Associations of common breast cancer susceptibility alleles with risk of breast cancer subtypes in BRCA1 and BRCA2 mutation carriers

Introduction More than 70 common alleles are known to be involved in breast cancer (BC) susceptibility, and several exhibit significant heterogeneity in their associations with different BC subtypes. Although there are differences in the association patterns between BRCA1 and BRCA2 mutation carriers and the general population for several loci, no study has comprehensively evaluated the associations of all known BC susceptibility alleles with risk of BC subtypes in BRCA1 and BRCA2 carriers. Methods We used data from 15,252 BRCA1 and 8,211 BRCA2 carriers to analyze the associations between approximately 200,000 genetic variants on the iCOGS array and risk of BC subtypes defined by estrogen receptor (ER), progesterone receptor (PR), human epidermal growth factor receptor 2 (HER2) and triple-negative- (TN) status; morphologic subtypes; histological grade; and nodal involvement. Results The estimated BC hazard ratios (HRs) for the 74 known BC alleles in BRCA1 carriers exhibited moderate correlations with the corresponding odds ratios from the general population. However, their associations with ER-positive BC in BRCA1 carriers were more consistent with the ER-positive associations in the general population (intraclass correlation (ICC) = 0.61, 95% confidence interval (CI): 0.45 to 0.74), and the same was true when considering ER-negative associations in both groups (ICC = 0.59, 95% CI: 0.42 to 0.72). Similarly, there was strong correlation between the ER-positive associations for BRCA1 and BRCA2 carriers (ICC = 0.67, 95% CI: 0.52 to 0.78), whereas ER-positive associations in any one of the groups were generally inconsistent with ER-negative associations in any of the others. After stratifying by ER status in mutation carriers, additional significant associations were observed. Several previously unreported variants exhibited associations at P <10−6 in the analyses by PR status, HER2 status, TN phenotype, morphologic subtypes, histological grade and nodal involvement. Conclusions Differences in associations of common BC susceptibility alleles between BRCA1 and BRCA2 carriers and the general population are explained to a large extent by differences in the prevalence of ER-positive and ER-negative tumors. Estimates of the risks associated with these variants based on population-based studies are likely to be applicable to mutation carriers after taking ER status into account, which has implications for risk prediction. Electronic supplementary material The online version of this article (doi:10.1186/s13058-014-0492-9) contains supplementary material, which is available to authorized users.


Introduction
Women who carry pathogenic mutations in BRCA1 or BRCA2 have markedly increased risks of developing breast cancer. The distributions of breast cancer tumor characteristics differ between BRCA1 mutation carriers, BRCA2 mutation carriers and those arising in the general population. The majority of breast tumors arising in BRCA1 carriers show low or absent expression of estrogen receptor (ER) [1][2][3], whereas the majority of BRCA2associated tumors are ER-positive [1,4,5].
Many common breast cancer susceptibility alleles identified through population-based genome-wide association studies (GWASs) have also been associated with breast cancer risk in BRCA1 and BRCA2 carriers [6,7]. Several of these variants are specifically associated with the ER status of the breast cancer in the general population [8,9]. Among the single-nucleotide polymorphisms (SNPs) that have been evaluated in mutation carriers so far, the variants found to be associated with breast cancer risk for BRCA1 carriers largely overlap with loci for which stronger associations with ER-negative breast cancer have been reported in the general population [8][9][10][11][12]. An important question for risk modelling and prevention studies is whether the effects of common variants on breast cancer risk in mutation carriers are mediated through a generic influence on the development of particular hormone receptor subtypes of breast cancer or through epistatic interaction with the BRCA1/2 mutation itself.
Previous studies by the Consortium of Investigators of Modifiers of BRCA1/2 (CIMBA) described the impact of 29 breast cancer susceptibility variants from nonhereditary breast cancer studies on ER-positive and ERnegative breast cancer risk in BRCA1 and BRCA2 carriers [6,7,[13][14][15]. These analyses demonstrated that, despite the lack of an association between some susceptibility variants and overall breast cancer risk for BRCA1 or BRCA2 carriers, residual associations exist with specific disease subtypes. In addition, the ER-specific associations in BRCA1 and BRCA2 carriers were mainly in the same direction and of a magnitude similar to the associations observed with breast cancer stratified by ER expression status in the general population. However, these studies were conducted on smaller numbers of mutation carriers than currently available and evaluated only a subset of the currently known breast cancer susceptibility alleles for their associations with ER-specific subtypes in carriers. Recently, 45 additional SNPs have been found to be associated with breast cancer risk in the general population [8][9][10]16]. Eighteen of these SNPs showed evidence of association with ER-positive breast cancer, but not with ER-negative breast cancer, and four loci (1q32.1 LGR6, 2p24.1, 16q12 and 20q11) were associated only with ER-negative breast cancer in the general population. These 45 newly discovered loci have not yet been evaluated for their associations with breast cancer risk for mutation carriers.
In the present study, we assessed the disease subtypespecific associations of all 74 previously reported breast cancer susceptibility variants in 15,252 BRCA1 and 8,211 BRCA2 carriers. We evaluated whether differences in associations of known breast cancer susceptibility variants between BRCA1 carriers, BRCA2 carriers and the general population are mediated by tumor ER status in mutation carriers. We also analyzed the associations of about 200,000 variants on the iCOGS genotyping array with subtype-specific breast cancer risk in carriers in an attempt to uncover previously unreported subtypespecific associations in women with BRCA1 and BRCA2 mutations. In addition to ER and progesterone receptor (PR) status, we report, for the first time to our knowledge, associations by HER2 status and with triplenegative disease (TN, referring to ER-, PR-and human epidermal growth factor receptor 2 (HER2)-negative), and we also describe associations with clinical features such as "ductal, no specified subtype" (hereafter referred to as ductal) and lobular morphologic subtypes, nodal status and histological grade.

