Antigenic and Genetic Evolution of Equine Influenza A (H3N8) Virus from 1968 to 2007^,

Abstract

Equine influenza virus is a major respiratory pathogen in horses, and outbreaks of disease often lead to substantial disruption to and economic losses for equestrian industries. The hemagglutinin (HA) protein is of key importance in the control of equine influenza because HA is the primary target of the protective immune response and the main component of currently licensed influenza vaccines. However, the influenza virus HA protein changes over time, a process called antigenic drift, and vaccine strains must be updated to remain effective. Antigenic drift is assessed primarily by the hemagglutination inhibition (HI) assay. We have generated HI assay data for equine influenza A (H3N8) viruses isolated between 1968 and 2007 and have used antigenic cartography to quantify antigenic differences among the isolates. The antigenic evolution of equine influenza viruses during this period was clustered: from 1968 to 1988, all isolates formed a single antigenic cluster, which then split into two cocirculating clusters in 1989, and then a third cocirculating cluster appeared in 2003. Viruses from all three clusters were isolated in 2007. In one of the three clusters, we show evidence of antigenic drift away from the vaccine strain over time. We determined that a single amino acid substitution was likely responsible for the antigenic differences among clusters.

INTRODUCTION

Influenza A viruses have a negative-sense RNA genome consisting of 8 segments. The viruses are subtyped according to the combination of two surface glycoproteins, hemagglutinin (HA) and neuraminidase (NA), and, to date, 16 HA and 9 NA influenza subtypes have been identified (1, 13, 16, 20, 30). Waterfowl appear to be the natural reservoir of all influenza A virus subtypes, and infections are generally nonpathogenic; however, these viruses can cause substantial morbidity and mortality upon transmission to other species (39).

Two subtypes of influenza A viruses have been isolated from horses, H7N7 and H3N8. In Prague in 1956, equine influenza A (H7N7) virus was isolated during a widespread epidemic of respiratory disease in horses in Eastern Europe (34). This subtype subsequently infected horses worldwide. Although there have been anecdotal reports of H7N7 outbreaks in Egypt and India and detection of H7-specific antibodies in reportedly unvaccinated horses elsewhere (23, 38, 42), the H7N7 subtype has not been isolated since 1979.

The currently circulating influenza A (H3N8) viruses were first isolated in 1963, when a group of thoroughbred horses exhibiting signs of respiratory disease arrived in Miami by air from Argentina. The causative agent was a previously unknown influenza virus, now known as influenza A (H3N8) virus (32). Subsequently, equine influenza A (H3N8) virus has caused episodic outbreaks in horse populations throughout the world, leading to substantial disruption to and economic losses for equestrian industries (3, 14, 17, 36). The Australian Bureau of Agricultural and Resource Economics estimated that costs during the initial response to an equine influenza outbreak in Australia in 2007 were AU$560,000 per day for disease control and AU$3.35 million per day in lost income for equestrian, farming, and recreation industries (4).

Initially, equine influenza A (H3N8) viruses evolved as a single phylogenetic lineage (19). Daly et al. (6) showed that two phylogenetic lineages, the American and Eurasian lineages, named according to the geographic origin of the isolates, emerged in the late 1980s. Using viruses isolated in North America, Lai et al. showed divergence of the American lineage into three sublineages: a South American lineage, a Kentucky lineage (corresponding to the American lineage demonstrated by Daly et al. [6]), and a Florida lineage (21). Further evolution of the Florida lineage (or sublineage, as it is now known) has resulted in the emergence of two groups of viruses with divergent HA sequences: Florida sublineage clade 1, containing influenza A/Equine/Wisconsin/03 (A/eq/Wisconsin/03)-like viruses, and Florida sublineage clade 2, containing A/eq/Newmarket/5/03-like viruses (24).

