model. Nearly half (47.92%) of the 48 combinations of vaccine and challenge viruses examined had bird protection of 80% or above. Most vaccinated groups had prolonged mean death times (MDT), and the virus-shedding titers were significantly lower than those of the sham-vaccinated group (P ≤ 0.05). The antibody titers in the prechallenge sera were not predictive of protection. Although vaccinated birds had higher titers of hemagglutination-inhibiting (HI) antibodies against the homologous vaccine antigen, most of them also had lower or no antibody titer against the challenge antigen. The comparison of all parameters and homologous or closely related vaccine and challenge viruses gave the best prediction of protection. Through additional analysis, we identified a pattern of epitope substitutions in the hemagglutinin (HA) of each challenge virus that impacted protection, regardless of the vaccine used. These changes were situated in the antigenic sites and/or reported epitopes associated with virus escape from antibody neutralization. As a result, this study highlights virus diversity, immune response complexity, and the importance of strain selection for vaccine development to control H5N1 HPAIV in the agricultural sector and for human prepandemic preparedness. We suggest that the engineering of specific antigenic sites can improve the immunogenicity of H5 vaccines.
IMPORTANCE The sustained circulation of highly pathogenic avian influenza virus (HPAIV) H5N1 A/goose/Guangdong/1996 (Gs/GD) lineage in the agricultural sector and some wild birds has led to the evolution and selection of distinct viral lineages involved in escape from vaccine protection. Our results using inactivated vaccine candidates from the human pandemic preparedness program in a chicken challenge model identified critical antigenic conformational epitopes on H5 hemagglutinin (HA) from different clades that were associated with antibody recognition and escape. Even though other investigators have reported epitope mapping in the H5 HA, much of this information pertains to epitopes reactive to mouse antibodies. Our findings validate changes in antigenic epitopes of HA associated with virus escape from antibody neutralization in chickens, which has direct relevance to field protection and virus evolution. Therefore, knowledge of these immunodominant regions is essential to proactively develop diagnostic tests, improve surveillance platforms to monitor AIV outbreaks, and design more efficient and broad-spectrum agricultural and human prepandemic vaccines.
INTRODUCTION
The highly pathogenic avian influenza virus (HPAIV) H5N1 A/goose/Guangdong/1996 (Gs/GD) lineage is widespread, producing infections in poultry, wild birds, and humans in 84 countries throughout Asia, Africa, Europe, and North America (1–3). Since the first report of H5N1 in China in 1996, the H5N1 Gs/GD lineage HPAIV has spread across continents, with successful eradication in many countries (4). However, the geographic isolation and maintenance of some H5N1 strains through endemic infections in the agricultural sector with spillover into natural ecosystems of migratory aquatic birds and genetic and antigenic drift have resulted in viral diversity, creating sublineages grouped into 10 genetically distinct virus clades (0 to 9) and multiple subclades (2). Continued circulation further resulted in some clades becoming extinct while new, antigenically distinct variants emerged. In total, four waves of intercontinental transmission of Gs/GD lineage H5Nx virus have been identified: (i) the 2005-2006 wave caused by clade 2.2 H5N1 HPAIV; (ii) the 2009-2010 wave caused by clade 2.3.2.1c H5N1 HPAIV; (iii) the 2014-2015 wave involving two separate lineages, 2.3.4.4a H5Nx and clade 2.3.2.1c H5N1, that differed antigenically from the 2009-2010 viruses; and (iv) the 2016-2017 wave caused by clade 2.3.4.4b H5Nx (5).
Ever since AIVs were identified as causes of fowl plague in 1955, more than 44 documented outbreaks of HPAIV have occurred (5). The Gs/GD lineage epizootic alone has affected more poultry in various countries than the other 43 HPAI outbreaks combined (5). Circulation of the H5N1 HPAIV Gs/GD lineages continues to threaten agricultural systems, natural ecosystems, and human health, i.e., One Health. Since June 2016, countries in both Europe and Asia have detected infections in migratory aquatic birds in natural ecosystems and/or domestic poultry in the agricultural sector with HPAIV H5N1 clade 2.3.4.4, showing mortality in wild birds (https://www.who.int/influenza/human_animal_interface/avian_influenza/riskassessment_AH5N8_201611/en/). Further spread into other countries can still occur due to possible spread along the migratory routes of aquatic birds. Although the likelihood of human infection with the clade 2.3.4.4 H5Nx virus is low, it should be noted that human infection with H5N6 HPAIV of clade 2.3.4.4 has already occurred (6). In addition, zoonotic infections with HPAIV H5N1 clades 2.1 and 2.2 have been reported, predominantly in Egypt and Indonesia, since 2014 (5, 7). Thus, although the risk of human infection is low, we cannot ignore its possibility—H5N1 continues to evolve and potentially acquire mutations that promote transmission to humans (6).
In countries where the Gs/GD lineage is endemic, the control of H5N1 HPAI in poultry by limited eradication programs is supplemented by vaccination, primarily through inactivated oil-emulsified adjuvanted whole-virus vaccines, with more limited use of live-vectored vaccines (8). The genetic diversification and rapid antigenic drift of field Gs/GD viruses in the agricultural sector have frequently led to inadequate protection. In addition, the prolonged vaccination programs adopted in countries where the virus is endemic, such as China, Egypt, and Vietnam, have subsequently seen variant strains resistant to inactivated vaccine seed strains emerge (7, 9–12). Thus, to maximize protection of both poultry and humans, the antigenic characteristics of vaccine seed strains should be periodically reevaluated by in vitro and in vivo assays against the currently circulating field viruses (13) to determine whether the available vaccine strains should be changed.
Currently, there is no vaccine seed strain capable of eliciting protective immunity across the various clades and subclades of the H5 Gs/GD lineage of HPAIV in circulation (http://www.who.int/influenza/vaccines/virus/candidates_reagents/H5N1_summary_a_h5n1_cvv_20180305.pdf?ua=1). There is a need for a protective H5 HPAIV vaccine that provides broad coverage of cocirculating antigenically distinct variants in poultry and as a prepandemic vaccine for humans.
