Genomic Analysis and Geographic Visualization of the Spread of Avian Influenza (H5N1)

Wild birds are known to carry all strains of influenza and, in theory, any of these strains could be the source of the next human pandemic. Most influenza infections in humans since 1968 have been attributed to influenza A, antigenic subtypes H3N2 or H1N1. However, due to recent and novel infections of humans with an avian strain of influenza (H5N1), a great deal of attention continues to be focused on the spread of H5N1. Key questions include which hosts have carried the virus from Asia to Russia, the Middle East, East and West Africa, and Europe, and which mutations of H5N1 have allowed it to be transmitted among various hosts, including humans. We have created an interactive genomic and geographic map using phylogenetic software and Google Earth (GE; earth.google.com) to reconstruct the evolution and spread of H5N1 influenza lineages over the past decade. Our results provide insight on competing hypotheses as to which avian hosts are responsible for the spread of H5N1. We show that the answers to these questions are temporally and geographically context specific. Various lineages of H5N1 spreading from southern China in the past decade have exploited diverse pathways in terms of avian host taxa and geography. By examining H5N1 phylogenies projected onto a globe and using character evolution to reconstruct host shifts, we studied, visually and statistically, whether key genotypes in viral proteins are correlated with the spread of the virus geographically and among hosts. We find that a key genotype (Lys-627 in polymerase basic protein 2, PB2) that allows for increased replication and virulence of H5N1 in laboratory mice (Subbarao et al., 1993; Shinya et al., 2004) is also significantly associated with mammalian hosts in the field (Table 1). In visualizations, this genotype appears to be prevalent in isolates of H5N1 circulating west of the East Asian–Australian flyway (Figure 1). However, this pattern is not supported statistically (Table 1). We also find that a genotype of a surface protein of H5N1 (Arg-110 in neuraminidase, NA) is correlated with geographically distinct clades but not strongly correlated with any host type. Screenshots from aiTrees.kmz found at systematicbiology.org. (a) Screenshot from an east (foreground) to west (background) perspective of a phylogenetic tree for 291 isolates. Branches of the tree are traced with color to represent the optimization of a character for taxonomic order of hosts. In this screenshot, only the hosts relevant for that point of view are provided in the key; greater detail can be found in aiTrees.kmz. (Images i to iv) A temporal series of screenshots showing the movement of the genotype Lys-627 in PB2, which increases the ability of the virus to replicate in mammalian hosts, as red branches on a phylogenetic tree for 351 isolates. Images i to iv show movement of Lys-627 in PB2 as of 2000, 2002, 2004, and 2006, respectively. (b) A close-up view of a description box containing an account of mutations for that isolate (see detail for all isolates and HTUs aiTrees.kmz). (c) A south (foreground) to north (background) view of avian influenza spread from East Asia, showing Lys-627 position in PB2 character optimization as colored branches. Earth background image sources: Google, TerraMetrics, and NASA. The correlation between phenotypes and various genotypes calculated using Maddison's (1990) concentrated changes test. To correct for multiple testing we set the significance level at CCT ≤ 0.0125. Significant associations are in bold, and nearly significant (0.0125 < CCT ≤ 0.05) associations are in italics. The correlation between phenotypes and various genotypes calculated using Maddison's (1990) concentrated changes test. To correct for multiple testing we set the significance level at CCT ≤ 0.0125. Significant associations are in bold, and nearly significant (0.0125 < CCT ≤ 0.05) associations are in italics. In this paper, we demonstrate a workflow using a variety of computational tools to integrate diverse data into interactive genomic and geographic maps. These maps are suitable for analyses of the spread of disease agent genotypes that confer important phenotypes such as drug resistance or the ability to infect certain hosts. The data sets we examine include phylogenetic topologies, substitutions in sequence data, and time and place of isolation of disease agents and host species. We use genomic and geographic maps to study the diversification of lineages of H5N1 influenza over the past decade. Wild aquatic birds (such as