SLCO1B1 polymorphisms and plasma estrone conjugates in postmenopausal women with ER+ breast cancer: genome-wide association studies of the estrone pathway

Abstract

Background

Estrone (E1), the major circulating estrogen in postmenopausal women, promotes estrogen-receptor positive (ER+) breast tumor growth and proliferation. Two major reactions contribute to E1 plasma concentrations, aromatase (CYP19A1) catalyzed E1 synthesis from androstenedione and steroid sulfatase (STS) catalyzed hydrolysis of estrone conjugates (E1Cs). E1Cs have been associated with breast cancer risk and may contribute to tumor progression since STS is expressed in breast cancer where its activity exceeds that of aromatase.

Methods

We performed genome-wide association studies (GWAS) to identify SNPs associated with variation in plasma concentrations of E1Cs, E1, and androstenedione in 774 postmenopausal women with resected early-stage ER+ breast cancer. Hormone concentrations were measured prior to aromatase inhibitor therapy.

Results

Multiple SNPs in SLCO1B1, a gene encoding a hepatic influx transporter, displayed genome-wide significant associations with E1C plasma concentrations and with the E1C/E1 ratio. The top SNP for E1C concentrations, rs4149056 (p = 3.74E–11), was a missense variant that results in reduced transporter activity. Patients homozygous for the variant allele had significantly higher average E1C plasma concentrations than did other patients. Furthermore, three other SLCO1B1 SNPs, not in LD with rs4149056, were associated with both E1C concentrations and the E1C/E1 ratio and were cis-eQTLs for SLCO1B3. GWAS signals of suggestive significance were also observed for E1, androstenedione, and the E1/androstene-dione ratio.

Conclusion

These results suggest a mechanism for genetic variation in E1C plasma concentrations as well as possible SNP biomarkers to identify ER+ breast cancer patients for whom STS inhibitors might be of clinical value.

Introduction

Estrogens are important drivers of the growth and proliferation of normal mammary tissue [1, 2] and of estrogen-receptor positive (ER+) breast cancer [3]. Estrone (E1) and estradiol (E2) are the two major active estrogens. E1 is more abundant than E2 in postmenopausal women, the group with the highest incidence of breast cancer in the United States [4]. E1 activates ER, and potentiates estrogenic effects such as the growth and survival of hormone-responsive breast cancer. However, since E1 can be converted to E2 by 17-beta-hydroxysteroid dehydrogenase, the estrogenic effects observed for E1 may be driven by E2 as it has a much higher affinity for ER [5]. There are two major precursors for E1, androstenedione and estrone-3-sulfate (E1S). In post-menopausal women, androstenedione is converted to E1 by aromatase (CYP19A1) in adipose tissue, the adrenal glands, liver, and in breast tumors. E1 can then be conjugated to form E1S [6]. E1S is biologically inactive, but is a “reservoir” for active estrogen since it can be hydrolyzed to E1 by steroid sulfatase (STS) [7]. Circulating E1S concentrations are an order of magnitude higher than those of E1 in post-menopausal women [8–10]. The major site of sulfation and desulfation of estrone is the liver, and this cycling of estrone is believed to be an intracellular mechanism for regulating estrogen activity [11–13].

Epidemiological studies have demonstrated a positive association between circulating E1 or E1S concentrations and breast cancer risk [14–18]. Furthermore, E1 concentrations in breast tumors are higher than in the plasma of postmenopausal women [10, 19, 20]. This may be due to the in situ biosynthesis of estrogens from both androgens and conjugated estrogen precursors. Aromatase inhibitors (AI), the major endocrine therapy for ER+ tumors in the adjuvant setting [21], target only one source of estrogens, i.e., their synthesis from androgens. That leaves another source uninhibited, i.e., synthesis from E1S catalyzed by STS. Therefore, drugs that target this source of estrogens, i.e., STS inhibitors [22–25], have been developed as an additional approach to the treatment of ER+ breast cancer.