In this study, we evaluated inactivated prepandemic vaccine strains in a One Health framework across human and agricultural and wildlife animal health, focusing on genetically and antigenically matched and mismatched clade challenge viruses from the agricultural sector and assessing protection in a chicken challenge model in a variety of protection parameters, including (i) survival, (ii) mean death time (MDT), (iii) decrease in virus-shedding titers and the number of birds shedding, and (iv) hemagglutination inhibition (HI) antibody levels measured against both the vaccine and challenge virus hemagglutinin (HA) proteins pre- and postchallenge in the context of genetic and antigenic differences between the HAs of vaccine and challenge viruses.
RESULTS
Protection of vaccinated birds.We examined the vaccination efficacies of six inactivated vaccine strains against challenge with eight H5N1 Gs/GD lineage HPAIVs in chickens (Table 1). Vaccine protection based on the percent bird survival varied greatly among the 48 homologous and heterologous combinations of vaccine strains and challenge viruses tested. We observed a survival efficacy rate of 90% or above for 16 out of the 48 possible combinations of vaccine strains and challenge viruses (Fig. 1; see Table S1 in the supplemental material). In contrast, survival rates of ≥80% and ≥70% were observed for a total of 23 and 28 combinations (Fig. 1; see Table S1), respectively. The homologous combination of vaccine 6 (7.1/Vn016/08) and challenge virus 7 (7.1/Vn016/08) provided 100% protection. Further, most of the vaccine strains and challenge strains from homologous clades had protection rates between 89% and 100% (Fig. 1; see Table S1), except for vaccine 2 (2.1.3.2/Indo/05) and challenge virus 2 (2.1.3/WJ29/07), vaccine 4 (2.3.2.1a/Hub/10) and challenge virus 6 (2.3.2.1b/Vn672/11), and vaccine 5 (2.3.4/Anh/05) and challenge virus 7 (2.3.4/HK8825/08), which had 20%, 60%, and 70% protection, respectively (Fig. 1; see Table S1).
Avian influenza virus vaccine strains and challenge viruses tested, clades, and accession numbers of HA genes
Survival curve of vaccinated and unvaccinated (Sham) chickens challenged at 6 weeks of age with different clades of H5N1 HPAIV at a dose of ∼106.0 EID50/0.1 ml. The inactivated vaccines used are shown in each panel. Each challenge virus used in the vaccinated and unvaccinated (positive-control) groups is identified by the same color but different symbols, as shown in the legend on the right.
To analyze the changes in protection of multiple vaccine strains against the same set of challenge viruses, we used the Wilcoxon matched-pairs signed-rank test, a nonparametric version of the dependent t test. The results showed statistically significant changes in protection between vaccine strains 2 (2.1.3.2/Indo/05) and 3 (2.2.1.1/E3300/08) (P = 0.0078), vaccine strains 2 (2.1.3.2/Indo/05) and 4 (2.3.2.1a/Hub/10) (P = 0.0078), and vaccine strains 2 (2.1.3.2/Indo/05) and 6 (7.1/Vn016/08) (P = 0.0234) (Fig. 2; see Table S2 in the supplemental material). Clearly, these analyses (Fig. 2; see Table S2) demonstrated that vaccine 2 (2.1.3.2/Indo/05) (Fig. 2, white box plot) had the lowest protection of all six vaccine strains tested. Since there were no statistical differences among the other five vaccine strains and vaccine strains 1 (1.1/Camb/07), 3 (2.2.1.1/E3300/08), 4 (2.3.2.1a/Hub/10), 5 (2.3.4/Anh/05), and 6 (7.1/Vn016/08) all had protection levels of ≥0% (Fig. 2, blue dashed line), our results suggest, for this study, a minimum protective level of 70% as the standard value for measuring the effectiveness of vaccines. However, if we consider a higher minimum protective value of ≥80% (Fig. 2, red dashed line), which is usually the minimum acceptable level used in licensing poultry AIV vaccines (14), our results show that none of the six individual vaccines we tested provides sufficient protection against all eight challenge HPAIVs. Despite that, we found that vaccine 3 (2.2.1.1/E3300/08) offered the broadest protection against different clades of Gs/GD viruses, providing protection against 6 out of the 8 challenge viruses (Fig. 1 and 2; see Table S1). Moreover, all six vaccines provided acceptable survival to meet the licensure requirement (≥80% protection level) against challenge viruses 3 (2.2.1/Egy102d/10) and 8 (7.1/V016/08) (Fig. 1 and 2; see Table S1).
Differences in vaccine protection against challenge with eight H5N1 HPAIVs from the Gs/GD lineage. To analyze changes in the protection provided by multiple vaccines against the same set of challenge viruses, we performed the Wilcoxon matched-pairs signed-rank test. Each vaccine is represented by a box plot and challenge viruses by numbers and squares. The box plot displays a summary of vaccine protection (%) including the median (horizontal line inside the box) as well the first and third quartiles. The lines extending from the boxes indicate the minimum and maximum protection values (whiskers). The white box plot represents the vaccine that was least efficient in protecting birds out of the six vaccines tested. The thresholds of 80% (red line) and 70% (blue line) vaccine protection are shown. *, P < 0.05.
All sham-vaccinated bird groups, the positive controls for the lethality of challenge viruses, had 100% mortality (Fig. 1), with MDTs ranging from 2 to 3.5 days (see Table S1). In all the combinations of vaccine strains and challenge viruses in which mortality was observed, MDT varied from 2 to 8.5 days (see Table S1). Further, for all the vaccine strains with low to intermediate survival rates (<70%), the average MDT was longer (the majority ranged from 4 to 8.8 days, except for one group in which it was 3 days) than for sham-vaccinated groups (MDT, 2.6 ± 0.4 days). In other words, in these birds, the vaccine delayed progression to death.
Evaluation of virus shedding.Virus shedding was detected at 2 days postchallenge (dpc) in oropharyngeal (OP) swabs from all the birds in the sham-vaccinated groups, independent of the challenge virus, with mean titers ranging between 4.6 and 7.0 log10 50% embryo infectious doses (EID50)/ml (Fig. 3; see Table S3 in the supplemental material). Birds that received one of the six vaccine strains had either no or very low levels of virus shedding, with mean virus titers ranging from 1.8 to 4.6 log10 EID50/ml (see Table S3). Moreover, our results showed that 44 of 48 combinations of vaccinated birds had statistically significantly lower OP titers than sham-vaccinated birds (P < 0.05) (Fig. 3). The four combinations that failed to decrease the level of shedding were related only to vaccine strains 2 (2.1.3.2/Indo/05) and 4 (2.3.2.1a/Hub/10) against challenge viruses 6 (2.3.2.1b/V672/11) and 7 (2.3.4/HK8825/08) (Fig. 3; see Table S3). Also, in 15 of 48 combinations, not only the titer, but also the number of birds shedding virus was significantly reduced compared to sham-vaccinated birds (see Table S3, asterisks).