the order Anseriformes) have been implicated as the source of influenza viruses isolated from domestic birds (such as the order Galliformes) and mammals (orders Artiodactyla, Carnivora, Cetacea, and Primates; Webster et al., 1992). The current H5N1 outbreak originated in 1996 among anseriforms and spread to humans and chickens in Hong Kong in 1997 (Shortridge, 1999). Between 1997, when there were chicken culls in Hong Kong, and 2002, the most common hosts of H5N1 were anseriform birds and the virus was restricted to China. Since 2003, various lineages of H5N1 have spread throughout South East Asia, Russia, the Middle East, Europe, and Africa, using a wide variety of hosts. Many avian taxa (Charadriiformes, Accipitriformes, Corvidae, Ardeidae, Columbidae, and Passeriformes) as well as primate, carnivore, artiodactyl, and arthropod hosts have been infected with H5N1 (Fig. 1a and aiTrees.kmz at http://www.systematicbiology.org). Direct human infection by avian strains of influenza A is considered rare (reviewed in Lipatov et al., 2004). However, the current outbreak of H5N1 influenza, thought to be strictly of avian origin (Li et al., 2004), has spread geographically and to novel hosts, including humans. Human H5N1 infections have been reported in Hong Kong in 1997–1999, in Vietnam, Cambodia, and Thailand in 2003–2006, in Indonesia in 2005–2007, in Laos and Nigeria in 2007, in China in 2003–2007, and in Azerbaijan, Turkey, Egypt, Iraq, and Djibouti in 2006 (WHO, 2007). As humans have little protective immunity to novel strains, continued transmission of avian strains to human populations and subsequent human-to-human transmission could have a devastating effect on public heath worldwide. One evolutionary pathway for direct avian to human transmission requires mutations of the receptor-binding domains of the hemagglutinin (HA) protein of the influenza virus—especially those that effect changes in viral specificity (from NeuAcα 2,3Gal sialo-oligosaccharides in avian cells to NeuAcα 2,6Gal sialo-oligosaccharides in human cells; Stevens et al., 2006). Another key surface protein of influenza A that plays a role in host range is neuraminidase. Neuraminidase (NA) is a glycoprotein enzyme important to viral replication. The NA protein permits new viruses to escape host cells by removing terminal sialic acid from oligosaccharides bound to the HA protein. As a pair, the receptor binding (HA) and destroying (NA) proteins exhibit concerted evolution that mediates the range of hosts that can be infected (Kobasa et al., 1999). In relation to the recent outbreak of avian influenza, Chen et al., (2006) proposed that specific genotypes in HA and NA are associated with H5N1 from anseriform birds from lakes in central China. Here we used phylogenetic and geographic visualization of large data sets of genomes of H5N1 collected over the last decade to formulate and test specific hypotheses about the movement of key genotypes of H5N1 as it adapts to various hosts. Specifically, we used the concentrated changes test (CCT; Maddison, 1990) to evaluate whether putative key mutations in HA and NA (Stevens, et al., 2006; Chen et al., 2006) and PB2 (Subbarao et al., 1993) were associated with various bird taxa or mammals. Avian influenza is not only complex and multidimensional in terms of biology but also raises several social and political issues. Pandemic influenza would have severe implications for public health, economic security, food safety, and wildlife conservation. For example, in the United States alone it is projected that 15% to 35% of the human population would be affected and the costs could range from $71.6 to $166.5 billion (Gerberding, 2005). Recent papers have pointed out gaps in our knowledge of the patterns of influenza transmission among domestic or wild birds (Olsen et al., 2006; Ducatez et al., 2006). Some advocate that wild birds spread H5N1 over long distances, whereas others contend that the globalization of trade in poultry and wildlife are responsible for the spread of H5N1 (Van Borm et al., 2005; Karesh et al., 2005; Chen et al., 2006; Kilpatrick et al., 2006). The debate among these groups is reflective of the interests of various stakeholders (Normile, 2006a). Rapidly addressing questions about the spread of H5N1 requires that all data are made available as quickly as possible to the global research community (Salzberg et al., 2006). However, there is a great deal of H5N1 sequence data in private hands (Enserink, 2006). Recently, a central database to share avian flu data after publication has been proposed (Bogner et al., 2006; http://www.gisaid.org) but