Inherited Variants in Mitochondrial Biogenesis Genes May Influence Epithelial Ovarian Cancer Risk

Epithelial ovarian cancer (EOC) is the ninth most commonly diagnosed female cancer and the most lethal gynecologic malignancy in the United States (1). Although the etiology remains largely unknown, there are 2 leading hypotheses. The incessant ovulation hypothesis (2) states that repeated ovulations overstimulate the ovarian surface epithelium, causing inflammation and increased cellular proliferation during epithelial repair, leading to DNA replication errors and malignant transformation. The gonadotropin hypothesis (3) posits that gonadotropins are responsible for this process. Supporting these hypotheses, epidemiologic studies have shown that factors which interrupt ovulation and/or lower gonadotropin levels, including pregnancy, oral contraceptive use, and lactation, are inversely associated with risk (4). However, factors unrelated to hormones, reproduction, or ovulation may also influence risk, suggesting alternate mechanisms may also contribute to the genesis of EOC (4).

Emerging experimental and epidemiologic evidence suggests EOC development and progression may be caused by oxidative stress (5–9), a phenomenon arising because of overproduction of reactive oxygen species (ROS) in mitochondria (10). Mitochondria are semiautonomous membrane-bound organelles that participate in free radical production, apoptosis, and energy metabolism (11). Aside from the nucleus, the mitochondrion is the only cellular organelle that contains its own genome (mtDNA) and genetic machinery. mtDNA is a maternally-inherited, 16.6 kb double-stranded, closed circular molecule, that encodes 37 genes (13 polypeptides, 22 tRNAs, and 2 rRNAs) mainly involved in oxidative phosphorylation (OXP), a process whereby molecular oxygen is oxidized to water, creating ATP and the by-product, ROS (11). Approximately 1,500 proteins encoded by nuclear DNA (nDNA) generate new mitochondria (known as mitochondrial biogenesis) and maintain mitochondrial structure and function by regulating processes such as OXP, apoptosis, and mtDNA replication, transcription, and translation (11). Importantly, interplay between the genomes can influence disease processes, and studies implicate nDNA-encoded genes involved in mitochondrial biogenesis as regulators of nuclear-mitochondrial interactions (5, 11, 12).

Although much attention has been focused on somatic mtDNA mutations (11, 13), less research has been directed to the influence of germline variants on mitochondrial dysfunction and cancer development. Single nucleotide polymorphisms (SNPs) in mtDNA and/or nDNA may enhance cancer risk through subtle changes in encoded proteins, altered OXP activity, and excess ROS production over time (Fig. 1; refs. 10, 13). In particular, variation involving mtDNA (13) and the following 3 categories or “pathways” of nDNA may alter cancer risk by promoting oxidative stress. Mitochondrial biogenesis (BIO) proteins represent transcription factors and coactivators that regulate mtDNA and nDNA involvement in OXP; because defective mitochondrial biogenesis may promote slower mitochondrial turnover and altered OXP activity, control of this process is integral for maintenance of energy production and prevention of ROS accumulation (12, 14, 15). Similar to their mtDNA-encoded counterparts, nDNA-encoded OXP proteins, when impaired, may alter OXP activity and contribute to diseases including hereditary cancer syndromes (16). Steroid hormone metabolism (HOR) proteins, primarily receptors localized to mitochondria, modulate mitochondrial gene expression and OXP activity (17, 18). SNPs in these pathways may also influence risk by interacting with 1 another or with nongenetic factors known to contribute to ROS accumulation (Fig. 1).

Few studies have examined potential associations between inherited mitochondrial SNPs and/or haplotypes and cancer risk (19–25). Only 1 study (25) investigated nDNA in addition to mtDNA, and only 1 small study in China (22) involved ovarian cancer. Thus, there is a need for large-scale studies that comprehensively evaluate inherited mitochondrial variation and EOC risk and incorporate epidemiologic risk factors into the analysis.

