Non-sharing of data: 1997–2005

Avian influenza (H5N1) has been a concern since 1997, when it began to infect humans in Hong Kong. H5N1 has spread from its origins in China to other regions in Asia, Russia, India, Pakistan, the Middle East, North and West Africa, and Europe (Chen et al., Reference Chen, Smith, Li, Wang and Fan2006; Janies et al., Reference Janies, Hill, Guralnick, Habib, Waltari and Wheeler2007; Hovmöller et al., Reference Hovmöller, Alexandrov, Hardman and Janies2010). Since 2003 there have been 489 human cases and 289 deaths attributed to H5N1 (WHO, 2009a).

The research environment has changed dramatically since 1997, when an outbreak of avian influenza (H5N1) among poultry and humans in Hong Kong alerted health officials to its dangers. In the early 2000s, genomic sequencing became highly mechanized and computational power began to improve dramatically via Beowulf clustering (Gee, Reference Gee2000). However, in the same period, genetic sequence data on avian influenza was not widely shared. Some data could be found in GenBank (http://ncbi.nlm.nih.gov), which is public domain. In parallel, a small group of influenza researchers (Nature, 2006) had access to a private database maintained on behalf of the World Health Organization (WHO) at Los Alamos National Laboratory (LANL) (http://flu.lanl.gov). The LANL database included data from GenBank, as well as private data submitted by WHO directly to LANL (Macken et al., Reference Macken, Lu, Goodman, Boykin, Osterhaus, Cox and Hampson2001). Furthermore, common practice by most biomedical researchers, including the influenza community, was to share data only after publication. In contrast, at the turn of the millennium, large publicly funded projects such as the Human Genome Project were charged with posting data rapidly after collection (Nature, 2006).

In 2007, the International Health Regulations were revised, thus requiring nations to assess and report disease outbreaks to the WHO within 48 h (Wilson et al., Reference Wilson, Tigerstrom and McDougall2008). Previously, governments were not willing to share information about diseases within their borders, as demonstrated in the severe acute respiratory syndrome (SARS) epidemic (FlorCruz, Reference FlorCruz2003). Each person and family affected by SARS represents a personal tragedy. At the level of international trade, perceptions of public health in a country can lead to economic tragedy. Estimates for the cost of SARS, in terms of lost productivity, are US $18 billion for East and Southeast Asian economies (ADB, 2003). If there had been ongoing public concern, the cost of SARS in Asia could have been US $60 billion or higher (ADB, 2003). SARS cost the Canadian economy an estimated CDN $519 million in lost revenues in 2003 (CBC, 2003). These losses are very high compared to the actual scale of the epidemic, representing over US $2 million per person infected in 2002–2003.

Along with influenza, the SARS epidemic represented an important event requiring changes to public health policy. SARS made it clear that novel infectious diseases can start in one country and quickly spread throughout the whole world. In light of SARS and influenza, many in public health informatics began to realize that data sharing and cooperation across the globe are vitally important to prepare for and respond to emergent pathogens (Nature, 2006). The CDC predicted in the United States alone that an H5N1 pandemic among humans would affect 15–35% of the population, resulting in $70–167 billion in lost productivity (Gerberding, Reference Gerberding2005).

The severity of the H1N1 pandemic that began in early 2009 appears to be much lower than projected (Presanis et al., Reference Presanis, De Angelis, Hagy, Reed, Riley, Cooper, Finelli, Biedrzycki and Lipsitch2009). As of 19 March 2010, pandemic H1N1 has caused at least 16,813 deaths worldwide (WHO, 2010). These deaths are heartbreaking for the families involved. Furthermore, H1N1 caused fear and absenteeism in some regions of the world such as Mexico (Stevenson, Reference Stevenson2009), Ukraine (Pan, Reference Pan2009), and Argentina (Valente, Reference Valente2009).

The economic aspects of the current H1N1 pandemic are still being determined. In the United States, any economic impacts of H1N1 will have been overshadowed by the trillions of dollars of costs due to the banking crisis and recession that began in 2007 and has continued into 2010 (Barr, Reference Barr2009).