For influenza vaccine strain selection, however, antigenic differences are accorded more weight in the decision-making process than genetic differences because the humoral immune response is mounted against the virus phenotype, not the genotype. The gene segment coding for HA is of major importance because HA is the primary target of the protective immune response (40) and immune system recognition is potentially the key selection pressure on influenza H3 virus in humans (15). Hence, HA is the focus of equine influenza virus surveillance and the primary component of equine influenza virus vaccines. Changes to the antibody binding sites of the HA protein can allow viruses to evade the host's immune response, a process called antigenic drift, and can lead to vaccine failure in the field (2, 6, 8, 25, 33, 43; J. A. Mumford, presented at the 4th International Meeting of OIE Experts on Control of Equine Influenza, Newmarket, United Kingdom, 2003).

As a result of antigenic drift, a vaccine strain may fail to induce protective immunity unless the vaccine is updated and the animal is revaccinated (43). Infection in individuals with partial immunity can lead to virus shedding, can occur in vaccinated animals in the absence of clinical signs, and is more likely to occur if the vaccine strain is antigenically dissimilar to the infecting isolate (8, 26, 27). Animal challenge studies are considered the gold standard to test the degree of clinical protection afforded by vaccine strains in horses (7, 12).

Antigenic changes in influenza virus are routinely measured using the hemagglutination inhibition (HI) assay (18). Antigenic cartography is a computational method that enables high-resolution quantitative analyses and visualizations of HI assay data (9, 22, 31, 33).

Here we quantify the antigenic and genetic evolution of equine influenza A (H3N8) virus by using a larger and more globally representative panel of viruses than has been previously analyzed. Integrating antigenic data with HA1 domain sequences, we examine the genetic basis for antigenic differences among circulating influenza A isolates and identify amino acid substitutions that might lead to immune escape and vaccine breakdown.

MATERIALS AND METHODS

Viruses.

All viruses used in this study (n = 107) were collected by World Organization for Animal Health (OIE) reference laboratories for equine influenza between 1968 and 2007. The viruses were propagated in the allantoic cavity of embryonated hens' eggs, except for A/eq/Moulton/98, which was isolated and grown in Madin-Darby canine kidney (MDCK) cells, as it had previously failed to replicate adequately in eggs. Harvested allantoic fluid was stored at −70°C.

Virus antigenic characterization.

HI assays using postinfection ferret antisera were performed to compare the antigenic properties of equine influenza A (H3N8) viruses as previously described (6). Ferret antisera were raised against 36 representative virus isolates, collected between 1963 and 2003, by intranasal instillation of infectious allantoic fluid. Sera were collected 3 weeks later. Each antiserum was treated with potassium periodate, followed by incubation at 56°C for 1 h, to remove nonspecific inhibitors of hemagglutination. Serial 2-fold dilutions, starting at a 1:8 dilution, were tested for their ability to inhibit the agglutination of chicken erythrocytes with 4 hemagglutinating units of equine influenza A (H3N8) virus. All HI assays were performed at least in duplicate. See the supplemental material for a list of viruses and reference sera.

Antigenic cartography.

The quantitative analyses of the antigenic properties of equine influenza A (H3N8) viruses were performed using antigenic cartography, previously used for human (H3) and swine (H3 and H1) influenza A viruses (9, 22, 33). In an antigenic map, the distance between antiserum point S and antigen point A corresponds to the difference between the log₂ of the maximum titer observed for antiserum S against any antigen and the log₂ of the titer for antiserum S against antigen A. Therefore, each titer in an HI table can be thought of as specifying a target distance for the points in an antigenic map. Modified multidimensional scaling methods are then used to arrange the antigen and antiserum points in an antigenic map to best satisfy all the target distances specified by the HI data. The result is a map in which the distance between points best represents antigenic distance as measured by the HI assay. Because antisera are tested against multiple antigens and antigens are tested against multiple antisera, many measurements are used to determine the positions of the antigen and antiserum points in an antigenic map, thus potentially increasing the accuracy of point placement beyond that of individual HI measurements.

