Associations between Polymorphisms in Glucuronidation and Sulfation Enzymes and Mammographic Breast Density in Premenopausal Women in the United States

Breast cancer is the most common cancer among women, with a lifetime probability of 1 in 8 in the United States (1). Aside from family history, the most well-established risk factors for breast cancer are those associated with hormonal and reproductive factors that result in greater lifetime exposure to estrogens and androgens (2), such as an extended reproductive life (resulting from an early age at menarche and late age at menopause), late age at first full-term pregnancy, and nulliparity. These observations, along with the finding that higher plasma concentrations of total and free estradiol (E₂) in the early follicular phase and total and free testosterone in both menstrual cycle phases are associated with an increased risk of breast cancer in premenopausal women (3), suggest that cumulative estrogen exposure is sufficient to alter breast cancer risk later in life.

Breast density may reflect lifelong hormone exposure and potentially could be used as a biomarker for breast cancer risk. Several studies have shown an inverse association between parity and mammographic density (4). Nulliparous women and women with a later age at first birth have higher estrogen levels than parous women and women with a younger age at first birth, respectively (5). Nulliparous women have denser breast tissue than parous women, and density decreases with increasing number of children (6). Moreover, among parous women, later age at first birth and fewer live births have been associated with a higher proportion of dense breast tissue and greater risk for breast cancer (6).

Glucuronidation catalyzed by UDP-glucuronosyltransferases (UGT; ref. 7) and sulfation catalyzed by sulfotransferases (SULT; ref. 8) are two pathways through which sex hormones are metabolized to less active compounds. Polymorphisms that alter enzyme function have been identified in UGT and SULT genes, and these may ultimately affect the clearance of, and therefore exposure to, endogenous and exogenous estrogens and androgens. Thus, individual variation in estrogen and androgen metabolism resulting from common genetic polymorphisms could be a risk factor for hormone-dependent diseases and may serve as genetic markers of differences in lifetime hormonal exposure. One approach to determine whether these hormone-metabolizing gene polymorphisms affect cumulative exposure to estrogens and androgens throughout life is to assess their relationship to mammographic breast density.

In a study population of premenopausal women, we assessed the associations of selected functional polymorphisms in the UGT1A and UGT2B gene families and SULT1A1 with mammographic breast density. We hypothesized that alleles with increased conjugating activity resulting in increased clearance of endogenous hormones and lower circulating hormone concentrations [UGT1A3(R11), UGT1A3(A47), UGT2B7(Y268), and UGT2B15(D85Y)] would be associated with decreased breast density (via lower lifelong hormone exposure). Likewise, alleles with decreased conjugating activity resulting in decreased clearance of endogenous hormones and higher circulating hormone concentrations [UGT1A1(TA7), UGT2B17(null), and SULT1A1(H213)] would be associated with increased breast density (via higher lifelong hormone exposure).

As described in detail elsewhere (9), women were recruited from within Group Health, a large integrated health plan in Washington State. Premenopausal women ages 40 to 45 years who have undergone a screening mammogram in the previous 10 months and who were not taking exogenous hormones were identified from the Group Health Breast Cancer Screening Program and recruited based on the Breast Imaging Reporting and Data System density score assigned to their most recent screening mammogram.

A total of 203 women attended a study clinic visit. At the time of consent, we asked each participant to indicate whether she was willing to have her stored biological samples used for future studies. A total of 189 (93%) study participants checked “yes” to this question, of whom 176 (93%) had a buffy coat available for genotyping. The major reason for those who did not have a buffy coat available was a problematic blood draw that resulted in no blood sample. We excluded one participant with a mammogram that was too dark to read, leaving a total of 175 women in our analyses. All study procedures were approved by the institutional review boards of the Fred Hutchinson Cancer Research Center and Group Health, and all study participants provided written informed consent.

Each participant's most recent routine Group Health X-ray screening mammogram before her study visit was digitized using a Lumysis 85 scanner. A single reader interpreted films using Cumulus for percent density, dense area size, and total area size as described in detail elsewhere (10).

