A Comprehensive Examination of CYP19 Variation and Breast Density

Introduction

Mammographic density is one of the strongest risk factors for breast cancer. Numerous lines of evidence suggest that mammographic density is influenced by estrogen exposures (1-5). CYP19 produces aromatase, the enzyme responsible for conversion of androgens to estrogens in all tissues, including the breast. This conversion of androgens to estrogens is important at all ages, especially after menopause when estrogen production takes place primarily in peripheral, nonovarian tissue (6). We report the results of comprehensive analyses of CYP19 polymorphisms and percentage mammographic density (PD) in 550 healthy Caucasian women. We examined whether genetic variation, or haplotypes, in CYP19 are associated with PD.

Materials and Methods

Study subjects were women without a history of cancer visiting Mayo Department of Internal Medicine for a medical exam. These women were controls for a breast cancer case-control study (described in ref. 7) who provided informed consent, a written questionnaire, and a blood sample. Postmenopausal status was defined as having no menstrual period for 12 months or having a hysterectomy or oophorectomy. Hormone therapy was categorized as “currently taking,” “taken in the past,” or “never used.”

Mammograms (craniocaudal or top-down view of one breast) closest to the questionnaire date were digitized using a Kodak Lumiscan 85 scanner with 12-bit grayscale depth as described elsewhere (8). PD (dense area divided by total area × 100) was estimated using a computer-assisted thresholding program that has been routinely used in several studies (9). All images were read by a programmer, trained in the estimation of density, who has shown an intraobserver variation of less than 10%.

Polymorphisms common in Caucasian samples were selected from all variants identified through gene resequencing using two selection methods (10). The first method, which aims to tag common haplotypes (11), selected 12 variants (0.02 minimum minor allele frequency, 0.01 minimum haplotype frequency, 90% minimum haplotype R²). The other method, which aims to tag common single nucleotide polymorphisms (SNP) correlated by linkage disequilibrium (12), also selected 12 SNPs (0.05 minor allele frequency, 80% correlation within bins). Six SNPs were chosen by both methods. Haplotype blocks within the gene were also identified (13). Genotyping was done as described previously (7). Genotyping quality was assessed by estimation of Hardy-Weinberg equilibrium and inclusion of 13 blinded duplicate samples. Most duplicates were concordant; however, the discrepancies that occurred resulted in at most two samples being discordant for any given SNP.

Single-variant analyses were done using multiple regression, and genotypes were modeled as having an additive relationship with PD. Analyses estimated the differences in PD associated with each additional copy of the variant allele. All models were adjusted for age and geographic region. Additional covariates that were significantly associated with PD in this sample were also included. PD values were skewed; therefore, they were square root transformed before analyses.

The associations between haplotypes and PD were assessed using a global test of association, followed by estimation of the linkage phase and the association of individual haplotypes with PD (14). Individual haplotype associations were considered statistically significant only if the global haplotype test was also significant.

Analyses were done in the SAS (SAS Institute, Cary, NC) and S-Plus (Insightful, Seattle, WA) statistical packages.

Results

Mammograms were available on 550 (75%) of the 732 women. Women without mammograms were younger and lived farther away from Mayo Clinic than those with mammograms. Otherwise, the groups were similar. Mean (untransformed) PD was inversely associated with age, number of children, and body mass index (BMI), and positively associated with duration of oral contraceptive use, alcohol intake, education level, and infertility (data not shown). With adjustment, only age, BMI, menopausal status, and parity/age at first birth remained significant.

Table 1 displays the PD estimates associated with each CYP19 variant. None of the variants, including the nonsynonymous cSNPs (W39R, R264C, and T201M), were significantly associated with PD. Table 2 presents the two sets of haplotype analyses conducted using the variants selected by the two methods [htSNP (11) versus tagSNP (12)]. These analyses also showed no association with PD, either overall, or stratified by menopausal status, BMI, use of hormone therapy (current versus former/never), or among women age 40 years or more (data not shown). Finally, analyses of haplotype blocks within CYP19 showed no association with PD (data not shown).

Discussion

Polymorphisms in aromatase (CYP19) have the potential to influence hormonally related traits, including PD. However, none of our analyses of individual variants or haplotypes showed an association between CYP19 variants and PD. The other report of CYP19 (including only two variants) and PD also found no association (15).

Tagging methods are an efficient way to select and genotype a maximally informative set of common variants within a candidate gene (11, 12). Results from the SNPs chosen by the two methods (11, 12) we used were similar. None of the variants or haplotypes selected by either method showed a significant association with density.

A strength of this study is the use of complete gene resequencing data to select the gene variants. Unlike the previous study, which examined only two CYP19 variants with density, we investigated all common variation within this gene. A second strength of our study is our power: with the sample available and using a Bonferroni-corrected error rate of 0.0028 (dividing 0.05 by the 18 unique SNPs examined), our study had 80% power to detect differences of square root–transformed density as low as 0.36 and 0.76 (or 0.12 and 0.58 untransformed) for SNPs with minor allele frequencies of 0.3 and 0.1, respectively. Finally, we use a quantitative estimate of PD that has been shown to be strongly associated with breast cancer (16).

Limitations of this study include (a) potential for reduced generalizability due to the primarily Caucasian population; (b) including only 13 duplicate samples that may have resulted in undetectable genotyping errors that could have affected study power and resulted in bias toward null values; and (c) using women recruited as controls for another study. However, the density values in these women were similar to those seen previously (8).

In summary, we found no evidence that genetic variation in SNPs or haplotypes of CYP19 was associated with PD in women.