Comprehensive Analysis of Hormone and Genetic Variation in 36 Genes Related to Steroid Hormone Metabolism in Pre- and Postmenopausal Women from the Breast and Prostate Cancer Cohort Consortium (BPC3)

In a sample of Caucasian women from prospective cohort studies, we find novel associations and confirm reported findings of genetic variation with sex steroid hormones.

Abstract

Context:

Sex steroids play a central role in breast cancer development.

Objective:

This study aimed to relate polymorphic variants in 36 candidate genes in the sex steroid pathway to serum concentrations of sex steroid hormones and SHBG.

Design:

Data on 700 genetic polymorphisms were combined with existing hormone assays and data on breast cancer incidence, within the European Prospective Investigation into Cancer and Nutrition (EPIC) and the Nurses' Health Study (NHS) cohorts; significant findings were reanalyzed in the Multiethnic Cohort (MEC).

Setting and Participants:

We analyzed data from a pooled sample of 3852 pre- and postmenopausal Caucasian women from EPIC and NHS and 454 postmenopausal women from MEC.

Main Outcome Measures:

Outcome measures were SHBG, testosterone, dehydroepiandrosterone (DHEAS), androstenedione, estrone (E1), and estradiol (E2) as well as breast cancer risk.

Results:

Globally significant associations were found among pre- and postmenopausal women combined between levels of SHBG and the SHBG gene and between DHEAS and the FSHR and AKR1C3 genes. Among postmenopausal women, serum E1 and E2 were significantly associated with the genes CYP19 and FSHR, and E1 was associated with ESR1. None of the variants related to serum hormone levels showed any significant association with breast cancer risk.

Conclusions:

We confirmed associations between serum levels of SHBG and the SHBG gene and of E1 and E2 and the CYP19 and ESR1 genes. Novel associations were observed between FSHR and DHEAS, E1, and E2 and between AKR1C3 and DHEAS.

Sex steroids play an important role in breast cancer development. For postmenopausal women, prospective cohort studies have shown that higher blood concentrations of estrogens [estrone (E1) and total and bioavailable estradiol (E2)] and androgens [total and bioavailable testosterone (TESTO)] and lower levels of SHBG are associated with an increased risk of breast cancer (1–4). An increased risk was also observed among premenopausal women with elevated blood levels of androgens, particularly dehydroepiandrosterone sulfate (DHEAS) and TESTO, or lower concentrations of serum progesterone (5, 6).

The association between breast cancer risk and genetic polymorphisms in genes related to hormone metabolism has been studied repeatedly (7–9). Fewer studies have examined the association of these polymorphisms with circulating hormone levels in women (10–13).

In a pooled analysis of 3852 pre- and postmenopausal Caucasian women drawn from the National Cancer Institute Breast and Prostate Cancer Cohort Consortium (BPC3), we performed a systematic and comprehensive examination of the associations between circulating levels of sex hormones and 700 single-nucleotide polymorphisms (SNPs) in 36 candidate genes that affect the synthesis, bioavailability, or metabolism of sex hormones (14). We used preexisting measurements of SHBG, E1, E2, TESTO, DHEAS, and Δ4-androstenedione (Δ4) in women participating in the European Prospective Investigation into Cancer and Nutrition (EPIC) and the Nurses' Health study (NHS). Observed associations were reexamined in samples from 454 women of five different ethnicities drawn from the Multiethnic Cohort (MEC). We furthermore tested whether any polymorphism associated with variation in hormone levels is associated with breast cancer risk as well.

Subjects and Methods

Study population

A total of 3852 breast cancer cases and controls from EPIC (n = 2772) and NHS (n = 1080) were analyzed. The EPIC sample consisted of 937 cases and 1835 controls. A total of 1037 women were premenopausal at the time of blood donation, and 1735 women were postmenopausal (1). The NHS sample consisted of 359 cases and 721 controls, all postmenopausal (2). The MEC sample consisted only of postmenopausal controls, and their self-reported ethnicities were as follows: 119 African-Americans, 84 Asians, 79 Caucasians, 70 Hawaiians, and 102 Latinos (15). In all three cohorts, women who used postmenopausal hormones at baseline or at blood draw were excluded. Informed consent was obtained from patients before sample collection for all cohort studies. Ethics approvals were obtained for all cohort studies.