there remain sociological impediments to data sharing. The underlying causes for the privacy of data include the need for researchers to receive credit for data and primary publications and the desire of governments and individuals to protect their economic interests. For example, subsistence duck and chicken farmers and vendors of live birds in markets may be unwilling to report H5N1 outbreaks in their flocks if they are not guaranteed compensation for culled birds (Shortridge, 1999). Thus far, the losses in the poultry sector in South East Asian countries that have had H5N1 epidemics have ranged between 1% and 2% of GDP due to culls and banned exports (Brahmbhatt, 2005). Nongovernmental organizations have their own interests and concerns regarding H5N1 avian influenza. Advocates for the conservation of wildlife fear that wild birds could be treated as primary or “scapegoat” carriers of H5N1 (Blythman, 2006) and subject to eradication programs. Given the biological and social complexity of the problem of avian influenza, it is important that diverse data be synthesized in a form communicable to a wide range of stakeholders. Our aim is to show the utility of visualization when used in with character evolution These can be used to diverse data on pathways of avian influenza spread and these results to a wide To these we phylogenetic of H5N1 genomes with novel visualization and using We geographic and temporal patterns of putative key mutations and host use of H5N1 (Fig. and aiTrees.kmz at http://www.systematicbiology.org). To examine avian influenza we a phylogenetic of the H5N1 from 291 of which were of our to new H5N1 genomes from Vietnam, Europe, the Middle East, and were These new data are the of the and were to after To our of H5N1 genomes from Europe, the Middle East, and Africa, we the new genomes and to a data set of 351 of for the 291 and 351 isolate data sets are provided as and at http://www.systematicbiology.org). A data set of isolates was created to the of et al., of isolates that to the of H5N1 that west of the East Asian–Australian of and acid data was with 2004) A of gaps were in for PB2, and by multiple The of and and gaps were not considered in tree but all and gaps were analyses were using of The tree used for data set of a tree in et al., the costs for and were As it is a of the isolates in at the of the H5N1 and is by a genomic the strain was used as the in the analyses of 291 and 351 isolates. The was used to a tree that of a tree with a sequence by and tree was of were for the data The data set was also to The and tree with are large for the in are provided as data and at http://www.systematicbiology.org). The and phylogenetic used to the data set were to that for the data For the data of were The and tree with for the data set are provided as data and at http://www.systematicbiology.org). In the data we isolates from west of the East Asian–Australian the isolates from China that were the to the isolates from the west to as the and the relevant new of et The of et relevant to the west were by isolates to the analyses and only the isolates that or to the west The tree of the data set of The and tree with for the data set are provided as data and at http://www.systematicbiology.org). For the data for of the a of mutations to branches by the tree was calculated using the in et al., 2005). The tree used in these was a of a tree using the et al., from the of The of mutations to for was into a suitable for with at http://www.systematicbiology.org). The tree and the was also used to character data such as host for the data set or acid genotype of the data The character and were used to branches with various and description used in the at The is can on multiple and is The allows for interactive visualization geographic and temporal for the The in was used to phylogenetic containing H5N1 isolates from the sequence of is for of temporal and spread of H5N1. In can be used in with in such as and geographic or for study of the spread of H5N1 in a variety of for all influenza strain and and host and place of were into a for into the is in the the may or and the point of view of the and terminal or in which to view mutations in description (Fig. and aiTrees.kmz at and of to their or To the of the H5N1 terminal and taxonomic was a set of in and data from for viral and an or we the and for terminal in our Our ability to the of isolation was by the of the For example, a description such as was for the of Hong The or HTUs were and on the geographic of their we calculated the of all the by an to and to that terminal taxa were an of the tree to the (Fig. The of HTUs was on the position from the of the phylogenetic is the HTUs are For example, any with only terminal taxa as is an of to HTUs was an of Thus for any is the of HTUs from the of the tree to the of In our visualizations, we for the the of any We for the which was in the of in the of These of a and were they provided the most visually for our We are the use of to represent data, such as phylogenetic A phylogenetic and are terminal taxa and are at level on the surface of the The or taxonomic and are the and are to their by the branches. is an of it has terminal is to is an of the of at or In at tree we a problem For example, strains had a of level data were this visualization would be and However, the of data provided in a we a to terminal the of the provided by the For we the terminal found the the and for terminal with an was in from is the of with the and is the of to the point of in this created an east to west of all isolates at the (Fig. One could also the position for data for all terminal and were tree branches between were using For in the we a between the the and of point into a were with to represent various character genotype or in the phylogenetic (see and aiTrees.kmz at http://www.systematicbiology.org). were using the of allows for of character data for isolates. We created a of that the to the host from which the strain was We used in description to and acid character as in the (Fig. The description the to with the tree and data to key mutations associated with host or that can be used to genotype viral isolates in an of These also include to public data such as We the of the sequence and the to the movement of the virus over the past decade. are available as data at http://www.systematicbiology.org). We use a test to examine hypotheses by the The concentrated changes test (CCT; Maddison, 1990) the of a character (such as changes to the branches of a as changes in character (such as by a in character are changes in of the on several of the branches of the the CCT significant character the changes in character of share or common the CCT significant 0.05) among the We the significance level for multiple testing to ≤ by by the data We calculated the CCT using and Maddison, with In of we use the to the in the several lineages of H5N1 were carried by anseriform and birds throughout Hong Kong and China (Fig. 1a and aiTrees.kmz). in diverse avian hosts from the taxa Accipitriformes, Corvidae, Ardeidae, and were infected in China. In were infected with H5N1 in China. the outbreak in Hong Kong in 1997, human were next reported by in (WHO, 2006). In and 2004, a of H5N1 and from China. in and arthropod hosts and In South was considered of H5N1. In 2006, there are of a novel outbreak in 2006). is not known at this time if the viruses that the 2006 outbreak in are to the lineages or a novel of H5N1 in among from southern distinct lineages of H5N1 in One in Vietnam, in and a can be found in Vietnam, and In 2004, and anseriforms et al., 2006; et al., 2006). Thus far, the that is in not human or mammalian hosts. However, H5N1 isolates from in of China are of a to the to The that of Asia mammals including humans. The Indonesia and a variety of In 2004, after H5N1 in Asia but in Europe, an isolate of the Thailand strain of H5N1 was carried by from Thailand to (Van Borm et al., 2005). However, this the a in our Google in which a an geographic in provide into the spread of H5N1 and when significant can be used to into the were H5N1 could infect hosts, the a pathway for spread of H5N1 to distinct from the of H5N1 in In several lineages of H5N1 west of east and out of the East Asian–Australian bird These lineages originated in anseriform birds from China and infected birds in China and In and 2006, these and lineages of H5N1 in those and their range to diverse birds and mammals in the Middle East, Africa, and A recent et al., HA and PB2 from viruses isolated from various birds in China in on this data, of these isolates are to isolates at the origin of the that west of east on our we that there is not only movement of this et al., 2006; Webster and 2006) but movement as to China and and aiTrees.kmz at http://www.systematicbiology.org). data from these of the be of use for the and host of this and H5N1 We a strongly supported between the genotype Lys-627 in PB2 and mammalian hosts in the and data sets (Table 1). genotype not in mammals but is of it is associated with increased replication and virulence of the H5N1 virus in laboratory mice (Subbarao et al., 1993; Shinya et al., 2004). In the data set the between Lys-627 in PB2 and anseriform hosts is