DNA for genotyping was extracted from the buffy coat fraction using the Qiagen blood kit. The concentration and purity were determined by spectroscopy at 260 and 280 nm. A total of 11 polymorphisms were genotyped (UGT1A1(TA6/TA7), UGT1A3(W11R), UGT1A3(V47A), UGT2B4(D458E), UGT2B7(H268Y), UGT2B15(D85Y), UGT2B17(null/not null), SULT1A1(R213H), SULT1E1(I169A>G), SULT1E1[I1(-73)G>C], and SULT1E1[I5(-10)C>G]) using a variety of PCR-based methods, including size-dependent separation, restriction fragment length polymorphism, sequencing, and fluorescent allelic discrimination (TaqMan). For SULT1E1, we selected three single nucleotide polymorphisms (SULT1E1(I169A>G), SULT1E1[I1(-73)G>C], and SULT1E1[I5(-10)C>G]) that were found by Adjei et al., which were used to distinguish the most common haplotypes (>5% allele frequency) in a Caucasian American population (11). Primers and probes for each polymorphism are shown in Table 1. Negative controls (no DNA template) and positive controls (cell line DNA and/or DNA samples of known genotypes) were run on every plate. The reliability/reproducibility of the genotyping assays was assessed by randomly selecting and re-assaying 5% of the samples for each run; no discrepancies were observed between initial and duplicate assays. Genotype calling was done both by machine and by one reader. If there was a discrepancy between the two calls, then an independent reader was brought in to resolve the difference. Samples for which we obtained an ambiguous result or did not obtain a genotyping result were repeated. Definitive results for repeated samples were obtained on the second attempt. Thus, we obtained genotyping results for the polymorphisms for all study participants.

Genotyping of the UGT1A1(TA6/TA7) (rs8175347) polymorphism was done as described previously (12). The fragments were analyzed using an ABI PRISM 3100 Genetic Analyzer and Genotyper 2.5 software (Applied Biosystems).

Genotyping of the two-residue amino acid substitutions W11R (rs3821242) and V47A (rs6431625) of UGT1A3 involved two steps (13). First, PCRs were done to amplify the region of interest. Second, the PCR amplicons were sequenced. The sequence data were analyzed using the Sequencher 4.1 (Gene Codes) software.

A difference of one nucleotide in UGT2B4 leads to a single amino acid change of aspartic acid to glutamic acid at position 458. PCR was used to amplify the fragment containing D458E (rs13119049) as described previously (14). We then performed restriction fragment length polymorphism using TaqI restriction enzyme on the PCR product and separated the fragments on a 2% NuSieve gel. The expected fragment sizes for the D458 allele were 232 and 32 bp, and the expected fragment size for the E458 allele was 264 bp.

We genotyped the polymorphisms in UGT2B7 (rs7439366), UGT2B15 (rs1902023), SULT1A1 (rs9282861; ref. 15), and SULT1E1 (rs3775768, rs4149530, and rs1220702; ref. 11) and a deletion in UGT2B17 (16) using TaqMan. Data were analyzed with SDS software (Applied Biosystems), and genotype calls were based on the level of fluorescence emission from the reporter dye. With the exception of UGT2B7 and UGT2B15, assays for the selected genotypes have been validated previously. Validation of the UGT2B7 and UGT2B15 assays was done using DNA samples (n = 20-100) from other studies that have been genotyped previously by restriction fragment length polymorphism and sequencing with no discrepancies between the results obtained from the two assays.

We assessed the genotypes at each locus for consistency with the Hardy-Weinberg equilibrium using a χ² test. Measures of central tendency and categorical distributions were calculated to describe the characteristics of the study population, and initial assessments were done using non-model-based approaches including simple means and t tests.

Strong linkage disequilibrium was observed between UGT1A3(W11R) and UGT1A3(V47A) polymorphisms (D′ = 0.98) and between UGT1A1(TA7) and UGT1A3(W11R) (D′ = −0.96) and UGT1A1(TA7) and UGT1A3(V47A) (D′ = 0.98) polymorphisms. Because the functional effect of UGT1A1(TA7) decreased UGT1A1 gene expression and therefore decreased glucuronidation and UGT1A3(W11R) and UGT1A3(V47A) results in increased glucuronidation, we inferred haplotypes involving these three loci for our study population. For each gene (UGT1A and SULT1E1), we performed a global test of all the haplotypes versus no haplotypes using a likelihood ratio test for mammographic breast density. We then fit a generalized linear model with additive haplotype effects under Hardy-Weinberg equilibrium to test for an association between each of the inferred UGT1A1-UGT1A3 and SULT1E1 haplotypes and mammographic breast density (17).

To extend results from Sillanpää et al., who reported an inverse association between SULT1A1(H213) alleles and breast cancer among premenopausal women with high parity only (18), we also explored the effect of number of pregnancies (0, 1-2, and ≥3) on the association between SULT1A1 genotypes and mammographic breast density. We used the SULT1A1(R213/R213) genotype as the reference group for each live birth category (0, 1-2, and ≥3) to compute the mean percent density values for the SULT1A1(R213/H213) and SULT1A1(H213/H213) genotypes. We performed a test for interaction between number of live births and SULT1A1 genotype using a likelihood ratio test, which tests the full model (contains the interaction) against the reduced model (no interaction).