Hormone measurement

Concentrations of DHEAS, Δ4, TESTO, E1, E2, and SHBG were measured in prediagnostic blood samples (Table 1). E1 and E2 were measured in postmenopausal women only, because of the large effect of menopause on their blood concentrations and because of the large variations in premenopausal women during the menstrual cycle. Previous studies have shown reasonably high intraclass correlations (ICCs) between measured concentrations of DHEAS (ICC > 0.85), Δ4 (ICC > 0.66), and TESTO (ICC > 0.73) in blood samples taken over time periods of 1 yr and longer among both pre- and postmenopausal women, indicating that even a single measurement can reasonably reflect longer-term serum concentrations of androgens and SHBG among pre- and postmenopausal women, and of the estrogens among postmenopausal women (16, 17). Δ4, DHEAS, SHBG, and TESTO were measured in both pre- and postmenopausal women for EPIC and for postmenopausal women in NHS/MEC. DHEAS was not available for MEC. TESTO and Δ4 were not analyzed in MEC, because no significant associations were identified in EPIC/NHS.

Table 1.

Characteristics of the study sample for sex steroid hormone levels and covariates are presented for EPIC, NHS, and MEC as number (and proportion, percent) of subjects with measured hormones as well as corresponding means and sd

n.a., Not applicable because of low sample size.

^a RIA, Radioimmunoassay.

^b Values taken for EPIC from Refs. 1 and 6, for NHS from Ref. 2, and for MEC from Ref. 4.

Genetic data

To identify tagging SNPs that capture most of the common variants within the candidate genes, we systematically resequenced, in a preliminary step, genes in breast and prostate cancer cases with different ethnic backgrounds. The resulting SNPs were genotyped in larger samples, and this data set combined with information from the HapMap project was used to select the subset of tagging SNPs that capture most of the common variants within the genes. The elected SNPs were genotyped using TaqMan and Illumina GoldenGate technology. We imputed 2671 SNPs using phased haplotypes from HapMap samples using the software MACH 1.0.

Details about study sample and the type of hormone assays used are given in Table 1; details about gene sequencing, genotyping, and quality control can be found in the Supplemental Materials and Methods and Supplemental Table 1 (published on The Endocrine Society's Journals Online web site at http://jcem.endojournals.org).

Statistical methods

Descriptive analysis

To adjust for differences in the absolute levels of sex steroid hormones, we calculated the residuals of the natural logarithm of the hormone levels regressed on breast cancer case-control status, body mass index (BMI), menopausal status (for pre- and postmenopausal women pooled), and age at blood donation. We adjusted further for assay batch to account for possible laboratory-induced differences in hormone measurement and thereby also adjusted for cohort membership. For the analysis of the MEC sample, we adjusted additionally for ethnicity.

Association tests

Association of genes with variation in hormone levels.

We performed regression of the hormone residuals on each SNP independently for four models: recessive, dominant, the additive model with one degree of freedom that corresponds to the usual trend test, and the additive model with two degrees of freedom. We combined the MAX test (18) with the step-down-min-p algorithm, a permutation-based algorithm to control the familywise error rate, to adjust for multiple testing, taking into account the correlation between the SNPs due to linkage disequilibrium (19). The adjusted P values were further Bonferroni corrected by the number of genes studied. The minimum adjusted P value among the SNPs for any gene was considered as the global P value of this gene.

We considered the association of a gene (and the corresponding SNP) with variation in hormone levels to be globally significant if the global P value was lower than the significance level α = 0.05. If the step-down-min-p algorithm yielded a P value of P = 0.0 based on 1000 permutations, it indicated P < 0.001. The corresponding global P value is P_global < 0.036, as calculated by Bonferroni correction for 36 genes. For the explorative analysis in the MEC samples, a more lenient threshold of P = 0.05 was used. We tested for gene-gene interactions using the likelihood ratio test.

Association of genes with breast cancer risk.

Genes associated with variation in hormone levels were tested for association with breast cancer risk in a logistic model, adjusted for age at blood donation, BMI, batch, and for menopausal status, if required. The same strategy as outlined above was used to correct for multiple testing and correlation between the test statistics.

Details about the statistical methods can be found in the Supplemental Materials and Methods.

Results

Correlation of hormone levels with age and BMI

In the pooled sample of EPIC and NHS, age was negatively correlated with Δ4 (Pearson correlation coefficient r = −0.41), DHEAS (r = −0.36), and TESTO (r = −0.30). BMI was negatively correlated with SHBG (r = −0.39) and positively correlated with E1 (r = 0.27) and E2 (r = 0.25) (Supplemental Table 2). DHEAS and TESTO were not correlated with BMI in the sample of pre- and postmenopausal women.

