The Emergence of Networks in Human Genome Epidemiology

Large-scale “big science” is advocated as an approach to complex research problems in many scientific areas.1 Epidemiologists have long recognized the value of large collaborative studies to address important questions that are beyond the scope of a study conducted at a single institution.2 We define networks (or, interchangeably, consortia) as groups of scientists from multiple institutions who cooperate in research efforts involving, but not limited to, the conduct, analysis, and synthesis of information from multiple population studies. Networks, by virtue of their greater scope, resources, population size, and opportunities for interdisciplinary collaboration, can address complex scientific questions that a single team alone cannot.3 There is a strong rationale for using networks in human genome epidemiology particularly. Genetic epidemiology benefits from a large-scale population-based approach to identify genes underlying complex common diseases, to assess associations between genetic variants and disease susceptibility, and to examine potential gene–environment interactions.4–6 Because the epidemiologic risk for an individual genetic variant is likely to be small, a large sample size is needed for adequate statistical power.7 Power issues are even more pressing for less common disease outcomes. Replication in different populations and exposure settings is also required to confirm and validate results. The adoption of common guidelines for the conduct, analysis, reporting, and integration of studies across different teams is essential for credible replication. Transparency in acknowledging and incorporating both “positive” and “negative” results is necessary to direct subsequent research. Furthermore, newer and more efficient genotyping technologies must be integrated rapidly into current and planned population studies.8,9 Networks can support studies with sample sizes large enough to achieve “definitive” results, promote spinoff research projects, and yield faster “translation” of results into clinical and public health applications. Networks can also foster interdisciplinary and international collaboration.10 Lastly, networks can assemble databases that are useful for developing and applying new statistical methods for large data sets.11 The experience of established networks provides an important knowledge base on which to develop recommendations for improving future efforts.12 The Human Genome Epidemiology Network (HuGENet) recently launched a global network of consortia working on human genome epidemiology.13 This Network of Investigator Networks aims to create a resource to share information, to offer methodologic support, to generate inclusive overviews of studies conducted in specific fields, and to facilitate rapid confirmation of findings. In October 2005, HuGENet brought together representatives from established and emerging networks to share their experiences at a workshop in Cambridge, U.K.14 In advance of the meeting, a qualitative questionnaire was distributed to workshop participants. The questionnaire elicited information on experiences and practices in building and maintaining consortia. This article reports on the numerous challenges and their possible solutions as identified by the workshop participants (summarized in Table 1) as well as new opportunities offered by the network approach to genetic and genomic epidemiology.TABLE 1: Challenges Faced by Networks of Investigators in Human Genome Epidemiology and Possible SolutionsSCIENTIFIC APPROACH Selection of Scientific Questions To date, most networks have targeted projects originating from preliminary evidence of specific associations or for the purpose of genetic linkage. In most consortia, projects are selected through group discussion and informal or semiformal (eg, voting) prioritization of candidate gene targets. Most networks try to focus on the best possible candidates to generate definitive evidence, but, given the large proportion of false-positives in genetic epidemiology,15 there is considerable uncertainty about the criteria for selecting such targets. Possible criteria include the number and consistency of published reports for a specific gene, the presence of a high-profile controversy in the literature, strong a priori biologic plausibility, potentially high population-attributable risk (eg, a common polymorphism) supporting linkage evidence from genomewide data, and candidates derived from genomewide association screens.16,17 Networks are often focused on candidate genes involved in pathogenesis of the disease outcome or in biologic pathways involving environmental exposures such as metabolism of carcinogens.18 For example, the WECARE consortium on genetics of cancer and radiation exposure19 has addressed individual genes that lie within pathways related to double-strand breaks caused by radiation damage. Consortia are increasingly used to replicate findings from hypothesis-free genomewide approaches. For example, consortia are attempting to replicate findings from 2-stage genomewide association studies of Parkinson disease20 and breast cancer.21 With decreasing genotyping cost and the expressed interest of funding agencies in genomewide association studies,22 some consortia are coordinating large-scale genotyping and replication of whole genome association designs.23 Prospective and Retrospective Components Networks use information and biologic specimens from ongoing or established