Phylogenetic Analysis Reveals

the Global Migration of Seasonal

Influenza A Viruses
Martha I. Nelson1,2, Lone Simonsen3, Cecile Viboud4, Mark A. Miller4, Edward C. Holmes1,2,4*
1 Center for Infectious Disease Dynamics, The Pennsylvania State University, University Park, Pennsylvania, United States of America, 2 Department of Biology, The
Pennsylvania State University, University Park, Pennsylvania, United States of America, 3 National Institute of Allergy and Infectious Diseases, National Institutes of Health,
Bethesda, Maryland, United States of America, 4 Fogarty International Center, National Institutes of Health, Bethesda, Maryland, United States of America

The winter seasonality of influenza A virus in temperate climates is one of the most widely recognized, yet least
understood, epidemiological patterns in infectious disease. Central to understanding what drives the seasonal
emergence of this important human pathogen is determining what becomes of the virus during the non-epidemic
summer months. Herein, we take a step towards elucidating the seasonal emergence of influenza virus by determining
the evolutionary relationship between populations of influenza A virus sampled from opposite hemispheres. We
conducted a phylogenetic analysis of 487 complete genomes of human influenza A/H3N2 viruses collected between
1999 and 2005 from Australia and New Zealand in the southern hemisphere, and a representative sub-sample of viral
genome sequences from 413 isolates collected in New York state, United States, representing the northern
hemisphere. We show that even in areas as relatively geographically isolated as New Zealand’s South Island and
Western Australia, global viral migration contributes significantly to the seasonal emergence of influenza A epidemics,
and that this migration has no clear directional pattern. These observations run counter to suggestions that local
epidemics are triggered by the climate-driven reactivation of influenza viruses that remain latent within hosts between
seasons or transmit at low efficiency between seasons. However, a complete understanding of the seasonal
movements of influenza A virus will require greatly expanded global surveillance, particularly of tropical regions where
the virus circulates year-round, and during non-epidemic periods in temperate climate areas.
Citation: Nelson MI, Simonsen L, Viboud C, Miller MA, Holmes EC (2007) Phylogenetic analysis reveals the global migration of seasonal influenza A viruses. PLoS Pathog 3(9):
e131. doi:10.1371/journal.ppat.0030131

temperate zones, recent studies show that tropical regions

Introduction
experience significant year-round influenza virus activity [11].
Influenza A virus is able to persistently re-infect human In theory, such a ‘‘tropical belt’’ could serve as a year-round
populations by continually evading host immunity through reservoir that harbors endemic populations of influenza virus
the rapid evolution of surface antigens (‘‘antigenic drift’’) [1]. that seasonally reintroduce viral isolates into temperate zones
Influenza virus epidemics strike temperate latitudes of the to trigger new epidemics [12,13]. Whereas population crashes
world each winter, from November to March in the northern at the end of seasonal epidemics create severe evolutionary
hemisphere and from May to September in the southern bottlenecks that limit genetic diversity, tropical zones may
hemisphere [2]. In the United States alone, these influenza function as permanent mixing pools for viruses from around
epidemics are associated with an annual average of 36,000 the world. Historically, Southeast Asia has been considered a
human deaths [3] and 226,000 hospitalizations [4]; globally, potential epicenter for emergence of pandemic viruses due to
the virus is associated with as many as half a million annual the proximity with which humans live with their domestic
deaths [5]. While rapid antigenic change is a hallmark of animals [14]. However, abundant data from these regions is
influenza evolution, recent studies have failed to detect currently unavailable, so the origins of influenza pandemics
antigenic drift over an epidemic season, suggesting that
and epidemics remain unclear.
important evolutionary processes may occur during non-
Given the ease and speed with which the influenza virus is
epidemic periods, either locally or perhaps elsewhere [6–8].
thought to spread between humans, it is generally accepted
However, surveillance during non-epidemic periods is not
that global chains of direct person-to-person transmission are
conducted routinely by the network of World Health
Organization influenza reference centers [9] and, conse-
quently, little is known about how and where the virus Editor: Bruce Levin, Emory University, United States of America
persists in the human population in between winter Received March 30, 2007; Accepted July 24, 2007; Published September 14, 2007
epidemics at low levels. A key question is therefore whether
This is an open-access article distributed under the terms of the Creative Commons
the virus remains locally within its host population in Public Domain declaration which stipulates that, once placed in the public domain,
between epidemics, perhaps persisting within hosts in a this work may be freely reproduced, distributed, transmitted, modified, built upon,
latent state [10], or whether the virus migrates afar to other or otherwise used by anyone for any lawful purpose.
reservoirs, such as the tropics, and is later reintroduced. Abbreviations: HA, hemagglutinin; ML, maximum likelihood; NA, neuraminidase;
Although influenza virus has long been regarded a ‘‘cold- NJ, Neighbor-Joining
weather’’ pathogen due to its marked winter epidemics in * To whom correspondence should be addressed. E-mail: [email protected]

PLoS Pathogens | www.plospathogens.org 1220 September 2007 | Volume 3 | Issue 9 | e131