Scatterplots of oropharyngeal virus shedding for vaccinated and sham-vaccinated groups. Viral titers are expressed as log10 EID50/ml, with error bars (standard deviation). The vaccinated versus sham-vaccinated groups are listed for each challenge virus (vaccines 1 through 6). The challenge viruses are represented by numbers, as shown in the legend on the right. *, P < 0.05.
Evaluation of virus shedding in the vaccinated groups also showed differences among all the challenge viruses used (see Fig. S1 in the supplemental material). In general, comparisons only in the vaccinated group showed that the majority of the birds challenged with virus 7 (2.3.4/HK8825/08) had significantly higher virus-shedding titers than did birds infected with the other challenge viruses used in this study (see Fig. S1). On the other hand, vaccinated birds challenged with virus 8 (7.1/Vn016/08) usually had the lowest levels of virus shedding for almost all the vaccine strains, with a few exceptions, such as challenge virus 4 (2.2.1.1a/Egy1063/10) for vaccine strains 1 (1.1/Camb/07) and 4 (2.3.2.1a/Hub/10) and challenge virus 3 (2.2.1/Egy102d/10) for vaccine strain 5 (2.3.4/Anh/05) (see Fig. S1).
Antibody levels in immunized birds pre- and postchallenge with HPAIV.Serum samples collected from all the birds before challenge (0 dpc) and the surviving birds after challenge (14 dpc) were analyzed by HI assay to measure humoral immune responses against homologous and heterologous antigens (see Fig. S2 in the supplemental material).
All the vaccinated birds had high titers of HI antibodies against homologous vaccine antigen (log2 geometric mean titers [GMT] ranging from 5.7 to 10) in the prechallenge sera, with the quantity maintained for the surviving birds after 14 dpc. The three exceptions were one bird in the group vaccinated with vaccine strain 1 (1.1/Camb/07) and challenge virus 1 (1.1/Vn118/08), one bird vaccinated with vaccine strain 1 (1.1/Camb/07) and challenge virus 6 (2.3.2.1b/Vn672/11), and one bird vaccinated with vaccine strain 2 (2.1.3.2/Indo/05) and challenge virus 6 (2.3.2.1b/Vn672/11). These three birds were excluded from further analyses (see Table S4 in the supplemental material).
To observe differences in the antibody responses, we tested sera from vaccinated birds with the specific challenge virus antigen (Fig. 4; see Table S4). A serum sample with an HI GMT of 3 log2 or more was considered seropositive. None of the sham-vaccinated birds had detectable HI antibody titers before challenge (data not shown). Most vaccinated birds had low or no antibody titer detected in the prechallenge sera, with GMT ranging from 2.0 to 3.2 log2, except for the homologous combination of vaccine strain 6 (7.1/Vn016/08) and challenge virus 8 (7.1/Vn016/08) (GMT, 6.9 log2). Thus, the antibody titer in the prechallenge sera was not associated with protection against mortality. However, the number of antibody-positive birds that survived was higher when at least 50% of the vaccinated birds in a group had detectable antibodies to the challenge virus in the prechallenge sera (see Table S4). All the vaccinated birds had statistically significantly higher HI titers in the postchallenge sera (GMT ranging from 3.4 to 8.5 log2) than in the prechallenge sera (P < 0.05) (Fig. 4), except for a few groups in which all the birds died (vaccine strain 2 [2.1.3.2/Indo/05] against challenge virus 7 [2.3.4/HK8825/08]) (Fig. 4B) or the number of live birds was insufficient for statistical analysis, such as with vaccine strain 2 (2.1.3.2/Indo/05) against challenge virus 2 (2.1.3/WJ29/07) (Fig. 4B) and vaccine strain 5 (2.3.4/Anh/05) against challenge virus 6 (2.3.2.1b/Vn672/11) (Fig. 4E).
Scatterplots of HI titers from individual vaccinated chicken prechallenge and postchallenge sera. The HI titers using challenge virus as the antigen are shown for each vaccine strain (vaccines 1 through 6). The HI titers were expressed as geometric mean titers (GMT-log2) with error bars included (standard deviation). Samples with titers below 3 log2 GMT (dotted horizontal line) were considered negative. ns, birds did not survive until 14 dpc, and samples were not collected. *, P < 0.05.
Molecular analysis of vaccine and challenge H5 avian influenza virus strains.The genetic divergence for the HA1 protein between vaccine strains and challenge viruses was evaluated (Fig. 5 and Table 2; see Fig S3 in the supplemental material), as well as specific amino acid changes in the antigenic sites (see Table S5 in the supplemental material) and predicted potential N-glycosylation sites (see Table S6 in the supplemental material).
Maximum-likelihood phylogeny of the hemagglutinin gene for each vaccine (VC) and challenge (CH) virus used in this study. The numbers along the branches indicate bootstrap values of >70%. Brackets indicate the genetic subclades. The scale bar indicates nucleotide substitutions per site.
HA1 amino acid identities between vaccine strains and challenge viruses
The paired HA1 amino acid identities between vaccine strains and challenge viruses varied from 84.7% to 100%, except for the homologous combination (vaccine strain 6 and challenge virus 8), 3 heterologous combinations (i.e., vaccine strain 1 and challenge virus 1, vaccine strain 3 and challenge virus 4, and vaccine strain 4 and challenge virus 5) had ≥97.2% homology (Table 2). Thus, a high HA1 amino acid identity between the vaccine strain and challenge virus alone did not provide a predictive value for protection.
In phylogenetic analysis, the vaccine and challenge strains of all the clades except 7.1 were not clustered together (Fig. 5; see Fig. S3A to F). The analysis of the antigenic sites in the HA proteins showed amino acid differences across the sites between strains, even from the same clade (see Table S5). Most of the amino acid changes were observed in the globular head of HA1 within or close to the antigenic-site regions (data not shown). Moreover, some of the changes in the antigenic sites correspond to the acquisition of potential N-linked glycosylation (see Table S6) that could potentially mask the regions from antibody access. Specifically, potential N-glycans were observed in antigenic sites B (amino acids [aa] 154 to 157) for vaccine strain 2 (2.1.3.2/Indo/05), D (aa 236 to 239) for both vaccine strain 3 (2.2.1.1/E3300/08) and challenge virus 4 (2.2.1.1a/Egy1063/10), E (aa 72 to 75) for vaccine strains 3 (2.2.1.1/E3300/08) and 6 (7.1/Vn016/08) and also challenge viruses 4 (2.2.1.1a/Egy1063/10) and 8 (7.1/Vn016/08), and between B and D (aa 195 to 198) for vaccine strain 6 and challenge virus 8 (7.1/Vn016/08) (see Table S6).