The study protocol was approved by the institutional review board at each center, and all study participants provided written informed consent. Data derived from 4 case–control studies of EOC: the Mayo Clinic Ovarian Cancer Study (MAY; Rochester, Minnesota), Duke University's North Carolina Ovarian Cancer Study (NCO; Durham, North Carolina), the University of Toronto Familial Ovarian Tumor Study (TOR; Ontario, Canada), and H. Lee Moffitt Cancer Center and Research Institute's Tampa Bay Ovarian Cancer Study (TBO; Tampa, FL). Study characteristics are summarized in Table 1. All studies recruited incident, pathologically confirmed primary EOC cases, either borderline or invasive, aged 20 and above. Three studies used population-based rapid ascertainment for the cases (NCO, TOR, TBO), and 1 study was clinic-based (MAY). All controls had at least 1 ovary intact at the reference date and were frequency-matched to cases on age group and race.

Demographic data and information on known and suspected EOC risk factors was collected from participants by using study-specific questionnaires. Data collected included race, ethnicity, age at diagnosis/interview, menstrual and reproductive history, exogenous hormone use, medical and surgical history, adult height and weight, smoking history, and family history of ovarian and breast cancer in first-degree relatives. To increase etiologic homogeneity, we excluded cases with nonepithelial or borderline tumors, nonwhites, known BRCA1 and BRCA2 mutation carriers, Jewish women (because of the likelihood of carrying a BRCA mutation), and women with a prior history of ovarian, breast, endometrial, or early-onset colorectal cancer (because of the likelihood of having a mismatch repair defect).

Blood served as the source of genomic DNA and was collected at each study site at the conclusion of the interview (NCO, TBO, and TOR) or in the course of medical care (MAY). Genomic DNA was isolated from whole blood by using PureGene DNA isolation reagents (Gentra Systems), resuspended in Tris EDTA buffer, and stored at 4°C. Samples were bar coded with a unique subject identification number to ensure accurate and reliable sample processing and storage.

All samples were genotyped by using the Illumina Infinium 610K Array at the Mayo Clinic Genotyping Shared Resource Facility (Rochester, MN) by laboratory personnel blinded to case–control status. Each 96-well plate contained 375 ng DNA of random mixtures of case and control samples, 2 blind duplicates, and 2 replicates of a CEPH family trio (mother, father, and child) from the Coriell Institute. A quality assurance (QA) panel of 96 SNPs was run on the Illumina Bead Express platform to test sample performance and ensure concordance of replicate samples.

Illumina's Genome Studio software was used to carry out automated genotype clustering and calling of the potential 550,000 beadtypes. SNPs were excluded from further analysis if (i) the call rate was less than 95%, (ii) they were monomorphic on manual clustering, or (iii) there were unresolved replicate errors. Among 81 pairs of replicate samples, the overall concordance rate was 99.93%.The overall genotype call rate was 99.7%.

We attempted to genotype unique samples from 4,150 eligible subjects. Of these, 394 were excluded because genotypes were generated at fewer than 95% of SNPs (i.e., sample call rate <95%); 81.4% of these exclusions (n = 321) were attributed to poor quality DNA from the TOR site. Other reasons for exclusion included sample failure under the QA panel (n = 15), ambiguous gender (n = 7), unresolved identical genotypes (n = 8), self-report as non-Caucasian (n = 2), and those predicted by structure (26) analysis to have less than 80% European ancestry (n = 9). This resulted in a final sample size of 3,715 subjects (1,815 cases and 1,900 controls).

