Influenza A viruses are negative-strand RNA viruses of the Family Orthomyxoviridae that infect a wide variety of warm-blooded animals, including domestic and wild birds and mammals (e.g., humans, pigs, and horses). The natural reservoir for influenza virus is thought to be wild waterfowl, and genetic material from avian strains episodically emerges in strains infectious to humans. These human viruses continually circulate in yearly epidemics (mainly during the winter months in temperate climates), and antigenically novel strains emerge sporadically as pandemic viruses [1,2]. In the United States, influenza is estimated to kill 30,000 people in an average year [3,4]. Every few years, influenza epidemics boost the annual mortality level above this average, causing 10,000–15,000 additional deaths. Occasionally, and unpredictably, global pandemics of influenza occur, infecting 20% to 40% of the population in a single year and raising death rates dramatically above normal levels. Pandemic influenza A viruses emerged three times during the last century: in 1918 (H1N1 subtype), in 1957 (H2N2), and in 1968 (H3N2) [2,5]. The recent circulation of highly pathogenic avian H5N1 viruses in Asia from 2003–2005 has caused at least 52 human deaths [6,7] and has raised concern about the development of a new pandemic [5]. How and when novel influenza viruses emerge as pandemic strains and their precise mechanisms of pathogenesis are still not understood.

While the risk of pandemic influenza poses a significant public health concern, inter-pandemic or epidemic influenza remains a major cause of morbidity and mortality. The influenza A surface glycoprotein hemagglutinin (HA) protein is under selective pressure for change in order to evade the host's immune system [8]. Antibodies against the HA protein inhibit receptor binding and are very effective at preventing reinfection with the same strain. However, HA can change to evade previously acquired immunity either by antigenic drift, whereby mutations of the currently circulating HA gene disrupt antibody binding, or by antigenic shift, in which the virus acquires an HA of a new subtype by reassortment of one or more gene segments. While it is generally accepted that drift is responsible for inter-pandemic influenza outbreaks and shift for pandemics, there are exceptions to this rule. For example, in 1977, an H1N1 virus re-emerged but failed either to cause a pandemic or to replace the prevailing H3N2 subtype [9].

The importance of predicting the emergence of new circulating influenza strains for subsequent annual vaccine development cannot be underestimated [10]. To this end, the global influenza surveillance network coordinated by the World Health Organization was established to select the candidate strains of influenza A and B for the yearly production of influenza vaccine in both the northern and southern hemispheres. The network characterizes the antigenic properties of influenza viruses using HA inhibition assays and sequencing of the HA1 domain (globular head) of HA of a select number of strains [11,12]. Antigenic, genetic, and epidemiologic data are then examined to make recommendations of candidate vaccine strains.

A number of retrospective studies have been performed using partial HA gene sequences to understand, and sometimes predict, the evolution of human H3N2 strains [13–17]. Accumulation of amino acid replacements in HA is clustered in five variable antigenic sites [18] around the receptor binding site. Phylogenetic analysis has revealed that 18 codons in the HA1 domain exhibit significantly more nonsynonymous nucleotide substitutions than synonymous ones, constituting a signature of strong, selectively driven, antigenic drift [15]. More recently the antigenic and genetic evolution of HA was compared through the construction of an antigenic map of H3N2 viruses [17]. Although antigenic evolution was found to be more punctuated than genetic evolution over the same time period, the two measures of HA drift were generally correlated.

Despite the wealth of data on the molecular evolution of influenza viruses, how the entire genome of influenza A virus evolves during epidemic years is unclear, particularly as past sample sizes have been inadequate. While antigenic drift of HA is clearly of vital importance in the survival of an influenza strain, other factors, including HA receptor binding specificity [19], antigenic drift of neuraminidase (NA) [20], matched activity between HA and NA [21–23], and the interaction of the other influenza proteins with each other and their host cells, are all likely to affect viral fitness in a polygenic manner. Similarly, it is unclear how many lineages of influenza A viruses persist between seasonal epidemics, particularly in genes other than that encoding HA.