Antigenic relationships were analyzed with a cartography map (see Fig. S4 in the supplemental material) generated from the HI data, and no clear association was observed between antigenic distance, percent amino acid identity, and clinical protection (see Fig. S5 in the supplemental material).
To identify mutations in HA that influence vaccine protection, we compared the positions of the amino acid changes in the vaccine and challenge strains that occurred only when bird survivability was below 80%. Our results showed a pattern of a few amino acid mutations in the HA of each challenge virus strain regardless of the vaccine clade/virus used (Fig. 6). These amino acid substitutions were localized in the globular head of the HA protein, mostly within antigenic sites (Fig. 6A and B). The decrease in vaccine protection associated with challenge viruses from clades 1.1 and 2.1.3 was mostly associated with changes in epitopes of antigenic site A or E or adjacent to this region, while challenge viruses from clades 2.2.1.1a and 2.3.2.1 had mutations within antigenic sites A and B and close to these sites. In contrast, the most recent H5N1 challenge virus clades, 2.3.2.1b and 2.3.4, had epitope changes in all five immune-dominant antigenic sites (A to E), affecting the level of vaccine protection. Similar results were observed when the cutoff value for vaccine protection was ≥70% (Fig. 6B, amino acid positions circled in yellow), excluding a few amino acid positions. Among all the amino acid changes, those at position 140 (H5 HA numbering) were observed in nearly all the challenge viruses, with a few exceptions (Fig. 6A and B), regardless of the vaccine.
Molecular analyses of H5N1 avian influenza virus hemagglutinin for vaccine strains and challenge viruses. (A) Table showing in silico analyses of the amino acid changes associated with loss of vaccine protection (80% or above) in each challenge virus compared to the vaccines used in this study. Each position is listed as vaccine strain amino acid-position (H5 HA numbering)-challenge strain amino acid. The protection obtained in the in vivo experiment for each vaccine strain-challenge virus combination is shown as a percentage and color coded, with high and low protection shown as shades of green and red, respectively. The amino acids located in antigenic sites A (cyan), B (orange), C (magenta), D (green), E (yellow), or none (X) are identified. The red Xs are the epitopes previously listed as important for virus escape, and the black Xs are unknown in the literature. (B) Ribbon diagrams and surfaces of H5 HA. Only monomers are shown. The amino acids listed in panel A are shown in the HA structure. Predicted antigenic sites A (cyan), B (orange), C (magenta), D (green), and E (yellow) are labeled for each challenge virus. The amino acid positions circled in yellow were observed only if the threshold for vaccine protection was 70% or above.
DISCUSSION
The use of vaccines to control H5N1 Gs/GD lineage HPAIV infection in poultry, where the virus is endemic, is one of the main measures to ensure livelihoods and food security and decrease the negative economic impact of AIV (1, 13). Due to the rapid antigenic evolution of HA, virologic surveillance and in vivo experimental data are important to ensure that the antigenic content of vaccines is updated in a timely fashion to protect individual birds and control AIV. However, vaccine protection against AIV is a complex multifactorial issue.
The results from this study showed ≥80% efficacy against mortality in nearly half (47.92%) of the combinations of vaccine strains and challenge viruses in the study (Table 1; see Table S1). The protection of these vaccines against challenge viruses was even higher (58.33%), with a survival rate of ≥70%. In general, vaccine strain 2 (2.1.3.2/Indo/05) was less protective against H5N1 Gs/GD lineage challenges. Although regulatory requirements vary from country to country, taking a ≥80% efficacy threshold of protection as the minimum acceptable level for licensing of poultry AIV vaccines (14), all the H5N1 Gs/GD vaccine strains tested failed to protect against challenges with 2.3.2.1b and 2.3.4 HPAIVs. Even in the 25 combinations with lower levels of survival, the MDT was prolonged for most vaccinated groups compared to the sham-vaccinated groups (see Table S1), suggesting partial protection, but inadequate to prevent death. For example, vaccine strain 2 (2.1.3.2/Indo/05) provided only a 20% survival rate against challenge virus 2 (2.1.3/WJ29/07) and zero against challenge virus 7 (2.3.4/HK8825/08) and was accompanied by prolonged MDTs of 5.5 and 6 days, respectively, compared to MDTs of 3 and 2.5 days for the respective sham-vaccinated groups. While none of the individual vaccine strains protected against all HPAIV challenges, the vaccine strain from clade 2.2.1.1 (vaccine strain 3 [2.2.1.1/E3300/08]) was able to protect against six virus clades, but not 2.3.2.1b (challenge virus 6) and 2.3.4 (challenge virus 7). According to previous studies, sera and antigens of clade 2.2 avian influenza viruses appeared to be optimal for detecting most of the HPAIV H5N1 viruses and sera, respectively (15). Although new distinct clusters had emerged, such as H5N1 viruses from clade 2.2.1.2 in Egypt, no marked difference in antigenicity and receptor preference was identified in this group of viruses before 2014 (16), probably explaining why some vaccine strains were able to recognize the clade before that period (17). Several studies have suggested a strong link between genetic and antigenic distances, even though other factors could be involved.
