Gene variations in oestrogen pathways, CYP19A1, daily 17β-estradiol and mammographic density phenotypes in premenopausal women

Introduction High mammographic density is an established breast cancer risk factor, and circulating oestrogen influences oestrogen-regulating gene expression in breast cancer development. However, less is known about the interrelationships of common variants in the CYP19A1 gene, daily levels of oestrogens, mammographic density phenotypes and body mass index (BMI) in premenopausal women. Methods Based on plausible biological mechanisms related to the oestrogen pathway, we investigated the association of single nucleotide polymorphisms (SNPs) in CYP19A1, 17β-estradiol and mammographic density in 202 premenopausal women. DNA was genotyped using the Illumina Golden Gate platform. Daily salivary 17β-estradiol concentrations were measured throughout an entire menstrual cycle. Mammographic density phenotypes were assessed using a computer-assisted method (Madena). We determined associations using multivariable linear and logistic regression models. Results The minor alleles of rs749292 were positively (P = 0.026), and the minor alleles of rs7172156 were inversely (P = 0.002) associated with daily 17β-estradiol. We observed an 87% lower level of daily 17β-estradiol throughout a menstrual cycle in heavier women (BMI >23.6 kg/m2) of rs7172156 with minor genotype aa compared with major genotype AA. Furthermore, the rs749292 minor alleles were inversely associated with absolute mammographic density (P = 0.032). Lean women with rs749292 minor alleles had 70 to 80% lower risk for high absolute mammographic density (>32.4 cm2); Aa: odds ratio (OR) = 0.23 (95% CI 0.07 to 0.75). Lean women with rs7172156 minor homozygous genotype had OR 5.45 for high absolute mammographic density (aa: OR = 5.45 (95% CI 1.13 to 26.3)). Conclusion Our findings suggest that two SNPs in CYP19A1, rs749292 and rs7172156, are associated with both daily oestrogen levels and mammographic density phenotypes. BMI may modify these associations, but larger studies are needed. Electronic supplementary material The online version of this article (doi:10.1186/s13058-014-0499-2) contains supplementary material, which is available to authorized users.


Introduction
Sex hormones, in particular oestrogens, are associated with breast cancer development in both pre-and postmenopausal women [1][2][3], and circulating oestrogens have been shown to influence oestrogen-regulating gene expression [4]. CYP19A1 is a member of the cytochrome P450 family and is involved in the bioconversion of androstenedione to oestrone and of testosterone to estradiol [5]. Human CYP19A1 is a protein commonly known as aromatase and is a gene product of CYP19A1, which is located on chromosome 15q21.2 [6]. In humans, aromatase is expressed in the gonads, adipose tissue and other sites, although the primary site of oestrogen production in premenopausal women is the ovaries [5]. Breast adipose tissue produces oestrogen locally, which may be increased in pre-and postmenopausal obese women [7], owing to higher levels of proinflammatory cytokines such as tumour necrosis factor, a known inducer of aromatase [8,9]. Importantly, the gene CYP19A1 is polymorphic, and the presence of single-nucleotide polymorphisms (SNPs) in the gene may alter aromatase activity, thereby causing variations in the levels of oestrogens [10].
Endogenous oestrogen has been shown to be both inversely [11,12], and positively [13][14][15] associated with mammographic density, and both high absolute and percent mammographic density have consistently been associated with breast cancer development [16,17]. Furthermore, mammographic density phenotypes are a strong heritable biomarker of breast cancer development, and twins studies suggest that genetic factors account for 30% to 60% of its variance [18,19]. In a recent meta-analysis including five genome-wide association studies, a variant (rs10995190) in the ZNF365 gene, which promotes genome stability during DNA damage, was associated with both breast cancer risk and mammographic density [20]. However, this SNP explains only 0.5% of the variance of mammographic density, and many other loci may be involved in predicting mammographic density phenotypes and breast cancer development [20].
Mammographic density is also influenced by several wellknown major breast cancer risk factors such as age, body mass index (BMI), parity, and hormone therapy [21]. Previous studies have observed an inverse association between BMI and premenopausal breast cancer development [22,23]. In contrast, weight gain in early adult life has been associated with postmenopausal breast cancer development [24], but the association between weight gain and premenopausal breast cancer development has not yet been clarified [25]. However, premenopausal abdominal adiposity has been associated with oestrogen receptornegative (ER−) breast cancer [26]. Studies also support excess weight being associated with higher oestrogen levels and ER+ postmenopausal breast cancer development [27,28]. In addition, we have previously shown that salivary estradiol concentrations are positively associated with BMI throughout the menstrual cycle in premenopausal women [29].
Few studies have been focused on genetic susceptibility, daily levels of oestrogen and premenopausal mammographic density, but plausible biological mechanisms may exist because functional genetic polymorphisms in the aromatase gene CYP19A1 have been associated with higher estradiol levels. Therefore, the main aim of the present study was to elaborate whether hypothesis-driven selected common variants in the CYP19A1 gene are associated with daily 17β-estradiol levels and mammographic density phenotypes among healthy premenopausal women and whether BMI modifies these associations.  [30]. Women meeting the following eligibility criteria were included: self-reported regular menstruation (normal cycle length, 22 to 38 days within the previous 3 months), no ongoing use of steroid contraceptives, no pregnancy or lactation in the previous 6 months, no history of gynaecological disorders and no chronic disorders (for example, diabetes, hypo-or hyperthyroidism). Two women were excluded because of missing mammographic data, resulting in 202 participants being included in the present study.