Cross-validation prediction experiments (in which a percentage of the titers were removed from the HI table and the map was used to predict the removed titers) were used to investigate the optimum number of dimensions of the map that was most suitable for the data set used here. The average magnitude of the prediction error for the two-dimensional map was 0.71 of an antigenic unit (AU) (standard deviation [SD], 0.90; correlation between predicted and measured values, 0.85 [P < 0.01]), where 1 AU is equivalent to a 2-fold difference in the HI assay titer. In addition to cross-validation prediction experiments using a 2-dimensional map, we also predicted the average magnitude of error from 3-, 4-, and 5-dimensional maps. The average magnitude of the prediction error for the three-dimensional map decreased to 0.67 AU (SD, 0.88; correlation between predicted and measured values, 0.86 [P < 0.01]). As the number of dimensions increased from 3 to 4 and 5 dimensions, the average magnitude of the prediction error, the average SD, and the correlation between predicted and measured values remained the same, indicating that distortion due to compression from higher dimensions to 3 dimensions was small. Thus, a 3-dimensional antigenic map was used for the remainder of these analyses. Clusters were defined with a k-means clustering algorithm using average weighting and a k value of 3 (33). See the supplemental material for further details on k-means clustering. The three-dimensional antigenic and genetic maps were displayed using the PyMOL molecular graphics system (10).

Genetic analyses: sequencing.

HA1 domain sequences were obtained for all 107 viruses in the data set: 58 were obtained from GenBank, 39 were generated by routine surveillance at the Animal Health Trust, and 10 were sequenced for this study. For the 10 sequences obtained as part of this study, viral RNA (vRNA) was extracted using a Qiagen QIAamp viral RNA kit according to the manufacturer's instructions. Reverse transcription-PCR (RT-PCR) was carried out with a Finnzymes kit (RobusT I RT-PCR) in one step, using sense (eqH3HA1/1, 5′-AGCAAAAGCAGGGGATATTTCTG-3′) and antisense (eqH3HA1/2, 5′-GCTATTGCTCCAAAGATTC-3′) primers and incubating the vRNA samples and a 1× final concentration reaction mixture at 50°C for 30 min and then at 94°C for 2 min, followed by 39 cycles of 94°C for 15 s, 50°C for 30 s, and 70°C for 1 min, and finally at 72°C for 7 min. PCR products were purified using the QIAquick PCR purification kit protocol and sequenced using BigDye v3.1 terminator chemistry (Applied Biosystems, CA) and an ABI Prism 3100 genetic analyzer according to the manufacturer's instructions. The sequences were edited and aligned using SeqMan (SeqMan 5.03; DNAStar, Inc.), EditSeq 5.03 (DNAStar Inc.), BioEdit (BioEdit sequence alignment editor 6.0.5; Isis Pharmaceuticals, Inc.), and ClustalW (23) (EMBL-EBI; www.ebi.ac.uk). All accession numbers are listed at the end of Materials and Methods.

Phylogenetic analysis.

A maximum-likelihood (ML) phylogenetic tree for the HA1 domain nucleotide sequences was inferred using PAUP* (version 4.0b10) (35) with GTR+I+Γ₄ (the general time-reversible model, with the proportion of invariant sites and the gamma distribution of among-site rate variation with four categories estimated from the empirical data) as determined by ModelTest (28). Global optimization of the tree topology was performed by tree bisection-reconnection branch swapping. The robustness of individual nodes of the tree was assessed using a bootstrap resampling analysis (100 replicates, with topologies inferred using the neighbor-joining method under the GTR+I+Γ₄ substitution model).

HA1 domain amino acid sequence alignments were used to calculate the numbers of amino acid substitutions between pairs of isolates. We made genetic maps using a method similar to that used for antigenic maps, except that the target distances are the numbers of amino acid substitutions between the antigen amino acid sequences and there are no antisera points (33).

To calculate the average rate of nucleotide substitutions in the HA1 domain, we measured the ML distance in the phylogenetic tree from A/eq/Zagreb/68 to each isolate. We calculated the number of amino acid substitutions from A/eq/Zagreb/68 to other isolates from HA1 domain sequence alignments. Linear models of genetic distance (response vector) and time (linear predictor for response) were fitted using the statistical package R, version 2.7.1 (29).

Analyses of antigenic evolution.