 Global Circulation of Influenza A Virus

Author Summary

The winter seasonality of influenza A virus in temperate climates is

one of the most puzzling epidemiological patterns in infectious
disease. To help resolve the issue of influenza seasonality, we
studied, using viral genome sequence data, the patterns of global
migration of influenza A virus, particularly between the northern
and southern hemispheres. A phylogenetic analysis of approx-
imately 900 complete genomes of the H3N2 subtype of human
influenza A virus sampled from New Zealand and Australia (southern
hemisphere), and New York state, United States (northern hemi-
sphere), revealed that cross-hemisphere migration frequently occurs
in both directions and involves multiple viral strains. Such global
viral traffic therefore contributes significantly to the introduction of
new influenza epidemics in both northern and southern hemi-
Figure 1. Models for the Global Evolution of Influenza Virus
spheres. These results also show that influenza A virus migrates afar
Two phylogenetic hypotheses depicting the evolution of A/H3N2
during non-epidemic periods, rather than persisting locally at low influenza virus by (A) global migration, in which isolates from adjacent
levels during the influenza ‘‘off-season’’. However, although this seasons in the northern and southern hemispheres are interspersed
represents the largest and first bihemisphere study of its kind to our topologically, or by (B) reactivation of latent virus, in which isolates from
knowledge, the results highlight the need for sampling from tropical the northern hemisphere give rise, in situ, to isolates that circulate in the
regions and during non-epidemic periods in temperate areas. same locality the next season, and thus are topologically linked (likewise
Studies of this kind are critical to fully understand the geographical for southern hemisphere viruses).
doi:10.1371/journal.ppat.0030131.g001
dispersal of influenza A virus and the role of climate in triggering
seasonal epidemics.
88 viral genome sequences from Australia (most commonly
Western Australia) from 1999 to 2005 (seven seasons), along
sufficient to maintain the influenza virus in the human with a carefully selected sub-sample of 52 isolates that are
population [15]. However, a complete understanding of how representative of the clades present in a larger sample of 413
the influenza virus transmits between humans is lacking [16], viruses from New York state, United States, collected between
and whether human-to-human spread alone accounts for the 1998 and 2005 and analyzed previously [7]. Given the relative
seasonal emergence of epidemics has been questioned [17]. geographic isolation and low population densities of New
The simultaneous appearance of influenza outbreaks sepa- Zealand’s South Island and Western Australia, as well as
rated by large longitudinal distances, as well as sporadic sampling limitations, this analysis provides a conservative
influenza cases during summer months, suggests that the virus estimate of the extent of cross-hemisphere migration occur-
may instead already be ‘‘seeded’’ and somehow reactivated by ring during this time period.
environmental stimuli. Thus, the alternating pattern of
northern and southern hemisphere bi-epidemics could, in Results
principle, also result from opposite climatic forces independ- Extensive Viral Genetic Diversity in Australia and New
ently reactivating viral activity in these two hemispheres at Zealand
alternating six-month intervals. Hence, instead of continually Populations of A/H3N2 influenza virus in Australia and
migrating across the equator, separate viral populations New Zealand from 1999 to 2005 exhibit extensive genetic
could persist locally in an asymptomatic latent state over diversity across the entire genome (Figures 2–4), comparable
the summer months until climatic stimuli sufficiently increase to the diversity observed previously in New York state [7] (in
host susceptibility and/or viral transmissibility to induce all phylogenies, clades from Australia are shaded blue, those
another epidemic. However, hypotheses of how climatic from New Zealand in green, those from New York state in
change may directly or indirectly influence viral activity orange, and global isolates in pink). In particular, multiple
and/or host susceptibility remain largely untested [18,19]. viral clades co-circulate during each influenza season in New
Crucially, some theories for influenza seasonality produce Zealand and Australia, defined as clusters of isolates with high
testable phylogenetic hypotheses. On one hand, if influenza A bootstrap support (.70%), or which are separated by
virus persists locally over the summer in a latent state, then exceptionally long branches (Table 1). As with the New York
isolates sampled over multiple seasons from a single locality state data, these viruses were collected in the context of
would cluster together on a phylogenetic tree, separate from seasonal surveillance efforts in Australia and New Zealand
isolates from other geographic regions (Figure 1). Alterna- and therefore likely provide a representative sample of the
tively, if the virus did not evolve in situ between epidemic overall genetic composition of the viral populations.
seasons, but rather traveled globally between epidemics, then On the neuraminidase (NA) tree (Figure 2), viruses from
the resulting phylogeny would show extensive intermixing of New Zealand and Australia fall into at least 15 distinct clades,
isolates from different localities. To determine whether some of which appear in multiple seasons: three clades
influenza virus migrates away between the northern and circulated in the 1999 season (clades v, vi, and ix; when data
southern hemispheres during non-epidemic summer months, was only available from Australia), five in 2000 (clades i, ii, iii,
or remains there latently, we conducted an extensive iv, viii), four in 2001 (clades i, iv, v, g), one in 2002 (clade a),
phylogenetic analysis of 399 whole-genome A/H3N2 influenza one in 2003 (clade b; when a new reassortant virus
A viruses sampled from New Zealand (most commonly predominated), two in 2004 (clades A and c), and three in
Canterbury, South Island) from 2000 to 2005 (six seasons), 2005 (clades A, B, E). Clade lettering reflects the three main

PLoS Pathogens | www.plospathogens.org 1221 September 2007 | Volume 3 | Issue 9 | e131