Methods
Participants' characteristics, including reproductive and lifestyle factors, were collected by one trained nurse using questionnaires and interviews at the time of recruitment. Recall and memory-probing aids, including a lifetime calendar, were used to date specific life events. Questionnaires (filled out by the participant and interviewer, administered by trained personnel) were used to collect information about birth weight, age at menarche, marital status, education, ethnicity, reproductive history, lifetime total physical activity, previous use of hormonal contraceptives and family history of cancer, smoking and alcohol. Dietary data were collected on 7 different days during the menstrual cycle (days 3 to 6 and 21 to 23) using a previously validated, precoded food diary [31]. Daily average energy and nutrient intake were computed.

Clinical parameters
Participants attended three study visits during one menstrual cycle: first visit, days 1 to 5 of the menstrual cycle, early follicular phase; second visit, days 7 to 12, late follicular phase; and third visit, days 21 to 25, late luteal phase. Measurements included height to the nearest 0.5 cm and weight (in light clothing) to the nearest 0.1 kg on a regularly calibrated electronic scale. BMI in kilograms per square metre was calculated for all participants. Fasting blood samples were drawn at all three scheduled visits during the menstrual cycle.

Assessment of oestrogen
Serum concentrations of 17β-estradiol were measured in fresh sera for all three collection points using a direct immunometric assay (Immuno-1; Bayer Diagnostics, Norway) at the Department of Clinical Chemistry, UNN [30]. The sensitivity was 0.01 nmol/L, and the coefficient of variation (CV) was 3.9%.
To assess the bioavailable fraction of 17β-estradiol, the participants collected daily saliva samples during one menstrual cycle, preferably in the morning, starting on the first day of menstrual bleeding according to previously established and validated collection protocols developed at the Reproductive Ecology Laboratory of Harvard University [32] and according to the manufacturer's protocol [30]. The samples were stored at −70°C. All samples were run in duplicate, and samples from the same cycles were run within the same assay. The assays were done in different batches at Harvard University. 17βestradiol concentrations were measured in daily saliva samples using a 125 I-based radioimmunoassay kit (no. 39100; Diagnostic Systems Laboratory, Webster, TX, USA). All cycles were aligned to the day of ovulation, based on the identification of the 17β-estradiol drop, which provides a reasonable estimate of the day of ovulation [33,34]. The midcycle 17β-estradiol drop could not be made for 14 of the included women, and their cycles were not aligned. Overall mean salivary 17β-estradiol concentration was calculated for all participants, whereas an additional index of mean menstrual estradiol on days −7 to +6 was calculated for the 188 women with aligned cycles. The sensitivity of the 17β-estradiol salivary assay was 4 pmol/L, and the average intra-assay CV was 9%. The measurements of 17β-estradiol had a higher CV at the start and end of the menstrual cycle, and the interassay CV ranged from 23% (low pool) to 13% (high pool). Furthermore, there were higher rates of missing data at the end of the cycle, so we included aligned measurements of salivary 17β-estradiol from day −7 to day +6 in this study.

