Post-GWAS gene–environment interplay in breast cancer: results from the Breast and Prostate Cancer Cohort Consortium and a meta-analysis on 79 000 women

Abstract

We studied the interplay between 39 breast cancer (BC) risk SNPs and established BC risk (body mass index, height, age at menarche, parity, age at menopause, smoking, alcohol and family history of BC) and prognostic factors (TNM stage, tumor grade, tumor size, age at diagnosis, estrogen receptor status and progesterone receptor status) as joint determinants of BC risk. We used a nested case–control design within the National Cancer Institute's Breast and Prostate Cancer Cohort Consortium (BPC3), with 16 285 BC cases and 19 376 controls. We performed stratified analyses for both the risk and prognostic factors, testing for heterogeneity for the risk factors, and case–case comparisons for differential associations of polymorphisms by subgroups of the prognostic factors. We analyzed multiplicative interactions between the SNPs and the risk factors. Finally, we also performed a meta-analysis of the interaction ORs from BPC3 and the Breast Cancer Association Consortium. After correction for multiple testing, no significant interaction between the SNPs and the established risk factors in the BPC3 study was found. The meta-analysis showed a suggestive interaction between smoking status and SLC4A7-rs4973768 (P_interaction = 8.84 × 10⁻⁴) which, although not significant after considering multiple comparison, has a plausible biological explanation. In conclusion, in this study of up to almost 79 000 women we can conclusively exclude any novel major interactions between genome-wide association studies hits and the epidemiologic risk factors taken into consideration, but we propose a suggestive interaction between smoking status and SLC4A7-rs4973768 that if further replicated could help our understanding in the etiology of BC.

INTRODUCTION

Several genome-wide association studies (GWAS) have identified single-nucleotide polymorphisms (SNPs) associated with breast cancer (BC) risk (1–17). The possible interplay between genetic variants and established epidemiologic BC risk factors is gradually being explored (18–22). Finding gene–environment interactions can be useful in several areas such as allowing a more specific risk assessment that could be useful for early detection or prevention strategies and moreover to further our understanding of biological pathways and mechanisms of disease etiology.

In a previous work conducted in a smaller set of cases and controls in the context of the National Cancer Institute (NCI)'s Breast and Prostate Cancer Cohort Consortium (BPC3), we have reported the lack of interactions between 17 GWAS SNPs and 9 epidemiologic risk factors for BC (21). The results from other groups were similar to what we found; however, a recent large study by Nickels et al. performed in the context of the Breast Cancer Association Consortium (BCAC) showed several highly significant gene–environment interactions (22). Given these new findings, we felt the need to extend our previous work and doubling our overall sample size, we studied a further 22 SNPs reported to show genome-wide statistically significant associations with BC risk. We also used updated and extended information on established BC risk factors (body mass index (BMI), height, age at menarche, parity, age at first full-term pregnancy, number of full-term pregnancies, age at menopause, smoking, pack-years of smoking, alcohol, family history of BC and ever use of oral contraceptives), as well as on prognostic factors of BC (estrogen receptor (ER) status, progesterone receptor (PR) status, tumor size, TNM stage, tumor grade and age at diagnosis) available in the central BPC3 database. Both genetic and non-genetic information were available for a total of 35 661 individuals (16 285 cases and 19 376 controls).

In this study, first, we analyzed BC risk SNPs stratifying by the established BC risk factors and BC prognostic factors. Second, we tested for gene–environment interactions and, taking advantage of the work done by Nickels et al., we combined the interaction ORs from BPC3 and BCAC (22) in a meta-analysis. This was the largest effort up to date, using data on up to 79 000 individuals in order to discover any possible interplay between genes and environment in relation to BC risk.

RESULTS

In total, 16 285 BC cases and 19 376 controls of European descent from BPC3 were included in the analysis of this study. The relevant characteristics of the study subjects are presented in Supplementary Material, Table S1. At the time of recruitment, 88% of the subjects in this study were peri- or postmenopausal (14 468 cases, 16 761 controls). For Nurses' Health Study (NHS), two SNPs (ZMIZ1-rs1045485 and 11q13-rs614367) showed departure from the Hardy–Weinberg equilibrium among the controls (P = 8.4 × 10⁻⁴ and 6 × 10⁻⁴, respectively). Therefore, for the analyses involving these two SNPs, subjects from NHS were removed. The genotyping success rate was 98.88% in the study population.

SNPs main effects

The SNPs included in the analyses are listed in Table 1. The results of the main effects analyses of the association between the 39 SNPs and BC risk are shown in Table 2. In the overall analysis (considering ER− and ER+ together), we found significant associations (at the conventional 0.05 level) with BC risk for 29 of the SNPs (P-values ranging from 0.035 to 5.81 × 10⁻³²). The directions of the associations were consistent with those reported in previous papers for all the SNPs. The results of the tests of heterogeneity of the main effects of SNPs and BC risk factors across cohorts were not significant (data not shown). The results of the main effects analyses for the epidemiologic risk factors are reported in Supplementary Material, Table S2.

Table 1.