Association of hormone levels

In the analyses combining pre- and postmenopausal women, four SNPs in the SHBG gene were associated with lower levels of serum SHBG (Table 2). Notable changes in levels of SHBG were found for SNP rs1619016 (−8.6% for heterozygotes and −26.1% for homozygotes minor allele) and for rs9913778 (−16.9% for heterozygotes and −21.6% for homozygotes minor allele). For DHEAS, we found novel globally significant associations in the FSH receptor (FSHR, with both lower and elevated hormone levels) and in the aldoketoreductase-1C3 gene (AKR1C3, with increased levels of DHEAS).

Table 2.

Results for the significant associations of hormone levels with the SNPs in the sex steroid pathway

Hormone/menopausal status	Gene	SNP	Major/minor allele	Locationa	MAF	Heterozygote				Homozygote minor allele				Test for association with hormone levels
						Estimate	sd	% differenceb	P valuec					R2 (%)d	P values for the four models				Globale	Model for minimum P value
										Estimate	sd	% differenceb	P valuec		Trend	Additive	Dominant	Recessive
Pre- and postmenopausal
SHBG	SHBG	rs1619016	T/C	Intergenic	0.12	−0.09	0.02	−8.6	0	−0.30	0.06	−26.1	0	1.0	<10−6	0.008370	<10−6	0.000005	<0.036	Trend
n = 3796		rs2955617	A/C	Intergenic	0.32	−0.09	0.02	−8.8	0	−0.21	0.03	−19	0	1.8	<10−6	0.000256	<10−6	<10−6	<0.036	Trend
R2 covariatesf: 0.283		rs9898876	G/T	Intergenic	0.18	−0.06	0.02	−6	0.001	−0.08	0.04	−7.4	0.067	0.3	0.000284	0.197217	0.000201	0.154507	<0.036	Trend
		rs9913778	C/T	Intron	0.07	−0.19	0.02	−16.9	0	−0.24	0.13	−21.6	0.053	1.6	<10−6	0.000564	<10−6	0.086601	<0.036	Dominant
DHEAS
n = 3547	FSHR	rs12713034	A/G	Intron	0.45	−0.02	0.02	−2.3	0.308	−0.11	0.03	−10.1	0	0.4	0.000567	0.085368	0.034019	0.000297	<0.036	Recessive
R2 covariatesf: 0.165		rs1290100	G/C	Intron	0.34	0.08	0.02	7.9	0	0.07	0.03	7.1	0.037	0.4	0.002194	0.100555	0.000308	0.393179	<0.036	Trend
														0.0
	AKR1C3	rs10752001	T/C	Intergenic	0.16	0.08	0.02	7.7	0.001	0.09	0.07	8.9	0.213	0.3	0.001015	0.135326	0.000741	0.339883	<0.036	Dominant
Postmenopausal
E1
n = 2433	CYP19	rs10046	G/A	Exon	0.52	0.05	0.02	5.2	0.005	0.11	0.02	11.1	0	1.2	<10−6	0.221788	0.000091	0.000021	<0.036	Trend
R2 covariatesf: 0.427		rs4646	C/A	Exon	0.26	−0.04	0.02	−3.8	0.008	−0.08	0.03	−7.7	0.005	0.5	0.000557	0.496994	0.001543	0.023862	<0.036	Trend
		rs6493494	G/A	Intron	0.42	0.04	0.02	4.5	0.006	0.10	0.02	10.9	0	0.9	0.000002	0.262544	0.000150	0.000092	<0.036	Trend
		rs727479	A/C	Intron	0.35	−0.04	0.02	−3.9	0.008	−0.12	0.02	−10.9	0	1.0	0.000001	0.232941	0.000137	0.000022	<0.036	Trend
		rs749292	G/A	Intron	0.43	0.06	0.02	5.8	0	0.11	0.02	11.3	0	1.1	<10−6	0.212815	0.000008	0.000172	<0.036	Trend
	ESR1	rs1884053	T/C	Intron	0.36	−0.06	0.02	−5.4	0	−0.03	0.02	−2.5	0.265	0.5	0.019190	0.454434	0.000696	0.827124	<0.036	Dominant
		rs2347871	A/G	Intron	0.35	−0.06	0.02	−5.3	0	−0.02	0.02	−2.4	0.29	0.5	0.019456	0.462545	0.000791	0.821801	<0.036	Dominant
		rs9341016	T/C	Intron	0.06	0.00	0.02	0	0.986	0.42	0.12	52.4	0	0.5	0.217255	0.461703	0.539285	0.000284	<0.036	Recessive
	FSHR	rs10495968	C/T	Intron	0.28	−0.05	0.02	−4.7	0.001	−0.06	0.03	−5.5	0.037	0.5	0.000902	0.486873	0.000485	0.177197	<0.036	Dominant
		rs11125215	G/A	Intron	0.23	−0.05	0.02	−4.8	0.001	−0.08	0.03	−7.8	0.012	0.6	0.000160	0.429278	0.000243	0.048757	<0.036	Trend
		rs1157876	C/T	Intron	0.26	−0.05	0.02	−5.2	0	−0.05	0.03	−4.7	0.092	0.6	0.000927	0.451640	0.000236	0.356689	<0.036	Dominant
		rs1394205	C/T	Exon	0.28	−0.05	0.02	−5	0.001	−0.06	0.03	−5.7	0.031	0.6	0.000485	0.446552	0.000217	0.178605	<0.036	Dominant
		rs4331540	C/T	Intron	0.28	−0.05	0.02	−5.2	0	−0.06	0.03	−5.7	0.03	0.6	0.000328	0.421135	0.000125	0.178605	<0.036	Dominant
		rs4971637	G/T	Intron	0.22	−0.04	0.02	−3.4	0.019	−0.10	0.04	−9.6	0.005	0.5	0.000810	0.502395	0.003579	0.012714	<0.036	Trend
		rs4971665	A/G	Intron	0.29	−0.05	0.02	−4.5	0.002	−0.07	0.03	−6.7	0.005	0.6	0.000307	0.454324	0.000387	0.038361	<0.036	Trend
		rs4971884	G/A	Intron	0.24	−0.04	0.02	−4.3	0.003	−0.08	0.03	−7.9	0.01	0.5	0.000332	0.469314	0.000698	0.038213	<0.036	Trend
		rs7606570	T/G	Intron	0.24	−0.05	0.02	−4.4	0.003	−0.08	0.03	−7.3	0.017	0.5	0.000414	0.477461	0.000621	0.062633	<0.036	Trend
E2
n = 2721	CYP19	rs10046	G/A	Exon	0.53	0.06	0.02	6.6	0.001	0.12	0.02	13.2	0	1.1	<10−6	0.099518	0.000010	0.000010	<0.036	Trend
R2 covariatesf: 0.733		rs6493494	G/A	Intron	0.42	0.04	0.02	4.1	0.02	0.12	0.02	13.1	0	1.0	0.000001	0.126579	0.000342	0.000004	<0.036	Trend
		rs727479	A/C	Intron	0.34	−0.05	0.02	−5.3	0.001	−0.12	0.03	−11.4	0	0.9	<10−6	0.139278	0.000015	0.000127	<0.036	Trend
		rs749292	G/A	Intron	0.43	0.04	0.02	4.3	0.014	0.13	0.02	14.4	0	1.2	<10−6	0.077097	0.000108	<10−6	<0.036	Trend
	FSHR	rs10454135	A/G	Intergenic	0.42	−0.05	0.02	−5.2	0.002	−0.06	0.02	−5.7	0.01	0.4	0.002522	0.407723	0.000671	0.194884	<0.036	Dominant