Study subjects
Data were obtained from 47 studies in 27 different countries in CIMBA [17]. Eligible study subjects were women who carry pathogenic mutations in BRCA1 or BRCA2. The majority were recruited through cancer genetics clinics and enrolled into national or regional studies. Written informed consent was obtained from all subjects. Each of the host institutions recruited under ethically approved protocols. A list of the local institutional review boards that provided ethical approval for this study is given in Additional file 1: Table S1. Eligibility was restricted to mutation carriers who were 18 years of age or older at recruitment. Data collected included year of birth, age at cancer diagnosis, personal history of bilateral prophylactic mastectomy and/or bilateral salpingo-oophorectomy, mutation description, tumor pathology and ethnicity.

Tumor pathology data
Breast tumor pathology data were gathered from a range of sources, specifically patient pathology reports, pathology review data, tumor registry records and tissue microarray results. These included information on ER, PR and HER2 status; morphologic subtype; lymph node involvement; and histological grade. For ER, PR and HER2, status was classified as negative or positive, with supplementary immunohistochemistry scoring or biochemical data and methodology provided when available. The vast majority of centers employed a cutoff of either ≥10% or ≥1% of tumor nuclei staining positive to define ER and PR positivity. Additional file 1: Table S2 lists the subtype definitions used by each study, which were not centrally reclassified, owing to the low proportion of records with supporting staining data. Similarly, HER2 status was determined using immunohistochemistry to detect strong complete membrane staining (with a score of 3+ considered positive) and/or in situ hybridization to detect HER2 gene amplification. To ensure consistency across studies, when information on the cells stained was available, we used the same cutoff to define ER-, PR-and HER2-positive tumors. The cutoffs used for the small number of cases where composite scoring methods based on the proportion and intensity of staining were available (Allred score, immunoreactive Remmele score) are given in Additional file 1: Table S2. Consistency checks were performed to validate receptor data against supplementary scoring information, if provided. Each cancer was assigned to a morphologic subgroup (ductal, lobular, medullary, other), which we confirmed using the World Health Organization International Classification of Diseases for Oncology (ICD-O) code for the classification of tumor type when sufficient information was provided [18]. Lymph node status, along with the number of nodes showing metastatic carcinoma, was provided when available. Histologic grade was assigned as grade 1, 2 or 3 by local pathologists who used a modified Scarff-Bloom-Richardson malignancy grading system.