Information on the selected SNPs

SNP	Gene	Chr	Location (bp) (hg19)a	Major allele/minor allele	References
rs11249433b	NOTCH2	1p11	***121280363	T/C	(9,15)
rs10931936	CASP8	2q33	202143678	C/T	(23)
rs1045485b	CASP8	2q33	202149339	G/C	(4)
rs13387042b	Intergenic	2q35	219905582	A/G	(2)
rs4973768b	SLC4A7	3p24	27415763	C/T	(8)
rs4415084b,c	Intergenic	5p12	44662265	C/T	(5)
rs10941679b	Intergenic	5p12	44706248	A/G	(5)
rs10069690	TERT	5p15	1279540	C/T	(12)
rs889312b	MAP3K1	5q11	56031634	A/C	(1)
rs17530068	Intergenic	6q14	82192859	T/C	(15)
rs13437553	Intergenic	6q14	82303985	T/C	(15)
rs1917063d	Intergenic	6q14	82322957	C/T	(15)
rs9344191e	Intergenic	6q14	82197935	T/G	(15)
rs2180341b,f	RNF146	6q22	127600380	A/G	(6)
rs3757318	Intergenic	6q25	151913863	G/A	(10)
rs9383938	Intergenic	6q25	151987107	G/T	(15,24)
rs2046210b	Intergenic	6q25	151948116	C/T	(7,14)
rs13281615b	Intergenic	8q24	128355368	A/G	(8)
rs1562430	Intergenic	8q24	128387602	T/C	(10)
rs1011970	CDKN2BAS	9p21	22061884	G/T	(16)
rs865686	Intergenic	9q31	110888228	T/G	(16)
rs2380205	Intergenic	10p15	5886484	C/T	(16)
rs10995190	ZNF365	10q21	68278432	G/A	(16,25)
rs16917302	ZNF365	10q21	64260948	A/C	(11,25)
rs1250003g	ZMIZ1	10q22	80846564	T/C	(16,26)
rs3750817b	FGFR2	10q26	123332327	C/T	(20)
rs2981582b	FGFR2	10q26	123352067	C/T	(8)
rs3817198b	LSP1	11p15	1908756	T/C	(1)
rs909116	LSP1	11p15	1941696	T/C	(10)
rs614367	Intergenic	11q13	69328514	C/T	(17)
rs999737b,h	RAD51L1	14q24	69034432	C/T	(9)
rs3803662b	TNRC9	16q12	52586091	C/T	(1,2)
rs2075555b	COL1A1	17q21	48274041	C/A	(3)
rs6504950b	COX11	17q22	53056221	G/A	(8)
rs12982178	USHBP1	19p13	17371318	T/C	(12)
rs8170	C19Orf62	19p13	17389454	G/A	(15)
rs2284378i	RALY	20q11	32587845	C/T	(15)
rs4911414	Intergenic	20q11	32729194	G/T	(15)
rs311499j	GMEB2	20q13	62217339	C/T	(11)

^aGenome Reference Consortium Human, build 37 (http://genome.ucsc.edu/cgi-bin/hgGateway).

^bThis SNP was included in the first analyses conducted in this dataset.

^c5p12-rs4415084 or surrogate 5p12-rs920329.

^d6q14-rs1917063 or surrogate 6q14-rs9344208.

^e6q14-rs9344191 or surrogate 6q14-rs9449341.

^fECHDC1, RNF146-rs2180341 or surrogate ECHDC1, RNF146-rs9398840.

^gZMIZ1-rs1250003 or surrogate ZMIZ1-rs704010.

^hRAD51L1-rs999737 or surrogate RAD51L1-rs10483813.

ⁱRALY-rs2284378 or surrogate RALY-rs6059651, RALY-rs8119937.

^jGMEB2-rs311499 or surrogate GMEB2-rs311498.

Table 2.