MAF, Minor allele frequency.

^aDerived from dbSNP.

^bEvaluated as the exp(b), where b is the estimate from the linear regression of the residuals on the SNPs under a codominant model.

^cP values from linear regression of the residuals on the SNP under a codominant model.

^dR² from linear regression of the residuals on the SNPs under a codominant model.

^eGlobal P value limited by 1000 permutations.

^fR² from linear regression of the residuals on breast cancer case-control status, BMI, menopausal status (for pre- and postmenopausal women pooled), assay batch, and age at blood donation.

In postmenopausal women, we found globally significant associations between serum levels of E1 and E2 with the genes encoding aromatase CYP19, FSHR, and, for E1, the estrogen receptor 1, ESR1 (Table 2). Four SNPs in CYP19 (rs10046, rs6493494, rs727479, and rs749292) were each associated with differences in geometric means of about 5% for the heterozygotes and about 11% for the homozygote minor alleles. Additionally, rs4646 was associated with E1, but with smaller differences (−3.8% for the heterozygotes and −7.7% for the homozygote minor alleles). SNPs rs10046 and rs4646 are both located in the 3′ untranslated region of CYP19, whereas rs727479 and rs749292 span the coding and proximal 5′ untranslated region of CYP19. For E1, SNP rs9341016 in ESR1 showed a change of 52.4% for the homozygote minor allele, corresponding to the recessive disease model (P_global < 0.036).

