Research Article

The Global Circulation of Seasonal Influenza A (H3N2) Viruses

See allHide authors and affiliations

Science  18 Apr 2008:
Vol. 320, Issue 5874, pp. 340-346
DOI: 10.1126/science.1154137


Antigenic and genetic analysis of the hemagglutinin of ∼13,000 human influenza A (H3N2) viruses from six continents during 2002–2007 revealed that there was continuous circulation in east and Southeast Asia (E-SE Asia) via a region-wide network of temporally overlapping epidemics and that epidemics in the temperate regions were seeded from this network each year. Seed strains generally first reached Oceania, North America, and Europe, and later South America. This evidence suggests that once A (H3N2) viruses leave E-SE Asia, they are unlikely to contribute to long-term viral evolution. If the trends observed during this period are an accurate representation of overall patterns of spread, then the antigenic characteristics of A (H3N2) viruses outside E-SE Asia may be forecast each year based on surveillance within E-SE Asia, with consequent improvements to vaccine strain selection.

Influenza A (H3N2) virus is currently the major cause of human influenza morbidity and mortality worldwide. On average, influenza viruses infect 5 to 15% of the global population, resulting in ∼500,000 deaths annually (1). Despite substantial progress in many areas of influenza research, questions such as when and to what extent the virus will change antigenically, and to what extent viruses spread globally, remain unanswered. A fundamental issue behind these questions is whether epidemics are the consequence of low-level persistence of viruses from the previous epidemic or whether they are seeded from epidemics in other regions and, if so, from where (28).

Addressing these issues of local persistence and global spread is vitally important for designing optimal surveillance and control strategies. If epidemics were regularly seeded from an outside region and if the source region of seed strains could be identified, it may be possible to forecast which variants would appear in epidemics in seeded regions and, consequently, to optimize vaccine strain selection. Alternatively, if viruses persist within a region, evolve, and reemerge to cause the next epidemic, intervention strategies targeting virus circulation between epidemics might be effective in minimizing subsequent epidemics.

The eight gene segments of influenza Aviruses reassort, leading to complicated phylogenetic patterns at the genomic scale (4, 810). The gene segment coding for the hemagglutinin (HA) is of major importance because the HA protein is the primary target of the protective immune response. Consequently, the HA is the focus of public health surveillance and the primary component of currently licensed influenza vaccines. We used antigenic and genetic analyses of the HA as a marker to investigate the global evolution and epidemiology of influenza A (H3N2) viruses from 2002 to 2007 and to determine whether influenza epidemics arise from locally persisting strains or whether epidemics are seeded from other regions.

Antigenic Evolution

Antigenic cartography has shown that the antigenic evolution of A (H3N2) virus, since its appearance in humans in 1968, can be represented by a two-dimensional (2D) antigenic map (11). Since 2002, the antigenic evolution of A (H3N2) viruses has roughly followed a line away from the A/Sydney/5/1997-like viruses that predominated in 1998 through the A/Fujian/441/2002-like viruses to the A/California/7/2004-like viruses to the A/Wisconsin/67/2005-like strains that predominated in 2006 and 2007. Figure 1A uses this directional bias to make a 2D plot to show spatial antigenic evolution and epidemiology in one plot.

Fig. 1.