Associations between selected SNPs and BC risk

SNP	Closest gene	Chr.	Cases			Controls			ORhet (95% CI)a	ORhom (95% CI)a	ORallele (95% CI)a	Ptrendb	P2d.f.c
			A/A	A/B	B/B	A/A	A/B	B/B
rs11249433	NOTCH2	1p11	4596	7232	2783	6097	8381	3055	1.14 (1.08–1.20)	1.21 (1.13–1.29)	1.10 (1.07–1.14)	7.06E−10	1.63E−09
rs10931936	CASP8	2q33	4470	3697	775	5762	4549	846	1.05 (0.99–1.12)	1.18 (1.06–1.31)	1.07 (1.03–1.12)	2.10 E−03	5.59 E−03
rs1045485	CASP8	2q33	9261	2690	200	10148	3058	236	0.96 (0.91–1.02)	0.94 (0.77–1.14)	0.96 (0.92–1.02)	1.75 E−01	3.98 E−01
rs13387042	Intergenic	2q35	4599	6945	3074	4618	8821	4203	0.79 (0.75–0.83)	0.73 (0.69–0.78)	0.85 (0.82–0.88)	2.66 E−24	3.04 E−26
rs4973768	SLC4A7	3p24	3647	7397	3572	4657	8847	4063	1.07 (1.01–1.13)	1.13 (1.06–1.21)	1.06 (1.03–1.10)	1.01 E−04	5.07 E−04
rs4415084d	Intergenic	5p12	2764	4125	1546	4218	5636	1964	1.11 (1.04–1.18)	1.19 (1.09–1.29)	1.09 (1.05–1.14)	1.69 E−05	8.33 E−05
rs10941679	Intergenic	5p12	7823	5750	1083	9912	6467	1075	1.13 (1.07–1.18)	1.29 (1.17–1.41)	1.13 (1.09–1.17)	2.42 E−11	2.00 E−10
rs10069690	TERT	5p15	7848	5652	1025	9810	6533	1197	1.08 (1.03–1.13)	1.07 (0.98–1.17)	1.06 (1.02–1.10)	2.19 E−03	3.21 E−03
rs889312	MAP3K1	5q11	7170	6042	1375	9155	7110	1388	1.09 (1.04–1.14)	1.26 (1.16–1.37)	1.11 (1.07–1.15)	3.41 E−09	1.37 E−08
rs17530068	Intergenic	6q14	8703	5769	928	10068	6285	976	1.06 (1.01–1.11)	1.11 (1.01–1.22)	1.06 (1.02–1.10)	2.69 E−03	1.09 E−02
rs13437553	Intergenic	6q14	3926	2492	400	5055	2978	460	1.08 (1.00–1.15)	1.15 (1.00–1.32)	1.07 (1.02–1.13)	9.42 E−03	3.42 E−02
rs1917063e	Intergenic	6q14	5780	3521	524	7385	4179	603	1.08 (1.02–1.14)	1.12 (0.99–1.27)	1.07 (1.02–1.12)	3.69 E−03	1.36 E−02
rs9344191f	Intergenic	6q14	4972	3566	645	6435	4317	741	1.07 (1.01–1.13)	1.14 (1.02–1.28)	1.07 (1.02–1.12)	4.04 E−03	1.60 E−02
rs2180341g	RNF146	6q22	5253	3244	529	7100	4583	722	0.95 (0.90–1.01)	0.99 (0.88–1.12)	0.97 (0.93–1.02)	2.51 E−01	2.54 E−01
rs3757318	Intergenic	6q25	7679	1443	66	9736	1645	56	1.13 (1.04–1.22)	1.40 (0.98–2.01)	1.14 (1.06–1.22)	5.05 E−04	2.10 E−03
rs9383938	Intergenic	6q25	7563	1568	104	9620	1841	86	1.08 (1.01–1.17)	1.56 (1.17–2.08)	1.11 (1.04–1.19)	1.60 E−03	1.57 E−03
rs2046210	Intergenic	6q25	5829	6700	2035	7304	7993	2313	1.05 (1.00–1.10)	1.10 (1.02–1.18)	1.05 (1.01–1.08)	4.35 E−03	1.68 E−02
rs13281615	Intergenic	8q24	2905	4306	1666	4231	5925	2164	1.06 (1.00–1.13)	1.13 (1.04–1.22)	1.06 (1.02–1.10)	2.46 E−03	1.01 E−02
rs1562430	Intergenic	8q24	5355	7123	2381	5549	8214	2995	0.90 (0.85–0.94)	0.82 (0.77–0.88)	0.90 (0.88–0.93)	8.35 E−10	5.71 E−09
rs1011970	CDKN2BAS	9p21	10500	4399	426	12020	4754	506	1.06 (1.01–1.12)	0.96 (0.84–1.09)	1.03 (0.99–1.08)	1.20 E−01	3.45 E−02
rs865686	Intergenic	9q31	6459	7072	1894	6814	8059	2486	0.92 (0.88–0.97)	0.81 (0.75–0.86)	0.90 (0.88–0.93)	1.11 E−09	4.81 E−09
rs2380205	Intergenic	10p15	5004	7435	2895	5388	8499	3412	0.94 (0.90–0.99)	0.91 (0.86–0.97)	0.95 (0.93–0.98)	3.44 E−03	1.07 E−02
rs10995190	ZNF365	10q21	11352	3663	285	12456	4463	361	0.89 (0.85–0.94)	0.83 (0.71–0.98)	0.90 (0.86–0.94)	2.73 E−06	1.49 E−05
rs16917302	ZNF365	10q21	7599	1574	86	9408	2060	101	0.94 (0.88–1.01)	1.04 (0.78–1.39)	0.96 (0.89–1.02)	1.86 E−01	2.58 E−01
rs1250003h	ZMIZ1	10q22	5637	7330	2345	6631	8074	2604	1.07 (1.02–1.12)	1.06 (1.00–1.14)	1.04 (1.01–1.07)	1.62 E−02	1.92 E−02
rs3750817	FGFR2	10q26	5510	6564	1963	5964	8330	2774	0.84 (0.80–0.89)	0.76 (0.70–0.81)	0.86 (0.84–0.89)	1.75 E−18	8.18 E−18
rs2981582	FGFR2	10q26	4712	7101	2774	6540	8431	2641	1.18 (1.12–1.24)	1.48 (1.39–1.59)	1.21 (1.17–1.25)	5.81 E−32	3.02 E−31
rs3817198	LSP1	11p15	4132	3881	947	5658	5463	1261	0.98 (0.92–1.03)	1.03 (0.94–1.14)	1.00 (0.96–1.04)	9.22 E−01	4.19 E−01
rs909116	LSP1	11p15	4320	7339	3198	4602	8264	3871	0.95 (0.90–1.00)	0.88 (0.83–0.94)	0.94 (0.91–0.97)	1.26 E−04	6.15 E−04
rs614367	Intergenic	11q13	9422	3567	411	9998	3359	348	1.13 (1.07–1.19)	1.25 (1.08–1.45)	1.12 (1.07–1.18)	5.15 E−07	3.30 E−06
rs999737i	RAD51L1	14q24	8557	4878	688	9979	6078	1010	0.93 (0.89–0.98)	0.80 (0.72–0.89)	0.91 (0.88–0.95)	2.81 E−06	9.31 E−06
rs3803662	TNRC9	16q12	6890	6177	1434	9290	6899	1311	1.21 (1.15–1.27)	1.49 (1.37–1.62)	1.21 (1.17–1.26)	2.41 E−28	3.14 E−27
rs2075555	COL1A1	17q21	6283	1987	175	8879	2740	226	1.04 (0.97–1.11)	1.09 (0.89–1.33)	1.04 (0.98–1.10)	2.01 E−01	4.37 E−01
rs6504950	COX11	17q22	7946	5664	1051	9245	7066	1348	0.93 (0.89–0.97)	0.91 (0.83–0.99)	0.94 (0.91–0.97)	6.88 E−04	2.13 E−03
rs12982178	USHBP1	19p13	6752	3308	388	8292	4052	520	1.00 (0.95–1.06)	0.93 (0.81–1.07)	0.99 (0.94–1.04)	6.15 E−01	5.50 E−01
rs8170	C19Orf62	19p13	10713	4780	557	12034	5359	654	1.00 (0.96–1.05)	0.96 (0.85–1.08)	0.99 (0.96–1.03)	7.82 E−01	7.59 E−01
rs2284378j	RALY	20q11	4354	3871	961	5286	4917	1170	0.96 (0.90–1.02)	1.01 (0.92–1.12)	0.99 (0.95–1.03)	6.20 E−01	2.85 E−01
rs4911414	Intergenic	20q11	4177	3954	1048	5132	5047	1311	0.97 (0.91–1.03)	1.00 (0.91–1.10)	0.99 (0.95–1.03)	6.03 E−01	5.04 E−01
rs311499k	GMEB2	20q13	7987	1162	66	9976	1505	70	0.96 (0.88–1.04)	1.13 (0.81–1.59)	0.98 (0.91–1.05)	5.57 E−01	4.64 E−01