In this era of “Precision Medicine,” understanding factors that govern individual variation in plasma concentrations of estrogens might help us understand their role in individual variation in risk for breast cancer occurrence and/or recurrence. Together with factors such as body mass index (BMI), age, and diet [14], genetics also influences circulating estrogen concentrations. Polymorphisms in genes such as CYP19A1, ESR1, and SHBG have all been shown to affect circulating estrogen concentrations [26–28]. However, application of an agnostic approach such as GWAS might help us to identify additional genetic factors. For example, Liu et al. [29] in a GWAS for E2 plasma concentrations in post-menopausal women with ER+ breast cancer, identified a novel genome-wide significant SNP signal in TSPYL5, and demonstrated that TSPYL5 encoded a transcription factor that regulated the expression of CYP19A1 in a SNP-dependent fashion. Most published GWAS for estrogens in post-menopausal women have focused on E2 [29, 30], rather than the more abundant E1, and its precursor E1S. No GWAS results have been published for plasma E1, E1S, or androstenedione in postmenopausal women. We hypothesized that genetic factors might contribute to variation in the concentrations of plasma hormones in the estrone biosynthesis and metabolism pathways, so we performed a comprehensive series of GWAS for plasma concentrations of these hormones in postmenopausal women with ER+ breast cancer prior to adjuvant AI therapy. We also performed GWAS of ratios of hormones and their precursors as phenotypes to “isolate” specific reactions in this pathway. The studies described subsequently identified genome-wide significant SNP signals associated with variation in plasma E1C and the E1C/E1 ratio across the SLCO1B1 gene, a gene encoding OATP1B1, an E1S transporter. These observations might ultimately help us identify individualized approaches for the therapy and/or prevention of ER+ breast cancer.

Materials and methods

The Mayo/MDA/MSK AI pharmacogenomics study

Details of the Mayo/MDA/MSK study were described previously [29, 31, 32]. Briefly, our study population consisted of 774 postmenopausal women with resected early-stage ER+ breast cancer. These patients were recruited from three Mayo Clinic locations (i.e., Rochester, MN, Jacksonville FL, and Scottsdale, AZ) as well as from MD Anderson Cancer Center and Memorial Sloan Kettering Cancer Center. Demographic information about the study population is listed in Table 1. Prior to the initiation of anastrozole therapy, plasma samples were collected and assayed for E2, E1, and total E1 conjugates (E1C), which included both sulfate and glucuronide conjugates (with E1 sulfate as the predominant conjugate), testosterone, and androstenedione using gas chromatography–tandem mass spectrometry. DNA from the subjects was genotyped for over 600,000 SNPs using the Ilumina Human610-Quad BeadChip. SNPs with minor allele frequencies (MAFs) <0.01 and/or Hardy–Weinberg equilibrium p values <1.0E–06 were excluded from the analyses. Approximately 8 million SNPs were imputed using reference populations from the 2010 release of the 1000 Genomes Project. Reference SNPs with MAF values <0.005 were removed prior to imputation performed using BEAGLE v3.3.1 [33]. Imputed SNPs with dosage r2 values <0.3 and/or MAF values <0.01 were excluded from analyses. See supplemental information for details.

Table 1.

Patient Characteristics for 774 postmenopausal women with ER+ breast cancer

Total	N = 774
Mean age, years (range)	67.9 (39.7–96.0)
Mean BMI, kg/m2 (range)	27.97 (16.0–57.2)
Current smoker
No	720 (93%)
Yes	43 (5.6%)
Unknown	11 (1.4%)
ER/PgR status
Positive/negative	152 (20%)
Positive/positive	614 (79%)
Negative/positive	7 (1.0%)
Positive/unknown	1 (0.1%)
HER2 status
Negative	669 (86.4%)
Positive	91 (11.8%)
Unknown	14 (1.8%)
Prior tamoxifen
No	692 (89%)
Yes	82 (11%)
Prior hormone replacement therapy
No	295 (38.1%)
Yes	364 (47.0%)
Unknown	115 (14.9%)
Prior chemotheraphy
No	463 (60%)
Yes	311 (40%)
Race
European ancestry	707 (91%)
African ancestry	52 (7%)
Asian ancestry	15 (2%)
Recruitment site
MCA	78 (10%)
MCF	32 (4%)
MCR	182 (24%)
MDA	349 (45%)
MSK	133 (17%)