Global patterns in antigenic and genetic evolution over time. (A) Antigenic distance from A/Sydney/5/1997 virus of all isolates plotted against time of isolation. One unit of antigenic distance corresponds to a twofold dilution of antiserum in the HI assay. Each point corresponds to a laboratory-confirmed influenza A (H3N2) isolate, with the color of the point indicating the region of geographic origin as shown in (E). The thick black line is the best-fit statistical model (a loess spline) fitted through all data points to show the trend over time. Points in advance of the spline are antigenically advanced, whereas strains behind the spline are antigenically lagging. (B) Genetic version of (A) for all sequenced strains with distance measured to the root, A/Wuhan/359/1995, of a maximum likelihood nucleotide phylogenetic tree (C). The thick black line is the same as in (A) but for the genetic data. (C) Phylogenetic tree of HA1 nucleotide sequences color-coded by geographic origin (E), including strain names and isolation dates. We constructed the initial tree with the PhyML software package version 2.4.5 (38), with 1567 nucleotide sequences (12) and A/Wuhan/359/1995 as the root, using 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 rate categories estimated from the empirical data) [determined by ModelTest (39)] as the evolutionary model. GARLI version 0.951 (40) was run on the best tree from PhyML for two million generations to optimize tree topology and branch lengths. Figure S3A provides a “zoom-able” version of this image. (D) Partial detail of (C). (E) Color-coded geographic setting for (A) to (D).

Previously, Smith et al. (11) showed that, from 1968 to 2003, the antigenic evolution of A (H3N2) virus was punctuated: Periods of relative stasis lasting from 3 to 8 years were followed by rapid antigenic change, resulting in transitions to new antigenic clusters that necessitated an update of the influenza virus vaccine strain. During this time, several clusters also exhibited intracluster antigenic evolution, necessitating a within-cluster update of the vaccine strain. We find a similar pattern of intracluster antigenic evolution from 2002 to 2007. During this time, antigenic evolution progressed away from A/Sydney/5/1997 at an average rate of 2.13 antigenic units per year (Fig. 1A). A distance of two antigenic units, representing a fourfold difference in hemagglutination inhibition (HI) assay titer, is generally considered as a sufficient antigenic difference to warrant a vaccine update. The A (H3N2) component of the influenza vaccine was updated four times during this period. A core component of vaccine strain selection involves identifying emerging antigenic variants. If an emerging variant is judged likely to cause epidemics in the upcoming influenza season, the vaccine is updated to contain a representative of the new strain.

We find differences in the amount of antigenic variation seen during an epidemic in different regions and from year to year in the same region (Fig. 1A). Despite such spatial and temporal heterogeneities, the antigenic evolution has been markedly homogeneous on a global scale. An explanation for this homogeneity could be that viruses circulate globally rather than persist and evolve locally.

To search for global patterns in the source of emerging variants, we measured which regions were leading or trailing antigenically and found that, from 2002 to 2007, newly emerged strains of the A (H3N2) subtype appeared in E-SE Asian countries, on average, ∼6 to 9 months earlier than they appeared in other regions, with long delays to South America, typically of an additional ∼6 to 9 months (Fig. 2A). Though A (H3N2) viruses in E-SE Asian countries are on average more antigenically advanced, there is sufficient variability from season to season (colored circles in Fig. 2A) such that no one particular country in the region is consistently most advanced. Thailand, Malaysia, and Japan are exceptions to the Asia-leading pattern, being less antigenically advanced than the rest of the region.

Fig. 2.

Evolutionarily leading and trailing regions. (A) Black circles indicate the average antigenic distance to the spline of Fig. 1A for all strains isolated in a region, and the thin horizontal black line indicates the SEM. Colored circles split this overall average by epidemic; circle color indicates time. The spline can also be interpreted as a function of time; thus, time is shown as a second x axis. (B) Similarto (A) but based on genetic distance to spline from Fig. 1B. (C) Genetic distance to trunk of the phylogenetic tree by region and season. We algorithmically defined the trunk of the tree in Fig. 1C (14) and calculated the tree distance of each strain to the trunk. Average distance to trunk was calculated per region and per season. The black circles indicate the overall average per region, the thin horizontal black line indicates the SEM, and colored circles indicate seasonal averages. The mean for E-SE Asia is different from that of Oceania (P < 0.00001), North America (P < 0.001), Europe (P < 0.01), and South America (P < 0.0001).