There are arguments, at both the international and individual levels, that data on infectious diseases should not be shared as some might profit from the data collected by others. For example, leaders in Indonesia have argued that isolates of avian influenza (H5N1) isolated in their country represent biological patrimony (Garrett and Fidler, Reference Garrett and Fidler2007). As such, Indonesia asked for a guarantee that any vaccine developed based on influenza isolates from their country should be in turn shared with their citizens (Wilson et al., Reference Wilson, Tigerstrom and McDougall2008). In a separate example, China did not share data because Western researchers did not give credit to Chinese researchers who collected the primary data (Zamiska, Reference Zamiska2006).

Data sharing: 2006–2009

The tradition in research has been that data are private until published (Brown, Reference Brown2006). However, in light of the public health concerns many have realized that this practice inhibits innovation (Bogner et al., Reference Bogner, Capua, Lipman, Cox and Lipmans2006). Data sharing is imperative for progress. When data are shared, researchers across many different disciplines can attack a problem. If data access is restricted to individuals well versed in established technologies and schools of thought, then the next generation of thinkers is limited in its ability to bring new technologies and approaches to the problem. Pandemic influenza presents challenges not easily met by molecular or cellular biology or epidemiology alone. Molecular and cellular biologists are concerned with the interactions of the host's and pathogens' genome(s). Epidemiologists are concerned with the occurrence of diseases and symptoms. Epidemiologists rarely consult genomic data, but rather count classes of symptoms such as ‘influenza-like illnesses’. Neither discipline has been able to develop an understanding of an outbreak of disease from both genomic and epidemiological points of view. Collaboration is necessary for a bridge between molecular biology and epidemiology to be built. For example, virologists, molecular biologists, evolutionary biologists, geographers, and computer scientists have collaborated to combine genomic sequences with geographic and temporal information. These collaborations have provided maps of the movement of pathogens and key mutations among pathogens, such as those that confer resistance to drugs or the ability to jump from animals to humans (Janies et al., Reference Janies, Hill, Guralnick, Habib, Waltari and Wheeler2007, Reference Janies, Habib, Alexandrov, Hill and Pol2008, Reference Janies, Treseder, Alexandrov, Habib, Chen, Ferreira, Çatalyürek, Varón and Wheeler2010a, Reference Janies, Voronkin, Studer, Hardman, Alexandrov, Treseder and Valsonb; Hill et al., Reference Hill, Guralnick, Wilson, Habib and Janies2009; Hovmöller et al., Reference Hovmöller, Alexandrov, Hardman and Janies2010). The next step is to turn genomic data and maps into forecasts of the spread of specific pathogens or mutations. We have taken a first step in this direction by using maps, phylogenetics, and studies of selective pressure for resistance to oseltamivir (Tamiflu®) on seasonal H1N1 lineages to develop early and spatially specific warning of the rise of resistance to oseltamivir in pandemic H1N1 lineages (Janies et al., Reference Janies, Voronkin, Studer, Hardman, Alexandrov, Treseder and Valson2010b). Further incorporation of genomic and geographic data into pathogen surveillance and public health intelligence will require data sharing as well as interaction among molecular and evolutionary biologists, mathematicians, statisticians, geographers, and public health scientists.

One bright spot in the area of data sharing has been the Influenza Genome Sequencing Project (IGSP) (http://www3.niaid.nih.gov/LabsAndResources/resources/gsc/Influenza). Since its origins in 2005, the IGSP has sequenced and released over 4650 influenza genomes into the public domain, pre-publication. The IGSP is under contract from the National Institute of Allergy and Infectious Diseases (NIAID) of the National Institutes of Health (NIH) of the United States to sequence genomes of viral isolates, collected by over 40 contributing laboratories around the world, and to rapidly put them in the public domain, via a special section of GenBank called the Influenza Virus Resource (Bao et al., Reference Bao, Bolotov, Dernovoy, Kiryutin, Zaslavsky, Tatusova, Ostell and Lipman2008). This practice is good for the research community at large but has hindered the IGSP from collaborating with those who prefer to maintain right to publish before data are released. In contrast, other groups have been successful in this model of data release and rapid publication. For example, Salzberg et al. (Reference Salzberg, Kingsford, Cattoli, Spiro, Janies, Aly, Brown, Couacy-Hymann, De Mia, Dung, Guercio, Joannis, Ali, Osmani, Padalino, Saad, Savić, Sengamalay, Yingst, Zaborsky, Zorman-Rojs, Ghedin and Capua2007) released data first and then published as an international team. Authorship credit was shared widely among the team who played various roles such as: providing the isolates, sequencing the isolates, analyzing the data, and writing the paper.

