Principal-component analysis for assessment of population stratification in mitochondrial medical genetics

Abstract

Although inherited mitochondrial genetic variation can cause human disease, no validated methods exist for control of confounding due to mitochondrial population stratification (PS). We sought to identify a reliable method for PS assessment in mitochondrial medical genetics. We analyzed mitochondrial SNP data from 1513 European American individuals concomitantly genotyped with the use of a previously validated panel of 144 mitochondrial markers as well as the Affymetrix 6.0 (n = 432), Illumina 610-Quad (n = 458), or Illumina 660 (n = 623) platforms. Additional analyses were performed in 938 participants in the Human Genome Diversity Panel (HGDP) (Illumina 650). We compared the following methods for controlling for PS: haplogroup-stratified analyses, mitochondrial principal-component analysis (PCA), and combined autosomal-mitochondrial PCA. We computed mitochondrial genomic inflation factors (mtGIFs) and test statistics for simulated case-control and continuous phenotypes (10,000 simulations each) with varying degrees of correlation with mitochondrial ancestry. Results were then compared across adjustment methods. We also calculated power for discovery of true associations under each method, using a simulation approach. Mitochondrial PCA recapitulated haplogroup information, but haplogroup-stratified analyses were inferior to mitochondrial PCA in controlling for PS. Correlation between nuclear and mitochondrial principal components (PCs) was very limited. Adjustment for nuclear PCs had no effect on mitochondrial analysis of simulated phenotypes. Mitochondrial PCA performed with the use of data from commercially available genome-wide arrays correlated strongly with PCA performed with the use of an exhaustive mitochondrial marker panel. Finally, we demonstrate, through simulation, no loss in power for detection of true associations with the use of mitochondrial PCA.

Figure 1

Autosomal Population Structure for the MGH-AIS Study Population structure of the MGH-AIS subjects (black dots), based on autosomal PCs 1 and 2, compared with reference populations from Phase 3 of the HapMap Project. CEU (red dots): residents of Northwestern European ancestry residing in Utah, US; TSI (green dots): Tuscans in Italy; ASW (yellow dots): African Americans in Southwestern USA; MEX (gray dots): Mexicans in Mexico City; CHB (purple dots): Han Chinese in Beijing.

Figure 2

Autosomal Population Structure and Mitochondrial Haplogroups in the MGH-AIS Cohort (A) Autosomal population structure as represented by plotting autosomal PCs 1 and 2 and corresponding individual mitochondrial haplogroup assignment in the MGH-AIS cohort. Mitochondrial haplogroups clearly distinguish individuals of Asian and African ancestry from European American individuals, but they fail to correctly identify Mexican American individuals (identified as Asian haplogroup carriers). (B) Autosomal population structure and mitochondrial haplogroup assignment in the European American MGH-AIS subjects. No intracontinental distribution of haplogroups is discernible within the European-ancestry cluster. Autosomal PCA was performed separately after the removal of individuals of non-European ancestry.

Figure 3

Mitochondrial Population Structure in MGH-AIS (A) Relationship between European mitochondrial haplogroups. Different colors identify haplogroups descending from the same ancestral haplogroup (preHV = green, JT = light blue, N = orange). (B) mtPCs 1 and 2 recapitulate haplogroup information in European Americans enrolled in the MGH-AIS. Colored contours identify haplogroups as in (A). Colored dots convey information about haplogroup assignment for each individual. (C) Plotting mtPCs 1 and 3 further separates haplogroups K, I, and WX from U as compared to the plot in (B). Colored dots convey information about haplogroup assignment for each individual as in (B). (D) Tridimensional plot of mtPCs 1–3 in European Americans enrolled in MGH-AIS assigns individuals to clusters that recapitulate mitochondrial haplogroup assignment.

Figure 4

Correlation Coefficients for Autosomal PCs and mtPCs Correlation between autosomal PCs and mtPCs is visualized as the distribution of correlation coefficients (absolute values) for each population enrolled in the participating studies. HGDP, Human Genome Diversity Panel; MGH-AIS, Massachusetts General Hospital Ischemic Stroke Study; MGH-ICH, Massachusetts General Hospital Intracerebral Hemorrhage Study; ISGS, Ischemic Stroke Genetics Study; SWISS, Siblings With Ischemic Stroke Study

Figure 5

Comparison of Haplogroup-Based and PCA-Based Control for PS in the MGH Mitochondrial Stroke Genetics Study Quantile-quantile plots for the mitochondrial GWAS of ischemic stroke in the MGH-AIS cohort and for corresponding mtGIF are shown for analyses adjusted for PS with the use of either haplogroup-based or mitochondrial PCA-based methods.