Bp, base pair; chr, chromosome; CI, confidence interval; OR_het, odds ratio of heterozygotes versus homozygotes for the major allele; OR_hom, odds ratio of homozygotes for the minor alleles versus homozygotes for the major allele; OR_allele, odds ratio of each increasing number of minor alleles; SNP, single E-nucleotide polymorphism; A, major allele in HapMap CEU subjects; B, minor allele in HapMap CEU subjects (http://hapmap.ncbi.nlm.nih.gov/).

^a Odds ratios have been adjusted for age and subcohort (defined by country in EPIC and study phase in NHS).

^bP-values for trend (two sided) were derived from Cochran–Armitage trend test (d.f. = 1).

^cP-values for the Cochran–Armitage trend test (two sided; d.f. = 2) were obtained by coding genotypes as three categories: major allele homozygotes (reference), heterozygotes and for minor-allele homozygotes.

^d5p12-rs4415084 or surrogate 5p12-rs920329.

^e6q14-rs1917063 or surrogate 6q14-rs9344208.

^f6q14-rs9344191 or surrogate 6q14-rs9449341.

^gECHDC1, RNF146-rs2180341 or surrogate ECHDC1, RNF146-rs9398840.

^hZMIZ1-rs1250003 or surrogate ZMIZ1-rs704010.

ⁱRAD51L1-rs999737 or surrogate RAD51L1-rs10483813.

^jRALY-rs2284378 or surrogate RALY-rs6059651, RALY-rs8119937.

^kGMEB2-rs311499 or surrogate GMEB2-rs311498.

Stratified and case–case analysis

In the stratified analysis, SNP C19Orf62-rs8170 showed an increased risk of BC in the ER− stratum (OR_allele = 1.20, 95% CI = 1.09–1.31, P_1d.f. = 1 × 10⁻⁴). In addition, 11q13-rs614367 showed a preferential association with ER+ BC (OR_allele = 1.16, 95% CI = 1.10–1.22, P_1d.f. = 9.6 × 10⁻⁸). The results of the stratified analyses are shown in Supplementary Material, Table S3.

The case–case analyses showed a significant difference in the distribution of alleles of C19Orf62-rs8170 between ER+ and ER− cases (P_1d.f. = 6.8 × 10⁻⁷) and a non-significant difference between PR+ and PR− cases (P_1d.f. = 5.4 × 10⁻⁴). For FGFR2-rs2981582, we observed significant differences in the distribution of alleles with respect to ER status (P_1d.f. = 1.2 × 10⁻⁵). This was also observed for 11q13-rs614367 (P_1d.f. = 1.1 × 10⁻⁴). A significant result was also observed for 5p12-rs10941679 with respect to PR status (P_1d.f. = 1.5 × 10⁻⁴). We did not observe any other statistically significant difference. The strongest, non-significant evidence for a difference in the case–case analysis was observed for USHBP1-rs12982178 with respect to ER status (P_1df = 5.6 × 10⁻⁴) and PR status (P_1df = 3.4 × 10⁻⁴). The results of the case–case analyses are shown in Supplementary Material, Table S4.

Gene–environment interactions

The results of the interaction analysis of the 39 SNPs and the established risk factors are presented in Supplementary Material, Table S5. After correcting for multiple testing, no interactions were significant considering the adjusted threshold (P < 1.34 × 10⁻⁴) in BPC3. The strongest interaction result in BPC3 was observed between 6q25-rs2046210 and alcohol consumption (P_interaction = 0.002) (see Table 3).

Table 3.