Genotyping and quality control
Genotyping was carried out using the iCOGS custom array. The array development and details of the genotyping and quality control for the CIMBA samples are described in detail elsewhere [6,7]. Briefly, genotyping for BRCA2 carriers was conducted at McGill University and Génome Québec Innovation Centre (Canada) and for BRCA1 carriers at the Mayo Clinic (USA). SNPs were excluded if they were located on the Y chromosome, if they were monomorphic, if they deviated significantly from Hardy-Weinberg equilibrium (P <10 −7 ) or if they had call rates <95%. Samples were excluded if they had a call rate <95%, if they were of non-European ancestry or if they demonstrated extreme heterozygosity. After quality control, we had 200,720 SNPs available for analysis in 15,252 BRCA1 samples and 200,908 SNPs available for analysis in 8,211 BRCA2 samples.

Statistical methods
We evaluated the associations of each genotype with risks of developing breast cancer or breast cancer subtypes defined by the tumor characteristics or morphology. The analyses were carried out within a survival analysis framework. Individuals were censored at the first of the following events: breast cancer diagnosis, ovarian cancer diagnosis, bilateral mastectomy or age at last follow-up. In order to account for non-random ascertainment of mutation carriers with respect to their disease phenotype, we used a retrospective likelihood approach that models the probability of observing the genotypes conditional on the disease phenotype [19,20]. It was assumed that the cancer incidence depends on the underlying SNP genotype through a Cox proportional hazards model: where λ 0 (t i ) is the baseline incidence and β is the logarithm of the per-allele hazard ratio (HR, under a multiplicative model). The association with overall breast cancer risk was evaluated by testing the hypothesis that β = 0 [20]. We evaluated the associations with the groups of each subtype class (for example, ER-positive and ER-negative), using an extension of the retrospective likelihood approach to model the simultaneous effect of each SNP on more than one tumor subtype [15]. Briefly, this involves modeling the conditional likelihood of the observed SNP genotypes and tumor subtypes, given the disease phenotypes. Within this framework, it is possible to estimate simultaneously the HRs for each tumor subtype and test for heterogeneity in the associations [15]. To maximize the available information, genotyped mutation carriers that were missing information on tumor characteristics were included in the analysis, and their disease subtype was assumed to be missing at random. In order to account for non-independence among relatives, a robust variance estimation approach was used [20]. Further details of the methods for evaluating the associations with overall breast cancer [20] and tumor subtypes have been described elsewhere [15]. We carried out association analyses by subtype for the following breast cancer characteristics: ER-positive and ER-negative, PR-positive and PR-negative, HER2positive and HER2-negative, TN breast cancer (that is, negative for ER, PR and HER2) and non-TN (that is, tumor positive for at least one of the three receptors), ductal morphologic subtype, lobular morphologic subtype, nodal involvement (no involved lymph nodes and at least one involved lymph node) and histological grade (high grade (grade 3) and non-high grade (grades 1 and 2)). Only samples with complete information on ER, PR and HER2 expression were included in the analysis for TN as well as non-TN breast cancer. The SNP associations by tumor morphologic subtype were evaluated by comparing ductal tumors to all others and, in a separate analysis, lobular tumors to all others. We are not reporting association analyses for risk of medullary morphologic subtype, owing to sparse data as well as the difficulties in diagnosing medullary breast tumors reliably [21,22]. All analyses were stratified by country of residence. The United States and Canada strata were further subdivided by reported Ashkenazi Jewish ancestry. For subtypes with small groups, strata of geographically close countries were combined to provide sufficiently large groups for estimation. All analyses used calendar year-and cohort-specific cancer incidences for BRCA1 and BRCA2. SNPs with minor allele frequencies <3% were excluded. The retrospective likelihood was modeled using custom-written functions implemented in the pedigree analysis software MENDEL [23].
When evaluating whether the known breast cancer susceptibility loci identified through population based studies also modify breast cancer risk in mutation carriers, a significance threshold of P <0.05 was used because of the strong prior evidence of association for these loci with disease risk. For the association analyses of all the approximately 200,000 variants on the iCOGS array with the breast cancer subtypes in mutation carriers, only associations with P < 5 × 10 −8 were considered significant. The discussion of findings and the tables were extended to associations at P <10 −6 .
For variants associated with ER-positive or ER-negative breast cancer with P <0.01, we evaluated whether the associations may have been affected by a possible survival bias due to inclusion of prevalent breast cancer cases in the analysis. For the sensitivity analysis, the association analysis by ER status was repeated after excluding mutation carriers diagnosed with breast cancer ≥5 years prior to study recruitment.
We evaluated the consistency between the breast cancer association estimates of previously reported breast cancer susceptibility variants in the general population (using published data) and the association estimates in BRCA1 and BRCA2 carriers using the intraclass correlation (ICC). We estimated ICC as outlined by Shrout and Fleiss [24] based on a one-way random-effects model and tested for agreement in absolute values of log HR. The same approach was used to evaluate the agreement between associations with ER-positive and/or -negative breast cancer in the general population and associations with ER-positive and/or -negative breast cancer in BRCA1 and in BRCA2 carriers. Furthermore, we carried out the same comparisons between associations for BRCA1 and associations for BRCA2 carriers.