Recruitment sites MCA Mayo Clinic Arizona, MCF Mayo Clinic Florida, MCR Mayo Clinic Rochester, MDA MD Anderson Cancer Center, MSK Memorial Sloan Kettering Cancer Center

Statistical analyses

The phenotypes analyzed were plasma concentrations of E1, E1C, androstenedione, and the ratios of E1/androstenedione and E1C/E1. These values were analyzed as continuous quantitative variables. Van Der Waerden transformations [34] were applied to phenotypes with skewed distributions to satisfy the assumption of a Gaussian distribution. To identify potential confounders, univariate linear regression analyses were performed with the hormone phenotypes and clinical variables such as BMI, age at diagnosis, recruitment site, smoking history, and eigenvectors from a principal component analysis (PCA) of population substructure. Covariates with the association of p values of 0.01 or lower were included in a multivariate linear regression model for the GWAS. For each phenotype, two analyses were performed, one using an unadjusted model without covariates, and another in which we adjusted for relevant covariates including eigenvectors from a PCA substructure analyses. The best model was determined by examining Quantile–Quantile (Q–Q) plots of the observed association of p values for adherence to the null distribution. This strategy was used for the genotyped SNP analyses. The best model emerging from the application of this approach was then applied to the analyses of the imputed SNPs. We used a p value of 5.0E–08 as the threshold for genome-wide significance. All analyses were performed using R statistical computing software (v3.0.2) and Plink (v1.07) [35].

Results

Associations with clinical variables

Associations among E1, E1C, androstenedione, E1C/E1, and E1/androstenedione and relevant clinical variables are listed in Table 2. E1 and E1C displayed significant positive associations with BMI as did the E1/androstenedione ratio. E1C/E1 was not associated with BMI. Age at the time of diagnosis was positively associated with E1 and the E1/androstenedione ratio, and was negatively associated with E1C/E1. Age did not appear to be associated with the plasma concentrations of androstenedione or E1C. Smoking status was only associated with E1C concentrations, with smokers having a lower mean E1C concentration than did non-smokers (p = 0.01). Women with prior exposure to hormone replacement therapy had lower mean E1C concentrations than did those who had not been treated with those agents (p = 0.01). With regard to race, the only significant association observed in our data involved the E1/androstenedione ratio. African-American women had a higher mean E1/androstenedione than did patients of European ancestry (p = 1.0E–03) (Table 2).

Table 2.

Association between plasma hormone concentrations or ratios of concentrations and clinical and demographic variables among 774 postmenopausal women with ER+ breast cancer

Variable	Androstenedione		Estrone		Estrone Conjugates		E1C/E1		E1/Androstenedione
			(E1)		(E1C)
	β	P	β	P	β	P	β	P	β	P
BMI	0.01	0.03	0.06	6.60E–24	0.04	3.30E–13	0.001	0.84	0.05	8.50E–19
Age at treatment	−0.01	0.2	0.01	3.00E–03	0	0.91	−0.01	0.01	0.02	1.60E–05
Her2 status
Positive versus negative	0.09	0.42	0.02	0.87	−0.11	0.33	−0.15	0.2	−0.13	0.26
Prior chemotherapy
Yes versus no	−0.09	0.2	−0.11	0.14	−0.02	0.8	0.07	0.35	−0.05	0.52
Prior tamoxifen use
Yes versus no	0.05	0.64	−0.08	0.51	0.14	0.23	0.3	0.01	−0.18	0.12
Prior HRT use
Yes versus no	−0.15	0.05	−0.15	0.05	−0.21	0.01	−0.12	0.15	0.01	0.95
Current smoker
Yes versus no	0.12	0.43	−0.25	0.11	−0.43	0.01	−0.33	0.04	−0.41	0.01
STRUCTURE-defined race
European ancestry	reference
African ancestry	−0.26	0.07	0.19	0.18	0.01	0.94	−0.15	0.29	0.48	1.00E–03
Asian ancestry	−0.23	0.39	−0.13	0.62	−0.06	0.81	−0.05	0.85	0.14	0.6

β values represent the coefficient from univariate linear regression analyses of the Van der Waerden transformed hormone quantitative phenotypes. Positive β values represent the positive associations between phenotype and covariate. Negative β values represent negative associations between phenotype and covariate. Bolded P values are statistically significant