One interpretation of this Asia-leading pattern is that new variants emerge first in E-SE Asia and subsequently seed other regions of the world. An alternative but more complex explanation is that this pattern is the product of independent local persistence in multiple regions and parallel evolution in which similar antigenic variants emerge independently worldwide as a result of similar selection pressures. To test between these two interpretations, we must answer the fundamental long-standing question of whether influenza viruses persist in a region, and could thus undergo parallel evolution, or whether regions are regularly seeded from external regions.

About 10% of the ∼13,000 A (H3N2) viruses analyzed antigenically were also analyzed genetically by sequencing the HA1 domain of the hemagglutinin (12). This subset was a representative sample of all ∼13,000 isolates (fig. S1B) and thus was suitable for investigating the ancestry of strains and the fundamental issue of local persistence versus seeding in the global circulation of A (H3N2) viruses. The genetic progression over time (Fig. 1B) shows similar average patterns to the antigenic data, with Asia leading [as previously shown for Taiwan (13)] and South America trailing (Fig. 2, A and B). However, there were important differences (China in 2005 and Oceania in 2005). For influenza vaccine strain selection, genetic-antigenic differences are resolved in favor of the antigenic data, because the humoral immune system “sees” the virus phenotype, not the genotype.

Persistence Versus Seeding

Source of inter-epidemic strains. The simplest test of persistence versus seeding is to examine the origin of strains isolated between epidemics. If viruses persist locally, at least some of the inter-epidemic strains would be descended from, and thus more closely related to, strains from the previous epidemic than to strains from outside the region. Alternatively, if there was no persistence, inter-epidemic strains would be more similar to strains from elsewhere.

We sequenced the HA1 domains of 52 inter-epidemic strains isolated in Oceania (primarily Australia and New Zealand), North America, and Japan from June 2002 to September 2006. None of these inter-epidemic strains was more similar to strains from the previous local epidemic than to externally circulating strains (Fig. 3). This result is evidence for external seeding and against local persistence.

Fig. 3.

The genetic relationship of interseasonal strains to strains in the previous local epidemic and to strains epidemic in other regions. Interseasonal strains are defined as strains isolated more than 1 month after the end of the previous local epidemic and more than 1 month before the beginning of the next local epidemic. For each interseasonal strain, the phylogenetic tree distance was calculated to the closest strain in the previous epidemic and to the closest strain found outside the region in the previous 4 months. The diagonal line is 1:1 and is included for reference.

Even when done well, inter-epidemic surveillance yields relatively small amounts of data that can never completely rule out the existence of local virus persistence between epidemics, especially of any low-pathogenic variants that produce subclinical infections. The test described in the next section has the advantage of using all available sequence data rather than being limited to inter-epidemic data. Nevertheless, all tests must take into account the effects of external introductions during local epidemics (14).

Evolutionary relationship of strains from one epidemic to the next in a region. As described by Nelson et al. (4, 8), if epidemic strains persist locally and give rise to the next local epidemic, those strains should be more closely related to one another than to strains isolated in other regions, and a phylogenetic tree of the data would look like that depicted in fig. S2A. Conversely, if epidemics were seeded from outside a region, the epidemic strains would be more similar to contemporary strains from outside that region than to strains from the previous local epidemic (fig. S2B).

Following the methodology of Nelson et al. (4, 8), we constructed a phylogenetic tree of the HA1 domain of the hemagglutinin from the sequenced subset of the global surveillance data (Fig. 1C and fig. S3A). In this tree, the HA1 of the viruses in each epidemic in a temperate region (four in North America, five in Oceania, four in Europe, and four in Japan) and each epidemic in a subtropical region (three in Hong Kong) descended from externally circulating strains, not from strains in the previous local epidemic (Fig. 1, C and D). The topology of this tree is more similar to that in fig. S2B than to that in fig. S2A. This result is also evidence for external seeding and against persistence. For other regions, including most tropical and subtropical regions, there were fewer sequences, and it was not possible to conclusively differentiate between persistence and seeding.

