The changing patterns of genetic diversity in viral isolates from New York state and New Zealand clearly reveal the seasonal dynamics of influenza A in individual temperate populations (). In the case of the better-sampled A/H3N2 subtype, the pattern exposed by our coalescent-based analysis is of an annual series of peaks in genetic diversity interspersed by strong genetic bottlenecks at the end of most influenza seasons. As expected, genetic diversity of the New York state epidemics peak in the Northern Hemisphere winter, whereas those of New Zealand are offset by approximately 6 months, corresponding to the Southern Hemisphere winter. (A similar pattern was observed when 123 genome sequences from Australia were added, suggesting that these patterns are common to other temperate populations, see Supplementary Information.) Furthermore, the genetic diversity of A/H3N2 is usually lower in New Zealand than in New York state, probably reflecting the smaller host population in New Zealand (most isolates were sampled from Canterbury, South Island). Differences in the size of the host population may also explain why A/H3N2 in New Zealand is sometimes less diverse than A/H1N1 in New York state, even though A/H3N2 is usually epidemiologically dominant. The absolute amount of genetic diversity, even at seasonal peaks, is small compared to other rapidly evolving viruses that infect far fewer people15,16
, suggesting that strong natural selection, in addition to periodic bottlenecks, reduces the level of diversity that co-circulates at any time. Simulations demonstrate that our reconstructions of genetic diversity are robust to the sampling protocol (Supplementary Information).
Population dynamics of genetic diversity in influenza A virus
More notable is that, in both populations, A/H1N1 seasons exhibiting highly defined peaks in diversity typically coincide with weakly defined peaks in A/H3N2 diversity (that is, the measures of the epidemic ‘peakedness’ for A/H3N2 and A/H1N1 are negatively correlated; Wilcoxon signed-rank test, W
= 348, n
= 32, P
< 0.002; Supplementary Figs 1
). This implies an evolutionary interaction between subtypes; for example, that A/H1N1 epidemics are suppressed by herd immunity when A/H3N2 is dominant, or that A/H3N2 out-competes A/H1N1, perhaps owing to greater replicative fitness. Consistent with our observation, A/H1N1 only dominates in seasons following unusually mild H3N2 epidemics17
, and infection with one subtype can protect against reinfection with the other in sequential epidemics18
. Furthermore, since 1977, A/H1N1 epidemics exhibit lower mortality rates than A/H3N2 epidemics (refs 19
) and are less spatially synchronized21-23
. In both hemispheres, A/H1N1 seems less prone to the seasonal genetic bottlenecks that characterize A/H3N2, suggesting that genetically diverse A/H1N1 lineages are better able to coexist. This might indicate that antigenic selection acts with less potency on A/H1N1, manifest as lower rates of amino acid fixation in HA1 (ref. 17
), so that the selective turnover of lineages occurs more slowly17,24
The persistence of viral diversity between epidemic peaks for A/H3N2 and A/H1N1 has two explanations: that chains of infection are surviving in each population and across inter-epidemic intervals, or that genetic diversity is imported into temperate populations each year. Phylogenetic evidence strongly weighs against the former, because there are few direct phylogenetic links between influenza A viruses sampled in successive seasons from specific locations, as expected if in situ
evolution was commonplace25,26
. Furthermore, a high risk of stochastic extinction makes the repeated survival of small chains of infection between successive epidemics unlikely. Additionally, the dates of segment ancestry in the New York state 1999-2000 season reappeared precisely in the 2002-2003 season, after a major bottleneck in the A/H1N1-dominant 2000-2001 season, strongly suggesting that diversity was maintained in a reservoir population.
Extrapolating the evolutionary dynamics of influenza in New York state and New Zealand to other comparable populations in the Northern and Southern Hemispheres leads us to propose a ‘source-sink’ model for the global ecology of influenza A virus (). Continual—but largely unidirectional—gene flow from a common source population (or a linked network of source populations24,27
) provides the viruses that ignite each epidemic in populations of the Northern and Southern Hemispheres. Although southern China has been proposed as the epicentre of influenza A virus28
, it is possible that tropical regions generally represent ideal source populations because of extended viral transmission2
. A necessary consequence of this model is that selection-driven antigenic drift will be much more efficient in the source population, the long-term effective size of which is maintained by a high background infection rate and by the absence of the severe population bottlenecks associated with temperate localities2
. Hence, the observation of antigenic change in temperate populations is a secondary effect of selection within, and gene flow from, the source population. This explains the paradox of there being frequent positive selection on HA across seasons at a global scale29-31
but little evidence for antigenic drift at the scale of individual seasonal epidemics in those temperate populations studied so far26,32
. To test this hypothesis, it will be essential to obtain more influenza virus samples from tropical regions, such as South-East Asia.
A ‘source-sink’ model for the evolution of influenza A virus