Statistical analysis

For the mitochondrial genome, pathway, complex, gene-based analysis, we used the sequence kernel association test (SKAT) [46, 47]. The SKAT_commonrare test is an omnibus procedure allowing for both rare and common variants to contribute to the overall test statistic. For this test, the minor allele frequency of threshold for rare variants was determined by sample size (T=12n) [47] (0.1%< MAF<1% for African, Asian, European, Latino ancestry groups; 1%<MAF<5% for Native Hawaiians). All analyses were adjusted for age, self-reported maternal race/ethnicity, and the first five principal components of genetic ancestry. The principal components of genetic ancestry were estimated from a panel of 128 ancestry informative markers [48, 49]. For single mtSNP and haplogroup analyses, we conducted unconditional logistic regression, adjusting for the same covariates listed above. Haplogroups were estimated using the HaploGrep software (http://www.haplogrep.uibk.ac.at) and based on Phylotree build 16 [50, 51]. Additional adjustment for family history of breast cancer, age at menarche, age at first birth, age at menopause, parity, hormone replacement therapy (HRT) use, body mass index (BMI), alcohol, and smoking did not notably alter the results. Thus, these covariates were not included in our final multivariate models. Stratified analyses were conducted by self-reported maternal race/ethnicity. All statistical tests presented are two-sided. A false discovery rate (FDR) was used to account for multiple hypothesis testing for the mitochondrial genome, OXPHOS pathway, complexes, genes, and haplogroup for the five racial/ethnic groups and overall. A false discovery rate q value [52] of 0.20 was used as a threshold to determine statistical significance, which is equivalent to a nominal p value of 6.7x10^-4 (0.2/300) for ~300 mtSNPs tested and p-value = 0.01 (0.2/20) for the ~20 mitochondrial gene-based tests.