Subtype patterns
The analyses included data from 15,252 BRCA1 carriers and 8,211 BRCA2 carriers. Among the breast canceraffected BRCA1 carriers, we had data on at least one disease characteristic of interest for 4,619 (59%) of the 7,797 affected women (Table 1). Data were available on tumor characteristics for 2,570 (59%) of the 4,330 affected BRCA2 carriers. Of the individuals with pathology information, 74% of the BRCA1 carriers and 75% of the BRCA2 carriers had data on ER status.

Single-nucleotide polymorphism associations
After quality control, genotype data were available for analysis for 200,720 SNPs for BRCA1 carriers and for 200,908 SNPs for BRCA2 carriers. After adjusting for sample size and excluding SNPs chosen for inclusion on the genotyping array based on reported associations in subsets of the current sample, the inflation coefficient λ 1000 values were 1.01 for ER-positive disease in BRCA1 carriers, 1.02 for ER-negative in BRCA1, 1.01 for ERpositive in BRCA2 and 1.02 for ER-negative in BRCA2 carriers (Additional file 1: Figure S1 and Figure S2). Similar patterns were observed for other tumor characteristics (results not shown). After excluding variants located at known breast cancer susceptibility loci, there was no evidence for an excess in associations by ER status beyond the number expected.
Associations of previously reported breast cancer susceptibility loci Associations with overall breast cancer and by tumor estrogen receptor status First, we considered the associations with risk for overall breast cancer and for tumor subtypes for the 74 breast cancer susceptibility variants that have been reported up to April 2013. In light of the strong prior evidence of association, we considered associations at P <0.05 as evidence that a previously reported breast cancer susceptibility allele also modifies overall or ER-specific breast cancer risk in mutation carriers. The associations with overall breast cancer risk and risk of breast cancer subtypes for all 74 variants are given in Table 2 and Additional file 1: Tables S4 to S10. Of the breast cancer susceptibility loci that had not previously been evaluated for an association in mutation carriers, SNPs at 5q33. 3, 8q24.21, 11q24.3, 12q22, 16q12.1, 22q13.1 were associated with overall breast cancer risk for BRCA1 carriers, and SNPs at 6p23, 11q24.3 and 16q12.1 were associated with breast cancer risk for BRCA2 carriers at P <0.05 (Table 2). Overall, 15 breast cancer susceptibility variants were associated with ER-negative breast cancer in BRCA1 carriers and 8 variants in BRCA2 carriers at P <0.05 (Table 2). Ten significant associations with ER-positive breast cancer in BRCA1 carriers and fourteen in BRCA2 carriers were found. The strongest association with ERpositive breast cancer was observed for rs2981579 in FGFR2 at 10q26.12 for both BRCA1 and BRCA2 carriers. SNP rs10069690 in TERT at 5p15.33 displayed the strongest association with ER-negative breast cancer for BRCA1 carriers and rs9348512 at 6p24.3 for BRCA2 carriers. We found significant differences in the associations by ER status for rs3803662 in TOX3 at 16q12.1 (P = 2 × 10 −4 ) and rs13387042 at 2q35 (P =0.002) for BRCA1 carriers, which were not previously seen. Both SNPs showed evidence of association with ER-positive breast cancer only. Similarly, six of the loci that did not show evidence of association with overall breast cancer were associated with ERpositive and two with ER-negative breast cancer in BRCA1 carriers. This included two of the loci not previously evaluated in mutation carriers: 3q26.1 and 6p25.3. In BRCA2 carriers, four of the variants lacking evidence of association with overall breast cancer were associated with ERnegative and three with ER-positive breast cancer. This included four loci not previously evaluated in mutation carriers: 2q24, 14q13. 3, 19q13.31 and 22q12.2. Of the breast cancer susceptibility loci that had not yet been evaluated for an association with breast cancer in mutation carriers, rs1011970 at CDKN2A/B and rs1292011 at 12q24.21 had significantly different associations with ERpositive and ER-negative cancer for BRCA1 carriers (P het = 0.009 and P het = 0.004, respectively, for the difference between ER-positive and ER-negative). SNP rs2236007 at 14q13.3 displayed differences by ER status for BRCA2 carriers (P het = 0.008). These three SNPs had associations in different directions for ER-positive and ER-negative tumors. When association analyses for ER-positive and -negative disease were repeated after excluding prevalent breast cancer cases (Additional file 1: Table S3), the HR estimates were consistent with the estimates from the complete sample but were associated with larger confidence intervals. Therefore, it is unlikely that our results are influenced by survival bias.