mtDNA content on the 610K array was determined by comparing an annotation file supplied by Illumina with the MitoMap database (27). Of 138 mtDNA SNPs from 22 genes and 8 regulatory regions included on the 610K array, 128 (92%) had call rates greater than 95%, but only 24 SNPs had minor allele frequency among controls (MAF) greater than or equal to 5% (Table 2). Information regarding nDNA proteins was derived from published literature and several databases, including MitoProteome (28), the Human Mitochondrial Protein Database (29), and the Kyoto Encyclopedia of Genes and Genomes Pathway database. A total of 3,064 SNPs in 142 nDNA genes of interest (35 BIO, 14 HOR, and 93 OXP) are on the 610K array and of these, 2,839 SNPs (93%) from 138 genes had call rates greater than 95% (Table 2). To evaluate evidence for SNP-level associations at nDNA markers that were not directly genotyped, imputation was carried out by using MACH version 1.0.16 by using phased data from HapMap release 22 (genome build 36) on individuals with European ancestry (CEU). We imputed 4,445 SNPs that were not genotyped as part of the 610K array that reside within the 138 nDNA-encoded genes.

Descriptive statistics were generated by using means and standard deviations for continuous variables and frequencies and percents for categorical variables. Distributions of covariates among cases and controls were compared within each study site by using t-tests and χ² tests for continuous and categorical variables, respectively. Genotype frequencies of nDNA SNPs among controls were tested for Hardy–Weinberg Equilibrium (HWE) by using χ² goodness-of-fit tests. Statistical analysis was conducted by using SAS Version 9.1 (SAS Inc.) and PLINK software (30).

Association testing was conducted for SNPs, haplotypes, genes, and pathways. To examine associations between each SNP and EOC risk, unconditional logistic regression was used to estimate OR and 95% CI between carriage of the minor versus major allele of maternally-inherited mtDNA SNPs and case status, and between genotypes and case status for nDNA SNPs. Log-additive, dominant, and recessive genetic models were fit for each nDNA SNP; the major allele was considered to be the reference allele during modeling. All models were adjusted for the design variables of age (continuous) and study site (indicator variables for NCO, TBO, and TOR). Because of the potential for population stratification, models were also adjusted for a quantitative variable for the first principal component representing European ancestry (26). Wald χ² tests were used to obtain P values for dominant and recessive SNP effects, and the Cochran Armitage trend test was used to estimate P for trend for log-additive effects. The best fitting model was represented by the 1 with the smallest P value. All P values were 2-sided, and a nominal P < 0.05 was considered the threshold of significance for SNP-level tests.

We permuted case–control status 100,000 times to generate point-wise empirical P-values (EMP1) by using PLINK. The EMP1 represents the proportion of permutations in which the minimum simulated P value was less than the observed P value. To predict the potential functional significance of risk-associated and strongly correlated (r² > 0.80) SNPs, we used the SNPinfo (31) and FastSNP (32) web-based tools.

To evaluate whether established or suspected EOC risk factors (lifetime ovulation, hormones, cigarette smoking) postulated to contribute to oxidative stress may confound the most promising SNP-EOC risk associations, we conducted analyses among subjects with complete data for variables of interest (N = 2,662; 1,217 cases, 1,445 controls, representing 72% of 3,715 total subjects). Similar to other studies that estimated lifetime ovulation (33), a composite variable for lifetime ovulatory cycles (LOC) was estimated by considering self-reported ovulatory and anovulatory periods by using the following modified formula:

LOC = {[index age – age at first menstrual period] – [(number of months pregnant + number of months of oral contraceptive use)/12 months]} × 13 cycles/year

where 13 cycles/year is assumed based on an average cycle length of 28 days. For postmenopausal women, index age was defined as age at menopause. For pre- or perimenopausal women, age at interview/diagnosis served as the index age. The LOC composite variable was subsequently categorized based on tertiles of the distribution among controls. Other factors that may influence LOC, such as breastfeeding, missed or irregular periods, or spontaneous or elective abortions, were not considered because pertinent data were not ascertained by or available from all participating sites. For each SNP investigated, the model included age (continuous), study site (indicator variable), LOC category (low, medium, high), duration of unopposed estrogen or combined estrogen–progestin hormone replacement therapy (HRT) use (continuous), and pack-years of cigarettes smoked (continuous).