 Global Circulation of Influenza A Virus

Figure 2. Phylogenetic Relationships of the NA Gene of A/H3N2 Influenza Viruses Sampled from New York State (n ¼ 52), New Zealand (n ¼ 51),
Australia (n ¼ 45), and Globally (n ¼ 22) from 1998 to 2005, Estimated Using ML
Viral isolates from New York state are highlighted in orange, isolates from New Zealand in green, isolates from Australia in blue, and global isolates in
pink. Light yellow rectangles identify viral clades, with numbers in white boxes giving bootstrap values for key nodes (.70%). To clarify reassortment
events, capital letters in blue refer to clades that appear in section I of the tree (denoted in pink) on the phylogeny of the concatenated six non-surface
glycoproteins; lowercase letters in red refer to clades contained within section II (denoted in light green); and lowercase roman numerals in dark green
refer to clades outside sections I and II. Bootstrap values highlighted in yellow identify nodes that define a cross-hemisphere migration event. The tree is
rooted by isolate A/New York/327/1999 from the 1998–1999 season (i.e., the earliest sampled isolate), and all horizontal branch lengths are drawn to a
scale of nucleotide substitutions per site.
doi:10.1371/journal.ppat.0030131.g002

sections of the phylogeny, based on topology and time: clades Differences in the number of clades among segments may
A–E contain viral isolates from 2003 to 2005 that fall within also be indicative of reassortment, especially involving the
section I of the tree (large pink rectangle in upper portion of HA gene, as previously demonstrated in New York state [7].
tree); clades a–e contain isolates from 2001–2003 that fall Indeed, several major reassortment events are immediately
within section II (large light green rectangle in middle evident from topological incongruities among these three
portion of tree); clades i–xiii contain isolates from 1998 to phylogenies. On the HA tree, clades b, c, and e fall in section I
2000 that fall outside of sections I and II. The hemagglutinin along with isolates from 2004 and 2005, while these clades fall
(HA) tree (Figure 3) contains at least 15 clades from New in section II amidst 2002 and 2003 isolates in the NA and
Zealand and Australia: two circulated in 1999 (clades v and x), concatenated six non-surface glycoprotein phylogenies (Fig-
four in 2000 (clades i, ii, iii, iv), four in 2001 (clades g, i, iv, ures 2–4). Clade c also falls in section I on the PB2 phylogeny,
xiii), two in 2002 (clades h and j), one in 2003 (clade b), two in showing a similar reassortment pattern as HA. However, aside
from this lone reassortment event, the phylogenies of the six
2004 (clades A and c), and three in 2005 (clades A, B, E)
non-surface glycoprotein segments are very similar, enabling
(Figure 3). Finally, the phylogeny of the concatenated six non-
us to study them as a single concatenated entity (phylogenetic
surface glycoprotein segments (PB2, PB1, PA, NP, M1, NS1)
trees for individual segments are provided as Figures S1–S6).
contains the largest number of clades, presumably because
this larger data set (9,636 bp) provides the greatest resolution. Intermixing among Viral Populations from Australia, New
Southern hemisphere isolates form at least 18 clades on this Zealand, and New York State
phylogeny: four were present in 1999 (clades v, vi, x, xii), three Isolates from the 52 representative New York state
in 2000 (clades i/ii, iii, iv), four in 2001 (clades g, i/ii, iv, xiii), genomes are clearly interspersed with southern hemisphere
three in 2002 (clades a, k, l), one in 2003 (clade b), three in isolates throughout our phylogenetic trees (Figures 2–4; Table
2004 (clades A, F, c), and three in 2005 (clades A, B, E) (Figure 1), indicating that these populations regularly intermix as a
4). result of cross-hemisphere migration. However, patterns of

PLoS Pathogens | www.plospathogens.org 1222 September 2007 | Volume 3 | Issue 9 | e131

 Global Circulation of Influenza A Virus

Figure 3. Phylogenetic Relationships of the HA Gene of A/H3N2 Influenza Viruses Sampled from New York State (n ¼ 52), New Zealand (n ¼ 51),
Australia (n ¼ 45), and Globally (n ¼ 13) from 1998 to 2005, Estimated Using ML
Color scheme, rooting, scale, and symbols are the same as those used in Figure 2.
doi:10.1371/journal.ppat.0030131.g003

viral intermixing are both variable and complex, as clades tropical regions, where influenza viruses typically circulate
from all three phylogenies contain an array of different year-round [12].
combinations of viral populations from New Zealand, Strikingly, even for those viral clades that do not exhibit
Australia, and/or New York state. For example, the phylogeny cross-hemisphere migration, there is very little evidence for
of the concatenated six non-surface glycoproteins contains a in situ evolution within specific localities. For example, in no
relatively even mix of mono-hemisphere and bi-hemisphere case on any phylogeny are clades of A/H3N2 from New York
clades (Table 1), with five New York state–only clades, seven state directly linked over multiple seasons. Rather, New
southern hemisphere–only clades, six clades that contain New Zealand and Australia viruses are always interspersed among
York state isolates and isolates from only one of the southern New York state clades from different seasons, indicating that
hemisphere countries, and three clades that contain isolates they are not evolving in geographic isolation across seasons.
from all three countries. Thus, on an annual basis some, but Most clades from New Zealand and Australia show equivalent
not all, viral populations mix with other populations from the patterns of discontinuous evolution, with very few clusters of
same and/or opposite hemisphere, and this number is likely southern hemisphere clades that are not separated by New
to increase with additional sampling. Indeed, a majority York state viruses (although in situ evolution cannot be ruled
(nine) of the 16 clades containing southern hemisphere out between a few 2004 and 2005 clades without additional
isolates also contained northern hemisphere isolates, suggest- sampling). Thus, even in relatively isolated areas of New
ing widespread viral traffic across the equator (Table 1). Zealand and Australia, viruses do not regularly evolve in
Migration between Australia and New Zealand is also geographic isolation. Rather, evolution appears to be shaped
extensive, as almost all clades containing isolates from New by frequent cross-hemisphere migration and recurrent
Zealand also contained Australian isolates, and vice versa, reintroduction.
except for the 1999 season, for which no New Zealand isolates
were available. Further, these figures are likely to be under- Bi-Directional Cross-Hemisphere Migration Occurs
estimates, as mono-national and mono-hemisphere clades between Influenza Seasons
may be at frequencies too low to be detected in the genome Importantly, our phylogenetic analysis suggests that sea-
collections currently available. Alternatively, clades could also sonal migration occurs from the northern hemisphere to the
have originated in areas not sampled in our study, such as southern hemisphere, as well as south-to-north. Although