Genotypes were coded on an ordinal scale [homozygous wild-type (wt/wt) = 0, heterozygous (wt/v) = 1, and homozygous variant (v/v) = 2] to model allele dosage effects with the wt/wt genotype as the reference category. Genotype was also examined by using a dichotomous variable to indicate whether the participant was a carrier (wt/v or v/v) of the variant allele if no gene dosage effect was observed. Adjusted generalized linear regression models were fit to determine mean percent breast density by genotype, and a test for trend was conducted between the ordinal genotype measures and breast density measures using adjusted linear regression.

Previous studies examining race/ethnicity and UGT and SULT genotypes have reported race/ethnicity to be associated with genotypes (14, 19). For example, Lampe et al. (12) showed that both allele and genotype frequencies of UGT1A1(TA6/TA7) varied by race (White versus Asian). Race/ethnicity has also been shown to be associated with mammographic breast density with the highest mean percent density reported for African American women and the lowest reported for Japanese women (20). Thus, ethnicity (categorized: Asian, White, Other) was included in our final models.

Mean percent breast density and 95% confidence intervals are presented. Data were analyzed using Stata/SE (version 9.0; StataCorp), and haplotypes were inferred using Hapstat (software for the statistical analysis of haplotype-disease association; Copyright © 2006-2008 Tammy Bailey, Danyu Lin, and the University of North Carolina at Chapel Hill). A two-sided P < 0.05 was considered statistically significant.

The mean (SD) age of the study participants was 42.4 (1.4) years, and the majority had one or more live births, had a history of hormone use (e.g., oral contraceptives, hormone patches, hormone injections, hormone implants, and intrauterine devices containing progesterone), and were nonsmokers (never or former), White, and highly educated (Table 2). The haplotype frequencies of UGT1A and SULT1E1, which satisfied Hardy-Weinberg equilibrium, and all genotype frequencies with nonsignificant χ² tests at P > 0.05 are presented in Table 3.

After adjusting for ethnicity, women with UGT1A3(W11/R11) and UGT1A3(R11/R11) genotypes had lower mean percent mammographic breast density compared with women with the wild-type genotype [UGT1A3(W11/W11); 35.1% and 31.6%, respectively, versus 40.9%; P_trend = 0.04; Table 3]. There was a nonstatistically significant inverse association between UGT1A1(TA7)-UGT1A3(R11)-UGT1A3(A47) haplotype and mammographic breast density compared with the more common UGT1A1(TA6)-UGT1A3(W11R)-UGT1A3(V47A) haplotype in this population (35.7% versus 40.6%, respectively; P = 0.07; Table 3). Mean mammographic breast densities for women with the SULT1A1(R213/H213) and SULT1A1(H213/H213) genotypes were lower compared with women homozygous for SULT1A1(R213) (34.9% and 25.7% versus 41.8%, respectively; P_trend = 0.001; Table 3). We also observed nonsignificant inverse associations between both the TA7 allele of UGT1A1 and the Y268 allele of UGT2B7 and mammographic density and a nonsignificant positive association between the E458 allele of UGT2B4 and mammographic density (Table 3).

For the polymorphisms for which we did not observe a dose-response relationship, a borderline statistically significant inverse association was shown between percent mammographic breast density and UGT1A3(V47A), with carriers of the A47 allele having a 6.9% lower percent density compared with noncarriers [percent density (95% confidence interval), 33.6% (29.3-37.8%) for carriers and 40.4% (35.0-45.8%) for noncarriers; P = 0.050; data not shown]. No statistically significant associations between percent mammographic breast density and either UGT2B15(D85Y) or UGT2B17(deletion) were observed.

Of the three SULT1E1 single nucleotide polymorphisms genotyped, one (rs1220702) did not occur as frequently relative to the other two (minor allele frequency = 11%) and did not contribute to the delineation of any common haplotype. Among the two remaining SULT1E1 single nucleotide polymorphisms, we identified three haplotypes in our study population. No significant association was shown between mammographic breast density and the SULT1E1 haplotypes. The likelihood ratio test comparing the model with the haplotype effects of SULT1E1 (full model) to the model with no haplotypes (reduced model) showed that the model without the haplotypes provided an adequate fit to the data.