To explore SNP–environment interactions, modifying effects of the following nongenetic variables were considered: LOC category (high and medium tertile vs. lowest tertile), HRT use (ever versus never), smoking history (ever versus never), and menopausal status (peri/pre versus post). Multiplicative interactions were evaluated by fitting logistic regression models with the corresponding SNP under a dominant effect, nongenetic variable, and interaction term. P values for testing the interaction effects were obtained by using Wald χ² tests. For interactions with P < 0.05, stratified analyses were conducted.

For each gene having multiple SNPs associated with risk, we estimated pairwise linkage disequilibrium (LD) by using Haploview v4.1 (34). For regions with high LD (r² > 0.80), we tested for associations between haplotypes and case–control status by using the Haplo.stats program and R software v2.10.0 according to methods proposed by Schaid and colleagues (35). Estimates of ORs, 95% CIs, and P-values were obtained, with adjustment for age and study site under the specified genetic model. Rarer haplotypes (frequencies <10%) were combined into a single category to minimize sparse cell counts. LD plots displaying the SNP correlation structure in selected regions are displayed in Supplementary Figure S1.

To complement single SNP analysis, we used Fisher's method (36) to combine association evidence from a group of SNPs within a gene for gene-level analysis. This method has been shown to have high statistical power in detecting associations (37). Briefly, the test statistic combines multiple results in the following form, |$F = - 2\sum\nolimits_{i = 1}^m {\ln P_i }$|

where P_i is the SNP level P value (adjusted for age and study site) for the ith SNP within each gene. To enhance power to detect associations, only common SNPs (MAF ≥ 5%) were included in this analysis. Because of the correlation structure of SNP data, the statistical significance for each gene was assessed by using 10,000 permutations with case/control labels permuted. The same statistical test was conducted for the pathway-level analysis; association tests for each of the 4 pathways (mtDNA, BIO, OXP, and HOR) were conducted. In a similar fashion, the statistical significance for each pathway was assessed by using 10,000 permutations. Similar to methods employed by Goode and colleagues (38), the results were interpreted hierarchically at the pathway, gene and SNP levels. Pathway- and gene-level results were not corrected for multiple testing because of the exploratory nature of this analysis.

The distributions of selected characteristics of the 1,815 EOC cases and 1,900 controls are summarized by study site and case–control status in Table 3. Despite frequency matching on age group, cases were older (60.0 ± 11.6 years) than controls (56.8 ± 12.0 years; P < 0.0001). As compared with controls, cases tended to have lower education levels (P < 0.0001), a higher number of LOC (P < 0.0001), were less likely to have used oral contraceptives (OC; P = 0.002), had longer duration of HRT use (P < 0.0001), and a higher number of pack-years of cigarettes smoked (P < 0.0001).

Results from association testing for mtDNA and the 3 nDNA pathways are presented in a tiered approach, beginning with evaluation at the pathway level, followed by gene- and SNP-level results. At the pathway and gene levels, we focused on results significant at empirical P < 0.10. At the SNP level, we focused on results significant at P < 0.05 and EMP1 < 0.05, and comment on in silico findings and imputated genotype results where appropriate. Adjustment for the first principal component representing European ancestry did not change parameter estimates and CIs appreciably, so results from simple models are presented. We report on haplotype effects with global P < 0.05.

Pathway analysis revealed that inherited variation in mtDNA (17 genes/regulatory regions, 24 SNPs analyzed) was not significantly associated with EOC risk (empirical P = 0.510), therefore, gene- and SNP-level results are interpreted with caution. At the gene level, only MT-CO1 (COX1) was associated with risk (empirical P = 0.006; Supplementary Table S1). At the SNP level, a synonymous SNP in MT-CO1, T6777C (OR = 0.68, 95% CI: 0.51–0.92, P = 0.006) seemed to decrease risk (Table 4), consistent with haplotype analysis (OR = 0.81, 95% CI: 0.70–0.94, P = 0.006; Supplementary Fig. S1A). Association results for all evaluated mtDNA genes and SNPs are in Supplementary Tables S1 and S2, respectively.