E1C GWAS and the SLCO1B1 SNP signal

In the GWAS for circulating E1C concentrations (Q–Q plots in Supplementary Fig. 1), we observed a genome-wide significant SNP signal across SLCO1B1, a gene that encodes a solute carrier (SLC) transporter that is highly expressed in liver (Figs. 1A, 2A, Supplementary Table 1). The minor allele (C) of the top SNP, rs4149056 (MAF = 0.14), was associated with increased E1C concentrations with a p value of 3.74E–11 (Table 3). This non-synonymous (ns) SNP (521T>C) resulted in an amino acid change from valine to alanine (Val174Ala) at residue 174 of the OATP1B1 protein encoded by SLCO1B1, resulting in a transporter with reduced activity [36]. OATP1B1 is involved in hepatic uptake from the systemic circulation of both endogenous molecules such as E1C and xenobiotics such as statins [37]. Given the reduced activity conferred by this SNP, patients with the C allele had higher concentrations of E1C in the circulation compared to those with the T allele (CC: mean = 873.6 pg/mL, TT: mean = 285.6 pg/mL) (Fig. 3). Higher circulating E1C results in greater systemic exposure to E1C by tissues such as the breast.

Fig. 1.

Manhattan plots showing GWAS results for plasma concentrations of (A). Estrone Conjugates (E1C), (B). the ratio of E1C/E1, (C). Estrone (E1) in 774 postmenopausal women with ER+ breast cancer prior to the initiation of endocrine therapy. Dashed lines represent thresholds for genome-wide significance (p ≤ 5.0E–08, green) and suggestive significance (p ≤ 5.0E–06, gray)

Fig. 2.

Regional plots of the genome-wide significant SNP signal across SLCO1B1 from (A). GWAS of E1C which shows the top SNP, rs4149056 (blue arrow), a missense variant that results in a plasma membrane transporter with reduced activity, (B) GWAS of E1C/E1 which shows the top SNP, rs10841753 (purple point), which is an intronic variant that is in high LD (R² = 1.00, CEU population, distance = 8443 bp), with another missense variant, rs11045819 (red arrow) in populations of European descent, but it is a signal independent from rs4149056 (R² = 0.013, CEU population, distance = 101179 bp). The black arrow highlights 3 SNPs (rs11045828, rs10841755, and rs11045834) that are cis-eQTLs for SLCO1B3 and are associated with breast cancer risk

Table 3.