cohort and case–control studies with data on phenotypes. Phenotype information may have been accumulated either retrospectively or prospectively depending on the study design. Participating teams with prospective designs usually continue collecting phenotype information. Regarding genotyping, several consortia perform meta-analyses of individual-level data using studies in which all genotyping has already been done and data have been published. Some consortia include additional genotyping from teams that have not yet done or published such genotyping; for other consortia, prospective genotyping represents the majority of the data. Increasingly, prospective genotyping is coordinated to test novel candidate gene variants or variants identified by genomewide approaches. Handling of Information From Nonparticipating Teams Many networks do not encompass all teams working on the disease or subject matter of interest. For some common diseases (eg, breast cancer), there are 2 or more organized multiteam consortia in addition to nonorganized teams.24–26 Some consortia attempt analyses that include outside data to examine the robustness of their findings. Integration of evidence across networks and across participating and nonparticipating teams remains a challenge in developing all-encompassing synopses of the evidence on specific gene–disease associations.27 LAUNCHING A NETWORK Network Characteristics Consortia in the Network of Investigator Networks are comprised of between 5 and 521 teams. Subject numbers range from 3,000 to over half a million. Elements deemed essential for launching a network are a strong scientific rationale, the agreement of all teams to work together and combine data on overarching research questions, and the ability to support initial communication, coordination, identification, and recruitment of partners. True integration of disciplines can be challenging because different disciplines are typically housed in discrete departments and have different scientific cultures. Interdisciplinary training is important for bridging these gaps. Established networks have coalesced through different processes. Frequently, the initiation of a network includes the gathering of information on available resources from several groups of investigators actively involved in research in the same field. Dissemination of information on integrated research aims, resources, and possible contributors ultimately leads to the identification of specific projects to be pursued. This process creates a forum for scientific exchange and more targeted collaborations.28 Networks tend to expand their membership over time and loss of partner teams is uncommon.29,30 Although network membership tends to be inclusive, there is concern that inclusion of flawed data jeopardizes the validity of the collaborative results. For this reason, some consortia have eligibility criteria based on appropriateness of study design and phenotypic accuracy. Organization and Coordinating Centers Networks use different models of steering and coordination. Working groups focused on specific topics are common within the largest networks. For example, the International Head and Neck Cancer Epidemiology (INHANCE) network32 requires all members to participate in at least one of 7 working groups that focus on scientific issues or projects such as age at cancer onset, nonsmokers and nondrinkers, tobacco and alcohol, genetics and DNA repair, human papilloma virus prognosis and survival, and occupational factors. The Genetics of Melanoma (GenoMEL) network33 has a Steering Committee, a Scientific Advisory Board, a Patient Advocacy Group and an Ethics Committee as well as several topic-specific working groups. Some networks have separate statistical, genetic, and clinical coordinating centers, whereas others centralize these functions. A primary coordinator or chair and a small steering group are usually essential for the network to operate efficiently. Sometimes it is difficult to trace in detail what happens at the local level of participating sites. Minimizing and streamlining administration to maximize the conduct of science is essential. Funding Funding sources include governmental and public health agencies as well as private foundations. Funding from for-profit companies and full partnership with industry-sponsored teams has been rare, although some consortia have partnered with private companies for specific projects. For example, the Colon Cancer Family Registry worked with specific companies to perform a systematic mutational analysis of the participants enrolled.34 Funding, especially for infrastructure, is a key limiting factor. Difficulties also exist occasionally for obtaining funding to support activities beyond the originally proposed specific projects despite demonstrated productivity of the network. Some consortia have a single source for primary funding (typically National Institutes of Health or European Commission grants), but most networks have diverse, sometime project-specific, sources of funding. For example, the Birth Cohorts Consortium had a total of 64 funders over the last 8 years. In some countries, participation in a consortium can constitute a strong leverage to obtain national funds. STANDARDIZATION WITHIN THE NETWORK Data Management Efficient and accurate data management is very important because poor-quality data from one or more teams may undermine an otherwise excellent collaboration. Data typically flow to one coordinating center, but some