Results of SNP-risk factors interaction analyses in BPC3, where interaction P-value < 0.01

Risk factora	SNP	Chr.	Nearest plausible gene	Cases	Controls	OR (95% CI)a	P_intb	Categories
Alcohol	rs2046210	6q25	Intergenic	14 169	17 026	1.05 (1.01–1.08)	1.95E−03	ALL
				4515	5722	0.96 (0.91–1.02)		0 g/day
				7436	8820	1.07 (1.02–1.12)		>0–<10 g/day
				1158	1414	1.17 (1.04–1.31)		10–20 g/day
				1060	1070	1.15 (1.02–1.30)		≥20 g/day
Height	rs2180341c	6q22	RNF146	9004	12 376	0.97 (0.93–1.02)	6.37E−03	ALL
				2206	3342	1.04 (0.95–1.13)		<1.60 m
				2563	3628	1.05 (0.97–1.14)		1.60–1.64 m
				2512	3206	0.87 (0.80–0.95)		1.65–1.69 m
				1723	2200	0.92 (0.83–1.02)		≥1.70 m
Oral contraceptives use	rs1562430	8q24	Intergenic	14 691	16 597	0.90 (0.87–0.93)	6.41E−03	ALL
				7610	8559	0.87 (0.83–0.90)		Never
				7081	8038	0.95 (0.90–0.99)		Ever
Height	rs311499d	20q13	GMEB2	9197	11 529	0.98 (0.91–1.06)	8.32E−03	ALL
				2291	2946	0.79 (0.68–0.92)		<1.60 m
				2709	3461	1.07 (0.94–1.23)		1.60–1.64 m
				2474	3011	1.09 (0.94–1.27)		1.65–1.69 m
				1723	2111	0.97 (0.81–1.15)		≥1.70 m
Smoking	rs4973768	3p24	SLC4A7	14 463	17 433	1.06 (1.03–1.10)	1.98E−02	ALL
				7286	9392	1.02 (0.98–1.07)		Never
				7177	8041	1.10 (1.05–1.16)		Ever

^aPer-allele OR of the SNP in various categories of the risk factor.

^bP-value of likelihood ratio test between models with and without interaction terms.

^cECHDC1, RNF146-rs2180341 or surrogate ECHDC1, RNF146-rs9398840.

^dGMEB2-rs311499 or surrogate GMEB2-rs311498.

In the meta-analysis with the BCAC data the pooled OR of the interaction between NOTCH2-rs11249433 and parity was significant (OR_meta = 1.13, 95% CI = 1.07–1.20, P_meta = 4.83 × 10⁻⁵). The interaction OR of LSP1-rs3817198 and number of full-term pregnancies, which was reported to be significant in BCAC (22) but not in our study, was not significant, considering the multiple testing, in the meta-analysis (OR_meta = 1.03, 95% CI = 1.01–1.05, P_meta = 0.0113). In addition, we observed heterogeneity between BPC3 and BCAC for the interaction ORs (P_{heterogeneity} = 6.39 × 10⁻⁵). This was also the case for the interaction between CASP8-rs1045485 and alcohol consumption (OR_meta = 1.14, 95% CI = 0.98–1.31, P_meta = 0.0813). The strongest evidence for interaction was observed between smoking and SLC4A7-rs4973768 (OR_meta = 1.08, 95% CI = 1.03–1.13, P_meta = 8.84 × 10⁻⁴) as shown in Table 4. Forest plots of the interactions shown in Table 4 are presented in Figure 1. Detailed results of the meta-analysis are shown in Supplementary Material, Table S6.

Table 4.

Interaction odds ratios from BPC3, BCAC and meta-analysis showing interaction P-value < 10⁻³

Study	SNP	Chr.	Nearest plausible gene	Location (bp) (hg19)a	BC risk factor	Cases	Controls	OR (95% CI)	Ptrendb	Phetc
BPC3	rs11249433	1	NOTCH2	121280363	Ever FTP	13 812	16 809	1.07 (0.95–1.20)	2.40E−01
BCAC						28 469	29 228	1.16 (1.08–1.24)	5.27E−05
META						42 281	46 037	1.13 (1.07–1.20)	4.83E−05	0.55
BPC3	rs4973768	3	SLC4A7	27415763	Ever smoke	14 463	17 433	1.09 (1.02–1.16)	1.21E−02
BCAC						16 737	18 263	1.07 (1.01–1.14)	2.76E−02
META						31 200	35 696	1.08 (1.03–1.13)	8.84E−04	0.96
BPC3	rs3817198	11	LSP1	1908756	Number of FTP	7591	10 679	0.97 (0.94–1.00)	6.26E−02
BCAC						23 064	22 151	1.06 (1.04–1.09)	2.38E−06
META						30 655	32 830	1.03 (1.01–1.05)	1.13E−02	6.39E−05
BPC3	rs1045485	2	CASP8	202149339	Alcohol	11 907	13 083	0.96 (0.81–1.15)	6.83E−01
BCAC						6081	9305	1.59 (1.24–2.05)	3.05E−04
META						17 988	22 388	1.14 (0.98–1.31)	8.13E−02	0.006

^aGenome Reference Consortium Human, build 37 (http://genome.ucsc.edu/cgi-bin/hgGateway).

^bPer-allele OR of the SNP in various categories of the risk factor.

^cHeterogeneity P-value between studies.

Figure 1.

Forest plots of the interaction odds ratios shown in Table 4.

DISCUSSION

An important extension of GWAS is to investigate whether genetic polymorphisms modify the effects of established BC risk factors and whether they show a stronger association in subgroups of BC cases. In this paper, we report findings from a consortium of large prospective studies on the possible interactions between 39 polymorphisms that have been associated previously with BC risk and established risk factors for the disease. Moreover, we conducted stratified analysis, considering the strata defined by the risk and prognostic factors and finally a case–case analysis considering the tumor prognostic factors alone. Data were examined using a nested case–control design within the BPC3. This work complements a previous report where a smaller set of cases and controls was analyzed for only 17 SNPs (21).

