Ovarian Cancer Risk and Polymorphisms Involved in Estrogen Catabolism

The “incessant ovulation” hypothesis states that the repeated cellular proliferation required to repair epithelial tissue after each ovulation may lead to ovarian neoplasms (1). The decreased risk of ovarian cancer associated with pregnancy and oral contraceptive use fits with this hypothesis; however, it has been shown that the risk reductions reported for these exposures are greater than would be expected (2). Some have suggested that there are additional mechanisms contributing to carcinogenesis, such as steroid hormones and growth factors (3). After ovulation, when epithelial cells are proliferating and DNA mutation is more likely, the ovulatory wound is exposed to follicular estrogen in concentrations 10,000 times higher than the level of circulating levels (4). Estrogen metabolites have been found to be a direct source of DNA damage, resulting in mutation and enhanced carcinogenesis in kidney and mammary tissues (5). Potentially genotoxic estrogen metabolites may play a role in the risk of epithelial ovarian cancer.

When estrogen is catabolized, intermediates called catechol estrogens can be formed via two pathways: 2-hydroxylation (2-OH) and 4-hydroxylation (4-OH). Within the ovary, the CYP1A1 enzyme is the primary catalyst for the 2-OH pathway, whereas the CYP1B1 enzyme is the primary catalyst for 4-OH (6). These catechol estrogens are inactivated by three means of conjugation: methylation by catechol-o-methyltransferase (COMT), glucuronidation by UDP-glucuronosyltransferases (UGT), and sulfonation by sulfotransferases (SULT; ref. 6). If catechol estrogens are not eliminated by conjugation, they go on to form semiquinones, which can be oxidized to quinones. Under normal circumstances, catechol estrogens have a short half-life because they are quickly O-methylated by COMT into their metabolites, 2-methoxyestradiol (2-MeO-E2) and 4-methoxyestradiol. This half-life may be extended if conjugation capacity is reduced or if there is an excess catechol load (7).

Catechol estrogens, particularly those from the 4-OH pathway, may elicit direct and indirect mutagenic effects (5, 8). Quinones create both stable adducts, which remain covalently bonded to DNA until removed by repair mechanisms, and depurinating adducts, which are spontaneously released from DNA by destabilization of the glycosidic bond. The 4-OH pathway favors formation of depurinating adducts, whereas 2-OH favors formation of stable adducts (6). Oxidization of semiquinones to quinones forms free radicals, which can modify DNA via base oxidation, oxidize proteins affecting cellular function, and initiate lipid peroxidation. 4-OH produces much higher levels of free radicals most likely because COMT preferentially catalyzes methylation of 2-OH catechol estrogens (6).

Variation in the genes associated with estrogen metabolism has been of interest for hormonal carcinogenesis. Other studies have examined ovarian cancer risk with respect to these genes, but the results have been inconclusive (9-14). We conducted this population-based case-control study of epithelial ovarian cancer to assess the risk associated with the presence of variants of estrogen catabolism genes. There are 11 polymorphisms included in this analysis: CYP1A1 m1 (T→C change in 3′ flank; rs4646903), m2 or Ile⁴⁶²Val (rs1048943), m3 (T→C change in 3′ flank; rs4986883), and m4 or Thr⁴⁶¹Asn (rs1799814); CYP1B1 Arg⁴⁸Gly (rs10012), Ala¹¹⁹Ser (rs1056827), Val⁴³²Leu (rs1056836), and Asn⁴⁵³Ser (rs1800440); COMT Val¹⁵⁸Met (rs4680); UGT1A1*28 A(TA)_nTAA; and SULT1A1 Arg²¹³His (rs9282861). We also explored potential gene-gene and gene-environment interactions that could influence this risk.