Assessment of mammographic density
Bilateral two-view mammograms were obtained from women during the second scheduled visit (between cycle days 7 and 12) at the Centre of Breast Imaging, UNN, using a standard protocol [30]. The left craniocaudal mammograms were digitised and imported into a computerised mammographic density assessment programme (Madena) developed at the University of Southern California School of Medicine (Los Angeles, CA, USA) [35,36]. The density measurements were conducted by one trained reader (GU), and the total breast area was determined by a research assistant trained by GU. The total breast area was defined using a special outlining tool, and the size of this area in square centimetres using the Madena software. To assess density, the reader outlined a region of interest (ROI), excluding the pectoralis muscle, prominent veins and fibrous strands. The reader applied a tinting tool to pixels considered to represent dense areas of the mammograms within the ROI. The Madena software calculated the size of this dense area in square centimetres. Absolute mammographic breast density represented this dense area, and the percentage mammographic density was the ratio of absolute mammographic breast density to total breast area multiplied by 100. The mammograms were read in four batches, with an equal number of mammograms included in each batch. A duplicate reading of 26 randomly selected mammograms from two of the batches showed a Pearson's correlation coefficient of 0.97. The reader was blinded to any characteristics of the study population.

Single-nucleotide polymorphism selection and genotyping
We analysed CYP19A1 genetic polymorphisms that encode the aromatase enzyme. Blood samples from 204 women in the EBBA-I study were frozen at −70°C. DNA was extracted from whole blood using a MagAttract DNA Blood Mini M48 kit (QIAGEN, Valencia, CA, USA) by the Department of Medical Genetics, UNN. DNA was genotyped on the Golden Gate Platform (Illumina, San Diego, CA, USA) at the Fred Hutchinson Cancer Research Center (Makar Lab), using the manufacturer's protocol. In brief, 250 ng of genomic DNA was divided into aliquots in 96-well plates, processed accordingly and scanned on the Illumina iScan reader using BeadStudio software.
We conducted a series of quality control procedures [37]. SNP call rates exceeded 99% for this study, with 100% concordance of blinded duplicates. The linkage disequilibrium select algorithm was employed to choose the tag SNPs via the Genome Variation Server [38,39]. The SNPs were selected using an r 2 threshold of 0.8 and a minor allele frequency >5%, representing variability in the white European population. Tag SNP coverage extended 2 kilobases (kb) upstream and 1 kb downstream of the gene, and 29 SNPs were covered. We further reduced the number of SNPs using power calculations and ended up with a final selection of eight common SNPs with minor allele frequency >0.2: rs10046, rs17703883, rs2414097, rs2445761, rs4646, rs7172156, rs727479 and rs749292 (see Additional file 1). None of the selected SNPs was monomorphic or significantly out of Hardy-Weinberg equilibrium.

Covariate analytes
Serum concentrations of total cholesterol were determined enzymatically using cholesterol esterase and cholesterol oxidase. High-density lipoprotein cholesterol (HDL-C) was quantified by direct assay using enzymes modified by polyethylene glycol and dextran sulphate.

Statistical methods
On the basis of the plausible biological mechanisms related to the oestrogen pathway, we investigated the associations between eight SNPs in the CYP19A1 gene, hormone levels (salivary midmenstrual 17β-estradiol and serum 17β-estradiol) and mammographic density phenotypes (total breast area, absolute mammographic density, percent mammographic density and nondense breast area) using multivariable linear regression models. Associations were assessed for the selected SNPs, and the selected SNPs were coded as AA = 0 (major homozygous), Aa = 1 (heterozygous) and aa = 2 (minor homozygous) and were included as ordinal variables in the models. We compared the linear response between the categories of genotypes by including indicator variables for Aa and aa, using AA as the reference.
Age, parity and BMI are known to be associated with mammographic density phenotypes, are possibly associated with hormone levels and/or CYP19A1 variants, and were therefore considered as potential confounders and included as covariates in all models [21]. Furthermore, the models with mammographic density as the dependent variable also included salivary 17β-estradiol and serum HDL-C, both of which are known to influence mammographic density [40,13]. In the final analyses, we focused on two selected SNPs (rs7172156 and rs749292) and stratified the women by major, heterozygous and minor genotypes. We then compared the genotype groups using different characteristics of the study population (lifestyle factors, anthropometric measures, serum blood sampling and salivary hormone sampling), and we used one-way analysis of variance for continuous variables and the χ 2 test for categorical variables.
The multivariable logistic regression models were run using median absolute mammographic density (32.4 cm 2 ) and median percent mammographic density (28.5%) as cutoff values. Mammographic density was used as a dependent variable, and rs7172156 and rs749292 were used as independent variables, adjusted for age, parity and BMI. In addition, we analysed in detail whether BMI variations influenced our results (that is, tertiles/dichotomised BMI), but only dichotomised BMI by median BMI values gave additional information and thus were included in the final analysis.
We used linear mixed models for repeated measures to study variations of daily salivary 17β-estradiol across the menstrual cycle for subgroups of women with major, minor homozygous or heterozygous genotypes in the SNPs rs7172156 and rs749292, and we then adjusted for age, BMI and parity. The Toeplitz covariance structure gave the best fit to the data and was used in all models.
Our candidate polymorphisms were based on plausible biological hypotheses, and all P-values were two-tailed and considered significant when the value was <0.05. The analyses were conducted with SPSS version 21.0 software (IBM, Armonk, NY, USA).