PLoS Pathogens | www.plospathogens.org 1223 September 2007 | Volume 3 | Issue 9 | e131

 Global Circulation of Influenza A Virus

Figure 4. Phylogenetic Relationships of the Concatenated Six Non-Surface Glycoprotein Segments of A/H3N2 Influenza Viruses Sampled from New York
State (n ¼ 52), New Zealand (n ¼ 51), and Australia (n ¼ 45) from 1998 to 2005, Estimated Using ML
Color scheme, rooting, scale, and symbols are the same as those used in Figure 2.
doi:10.1371/journal.ppat.0030131.g004

inferring the direction of viral migration is in part dependent intervals between the northern and southern hemispheres, it
on sample composition, definitive evidence of a migration was also possible to determine, within the confines of
event are clades containing a single population of northern sampling, the timescale and hence direction of cross-hemi-
hemisphere viruses and a single population of southern sphere migration. The inferred directionality of 11 cases of
hemisphere viruses supported by a high level of bootstrap definitive cross-hemisphere migration evident on the HA, NA,
support (.70%). and non-surface glycoprotein phylogenies are summarized in
Because winter influenza seasons alternate by six-month Table 2. Of these, all but two involve the concatenated non-
surface glycoproteins, which provide a more reliable phylog-
eny as previously described. Eight of these migration events
Table 1. Clades in the HA, NA, and Concatenated Six Non- occur in a north-to-south direction, versus three in a south-to-
Surface Glycoprotein Segment Phylogenies Comprised of Viruses north direction, suggesting that viruses may migrate more
from New York State, New Zealand, Australia, and Globally frequently from New York state to the southern hemisphere
than in the opposite direction, although this will need to be
Locality HA NA Non-Surface confirmed with larger sample sizes. For example, on the
Glycoproteins concatenated gene tree, viral isolates from the 2003–2004
season in New York state form a single well-supported
NY only C, Da, e, d, f, vi, xi (7) C, Da, ea, d, f, viia (6) C, D, e, d, f (5)
NZ only i (1) i, viii (2) (0)
phylogenetic clade (clade b, 100% bootstrap support) with
AUS only c, g, xb, xiii (4) c, g, ixb (3) g, x, xiib, xiiib (4) viruses from 2003 from New Zealand and Australia (Figure 4).
NY, NZ ii (1) iia (1) F, i/ii (2) Since the 2003 New Zealand and Australia viruses predate the
NY, AUS va,b (1) vib, va,b (2) c, vb, vib (3) 2003–2004 New York state viruses (i.e., the northern hemi-
NZ, AUS Aa, B, E, iii, iv (5) Aa, B, E, a, iii, iva (6) A, B, E, a, iv (5)
NY, NZ, AUS ba, h, j (3) ba (1) b, k, l, iii (4) sphere winter), we infer that the lineage that gave rise to these
southern hemisphere viruses migrated northward to infect
New York state between the 2003 southern hemisphere winter
Clade letters/roman numerals refer to Figures 2–4. Numbers in parentheses refer to the
total number of clades in each category. (May to October) and the 2003–2004 winter in New York state.
a
b
Clades that also contain global isolates (HA and NA only). It is also notable that cross-hemisphere migration does not
Clades found in 1999 for which no New Zealand sequences were available.
AUS, Australia; NY, New York state; NZ, New Zealand; SA, South Australia.
follow any clear pattern. In addition to occurring in both
doi:10.1371/journal.ppat.0030131.t001 north-to-south and south-to-north directions, migration

PLoS Pathogens | www.plospathogens.org 1224 September 2007 | Volume 3 | Issue 9 | e131

 Global Circulation of Influenza A Virus

Table 2. Direction of 11 Migration Events between New York State in the Northern Hemisphere and Australia and New Zealand in the
Southern Hemisphere

Clade Gene Northern Hemisphere Direction Southern Hemisphere

Clade F Non-surface glycoproteins NY 2004–2005 (major) ‹ NZ 2004 (minor)

Clade b HA, NA, non-surface glycoproteins NY 2003–2004 (major) ‹a AUS/NZ 2003 (major)
Clade c Non-surface NY 2003–2004 (minor) fi AUS 2004 (minor)
Clade h HA NY 2001–2002 (major) fi AUS/NZ 2002 (minor)
Clade j HA NY 2001–2002 (major) fi AUS/NZ 2002 (major)
Clade k Non-surface glycoproteins NY 2001–2002 (minor) fi AUS/NZ 2002 (minor)
Clade l Non-surface glycoproteins NY 2001–2002 (major) fi AUS/NZ 2002 (minor)
Clade ii HA, NA, non-surface glycoproteins NY 1999–2000 (major) fi NZ 2000 (minor)
Clade iii Non-surface glycoproteins NY 1999–2000 (major) fi NZ/AUS 2000 (minor)
Clade v HA, non-surface NY 1998–1999 (minor) fia AUS 1999 (major)
Clade vi Non-surface glycoproteins NY 1999–2000 (minor) ‹ AUS 1999 (minor)