Study subjects were enrolled in a multicenter, population-based ovarian cancer case-control study (15). Ovarian cancer cases were identified from metropolitan Atlanta, Seattle, and Detroit population-based tumor registries, which are operated as part of the National Cancer Institute's Surveillance, Epidemiology, and End Results program. Eligible individuals were English speaking, white or black women, and ages 35 to 54 years who were diagnosed with a borderline or invasive ovarian tumor between July 1994 and June 1998. Of 547 women eligible for the study, 406 (74.2%) were interviewed. After interview, 28 cases were excluded because they did not meet either of two eligibility criteria for controls: U.S. birth and no personal history of breast cancer (see below). Of the remaining 378 cases, we obtained peripheral blood leukocyte samples from 322 (85.2%). We further limited our analysis to the 310 (96.3%) women with epithelial tumors.

Controls were women who participated in the Women's Contraceptive and Reproductive Experiences (CARE) study, a multicenter, population-based, breast cancer study that was done contemporaneously with the ovarian cancer case-control study (16). CARE controls were identified through random digit telephone dialing using methods that have been described elsewhere (16). Eligible CARE controls were English speaking, U.S. born, white or black women ages 35 to 64 years at the reference date, and with no history of breast cancer. The CARE controls were frequency matched to CARE breast cancer cases by age, study center, and race. For cost and design reasons, the CARE study attempted to collect blood from all controls who had a first-degree relative with breast cancer and a randomly selected sample of those with no such history, with the goal of drawing blood for 30% of the interviewed controls. Blood was obtained from 75% of the controls identified for blood draw. These CARE subjects included 78% of the women who had a first-degree relative with breast cancer and 27% of the women without such a history.

From the overall CARE controls, the potential controls for our study were a subset of women ages 35 to 54 residing in metropolitan Atlanta, Seattle, and Detroit (n = 2,228). After interview of 1,828 (82%) of these women, those with bilateral oophorectomy (n = 176), a history of ovarian cancer before the reference date (n = 6), or uncertainty about whether they had at least one ovary (n = 9) were excluded, for a total of 1,637 potential controls. From these potential controls, our analysis was limited to the 577 (35.2%) from whom blood was collected.

Cases and controls completed detailed in-person interviews covering demographic, reproductive, contraceptive, medical, lifestyle, and family cancer history. Blood specimens were processed and buffy coat aliquots were frozen before DNA extraction and genotyping.

Genotype determination was carried out in Dr. Chu Chen's laboratory. The 10 single-nucleotide polymorphisms (SNP) included in the study were assayed using 5′-nuclease (Taqman) technology. The number of UGT1A1 “TA” microsatellite repeats was determined by first amplifying the “TA” repeat via PCR to create a series of clones from in-house laboratory samples. Isolated and purified cloned PCR products were then sequenced to determine the number of “TA” repeats. Mobility values (relative mobility units) of the clone standards were used in Genotyper version 3.7 software to create size bins. Peaks from case and control samples falling within these size bins were designated with repeat lengths identical to the clone standard. Specific primers, probes, and annealing temperatures are available on request. Negative controls (reagent mixture without DNA) and positive controls with known genotypes (homozygous wild-type, heterozygous, and homozygous variant) were included in each 96-well plate of samples. Positive controls for the UGT1A1 microsatellite repeats were cloned plasmids containing 5, 6, 7, or 8 repeats that had been confirmed by direct sequence analyses and two in-house DNA samples with known number of repeats. In addition to these negative and positive controls, 150 randomly selected DNA specimens from study subjects were assayed twice. DNA extraction and pre-PCR experimental steps versus post-PCR procedures were conducted in separate rooms.