As a first step, we tested the main effect of the SNPs on BC risk and we found associations for all of the previously reported ER+ BC risk SNPs at the 0.05 significance level (P-values ranging from 0.035 to 5.81 × 10⁻³²) except for CASP8-rs1045485, RNF146-rs2180341, ZNF365-rs16917302, LSP1-rs3817198, COL1A1-rs2075555 and GMEB2-rs311499 (1,3,6,10,11,16,21). The possible involvement of LSP1-rs3817198 and CASP8-rs1045485 with BC risk was investigated in several studies and their association was consistently found in case–control studies (1,4,7,10,22) but not in prospective studies (9,18,21). The other SNPs were either found in small studies (COL1A1-rs2075555) (3) or selected populations such as Ashkenazi Jews (RNF146-rs2180341) (6) or BRCA2 mutation carriers (GMEB2-rs311499, ZNF365-rs16917302) (11).

As a second step, we investigated the possible differential association between SNPs and prognostic factors. The stratified and case–case analyses in groups determined by prognostic factors showed a preferential association of C19Orf62-rs8170 with ER− BC, which confirms previous findings (15). With respect to receptor-specific BC risk, our results were in general agreement with previous reports, suggesting that several SNPs are predominantly associated with ER+ BC: NOTCH2-rs11249433 (9), 2q35-rs13387042 (2,27), TNRC9-rs3803662 (2), 5p12-rs4415084, 5p12-rs10941679 (5), FGFR2-rs2981582 (1), FGFR2-rs3750817 (20) and MAP3K1-rs889312 (28). Others are more predominantly associated with ER− or PR− BC: C19Orf62-rs8170, 6q14-rs17530068 and 6q14-rs13437553 (15), whereas we did not replicate the preferential association of 20q11-rs4911414 with ER− BC (15). Our results are also consistent with previous reports of SNPs C19ORF62-rs8170, USHBP1-rs12982178 and TERT-rs10069690 being specifically associated with ER− BC risk (12,15), while we could not replicate the association of RALY-rs2284378 with ER− BC (15). A possible explanation for this is lack of statistical power in the ER− group, which included 2127 cases in our study. Since part of the individuals used in this report overlap with Refs. (12,15), the results presented here cannot be considered an independent replication.

In the last years, there has been a keen interest in investigating gene–environment interactions using SNPs identified by GWAS and established risk factors, especially in common cancer types such as breast, for which multiple susceptibility loci have been identified and several risk factors are known. Finding gene–environment interactions can be useful in two areas: in order to allow a more specific risk assessment and aid targeted early detection or prevention strategies and to further our understanding of biological pathways and mechanisms of disease etiology (29). Despite the vast international effort, only a few established examples of gene–environment interactions exist, such as the one between NAT2 polymorphisms and smoking in relation to bladder cancer risk (30), and between ALDH2 polymorphisms and alcohol in relation to esophageal cancer risk (31,32). For BC, several large studies, including our own (18,19,21), focusing on GWAS loci have reported no gene–environment interactions. We considerably expanded our previous study (21). Moreover, taking advantage of the work of Nickels et al., who reported significant interactions between NOTCH2-rs11249433 and parity as well as between LSP1-rs3817198 and number of full-term pregnancies (22), we conducted a meta-analysis of the results from BCAC and BPC3 in order to investigate gene–environment interactions on up to ∼79 000 individuals. In BPC3, we did not observe any significant interactions between the selected SNPs and any of the epidemiologic risk factors, when we considered the adjusted significance threshold (P < 1.34 × 10⁻⁴). In the meta-analysis, we observed that our estimate of the interaction between NOTCH2-rs11249433 and parity (OR_interaction = 1.07 95% CI = 0.95–1.20, P_trend = 0.24) pointed towards an increased risk, as did the one reported by BCAC (OR_interaction = 1.16, 95% CI = 1.08–1.24, P_trend = 5.27 × 10⁻⁵), but our estimate did not reach statistical significance. However, our result does not weaken the meta-analysis estimate, on the contrary, it makes the association even stronger (OR_meta = 1.13, 95% CI = 1.07–1.20, P_meta = 4.83 × 10⁻⁵). It is therefore plausible that the interaction observed by BCAC is true, although modest, but not observed in BPC3 because of insufficient sample size. We did not observe the interaction reported from Nickels et al. for LSP1-rs3817198 with the number of FTP, which is not surprising since we did not replicate the association with BC risk that was reported by BCAC for this SNP either. The discrepancy in the findings between cohort studies and case–control studies seem to be consistent since also the previous studies from BPC3 (21) and BCAC showed discordant results. Nickels et al. suggest that the difference might be due to a misclassification of parity in the cohorts, considering that the information was collected only at the time of enrolment. This seems unlikely considering the age of enrolment of the women in the BPC3 cohorts and therefore the most likely explanation is that our power was limited to detect the modest magnitude of the interaction reported by BCAC.