Associations with other subtypes and clinical features
The pattern of associations of previously reported breast cancer susceptibility variants by PR and TN status were very similar to that by ER status (Additional file 1: Tables S4 and S6), but fewer associations were observed at P <0.01. SNP rs720475 at 7q35 was the only variant that was associated with HER2-positive disease (HR = 1.45 and P = 0.003 for HER2-positive, P het = 9 × 10 −4 in BRCA2 carriers) (Additional file 1: Table S5).

Comparison of patterns of associations by breast cancer estrogen receptor status between BRCA1 and BRCA2 carriers and the general population
We compared the log HR estimates for the breast cancer association of known breast cancer susceptibility variants for BRCA1 carriers, BRCA2 carriers, and for the general population using published data from the Breast Cancer Association Consortium (BCAC) [8]. The resulting ICC coefficients for log HR/OR estimates for all comparisons are shown in Table 3. Log HR estimates for overall breast cancer risk in BRCA2 carriers were very similar to the log odds ratios (ORs) from the general population (ICC = 0.63, 95% CI: 0.47 to 0.75) (Additional file 1: Figure S3B), whereas there was only moderate correlation between the log HR estimates for BRCA1 carriers and the log HR estimates from both other groups (ICC: BRCA1-BCAC estimates = 0.43, BRCA1-BRCA2 = 0.46) (Additional file 1: Figure S3A,C). When comparing ERpositive specific associations, we found stronger agreement between the log HR/OR estimates than for overall breast cancer. The ICC estimates ranged from 0.61 (95% CI: 0.45 to 0.74) ( Figure 1B) for BCAC-BRCA1 to 0.69 (95% CI: 0.55 to 0.79) ( Figure 1C) for BCAC-BRCA2. The ER-negative breast cancer log HR estimates in BRCA1 carriers and the corresponding BCAC estimates were strongly correlated (ICC = 0.59, 95% CI: 0.42 to      0.72) ( Figure 1E). However, the ER-negative breast cancer log HR estimates in BRCA2 carriers were less strongly correlated with the corresponding estimates in BRCA1 carriers (ICC = 0.46, 95% CI: 0.25 to 0.62) ( Figure 1D) and in BCAC (ICC = 0.28, 95% CI: 0.05 to 0.48) ( Figure 1F). There was no evidence that the ICC was different from 0 for the comparison between ER-positive associations in BRCA1 carriers with ER-negative associations in BRCA2 carriers and vice versa (Additional file 1: Figure S4B,C). Similarly, there was no significant correlation between log OR estimates for ER-positive breast cancer in BCAC with log HR estimates for ER-negative breast cancer in BRCA1 and BRCA2 carriers (Additional file 1: Figures S5B and  S6B). There was only moderate correlation between log OR estimates for ER-negative breast cancer in the general population and log HR estimates for ER-positive breast cancer in BRCA1 and BRCA2 carriers (ICC = 0.39 and ICC = 0.34, respectively) (Additional file 1: Figures S5C  and S6C).