SNP genotype frequencies were examined for Hardy-Weinberg equilibrium using the χ² statistic. Data were analyzed using unconditional logistic regression to calculate odds ratios (OR) as estimates of the relative risk of ovarian cancer associated with a given polymorphism. For individual polymorphisms, we used both dominant and codominant models excepting the polymorphic loci for which no individuals were homozygous for the variant genotype. For the UGT1A1 repeat polymorphism analysis, we excluded eight individuals with alleles that contained only five TA repeats [A(TA)₅TAA]. The reference group had two alleles with six TA repeats [A(TA)₆TAA]. The two comparison groups were (a) individuals with one A(TA)₆TAA allele and one allele with A(TA)₇TAA or A(TA)₈TAA and (b) individuals with two A(TA)₇TAA or A(TA)₈TAA alleles. We included the design variables of age, study center, and first-degree relative with breast cancer in all regression models. Due to selection methods resulting in different proportions of cases and controls in the design variable categories, we investigated the effect of an additional weighted offset term (the inverse of the likelihood of being selected) in the adjusted analysis. We found that the offset term as described in other studies using these data (17) did not appreciably alter our risk estimates and chose not to include it in the final adjusted analysis. We conducted separate analyses that either stratified by or adjusted for race. We also assessed possible confounding effects of education, parity, oral contraceptive use, tubal ligation, body mass index, age at menarche, and smoking. As adjustment for none of these variables appreciably altered risk estimates, we did not include them as covariates. All analyses were done using the STATA statistical package version 8 (STATA Corp., College Station, TX).

In addition to estimating ORs for individual polymorphisms, associations with CYP1A1 and CYP1B1 haplotypes were examined. To analyze haplotypes, we used HPlus software version 2.1 (Fred Hutchinson Cancer Research Center, Seattle, WA), which infers haplotypes based on expectation-maximization with a modified progressive ligation computational algorithm (18). Diplotype (haplotype pair) analysis was done using PHASE software version 2.0 (University of Washington, Seattle, WA), which, based on a Bayesian coalescence theory, generates all possible haplotype pairs and the probability of that pair for each individual (19). STATA software was used to calculate ORs for individuals from a clustered logistic regression analysis that was weighted by diplotype probabilities (20).

We examined associations with genetic factors according to defined tumor histology (mucinous, serous, endometroid, clear cell, and these last three combined into a “nonmucinous” group) and tumor behavior (invasive versus borderline). We also assessed the possibility of an interaction between genotype and smoking (never versus current/former), parity (nonparous versus parous), lifetime duration of oral contraceptive use (≤6 months versus >6 months use), and lifetime extent of ovulation (<235, 235-375, and >376 cycles). Calculation of lifetime ovulatory cycles was based on the following formula (21): [(age at last menstrual period before reference date − age at menarche) × 13] − {[(# months lactation + # months pregnant + # months oral contraceptive use) × 13] / 12}. Each potential modifier was examined separately under a dominant model controlling for age, study center, family history of breast cancer, and race. Models with and without interaction terms were compared using the likelihood ratio test. Potential environmental modifiers with CYP1A1 and CYP1B1 haplotypes were also examined using HPlus software (22).

Logistic regression models were used to explore combinations of COMT and SULT1A1 genotypes, COMT and UGT1A1 genotypes, and COMT genotypes with combinations of UGT1A1/SULT1A1 genotypes. Likelihood ratios were used to compare models with and without gene interactions. We assessed risk estimates of the CYP1A1 and CYP1B1 genotypes by COMT Val¹⁵⁸Met genotypes corresponding to the three levels of enzyme activity: high (Val/Val), moderate (Val/Met), and low (Met/Met). To further investigate potential gene-gene interactions, we tested the oligogenic model reported by Sellers et al. (9) within our Caucasian population. We assigned our individuals into the same risk groups as outlined by Sellers for all possible genotype combinations of CYP1A1 m2, CYP1B1 Val⁴³²Leu, COMT Val¹⁵⁸Met, and SULT1A1 Arg²¹³His polymorphisms and then determined ORs by risk group.