Of the 3 nDNA pathways studied, only the 25 genes (1,051 SNPs) involved in BIO had a pathway-wide association with EOC risk (empirical P = 0.050). At the gene level, nuclear respiratory factor 1 (NRF1) had the strongest result (empirical P = 0.041). Four other genes also showed gene-wide associations, including mitochondrial transcription termination factor (MTERF) peroxisome proliferator-activated receptor gamma, coactivator 1 alpha (PPARGC1A), estrogen-related receptor alpha (ESRRA), and calcium/calmodulin-dependent protein kinase D (CAMK2D), with empirical P values of 0.044, 0.060, 0.066, and 0.067, respectively (Supplementary Table S1).

Association results for BIO SNPs significantly associated with EOC risk are displayed in Table 4. For NRF1, the strongest association with EOC risk was observed under a recessive model with rs10245560 [OR (95% CI) = 0.80 (0.68–0.95), P = 0.010; Supplementary Fig. S1B]. MTERF includes several sets of correlated risk-associated SNPs (Supplementary Fig. S1C), with ORs ranging from 0.89 to 2.90. Of all BIO genes, PPARGC1A contains the most risk-associated SNPs, the majority of which flank the 5′ or 3′ UTR, including several correlated SNP pairs (Supplementary Fig. S1D). ESRRA contains rs11600990, a SNP associated with decreased risk under a dominant model [OR (95% CI) = 0.83 (0.72–0.97), P = 0.016] that was predicted by SNPinfo to reside in a putative transcription factor binding site (TFBS). Finally, we identified several intronic SNPs in CAMK2D, including rs2040742, a predicted intronic enhancer by FastSNP associated with decreased risk (Supplementary Fig. S1E). After adjustment for lifetime ovulation, HRT use, and pack-years smoked, similar magnitudes of association were observed for the majority of BIO SNPs depicted in Table 4, although associations seemed to strengthen for several SNPs in MTERF, PPARGC1A, and CAMK2D.

Imputed genotypes reinforced SNP-level associations for several BIO genes. For example, rs10954252 and rs9929 are imputed SNPs within NRF1 and CAMK2D, respectively, that were most strongly associated with EOC risk [OR (95% CI) = 1.14 (1.04–1.25), P = 0.006 for rs10954252 and OR (95% CI) = 1.36 (1.12–1.66), P = 0.002 for rs9929; data not shown]. Imputation also revealed previously undetected signals for mitogen activated protein kinase 1 (MAPK1) and transcription factor B1, mitochondrial (TFB1M), with the strongest signal identified for TFB1M rs721101 [OR (95% CI) = 0.86 (0.77–0.96), P = 0.006; data not shown).

Interindividual variation within the HOR pathway (9 genes, 532 SNPs analyzed) was not significantly associated with EOC risk (empirical P = 0.35), so gene- and SNP-level results are interpreted cautiously. Estrogen receptor 2 (ESR2) is the only HOR gene that was associated with EOC risk (empirical P = 0.026; Supplementary Table S1). Four intronic ESR2 SNPs were individually associated with EOC risk (Table 4), and a haplotype block including rs1256062, rs1256061, and rs12435857 was associated with a slightly increased risk under an additive model (OR = 1.22, 95% CI: 1.01–1.49, global P = 0.029; Supplementary Fig. S1F). Imputation reinforced evidence of associations within ESR2, and identified 17 new signals for NR3C1/GR (data not shown).

Overall, the OXP pathway (73 genes, 1,043 SNPs) was not statistically significantly associated with EOC risk (empirical P = 0.68; Supplementary Table S1). Individually, 7 OXP genes showed statistically significant gene-level empirical P values, and Tu translation elongation factor (TUFM) had the strongest gene-level result (empirical P = 0.014, Supplementary Table S1). At the SNP level, TUFM rs9972768 resides in a putative TFBS and is associated with an 11% increased risk with each copy of the minor allele (Table 4).