To this end the National Institute of Allergy and Infectious Diseases of the National Institutes of Health has funded the Influenza Genome Sequencing Project with several partners [24,25] including the Institute for Genomic Research. To date, 156 genomes of human H3N2 viruses collected between 1999 and 2004 from New York State have been completely sequenced and deposited in GenBank. We have performed an initial analysis of these viruses and have found evidence for both the existence of multiple clades of viruses co-circulating at the same time point and for multiple reassortment events among these clades. One of these reassortment events was the likely progenitor of the A/Fujian/411/2002-like drift epidemic of the 2003–2004 influenza season, in which there was a poor match between the vaccine strain and the predominant circulating viruses of that year [26,27]. This report extends recent observations of reassortant H3N2 influenza A viruses from the southern hemisphere during this same time period [28].

Our analysis of whole genomes of H3N2 influenza A viruses sampled during 1999–2004 has identified two key evolutionary patterns. First, although the majority of viruses isolated after 2002 fall into a single phylogenetic group (clade A), multiple, co-circulating viral lineages are present at particular time points. The genetic diversity of influenza A virus is therefore not as restricted as previously suggested, particularly when genes other than that encoding HA are analyzed. This co-circulation of lineages is most apparent with the identification of three clades of H3N2 viruses that appear to infect the same populations until 2002, after which they acquired a common HA gene through reassortment. Second, and more dramatically, these multiple, co-circulating lineages may have complex genealogical histories and interact through reassortment. Indeed, we have documented two reassortment events involving the HA gene of clade B: one in which it was acquired by the clade A viruses and another in which it was independently acquired by those isolates assigned to clade C. Two further reassortment events involving the PB2 and PA genes were also evident from our phylogenetic analysis. Given that we are only able to reliably detect reassortment when it is associated with major changes in tree topology, it is likely that reassortment among closely related lineages is also commonplace in influenza A viruses.

Reassortment between influenza A viruses has been described in both human and animal viruses [1,29]. Notably, antigenic shift by reassortment between human and avian influenza A viruses has been documented in the formation of the 1957 H2N2 and 1968 H3N2 pandemics [30–32]. Other recent examples of reassortment between human and animal influenza A viruses have resulted in the emergence of novel H3N2 and H1N2 swine viruses in North America and Europe [33,34] and the evolution of H5N1 viruses in Asia from 1997 to the present [35]. Reassortment between co-circulating lineages of human influenza A and more recently influenza B viruses following mixed infection has also been described [36–41]. For example, human H2N2 viruses formed two distinct clades in the 1960s prior to the emergence of the 1968 H3N2 pandemic virus, with one virus a reassortant containing genes of both clades [42]. Similarly, the early H3N2 viruses (1968–1972) acquired the H3 HA and the PB1 gene via reassortment with an avian virus [30,31]. Reassortment between H3N2 and H2N2 viruses may therefore have assisted successful cross-species transmission [42].

Reassortant viruses were also described following the re-emergence of the H1N1 subtype in 1977 that did not replace the previously circulating H3N2 viruses. In this case, co-circulation of influenza viruses of both subtypes continued, and co-infection with both subtypes was reported [43]. While reassortant H3N1 strains were not isolated, H1N1 strains containing reassorted internal protein-encoding gene segments from H3N2 viruses were observed [44,45]. Occasional isolates of H1N2 viruses were also detected after the re-emergence of H1N1 [46,47]. More recently, the widespread circulation of viruses with the H1N2 subtype has been documented [41]. These viruses contained the HA segment of contemporary H1N1 viruses reassorted onto an H3N2 background, a 7:1 reassortment pattern similar to that observed with the sporadically circulating H1N2 viruses of the 1980s and early 1990s [47] and to the dominant reassortments described in this analysis. Since the H1 and N2 subtype proteins were antigenically and genetically similar to co-circulating H1N1 and H3N2 subtype viruses, the emergence of this new subtype did not result in an epidemiologically significant event [41]. Reassortment among co-circulating clades of H3N2 viruses like that observed in the current study has also been previously described, including reassortment of the NA gene segment [48] and the core protein-encoding segments [49].