Humoral responses to influenza vaccines are usually assessed by measuring antibodies against the circulating field virus HA protein by HI assay, and such results are interpreted as predictors of protection. Some studies have suggested that the presence of predetermined levels of HI antibodies would predict protection against the challenge virus used in HI assays (18, 19), but other studies also report that birds with low or no detectable levels of HI antibodies can survive some challenge viruses (11, 20–22). In our study, high HI antibody levels to the vaccine were predictive of survival for genetically identical or closely related vaccine strain-challenge virus combinations but not predictive of survival when the challenge virus was antigenically divergent from the vaccine seed strain. Furthermore, in a previous study (14), the detection of HI antibodies using challenge virus as an HI antigen was associated with protection, i.e., it had positive predictive value (PPV). Similar PPV results against mortality using challenge virus were observed in the present study when the HI titers were ≥32 (GMT, 5 log2) (data not shown), which is in agreement with the World Organization for Animal Health (formerly Office International des Epizooties [OIE]) efficacy requirement for new vaccines in terms of achieving mortality reduction (23). In contrast, some divergent vaccine strain-challenge virus combinations lacked detectable HI antibodies, but birds were still clinically protected (i.e., poor negative predictive value [NPV] for lack of HI titers and mortality), suggesting possible protection by humoral non-HI antibodies, cellular immunity, or a combination of the two. Although inactivated vaccines are poor inducers of cellular immunity (13), some minor induction of cellular immunity may have contributed to protection. A previous study found that an inactivated H5N2 North American low-pathogenic avian influenza virus (LPAIV) vaccine formulation provided efficient vaccine protection and reduced virus shedding against challenge with a Gs/GD lineage clade 2.2.1 virus from Egypt (21) despite a low level of antigenic cross-reactivity by HI. In contrast, a comprehensive study with H5 vaccines against several genetically and antigenically dissimilar Indonesian H5 Gs/GD lineage 2.1.3 clade field strains concluded that the presence of HI antibodies against the vaccine antigen was a poor positive predictor of survival after lethal challenge (14). Also, several studies noted a lack of association between vaccine-induced HI titers and protection in chickens for H5 (11, 24) and H7 (20, 25) and mice for H6 influenza A viruses (26). Therefore, even though detection of HI antibodies against a challenge virus is usually reported as a proxy for protection (i.e., PPV), HI antibodies are not the only active immune effectors against infection, and other parameters, such as humoral Fc-mediated functions and nonhumoral immunity to other conserved influenza virus proteins, may play some role in protection (13, 27, 28).
In this study, we also tested 78 serum samples by virus neutralization assay (data not shown) because the HI assay alone does not detect all protective antibodies, such as those against the fusion domain of HA or other proteins, such as neuraminidase. However, no clear association of the presence of neutralizing antibodies and survival efficacy could be found, as several birds with neutralizing antibodies (even with titers as high as 256) died after challenge. Based on our limited sample set, the results of the neutralization assays by themselves were not enough to explain why some groups lacking neutralizing antibodies were protected and others were not protected despite the presence of neutralizing antibodies. Thus, they reinforce the complexity of the immune response required to provide full protection after challenge with antigenic and genetically diverse, rapid, and systemic replicating HPAIVs. Similar results with a lack of correlation of the HI or neutralizing antibody data with protection have been reported in humans against H5N1 infection (29, 30).
In general, homologous or closely related vaccine virus-challenge virus pairs in the same genetic clade had the best protection across multiple metrics (Fig. 5). The homologous vaccine and challenge performed with 7.1/V016/08 had a 100% survival rate, and only two birds shed low levels of virus. Vaccine strain 3 (2.2.1.1/E3300/08) protected against mortality and significantly decreased virus shedding and the number of birds shedding with the two most closely related challenge viruses, 3 (2.2.1/Egy102d/10) and 4 (2.2.1.1a/Egy1063/10). Birds vaccinated with 2.3.2.1a/Hub/10 (vaccine strain 4) were protected against mortality and presented a significant decrease in virus shedding when challenged with viruses 6 (2.3.2.1b/Vn672/11) and 5 (2.3.2.1/Vn398/10). Vaccine strain 1 (1.1/Camb/07) protected against mortality and decreased virus shedding after challenge with virus from the same clade (1.1/Vn118/08) but did not significantly decrease the number of birds shedding. Despite these results, not all the most closely related vaccine strain-challenge virus combinations provided acceptable protection. Birds vaccinated with 2.1.3.2/Indo/05 (vaccine strain 2) and challenged with 2.1.3/WJ29/07 (challenge virus 2) (96.3%) had low survivability (20%), and although the titer of shed virus was significantly decreased at 2 days postinfection (dpi), the number of birds shedding was equivalent to that for the sham-vaccinated controls. Although our study demonstrated that most vaccine seed strains reduced challenge virus shedding at 2 dpi, the peak of virus replication and shedding, similar trends in reduction of oropharyngeal shedding have been reported for samples collected at later times postinfection (13, 14, 19, 31–33).
While the levels of HA1 amino acid identity between vaccine and challenge strains for all the combinations ranged from 84.7% to 100%, the level of HA amino acid sequence identity was not consistently a good predictor of survival (except for combinations with ≥97.2% identity). Ten combinations that had low survival rates (0 to 40%) had identities between vaccine and challenge strains ranging from 87.2% to 96.3%. This finding was not unexpected, since previous studies had similar results (24). For example, chickens vaccinated with vaccine strain 2 (2.1.3.2/Indo/05) and challenged with the closely related HPAIV 2 (2.1.3/WJ29/07; 96.3% HA1 homology) had only a 20% survival rate. In contrast, a 90% survival rate was obtained with vaccine strain 4 (2.3.2.1a/Hub/10) followed by challenge with virus strain 8 (7.1/Vn016/08) (85.9% homology) and a 100% survival rate with vaccine strain 5 (2.3.4/Anh/05) when challenged with strain 8 (7.1/Vn016/08) (86.6% homology). This suggests that specific changes in critical antigenic sites might be a better predictor of protection than the overall sequence identity of HA proteins. The results of in silico analyses associated with in vivo vaccine strain-challenge virus combinations highlighted the importance of specific polymorphisms observed during H5N1 AIV evolution. Some important positions for antigenic-drift mutations that allow the 2002 H5N1 virus to escape antibody neutralization, such as positions 86, 94, 124, 140, and 189 (34), were detected in our study as important epitopes, alone or in combination, that affect vaccine protection regardless of the clades of the H5N1 challenge viruses. In general, we found that mutations at position 140 (H5 HA numbering) were predominant and strongly associated with loss of vaccine protection in almost all combinations of vaccine and challenge strains. The amino acid at position 140, located in antigenic site A, was previously characterized in different H5N1 strains and a panel of chimeric and mouse monoclonal antibodies that target the receptor binding site (RBS) in HA (35–39). This position is important, as it impacts the H5 HA structure involving specific monoclonal antibody recognition and mediates neutralization escape (35–39). Similarly, changes in position 162 were also detected and were negatively associated with vaccination protection. Other studies showed amino acid 162 in HA to be part of a neutralization epitope outside the RBS and in proximity to site B (40, 41).