consortia have multiple data coordinating centers with complementary functions. Networks use various data quality assurance practices and checks for logical errors and inconsistencies. Networks that have invested heavily in quality assurance believe that the effort was worthwhile, because errors may occur even under the best circumstances.35 Logical errors (inconsistencies in the contributed data) are usually easy to identify and readily solved through communication with the team investigators. Examples include out-of-range values, inversion of coding of phenotypes, improper or inconsistent allele calling, and inconsistent crosscoding in databases. Logical errors may reveal deeper problems with contributed data. Queries regarding missing data may yield additional information with some additional effort from the team. Some consortia have instituted in-person training for collecting genotype and phenotype data in addition to ongoing quality control checks. Some networks have developed and published explicit policies of quality assurance for phenotype or genotype data.25 Standardization or Harmonization of Phenotypes and Other Measurements Data standardization is best implemented at the beginning of a “de novo” collaborative study, when tools for data collection and definition of data items are developed. Data standardization achieves agreement on common data definitions to which all data layers must conform. Each data item is given a common name, definition, and value set or format. When standardization is not possible (eg, different questionnaires or criteria have been used historically by different teams), harmonization of data items is suggested—and sometime required by the funding agencies. Data harmonization is useful when data sets are already collected from originally independent studies focusing on similar questions or field of inquiry. The harmonization process seeks to maximize the comparability of data from 2 or more information systems with the goal of reducing data redundancy and inconsistencies as well as improving the quality and format of data. Standardization or harmonization is crucial for a network to perform better than single studies, and these processes increase the credibility of the derived evidence. Phenotypes and other nongenetic measurements may be difficult to standardize across teams. For example, Parkinson disease has several sets of accepted diagnostic criteria and teams may use different criteria that have high concordance. It is often challenging to reassess phenotype using alternative criteria. In some diseases, there may be no consensus regarding the most important phenotypes to study. For example, 21 pharmacogenetic studies in asthma analyzed 483 different end points.36 Conversely, the assembled data of some networks have been used to define subphenotypes of disease that would not have been evident with lower statistical power.37 Networks may help achieve harmonization, even when single-team studies have been inconsistent in preferred definitions and outcomes. For example, in the HIV consortium, access to primary data allowed for harmonized definitions of seroconverter and seroprevalent subjects and for the outcome (clinical AIDS),31 although these variables had been defined inconsistently by the teams. In contrast, the InterLymph consortium standardizes the diagnosis of lymphoma subtypes through a coordinated review of a subset of slides from each numbered study.38 One criterion of the importance and success of a network may be its ability to adopt standards for phenotypes and covariates to prevent the use of inconsistent definitions in subsequent studies. In some networks, phenotypes are assessed in prospectively ascertained cases or through an extensive reexamination of phenotypes of existing cases. Consortia also use training sessions on phenotyping, photographs (eg, for moles in melanoma family members), and central review to enhance consistency of data. Standardization of Genotypes Most networks have not performed central genotyping of all samples, but exceptions exist.32,39 Shipping specimens is sometimes challenging in collaborations among geographically dispersed teams and regulatory considerations may also prohibit centralized genotyping. For example, some teams are prohibited from shipping specimens by their protocol, local legislation, or their funding agency. Several networks use a semicentralized approach in which some teams ship their samples to a central laboratory, whereas others perform onsite genotyping. Quality control of genotype results is usually straightforward, but additional checks are required in a multiteam collaboration. Some networks use published genotype data without quality checks beyond what each individual team implemented in their laboratory (eg, repeat genotyping of a random sample of specimens). In the absence of centralized quality control, consortia must depend on post hoc analyses such as deviation from Hardy-Weinberg equilibrium proportions in the controls,40 to identify possible genotyping (or other) errors. Large between-study heterogeneity in the final analyses may also reflect measurement errors. However, sizeable errors may still be missed with these methods. Several networks, including the Public Population Project in Genomics (P3G), check genotype results through exchange of blinded samples between groups. Another approach is to ship samples of known (ideally sequence-verified) genotypes to all participating laboratories. Alternatively, a sample of