Associations by subtype with all single-nucleotide polymorphisms on iCOGS
Variants in or near the known breast cancer susceptibility loci TERT, ESR1 and 19p13.11 showed strong associations (P <10 −6 ) with at least one of the categories in all subtype analyses in BRCA1 carriers. The same was true for SNPs in FGFR2 and TOX3 in BRCA2 carriers. Variants on the iCOGS array that exhibited associations at P <10 −6 with any of the breast cancer subtypes of ER-, PR-, HER2-positive or -negative and TN breast tumors are shown in Table 4. All the variants associated with ERpositive or ER-negative breast cancer at P <10 −6 were located within known breast cancer susceptibility loci. Similar associations were observed with PR-positive and PR-negative breast cancer for BRCA1 carriers. A previously unreported SNP at 2p13.2 was associated with HER2-positive breast cancer in BRCA1 carriers. In BRCA2 carriers, two previously unreported variants showed evidence of association in the analysis by PR status.
Furthermore, SNPs at 8q12.1 near TOX were associated with HER2-positive cancer in BRCA2 carriers. Only SNPs in the known breast cancer susceptibility loci FGFR2 and TOX3 were associated with non-TN breast cancer in BRCA2 carriers, and none were associated with TN breast cancer at P <10 −6 .
A SNP at 7q36 was associated with ductal subtype for BRCA1 carriers (Table 5). Two loci were associated with lobular breast cancer for BRCA2 carriers: 11q23.3 and Xp11.23.
There was one novel association with high-grade tumors in BRCA1 carriers (Table 6). Three previously unreported variants were associated with breast cancer nodal status in BRCA1 carriers: SNPs at 4q24 in the TET2 gene, at 5q32 in the SH3RF2 gene and at 7p22 in an intron of NXPH1. For BRCA2 carriers, only SNPs in FGFR2 and TOX3 exhibited associations at P <10 −6 with breast cancer nodal status or histological grade.

Discussion
This is the first comprehensive report, to our knowledge, of the associations of genetic variants with risk of developing breast cancer by tumor subtypes in BRCA1 and BRCA2 carriers. We evaluated the associations with ER, PR and HER2 status; morphologic subtype (ductal or lobular); histological grade; and lymph node status.