This evidence for external seeding and against persistence agrees with full-genome analyses of New York state, Australia, and New Zealand data that show global migration of A (H3N2) viruses rather than local persistence (4, 810, 15). In addition, Nelson et al. (8) find evidence compatible with either northern-to-southern hemisphere migration or migration from tropical regions, including Southeast Asia.

Source of Seeding

Given this evidence for seeding and against local persistence, we are left with the interpretation that the E-SE Asia–leading pattern (Fig. 2, A and B) implies that new variants emerge first in E-SE Asia and then seed the rest of the world. The phylogenetic tree provides further evidence to support this interpretation, showing that the ancestors of strains in temperate regions typically originate in E-SE Asia, with the “trunk” of the phylogenetic tree typically occupied by E-SE Asian strains (Fig. 1, C and D) and with, on average, E-SE Asian strains closer to the trunk (P < 0.001) than strains from all other regions (Fig. 2C).

The above analyses, in addition to being evidence for an “out of E-SE Asia” hypothesis, are also evidence against several other long-standing unresolved hypotheses for the global circulation of influenza viruses, as follows.

Out-of-China. If China alone were the source of all new variants and effectively seeded the rest of the world (16), then Chinese strains would be (i) closer to the trunk than strains from all other regions each year and (ii) consistently antigenically and genetically advanced relative to strains from other regions, both of which we do not find (Fig. 2). Commensurate with previously published studies, we find new variants are sometimes first detected in China (15, 17, 18); however, we also find that new variants are sometimes first detected in other countries in E-SE Asia (Fig. 2).

Out-of-tropics. In this hypothesis, epidemics in regions outside the tropics are seeded from the tropics (7, 19). If true, A (H3N2) viruses would need to circulate continually in the tropics and would give rise to one of two evolutionary patterns. One pattern would arise if tropical Asia, Africa, and South America were well connected epidemiologically to one another; in this case, all three regions would be similarly antigenically and genetically advanced and closer to the trunk than nontropical regions. The other pattern would arise if the three tropical regions were poorly connected epidemiologically to one another; in this case, there would be independent genetic lineages for each tropical region. There is currently not enough data to include tropical Africa in this analysis. However, for tropical Asia and South America, neither of these patterns is observed (Figs. 1C and 2 and fig. S3A) (14).

Seeding by strains moving between the northern and southern hemispheres. If this hypothesis were true, then each year epidemics in the northern hemisphere would be seeded by viruses from epidemics in the southern hemisphere and vice versa (2022). The phylogenetic tree for the HA1 domain (Fig. 1C and fig. S3A) shows no evidence for epidemic strains in the northern hemisphere being descendant from strains epidemic in South America or Africa. The tree shows limited evidence for Oceania playing a role in seeding a minority of epidemics in the northern hemisphere (14) but at a level insufficient to support this hypothesis as the dominant mechanism of the global circulation of influenza viruses. This hypothesis also fails to explain how viruses from Asia, which is almost entirely contained within the northern hemisphere, lead antigenically and genetically and are closest to the trunk of the phylogenetic tree.

Local persistence with seeding only at cluster transitions. Though there was no major cluster transition during our study period, there has been an average of 2.13 units of antigenic evolution each year. Even so, in temperate regions, no local persistence could be detected, and each epidemic was seeded by exogenous strains. The drift variants observed in this study have emerged from E-SE Asia; however, because we have not seen a major cluster transition of the magnitude of Wuhan 1995 to Sydney 1997 (∼4.7 antigenic units), we can neither exclude that in such a case a new variant could emerge outside E-SE Asia nor that it could affect seeding patterns.

The E-SE Asian Circulation Network

For E-SE Asia to seed epidemics in multiple regions of the world, influenza virus must circulate continually in E-SE Asia. But how?