Because of the global significance of the BIO pathway, exploratory models were fitted to test interactions between the 39 risk-associated BIO SNPs (Table 4) and a set of a priori determined nongenetic variables linked to oxidative stress as lifetime ovulation, HRT use, cigarette smoking, and menopausal status. This analysis was limited to a dominant genetic effect, for a total of 156 tests (39 SNPs × 4 nongenetic factors). There were 8 interactions significant at P < 0.05, which is the same as expected based on chance alone, and they involved SNPs from MTERF, PPARGC1A, and CAMK2D. Most noteworthy, stratified analyses showed MTERF rs10488506 was a risk factor among ever smokers [OR (95% CI) = 1.44 (1.09–1.91), P = 0.011] but not never smokers [OR (95% CI) = 0.98 (0.77–1.24), P = 0.87]. PPARGC1A rs1509241 was a risk factor among pre-/perimenopausal women [OR (95% CI) = 0.71 (0.56–0.90), P = 0.004] but not postmenopausal women [OR (95% CI) = 0.99 (0.83–1.19), P = 0.95]. HRT use seemed to modify the association between CAMK2D rs13107662 and EOC risk, with an inverse association observed among ever users [OR (95% CI) = 0.72 (0.58–0.89), P = 0.003] but no association among never users [OR (95% CI) = 0.98 (0.82–1.16), P = 0.80].

Multiple lines of evidence, including animal, in vivo, and epidemiologic studies, point to a role for oxidative stress in the etiology and/or progression of EOC (5–9). This report explored the hypothesis that inherited variants in mitochondrial-related genes may influence EOC risk, possibly by contributing to oxidative stress, by using a large, homogeneous study population of 1,815 incident pathologically-confirmed Caucasian EOC cases and 1,900 controls. The analysis focused on 128 mtDNA-encoded SNPs from 22 genes and 8 regulatory regions and 2,839 nDNA-encoded SNPs from 138 genes having functions related to BIO, HOR, and OXP. We considered various genetic models, haplotypes, gene and pathway-level analyses, in silico findings, and evaluation of potential confounders and effect modifiers. Recognizing the large number of statistical comparisons, the data suggest that inherited variants in key genes involved in mitochondrial biogenesis may represent contributors to EOC susceptibility and therefore merit efforts to replicate in additional study populations.

BIO is important because of the DNA damage endured by these organelles over a lifetime which can affect aging, neuromuscular diseases, ovarian insufficiency, and risk of cancer (39–42). This process is influenced by environmental stimuli (i.e., temperature, nutrients, hormones, and exercise) that change the energetic and physiologic conditions of the cell and in turn increase intracellular calcium concentrations which stimulate kinases to activate mitochondrial gene transcription (14). Because BIO is repressed in cancer cells (40) and its reactivation can induce apoptosis in cancer cells (39), knowledge gained regarding genetic factors that influence BIO in EOC may have implications for biomarker development and therapeutics.

At the transcriptional level, 2 classes of nDNA-encoded regulatory proteins direct nuclear–mitochondrial interactions and promote mitochondrial biogenesis, and our study highlighted genes from each class (14, 15, 41). MTERF regulates OXP by acting as an mtDNA transcript terminator, initiator, and controller of mtDNA replication (43). We observed what seemed to be independent association signals from different MTERF gene regions, many of which strengthened after accounting for confounding by and/or interaction with HRT use and smoking history. To our knowledge, this is the first report that has linked MTERF to cancer development. In response to oxidative stress, NRF1 activates transcription of antioxidant and detoxification genes (44). Our findings of NRF1 gene- and SNP-level associations (and a high proportion of associated NRF2 SNPs) seem promising in light of accumulating data regarding the role for nuclear respiratory factors, especially NRF2, in chemoprevention and therapeutics (45). PPARGC1A modulates the function of most mitochondrial proteins (12) and serves as a coactivator for ESRRA, an inducer of BIO and regulator of OXP and oxidative stress defenses (12, 46). In ovarian tumors, a PPARGC1A-ESRRA pathway regulates expression of vascular endothelial growth factor (VEGF), a positive regulator of angiogenesis (47). Overexpression of PPARGC1A and ESRRA has been associated with poor EOC prognosis, suggesting these genes may be useful therapeutic targets (48, 49). CAMK2D is a serine/threonine kinase involved in calcium signaling (15); these kinases are dysregulated in many malignancies and participate in cell growth, apoptosis, and angiogenesis (50).