Sequence alignments of representative viruses from each clade and sequence variations mapped onto the H5N1 HA structure have demonstrated that the most variable interclade positions are largely located on the membrane distal globular head close to the RBS (42). The identification of these critical antigenic epitopes, or hot spots, highlighted the complexity of the conformational epitopes on the H5 HA structure involved in antibody recognition and escape neutralization. Even though several groups have reported epitope mapping in H5 HA, much of the information pertains to antigenic epitopes reactive to mouse antibodies rather than human or chicken antibodies (42). Therefore, knowledge of human and chicken immunodominant regions is important to proactively develop diagnostic tests and improve surveillance platforms to monitor AIV outbreaks and design more efficient and broad-spectrum preoutbreak vaccines.
To characterize the antigenic relationships among vaccine strains and challenge viruses, we used antigenic cartography based on HI assay reactivity (see Fig. S4). Viruses from within each clade cluster close to each other, although variation was still observed with each clade and subclade, as in previous studies (43, 44). However, it was not possible to establish a clear association between antigenic distance or location on the map and the prediction of clinical protection or high HA1 amino acid homology in these in vivo studies (see Fig. S5) with the specific panel of chicken antisera used. Previously, with a few exceptions, a strong association between antigenic distances for vaccine and challenge viruses and clinical protection against challenge for distances of less than 4 antigenic units were reported for HPAI in chickens (45). Other studies also suggested that antigenic cartography could be an important tool to characterize and map virus diversity (46), but a lack of association between clinical protection and specific antigenic differences, visualized by antigenic cartography alone, has been previously reported for H5, H7, and H9 AIVs in chickens (20, 31, 47) and ducks (27), and the absence of HI reactivity, especially, has been a predictor of lack of protection.
Prior research has shown that amino acid changes in the antigenic sites and glycosylation composition in HA can alter the antigenicity of AIV (46, 48–50). The evaluation of the antigenic sites displayed specific amino acid changes across sites A through E (35, 36, 51) in the vaccine and challenge strains used in this study. However, no direct association was observed between the complete antigenic site sequence and survivability—not even a decrease in oropharyngeal virus shedding. However, some amino acid changes were associated with additional N-glycosylation sites, which also included positions 72 to 75, 154 to 157, 195 to 198, and 236 to 239 (H5 HA numbering) within the antigenic sites (Fig. 6). To our surprise, vaccine strain 2 (2.1.3.2/Indo/05) was the only tested vaccine strain with N-linked glycosylation at the in silico-evaluated antigenic site B (positions 154 to 157) (see Table S6). This vaccine strain demonstrated the lowest protection rate against the tested challenge viruses, regardless of the high protein sequence identity. It has been recognized that the presence of N-glycosylation on the globular head of HA can mask or modify antigenic sites recognized by neutralizing antibodies (49). In a previous study with ferrets, masking of the antigenic epitopes at positions 154 to157 in the HA globular head of A/Vietnam/1203/2004(H5N1) vaccine virus by N-linked glycosylation affected virus antigenicity, resulting in a weaker antibody response (52). The presence of N-linked glycosylation at the tested antigenic site at positions 154 to 157 observed in the 2.1.3.2/Indo/05 vaccine strain could be a factor involved in the poor protection seen against some challenge viruses with high protein sequence homology with the vaccine strain. The addition of three N-glycosylation sites on a 2015 emergent H7N3 HPAI field virus was associated with a decrease in protection from a recombinant fowl poxvirus vaccine strain containing a 2002 H7 HA gene insert (32). This vaccine strain gave good protection against the 2012 H7N3 HPAIV, which had four glycosylation sites, as does the 2002 vaccine strain (32). Therefore, we should consider the potential N-glycosylation sites to select AIV vaccine candidates and vaccination effectiveness against AIV field strains. Future studies are necessary to understand how the predicted N-glycosylation sites identified in this study can influence the efficacy of protection.
Before the widespread and long-term use of AI vaccines in poultry, researchers had suggested that an HA identity of >85% could be a good predictor of protection (53). However, virus evolution and the massive use of vaccination could have contributed to virus evasion mechanisms. This would have been possible with only a few critical genetic and antigenic changes in specific sites of HA in order to overcome immunity. Our findings suggested that a combination of amino acid mutations (in contrast to a single substitution) are usually involved in overcoming vaccine immunity. Most of the clade-specific sequence variations identified in the HA of H5 Gs/GD were localized at the antigenic sites and reported epitopes associated with virus escape from antibody neutralization (42). Our data show differences in the patterns of amino acid substitutions between Gs/GD virus clades, which highlighted important regions of immune pressure in these H5 field HPAIVs. Studies have suggested that antigenic evolution in AIV involves transitions from one antigenic cluster to another (44, 54), as we observed in these results. Further, the sustained circulation of H5N1 strains has led to the evolution and selection of distinct viral lineages with significant variation in the HA sequence (5, 15, 34, 44), which also explained the increase of epitopes involved in the escape of vaccine protection observed in clade 2.3.2.1b and 2.3.4 AIVs.
Our goal with this study was to test vaccine candidates originally included in the human pandemic preparedness program and to evaluate whether they could protect a chicken model against challenge viruses relevant for poultry in regions of endemicity. We conclude that the effectiveness of the vaccine is dependent on the challenge virus, especially specific epitopes in the antigenic sites. Although most vaccines could reduce oropharyngeal virus shedding, only half of the vaccine strain-challenge virus combinations had survivability of ≥80% (Table 1). The lack of consistent association between HI titers and protection for all vaccine strain-challenge virus combinations, as well as homology of HA1 and antigenic sites with clinical protection, reinforces that protection induced by killed vaccine strains against the H5N1 HPAIV Gs/GD lineage in chickens involves a complex process. Several variables are involved, and a single criterion, such as specific genetic and/or antigenic differences or HI results, cannot by itself predict clinical protection. On the other hand, we were able to identify HA epitopes in the challenge viruses that correlated with vaccine protection. Most of the mapped high-mutation sites observed were previously associated with virus escape from antibody neutralization, which supports the direct correlation of our findings connecting the in vivo and in silico results. Our data also reinforce the need for further studies to better understand the antigenic-drift mutations in the immune-dominant regions in HA and other AIV proteins involved in virus escape from vaccine protection.