We assessed the interaction of SULT1A1 genotypes and number of live births on mammographic breast density. Mammographic density decreased with increasing number of H213 alleles within each category of live births (0, 1-2, and ≥3 live births). The reduction in percent density between the R213/H213 and H213/H213 genotypes and the reference genotype (R213/R213) was more pronounced in women who had no live births [absolute differences: −8.0% (R213/H213) and −18.9% (H213/H213); P_trend = 0.049] compared with women with 1-2 and ≥3 live births (Table 4). However, no statistically significant interaction between SULT1A1 genotypes and number of live births was shown (χ² = 1.48; P = 0.83).

In this well-characterized population of healthy, premenopausal women, we assessed the associations between percent mammographic breast density and polymorphisms in the UGT1A, UGT2B, SULT1A1, and SULT1E1 genes. We observed a strong significant inverse association between percent mammographic breast density and SULT1A1(H213/H213) carriership. This finding is counterintuitive given that the protein coded by SULT1A1(H213/H213) has been shown to be associated with lower enzyme thermostability, lower enzyme activity, and lower capacity to sulfate E₂ and catechol estrogens compared with the wild-type [SULT1A1(R213/R213); ref. 21]. Our results could, however, reflect the role of catechol estrogens (2- and 16α-hydroxyestrone). 2-Hydroxyestrone is conjugated to anticarcinogenic methoxylated metabolites (e.g., 2-methoxyestrone and 2-methoxyestradiol) and is hypothesized to protect against breast cancer (22). In contrast, 16α-hydroxyestrone is a potent estrogen, has been shown to form covalent bonds with estrogen receptors, and appears to be genotoxic (23). Findings reported by others (24, 25) suggest that these catechol estrogen metabolites may be involved in the etiology of breast cancer and this effect may be mediated, in part, by percent breast density. Hui et al. (26) recently showed that human SULT enzymes are capable of sulfating catechol estrogens and methoxyestrogens in breast cancer cells and human mammary epithelial cells. Therefore, it is possible that the lower conjugating activity of SULT1A1(H213/H213) might increase the availability of estrogens for conversion to catechol estrogens and subsequent conjugation to methoxyestrogens. In addition, given that SULT1A1 has been shown to be an efficient and selective catalyst of 2-methoxyestradiol sulfation (27), it is possible that SULT1A1 could modify the effects of 2-methoxyestradiol. Women with low activity SULT1A1(H213/H213) genotype could have higher levels of the unconjugated form of 2-methoxyestradiol. Consequently, the potential protective effects of this metabolite may be prolonged in women with low sulfation capacity compared with women with high sulfation activity.

Based on our findings of lower mammographic breast density with increasing numbers of the H213 allele, we might expect premenopausal women with the H213 allele to have a decreased risk of breast cancer. Two studies have examined the association between SULT1A1 genotypes and breast cancer risk in premenopausal women and neither found a significant genotype effect on overall breast cancer risk (18, 28). However, Sillanpää et al. showed an inverse association of this allele with breast cancer in premenopausal women with high parity, suggesting a modifying effect of full-term pregnancies (18). In contrast, our findings suggested that the inverse association between the H213 allele and breast density was most pronounced in women with no pregnancies, and the inverse trend became weaker as the number of live births increased; however, our finding was not statistically significant and our study was underpowered for the analysis of this interaction (power = 0.093). Nonetheless, if confirmed in other studies, our results suggest that the effect of this polymorphism may be strongest when the substrate is highest, given that low parity results in higher lifelong estrogen exposure, whereas higher parity results in lower lifelong estrogen exposure.

We did not observe significant associations between mammographic breast density and polymorphisms in UGT2B15 and UGT2B17, although there was a nonsignificant inverse trend with the UGT2B15(D85Y) allele in the hypothesized direction. We also did not observe an association between mammographic breast density and common SULT1E1 haplotypes. It is possible that the single nucleotide polymorphisms identified by Adjei et al. (11) may not have any functional consequences for steroid hormone sulfation or that there is a true effect but we did not have the statistical power to detect it in our study.

In vitro studies of the TA7 allele have reported a 30% reduction in UGT1A1 gene transcription and decreased UGT1A1 gene expression (29-31), and individuals deficient in UGT1A1 due to a deletion encompassing the promoter and first exon exhibit a 70% decrease in the glucuronidation of E₂ (32). Thus, we hypothesized that carriers of the TA7 allele should have higher lifelong estrogen exposure and higher risks of estrogen-related conditions, including increased mammographic density. Our finding and that of Haiman et al. (33) that percent breast density was 8% and 16% lower, respectively, in premenopausal women with the UGT1A1(TA7/TA7) genotype compared with those with the UGT1A1(TA6/TA6) genotype appear to contradict this hypothesis. However, strong linkage disequilibrium exists between the UGT1A1(TA7) allele and apparently functional polymorphisms in multiple other UGT1A family genes [including UGT1A6 (12, 34), UGT1A3 (13), and UGT1A7 (35)]. To date, no studies have looked at associations between estrogen glucuronidation and UGT1A haplotypes. Although not statistically significant, we found that mean percent breast density was lowest for women who carry at least one copy of the UGT1A1(TA6)-UGT1A3(R11)-UGT1A3(V47A) haplotype.