The H1N1 pandemic that began in 2009 has brought several challenges and has evoked renewed responses by the research community for data sharing. Foremost among the responses is the Global Initiative for Sharing of All Influenza Data (GISAID). Due to the efforts of GISAID, the Centers for Disease Control (CDC) in the United States, and the WHO, sequence data for the 2009 pandemic strain of H1N1 were released swiftly into a publicly accessible database on 25 April 2009 (the EpiFluDB™ database section of gisaid.org), pre-publication (WHO, 2009b). Large amounts of data also flowed rapidly into GenBank as of 27 April 2009 (e.g. http://www.ncbi.nlm.nih.gov/genomes/FLU/SwineFlu2009.html).

Studies on these early sequences were rapidly produced to examine the host and origins of pandemic H1N1 (Garten et al., Reference Garten, Davis, Russell, Shu, Lindstrom, Balish, Sessions, Xu, Skepner, Deyde, Okomo-Adhiambo, Gubareva, Barnes, Smith, Emery, Hillman, Rivailler, Smagala, de Graaf, Burke, Fouchier, Pappas, Alpuche-Aranda, López-Gatell, Olivera, López, Myers, Faix, Blair, Yu, Keene, Dotson, Boxrud, Sambol, Abid, St. George, Bannerman, Moore, Stringer, Blevins, Demmler-Harrison, Ginsberg, Kriner, Waterman, Smole, Guevara, Belongia, Clark, Beatrice, Donis, Katz, Finelli, Bridges, Shaw, Jernigan, Uyeki, Smith, Klimov and Cox2009; Novel Swine-Origin Influenza A H1N1 Virus Investigation Team, 2009). The actions of GISAID/CDC/WHO and the IGSP are watershed events for infectious disease genomics and public health that hopefully have a lasting impact on attitudes for data sharing.

As of late April 2009, no longer did the world have to wait for any particular scientist or organization to decide that their H1N1 data were ready to be shared. Public health organizations around the world followed the lead of GISAID/CDC/WHO and IGSP and quickly began to submit pre-publication sequence data to GISAID and GenBank. Currently, GISAID has the largest collection of sequence data for the H1N1 pandemic lineage, and there is little to no unique data in Genbank. However, at the same time, sequence data from other strains and the IGSP have continued to be submitted to NIH's GenBank. These multiple data sources can create a challenge for the researcher who is concerned with comprehensive sampling of influenza data. For example, in a phylogenetic analysis, the sampling of the sequence diversity is very important to increase the accuracy of the study (Zwickl and Hillis, Reference Zwickl and Hillis2002). In our laboratory, we found it useful to check various databases and create a set of non-overlapping sequences. Our solution is presented in the Methods section below. Recently, GISAID started to mirror sequence data from GenBank and other repositories. This mirroring has abated our concerns of not having a complete dataset. However, the speed of mirroring is very important. For those public health scientists charged with disease surveillance, having timely access to sequence data is critical. Despite all the progress in data sharing, GISAID's independence was temporarily in peril in the summer of 2009, requiring the Max-Planck-Institute for Informatics (MPI) in Saarbrücken, Germany, to step in as the new host for the EpiFlu™ database, following a challenge over its control by the Swiss Institute of Bioinformatics (SIB) (Butler, Reference Butler2009a, Reference Butlerb). In collaboration with GISAID's user community, WHO Collaborating Centers for Influenza, and veterinary reference laboratories of the Food and Agriculture Organization (FAO) of the United Nations and the World Organization for Animal Health (OIE), MPI promptly developed and offered users a replacement to SIB software. Subsequently, in October 2009, the German Government announced it had entered into an agreement with GISAID to become the long-term host of its EpiFlu™ database in Bonn, from 2011 onwards (Federal Ministry of Food, Agriculture and Consumer Protection, Germany, 2009). The agreement also calls for the curation of GISAID data by the Friedrich-Löffler-Institute, the national research center for animal health of Germany, to improve the overall quality of data offered to GISAID users.