List of top and selected SNP signals from the GWAS analyses

Phenotype	SNP	CHR	BP	Minor allele	MAF	Effect size	Direction of effecta	SE	P	Gene	Gene location	SNP function	Cis-eQTL
E1C	rs4149056	12	21222816	C	0.136	0.50	↑	0.075	3.74E–11	SLCO1B1	Exon	Missense	–
E1C	rs2900478	12	21368797	A	0.14	0.48	↑	0.077	6.87E–10	SLCO1B1	Intron	–	–
E1C	rs4149083	12	21380630	T	0.14	0.48	↑	0.077	7.29E–10	SLCO1B1	Intron	–	–
E1C	rs12367888	12	21347021	T	0.15	0.45	↑	0.073	1.40E–09	SLCO1B1	Intron	–	–
E1C	rs58310495	12	21357711	T	0.15	0.45	↑	0.073	1.41E–09	SLCO1B1	Intron	–	–
E1C/E1	rs10841753	12	21321370	C	0.208	−0.43	↓	0.061	2.95E–12	SLCO1B1	Intron	–	–
E1C/E1	rs11045828	12	21338231	A	0.307	−0.35	↓	0.052	5.33E –11	SLCO1B1	intron	–	SLCO1B3
E1C/E1	rs10841755	12	21338889	A	0.306	−0.35	↓	0.052	8.13E–11	SLCO1B1	Intron	–	SLCO1B3
E1C/E1	rs11045834	12	21232363	A	0.302	−0.34	↓	0.053	3.08E –10	SLCO1B1	intron	–	SLCO1B3
E1C/E1	rs4149056	12	21222816	C	0.136	0.42	↑	0.071	8.67E–09	SLCO1B1	Exon	Missense	–
E1C/E1	rs11045819	12	21329813	A	0.158	−0.38	↓	0.068	3.87E–08	SLCO1B1	Exon	Missense	–
E1/Andr	rs10953024	7	90504067	C	0.088	−0.49	↓	0.100	1.45E–06	CDK14	Intron	–	STEAP1
E1/Andr	rs1453308	2	237186325	A	0.198	−0.33	↓	0.069	2.06E–06	ASB18	5′upstream	–	–
E1/Andr	rs16952550	18	7807948	C	0.171	−0.33	↓	0.069	2.09E–06	PTPRM	Intron	–	–
E1/Andr	rs10224911	7	90539427	A	0.083	−0.49	↓	0.102	2.11E–06	CDK14	Intron	–	–
E1/Andr	rs17601876	15	51553909	A	0.500	0.24	↑	0.052	3.50E–06	CYP19A1	Intron	–	CYP19A1
E1	rs12586722	14	70890140	A	0.155	0.30	↑	0.063	1.48E–06	LOC105370555	5′ upstream	–	ADAM20P1, AL357153.1
E1	rs12436808	14	70904607	C	0.156	0.30	↑	0.063	2.19E–06	LOC105370555	Intron	–	ADAM20P1, AL357153.1
E1	rs3764182	14	70919161	A	0.156	0.30	↑	0.063	2.21E–06	ADAM21	Intron	–	ADAM20P1, AL357153.1
E1	rs114864695	14	70924955	A	0.162	0.31	↑	0.064	2.55E–06	ADAM21	Exon	Missense	ADAM20P1, AL357153.1
Andr	rs57712673	19	14885800	A	0.349	0.36	↑	0.067	1.39E–07	EMR2/ADGRE2	Intron	–	EMR2
Andr	rs1108015	19	14886784	C	0.354	0.35	↑	0.067	2.10E–07	EMR2/ADGRE2	Intron	–	EMR2
Andr	rs59964204	7	36939322	A	0.096	−0.45	↓	0.087	3.19E–07	ELMO1	Intron	–	–
Andr	rs34346910	10	88427209	C	0.111	0.49	↑	0.097	5.67E–07	LDB3	5′upstream	–	–
Andr	rs6988985	8	144007104	C	0.461	0.25	↑	0.050	6.65E–07	CYP11B2	5′ upstream	–	LYNX1, LY6E
Andr	rs3802230	8	143992864	A	0.433	0.24	↑	0.047	6.88E–07	CYP11B2	3′UTR	–	LYNX1, LY6E

Expression quantitative trait loci (eQTL) information was obtained from the GTEx database. An eQTL is a variant or SNP that is associated with the mRNA expression of a given gene

^aDirection of effect = (↑) indicates that the minor allele of the SNP is associated with an increase in the hormone or hormone ratio. Direction of effect = (↓) indicates that the minor allele of the SNP is associated with a decreased in the hormone or hormone ratio

Fig. 3.

Mean plasma E1C concentrations for each genotype of rs4149056 (with 95% confidence intervals shown) in postmenopausal women with ER+ breast cancer. Red points with red horizontal bars indicate the mean values for each genotype

As anticipated, there was a strong association between plasma concentrations of E1C and those of its precursor, E1 (Pearson’s r = 0.63, p = 5.15E–83). However, regression analyses showed that there was no direct effect of the rs4149056 nsSNP on E1 concentrations, but the positive relationship between E1 and EIC concentrations was modified by rs4149056 (Supplementary Table 2). Specifically, individuals homozygous for the variant allele had, on average, a smaller increase in E1 per unit increase in E1C concentrations, whereas those homozygous for the wildtype genotype had, on average, a greater increase in E1 per unit increase in E1 (Supplementary Fig. 2)—compatible with the functional effect of the rs4149056 C allele. This suggests that even when there was a similar quantitative increase in E1C among our subjects, the subsequent increase in E1 was smaller in women who were homozygous variant for the SLCO1B1 missense SNP. Perhaps, this might be due to fewer EIC molecules entering the liver for desulfation in subjects with decreased OATP1B1 transport activity.