This is consistent with our observation of a statistically significant inverse association between mammographic breast density and the UGT1A3(R11) allele. The difference in breast density that we observed according to UGT1A3 genotype is consistent with the greater clearance of estrogen expected in women with the R11 allele. However, the R11 allele has little effect on mammographic density in combination with the UGT1A3(A47) allele and only a small effect when inherited with the UGT1A1(TA7) allele. These findings suggest that, at the UGT1A locus, the coinheritance of UGT1A3(R11) and UGT1A3(V47A), or other variation on the haplotype containing these alleles, has the strongest influence on mammographic density. Thus, the reduced mammographic density that we and Haiman et al. have observed associated with the UGT1A1(TA7) allele appears to be due to the UGT1A haplotype of variants located in the UGT1A1 and UGT1A3 genes. It remains to be determined whether these alleles or others that are in linkage disequilibrium with the TA7 allele have enhanced glucuronidation activities toward estrogens.

Our results for the UGT2B4(E458) allele, although not statistically significant, suggest that it may be associated with a somewhat higher breast density. We also observed a nonsignificant inverse association between mammographic breast density and the UGT2B7(Y268) allele. This enzyme, expressed in breast tissue, has been shown to glucuronidate catechol estrogens, particularly 4-hydroxyestrone, which is a major metabolite of E₂ and has been shown to be carcinogenic in breast and uterine tissues (36). Thibaudeau et al. (37) evaluated the effects of the UGT2B7(Y268) allele on the formation of 4-hydroxyestrone and 4-hydroxyestradiol glucuronides in human embryonic kidney cells and showed that the Y268 allele was associated with a significant 2-fold increase in clearance of these glucuronides compared with the wild-type. Because higher circulating E₁ and E₂ and free E₂ concentrations have been shown to be associated with higher percent mammographic density (38), we hypothesized that the UGT2B7(Y268) allele would be associated with lower percent density. Our results, although not statistically significant, as well as those from previous experimental studies (37, 39) support this hypothesis.

There are several strengths and limitations of our study. Premenopausal women tend to have high breast density, so study participants were sampled based on a Breast Imaging Reporting and Data System classification score, which allowed us to obtain a wide range of breast densities (10). However, because most women were White and of high socioeconomic status and all were members of a health plan, our findings may be generalizable only to similar populations of women. Another limitation is our small sample size, which restricted our ability to examine rare genotypes or interactions between genotypes. We may not have had adequate power to detect differences in mammographic density measures for most of the genotypes. Post hoc power calculations, based on the distribution of breast density measures observed and sample sizes obtained in this study, showed that we had <80% power to detect differences between genotypes for seven of the eight genes in our study. Given that many comparisons were made, it is possible that some of the statistically significant findings may have occurred by chance. Finally, there may have been bias due to nonparticipation, and although it is conceivable that an association might exist between mammographic breast density and willingness to participate, it is unlikely that the genetic polymorphisms would be differentially associated with those who participated and those who did not.

Measuring all the polymorphisms involved in steroid hormone metabolism was beyond the scope of this study. Few studies have evaluated relationships between the UGT and SULT polymorphisms and hormonal biomarkers in healthy, premenopausal women, and results from our study can be used as important preliminary data for determining approaches for future, larger-scale molecular epidemiologic studies that aim to capture all the relevant sex hormone metabolizing enzymes.

In summary, in this population of premenopausal women, mammographic breast density was significantly associated with polymorphisms in SULT1A1. Given that only one other study has examined the association between a UGT polymorphism and a biomarker of risk of hormone-dependent conditions in premenopausal women (33), larger studies examining the role of polymorphisms in steroid hormone pathway genes as predictive markers of mammographic breast density are needed. If the discovery of susceptibility genes is successful, the identification of high-risk women for prevention efforts by the use of multigenic models of breast cancer susceptibility may be possible (40).

No potential conflicts of interest were disclosed.

We thank Kelly Ehrlich, Kathy Plant, and the Group Health Research Institute for screening interviews, clinic visits, and study coordination and all of the study participants.

Grant Support: NIH grants R03CA121872 and R01CA97366 and the Fred Hutchinson Cancer Research Center interdisciplinary fellowship.

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.