Cases and controls were similar in age (mean in cases, 44 years; mean in controls, 45 years). Cases were less likely to be black or to report a family history of breast cancer (Table 1). The distribution of the design variables across the three study centers was different among the cases and controls. Differences observed in race, reported family history of breast cancer, and study center were all expected consequences of study design and accounted for in the analysis. Risk estimates observed for epidemiologic risk factors with ovarian cancer, such as the decreased risk associated with pregnancy and oral contraceptive use, were similar to findings reported in past studies (23). Of the 310 cases, 188 (60.6%) had invasive tumors whereas 122 (39.4%) had borderline tumors. The histologic breakdown was as follows: 181 (58%) serous, 51 (16%) mucinous, 42 (14%) endometroid, 11 (4%) clear cell, and 25 (8%) other subtypes.

Genotyping was successful for 100% of the cases for all polymorphisms assessed. Nine (2%) controls did not have complete genotyping. Within the Caucasian controls, all SNPs included in this study were statistically consistent with the Hardy-Weinberg equilibrium. Within African-American controls, CYP1B1 Arg⁴⁸Gly showed a deviation from Hardy-Weinberg equilibrium (P = 0.004). CYP1B1 Ala¹¹⁹Ser, which is in strong linkage disequilibrium with CYP1B1 Arg⁴⁸Gly, also showed a deviation from Hardy-Weinberg equilibrium (P = 0.04) in this group. For both of these polymorphisms, there were more homozygous variants observed than expected by Hardy-Weinberg equilibrium (data not shown). Genotype frequencies among Caucasian and African-Americans were consistent with published genotype distributions for all SNPs under study (Table 2).⁴

Examining individual polymorphisms, we found that only the CYP1B1 Leu⁴³² allele was modestly associated with epithelial ovarian cancer, with Leu⁴³² carriers having a 50% higher risk than other women [95% confidence interval (95% CI) for the OR, 1.1-2.3; Table 2]. This association was present to a similar degree both in Caucasian and African-American women. Risk increased with the number of copies of the Leu⁴³² allele, with adjusted ORs of 1.4 (95% CI, 0.9-2.0) for Leu/Val genotypes and 1.9 (95% CI, 1.2-3.0) for Leu/Leu genotypes, among Caucasians and African-Americans combined.

Within inferred CYP1A1 haplotypes, comparison haplotypes usually contained only one variant and were found in low frequencies, which limited findings beyond those observed in the single SNP assessment (Table 3). The frequency of the reference CYP1A1 haplotype (“wild-type” m1, m2, m3, and m4; or T-A-T-C) was 86% among Caucasians and 69% among African-American controls. Observed ORs for haplotypes containing the m1 “C” allele variant (C-A-T-C) and the m4 Asn⁴⁶¹ allele (T-A-T-A) were very similar to OR observed for m1 “C” allele and Asn⁴⁶¹ carriers in the single SNP analyses. Carriers with the haplotype containing the m3 polymorphic “C” allele variant (T-A-C-C), which is found only in those of African descent, were not at altered risk.

Inferred CYP1B1 haplotypes allowed for greater examination of the potential association between different combinations of alleles at these four polymorphic loci with risk of ovarian cancer. Our control frequencies of both Caucasian and African-American inferred CYP1B1 haplotypes closely match published frequencies of similar populations (Table 4; refs. 24, 25). The reference haplotype, Arg⁴⁸-Ala¹¹⁹-Val⁴³²-Asn⁴⁵³, was the most common, with a frequency of >44% for the controls of both populations. Within Caucasians, the three remaining haplotypes with frequencies >5% all contained Leu⁴³². Among the combined populations of Caucasians and African-Americans, haplotypes that contained Leu⁴³² all showed a positive association with ovarian cancer, with adjusted ORs ranging from 1.3 to 1.5. Among African-Americans, one additional haplotype, Gly⁴⁸-Ser¹¹⁹-Val⁴³²-Asn⁴⁵³, was found in >25% of cases and controls and was not associated with disease occurrence.