E1C/E1 GWAS and the SLCO1B1 SNP signal

The Manhattan plot for the GWAS of the E1C/E1 ratio is shown in Fig. 1B (Q–Q plots in Supplementary Fig. 3). Numerous SNPs in this GWAS achieved genome-wide significance (Supplementary Table 3) and they all mapped to SLCO1B1 (Fig. 2). The linkage disequilibrium (LD) structure among the top SNPs suggested the existence of two independent signals in the SLCO1B1 gene. The rs4149056 nsSNP that was the top GWAS signal for E1C concentration also displayed a genome-wide significant association with E1C/E1 (p = 8.67E–09) (Table 3). However, this SNP was not in LD (r² = 0.013) with the top SNP for the E1C/E1 GWAS, rs10841753 (p = 2.95E–12), (see Fig. 2B) an intronic SNP with a minor allele C that was associated with decreased E1C/E1. This SNP was not an expression quantitative trait locus (eQTL) for SLCO1B1, but it was in perfect LD (r² = 1) with another SLCO1B1 missense variant, rs11045819, that was also associated with E1C/E1 (p = 3.87E–08) (Table 3). This second missense variant, rs11045819 (463C>A), resulted in a proline to threonine (Pro155Thr) change in the protein encoded by SLCO1B1 and had a minor allele (A) frequency of 15% in our sample of women who were primarily of European ancestry. This variant was absent in populations of East Asian ancestry and had a MAF of 6% in populations of African ancestry based on 1000 Genomes data [38]. Unlike rs4149056, rs11045819 does not appear to alter the activity of the OATP1B1 transporter [36].

Three of the top SNPs from this GWAS, rs11045828 (p = 5.33E–11), rs10841755 (p = 8.13–11), and rs11045834 (p = 3.08E–11) (Table 3), mapped to an SLCO1B1 intron and were significant eQTLs for the nearby SLCO1B3 gene, but not for SLCO1B1 (GTEx) [39]. For each of these SNPs, the variant allele was associated with decreased E1C/E1 and with decreased expression of SLCO1B3 (rs11045828: effect size for expression = −0.4, p = 1.6E–06; rs10841755: effect size = −0.4, p = 1.7E–06; 11045834: effect size = −0.38 p = 4.7E–06). SLCO1B3 also encodes a hepatic influx transporter, maps 200 kb upstream of SLCO1B1, and transports E1Cs [37]. All three of the SNPs that were eQTLs for SLCO1B3 were in high LD [r² = 1.0] with each other. In addition to being associated with E1C/E1, these SNPs were also significantly associated with decreased E1C concentrations (rs11045828: p = 5.68E–06, rs10841755: p = 8.68E–06, 11045834: p = 1.61E–06) but not with E1 concentrations. Since SLCO1B3 encodes an influx transporter one might anticipate that SNPs associated with decreased expression of SLCO1B3 might be associated with increased E1C concentrations, but our data clearly demonstrate the opposite.

E1 GWAS

We also performed a GWAS for E1 concentrations (Fig. 1C, Supplementary Fig. 4) and the top SNP was rs12586722 which mapped to LOC105370555 on chromosome 14 (Table 3, Supplementary Table 4), but this SNP did not achieve genome-wide significance. The minor allele A (MAF = 0.16) for this SNP was associated with increased E1 concentrations (p = 1.48E–06). This SNP was a cis-eQTL for ADAM20P1, a nearby processed pseudogene [39]. The minor allele for the SNP was associated with decreased expression. Another SNP of possible interest in this region was rs114864695 (p = 2.55E–06), a missense variant in exon 2 of ADAM21 (Table 3). It resulted in a 739A>C change in the genomic sequence and a Lys247Gln change in the encoded amino acid sequence. The functional consequences of this missense variant are unknown. According to GTEx, this SNP is also a cis-eQTL for ADAM21P1, and the C allele is associated with increased expression of this putative pseudogene (effect size = −0.49, p = 1.8E–09).