Not all substitutions are necessarily responsible for antigenic changes in HA. An amino acid substitution of X to Y at location L is considered a “cluster difference” substitution between clusters A and B if all (or all but one) isolates in cluster A have amino acid X at location L and all (or all but one) isolates in cluster B have amino acid Y at location L (33). We used this classification and the HA1 domain amino acid alignments described above to determine which amino acids likely defined the differences among clusters (see Fig. S1 in the supplemental material for alignment).

To assess antigenic drift away from a representative vaccine strain for each cluster, we measured the antigenic distance from the vaccine serum position in the map to the viruses in that cluster and fitted linear models of antigenic distance from serum and time using R version 2.7.1 (29).

Nucleotide sequence accession numbers.

The sequences in this study were deposited into GenBank under accession numbers JN084397 to JN084445 a complete list is in the supplemental material.

RESULTS

The antigenic relationships among the 107 equine influenza A (H3N8) viruses in the global panel are shown in Fig. 1A. We found that the viruses group into three distinct antigenic clusters, with each antigen colored according to the cluster to which it belongs. Viruses forming the blue cluster were isolated throughout the study period from 1968 to 2007. The first isolate in the green cluster appeared in 1989, with viruses from the cluster isolated to the end of the study period. The first viruses in the third antigenic cluster, the red cluster, were first isolated in 2003 and were also isolated to the end of the study period. Figure 1B shows current OIE vaccine recommendations and the antigenic positions of the two previous vaccine recommendations.

Fig. 1.

(A) Antigenic map of equine influenza A (H3N8) viruses isolated between 1968 and 2007. The relative positions of isolates (colored spheres) and antisera (gray cubes) were computed so that the distances between the isolates and the antisera in the map correspond, with the least error, to measurements in the HI assay (33). The orientation of the map within the axes is free because only the relative positions of the antigens can be determined. The color of each isolate represents the antigenic cluster to which it belongs. The scale bar represents 1 unit of antigenic distance, corresponding to a 2-fold dilution of antiserum in the HI assay. (B) Antigenic map of equine influenza A (H3N8) viruses isolated between 1968 and 2007. The map is rotated through 180 degrees from the orientation of panel A and shows the two current OIE vaccine recommendations, A/eq/Ohio/03 (indicated by the number 1) and A/eq/Richmond/1/07 (indicated by the number 2), and two previous recommendations, A/eq/Newmarket/2/93 (indicated by the number 3) and A/eq/Newmarket/1/93 (indicated by the number 4).

The ML phylogenetic tree (Fig. 2) shows that the genetic evolution of the same equine influenza A (H3N8) viruses used in the antigenic analyses resulted in three primary clades. The clades previously had geographical names; however, because the clades are now geographically mixed, we have identified them as clades 1, 2, and 3. Isolates from the single, or “predivergence,” lineage noted by Kawaoka et al. (19) are in clade 1, the Eurasian lineage isolates identified by Daly et al. (6) are in clade 2, and the isolates in clade 3 correspond to the American (Kentucky), Florida, and South American lineages noted by Lai et al. and Daly et al. (6, 32). We calculate that the average rate of change from the maximum-likelihood (ML) tree distance is 2.1 nucleotide substitutions per year since 1968 (Fig. 3) and that it is approximately constant over time.

Fig. 2.

Maximum-likelihood phylogenetic tree based on HA1 domain nucleotide sequences of 107 equine influenza A (H3N8) viruses isolated between 1968 and 2007. Numbers on the trunk indicate bootstrap values for key nodes. The tree is rooted by A/eq/Zagreb/68 (not shown). The isolate names in the tree are colored according to antigenic cluster. To show the temporal isolation patterns in the tree, the years in which recent isolates in the tree were isolated are indicated by horizontal bars to the right of the tree that are drawn at the same vertical position as the position of the isolate in the tree. The horizontal bars are also colored by antigenic cluster.

Fig. 3.

For each of the equine influenza virus isolates in Fig. 1, the genetic distance to A/eq/Zagreb/68 was calculated from the maximum-likelihood phylogenetic tree (Fig. 2) and was plotted as a function of time. The points are colored according to antigenic cluster. The regression line has a slope of 0.002 and an x-intercept of −4.08 (standard error [SE] = 0.01). The slope gives the rate of evolution of nucleotide substitutions per year, and the x-intercept gives the retrospective estimate of the year of introduction of equine influenza H3 viruses in horses as 1959, assuming the same rate of nucleotide substitution.