MATERIALS AND METHODS
Vaccine strains and challenge viruses.All the vaccine seed strains used in this study (Table 1) are H5N1 vaccine strains produced by reverse genetics with PR8 internal gene segments. The vaccines were obtained from the U.S. Centers for Disease Control and Prevention (CDC), Influenza Division, Atlanta, GA, USA, and the National Institute for Biological Standards and Controls, Hertfordshire, United Kingdom, as part of the WHO prepandemic vaccine preparedness program. The isolates were propagated in specific-pathogen-free (SPF) 9- to 11-day-old embryonated chicken eggs (ECE) by following standard procedures (55). Allantoic fluid harvested from infected eggs was inactivated with 0.1% beta-propiolactone and used as an antigen for vaccine production and serological assays. All the vaccines were adjuvanted by oil-in-water emulsion formulation using mineral oil, as previously described (56). The antigens were adjusted to have all the vaccine with the same antigen concentration as measured in hemagglutinating units (HAU) (512 HAU per 0.5-ml dose). Use of between 256 and 512 HAU per dose in poultry vaccine is standard across experimental studies and in veterinary biological manufacturing. Sham-vaccinated group vaccines were produced using sterile allantoic fluid from uninfected embryonated chicken eggs and the same formulation.
Eight H5N1 HPAIV isolates (Table 1) were selected as challenge viruses based upon distinct clades of Gs/GD from poultry found in some areas of endemicity, such as Asia, Africa, and the Middle East. The isolates were propagated using the same procedures described for the vaccines.
Animals and housing.Vaccinations were performed in animal biosafety level 2 (ABSL2) facilities at the Southeast Poultry Research Laboratory (SEPRL), Agricultural Research Service, U.S. Department of Agriculture (ARS/USDA). The birds were transferred before challenge to the animal biosafety level 3 enhanced (ABSL3E) facilities at SEPRL, where they were housed in negative-pressure HEPA-ventilated isolation cabinets and had ad libitum access to food and water throughout the experiment. All the studies were reviewed and approved by the U.S. National Poultry Research Center (USNPRC) Institutional Animal Care and Use Committee (IACUC) and conducted with appropriate measures to maintain biosecurity and biosafety.
Experiment design.Experiment design and sampling were done using animal, vaccination, and challenge protocols under conditions identical to those previously published by our group. In summary, a total of five experiments were performed using six different vaccine strains against eight Gs/GD H5N1 HPAIV challenge viruses (a total of 48 combinations with an initial 10 birds per group) (Table 1). A sham-vaccinated group was included in each experiment as an unvaccinated challenge control, and all data comparisons were performed using the specific sham-vaccinated group for the appropriate challenge virus. All the birds were 3-week-old SPF white leghorn chickens from SEPRL in-house flocks. The vaccines were administered via the subcutaneous route in the nape of the neck as a 0.5-ml dose per bird. Three weeks postvaccination, the birds were bled to evaluate the serological response and challenged by the intranasal route via the choanal cleft with a target of 106 EID50 in 0.1 ml per bird. The inoculum titer was verified by back titration in ECE for challenge virus (different studies were performed, so back-titration titers are provided as ranges for the challenge viruses that had different back titers in different studies): (i) 1.1/Vn118/08, 105.9 to 106.1 EID50/0.1 ml; (ii) 2.1.3/WJ29/07, 105.9 to 106.1 EID50/0.1 ml; (iii) 2.2.1/Egy102d/10, 105.9 EID50/0.1 ml; (iv) 2.2.1.1a/Egy1063/10, 105.9 to 106.1 EID50/0.1 ml; (v) 2.3.2.1/Vn398/10, 105.9 to 106.1 EID50/0.1 ml; (vi) 2.3.2.1b/Vn672/11, 105.9 to 106.1 EID50/0.1 ml; (vii) 2.3.4/HK8825/08, 105.7 to 105.9 EID50/0.1 ml; (viii) 7.1/Vn016/08, 106.1 EID50/0.1 ml. Two days postchallenge, OP swabs were collected and placed in Becton Dickinson BBL brain heart infusion (BHI) medium with 2× antibiotics (penicillin-streptomycin-Fungizone; HyClone, Logan, UT, USA) and stored at −80°C until they were tested to determine shed virus titers. All the birds were observed daily, and clinical signs and mortality were recorded from 0 to 14 dpc. Birds presenting severe clinical signs, e.g., unresponsive or unable to reach food and water, or other HPAI infection signs were euthanized for humane reasons due to severe disease and were counted as deaths on the following day for MDT calculations. Fourteen days after challenge, all the surviving birds were bled to evaluate antibody titers postchallenge and euthanized according to the approved IACUC protocol.
Virus shedding.OP swabs collected at 2 dpc were processed for quantitative real-time PCR (qRT-PCR) to determine the virus shed from the oropharynx after challenge. We used a standard protocol that demonstrated the high correlation between qRT-PCR and the infectious titer because we (i) used only influenza A challenge viruses that were chicken adapted and propagated in ECE; (ii) used a low multiplicity of infection (MOI) to inoculate ECE in propagating the viruses to generate the challenge inoculum, which minimizes defective interfering (DI) RNAs; and (iii) used the same specific challenge virus stock to generate the standard curve with each dilution point, directly comparing qRT-PCR results and infectious titers (57). This methodology has been used in several published veterinary influenza vaccine studies (57–59). Briefly, RNA was extracted using a MagMax 96 AI/ND viral RNA isolation kit (ThermoFisher Scientific, Carlsbad, CA, USA) following the manufacturer’s instructions. Further, qRT-PCR targeting the matrix gene of avian influenza virus was performed with an AgPath-ID OneStep RT-PCR kit (ThermoFisher Scientific) using a 7500 FAST real-time PCR system (Applied Biosystems, Foster City, CA), as previously described (60). The virus quantity was established with a standard curve from RNA extracted from 10-fold dilutions of the challenge virus in duplicate. The limits of detection obtained for each challenge virus were as follows: (i) 1.1/Vn118/08, 2.5 log10 EID50/ml; (ii) 2.1.3/WJ29/07, 1.7 log10 EID50/ml; (iii) 2.2.1/Egy102d/10, 2.3 log10 EID50/ml; (iv) 2.2.1.1a/Egy1063/10, 2.5 log10 EID50/ml; (v) 2.3.2.1/Vn398/10, 2.5 log10 EID50/ml; (vi) 2.3.2.1b/Vn672/11, 2.5 log10 EID50/ml; (vii) 2.3.4/HK8825/08, 2.3 log10 EID50/ml; and (viii) 7.1/Vn016/08, 1.9 log10 EID50/ml. For statistical analysis, negative samples were considered lower than the limit of detection mentioned above for each specific challenge virus tested.