In the MEC sample, we reanalyzed the genes for which we found significant associations in EPIC/NHS (Supplemental Table 3). Two SHBG SNPs, rs1619016 and rs2955617, were replicated in the Caucasian subsample (unadjusted P = 0.021 and P = 0.012, respectively). In the other ethnic groups, some of the associations of variants in the SHBG gene with SHBG, and of variants in CYP19 with E1 and E2, could be replicated.

Imputation and SNP × SNP interaction in EPIC and NHS

We imputed 2671 nonmeasured SNPs (Supplemental Table 1). Their analysis did not reveal further evidence for additional genes to be associated with variation in circulating hormone levels (Supplemental Fig. 1). Within the genes found to be associated, the signals of the imputed SNPs were at nearly the magnitude of the most significant measured SNPs (Supplemental Fig. 2).

When testing pair-wise SNP × SNP interactions for each hormone, the lowest P value, P = 1.8 × 10⁻⁷, was observed for TESTO in the combined sample of pre- and postmenopausal women for the SNPs rs985192 in ESR1 and rs6755901 in LHCGR (which encodes the receptor for both LH and choriogonadotropin). However, in none of the models did the likelihood ratio test identify a statistically significant interaction, defined as the deviation form the log-additive model.

Association with breast cancer risk

None of the five genes that significantly associated with variation in hormone levels were significantly associated with breast cancer risk after adjustment for multiple tests and correlation between the test statistics (Supplemental Fig. 3 and Supplemental Table 4).

Discussion

We confirmed previously reported associations between serum levels of SHBG and SHBG SNPs and between circulating levels of E1 and E2 and polymorphisms in CYP19 and ESR1 (9–13, 20). In addition, novel associations were observed between SNPs in FSHR and DHEAS (minimum unadjusted P value P_μ = 0.000308), E1 (P_μ = 0.000125), and E2 (P_μ = 0.000671) and between polymorphisms in AKR1C3 and DHEAS (P_μ = 0.000741).

Although genetic variation in the FSH receptor gene (FSHR) has not been observed to influence serum levels of DHEAS, E1, or E2 in previous studies, FSH itself has major effects on these hormones. FSH stimulates estrogen synthesis by ovarian follicles, and activating mutations in FSHR have been linked to ovarian hyperstimulation syndrome and to a higher responsiveness to FSH, whereas inactivating mutations, by contrast, have been linked to premature ovarian failure (21).

AKR1C3 (also known as 17β-hydroxysteroid dehydrogenase type 5, HSD17B5) is highly expressed in the mammary gland where it is thought to promote a proestrogenic environment and in adipose tissue. In the mammary glands, the combined actions of CYP19 aromatase and AKR1C3 result in the generation of E1 and E2 and consequent activation of the estrogen-binding receptors ERα and ERβ. Recently, differential expression of AKR1C3 in malignant breast tissues compared with normal tissue was found (22, 23).

The analysis of ethnicity strata in MEC revealed multiple significant findings for E1 in the Hawaiian sample in the ESR1 gene, for the Whites and the Asians in the FSHR gene, and for E2 in the Latinas for SNPs in the FSHR gene (Supplemental Table 2). These results may reflect true allelic heterogeneity revealed some heterogeneity, in concordance with previous results (24), or a lack of power due to the small sample sizes within the strata.

In the present analysis, BMI, menopausal status, age, cohort, and laboratory batch explained a large part of between-subject variance in circulating hormone levels. The unadjusted coefficient of determination (R²) values were 28.3% for SHBG, 42.7% for E1, 73.3% for E2, and 16.5% for DHEAS (Table 2). The inclusion of the SNPs did not notably increase the variance explained. Only for the SNPs rs2955617 and rs9913778 in SHBG with respect to SHBG (1.8% at max), and rs10046 and rs749292 in CYP19 with respect to E1 and E2 (1.2% at max) did the corresponding unadjusted R² exceed 1%. This low range of R² values for genetic associations is consistent with previous reports (9, 13).

In our analyses, none of the SNPs that were significantly associated with serum hormone levels were found to be significantly associated with breast cancer risk. These observations may seem contradictory, because serum levels of the same hormones have been systematically found to be associated with cancer risk in a number of prospective cohort studies, including EPIC and the NHS. Most other large-scale analyses, however, also showed a lack of association of breast cancer risk with polymorphic gene variants that did have a significant association with serum hormone concentrations (7). Because of the intra-individual variation in hormone levels over time, the true proportion of explained variation might be slightly higher than that estimated (at best 1–2%). Nevertheless, the variation in hormone levels explained by the SNPs may be too small to also observe a significant association of the gene variants with breast cancer risk, assuming that such association would indeed be mediated exclusively by changes in circulating levels.