When we color the ML phylogenetic tree (Fig. 2) according to the antigenic cluster to which the isolate belongs (Fig. 1), we find that the antigenic cluster cannot always be predicted from the phylogenetic clade. For example, although the majority of clade 2 isolates form the green cluster, the blue cluster, made up of all clade 1 isolates and a subset of clade 3 isolates, is more phylogenetically diverse. The red cluster consists of the remaining subset of clade 3 isolates. Since 2001, only viruses from clade 2 and clade 3 have been isolated, and most are from clade 3. Isolates from Europe and North America are in two topologically distinct clade 3 subclades, corresponding to the blue (European) and red (North American) antigenic cluster subsets observed above.

To investigate the molecular basis of the antigenic clusters, we aligned the amino acid sequences used in this study (see Fig. S1 in the supplemental material), grouped and color coded them based on antigenic cluster, and marked the cluster difference amino acid substitutions. This alignment shows that one amino acid at position 189 distinguishes the green cluster (189K) from the blue or red cluster (189N, -D, -Q, or -E) for all antigens (Fig. 4A). The amino acid at position 159 distinguishes the blue or green cluster (159N) from the red cluster (159S). The amino acids that distinguish clusters when highlighted on the HA structure (Fig. 4B) are close to the receptor binding site.

Fig. 4.

(A) Antigenic map showing the amino acid substitutions that potentially define the antigenic properties of each cluster. (B) Two views of the three-dimensional trimer structure of the H3 hemagglutinin. One trimer is colored for clarity. The numbers indicate the amino acid positions that distinguish clusters that are close to the receptor binding site. (C) Genetic map of equine influenza A (H3N8) viruses isolated between 1968 and 2007 that was generated from the numbers of amino acid substitutions between strains in the antigenic map. The spheres are colored according to the year of isolation. The black scale bar corresponds to 5 amino acid substitutions.

To further test this observed molecular basis for the antigenic clusters, we investigated strains that might be within the same phylogenetic clade but that differ antigenically. Although antigenic clusters correspond to phylogenetic clades for the majority of isolates, for five isolates, their positions in the phylogenetic tree (Fig. 2) and in the genetic map (Fig. 4C) do not correspond to their antigenic cluster. Of these, two isolates (A/eq/Ella/89 and A/eq/Cagnes-sur-Mer/00) are phylogenetically in clade 2, which is predominantly comprised of isolates from the green antigenic cluster, but antigenically in the blue cluster. These isolates have 189E and not 189K, which is seen in all isolates in the green antigenic cluster. Conversely, A/eq/Osgodsby/92, A/eq/Arundel/12369/91, and A/Pennsylvania/1/07 are in phylogenetic clade 3, but they antigenically map in the green cluster and have lysine (K) at position 189. As these isolates are key to identifying the amino acid substitutions that define the antigenic properties of a cluster, we tested for laboratory errors by resequencing four of the five isolates (A/eq/Ella/89 and A/eq/Cagnes-sur-Mer/00, antigenically in the blue cluster, and A/eq/Arundel/12369/91 and A/eq/Pennsylvania/1/07, antigenically in the green cluster) and regenerated the HI assay data with similar results. No virus stocks were available for the fifth isolate (A/eq/Osgodsby/92). Thus, one amino acid substitution changes the antigenic properties of equine influenza A viruses sufficiently to move them from one antigenic cluster to another.