The most interesting result from the meta-analysis is the interaction of smoking status and SLC4A7-rs4973768. This association, although not significant, has biological plausibility. The SLC4A7 gene affects bicarbonate transport but is also thought to be responsible for the influx of lead into erythrocytes (33). It is well known that smokers have a higher concentration of lead in the blood than non-smokers (34,35), since lead is present in cigarette tar. In addition, several studies suggest a positive association between lead exposure and risk of several kinds of cancer, including breast (36–39). Moreover, rs4973768 has been reported to have a functional impact on the SLC4A7 gene (33). To follow this up, we have used the regulomeDB web site (40), observing that the polymorphic variant rs552647, which is in high LD (r² = 0.963) with rs4973768, is predicted to affect binding with the transcriptional repressor CTCF, a master regulator of gene expression. It is therefore conceivable that this polymorphic variant could modify the ability of SLC4A7 to introduce deleterious lead inside cells. Since lead is not normally present in the organism and is introduced by environmental exposure it is reasonable that a polymorphic variant that influences SLC4A7 functionality might exert its effect in subjects with an increased exposure to lead such as smokers. In the BPC3, we also note that this SNP is significantly associated with BC risk among smokers (OR_{allele, smokers} = 1.10, 95% CI = 1.05–1.16, P_trend = 2.1 × 10⁻⁵, OR_{allele, non-smokers} = 1.02, 95% CI = 0.98–1.07, P_trend = 0.28). Therefore, this statistical interaction, although not significant, is highly suggestive, since it could describe a biological process which may contribute to breast carcinogenesis. This effect could be an example of a pure or qualitative interaction (i.e. there is an effect only in the presence of both the susceptible genotype and the environmental factor, as described by Thomas (29)), since it seems that the effect of SLC4A7 genotypes is present only in smokers. A similar finding has been described for polymorphisms in the NAT2 gene in bladder cancer: in fact, several studies have consistently shown that slow acetylator phenotype increases the risk of developing the disease only among smokers (30).

BC is a complex disease involving multiple environmental and genetic risk factors and, in this analysis, we have considered only a small fraction of the loci that potentially could influence the disease risk. A possible alternative strategy to pursue this route is to perform gene–environment-wide interaction studies, i.e. using the information of a GWAS to agnostically find novel gene–environment interactions. This innovative approach has shown some promising results (32,41–46), although several questions on how to apply this strategy still remain unsolved (29,47). In addition, our data confirm what has been suggested by others, namely that one possible way to further our understanding of BC would be to conduct GWAS in more homogeneous groups of cases defined by receptor status or histological subtype (2,5,15,16,48,49). Continuing to implement gene–environment interaction analysis, possibly using more comprehensive statistical approaches and study designs, alongside novel genotyping techniques, will still be useful. More importantly, it will be necessary not to restrict the field to GWAS hits and established epidemiologic risk factors.

In conclusion, in this study of up to almost 79 000 women we confirmed several known associations between polymorphic variants and BC risk. Considering the significance threshold adjusted for multiple testing, we did not observe any novel interaction between genetic variants and BC risk factors nor any novel difference in the association between risk SNPs and tumor characteristics. The results obtained in this paper strongly suggest that there are no biologically relevant interactions for the majority of the current GWAS variants and the epidemiologic risk factors taken into consideration. However, we did observe a suggestive novel gene–environment interaction (SLC4A7 genotypes and smoking behavior) that if confirmed could further our understanding on BC susceptibility. Finally, our results are compatible with a significant interaction between NOTCH2-rs11249433 and parity reported by Nickels et al. The results obtained in this paper are conclusive in excluding any other major interactions between the current GWAS hits and the epidemiologic risk factors taken into consideration.

MATERIALS AND METHODS

Study subjects

The BPC3 has been described extensively elsewhere (50). Briefly, it consists of large, well-established prospective cohorts assembled in Europe, Australia and the United States that have both DNA samples and extensive questionnaire information collected before BC diagnosis. The cohorts used in this analysis are: the European Prospective Investigation into Cancer and Nutrition (EPIC) (51), the Women's Health Initiative (WHI) (52), the Melbourne Collaborative Cohort Study (53), the NHS (54), the Women's Health Study (55), the American Cancer Society Cancer Prevention Study II (56), the Prostate, Lung, Colorectal, Ovarian Cancer Screening Trial (57) and the Multiethnic Cohort (58).

Cases were women who had been diagnosed with invasive BC after enrolment in one of the BPC3 cohorts. The diagnosis was confirmed by medical records or tumor registries (the method varied among cohorts). Subjects were considered eligible controls if they were free of BC until the follow-up time for the matched case subject. Matching criteria were: age at baseline, menopausal status at baseline and cohort. All study subjects were of Caucasian ethnicity. Relevant institutional review boards from each cohort approved the project and informed consent was obtained from all subjects.

SNP selection and genotyping

The SNPs included in the analyses (Table 1) were reported to show a genome-wide statistical significant association with BC risk (P < 5 × 10⁻⁷). In this study, either the SNP from the original publication or a surrogate in complete linkage disequilibrium (r² = 1 in HapMap CEU) was genotyped. In particular, for the following SNPs, we have genotyped either the original SNP or the surrogate: rs4415084 (surrogate rs920329), rs9344191 (surrogate rs9449341), rs1250003 (surrogate rs704010), rs999737 (surrogate rs10483813), rs2284378 (surrogates rs8119937 and rs6059651), rs2180341 (surrogate rs9398840), rs311499 (surrogate rs311498) and rs9344208 (surrogate rs1917063). The BC susceptibility SNPs recently reported by BCAC were not included in this analysis (59).

Genotyping was performed using TaqMan assays (Applied Biosystems, Foster City, CA, USA) as specified by the producer. Genotyping of the BC cases and controls was performed in four laboratories (German Cancer Research Center (DKFZ), University of Southern California, U.S. NCI, Harvard School of Public Health). Laboratory personnel were blinded to whether the subjects were cases or controls. Duplicate samples (∼8%) were included and concordance of these samples was >99.99%.

Data filtering and statistical analysis

The study subjects were BC patients and controls for which at least 90% of the SNPs had been successfully genotyped (35 661 subjects in total). Each SNP was tested for Hardy–Weinberg equilibrium among the controls. All unconditional statistical models that were used in this study have been adjusted for age at recruitment and cohort (defined as country in EPIC and study phase in NHS).

We investigated the association between genetic variants and BC risk by fitting an unconditional logistic regression model. The genotypes were treated either as nominal variables, comparing heterozygotes and minor allele homozygotes to the reference group of major allele homozygotes (co-dominant model), or as interval variables in a log-additive model. This was also carried out separately for each cohort and tests for heterogeneity between cohorts were performed.

In order to investigate the possible interactions between SNPs and risk factors, two models for each pair of SNP and risk factor were explored: one with only SNP and risk factor and one including additional SNP-risk factor interaction term(s). The genetic variants were treated as interval variables (counts of minor alleles) and the non-genetic risk factors were treated as continuous (BMI, height, age at menarche, age at menopause, pack-years of smoking, alcohol in g/day, age at first full-term pregnancy and number of full-term pregnancies) or dichotomous (parity, family history, ever use of oral contraceptives). For smoking status, we used three categories (current, former and never smoker).

We then applied the likelihood ratio test to compare the two models and to assess departures from the log-additive model for the joint effect of SNP and risk factor. For BMI, the interaction analysis was performed separately for pre- and postmenopausal women.

The interaction of genes and use of hormone replacement therapy will be explored in a separate study and was not examined in this analysis.

Stratified, unconditional analyses using cases and controls were performed for risk factors of BC. Cochrane's Q-test was used to test for heterogeneity between the strata. The strata for the risk factors were defined as: BMI (BMI < 25, 25 ≤ BMI < 30, BMI ≥ 30), height (height ≤ 1.60, 1.60–1.65,1.65–1.70 m, height ≥ 1.70 m), age at menarche (early, ≤11 years; intermediate, 12–13 years; late, ≥14 years), age at menopause (early, ≤44 years; intermediate, 45–49 years; late, ≥50 years), smoking (never smoker, ever smoker), pack-years of smoking (0, >0–<10, ≥10–<20, ≥20), alcohol (0, >0–<10, ≥10–<20, ≥20 g/day), ever full-term pregnancy (yes, no), age at first full-term pregnancy (<20, 20–24, 25–29, >29), number of full-term pregnancies (0, 1, 2, 3, ≥4), family history (mother diagnosed with BC or not, as information on first-degree relatives was sparse), ever use of oral contraceptives (yes, no).

Additional case–case analyses using an unconditional logistic regression model were performed to test for the effect ***of heterogeneity of prognostic factors. The prognostic factors were defined as follows: TNM staging (Stage 1, Stage 2, Stages 3–4), grade (well differentiated, moderately differentiated and poorly differentiated), tumor size (<2, 2–5 cm, >5 cm), age at diagnosis (younger than 55, older than 55) and ER, PR status (negative, positive). Using the matching to select controls, we also performed subgroup analyses for the prognostic factors in order to have a complete assessment of the preferential association of the SNPs with the tumor characteristics.

A fixed-effects meta-analysis of the interaction ORs was performed to combine the results from our study and that of the BCAC study (22) giving us a final sample size of 79 534 subjects (34 817 BC cases and 44 717 controls). We analyzed 19 SNPs that were used in both studies, the non-genetic variables in the BPC3 study were recoded so as to correspond to the categories used in the BCAC study (22). Before doing the meta-analysis, heterogeneity between the studies was investigated using Cochrane's Q-test.

The significance threshold was adjusted, taking into account the large number of tests carried out. Since some of the SNPs map to the same regions and might be in linkage disequilibrium, for each locus we calculated the effective number of independent SNPs, M_eff, using the SNP Spectral Decomposition approach (simpleM method) (60). The study-wise M_eff obtained was 31. For the interaction analyses, the P-value threshold was obtained by dividing the conventional significance threshold of 0.05 by the product of M_eff and the number of risk factors (n = 12). Thus, for the interaction analyses and for the heterogeneity tests the threshold for statistical significance was 1.34 × 10⁻⁴ (0.05/(31 × 12)), for the stratified analyses it was 0.05/(31 × 18) = 9 × 10⁻⁵ and for the case–case analyses it was 0.05/(31 × 6) = 2.7 × 10⁻⁴.

All statistical tests were two sided, and all statistical analyses were performed with SAS version 9.2.

SUPPLEMENTARY MATERIAL

Supplementary Material is available at HMG online.

FUNDING

This work was supported by U.S. National Institutes of Health, National Cancer Institute (U19-CA148065, cooperative agreements U01-CA98233-07 to D.J.H., U01-CA98710-06 to M.J.T., U01-CA98216-06 to E.R. and R.K., and U01-CA98758-07 to B.E.H.) and Intramural Research Program of National Institutes of Health and National Cancer Institute, Division of Cancer Epidemiology and Genetics. The WHI program is funded by the National Heart, Lung, and Blood Institute, National Institutes of Health, U.S. Department of Health and Human Services through contracts HHSN268201100046C, HHSN268201100001C, HHSN268201100002C, HHSN268201100003C, HHSN268201100004C, and HHSN271201100004C. A full listing of WHI investigators can be found at: https://cleo.whi.org/researchers/Documents%20%20Write%20a%20Paper/WHI%20Investigator%20Short%20List.pdf. The Founding Sources had no role in the study design; in the collection, analysis and interpretation of data and in the decision to submit the paper for publication.