the ≤ significance level that we have We genotype that is significantly associated with certain host in the acid of the surface proteins and NA) in the or data of HA in acid and which a from avian to human specificity in influenza strains of et al., are at and among the 291 and 351 isolates of H5N1 that we in NA was proposed as a for H5N1 to et al., this genotype is not significantly correlated with any host (Table 1). In genotypes of HA acid and we from the genotype the East Asian–Australian flyway in the and data that a isolates have or at position To the west of East Asian–Australian the HA genotype is with the of a of the tree isolates of H5N1 from from Thailand to in (Van Borm et al., 2005). The of in the west is statistically significant at the CCT ≤ level but is at the CCT ≤ level (Table 1). We found correlation in the or data sets between HA acid and and of anseriform and mammalian or avian host (Table 1). of the surface protein NA is significantly correlated with viruses isolated west of the East Asian–Australian flyway for at the data set ≤ in data set but CCT ≤ in data 1). The correlation of the genotype of NA with isolates west of the East Asian–Australian flyway is nearly significant for the data set but for the data the of a correlation (Fig. the CCT not a correlation between Lys-627 in PB2 and the of isolation or in viruses isolated west of the East Asian–Australian Avian influenza data are and the and can be We used and to examine important visualization provide insight as to which bird taxa spread H5N1 et al., 2006; Borm et al., 2005; and 2006; Kilpatrick et al., we distinct temporal and patterns in key mutations in The mutations we to are thought to be important to infection and replication of H5N1 in various hosts such as or birds 1993; Chen et al., 2006; Stevens et al., 2006). Our results that the avian taxa responsible for the spread of H5N1 are temporally and Recent has a to the of H5N1 spread in the terms of whether or birds or are responsible for movement of the disease (Van Borm et al., 2005; Karesh et al., 2005; Chen et al., 2006; 2006). However, the associated with public genomic data on bird or In the associated with a genomic include trade and from those we can be certain that H5N1 has of the host For example, the of in in is (Van Borm et al., 2005). However, putative trade plays a role in the spread of H5N1 are to and maps of trade are to due to the of the underlying data 2006). Recently, Kilpatrick et al., (2006) have made an to trade data from the and of the United and a of sequence data of the from the HA in order to spread of H5N1. However, the from to the data is at and not the problem of associated with the genomic genomic of receptor and of the movement the of the of various of data to be if we are to about the global of this Our demonstrate that it is possible to reconstruct specific genomic changes in the virus in a but we are by the of the available with as there has been a for public data et al., 2006; et al., and eradication for from diverse and with genomic of include common for hosts, and for viral data on the community or of host and data from trade and of host et al., 2006; Borm et al., 2005). The genotypes for surface proteins and NA) that we are or have distinct geographic and temporal patterns in their These genotypes in surface proteins not statistically or visually to host use by H5N1 as by our character the in the and specificity of viral surface proteins with on various host cells (Stevens, 2006; 2006). In we correlation between Lys-627 in PB2 and mammalian hosts. genotype is not to H5N1 in it also in H5N1 lineages circulating in In of key genotypes for the spread of H5N1 to various hosts on represent a on mutations we be Our study of the geographic of H5N1 mutations and host these in a context and these the data from isolates circulating in the As by et al., 2005; et al., are using However, our ability to and large genomic data sets in and is The spread of disease is of the most important the food and there is significant and in genomic and geographic analyses of these results are complex and to to a wide As Google Earth is any can the provided as to view H5N1 diversification by host or time of We have a workflow that can of and As data on avian influenza and analyses and can be to specific mutations over hosts, and geography. We this is supported or in the and the We and of the of for their in of the and the of the of and the The viral isolates and of the sequence data were by individuals and we for their data the of of key Avian isolates was by the of of and and at the for We and for their in the and