Equine vaccines are often updated on evidence of vaccine breakdown in the field. Here we showed that vaccine updates corresponded to an antigenic cluster and that often the recommendation to update vaccines was made just after the emergence of the new cluster or when it was recognized that two different antigenic variants, or two clusters, were cocirculating. However, it is possible that continuous antigenic drift within a cluster, away from the vaccine strain used to protect against that cluster, occurs over time. To investigate the potential for within-cluster antigenic evolution, we first measured the antigenic distances from a serum raised against the prototype (H3N8) virus (A/eq/Miami/63, blue cluster) to all clade 1 strains isolated up to, but not including, the year of the next vaccine update (Fig. 5). There was no evidence of continuous antigenic drift. In 1989, clade 2 arose and a vaccine update to A/eq/Suffolk/89 (green cluster) was recommended. When we measured the antigenic distance from the A/eq/Suffolk/89 serum to all strains within the clade 2 cluster, we observed that although some strains are close to the serum and some strains are further away, there is no discernible antigenic drift within the cluster over time. Following the emergence of this second cluster, the blue and the green clusters cocirculated. Because of an outbreak occurring in horses vaccinated solely with the strain representing the green cluster, an additional vaccine recommendation was made to include a representative of each antigenic cluster. When we measured the distances from the vaccine strain update for the blue cluster (A/eq/Newmarket/1/93) to all strains within the cluster isolated between 1993 and the end of the study period, we did observe antigenic drift within the cluster over time. Following the more recent emergence of the red cluster strains, we measured the distances from the vaccine update (A/eq/South Africa/4/03) to strains within this cluster and we again did not observe antigenic drift away from the vaccine recommendation. Thus, for two of three clusters, influenza A (H3N8) viruses in horses were characterized by punctuated evolution, as each cluster emerged with no discernible antigenic evolution within the cluster in the intervening period. Within one cluster which has circulated for the longest period of time, however, we observed antigenic drift away from the vaccine strain.

Fig. 5.

Antigenic distance from a vaccine serum to circulating viruses by year of isolation. The points are colored according to the vaccine used to protect against them. The linear regression lines (dotted) indicate whether there is antigenic drift away from a vaccine serum over time. Blue, slope = 0.09, P = <0.001, adjusted R² = 0.18; pink, slope = 0.05, P = 0.13, adjusted R² = 0.27; green, slope = −0.01, P = 0.80, adjusted R² = −0.05; red, slope = −0.0002, P = 0.99, adjusted R² = −0.11.

DISCUSSION

The antigenic relationships among the equine influenza A (H3N8) viruses in the global panel show that the viruses group into three clusters. The clusters have spent extensive periods of time cocirculating with no apparent global replacement of one antigenic cluster with another, and antigenic clusters are mostly, but not entirely, defined by genetic clades. Equine influenza A virus evolution is punctuated, and the antigenic clusters that developed from 1968 to 2007 have been caused by a single unique amino acid substitution.

In this study, a new antigenic cluster arose 21 years after the first cluster and a third cluster arose 14 years after the second. In contrast with the three equine antigenic clusters, over the same time period, human influenza A (H3N2) viruses evolved 11 antigenic clusters (33). We observed punctuated antigenic evolution despite a continuous rate of genetic evolution over time. The average rate of genetic change of the equine influenza H3 virus was 2.1 nucleotide substitutions per year since 1968. This rate is comparable with the rate noted in 1996 of 2.5 nucleotide substitutions per year (5) despite the emergence of additional phylogenetic lineages. However, this rate is less than half the rate for swine and human influenza H3 viruses (4.6 and 5.9 nucleotide substitutions per year, respectively) (9). The reason for this lower rate of genetic evolution at the nucleotide level is not currently clear. For two of three antigenic clusters, no within-cluster antigenic evolution was observed. However, there was within-cluster antigenic drift for one antigenic cluster. The OIE expert surveillance panel for equine influenza annually reviews the strains used in the equine influenza virus vaccine by considering antigenic, genetic, and epidemiological data on currently circulating strains. This change in the antigenic characteristics of the clade 2 viruses from those of the older clade 2 viruses within a cluster, together with evidence of a reduction in vaccine efficacy, coincided with a recommendation from the OIE to update the vaccine strain (6).