Clade letters/roman numerals refer to Figures 2–4. Red arrows signify north-to-south migration events; blue arrows denote south-to-north migration. Viral populations are classified as
belonging to the major clade of viruses that circulated each season in a given locality or to a minor clade.
a
The additional presence of global strains in a clade.
AUS, Australia; NY, New York state; NZ, New Zealand.
doi:10.1371/journal.ppat.0030131.t002

events also appear to involve minor clades as frequently as either the previous or following season in the same location.
major clades, assuming that our study sample is generally Thus, viral populations do not appear to ‘‘over-summer’’
representative of the viral population structure (Table 2). locally, where they would evolve in situ and give rise to the
Furthermore, the populations of southern hemisphere viruses next season’s epidemic. Rather, cycles of viral migration and
that migrate northward are a mix of compositions, including recurrent introduction have clearly played a significant role
some isolates only from New Zealand, some only from in generating the genetic diversity that characterizes influ-
Australia, and populations with a mix of isolates from both enza A virus in both hemispheres. Importantly, given the
countries (Tables 1 and 2). Finally, two clades contain global sample composition of our sequence data set, the extent of
strains, but because the dates of these global isolates are not cross-hemisphere migration observed here undoubtedly
recorded, it is impossible to accurately determine the represents a conservative estimate. Hence, including data
direction of migration events. from more populated areas could only reveal more instances
The 11 cases presented in Table 2 represent only the of cross-hemisphere migration.
strongest examples of cross-hemisphere migration, using the In addition, our finding that the virus migrates globally
strictest criteria to infer migration events from the phyloge- between epidemics and is reintroduced is clearly compatible
netic data. Relaxing this stringency allows for the possibility with tropical regions, including Southeast Asia, playing a key
of greater bi-directional cross-hemisphere migration, espe- role in the genesis of new clades and the global spread of
cially involving clades containing more than one viral these novel influenza virus variants. Thus, while limitations in
population from the southern hemisphere. Although the global genome sampling necessarily means that the current
direction of migration is less certain when populations of study is directed toward testing hypotheses of viral migration
viruses from three geographical regions are present, the versus latency, equivalent data from tropical regions would
relative frequency of migration observed under the more undoubtedly enable us to conduct a more refined analysis of
stringent criteria suggests that cross-hemisphere migration global migration patterns and their determinants. Specifi-
likely operates in these cases as well. Finally, while it is likely cally, if tropical regions serve as year-long influenza reser-
that a more intensive sampling regime will increase clade voirs, we would expect to observe phylogenies in which
diversity, and in doing so affect the inference of the direction tropical isolates display the greatest genetic diversity and are
of migration, the complexity of the patterns observed positioned basal to viruses sampled from temperate regions.
strongly argues for frequent bi-directional migration. Consequently, complete genome sampling from tropical
regions where influenza viruses circulate year-round, includ-
ing a record of the precise date of collection, is of key
Discussion
importance for understanding the global epidemiology of the
Our large-scale phylogenetic analysis of A/H3N2 influenza influenza virus.
virus populations from opposite geographic hemispheres Notably, the viral migration we observe does not appear to
provides evidence for regular bi-directional cross-hemi- follow any clear pattern, but rather occurs in all directions,
sphere viral migration between seasons, even among localities involves all genes, and involves clades of all sizes and
as distantly separated as New York state and Australia and as geographic compositions. This argues against a role of
relatively geographically isolated as New Zealand’s South immune selection in determining which viral clades are able
Island. Multiple genetic variants of influenza virus co- to migrate among localities, although it does not preclude a
circulate each season, even in geographically remote areas, role for natural selection as the sieve that determines which
and many of these viral clades are more closely related to clades are able to survive in specific host populations.
isolates from the opposite hemisphere than to isolates from Similarly, the observation that migration patterns vary to

PLoS Pathogens | www.plospathogens.org 1225 September 2007 | Volume 3 | Issue 9 | e131