Ethical considerations
All participants underwent informed consent procedures and signed a consent form. The study was approved by the Norwegian Data Inspectorate and the Regional Committee for Medical Research Ethics.

Results
The participating premenopausal women had mean values (standard deviation (SD)) for age of 30.7 (3.07) years and BMI of 24.4 (3.77) kg/m 2 (Table 1). When we stratified the women into groups for rs749292 and rs7172156 by major homozygous, heterozygous and minor homozygous genotypes, we observed no differences in lifestyle factors, anthropometric measures or serum analytes (Table 1).
We observed an association between two SNPs (rs749292, rs7172156) and both salivary estradiol and absolute mammographic density. Moreover, a positive association was observed between rs749292 and midmenstrual salivary 17β-estradiol (P = 0.026), and an inverse association between rs7172156 and midmenstrual salivary 17β-estradiol (P = 0.002), after adjustment for age, BMI and parity ( Table 2). We also observed a negative association between rs749292 and absolute mammographic density (P = 0.032) after adjusting for age, BMI, parity, salivary midmenstrual 17β-estradiol and serum HDL-C.
No clear association was observed between any of these SNPs and serum levels of 17β-estradiol at any of the three measured time periods (early follicular, late follicular or late luteal phase) of the menstrual cycle. In the mixed linear regression models, we found that women with different genotypes of rs7172156 varied in the levels of average midmenstrual salivary 17β-estradiol (P = 0.001). Among women with genotype AA and genotype Aa, compared with women with genotype aa, 57% and 56% higher mean 17β-estradiol levels were observed, respectively ( Figure 1d). This association was even more marked when we dichotomised the data by median split of BMI (23.6 kg/m 2 ). We observed an 87% lower level of mean 17β-estradiol throughout a menstrual cycle in heavier women (BMI >23.6 kg/m 2 ) with minor genotype aa of rs7172156 compared with those with major genotype AA (Figure 1f). Among women with genotype AA, heavier women had a 33% higher level of 17β-estradiol compared to leaner women. However, in genotype aa, there was no increase in 17β-estradiol levels when we compared leaner and heavier women. When comparing mean 17β-estradiol levels in lean women (BMI ≤23.6 kg/m 2 ) with rs749292 major genotype AA with heavier women (BMI >23.6 kg/m 2 ) with rs749292 minor genotype aa, a 52% higher mean 17β-estradiol level was observed ( Figure 1).

rs749292 and rs7172156 and mammographic density phenotypes
The association between the SNPs and mammographic density phenotypes was studied with multivariable linear regression models. For rs749292, we observed an inverse association between minor alleles (Aa, aa) and absolute mammographic density (Table 4). We observed a positive association between rs7172156 minor genotype aa and absolute mammographic density. After dichotomising by median split of BMI, we found that rs749292 minor alleles were inversely associated with absolute mammographic density (Aa: β = −13.0, P = 0.006; aa: β = −14.1, P = 0.010) in lean women (≤23.6 kg/m 2 ), but not in women with a BMI >23.6 kg/m 2 . Among lean women (≤23.6 kg/m 2 ) with rs7172156 genotype aa, we observed a positive association with absolute mammographic density (aa: β = 18.2, P = 0.005) ( Table 4).
For rs7172156, lean women with a minor homozygous genotype had a 5.45 higher OR for high absolute mammographic density (aa: OR = 5.45 (95% CI, 1.13 to 26.3)). Similar associations were observed for rs7172156 and percent mammographic density (Table 5).