In the CYP1B1 diplotype analyses, we designated women homozygous for Arg⁴⁸-Ala¹¹⁹-Val⁴³²-Asn⁴⁵³ as the referent group (Table 5). In an analysis of Caucasian and African-American populations combined, heterozygotes with one Arg⁴⁸-Ala¹¹⁹-Val⁴³²-Asn⁴⁵³ allele and one allele containing a haplotype that included Leu⁴³² all showed a tendency toward an increased risk of ovarian cancer with adjusted ORs ranging from 1.3 to 1.6. Diplotypes that contained Leu⁴³² in both haplotypes showed a somewhat stronger tendency toward increased risk of ovarian cancer, with ORs ranging from 1.6 to 2.1. The number of women in each diplotype category was fairly small, resulting in somewhat imprecise risk estimates.

There were no clear differences in risk estimates between borderline and invasive tumors for any of the polymorphisms examined. There was no association with epithelial ovarian cancer for CYP1A1, COMT Val¹⁵⁸Met, and UGT1A1 A(TA)_nTAA polymorphisms when assessed within subgroups of tumor histology (mucinous versus nonmucinous). The increase in risk of ovarian cancer for individuals carrying CYP1B1 Leu⁴³² was apparent only in nonmucinous histologic subtypes of epithelial ovarian cancer. For Leu⁴³² carriers, the adjusted ORs were 1.8 (95% CI, 1.2-2.6) for nonmucinous subtypes and 0.9 (95% CI, 0.4-1.8) for mucinous tumors. Similarly, all CYP1B1 haplotypes containing Leu⁴³² were associated with nonmucinous subtypes of ovarian cancer but not with mucinous tumors (data not shown).

We found no evidence of gene-environment interaction with parity, oral contraceptive use, or lifetime ovulatory cycles with any of the polymorphisms and haplotypes studied (data not shown). Although in the single polymorphism assessment there did not seem to be any difference in risk estimates for CYP1B1 Leu⁴³² carriers between smokers and never smokers, we found that the Arg⁴⁸-Ala¹¹⁹-Leu⁴³²-Asn⁴⁵³ haplotype (versus Arg⁴⁸-Ala¹¹⁹-Val⁴³²-Asn⁴⁵³) had adjusted ORs of 2.1 (95% CI, 1.3-3.5) in smokers and 1.0 (95% CI, 0.6-1.8) in nonsmokers (P_interaction = 0.05).

Combinations of conjugation gene polymorphisms [COMT Val¹⁵⁸Met, SULT1A1 Arg²¹³His, and UGT1A1 A(TA)_nTAA] did not reveal clear high- or low-risk genotype groups nor were there any observed gene-gene interactions (data not shown). Because there seemed to be no interaction within these genes and COMT is the dominant catechol estrogen inactivation enzyme among the three, we chose to stratify only COMT genotypes against activating genes (CYP1A1 and CYP1B1). When CYP1B1 genotypes were stratified across COMT genotypes, increased ORs associated with carriage of CYP1B1 Leu⁴³² were only seen in COMT Met¹⁵⁸ carriers (Table 6). Among COMT Met¹⁵⁸ homozygotes, carriers of CYP1B1 Leu⁴³² had a 2.6-fold increase in risk relative to CYP1B1 Val⁴³² homozygotes (95% CI, 1.1-5.9). For carriers of haplotypes containing Leu⁴³² (versus Arg⁴⁸-Ala¹¹⁹-Val⁴³²-Asn⁴⁵³), ORs ranged from 0.9 to 1.2 for individuals that were homozygous for the COMT Val allele whereas ORs ranged from 1.2 to 1.8 for individuals who had COMT genotypes Val/Met or Met/Met. When our cases and controls were assigned into the same risk groups outlined by Sellers (9), we found the following ORs for each risk group: “low risk” (43 cases, 84 controls), 1.1 (95% CI, 0.6-1.8); “intermediate low risk” (49 cases, 63 controls), 1.5 (95% CI, 0.9-2.6); “neutral” (62 cases, 111 controls), 1.0 (reference group); “intermediate high risk” (65 cases, 89 controls), 1.4 (95% CI, 0.9-2.3); and “high risk” (58 cases, 94 controls), 1.2 (95% CI, 0.7-1.9).