 Global Circulation of Influenza A Virus

some extent among the HA, NA, and concatenated non- Materials and Methods
surface glycoproteins must reflect the effect of widespread
Influenza viruses used in this study. All influenza A (H3N2) virus
genomic reassortment [7,20]. Frequent reassortment compli- complete genome sequence data were collected from the National
cates the analysis of migration patterns, as individual viruses Institute of Allergy and Infectious Disease’s Influenza Genome
can carry genomic segments with differing phylogenetic, and Sequencing Project (http://www.niaid.nih.gov/dmid/genomes/mscs/
influenza.htm) for the period 1998–2005 [28]. Influenza A/H3N2
hence geographical, histories. Consequently, the analysis of viruses were sampled by a network of participating general
migration patterns based on single gene segments may paint a practitioners. Viruses from all 11 regions in New York state were
misleading picture. collected by the Virus Reference and Surveillance Laboratory at the
Wadsworth Center, New York State Department of Health. Influenza
Although the transmission of the influenza virus through viruses from both the North and South Islands of New Zealand were
population movements has been studied extensively, partic- collected by Canterbury Health Laboratories in Christchurch, New
ularly for the spread of pandemic isolates across the globe by Zealand. In Australia, viruses from Western Australia were collected
air travel [21,22], neither the routes nor the mechanisms of by PathWest Laboratory Medicine, Western Australia; viruses from
New South Wales were collected by the Prince of Wales Hospital, New
the virus’s geographical spread have been fully resolved. South Wales; viruses from South Australia were collected by the
Several recent studies have used empirical data to investigate Institute of Medical and Veterinary Sciences, South Australia; and
the role of population movements on the spatial diffusion of viruses from Queensland were collected by the Queensland Health
Science Services, Queensland. All sequence data were downloaded
seasonal epidemics, including an intricate analysis of the from the National Center for Biotechnology Information (NCBI)
regional spread of influenza epidemics across the United Influenza Virus Resource (http://www.ncbi.nlm.nih.gov/genomes/FLU/
States, which was strongly correlated with adult workflow FLU.html). For Australia, 88 genome sequences from the 1999–2005
seasons were compiled, while for New Zealand, 399 genome
movements [23]. A previous epidemiological study comparing sequences A/H3N2 sequences from the 2000–2005 seasons were
the synchronicity with respect to timing of influenza collected. For New York state, United States, 52 phylogenetically
epidemics between the United States, France, and Australia representative genome sequences from the 1998–1999 to 2004–2005
suggested that the inter-hemispheric circulation of epidemics seasons were carefully selected from a larger data set of 413 sequences
from 1997–2005 analyzed previously [7] (excluding 2000–2001, for
follows an irregular pathway, with recurrent changes in the which few H3N2 sequences were available in an H1N1-dominant
leading hemisphere [24], in accordance with the phylogenetic season). GenBank accession numbers for all sequences used in this
analysis presented here. More fine-scaled analyses of discrete study are listed in Table S1.
Phylogenetic analysis. Sequence alignments were manually con-
viral populations have shown that frequent introduction of structed for the major coding regions of each of the eight genomic
‘‘foreign’’ viruses significantly impacts the viral population segments. In addition to alignments for the HA (1,698 bp) and NA
structure and geographic spread at local levels. For example, (1,407 bp), an alignment of the concatenated six non-surface
glycoproteins segments (PB2, PB1, PA, NP, M1, NS1) was also
the rapid timescale of global mixing of influenza drowns out compiled (9,636 bp), as these are expected to evolve differently from
any impact of local heterogeneities on the spread of the the HA and NA surface glycoproteins. Because the minor M2 and NS2
epidemics through France [25]. Similarly, the seasonal proteins are involved in overlapping reading frames, they were
importation of multiple global isolates appears to be a excluded from the analysis.
Initial phylogenetic trees were inferred for sequences of the HA,
greater contributor to the genetic diversity of the influenza NA, and concatenated non-surface glycoproteins from New York
virus population in New York state from 1997 to 2005 than state, New Zealand, and Australia under the HKY85 (Hasegawa-
local in situ evolution [7]. While our findings confirm that Kishino-Yano) model of nucleotide substitution using the Neighbor-
Joining (NJ) method available in PAUP* [29]. Due to the very large
human population movements play a role in introducing new size of all data sets, and the provisional nature of the analysis, the
viral variants at the start of an epidemic, some aspect of nearest-neighbor-interchange branch-swapping method was em-
climate is clearly of importance in triggering epidemics. ployed in this case. To assess the robustness of individual nodes on
Additional research is required to define how human these phylogenetic trees, we performed a bootstrap resampling
analysis (1,000 replications) using the NJ method. From these three
susceptibility to infection and viral transmissibility fluctuate starting phylogenetic trees, ‘‘major’’ clades (which contained the
under varying climate conditions and why influenza virus is majority of isolates from a season) and ‘‘minor’’ clades of genetically
absent in summer in temperate climates but exists year-round related viruses were identified by exceptionally long branch lengths
and/or high bootstrap values (.70%). A subset of sequences for the
in tropical zones. concatenated non-surface glycoproteins was constructed with 51
Although the underlying cause of the seasonality of the sequences from New Zealand, 45 sequences from Australia, and 52
influenza virus remains uncertain, even in reservoir avian from New York state (see above) for a total data set of 148 sequences.
For the HA gene, these 148 isolates were placed in a more global
species [26], our findings illustrate the critical importance of context with the addition of 13 genetically unique HA sequences
expanding surveillance to elucidate the geographical move- sampled from this time period available on GenBank, to produce a
ments and evolution of this virus throughout its entire annual total of 161 HA sequences. Likewise, 22 global NA sequences were
cycle. The traditional focus on epidemic influenza may combined with the original 148 from New York state, New Zealand,
and Australia for a total of 170 NA sequences. Maximum likelihood
detract from the equally important epidemiological question (ML) phylogenetic trees were then inferred using the PAUP* package
of why influenza A virus does not circulate in humans for so [29] for these three new data sets: 161 HA sequences, 170 NA
many months of the year in temperate areas, especially given sequences, 148 concatenated sequences. ML trees were also inferred
for each of the six non-surface glycoprotein segments to ensure that
its apparent ability to infect humans in tropical areas year- all exhibit similar tree topologies (Figures S1–S6). In each case, the
round. Attempts to predict, model, or contain the spread of best-fit model of nucleotide substitution was identified by MOD-
the influenza virus require a unified understanding of how ELTEST [30] as the general reversible GTRþIþC4 model, with the
frequency of each substitution type, proportion of invariant sites (I),
the virus’s spatial-temporal dynamics, antigenic evolution, and the gamma distribution (C) of among-site rate variation with four
and seasonal emergence interrelate [27]. Although this study rate categories (C4) estimated from the empirical data. In all cases
is limited to only the three countries for which we have tree bisection-reconnection branch-swapping was utilized to deter-
extensive data, our analysis exemplifies the capacity of mine the optimal tree. Finally, a bootstrap resampling process (1,000
replications) using the NJ method was used to assess the robustness of
phylogenetic analysis to elucidate challenging epidemiolog- individual nodes on the phylogeny, incorporating the ML substitu-
ical questions by providing a level of finer resolution. tion model.