It is generally considered that influenza viruses continually circulate in tropical countries (7, 2327) and, if this were true, it would explain how influenza viruses could persist in tropical Asia. Indeed, circulation in an endemic core area that seeds satellite areas has been shown to be a key epidemiological process for the continual circulation of antigenically stable pathogens (28, 29). Reports based on influenza-like illness (ILI) or influenza and pneumonia mortality (IPM) data describe continual circulation in the tropics (23). However, several viruses other than influenza can cause ILI and IPM, and studies from tropical countries based on viral isolations show a marked seasonality for influenza epidemics, with peaks usually occurring during periods of high rainfall (3035). In agreement with these studies based on virus isolation, our virus isolation study also finds that influenza has clear epidemic peaks and deep troughs in all regions, including the four tropical and four subtropical E-SE Asian countries for which there are sufficient data to detect an epidemic signal. Thus, continuous circulation in individual tropical countries is unlikely to be the mechanism for persistence in E-SE Asia. However, more data from a wide diversity of locations are needed to fully understand seasonal forcing and to definitively exclude local persistence as an element of transmission dynamics in tropical and subtropical areas of E-SE Asia.

Another possibility for continual circulation is that viruses pass from epidemic to epidemic among countries via the mobile human population. Figure 4, A and C, shows that there is sufficient variability in the timing of epidemics within E-SE Asia such that the virus could circulate continuously in this way as a result of the temporal overlap of epidemics. Much of the variability in the timing of epidemics is likely to be linked to the heterogeneity in the timing of lower temperatures and rainy seasons (19, 33, 36). We thus hypothesize that the variability of epidemics, combined with the interconnectedness of E-SE Asian countries, forms an E-SE Asian circulation network that maintains influenza virus in the region by passing from epidemic to epidemic.

Fig. 4.

Synchrony of epidemics in east Asia and the South Pacific. (A) Epidemics in east Asia. The y axis shows laboratory-confirmed H3N2 infections per 2 weeks as a proportion of the total number of laboratory-confirmed H3N2 infections over the study period in each location. (B) Strains on the trunk of the phylogenetic tree are of particular evolutionary importance in testing for virus migration among countries. In (B) and (E), there is a circle for every strain on the trunk of the phylogenetic tree (figs. S3A and S4). The purpose is to show where the trunk strains were isolated [top row color code from (A), bottom row from (C)] and when they were isolated, to assess the epidemiological activity at the time of isolation. Cyan circles represent E-SE Asian strains but in locations not shown in (F). (C) Same as (A) but for tropical Southeast Asia. (D) Same as (A) but for Australia and New Zealand. (E) Same as (B), but the top row are Oceanian strains [cyan circles represent strains from cities in Australia not shown in (F)], and the bottom row are strains from North America (blue) and Russia and Ukraine (yellow). (F) Geographic setting for (A) to (E).

If such a network existed, we would expect a temporal and phylogenetic progression of E-SE Asian viruses on the trunk of phylogenetic tree as viruses pass from epidemic to epidemic within the network. Figure S4 shows such a progression, and Fig. 4, B and E, shows the relationship of these trunk strains to the timing of epidemics. Trunk strains were isolated in temperate, subtropical, and tropical regions of E-SE Asia, indicating that all three climatic regions of E-SE Asia are part of the circulation network. To test whether the non–E-SE Asian strains on the trunk indicate that the circulation network includes countries outside of E-SE Asia or whether they represent one-way seeding events out of E-SE Asia, we examined the phylogenetic tree (Fig. 1C and fig. S3A). We found only limited instances of such seeding back into E-SE Asia, with clear evidence that most E-SE Asian strains were directly descendent from other E-SE Asian strains (14)—thus indicating that the temporally overlapping epidemics in E-SE Asia form a circulation network that, during the study period, has been mostly closed to external reseeding.

