Association testing of the mitochondrial genome using pedigree data

Abstract

In humans, mitochondria contain their own DNA (mtDNA) that is inherited exclusively from the mother. The mitochondrial genome encodes 13 polypeptides that are components of oxidative phosphorylation to produce energy. Any disruption in these genes might interfere with energy production and thus contribute to metabolic derangement. Mitochondria also regulate several important cellular activities including cell death and calcium homeostasis. Aided by sharply declining costs of high-density genotyping, hundreds of mitochondrial variants will soon be available in several cohorts with pedigree structures. Association testing of mitochondrial variants with disease traits using pedigree data raises unique challenges because of the difficulty in separating the effects of nuclear and mitochondrial genomes, which display different modes of inheritance. Failing to correctly account for these effects might decrease power or inflate type I error in association tests. In this report, we sought to identify the best strategy for association testing of mitochondrial variants when genotype and phenotype data are available in pedigrees. We proposed several strategies to account for polygenic effects of the nuclear and mitochondrial genomes and we performed extensive simulation studies to evaluate type I error and power of these strategies. In addition, we proposed two permutation tests to obtain empirical P values for these strategies. Furthermore, we applied two of the analytical strategies to association analysis of 196 mitochondrial variants with blood pressure and fasting blood glucose in the pedigree rich, Framingham Heart Study. Finally, we discussed strategies for study design, genotyping, and data cleaning in association testing of mtDNA in pedigrees.

Figure 1

Covariance structures for polygenic effects from nuclear and mitochondrial genomes, respectively, in a nuclear family of size five (two parents and three offspring). nDNA = nuclear DNA, mDNA = mitochondrial DNA.

Figure 2

Type I error rates estimated at α = 0.001 when $h_g^2=0\%$ for $h_G^2+h_N^2=50\%$. The y-axis is plotted on a log10 scale, therefore, 0.001, 0.002, 0.003, 0.004, 0.005 correspond to −3.0, −2.70, −2.52, −2.40, and −2.30 on the log10 scale, respectively. The x-axis is $h_g^2(\%)$. The five strategies considered were as follows: 1) UI - unrelated individuals (i.e., founders or singletons) were analyzed using simple linear regression. 2) NF - all individuals were included and familial correlation (Figure 1A) due to autosomal polygenic component among nuclear families was accounted for in a linear mixed effects (LME) model. 3) MO - fathers were excluded from the sample, but mother and offspring were used for analysis. This reduced covariance structure (without father in Figure 1A or 1B) could be accommodated by LME. 4) ML - all individuals were used. Random effect in SSML (Figure 1B) was accounted for in a LME model. 5) BO - all individuals were used. Both coefficient structures of nDNA (Figure 1A) and mtDNA (Figure 1B) were accounted for in a LME model.

Figure 3

Empirical type I error rates corresponding to α = 0.001 for the two permutation tests when $h_G^2+h_N^2=50\%$. The y-axis is plotted on a log10 scale. Therefore, 0.001, 0.0001, and 0.0001 correspond to −3.0, −4, and −5 on the log10 scale, respectively. The x-axis is $h_g^2(\%)$. Please see footnote in Figure 2 for description of the analytical strategies. In the parentheses of the legend, “M” stands for maternal permutation or PERM_M and “F-M” stands for father-mother pair permutation or PERM_F-M.

Figure 4

Power estimated using α_{PERM_F_M_0.001} and α_{PERM_M_0.001} when $h_G^2+h_N^2=50\%$. Please see footnote in Figure 2 for description of the analytical strategies. In the parenthesis of the legend, “M” stands for maternal permutation or PERM_M and “F-M” stands for father-mother pair permutation or PERM_F-M.