The CYP1A1 gene encodes the enzyme that is the primary catalyst for the 2-OH of estrogen. Genotypic variation resulting in a phenotype that increases 2-OH may decrease risk of carcinogenesis by two possible mechanisms: competitively reducing potentially genotoxic 4-OH intermediates and increasing 2-OH intermediates, which can be methylated to 2-MeO-E2, an antiangiogenic metabolite (26). Functional studies have shown increased enzyme activity with m1 “C,” m2 Val⁴⁶², and m3 “C” alleles and reduced activity with m4 Asn⁴⁶¹ allele (27-30). Similar to results of previous studies, we observed no association between the CYP1A1 m1, m2, and m4 polymorphisms and epithelial ovarian cancer (9-11). For the m4 polymorphism, there was a modest increased risk for Asn⁴⁶¹ carriers, but the association was well within the bounds of chance given no true association. We believe this is the first study that has examined the association between ovarian cancer and the CYP1A1 m3 polymorphism, which is found only in individuals of African descent. Although we were underpowered to make precise risk estimates, we did observe a tendency toward a decreased risk among the m3 variant “C” allele carriers among African-Americans in the single locus model. This relationship, however, was not reinforced by the multiloci haplotype model.

We postulated that genetic variants within CYP1B1, the gene encoding the primary catalyst for 4-OH of estrogen, would alter levels of 4-OH catechol estrogens and thus increase or decrease the risk of ovarian cancer accordingly. We observed a positive association with ovarian cancer for CYP1B1 Leu⁴³² carriers. The frequencies of the CYP1B1 haplotypes found in our population led us to designate the Arg⁴⁸-Ala¹¹⁹-Val⁴³²-Asn⁴⁵³ as the referent haplotype. Because this haplotype contained Val⁴³², we used Val⁴³² as our referent allele in the single polymorphism analysis. Other studies have designated their referent allele as Leu⁴³²; thus, comparisons with prior studies should be made carefully. For comparison purposes, we can state that, for Val⁴³² carriers, we found a significant decreased risk of ovarian cancer compared with Leu⁴³² homozygotes (OR, 0.6; 95% CI, 0.4-0.6). This is in contrast to the findings in a smaller study (129 cases and 144 controls) conducted by Goodman et al. (10) that observed an increased risk of epithelial ovarian cancer among Val⁴³² carriers, with an adjusted ORs of 1.8 (95% CI, 1.0-3.3) for Val/Leu carriers and 3.8 (95% CI, 1.2-11.4) for Val/Val carriers. The Goodman study was composed of 45% Asians, 27% Native Hawaiians/Samoans, and 28% Caucasians, whereas our study was 82% Caucasian and 18% African-American. Because haplotype frequencies and linkage disequilibrium patterns vary significantly across populations, the difference in our findings could be attributed to the distinctly different populations of the two studies. Whereas both studies adjusted for race, the Goodman study did not present risk estimates by race. In addition, one or more of the results in either study could be false positives. In a larger population-based study (503 cases and 609 controls) with a more comparable population (92% Caucasian and 8% African-American) done by Sellers et al. (9), no association with the CYP1B1 Arg⁴⁸Gly, Ala¹¹⁹Ser or Leu⁴³²Val polymorphisms and ovarian cancer risk was seen in single polymorphism models for either racial group. Another population-based study of 223 cases and 280 controls done in Italy also found no association between the CYP1B1 Val⁴³²Leu polymorphism and epithelial ovarian cancer (12).