E-SE Asia's strong travel and trade connections to Oceania, North America, and Europe (14, 37) facilitate the rapid movement of new influenza virus variants into those areas and thus explain the relatively small lag in antigenic and genetic advancement seen in those regions (Fig. 2, A and B). Also, though it is unclear how much travel there must be between two locations for them to be epidemiologically well-connected, South America's 6- to 9-month antigenic lag (Fig. 2A) may be attributable to its paucity of direct connections with E-SE Asia (fig. S5). South America's strong travel connections to Europe and North America, but not to E-SE Asia, could result in a seeding hierarchy where strains are first seeded into North America and Europe and from there to South America (Fig. 5). Most strains appear to circulate in this simple hierarchy, and even those strains that circulate in a more complex hierarchy still originate in E-SE Asia (14). Thus, the extinction of many H3 lineages—a key characteristic of the H3 phylogeny—may, in addition to the accumulation of deleterious mutations (25), also be due to reaching the end of this hierarchy.

Fig. 5.

Schematic of the dominant seeding hierarchy of seasonal influenza A (H3N2) viruses. The structure of the network within E-SE Asia is unknown.

Surveillance and Vaccine Strain Selection

A major practical function of WHO's Global Influenza Surveillance Network is to assist regulatory authorities to recommend which strains should be included in influenza vaccines. Expanding surveillance within the E-SE Asian circulation network will aid the early detection of the emergence and spread of new variants and could help to more precisely define the network. Such surveillance is crucial for optimizing vaccine strain selection for countries within the network and for forecasting which variants will seed epidemics in the rest of the world, consequently increasing vaccine efficacy and ultimately reducing influenza morbidity and mortality worldwide.

Given the importance of the HA, it is the only portion of the virus genome that is currently sequenced routinely within the WHO Global Influenza Surveillance Network. Recently, whole-genome sequencing initiatives have provided important insights into the genesis and spread of reassortment viruses, their rapid migration, and the cocirculation of multiple lineages (4, 8, 9, 15). Expanded sequencing of whole genomes will provide additional markers for tracking the global migration of viruses and reveal potential differences between the global evolution of the HA and the other gene segments. Such sequencing efforts should include strains from E-SE Asia and be linked with antigenic data on HA and, in the longer term, with phenotype changes determined by other virus genes to fully understand the selection pressures on influenza viruses and their epidemiology.

The data used in this study were generated by the WHO Global Influenza Surveillance Network. Although there are biases in surveillance data, these biases do not have a substantial effect on the results (14). The methods we have used are generic and, although applied here to human influenza A (H3N2) viruses, are broadly applicable to influenza viruses in other species and to other pathogens.


We present evidence from antigenic and genetic analyses of HA that, from 2002 to 2007, influenza A (H3N2) virus epidemics worldwide were seeded each year by viruses that originated in E-SE Asia. We find evidence that temporally overlapping epidemics in E-SE Asia create a circulation network in which influenza A (H3N2) viruses continually circulate within the region by passing from epidemic to epidemic. E-SE Asia'sstrong travel and trade connections with Oceania, North America, and Europe, coupled with weak connections to South America, could explain the seeding hierarchy observed in this period where new virus variants first seed epidemics in Oceania, North America, and Europe and later in South America. The mostly one-way nature of this hierarchy suggests that, once A (H3N2) viruses leave E-SE Asia, they are unlikely to contribute to long-term viral evolution. If the trends observed during this period are an accurate representation of overall patterns of spread, then the antigenic characteristics of A (H3N2) viruses outside E-SE Asia may be forecast each year based on surveillance within E-SE Asia, with consequent improvements to vaccine strain selection and reductions in influenza A (H3N2) morbidity and mortality. Intensified surveillance, including whole-genome sequencing, and better understanding of the evolutionary selection pressures in E-SE Asia would further improve vaccine strain selection worldwide and potentially make influenza virus evolution more predictable.

Supporting Online Material

Materials and Methods

Figs. S1 to S6

Table S1


References and Notes

Stay Connected to Science

Navigate This Article