specimens that were genotyped locally may be shipped to a central laboratory for confirmation. Experience suggests that the reliability of each laboratory should not be taken for granted. Serious errors have occurred (eg, inverse reporting of genotype results that produces an inverse association) that could only be detected by rigorous quality control mechanisms. Error may be considerable even for single and can depend on a and This is because most gene–disease associations have sizes that could be by small laboratory errors. and Networks use and to with an for communication promote and information on the and (eg, access for more information, which is essential to communication within and between teams as well as for private scientific with Some networks have developed as of data from multiple groups and their data management is such as the on DNA and cancer and review and policies are best established in the of a network to For each a team is essential for developing an initial and incorporating from Most consortia use which in a long of The is typically the of the specific Some networks use and separate of additional contributors and separate Group may also be but errors in in and the may may also be an in consortia. A agreement involving all should be at the and on are for of investigators. In the long networks likely for by interdisciplinary and opportunities for developing scientific but in emerging consortia, more investigators tend to and the and funding. Some consortia have developed explicit policies of opportunities for investigators. in funding and are needed to support consortia as a for both the rapid of scientific knowledge and the of new independent to Data and of Data data and resources should be to the scientific and networks should develop policies that support this Standardization of policies is needed and could be by and policies by funding It is important that both “positive” and “negative” results be to their very networks may be the last of reporting and and should to identify and include but data. Interdisciplinary science requires interdisciplinary of scientists and of initial review groups with interdisciplinary are to the of consortia Interdisciplinary research teams time to assemble and funding may be especially to for emerging consortia. for of productivity by funding agencies should into the and time to the necessary Networks to address emerging scientific should data and support of research conducted by multiple investigators at different institutions in different Examples of to be in such have been published and by some existing However, the of at different institutions in the of these and the heterogeneity of at the and international level may data and in large The participants identified a number of additional For example, criteria are challenging and should be quality teams should be to their research and their results may be by the consortium at “negative” results should be in an access to of effort and in the field. may also reflect the of multiple networks in the same field with similar or very different of membership may and maximize and replication of results. study has been for clinical studies and already has or more in its of genomewide association studies is planned by the National Institutes of Health as a to and reporting maximize and data access rapidly advance and maximize funding and integration of genomic and other technologies is a This may the adoption of centralized which may be by such as the should into the that laboratory are rapidly and to on a large-scale The and standardization across teams of biologic (or are a The goal is to maximize through various such as whole genome or as deemed Many of the challenges networks, may yield as in Table for Networks to to in Human Genome The HuGENet Network of Investigators Networks seeks to an forum for communication and of in statistical and laboratory and among consortia. Consortia are to create a that would include information on their participating teams and on the of their studies and This knowledge base would in studies and for faster replication of results Another HuGENet Network of Investigator Networks effort aims at developing an of genomic maintaining information on results from ongoing studies. of evidence are for several diseases, with various that would be and enough to the of a rapidly developing interdisciplinary science” human genome epidemiology is to funding and scientific must and that support team science individual This a which is already from a research of to one of team From the of Cancer and Population National Cancer the of Genomics and Centers for and the of Cancer Epidemiology and National Cancer the Human Genome National Institutes of the in Human Genome of Epidemiology and of the of Cambridge, Cambridge, the Health Genetics Cambridge, the of Epidemiology and Cancer the for on the of of Cancer the for Human of and of Public of of and the of the of Epidemiology and Population of and the of Public Health and of Cambridge, Cambridge, the of of of the Health the of The of Public the of Epidemiology and Public the of Public Health and of the of the of the Epidemiology the of the of and Epidemiology the the the Epidemiology Cambridge, the and Epidemiology of and of of and for and and the of of

Récupéré en direct depuis OpenAlex et désinversé. Les résumés ne sont pas conservés dans cette base de données : les index inversés représentent 8,6 Go des 9,3 Go de texte de la base, et le serveur dispose de 13 Go libres.