Most prior phylogenetic studies of human influenza A have suggested that inter-pandemic evolution may be essentially described as a series of successions by variants of the previous season's dominant strain. These successions are largely determined by strong positive selection acting on the abundance of mutational diversity in the HA of the dominant strain. However, we found that at least four reassortment events occurred among human viruses during the period 1999–2004 and that two of these involved a major change in HA. Recently, Barr et al. independently provided phylogenetic evidence of the clade A–clade B reassortment described here in an analysis of predominantly southern hemispheric influenza A H3N2 isolates collected during the same period [28]. To our knowledge, these analyses are the first demonstrations of the emergence of a major antigenically variant virus derived by reassortment between two distinct clades of co-circulating H3N2 viruses rather than by antigenic drift. These findings suggest that the ongoing evolution of human influenza A virus is likely to be more complex than depicted in standard models of antigenic drift; multiple lineages of antigenically distinct viruses can persist within populations and, through their reassortment, produce major changes in antigen space. Similarly, the persistence of multiple lineages of H3N2 within a single population indicates that human populations represent a larger reservoir of genetically distinct viruses than previously anticipated. Indeed, it is possible that key changes in antigen type, depicted as jumps between cluster types [17], could be strongly influenced by reassortment among co-circulating human strains. Crucially, the real importance of both lineage persistence and reassortment in influenza A virus evolution could not be determined until a representative sample of full-genome sequences was collected from a single population.

In the 2003–2004 influenza season, a major drift variant emerged in both the northern and southern hemispheres [50–52]—the A/Fujian/411/2002-like variant. In the United States, the 2001–2002 influenza was an H3N2-predominant season, and all antigenically characterized isolates matched the A/Moscow/10/1999 vaccine strain [53]. In the 2002–2003 season, which was an H1- and influenza B–dominant season in the United States, a minority of antigenically characterized H3N2 isolates were different from the Moscow/10/1999-like vaccine strain [54], probably coinciding with the emergence of the Fujian strain. Indeed, in our phylogenetic analysis two of five H3N2 viruses from the 2002–2003 season fall into clade B, along with two newly sequenced isolates that were not included in our analysis (A/New York/201/2003 and A/New York/203/2003). Consequently, clade B viruses made up a significant fraction of the few H3N2 isolates in that season. In contrast, only a small minority of influenza H3N2 viruses sampled in Europe at this time were antigenically characterized as Fujian-like [50], while in the southern hemisphere's 2003 influenza season the Fujian strain was predominant [28,51]. The 2003–2004 influenza season in the northern hemisphere was also dominated by the Fujian strain [55], although the vaccine contained the H3N2 (A/Panama/2007/1999) from the previous year. This strain was a poor antigenic match to the Fujian strain [26,27], which in turn led to reduced vaccine effectiveness. Thirteen amino acid changes distributed across the five antigenic sites of the HA1 domain distinguish the A/Panama/2007/1999-like and A/Fujian/411/2002-like strains. While these replacement changes appear to have phenotypic consequences (e.g., replication efficiency in eggs), only two residues, 155 and 156, are responsible for the major antigenic differences between the strains [27].