Other issues adding complexity for researchers are the various data access agreements of GISAID and GenBank. GenBank, the DNA Data Bank of Japan (DDBJ), and the European Molecular Biology Laboratory (EMBL) form the International Nucleotide Sequence Database Collaboration (INSD) (http://www.insdc.org). Highlights of the International Nucleotide Sequence Database Collaboration Policy (Brunak et al., Reference Brunak, Danchin, Hattori, Nakamura, Shinozaki, Matise and Preuss2002) include: ‘The INSD has a uniform policy of free and unrestricted access to all of the data records their databases contain’ and ‘The INSD will not attach statements to records that restrict access to the data, limit the use of the information in these records, or prohibit certain types of publications based on these records. Specifically, no use restrictions or licensing requirements will be included in any sequence data records, and no restrictions or licensing fees will be placed on the redistribution or use of the database by any party.’

In contrast, a GISAID user must sign an agreement to become an authorized user (GISAID, 2008). A passage of this agreement states: ‘Distribution of Data. You agree that Data may be provided to all GISAID Data users and providers that have agreed to be bound by this Agreement (or an Agreement with substantially similar terms) and continue to abide by its terms (collectively “Authorized Users”). However, subject to applicable law, You agree not to distribute Data to any third party other than Authorized Users, except for Data that has been expressly and lawfully placed in the public domain by GISAID in accordance with subparagraph (a), by You, or by any other party having the right to do so.’

The key distinction is that GenBank data can be repurposed and distributed by anyone at will. GISAID data can only be repurposed and distributed to GISAID users who sign the agreement. This distinction on how data can be repurposed has important motivations and implications. GISAID intends the user agreement to be a vehicle to enable data sharing by allowing access to those users who agree to collaborate with and acknowledge data providers. Acknowledgement of data providers is something that is not often done in the context of use of data from GenBank, other than citation of the accession numbers for the sequences used. Irrespective of the database from which a sequence was drawn, the basic scholarship and public policy (treated below) that calls for citation of papers can easily be extended to the acknowledgment of the originating laboratories and scientists that provided the isolates, sequence data, and metadata. This acknowledgement can be accomplished via a table of supplemental data (e.g. http://www.ij-healthgeographics.com/qc/content/9/1/13/suppl/S9) or co-authorship (e.g. Salzberg et al., Reference Salzberg, Kingsford, Cattoli, Spiro, Janies, Aly, Brown, Couacy-Hymann, De Mia, Dung, Guercio, Joannis, Ali, Osmani, Padalino, Saad, Savić, Sengamalay, Yingst, Zaborsky, Zorman-Rojs, Ghedin and Capua2007). Common practice among scientists who publish results based on sequence data, such as posting a multiple sequence alignment containing GISAID data, can be done in the ‘wiki’ section, which is in the password protected section of GISAID.org, without violating the agreement. In contrast, an alignment of GenBank data can be posted openly (e.g. http://supramap.osu.edu/sm/supramap/publications). The practice of making raw data and research results that contain raw data public after publication is important as it allows any scientist to repeat the experiments. For NIH-funded investigators in the United States, the practices of timely sharing of data and acknowledgement of sources are codified (NIH, 2003). Furthermore, timely public dissemination (within a year) of papers based on NIH-funded research has recently been made a law in the United States (NIH, 2008). Similar policies have been enacted or considered around the world (Alliance for Taxpayer Access, 2006).

All of the analytical applications on which we have published in the peer-reviewed literature (Janies et al., Reference Janies, Treseder, Alexandrov, Habib, Chen, Ferreira, Çatalyürek, Varón and Wheeler2010a, http://supramap.osu.edu; Hovmöller et al., Reference Hovmöller, Alexandrov, Hardman and Janies2010, http://routemap.osu.edu; and Janies et al., Reference Janies, Voronkin, Studer, Hardman, Alexandrov, Treseder and Valson2010b, http://pointmap.osu.edu) are available freely over the web. These applications do not provide data. They are analytical and visualization tools to which users must supply datasets. However, to some users, data security is very important and a web-based application is unsuitable. In these cases, users can compile and run a binary of POY (Varón et al., Reference Varón, Vinh and Wheeler2009) enabled for Supramap analysis on computers in a local secure environment without the use of a remote cluster and without transmission of their data outside of their organization. We include instructions for the stand-alone application here: http://supramap.osu.edu/sm/supramap/tutorials#section2.

