POIROT: a powerful test for parent-of-origin effects in unrelated samples leveraging multiple phenotypes

Abstract

Motivation: There is widespread interest in identifying genetic variants that exhibit parent-of-origin effects (POEs) wherein the effect of an allele on phenotype expression depends on its parental origin. POEs can arise from different phenomena including genomic imprinting and have been documented for many complex traits. Traditional tests for POEs require family data to determine parental origins of transmitted alleles. As most genome-wide association studies (GWAS) sample unrelated individuals (where allelic parental origin is unknown), the study of POEs in such datasets requires sophisticated statistical methods that exploit genetic patterns we anticipate observing when POEs exist. We propose a method to improve discovery of POE variants in large-scale GWAS samples that leverages potential pleiotropy among multiple correlated traits often collected in such studies. Our method compares the phenotypic covariance matrix of heterozygotes to homozygotes based on a Robust Omnibus Test. We refer to our method as the Parent of Origin Inference using Robust Omnibus Test (POIROT) of multiple quantitative traits.

Results: Through simulation studies, we compared POIROT to a competing univariate variance-based method which considers separate analysis of each phenotype. We observed POIROT to be well-calibrated with improved power to detect POEs compared to univariate methods. POIROT is robust to non-normality of phenotypes and can adjust for population stratification and other confounders. Finally, we applied POIROT to GWAS data from the UK Biobank using BMI and two cholesterol phenotypes. We identified 338 genome-wide significant loci for follow-up investigation.

Availability and implementation: The code for this method is available at https://github.com/staylorhead/POIROT-POE.

Figure 1

QQ plots of P-values for proposed parent-of-origin effect test under the null hypothesis ${\mathit{\beta }}_{\mathit{M}}={\mathit{\beta }}_{\mathit{P}}=\mathbf{0}$ using a series of 50 000 simulations of 5000 individuals using 3 (left column), 6 (middle column), or 10 (right column) continuous normal phenotypes. MAF is assumed to be 0.25. Horizontal panels depict level of pairwise-trait correlation (low, medium, high). QQ, quantile–quantile; MAF, minor allele frequency.

Figure 2

QQ plots of P-values for proposed parent-of-origin effect test under the null hypothesis ${\mathit{\beta }}_{\mathit{M}}={\mathit{\beta }}_{\mathit{P}}=\mathbf{0}$ using a series of 50 000 simulations of 5000 individuals using 3 (left column), 6 (middle column), or 10 (right column) continuous non-normal phenotypes. MAF is assumed to be 0.25. Horizontal panels depict level of pairwise-trait correlation (low, medium, high). QQ, quantile–quantile; MAF, minor allele frequency.

Figure 3

Power of POIROT to identify POEs assuming K = 3, 6, or 10 normal phenotypes (horizontal panels) compared to univariate test. We assume either 1, 2, or 3 of the phenotypes harbor POEs at the locus (vertical panels). We performed 5000 simulations for each scenario. We calculated power at significance level 0.0005 for our multitrait test and 0.0005/K (Bonferroni correction) and 0.0005/K_eff for the univariate test, where K_eff is the number of PCs needed to explain 90% phenotypic variation. ${\beta }_{Mk}=0.75$ for POE traits, MAF = 0.25, and sample size = 5000. POE, parent-of-origin effect; MAF, minor allele frequency; PCs, principal components.

Figure 4

Manhattan plot of parent-of-origin effects analysis using POIROT with BMI, HDL cholesterol, and LDL cholesterol phenotypes from the UK Biobank. The dashed line represents Bonferroni-adjusted genome-wide significance of 1.5 × 10⁻⁷. BMI, body mass index; HDL, high-density lipoprotein, LDL; low-density lipoprotein.