Evaluation of methods for meta-analysis of genetic linkage studies for complex diseases and application to genome scans for asthma and adult height
Linkage genome scans for genetically complex diseases have low power with the sample sizes that were often used in the past, and hence meta-analysis of several scans for the same disease might be a promising approach. Appropriate data are now becoming accessible as many groups worldwide investigate...
|PDF Full Text
No Tags, Be the first to tag this record!
|Linkage genome scans for genetically complex diseases have low power with the sample sizes that were often used in the past, and hence meta-analysis of several scans for the same disease might be a promising approach. Appropriate data are now becoming accessible as many groups worldwide investigate common diseases. The aim of this thesis is to extend and evaluate statistical methodology for meta-analysis. In addition, two meta-analyses of linkage genome scans for the complex phenotypes asthma and adult stature are performed and discussed.
In the first part of this thesis, an overview of available statistical methods and current applications is given. A new meta-analysis method is introduced which is based on a weighted combination of non-parametric linkage scores. Its relationship to traditional fixed effects meta-analysis of combining parameter estimates from different studies weighted by the inverse of their respective variances is described. Recombination and low informativity of markers lead to a reduction of the effective sample size in multipoint linkage analysis. A locus specific weighting of individual studies with this effective sample size is therefore proposed. In a simulation study, the power of different methods to combine multipoint linkage scores, namely Fisher’s p-value combination (Fisher 1932), the truncated product method (Zaykin et al. 2002, a variant of Fisher's method), the Genome Search Meta-Analysis (GSMA, Wise et al. 1999) method and the proposed weighting methods were compared. In particular, the effects of different genetic marker sets and sample sizes between genome scans were investigated. The weighting methods explicitly take those differences into account and have higher power in the simulated scenarios than the other methods.
The proposed meta-analysis method was applied to four linkage genome scans for the phenotype asthma and five studies of a candidate genetic region. Multipoint nonparametric linkage analysis is performed and different weighting schemes are used to combine the score statistics of individual studies to an overall statistic. For comparison, the GSMA method is also applied to the same data sets. For meta-analysis of linkage studies, a common map of genetic markers is necessary to align results obtained in different studies with different markers. In this meta-analysis, the effects of map uncertainties were evaluated. The latest versions of available combined physical and linkage maps are very precise and the small potential map errors that are left do not have relevant impact. This meta-analysis of nine asthma linkage studies does not identify significant regions of genetic linkage to asthma. A still rather small size of the combined samples may be the reason for low power to identify susceptibility genes for the complex trait asthma.
The statistical methods that can be applied for a meta-analysis of linkage studies depend crucially on the available data, especially any additional information besides the usually reported linkage statistics. For the meta-analysis of linkage genome scans for the highly heritable trait adult height, only LOD scores from variance components linkage analysis, which are measures of significance and not effect estimates, could be obtained. Thus, Fisher’s method and a weighted and unweighted variant of the inverse normal method were applied. Initially, a linkage genome scan for this quantitative trait was performed in the extended pedigrees of the Framingham Heart Study. A variance components linkage analysis in this sample unselected for height gave evidence for linkage in several regions. All markers showing a LOD score greater than 1 in this analysis correspond to previously reported linkage regions, including chromosome 6q with a maximum LOD score of 2.45 and chromosomes 9, 12, 14, 18 and 22. Following this observation, a meta-analysis of all previously published genome scans for adult stature was planned. Genome scan results of 17 separate samples reported in seven publications and comprising more than 14000 phenotyped and genotyped individuals could be obtained in sufficient detail to be included in the meta-analysis. The comparison of meta-analysis results with individual studies shows that only a formal meta-analysis can exactly quantify the combined evidence for linkage and is superior to an informal classification of results as replication or non-replication. Significant linkage of stature is observed on chromosomes 6, 7, 9 and 12 (LOD scores >4) and suggestive linkage with LOD scores >2 is obtained in six additional genetic regions. This is well compatible with the concept of height as a mostly polygenic trait for which also some major genes exist. Candidate genes in the linkage regions are discussed.