Although data are insufficient for precise determination of the timing of these two critical mutations, the available data are most consistent with these changes occurring in a relatively short time period before the reassortment event. The histidine to threonine change at site 155 and the glutamine to histidine change at site 156 are present in all the clade A reassortant isolates as well as in clade B isolates from 2003–2004, thus suggesting that they occurred prior to the reassortment event. No available clade B isolates prior to 2002–2003 have either of these mutations, and we were able to identify only three “intermediate” isolates from 2002–2003 (A/Kwangju/219/2002, A/Kwangju/243/2002, and A/Cheonnam/340/2002) with the replacement at site 155 but not at site 156. Overall, the data presented here, coupled with those recently reported [28], reveal that the HA segment of the H3N2 clade B viruses, present in low frequencies at least since 1999, was reassorted into clade A of H3N2, probably in 2002 soon after the appearance of these two critical mutations, and that this reassortment was central to the production of antigenically variant strains that were poorly matched to the vaccine strain in the 2003–2004 season [27]. In addition, presumably because of the high rate of reassortment and the fitness advantage conferred by these two mutations, this clade B HA segment also appeared to replace the HA segment of the clade C strains. Finally, though previously present only in low frequencies, recent sequencing of HA and NA by investigators in Denmark [56] and published phylogenetic trees from Australia [28,51] show the existence of clade B virus, suggesting that it continued to have a global distribution and sometimes at appreciable frequency.

Several questions remain unanswered by our study. Since the HA donated by clade B led to a major expansion of the reassorted clade A, it is uncertain why clade B did not initially out-compete clade A without reassortment. One possibility is that the HA of clade B had an intrinsically higher fitness than other HAs circulating at the same time but was unable to reach a high frequency in the New York population owing to linkage to mutations located in other segments that reduced the overall fitness of this genotype. According to this hypothesis, it was not until it was placed by reassortment into a more favorable genetic background, in this case the clade A viruses, that its fitness advantage was realized. Since clade B itself appeared to proliferate in other regions, it will be useful to analyze whole-genome sequence from these isolates when they are available.

More generally, it is clear that the genotypic basis to viral fitness has not been entirely elucidated. In particular, it is likely that interactions among viral proteins and between viral proteins and host factors play a key role. In this respect it is notable that of the 48 amino acid differences that distinguish the clade B viruses, nine fall in NP and 14 fall in NA (see Table 1). In NP, the replacement at residue 425 falls within an HLA-B35-restricted cytotoxic T lymphocyte (CTL) epitope and is not commonly seen in human populations [57]. Two other changes, at residues 27 and 197, have also been identified as being contained in two HLA-A11-restricted CTL epitopes [58]. Another unusual change in NP is M136I: a change to methionine at this site was proposed as one of six human adaptive changes distinguishing the 1918 NP from avian NPs [59]. Indeed, all 45 available NP sequences from human H2N2 viruses, and all but four of approximately 250 human H3N2 NP sequences, maintain this methionine. Only the three New York clade B viruses, as well as A/Taiwan/1/71, have a change to an isoleucine at this residue. Of the changes in NA, six of them lie in five regions previously identified as being phylogenetically important regions, residues 40, 42, 143, 199, 307, and 385 [60], and therefore play a role in virus–host interaction. Three further changes, at residues 172, 199, and 307, map to antigenic sites [20], while a change at residue 18 has been mapped as an HLA-A2-restricted CTL epitope [58]. Similarly, two changes in M2, at residues 51 and 56, map to an HLA-A11-restricted CTL epitope, while residue 82 of NS1 maps to an HLA-A2-restricted CTL epitope [58]. Another change between clade A and B viruses, at residue 226 of PA, also maps to an HLA-A2-restricted CTL epitope [58]. It is possible that some of the mutations that fall in CTL epitopes assist the persistence of clade B by elongating the viral infectious period [61]. However, any combination of the constellation of amino acid changes may have altered the fitness of the clade B viruses in a way that we do not have the ability to understand from sequence analysis alone. Interestingly, Gulati et al. [62] have recently shown that Fujian-like strains have a mismatch in their HA and NA activities that is probably the result of the reassortment event described in the current study. The significance of this for the pathophysiology of the virus is currently unknown.

In summary, our study clearly demonstrates the utility of whole-genome analyses of influenza A viruses, and further makes clear that additional whole-genome analyses are required to understand fully the evolutionary mechanisms and epidemiological dynamics of this virus. While antigenic variance of HA is still the dominant selective pressure on human influenza A virus evolution, the finding that antigenically novel clades emerge by reassortment among persistent viral lineages rather than via antigenic drift is of major significance for vaccine strain selection.