Discussion
In the present study of premenopausal women, two SNPs (rs749292, rs7172156) of eight studied in the CYP19A1 gene were related to both daily salivary 17β-estradiol and mammographic density phenotypes. The association with mammographic density was revealed when we used salivary 17β-estradiol as a covariate, and similar results were observed for absolute and percent mammographic density. Furthermore, our results suggest that body weight may modify these associations. We observed an 87% lower level of daily 17β-estradiol throughout a menstrual cycle in heavier women (BMI >23.6 kg/m 2 ) with minor genotype aa (17β-estradiol 12.3 pmol/L) of rs7172156 compared with major genotype AA (17β-estradiol 22.4 pmol/L). Furthermore, lean women with rs7172156 minor homozygous genotype aa had a fivefold higher OR for high absolute mammographic density compared with major homozygous genotype AA. Lean women who had rs749292 minor alleles had 70% to 80% lower risk for high absolute and high percent mammographic density compared with major homozygous genotype AA. The CYP19 activity is responsible for the bioconversion of androgens to oestrogens [5,6], and to our knowledge, there have been few studies related to CYP19A1 SNPs, daily levels of oestrogen throughout an entire menstrual cycle and mammographic density phenotypes in premenopausal women. It is not clear why and how noncoding SNPs influence the gene activity, but previous genome-wide association studies have shown intronic SNPs to be important breast cancer risk loci [41]. This does not necessarily imply that the SNPs are causal, but it may help to identify novel susceptibility loci. In addition, intronic SNPs may regulate gene expression through endogenous trans-acting factors, epigenetics and chromosome conformation [42]. Our results are in part supported by a previous report [10] that rs749292 minor alleles were associated with a 10% to 20% increase in oestrogen levels among postmenopausal women in a combined analysis of five cohort studies. Other SNPs in the CYP19A1 gene have also been studied, and in one study on postmenopausal women with a mean age of 57 years and a mean BMI of 24.2 kg/m 2 , researchers found an association with circulating oestrogen levels, but only among women with BMI >25 kg/m 2 [43]. Interestingly, functional genetic polymorphisms may also influence the level of estradiol in women undergoing inhibitory treatment, as two CYP19A1 SNPs were associated with higher estradiol levels, particularly after initiation of aromatase inhibitors [44]. These findings imply that CYP19A1 SNPs may be of clinical interest, as aromatase inhibitor treatment has been shown to be one of the most effective modern antihormonal breast cancer treatment regimens. To our knowledge, no clear associations have been observed between CYP19A1 SNPs and mammographic density [45], and researchers in one study found no associations of oestrogen synthesis or oestrogen metabolism genes with mammographic density in a mixed population of perimenopausal, young postmenopausal and postmenopausal women [45]. Few known genetic variants predict both mammographic density and breast cancer risk, but Lindstrom et al. found an association between common variants in the ZNF365 gene, which promotes genome stability under DNA damage, with both mammographic density and breast cancer development [20]. In addition, SNPs in the inflammatory gene interleukin-6 (IL-6) have recently been associated with premenopausal percent mammographic density [46]. Despite the clear association of endogenous oestrogens with breast cancer development [1], results have been inconsistent regarding associations between CYP19A1 variants and risk for breast cancer [10,[47][48][49], but rs1008805 [50] and, recently, rs10046 were observed to be associated with breast cancer susceptibility among premenopausal women [51].
Elevated BMI has been related to higher levels of sex hormones in both premenopausal [30] and postmenopausal women [52], and weight loss through diet and exercise may reduce sex steroid hormone levels in premenopausal [53] and postmenopausal women [54]. We previously observed that CYP17 polymorphisms were associated with 17βestradiol levels, especially in women with unfavourable metabolic profiles [55]. Interestingly, in the present study, an inverse association was observed between rs749292 minor alleles and absolute mammographic density among lean women, but this association disappeared in heavier women. Furthermore, rs7172156 minor alleles were associated with higher absolute mammographic density among lean women. In contrast, we found that among women with minor alleles and high BMI, rs7172156 may be a protective polymorphism associated with lower 17β-estradiol and lower OR for having above-median percent mammographic density (>28.5%) and absolute mammographic density (>32.4 cm 2 ). Similar mammographic threshold estimates of 25% mammographic density and 32-cm 2 absolute mammographic density have been shown to predict a two-to threefold risk of breast cancer development within 5 to 10 years [56,57]. Interestingly, a previous study observed an association between rs7172156, rs749292 [58] and serum levels of hepatocyte growth factor (HGF). HGF is a cytokine derived from adipose tissue [58] that promotes cell migration, proliferation and invasion, and previous studies have found associations between HGF levels and development from benign breast disorders to preinvasive, basal-like breast cancer [59], as well as further correlations with poor prognosis. These findings lead us to hypothesize that there may be a biological rational for the associations we observed for two SNPs in CYP19A1: rs7172156 and rs749292.
Our study has several strengths. These are inclusion of premenopausal women; clinical measurements carefully timed to the menstrual cycle, including mammographic density phenotypes and serum and daily saliva 17βestradiol; and a validated computer-assisted method for quantifying mammographic density. In contrast, we did not observe the same associations between these two SNPs in CYP19A1 and serum 17β-estradiol as we did for salivary 17β-estradiol. Importantly, salivary 17β-estradiol was assessed daily, is the free bioavailable fraction and is not bound to albumin or sex hormone-binding globulin, in contrast to the serum 17β-estradiol levels, and may in part explain the variations observed [30,33]. Previous research has indicated that single measurements of serum oestrogen do not accurately reflect women's long-term oestrogen levels [3], whereas multiple measurements of unbound bioavailable levels probably give a picture of the real endogenous cumulative exposure over time. This means that single measurements are likely to be an   underestimate because they do not capture the premenopausal cyclical changes and will be imperfect estimates of the true pattern [3,60]. Thus, use of exploratory, noninvasive, repeated sampling of salivary hormones may provide new knowledge on the true association between hormones and breast cancer. Moreover, this may in part explain why circulating oestrogen levels consistently have been observed to increase risk, as well as risk prediction for invasive postmenopausal breast cancer [61], but the association between endogenous oestrogen levels and breast cancer among premenopausal women is less clear [3]. Today, liquid chromatography-tandem mass spectrometry, as compared to the immunoassay method, is a more efficient way of analysing salivary hormones with higher specificity and sensitivity. However, previous studies on estradiol measurements, specifically, have shown a correlation of 0.969 between mass spectrometry and immunoassays [62]. However, our sample size was small, and associations could have been missed by chance. Furthermore, the population was a sample of volunteer participants and therefore may not be representative of the source population, but their average BMI and other lifestyle-related factors and lipid profiles are in accordance with the population of premenopausal Norwegian women [63]. A limited number of SNPs were examined, based on the biological hypothesis that polymorphisms in the CYP19 gene may influence 17β-estradiol levels and mammographic phenotypes. Even though only eight SNPs were examined, there is a risk of false-positive results. Nevertheless, our findings are intriguing and support future research in larger sample sizes.