E1/androstenedione ratio GWAS

Although there were no SNPs that reached genome-wide significance in the E1/androstenedione GWAS (Supplementary Fig. 5), there were several SNPs with suggestive significance. Those polymorphisms included SNPs in or near CDK14, ASB18, PTPRM, and CYP19A1 (Table 3, Supplementary Table 5). CYP19A1 was of interest since it encodes aromatase, the enzyme that catalyzes the conversion of androstenedione to E1 and of testosterone to E2. The minor allele (G) for the top SNP in this gene, rs17601876 (p = 3.50E–06), was associated with an increased E1/androstenedione (Table 3). Several of the SNPs from this GWAS were cis-eQTLs for CYP19A1, including rs7175531 (MAF = 0.34, p = 6.18E–05) and rs2414095 (MAF = 0.34, p = 7.02E–05). Minor alleles for both SNPs were associated with decreased E1/androstenedione among our patients and with decreased CYP19A1 expression [39]. Since decreased CYP19A1 would result in decreased conversion of androstenedione to E1, the directional effects for both relationships were consistent.

Androstenedione GWAS

In our androstenedione GWAS (Supplementary Fig. 6c), the main signal included intronic SNPs for EMR2/ADGRE2, with rs57712673 having the lowest p value (p = 1.39E–07) (Table 3). The minor allele for this SNP (A) was associated with increased androstenedione concentrations as well as decreased expression of EMR2 (effect size = −0.2, p = 8.0E–07). Another signal of interest involved a cluster of SNPs on chromosome 8 within and flanking CYP11B1 and CYP11B2 (Supplementary Fig. 6c, Supplementary Table 6), genes encoding enzymes in the steroid hormone biosynthesis pathway. Within this region, the minor allele (T) of the SNP with the lowest p value, rs6988985, was associated with increased androstenedione (p = 6.65E–07) (Table 3). CYP11B1 expression is regulated by adrenocorticotrophic hormone (ACTH) which regulates the quantity of 17-hydroxyprogesterone available for conversion to androstenedione in the adrenal glands.

Discussion

The major signal observed in our E1C GWAS was a missense variant in SLCO1B1, rs4149056. GWAS for the ratio of E1C/E1 yielded the same signal, as well as an additional independent signal that included another SLCO1B1 missense variant, rs11045819. These two nsSNPs are approximately 1700 bp apart and map to adjacent SLCO1B1 exons, 5 and 6, but they are not in LD with each other in European, African, or Asian populations. The variant allele for rs4149056 (521T>C) results in a Val174Ala amino acid change in the fourth transmembrane segment of the OATP1B1 protein. This variant is a well-studied pharmacogenomic marker identified initially during a GWAS for statin-induced myopathy [40]. This variant has also been associated with concentrations of bilirubin [41], sex hormone-binding globulin [42], and various metabolites [43]. OATP1B1, the transporter encoded by SLCO1B1, is expressed primarily in hepatic tissue, and transports numerous endogenous compounds, including E1Cs (E1S), from the systemic circulation into hepatocytes [37]. Functional studies have shown that the rs4149056 variant allele is associated with decreased membrane expression of the OATP1B1 transporter, and thus, reduced transport activity [36]. Reduced transport would increase the plasma concentration and systemic exposure to OATP1B1 substrates, which is compatible with our observations. In our patients, those with one or more copies of the rs4149056 C variant allele had higher mean plasma concentrations of E1C than patients homozygous for the major allele (CC: 873.6 pg/mL, TT: 285.6 pg/mL) (Fig. 3). Thus, women who carried the minor allele had increased systemic exposure to E1Cs.

The second SLCO1B1 missense variant, rs11045819 (463C>A), resulted in a Pro155Thr change in the extra-cellular loop region of OATP1B1. However, in vitro studies failed to show that 463C>A affects either protein expression for the encoded transporter or its uptake activity [32]. In our patients, the variant allele (A) for this SNP was associated with decreased E1C/E1, and thus a lower exposure to circulating E1C. There were no patients in our sample who were homozygous for variants for both mis-sense SNPs. The 463C>A variant might be significantly associated with E1C/E1 because it is in modest LD (r² = 0.517) with 3 intronic SNPs that displayed statistically significant associations with the E1C/E1 ratio, and which were also cis-eQTLs for SLCO1B3. That gene encodes OATP1B3, a solute carrier transporter that is also capable of transporting E1Cs, but with lower transport efficiency than OATP1B1 [44].