HRs for ER-positive
HRs for ER-negative

BRCA1-BRCA2
BCAC-BRCA1 BCAC-BRCA2 Association results for the general population were taken from reports published by the Breast Cancer Association Consortium (BCAC) [8][9][10]16].   Reference allele. d Effect allele. e Variant located in previously reported breast cancer susceptibility locus. f P-value for the difference in association between subtype positive and subtype negative breast cancer (for example, ER-positive vs ER-negative). g Non-triple negative was considered as "positive" and triple-negative as "negative."   After stratifying by ER status, we observed additional associations that were not seen for overall breast cancer. Among the 43 susceptibility variants that were evaluated in mutation carriers for the first time, we identified two additional associations in BRCA1 carriers when stratifying by ER status (3q26.1, 6p25.3) and four in BRCA2 carriers (2q24, 14q13.3, 19q13.3, 22q12.2). Population-based studies have shown that seven of the 74 breast cancer susceptibility variants display stronger associations with ERnegative disease in the general population [9]. Consistent with these findings, SNPs at 1q32.1 (MDM4), 5p15.33, 6q25.1 and 19p13 were associated with ER-negative breast cancer in BRCA1 carriers and SNPs at 2p24.1, 5p15.33 and 6q25.1 in BRCA2 carriers. No data were available for the SNP at 20q11.
We were able to confirm most of the associations with ER and PR subtypes of the 12 SNPs reported in the previous smaller CIMBA study [15]. In addition, two variants from that analysis now displayed evidence at P <0.05: rs13387042 at 2q35 with ER-positive breast cancer in BRCA1 carriers and rs2046210 at 6q25.1 with ER-negative breast cancer in BRCA2 carriers.
We also evaluated the associations of the 74 previously reported breast cancer susceptibility loci with other breast cancer subtypes. Variants at 5p15.33 (TERT), 6q25.1 (ESR1) and 19p13.11 showed associations in all subtype analyses for BRCA1 carriers. These are the most strongly associated loci for overall breast cancer in BRCA1 carriers, but they had not previously been investigated for their roles in subtypes other than ER. Variants at these three loci were associated with ER-, PR-and HER2-negative and TN subtypes. These variants were also associated with risk of high-grade tumors, with some suggestive evidence that this association was different from the association with grades 1 and 2 tumors for SNPs at ESR1 and 19p13. The three loci were associated with ductal as well as nonductal subtypes and node-positive as well as nodenegative breast cancer. For BRCA2 carriers, SNPs at loci 10q26 (FGFR2) and 16q12.1 (TOX3) were associated with all subtypes of breast cancer. SNPs at FGFR2 and TOX3 have consistently been associated with overall and with ER-positive breast cancer in population-based cases [25] as well as in BRCA1/2 carriers [15]. Furthermore, SNPs at these two loci were associated with PR-positive and HER2-negative disease. There was no evidence for a difference in HR estimates by tumor grade, nodal involvement or morphologic subtype (ductal).
It is important to note that for each of the 74 known loci considered, we evaluated only the associations for the specific SNPs that have been reported by the BCAC. We have not considered all genetic variants within a given region. Therefore, we cannot rule out the possibility that more strongly associated variants exist at these loci than the SNPs reported here. Future fine-mapping efforts in conjunction with BCAC analyses should clarify this. Such studies may also identify the causal variant and together with subsequent functional studies gather insights about the functional mechanisms causing these association signals. This in turn may yield insights about the etiology of breast cancer in BRCA1 and BRCA2 carriers.
We compared HRs for the association with overall breast cancer and ER-positive and ER-negative breast cancer for all the 74 known breast cancer susceptibility variants between BRCA1 carriers, BRCA2 carriers and population-based studies using published data from BCAC. Although only some of these variants were associated at P <0.05 with breast cancer in mutation carriers as outlined above, there was nevertheless strong correlation between the HRs for overall breast cancer from the general population and those from BRCA2 carriers, and moderate correlation between the HRs from BRCA1 carriers and from BCAC. These results suggest that many of these variants may also be associated with breast cancer risk for mutation carriers, but the power to detect statistically significant associations in mutation carriers is low. These variants could be employed in risk prediction models for mutation carriers. Future studies should be aimed at assessing the associations of the combined effects of the SNPs in mutation carriers in terms of polygenic risk scores.
We used these comparisons to assess the hypothesis that observed differences in the associations between BRCA1 and BRCA2 carriers and the general population are mediated by ER status. The smaller correlation between the association estimates for BRCA1 carriers and both BCAC and BRCA2 carriers compared with those between BCAC and BRCA2 carriers is consistent with this hypothesis. Moreover, we found stronger correlations between the HRs for ER-positive disease in all three two-way comparisons (ICC = 0.61 to 0.69). The correlation between ORs for ER-negative disease from BCAC and HRs for ER-negative disease from BRCA1 carriers was also strong (ICC = 0.59). Correlations diminished when comparing HR estimates for ER-positive with estimates for ER-negative breast cancer; most of them were not significantly different from zero. This finding suggests that, to a large extent, the difference in SNP association patterns is due to mediating effects of tumor ER status. Under such a model, the effects of common breast cancer susceptibility variants and of BRCA1 and BRCA2 mutations on breast cancer risk would be multiplicative, after taking into account tumor ER status. As BRCA1 carriers are more likely to develop ER-negative disease, SNPs associated with this subtype will be more informative in models to predict overall breast cancer risk in these women, whereas SNPs associated with ERpositive disease will be more useful in BRCA2 carriers. Furthermore, ER-specific SNP associations could be used to provide separate estimates of ER-negative and ERpositive breast cancer risk for mutation carriers. This understanding allows the development of more refined models and is critical to the provision of accurate information to women considering more targeted preventive options.
However, the fact that the correlations between the HR estimates matched for ER status were smaller than 1 implies that there were still some differences in the associations after accounting for ER status. This could be due to sampling error or to real differences in genetic associations between BRCA1 and BRCA2 carriers and the general population. There are examples for such differences: BRCA1 and BRCA2 carrier-specific modifiers, such as the recently identified variant at 6p24, which was associated with breast cancer risk only in BRCA2 carriers [6], and the ovarian cancer susceptibility locus 4q32, which appeared to modify ovarian cancer risk only for BRCA1 carriers [7]. In addition, an ovarian cancer susceptibility locus 17q11.2 identified through populationbased data has been shown to display a consistent association in BRCA2 carriers, whereas an association of similar magnitude has been ruled out in BRCA1 carriers [26]. The extent to which genetic susceptibility to breast cancer in mutation carriers and in the general population is shared, as well as the extent to which it is mediated by ER status, need to be quantified systematically by future studies.
We also assessed the associations of over 200,000 SNPs on the iCOGS array. We identified several variants not previously reported that were associated with breast cancer at P <10 −6 in the analyses by PR status, HER status, TN phenotype, histological grade, nodal involvement, ductal and lobular morphologic subtypes. In the absence of P-values at genome-wide significance levels for these associations to account for multiple testing, these associations require confirmation through by gathering additional data.
Although this is the largest study of its kind, the statistical power to detect associations of variants conferring small effects with specific tumor characteristics may be low, owing to the limited data available. Future studies of additional BRCA1 and BRCA2 carriers with detailed tumor pathology information on new and previously recruited mutation carriers are needed. In this study, tumor pathology information was retrieved primarily from medical records. Despite extensive efforts, it is difficult to control the quality of these data. If there is low reproducibility in the classification of tumor characteristics for some samples, this could potentially add to the sampling error and make it more difficult to detect subtype-specific associations.