Conclusion
In the present study, we found associations between two CYP19A1 SNPs (rs7172156 and rs749292) and both daily 17β-estradiol throughout an entire menstrual cycle and both absolute and percent mammographic density in premenopausal women, and the results differed between lean and heavier women. This observation suggests that there may be genetic influences on these breast cancer biomarkers and also that the effect of body size may play a major role. Future research on genetic control of mammographic density phenotypes and sex hormones should include exploratory salivary hormone measurements and take body size and adiposity into account.

Additional file
Additional file 1: Four supplementary tables. Table S1. Allele frequencies and distributions of selected single-nucleotide polymorphisms (SNPs) in CYP19A1: The Norwegian EBBA-I study. Table S2. Population frequencies of single-nucleotide polymorphisms (SNPs) in selected single-nucleotide polymorphisms in CYP19A1. Table S3. Associations between each of eight selected single-nucleotide polymorphisms (SNPs) in the CYP19A1 region and mammographic density (total breast area, absolute density, percent density and nondense breast areas). Table S4. Associations between each of eight selected single-nucleotide polymorphisms (SNPs) in the CYP19A1 region and estradiol.

Competing interests
The authors declare that they have no competing interests.
Authors' contributions IT and ASF conceived of and designed the study and collected the clinical data. PTE carried out the radioimmunoassay and measured salivary hormone levels. GU digitized and measured the mammographic density data. KWM and AM performed the single-nucleotide polymorphism analysis. VGF, HF, TL, TE, TW, AI, EAW and JCC performed statistical analyses. VGF, ASF, AM, HF, GU, AI, TL, PTE, EAW, TE, TW, KWM, JCC and IT interpreted the results. VGF drafted the manuscript in cooperation with IT. All authors contributed to critical revision of the manuscript, editing of the final version of the manuscript and agree to be accountable for the accuracy and integrity of the work. All authors read and approved the final manuscript.