PLoS Pathogens | www.plospathogens.org 1226 September 2007 | Volume 3 | Issue 9 | e131

 Global Circulation of Influenza A Virus

The analysis of the frequency and directionality of migration was Color scheme, rooting, scale, and symbols are the same as those used
undertaken through a visual inspection of the topological position of in Figure S1.
individual clades on each tree and in consideration of their time of Found at doi:10.1371/journal.ppat.0030131.sg003 (970 KB EPS).
sampling. Although more quantitative methods for determining
migration patterns from gene sequence data have been established, Figure S4. Phylogenetic Relationships of the NP Gene of A/H3N2
particularly those based on parsimony reconstructions of changes in Influenza Viruses Sampled from New York State (n ¼ 52), New
character state (i.e., geographical locality) [31], these were considered Zealand (n ¼ 51), and Australia (n ¼ 45) from 1998–2005, Estimated
inappropriate for the current study because they ignore the temporal Using ML Method
structure of the influenza virus genome sequence data. Specifically, Color scheme, rooting, scale, and symbols are the same as those used
we reasonably assume that older sampled clades give rise to younger in Figure S1.
sampled clades if they fall basal to them on phylogenetic trees.
Found at doi:10.1371/journal.ppat.0030131.sg004 (951 KB EPS).
Figure S5. Phylogenetic Relationships of the M1 Gene of A/H3N2
Supporting Information Influenza Viruses Sampled from New York State (n ¼ 52), New
Figure S1. Phylogenetic Relationships of the PB2 Gene Segment of A/ Zealand (n ¼ 51), and Australia (n ¼ 45) from 1998–2005, Estimated
H3N2 Influenza Viruses Sampled from New York State (n ¼ 52), New Using ML Method
Zealand (n ¼ 51), and Australia (n ¼ 45) from 1998–2005, Estimated Color scheme, rooting, scale, and symbols are the same as those used
Using ML Method in Figure S1.
Viral isolates from New York state are highlighted in orange, isolates Found at doi:10.1371/journal.ppat.0030131.sg005 (3.9 MB EPS).
from New Zealand in green, isolates from Australia in blue, and
global isolates in pink. Light yellow rectangles identify viral clades, Figure S6. Phylogenetic Relationships of the NS1 Gene of A/H3N2
with numbers in white boxes giving bootstrap values for key nodes Influenza Viruses Sampled from New York State (n ¼ 52), New
(.70%). Capital letters in blue refer to clades that appear in section I Zealand (n ¼ 51), and Australia (n ¼ 45) from 1998–2005, Estimated
of the tree (denoted in pink) on the phylogeny of the concatenated Using ML Method
six non-surface glycoprotein segments; lowercase letters in red refer Color scheme, rooting, scale, and symbols are the same as those used
to clades contained within section II (denoted in light green); in Figure S1.
lowercase roman numerals in dark green refer to clades outside
sections I and II. Bootstrap values highlighted in yellow identify nodes Found at doi:10.1371/journal.ppat.0030131.sg006 (830 KB EPS).
that define a cross-hemisphere migration event. The tree is rooted by Table S1. Complete Genome Sequences of A/H3N2 Influenza Viruses
isolate A/New York/327/1999 from the 1998–1999 season (i.e., the Sampled from Australia, New Zealand, and New York State Used in
earliest sampled isolate), and all horizontal branch lengths are drawn This Study
to a scale of nucleotide substitutions per site.
Found at doi:10.1371/journal.ppat.0030131.st001 (535 KB DOC).
Found at doi:10.1371/journal.ppat.0030131.sg001 (975 KB EPS).
Figure S2. Phylogenetic Relationships of the PB1 Gene Segment of A/
H3N2 Influenza Viruses Sampled from New York State (n ¼ 52), New Acknowledgments
Zealand (n ¼ 51), and Australia (n ¼ 45) from 1998–2005, Estimated We thank all those who have generously denoted their samples of
Using ML Method influenza virus to the Influenza Genome Sequencing Project.
Color scheme, rooting, scale, and symbols are the same as those used Author contributions. MIN and ECH conceived the study. MIN
in Figure S1. performed the genome sequence analysis. MIN, LS, CV, MAM, and
Found at doi:10.1371/journal.ppat.0030131.sg002 (960 KB EPS). ECH wrote the paper.
Funding. This work was supported by National Institutes of Health
Figure S3. Phylogenetic Relationships of the PA Gene Segment of A/ grant number GM080533-01.
H3N2 Influenza Viruses Sampled from New York State (n ¼ 52), New Competing interests. The authors have declared that no competing
Zealand (n ¼ 51), and Australia (n ¼ 45) from 1998–2005, Estimated interests exist.
Using ML Method