Hemagglutination inhibition assay.Sera collected pre- and postchallenge were evaluated for specific antibodies using the appropriate vaccine or challenge virus antigens in the HI assay. The antigens were prepared as previously described (61), and the HI assay was performed following standard procedures (62). Titers were calculated as the reciprocal of the last HI positive serum dilution and were converted to log2. Titers were expressed as log2 GMT. Samples were considered positive for the presence of AIV antibodies at the limit of detection (1:8 dilution, or 3 log2). Negative samples were assigned as 2 log2 GMT for statistical purposes.
Antigenic cartography.All the vaccine strains and challenge viruses plus additional reference strains were antigenically characterized and mapped with statistical robustness tests as previously described (63). The antigenic maps were produced from HI assay data. The HI assays were performed as previously described (62) using isolate-specific polyclonal chicken sera produced in house to serve as monospecific reference antibodies. The antiserum was produced by vaccinating chickens with an oil emulsion vaccine prepared with Montanide ISA 50V adjuvant (SEPPIC, Inc., Paris, France) using infectious allantoic fluid inactivated with 0.1% beta-propiolactone. Five-week-old SPF chickens were vaccinated with 0.5 ml of the vaccine by the subcutaneous route, and sera were collected 3 weeks after vaccination. The sera were treated with 5% chicken red blood cells for 30 min at room temperature to decrease nonspecific reactions.
Molecular analysis of HA.The HA segments of vaccine and challenge viruses used in this study were aligned using MUSCLE (64), and amino acid sequences at the previously reported antigenic sites of HA1 were evaluated (31, 46). All available HPAIV clade 1.1, 2.1.3, 2.2.1, 2.3.2.1, 2.3.4, and 7.1 HA gene sequences were downloaded from the Influenza Virus Resource database (https://www.ncbi.nlm.nih.gov/genomes/FLU/FLU.html) on 22 February 2017. The complete HA sequences of clade 2 viruses were further pruned using the cd-hit software (65) to remove redundant sequences at a 99.3% to 99.7% similarity level. The final data set consisted of 65 taxa for clade 1.1, 119 for clade 2.1.3, 112 for 2.2.1, 115 for 2.3.2.1, 106 for 2.3.4, and 82 for 7.1.
The maximum-likelihood (ML) tree of each clade (see Fig. S3) was generated with RAxML (66) using the general time-reversible (GTR) model of nucleotide substitution with gamma-distributed rate variation among sites (with four rate categories). Statistical analysis of the phylogenetic tree was performed by bootstrap analysis with 1,000 replicates. Potential N-glycosylation sites were predicted using NetNGlyc server 1.0 (http://www.cbs.dtu.dk/services/NetNGlyc).
We created a series of visualizations in the R software (version 3.5.0) using the ggplot2 package (67) to help identify the amino acid changes that were correlated with a loss of effectiveness in vaccine protection using thresholds of 80% and 70% survival. Each challenge virus strain was compared to the vaccine strains to identify the amino acid substitutions. Only amino acid changes that occurred in the combinations of vaccine and challenge strains with bird survival below 80% or 70% were considered. The amino acid mismatch was excluded from the final analyses (see Fig. 6A) if it was observed simultaneously in the effective and ineffective vaccinations. The H5 HA subtype numbering conversion was obtained in the Influenza Research Database using the closest reference sequence (A/Vietnam/1203/04; HPAI) suggested by the website.
The HA structure was modeled using the H5 HA template (Protein Data Bank [PDB] accession number 2FK0) in the SWISS-MODEL server (68, 69). The three-dimensional (3D) molecular HA structures were visualized using the PyMOL Molecular Graphics System (version 2.0; Schrödinger LLC).
Statistical analysis.Statistical analyses were performed using Prism 7 (GraphPad Software, San Diego, CA, USA). The survival rate data were analyzed using the Mantel-Cox log rank test. Fisher’s exact test was used to analyze the statistical significance of virus shedding comparisons among vaccinated and sham-vaccinated groups in experiments. Also, one-way analysis of variance (ANOVA) using Kruskal-Wallis and Dunn's multiple-comparisons tests was performed to evaluate virus shedding between vaccinated groups. Statistical differences in mean viral titers and antibody levels between groups were analyzed using the Mann-Whitney test and Tukey one-way ANOVA. To analyze changes in the protection of multiple vaccines against the same set of challenge viruses, we performed the Wilcoxon matched-pairs signed-rank test, a nonparametric version of the dependent t test. A P value of <0.05 was considered significant.
Data availability.All the hemagglutinin gene sequences used in this study were downloaded from the Influenza Virus Resource Database (https://www.ncbi.nlm.nih.gov/genomes/FLU/FLU.html). The avian influenza virus vaccine strains and challenge viruses tested are available in GenBank under accession numbers listed in Table 1. The H5 HA structure template is available in PDB under accession number 2FK0.
ACKNOWLEDGMENTS
We thank Kira Moresco, James Doster, and Scott Lee for excellent technical support. Also, we thank Roger Brock and Ronald Graham for excellent animal care assistance.
This research was supported by ARS/USDA (projects 6040-32000-063-00D and 6040-32000-066-00D) and Centers for Disease Control and Prevention agreements (grants 13FED1310143, 17FED1712072, and 18FED1812056IPD). Xiu-Feng Wan was supported by the National Institutes of Health (grant 1R01AI116744-01). The funders had no role in study design, data collection and interpretation, or the decision to submit the work for publication.
The contents of this article are solely our responsibility and do not necessarily represent the official views of the USDA or NIH. Mention of trade names or commercial products in this publication is solely for the purpose of providing specific information and does not imply recommendation or endorsement by the U.S. Department of Agriculture.
We declare that we have no conflict of interest.
FOOTNOTES
- Received 24 April 2020.
- Accepted 22 September 2020.
- Accepted manuscript posted online 30 September 2020.
Supplemental material is available online only.
This is a work of the U.S. Government and is not subject to copyright protection in the United States. Foreign copyrights may apply.