Given that the HOR, OXP, and mtDNA pathways were not associated with EOC risk in our study, caution should be taken when interpreting gene- and SNP-level results. Consistent with our finding of gene- and SNP-level associations between ESR2 and EOC risk, studies support a role for ESR2 in mitochondrial dysfunction, oxidative stress, and ovarian carcinogenesis (18, 51). On the basis of data suggesting cross talk between ESR2 and ESRRA (52), the possibility remains that stimulation of mitochondrial proteins by steroid hormones may cause altered BIO and function, leading to hormone-related cancers. All nDNA-encoded subunits of the OXP pathway were encompassed on the 610K array; collectively this pathway may not have been associated with EOC risk because expression of these enzymes is most critical in tissues or organs with high energy demands (i.e., muscle, brain, and heart). However, given the role for OXP enzymes in ovaries and female reproductive function, additional research may be warranted in this area (53). Most mtDNA variants are rare (MAF < 1%), possibly because of negative selection of germline changes during evolution (54), limiting power to detect SNP level and higher associations. Although our coverage of mtDNA genes and SNPs is greater than other studies (19, 23, 25), the array did not encompass 1 polypeptide (ATPase 8) and 14 of 22 tRNAs. Further research is necessary to clarify the role of mtDNA-encoded genes in EOC.

The current study did not evaluate other types of nDNA-encoded SNPs that may impact oxidative stress, such as those that influence antioxidant defenses (i.e., MPO and GSTM1), DNA repair (i.e., MGMT and XRCC1), carcinogen metabolism (i.e., CYP1A1 and GST), and inflammation (i.e., TNF-α and IL-6), because they have been previously examined (55, 56). Mitochondrial variants involved in intrinsic apoptosis (i.e., caspases and APAF1) may also alter EOC risk and should be explored. Another limitation pertains to our subanalysis in which we evaluated whether lifetime ovulation, HRT, and cigarette smoking may confound or modify observed SNP-EOC risk associations. Because we generated crude estimates of lifetime ovulation that did not consider anovulatory periods because of breastfeeding, menstrual irregularities, or incomplete pregnancies, this measurement lacks precision, and observed associations may be attenuated. Furthermore, we lacked data on HRT formulation and could not consider effects of unopposed estrogen therapy separately from combined estrogen–progestin HRT. Although subanalyses were based on only 72% of our study sample, their profile did not seem to systematically differ from the entire study population (data not shown), and comparison of parameter estimates for individual SNPs by using simple and full models generally showed magnitudes of association similar in direction and strength (Table 4).

In summary, this study reports on the largest number of inherited mitochondrial polymorphisms, genes, and pathways to date for associations with EOC risk, and is the first to suggest that SNPs in BIO genes may represent novel EOC susceptibility loci. On the basis of internal consistency and biologic plausibility, functional validation of the most promising SNPs identified and characterized as part of this research is warranted. Furthermore, because of the heterogeneous nature of EOC (4), examination of histology-specific effects in larger sample sizes is needed.

No potential conflicts of interest were disclosed.

We thank all of the women who participated in this research along with all of the researchers, clinicians, and staff who have contributed to the participating studies.

This work was supported by the National Cancer Institute (R01 CA114343 to T. A. Sellers, R01 CA122443 to E. L. Goode, R01 CA 76016 to J. M. Schildkraut, R01 CA 106414 to R. Sutphen, and P30 CA15083 to the Mayo Clinic Genotyping Shared Resource).