Another area for improvement in influenza informatics is sharing of metadata such as host, place of isolation, date of isolation, and clinical information (Janies et al., Reference Janies, Hill, Guralnick, Habib, Waltari and Wheeler2007; Butler, Reference Butler2008). Metadata standards have improved in both GISAID and GenBank. Binomial names are now being used for many host species (e.g. Anas crecca, Anas platyrhynchos, Anas boschas, Anas clypeata, and Anas acuta). Significant improvements in GISAID include important fields for metadata of animal hosts ‘domestic (or wild) status’, ‘health status’, ‘vaccination status’, and ‘specimen source (type of tissue or swab)’. Nevertheless, there is still much room for improvement in the data that providers submit. For example, in a recent study, metadata associated with NA sequences for H1N1 data contained many common names for host species (e.g. ‘duck’, ‘goose’, ‘swan’, and ‘chicken’) as well as some unhelpful names for hosts (e.g. ‘other avian’ and ‘other mammal’).

There are also other databases that specialize in phenotype prediction and vaccine design. The Influenza Research Database (fludb.org) includes drug resistance and virulence genotypes, information on epitope variation important for vaccine design, clinical metadata, and surveillance information. GISAID provides discrete metadata required by the collaborating centers of the WHO that are part of the vaccine strain selection process. This has permitted the raw data of candidate strains to be available to GISAID users, but their candidacy for a vaccine is unknown, except to the collaborating centers.

Beyond data sharing: 2010 to the future

Large-scale sequencing and rapid global sharing of influenza genomes have been the breakthrough events of the past decade. However, raw sequence data cannot, on their own, provide the information needed by public health officials. Public health scientists must integrate knowledge of the genomes of pathogens with host biology as well as societal and environmental factors to understand the etiology epidemics and to anticipate their trajectories. One of the most effective ways to integrate diverse information is to put the data together in a geographic information system (GIS) (Fig. 1). A GIS provides an interdisciplinary framework for hypothesis generation and testing and can be used to communicate results. To these ends, we have developed several mapping applications to integrate genomic data with geographic and phenotypic information in interactive visual environments that allow public health scientists and policy makers to determine: (i) the number and type of distinct pathogen strains that circulate in a region (Janies et al., Reference Janies, Hill, Guralnick, Habib, Waltari and Wheeler2007); (ii) from which geographic sources and hosts, the pathogens originate (Fig. 1) (Janies et al., Reference Janies, Hill, Guralnick, Habib, Waltari and Wheeler2007, Reference Janies, Habib, Alexandrov, Hill and Pol2008; Hovmöller et al., Reference Hovmöller, Alexandrov, Hardman and Janies2010; http://routemap.osu.edu); and (iii) whether these strains are becoming resistant to drugs (Hill et al., Reference Hill, Guralnick, Wilson, Habib and Janies2009; Janies et al., Reference Janies, Voronkin, Studer, Hardman, Alexandrov, Treseder and Valson2010b; http://pointmap.osu.edu) or adapting from avian to mammalian hosts (Janies et al., Reference Janies, Hill, Guralnick, Habib, Waltari and Wheeler2007, Reference Janies, Treseder, Alexandrov, Habib, Chen, Ferreira, Çatalyürek, Varón and Wheeler2010a; http://supramap.osu.edu). The major advantage of genetic and phenotypic maps over syndromic approaches (Brownstein et al., Reference Brownstein, Freifeld, Reis and Mandl2008) is that genetic and phenotypic maps allow public health officials to make decisions based on the biology of specific pathogens and hosts, rather than the occurrence of symptoms, which can arise from unrelated diseases.

Fig. 1.

Screen capture of an interactive phylogenetic map of the emergence in early 2009 of the pandemic lineage of H1N1 influenza. The tree is based on nucleotide sequences for the neuraminidase segment. The colors of branches of the tree indicate the sample (animal host or environment) from which the virus was isolated. Mutations at each node of the tree can be viewed in pop-up windows. This phylogenetic map was created with Supramap software (Janies et al., Reference Janies, Treseder, Alexandrov, Habib, Chen, Ferreira, Çatalyürek, Varón and Wheeler2010a) and is available for download as a keyhole markup file (KML) compatible with Google Earth™ at http://supramap.osu.edu/sm/supramap/publications.