Both the non-synonymous and other variants in SLCO1B1 associated with elevated plasma E1C concentrations may be clinically important since elevated E1C concentrations have previously been linked to increased breast cancer risk in postmenopausal women [15, 16]. In addition, ten SNPs in SLCO1B1 were previously shown to be associated with increased risk for breast cancer in postmenopausal women in the California Teachers Study (CTS) [45]. Although none of those SNPs showed genome-wide significant associations with E1C concentrations in our data, seven were nominally associated, and one SNP (rs4149058) exhibited suggestive statistical significance (p = 5.02E–06). That SNP was associated with a 23% increase in breast cancer risk in the CTS and with increased E1C concentrations in our study. Rs4149058 was in modest LD (r² = 0.59) with the rs4149056 missense variant that was the top SNP signal in our E1C GWAS. Rs4149056 was not genotyped in the CTS, nor were most of our variants that displayed genome-wide significant associations with E1C or E1C/E1.

The increase in systemic exposure to E1C by ER+ breast cancer patients who were homozygous variant for rs4149056 may mean that their tumors take up more E1C, a precursor for E1. Since SLCO1B1 is not expressed in either normal or cancerous breast tissue according to the GTEx (39) and TCGA databases (46–47), one can assume that this missense SNP will not affect E1C transport into breast tumor tissue. Other OATPs that have the capacity to transport E1S (33), such as OATP2B1, OATP3A1, and OATP4A1, are highly expressed in breast tumors. Those OATPs tend to be ubiquitously expressed, but have lower hepatic expression than does OATP1B1 (35). Furthermore, TCGA data indicate that the breast tumor expression of these genes is comparable to key breast cancer genes such as ESR1 and CCND1 [46, 47]. Hypothetically, elevated circulating E1Cs could indirectly augment the proliferation of breast tumors that have the capacity to take up and hydrolyze E1Cs to form E1. In kinetic studies, ER+ cell lines have much higher rates of E1C uptake than do normal breast cell lines [48, 49]. Furthermore, ER+ cell lines had lower Km values for OATP-mediated transport of E1Cs compared to other breast cancer cell lines, indicating that ER+ cell lines may have greater transport efficiency [49]. Higuchi et al. [50] demonstrated up-regulation of the expression of the OATP1B1, OATP1A2, OATP4A2, and OATP5A1 transporters in a letrozole-resistant MCF7 breast cancer cell line. Those cell lines were shown to proliferate preferentially in E1C supplemented media, a situation analogous to elevated circulating plasma concentrations. They also showed that treatment with an STS inhibitor reduced proliferation of the letrozole-resistant cell lines, results that support the possible clinical application of STS inhibitors. Further highlighting the biological and potential clinical importance of the STS-mediated pathway for estrogen production, Utsumi et al. [51] demonstrated that patients with high steroid sulfatase [STS] expression in breast tumors had a higher risk of recurrence than those with low expression. This implies that higher plasma E1C exposure, higher tumor E1C uptake, and higher STS expression might contribute to risk for tumor recurrence. Therefore, STS inhibitors might prove beneficial for ER+ breast cancer patients with high circulating E1C during adjuvant AI therapy. Selection of patients for STS inhibitor trials has been a challenge, and it is possible that rs4149056 could help address this issue. We assume that patients with high circulating concentrations of E1C might benefit from STS inhibition. Thus, patients who are homozygous variant for rs4149056 and are on an aromatase inhibitor could be selected for clinical trials of STS inhibitor as add-on therapy.

In summary, our GWAS of plasma E1 and its major precursors in postmenopausal women with ER+ breast cancer identified SNP signals across SLCO1B1 that point to a mechanism that might contribute to increased tumor exposure to estrogen. The SLCO1B1-encoded hepatic influx transporter regulates circulating concentrations of E1C, a “reservoir” for E1 that can increase the exposure of patients to active estrogens after hydrolysis of the conjugates. This pathway remains active in patients who are on aromatase inhibitor therapy. We found that two independent missense variants in SLCO1B1 were associated with either increased or decreased circulating E1C concentrations and/or E1C/E1 ratio—results that require replication in future studies. These variants may provide a means by which to identify ER+ breast cancer patients on AI therapy who are at increased risk for tumor recurrence because of elevated tumor E1C exposure, and who might benefit from additional therapy such as STS inhibition.