To our knowledge, this was the first study to assess inferred CYP1B1 haplotypes in relation to ovarian cancer. If multiple polymorphisms in a single gene affect enzyme function in an epistatic fashion, such an analysis may have a relatively greater ability to identify an association with cancer risk. Prior in vitro enzyme function studies have shown differences in estrogen 4-OH rates across the CYP1B1 haplotypes with conflicting results (25, 31-34). Aklillu et al. (35), using a yeast expression system, and Hanna et al. (31), using bacterial expression systems, both showed higher catalytic efficiency for 4-OH of estradiol with Arg⁴⁸-Ala¹¹⁹-Leu⁴³²-Asn⁴⁵³ and Gly⁴⁸-Ser¹¹⁹-Leu⁴³²-Asn⁴⁵³ haplotypes compared with the reference Arg⁴⁸-Ala¹¹⁹-Val⁴³²-Asn⁴⁵³ haplotype. If Leu⁴³² haplotypes actually increase levels of 4-OH catechol estrogens, then our observations of a modest increase in risk for women with Leu⁴³² haplotypes would be predicted. However, Shimada et al. (33) and Li et al. (34) both found decreased catalytic efficiency for Arg⁴⁸-Ala¹¹⁹-Leu⁴³²-Asn⁴⁵³ versus Arg⁴⁸-Ala¹¹⁹-Val⁴³²-Asn⁴⁵³ haplotypes. Future studies could benefit from better understanding of the phenotypic effect of these haplotypes on 4-OH of estrogen in vivo.

The COMT genotypes included in this study have been shown to express three levels of enzyme function: high activity Val/Val, intermediate activity Val/Met, and low activity Met/Met (36). The number of UGT1A1 A(TA)_nTAA repeats seems to be inversely associated with transcription (37). The His²¹³ allele (SULT1A1*2) is associated with ∼2-fold lower enzyme activity and is less thermostable (38). Carriers of the higher enzymatic activity allele of COMT, UGT1A1, and SULT1A1 [e.g., COMT Val¹⁵⁸, UGT1A1 A(TA)₆TAA, and SULT1A1 Arg²¹³] could, in theory, have lower levels of 4-OH catechol estrogens and decreased ovarian cancer risk. In addition, the COMT Val¹⁵⁸ genotype could, hypothetically, decrease ovarian cancer risk by increasing levels of the antiangiogenic 2-MeO-E2 conjugates. The lack of association seen between the COMT Val¹⁵⁸Met, SULT1A1 Arg²¹³His, and UGT1A1 A(TA)_nTAA polymorphisms and ovarian cancer is consistent with the findings of other studies (9, 10, 13, 14).

Lack of association between the polymorphisms included in this study and ovarian cancer risk may be because of incomplete measurement of genetic variation. For SULT1A1, it has recently been reported that polymorphisms within the 5′-flanking region affect enzyme activity (39). Furthermore, these promoter polymorphisms were in linkage disequilibrium with the SULT1A1 Arg²¹³His polymorphism, which failed to predict enzyme activity after promoter polymorphisms had been accounted for. For all of the genes, a haplotype approach using tag SNPs identified through recent resequencing studies may be a better alternative for capturing the complex phenotypic variability of these genes (40). For example, results from a functional analysis of UGT1A1 haplotypes within a Japanese population have suggested that there may be novel polymorphisms that predict phenotype in addition to UGT1A1 A(TA)_nTAA genotypes (41).