References 13. Alonso WJ, Viboud C, Simonsen L, Hirano EW, Daufenbach LZ, Miller MA
1. Fitch WM, Bush RM, Bender CA, Cox NJ (1997) Long term trends in the (2007) Seasonality of influenza in Brazil: A traveling wave from the Amazon
evolution of H(3) HA1 human influenza type A. Proc Natl Acad Sci 94: to the subtropics. Am J Epidemiol 165: 1434–1442.
7712–7718. 14. Webster RG, Bean WJ, Gorman OT, Chambers TM, Kawaoka Y (1992)
2. Hope-Simpson RE (1981) The role of season in the epidemiology of Evolution and ecology of influenza A viruses. Microbiol Rev 56: 152–79.
influenza. J Hyg 86: 35–47, 35–47. 15. Cox NJ, Subbarao K (2000) Global epidemiology of influenza: Past and
3. Thompson WW, Shay DC, Weintraub E, Brammer L, Cox N, et al. (2003) present. Annu Rev Med 51: 407–421.
Mortality associated with influenza and respiratory syncytial virus in the 16. Bridges CB, Kuehnert MJ, Hall CB (2003) Transmission of influenza:
United States. JAMA 289: 179–186. Implications for control in health care settings. Healthcare Epidemiol 37:
4. Thompson WW, Shay DC, Weintraub E, Brammer L, Bridges CB, et al. 1094–1101.
(2004) Influenza-associated hospitalizations in the United States. JAMA 292: 17. Hope-Simpson RE, Golubev DB (1987) A new concept of the epidemic
1333–1340. process of influenza A virus. Epidemiol Infect 99: 5–54.
5. Stohr K (2002) Influenza—WHO cares. Lancet Infect Dis 2: 517. 18. Dowell SF (2001) Seasonal variation in host susceptibility and cycles of
6. Lavenu A, Luruez-Ville M, Chaix ML, Boelle PY, Rogez S, et al. (2005) certain infectious diseases. Emerg Infect Dis 7: 369–374.
Detailed analysis of the genetic evolution of influenza virus during the
19. Lofgren E, Fefferman N, Naumov YN, Gorski J, Naumova EN (2007)
course of an epidemic. Epidemiol Infect 134: 514–520.
Influenza seasonality: Underlying causes and modeling theories. J Virol 81:
7. Nelson MI, Simonsen L, Viboud C, Miller MA, Taylor J, et al. (2006)
5429–5436.
Stochastic processes are key determinants of the short-term evolution of
20. Holmes EC, Ghedin E, Miller N, Taylor J, Bao Y, et al. (2005) Whole genome
influenza A virus. PLoS Pathog 2: e125.
8. Wolf YI, Viboud C, Holmes EC, Koonin EV, Lipman DJ (2007) Long analysis of human influenza A virus reveals multiple persistent lineages and
intervals of stasis punctuated by bursts of positive selection in the seasonal reassortment among recent H3N2 viruses. PLoS Biol 3: e300.
evolution of influenza A virus. Biol Direct 1: 1–19. 21. Grais RF, Ellis JH, Glass GE. (2003) Assessing the impact of airline travel on
9. Kitler ME, Gavinio P, Lavanchy D (2002) Influenza and the work of the the geographic spread of pandemic influenza. Eur J Epidemiol 18: 1065–
World Health Organization. Vaccine 20: S5–S14. 1072.
10. Hope-Simpson RE, Golubev DB (1987) A new concept of the epidemic 22. Rvachev LA, Longini IM Jr (1985) A mathematical model for the global
process of influenza A virus. Epidemiol Infect 99: 5–54. spread of influenza. Math Biosci 75: 3–22.
11. Peiris M, Wong CM, Yang L, Chan KP, Leung G, et al. (2006) Influenza 23. Viboud C, Bjornstad ON, Smith DL, Simonsen L, Miller MA, et al. (2006)
associated weekly hospitalization in a subtropical city. PLoS Med 3: e121. Synchrony, waves, and spatial hierarchies in the spread of influenza.
12. Viboud C, Alonso WJ, Simonsen L (2006) Influenza in tropical regions. Science 312: 447–451.
PLoS Med 3: e89. 24. Viboud C, Boelle PY, Pakdaman K, Carrat F, Valleron AJ, et al. (2004)

PLoS Pathogens | www.plospathogens.org 1227 September 2007 | Volume 3 | Issue 9 | e131

 Global Circulation of Influenza A Virus

Influenza epidemics in the United States, France, and Australia, 1972–1997. (2005) Large-scale sequencing of human influenza reveals the dynamic
Emerg Infect Dis 10: 32–39. nature of viral genome evolution. Nature 437: 1162–1166.
25. Bonabeau E, Toubiana L, Flahault A (1998) The geographical spread of 29. Swofford DL (2003) PAUP*: Phylogenetic analysis using parsimony (*and
influenza. Proc R Soc Lond B 265: 2421–2425. other methods) version 4.0 [computer program]. Sunderland (Massachu-
26. Halvorson DA, Kelleher CJ, Senne DA (1985) Epizootiology of avian setts): Sinauer Associates.
influenza: Effect of season on incidence in sentinal ducks and domestic 30. Posada D, Crandall KA (1998) MODELTEST: Testing the model of DNA
turkeys in Minnesota. Appl Environ Micro 49: 914–919. substitution. Bioinformatics 14: 817–818.
27. Nelson MI, Holmes EC (2007) The evolution of epidemic influenza. Nat Rev 31. Carrington CVF, Foster JE, Pybus OG, Bennett SN, Holmes EC (2005)
Gen 8: 196–296. Invasion and maintenance of dengue virus type 2 and type 4 in the
28. Ghedin E, Sengamaly NA, Shumway M, Zaborsky J, Feldblyum T, et al. Americas. J Virol 79: 14680–14687.

PLoS Pathogens | www.plospathogens.org 1228 September 2007 | Volume 3 | Issue 9 | e131