Simultaneous genome-wide association studies of anti-cyclic citrullinated peptide in rheumatoid arthritis using penalized orthogonal-components regression.

Genome-wide associations between single-nucleotide polymorphisms and clinical traits were simultaneously conducted using penalized orthogonal-components regression. This method was developed to identify the genetic variants controlling phenotypes from a massive number of candidate variants. By investigating the association between all single-nucleotide polymorphisms to the phenotype of antibodies against cyclic citrullinated peptide using the rheumatoid arthritis data provided by Genetic Analysis Workshop 16, we identified genetic regions which may contribute to the pathogenesis of rheumatoid arthritis. Bioinformatic analysis of these genomic regions showed most of them harbor protein-coding gene(s).

Background

Most of the available statistical approaches for genome-wide association study (GWAS) have focused on studying one single-nucleotide polymorphism (SNP) at a time [1], thereby ignoring the multigenic nature of complex diseases [2,3] and the strong correlation between some SNPs due to linkage disquilibrium (i.e., some SNPs are inherited together in blocks of DNA). As pointed out by Waldron et al. [4], association studies using multiple SNPs have substantial advantages over those based on SNP associations. To capture the correlation between SNPs in regions of low recombination, haplotype-based methods that recognize the existence of linkage disequilibrium between genetic markers have been employed for multi-SNP analyses. However, such methods introduce additional problems, including the need to infer haplotypes, the impact of including (or excluding) rare haplotypes, and the need to define haplotype block boundaries. Even with haplotype-based association methods, high false-positive rates and a lack of reproducibility remain major concerns.

Because hopes remain high for the value of GWAS, other approaches need to be pursued that account for the correlation structure among SNPs. Here we have developed a method for GWAS that incorporates regression of a phenotype simultaneously on all available SNPs, i.e., considering a multiple linear regression,

where Y is the phenotypic value, Xcounts one of the two alleles at the jth SNP, and βis the additive effect of that allele, j = 1,..., p. To account for the issues raised by large p, such as lack of independence between SNPs due to linkage disequilibrium, we will conduct the GWAS using the penalized orthogonal-components regression (POCRE) [5]. POCRE offers a fast and efficient way to identify significant SNPs simultaneously from a large number of candidates.

Methods

GWAS usually entails the collection of a massive amount of SNPs (i.e., large p) for only a small number of biological individuals (i.e., small n). Therefore, identifying the few genetic variants underlying disease risk is equivalent to the task of finding "a very few needles in a haystack", and poses a challenging statistical issue. Zhang et al. [5] recently described the POCRE approach, which sequentially constructs sparsely loaded orthogonal components with proper regularization. They demonstrate this approach works well when fitting regression models with n <