Since 1989, two antigenic clusters have circulated at the same time in the equine population. There is well-documented evidence that initially, one antigenic cluster arose and circulated in North American horses and one arose and circulated in European horses (6). The introduction of the American lineage of viruses into Europe, once in 1993 and again in 2003, led to cocirculation of two antigenic clusters in Europe. Incursion of influenza into equine populations can occur as a result of quarantine failure when horses travel internationally for competition or sales (4, 6; A. J. Guthrie, presented at the 9th World Equine Veterinary Association Congress, Marrakech, Morocco, 22 to 26 January 2006). However, there have only been two reported incidences of infection in North America with a virus originating in Europe: with A/eq/Saskatoon/90, which was isolated in Canada (21), and with A/eq/Kentucky/4/2004, which was isolated from a horse under U.S. quarantine (T. M. Chambers, personal observation). That European viruses have not become established in the North American horse population is likely due to differences in quarantine practices (11, 37).

However, only a small proportion of the global equine population consists of competition and breeding horses that travel internationally and that have potential to transmit viruses intercontinentally, either through quarantine or vaccine failure. The majority of the equine population, whether general recreational riding horses or animals used for draft work, rarely, if ever, mix with such competition horses. Given the spatial heterogeneity of horse populations worldwide, one would, perhaps, expect substantially more virus diversity among the equine influenza H3 viruses than the two cocirculating lineages observed in this study. As a comparison, there are two cocirculating lineages of influenza B viruses in humans, with substantially greater connectivity among human populations worldwide than among equine populations. Hypotheses for this surprising homogeneity of equine influenza H3 viruses include that the level of horse movement or mixing needed to maintain homogeneity is lower than previously thought or that there is greater horse movement than we realize. A further hypothesis is that equine influenza outbreaks might be relatively short-lived and spatially isolated, and therefore, the epidemiology of the equine influenza virus might offer it little opportunity to evolve, so we observe little diversity. Given the potential for spatial heterogeneity to lead to heterogeneities in the evolution of the virus, it is important to expand surveillance efforts in horse populations not currently being tested to monitor the emergence of new antigenic variants.

Vaccination is the primary method of controlling equine influenza. Vaccine updates are recommended after an assessment of antigenic, genetic, and field data. Wilson and Cox observed that vaccine updates occurred for human influenza viruses when there were 4 or 5 amino acid substitutions in at least 2 antigenic sites (41). Antigenic sites for equine influenza H3 virus have not been demonstrated but are likely similar to those of human influenza H3 viruses. However, here we show that just one amino acid substitution results in antigenic distances between the immunizing virus and the challenge virus that would result in a loss of vaccine protection.

Equine vaccine cross-protection studies have previously showed that vaccination with a strain from one cluster does not prevent infection with a strain from the other cocirculating cluster (8). We can quantitatively relate the ferret serum-derived antigenic data back to the immunological response to vaccination in the horse and to the protection against different clusters that a particular vaccine strain might afford because equine influenza virus vaccine updates are accompanied by clinical trials. Measuring the distance between strains used in vaccination and challenge studies in each cluster that do not provide cross-protection shows that, on average, antigenic distances greater than 3 antigenic units when characterized using ferret antisera would likely result in vaccine breakdown. Relatedly, the antigenic distance from one vaccine update recommendation to the next is, on average, 4.7 antigenic units when antigenic differences between isolates are characterized using ferret antisera, thus supporting the need for these equine influenza virus vaccine updates.

Given that we found that a single amino acid substitution is sufficient to result in the emergence of a new antigenic cluster and also to potentially result in vaccine failure, it is thus imperative to antigenically and genetically characterize new equine influenza virus variants to be able to make the best recommendations for a vaccine update. Coupling antigenic and genetic analyses with cross-protection data in the natural host links HI cross-reactivity, derived using a small-animal model, such as a ferret, with that which is significant for immunological protection in the horse. Here we describe and quantify the evolution of influenza A (H3N8) virus in horses and compare its evolution with that of influenza viruses circulating in other hosts. Such comparative analyses can reveal surprising similarities and differences, such as the similarity of influenza A (H3N8) virus evolution in horses to influenza B virus evolution in humans despite the large differences between the mixing patterns among horses and the mixing patterns among humans. Such information is potentially important to increase our understanding of what governs the underlying evolutionary mechanisms of how influenza virus evolves and thus may improve control measures for influenza A viruses.