Arthritis & Rheumatism, Volume 63,
November 2011 Abstract Supplement
Abstracts of the American College of
Rheumatology/Association of Rheumatology Health Professionals
Annual Scientific Meeting
Chicago, Illinois November 4-9, 2011.
Polygenic Modeling of Genome-Wide Association Study Data Reveals Hidden Heritability of Rheumatoid Arthritis Risk.
Stahl1, Eli A., Wegmann2, Daniel, Kraft3, Peter, Kallberg4, Henrik, Kurreeman5, Fina, Gregersen6, Peter K., Alfredsson7, Lars
Brigham and Women's Hospital, Boston, MA
Department of Ecology and Evolutionary Biology, University of California, Los Angeles, CA
Department of Biostatistics, Harvard School of Public Health, Boston, MA
Rheumatology Unit, Department of Medicine, Karolinska Institutet at Karolinska University Hospital Solna, Stockholm, Sweden
Division of Rheumatology Immunology and Allergy, Brigham and Women's Hospital, Boston, MA
Feinstein Institute Medical Reschearch, Manhasset, NY
Institute of Environmental Medicine, Unit of Cardiovascular Epidemiology, Karolinska Institutet, Stockholm, Sweden
Mount Sinai Hospital, Toronto, ON
University of Manchester, Manchester, United Kingdom
Genetic studies of rheumatoid arthritis (RA) susceptibility have identified ~40 loci to date that explain approximately 18% of disease liability, whereas >50% is thought to be genetic. We hypothesized that much of the missing heritability is due to causal variants that are tagged by common SNPs on contemporary genome-wide association study (GWAS) arrays.
We estimated missing heritability and modeled its underlying genetic architecture by analyzing six GWAS datasets totaling 5,485 seropositive rheumatoid arthritis cases and 22,609 controls of European ancestry. After removing known RA risk loci, we assessed whether the remaining >2 million SNPs in aggregate could predict risk in independent case-control collections. We then used polygenic modeling and approximate Bayesian computation to estimate distributions of the number, minor allele frequency and effect size of undiscovered common variants. Heritability analysis using mixed linear model regression analysis was used to corroborate our results. In order to understand whether these undiscovered associations tag rare or common causal variants, we simulated hypothetical case-control datasets using 1000 Genomes Project data.
We found that polygenic risk scores of additive, log-odds weighted risk allele counts at independent SNPs achieving P<0.05 in discovery GWAS are consistently associated with RA case-control status in independent validation data (P= 1×10-9). Both methods estimated that an additional 20% of rheumatoid arthritis disease variance is explained by at least hundreds of SNPs in contemporary GWAS. Simulations of causal and marker variants revealed that an underlying genetic model where most of the causal alleles are common is much more consistent with our observations than models where most of causal alleles are rare.
We conclude that hundreds of causal variants, most of which are common in general population but with a small effect on disease risk, explain an additional 20% of variance in RA risk. Many of these causal variants are discoverable by larger GWAS. Our approach can be applied to understand genetic architecture of many complex traits where GWAS data are available.
To cite this abstract, please use the following information:
Stahl, Eli A., Wegmann, Daniel, Kraft, Peter, Kallberg, Henrik, Kurreeman, Fina, Gregersen, Peter K., et al; Polygenic Modeling of Genome-Wide Association Study Data Reveals Hidden Heritability of Rheumatoid Arthritis Risk. [abstract]. Arthritis Rheum 2011;63 Suppl 10 :167