Conclusions
We have identified additional genetic modifiers of breast cancer risk for mutation carriers among reported breast cancer susceptibility loci. Large differences in absolute risk are expected between mutation carriers who carry many and mutation carriers who carry few risk alleles of modifying variants [6,7]. Therefore, in combination with previously identified modifiers, these variants may be of value for cancer risk prediction. Moreover, our results show that, to a large extent, the differences in breast cancer associations of known breast cancer susceptibility loci between BRCA1 and BRCA2 carriers and the general population are due to differences in the prevalence of tumor subtypes in BRCA1 and BRCA2 carriers. Estimates of the risks associated with these genetic variants based on large population-based association studies are likely to be applicable also to mutation carriers after taking ER status into account. Our results thus have implications for developing risk prediction models for breast cancer subtype-specific risks in mutation carriers that incorporate the effects of these SNPs.

Additional file
Additional file 1: Table S1. Ethics committees that granted approval for the access and use of the data for this study. Table S2. Hormone receptor definitions used by the studies. Table S3. Associations of breast cancer susceptibility variants with breast cancer by tumor ER status after excluding prevalent cases. Table S4. Associations of breast cancer susceptibility variants with breast cancer by tumor PR status. Table S5. Associations of breast cancer susceptibility variants with breast cancer by tumor HER2 status. Table S6. Associations of breast cancer susceptibility variants with breast cancer by tumor triple-negative status. Table S7. Associations of breast cancer susceptibility variants with ductal breast cancer. Table S8. Associations of breast cancer susceptibility variants with lobular breast cancer. Table S9. Associations of breast cancer susceptibility variants with breast cancer by tumor grade. Table S10. Associations of breast cancer susceptibility variants with breast cancer by nodal involvement. Figure S1. QQ plots for SNP associations with ER-positive and -negative breast cancer in BRCA1 carriers. Figure S2. QQ plots for SNP associations with ER-positive and -negative breast cancer in BRCA2 carriers. Figure S3. Comparison of the breast cancer associations of breast cancer susceptibility loci in BCAC and in BRCA1 and BRCA2 carriers. Figure S4. Comparison of the associations of breast cancer susceptibility loci by ER status in BRCA1 and in BRCA2 carriers. Figure S5. Comparison of the associations of breast cancer susceptibility loci by ER status in population-based studies (BCAC) and in BRCA1 carriers. Figure S6. Comparison of the associations of breast cancer susceptibility loci by ER status in population-based studies (BCAC) and in BRCA2 carriers.

Competing interests
The authors declare that they have no competing interests. Fellowship scheme. KOHBRA is supported by a grant from the National R&D 65