Both CYP1A1 and CYP1B1 enzymes are induced and metabolically activated by numerous polycyclic aromatic hydrocarbons and aryl amines, whereas SULT1A1 and UGT1A1 can activate and detoxify some hydroxylated metabolites of these same carcinogens (42-44). It has been proposed that polymorphisms that alter levels of estrogen metabolites may also alter levels of carcinogen metabolites, but functional evidence supporting this hypothesis is incomplete. Because of these relationships between these genes and carcinogens, exposure to tobacco smoke may influence risk estimates. For example, an increased association observed with a CYP1B1 polymorphism may be especially apparent in smokers because of the increased induction via benzo(a)pyrene found in tobacco and/or the metabolic activation of benzo(a)pyrene as a carcinogen. Potential interactions between smoking and CYP1A1 with risk of ovarian cancer have been reported for both the m1 (10) and m2 (9) polymorphisms, but these finding were not replicated by our study or other studies that included smoking in their analysis (9-11). Thus far, neither our study nor any other epidemiologic studies have reported evidence of any appreciable interaction between CYP1B1 Arg⁴⁸Gly, Ala¹¹⁹Ser, Val⁴³²Leu, Asn⁴⁵³Ser; UGT1A1 A(TA)_nTAA, or SULT1A1 Arg²¹³His polymorphisms and smoking on risk of ovarian cancer (9, 10).

Within our limited set of polymorphisms, we did see some evidence of a joint influence between CYP1B1 Leu⁴³² and COMT Met¹⁵⁸ alleles. However, the association observed was opposite in direction to that observed by Goodman et al. (10), who reported significantly higher ORs for CYP1B1 Val⁴³² carriers in COMT Met¹⁵⁸ carriers. We also conducted an oligogenic analysis of CYP1A1 Ile⁴⁶²Val, CYP1B1 Val⁴³²Leu, COMT Val¹⁵⁸Met, and SULT1A1 Arg²¹³His polymorphisms described by Sellers et al. (9) but observed no evidence of increasing ORs by increasing model-based risk group.

Estrogen catabolism is complex and tissue specific; we do not yet have a complete picture of the pathway as it relates to the ovary and ovarian carcinogenesis (7). It may be that the CYP1A1 enzyme plays a minor role in estrogen catabolism within the ovarian epithelium or that the overall high capacity of COMT to detoxify catechol estrogens nullifies the effect of observed differences in enzyme activity phenotypes (45). Tight regulation of activation and detoxification enzymes could possibly override the phenotypic effects of a single gene on estrogen metabolite levels (42). Among conjugation enzymes, which share affinity for a wide array of substrates, competition could factor into how fast catechol estrogens are conjugated and into what form. It has been shown that UGT1A1 preferentially catalyzes glucuronidation of 2-OH catechol estrogens; perhaps, a low activity UGT1A1 genotype would result in more 2-MeO-E2 via less competition with COMT (46, 47). SULT1A1 also catalyzes sulfonation of 2-MeO-E2, but the effects of SULT1A1 genetic variants on modulating the levels of 2-MeO-E2 are unknown (48). These intricacies of estrogen catabolism suggest that hypotheses based on individual enzyme effects may be overly simplistic.

One of the strengths of this study was inclusion of genetic polymorphisms that have been well characterized and occur at frequencies high enough to have measurable attributable risks. Assessment of functional haplotypes allowed corroboration of findings observed in single loci models and led us to a more thorough picture of the role of CYP1B1 polymorphisms in ovarian carcinogenesis. Another strength of this study was the relatively young age of the cases and controls. This could possibly increase the likelihood of observing an association with genetic variants because (a) in most subjects the ovarian epithelium is still being exposed to the potential genotoxic properties of high levels of estrogen via ovulation and (b) genetic variants that increase the risk of other forms of cancer do so at a relatively greater degree at younger ages. Although we did have a large enough sample size to examine associations with individual polymorphisms, our study was underpowered to produce precise risk estimates for histologic subgroups, diplotype groups, gene-environment interactions, and gene-gene interactions.

We observed a moderate increase in risk of ovarian cancer among carriers of CYP1B1 Leu⁴³² in single SNP, haplotype, and diplotype models. However, the results of the four studies that have examined this question are inconsistent, and at present, no conclusions can be drawn. Inclusion of more polymorphic loci and haplotypes may improve measurement of functional variation found within these genes. In addition, a larger study could better explore the potential interaction with genes encoding the conjugation enzymes and possible differences in risk estimates for mucinous and nonmucinous histologic subtypes.