Characterising the effects of sex interaction, pleiotropy and local population structure on ALS GWAS
Citation:
Byrne, Ross, Characterising the effects of sex interaction, pleiotropy and local population structure on ALS GWAS, Trinity College Dublin.School of Genetics & Microbiology, 2021Download Item:
Abstract:
In the past decade, increasingly large-scale genome wide association studies (GWAS) have gleaned insights into the genetic architecture of a range of human traits. Widespread sharing of summary statistics for these GWAS in large public repositories like the GWAS Catalog has enabled a shift in the paradigm from the study of individual traits to analysis of shared genetic architecture in families of related traits. This has greatly improved our understanding of the landscape of relationships between diseases. However, several important issues remain in the standard GWAS design which may impact downstream applications such as drug discovery and genomic prediction. The most concerning issue is the emerging evidence that residual confounding from recent population structure often persists in spite of current correction methods, leading to statistical inflation and biased effect size estimates. Advances in detecting recent population structure using haplotype sharing methods may hold the solution to this problem, however existing methods are often too computationally costly. Another pertinent GWAS design issue is the general assumption that risk variants have homogeneous effects across all individuals, which may be violated in biologically meaningful ways, such as where sex interacts with genetic risk. For traits with clinical sex-differences in presentation and prevalence, sex-partitioned analyses will likely improve power.
Work in this thesis revisits the analysis of a large published GWAS dataset (N=36,052) for Amyotrophic Lateral Sclerosis (ALS), a rare neurodegenerative disease, to address these issues of residual confounding from recent population structure and the existence of sex-differences, while exploring the opportunities presented by multi-trait analyses with public data. The central aim is to better understand the genetic risk factors underlying ALS, and improve the design of future GWAS.
Given the established behavioural and cognitive secondary symptoms in ALS patients we ran multi-trait analyses of ALS with GWAS summary statistics for psychiatric traits and cognitive performance. This analysis revealed a novel genetic correlation between ALS and bipolar disorder, identified several pleiotropic loci shared between ALS and secondary psychiatric and cognitive traits and implicated ALS genetic risk as a causal factor for lowered cognitive performance. Our findings reveal the role of genetic factors in these extra-motor symptoms, with implications for treatment.
Motivated by clinical observations of sex differences in ALS prevalence and presentation we examined the possibility of sex-differences in the genetic risk for ALS. We partitioned the GWAS into male- and female-only subsets and assessed local and global differences in the genetic architecture of the disease. We identified a significant interaction between sex and SNP-heritability, with lower heritability and polygenicity in males than females, and uncovered several novel and known loci that show sex-differences in association with ALS. These findings build on clinical observations of sex differences in ALS, and indicate a genetic component may be partially responsible.
We carried out detailed population genetic analysis on subsets of the data from Ireland (Chapter 3) and the Netherlands (Chapter 4), exploring haplotype sharing profiles within each country and with a wider European reference dataset. Both countries demonstrated significant finescale population structure, which correlated strongly with local geography. Our analyses identified genetic evidence of several recent historical events, including signatures of the Viking and Norman invasions in Ireland and evidence of the population crash caused by the Black Death in the Netherlands. Strikingly in both countries, principal components (PCs) of these haplotype sharing matrices demonstrated significantly better correlation with geography than standard SNP based PCs, suggesting they may be more suitable for correction of population structure in a GWAS setting.
Building on this observation we analysed haplotype sharing in the full GWAS dataset, using a fast version of ChromoPainter based on the Positional Burrows Wheeler Transform (PBWT-paint) to overcome computational barriers. GWAS run with PCs from this haplotype sharing matrix demonstrated reduced confounding as indicated by a lower LD-score intercept, while retaining the power to detect key ALS loci. Polygenic risk scores calculated using this correction method showed less evidence of bias from population structure, suggesting our method may lead to more robust genomic prediction. Finally we validated a novel pipeline for quickly identifying population clusters in large datasets which successfully identified meaningful genetic clusters in Dutch, British and multi-population data. This should enable application of finescale population genetics to growing datasets.
Sponsor
Grant Number
Motor Neurone Disease Association (MNDA)
Science Foundation Ireland (SFI)
Description:
APPROVED
Author: Byrne, Ross
Advisor:
McLaughlin, RussellPublisher:
Trinity College Dublin. School of Genetics & Microbiology. Discipline of GeneticsType of material:
ThesisCollections
Availability:
Full text availableMetadata
Show full item recordLicences: