|Home | About | Journals | Submit | Contact Us | Français|
Recently, several genome-wide association studies have identified various genetic susceptibility loci for breast cancer. Relatively little is known about the possible interactions between these loci and the established risk factors for breast cancer.
To assess interactions between single-nucleotide polymorphisms (SNPs) and established risk factors, we prospectively collected DNA samples and questionnaire data from 8576 breast cancer case subjects and 11892 control subjects nested within the National Cancer Institute’s Breast and Prostate Cancer Cohort Consortium (BPC3). We genotyped 17 germline SNPs (FGFR2-rs2981582, FGFR2-rs3750817, TNRC9-rs3803662, 2q35-rs13387042, MAP3K1-rs889312, 8q24-rs13281615, CASP8-rs1045485, LSP1-rs3817198, COL1A1-rs2075555, COX11-rs6504950, RNF146-rs2180341, 6q25-rs2046210, SLC4A7-rs4973768, NOTCH2-rs11249433, 5p12-rs4415084, 5p12-rs10941679, RAD51L1-rs999737), and odds ratios were estimated by logistic regression to confirm previously reported associations with breast cancer risk. We performed likelihood ratio test to assess interactions between 17 SNPs and nine established risk factors (age at menarche, parity, age at menopause, use of hormone replacement therapy, family history, height, body mass index, smoking status, and alcohol consumption), and a correction for multiple testing of 153 tests (adjusted P value threshold = .05/153 = 3 × 10−4) was done. Case–case comparisons were performed for possible differential associations of polymorphisms by subgroups of tumor stage, estrogen and progesterone receptor status, and age at diagnosis. All statistical tests were two-sided.
We confirmed the association of 14 SNPs with breast cancer risk (Ptrend = 2.57 × 10−3 –3.96 × 10−19). Three SNPs (LSP1-rs3817198, COL1A1-rs2075555, and RNF146-rs2180341) did not show association with breast cancer risk. After accounting for multiple testing, no statistically significant interactions were detected between the 17 SNPs and the nine risk factors. We also confirmed that SNPs in FGFR2 and TNRC9 were associated with greater risk of estrogen receptor–positive than estrogen receptor–negative breast cancer (Pheterogeneity = .0016 for FGFR2-rs2981582 and Pheterogeneity = .0053 for TNRC9-rs3803662). SNP 5p12-rs10941679 was statistically significantly associated with greater risk of progesterone receptor–positive than progesterone receptor–negative breast cancer (Pheterogeneity = .0028).
This study does not support the hypothesis that known common breast cancer susceptibility loci strongly modify the associations between established risk factors and breast cancer.
Genome-wide association studies have identified many single-nucleotide polymorphisms (SNPs) that are associated with breast cancer risk. Several established epidemiological factors are also associated with breast cancer risk. However, it is not well understood whether the interactions between the SNPs and established risk factors can modify breast cancer risk.
Prospective nested case–control study within the National Cancer Institute’s Breast and Prostate Cancer Cohort Consortium (BPC3) to analyze the associations of 17 germline SNPs with breast cancer risk. Interactions between these 17 SNPs and nine established risk factors (age at menarche, parity, age at menopause, use of hormone replacement therapy, family history, height, body mass index, smoking status, and alcohol consumption) were tested.
Of the 17 SNPs, 14 showed association with breast cancer risk. After correction for multiple testing, no statistically significant interactions between the 17 SNPs and the nine risk factors were detected.
The common polymorphisms associated with breast cancer risk tested in this study did not modify the association between established risk factors and breast cancer risk.
Majority of the white subjects were of European descent, so analyses in other race or ethnicity were of limited statistical power.
From the Editors
Recently, multiple breast cancer susceptibility loci have been identified by several genome-wide association studies (GWAS) or studies of specific candidate single-nucleotide polymorphisms (SNPs) (1–10). Genetic variants that showed strong statistically significant associations with breast cancer risk (odds ratios [ORs] = 1.15–1.45, P < 5 × 10−7) were identified in fibroblast growth factor receptor 2 (FGFR2). The vicinity is referred to all genes, not only FGFR2. In other words, some of the SNPs mentioned in this paragraph are located directly within the mentioned genes, some others are near the genes. TOX high mobility group box family member 3 (TOX3; also known as TNRC9), mitogen-activated protein kinase kinase kinase 1 (MAP3K1), caspase 8 (CASP8), lymphocyte-specific protein 1 (LSP1), collagen type I alpha 1 (COL1A1), cytochrome c oxidase assembly homolog 11 (COX11), ring finger protein 146 (RNF146), solute carrier family 4 member 7 (SLC4A7), neurogenic locus notch homolog protein 2 (NOTCH2), RAD51-like 1 (RAD51L1) genes, or in the vicinity of these genes, as well as in gene-poor regions on chromosomes 2q35, 8q24, 6q25, and 5p12 (1–10).
These SNPs are genetic markers and do not necessarily represent the functional variants responsible for the association with breast cancer risk. Relatively little is known about the possible interplay between established epidemiological and genetic risk factors for breast cancer risk (11–19). Previous reports suggest that specific SNPs in FGFR2 (rs3750817 and rs1219648) modify the association between hormone replacement therapy (HRT) and breast cancer risk (13,14), although a large epidemiological study that included women from the Million Women Study did not confirm these findings (18).
Large-scale prospective data are needed to test reliably for interactions, defined here as departures from a multiplicative odds ratio model for the joint association of the SNPs and the established risk factors. In this study, we estimated interactions between 17 SNPs, previously reported to be associated with breast cancer risk and reaching genome-wide statistical significance in at least one previous study (FGFR2-rs2981582, FGFR2-rs3750817, TNRC9-rs3803662, 2q35-rs13387042, MAP3K1-rs889312, 8q24-rs13281615, CASP8-rs1045485, LSP1-rs3817198, COL1A1-rs2075555, COX11-rs6504950, RNF146-rs2180341, 6q25-rs2046210, SLC4A7-rs4973768, NOTCH2-rs11249433, 5p12-rs4415084, 5p12-rs10941679, RAD51L1-rs999737) (1–10), and nine established risk factors (age at menarche, parity, age at menopause, use of HRT, family history, height, body mass index (BMI), smoking status, and alcohol consumption), in the National Cancer Institute’s Breast and Prostate Cancer Cohort Consortium (BPC3), a large consortium of prospective cohort studies from Europe and the United States.
The BPC3 has been described in detail elsewhere (20). Briefly, the consortium includes large well-established cohorts assembled in the United States and Europe, which have both DNA samples and extensive questionnaire information. These cohorts are the American Cancer Society Cancer Prevention Study II (CPS-II) (21), the European Prospective Investigation into Cancer and Nutrition (EPIC) (22), the Nurses Health Study (NHS) (23), the Women’s Health Study (WHS) (24), the Prostate, Lung, Colorectal, Ovarian (PLCO) Cancer Screening Trial (25), and the Multiethnic Cohort (MEC) (26).
Case subjects were identified in each cohort by self-report with subsequent confirmation of the diagnosis from medical records or tumor registries and/or direct linkage with population-based tumor registries (method of breast cancer case confirmation varied by cohort). Control subjects were matched with case subjects by ethnicity and age and, in some cohorts, additional criteria, such as country of residence in EPIC. The requirement for each control subject was to be free of cancer up to the duration of follow-up of the matched case subject.
Most of the subjects were white and of European descent. One cohort (MEC) provided most of the DNA samples from nonwhite subjects. In total, we genotyped (described below) 8576 case subjects and 11892 control subjects, of whom 7023 case subjects and 10065 control subjects were white (of European descent), 389 case subjects and 423 control subjects were Latino, 430 case subjects and 471 control subjects were African American, 552 case subjects and 580 control subjects were Asian American (mostly of Japanese origin), and 148 case subjects and 297 control subjects were Native Hawaiian. Table 1 describes the study populations in detail.
Informed consent was obtained from all subjects, and the project was approved by the relevant institutional review boards for each cohort.
We selected SNPs that were reported to be associated with breast cancer risk and reached a commonly accepted threshold for genome-wide statistical significance (P < 5 × 10−7) (27) in at least one previous study. For two loci, we genotyped either the SNP reported in the original study or a surrogate in a complete or nearly complete linkage disequilibrium, based on data from the International HapMap Project (28,29). Namely, we genotyped rs4415084 or surrogate rs920329 (correlation coefficient r2 = 0.981 in HapMap Centre d’Etude du Polymorphisme Humain [CEPH] or CEU Utah residents with ancestry from Northern and Western Europe); likewise, we genotyped rs999737 or surrogate rs10483813 (r2 = 1 in HapMap CEU). We selected two SNPs from the locus on chromosome 5p12. Both were reported to be strongly associated with breast cancer risk (6), but there was only moderate linkage disequilibrium between the two SNPs (r2 = 0.5 in the Icelandic population of the original study; r2 = 0.51 in HapMap CEU subjects), and they could possibly represent two distinct susceptibility loci.
Genotyping assays were designed and performed using Taqman assays with reagents by Applied Biosystems (Foster City, CA). Details of primers and probes are available upon request. Genotyping of the breast cancer case subjects and control subjects was performed in four laboratories (located at the University of Southern California, the US National Cancer Institute, Harvard School of Public Health, and the German Cancer Research Center, DKFZ). Laboratory personnel were blinded to case–control status. Within each study, blinded duplicate samples (approximately 5%) were also included and concordance of these samples was greater than 99%.
The genotyping success rate was 94.14% overall (range 93.09%–95.62%). It was 97.11% (range 95.28%–98.86%) for white subjects of European descent.
DNA samples were excluded from further analysis if more than 25% of the SNPs in the samples failed genotyping. Genotype frequencies of each SNP were checked for deviation (P < 10−3) from the expected Hardy–Weinberg proportions among control subjects of European ancestry within a cohort or overall.
We examined whether each of the 17 selected SNPs (1–10) was associated with risk of breast cancer by fitting for each SNP an unconditional logistic regression model involving the SNP and adjustment for age at baseline, study, ethnicity (within MEC), and country (for EPIC). An additional analysis was run with a model including also the established risk factors as adjustment variables. Genotypes were coded either as allele count (trend test) or as three categories: for major-allele homozygotes (reference category), for heterozygotes, and for minor-allele homozygotes (two df test). We calculated Ptrend for each SNP as P value when coding minor alleles as trend variable. We performed these analyses in all subjects, separately for each ethnicity, and (for white subjects of European descent and African Americans) separately for each study within ethnicity.
We investigated the hypothesis that the odds ratios associated with nine established risk factors for breast cancer (age at menarche, parity, age at menopause, use of HRT, family history, height, BMI, smoking status, and alcohol consumption) could be modified by any one of 17 SNPs. Information on these established risk factors was recorded prospectively at the time the women joined the study or provided a blood sample (ie, before the diagnosis of breast cancer among the case subjects and at an equivalent time for the control subjects).
To test for interactions between SNPs and the established risk factors, we analyzed two models for each SNP—one with terms for the SNP and the covariate of interest, the other including additional interaction term(s) between the SNP and the covariate. SNPs were coded as counts of minor alleles (trend variable), and other risk factors were coded in categories. Both models were adjusted for age at baseline, study, ethnicity (within MEC), and country (for EPIC). We then computed the likelihood ratio test between the two models to test for a particular form of interaction, namely, departure from a multiplicative odds ratio model for the joint association of a genetic marker and an established risk factor. We did this for each SNP–covariate pair. The non-SNP variables were grouped as follows: 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); BMI (BMI < 25 kg/m2; 25 kg/m2 ≤ BMI < 30 kg/m2; BMI ≥ 30 kg/m2, separately for pre- and postmenopausal women); alcohol intake (non-drinker, <1 g alcohol per day; moderate drinker, <14 g alcohol per day; regular drinker, 14 g alcohol per day); height (<1.63 m or ≥1.63 m); use of HRT (never or ever use of any type of HRT, never or ever use of estrogen-only HRT, never or ever use of combined estrogen plus progestin HRT); smoking status (never, former, or current smoker); family history (presence or absence of first-degree relatives diagnosed with breast cancer); and parity (nulliparous or parous). We also computed stratum-specific odds ratios and 95% confidence intervals (CIs) for each SNP. An additional analysis was performed with models including not only each covariate of interest but also all other established risk factors. To correct for multiple testing of the 153 (17 × 9 = 153) tests performed, we evaluated statistical significance at an adjusted P value threshold (P = .05/153 = 3 × 10−4).
We performed case-only analyses to test for differences in the associations of SNP with breast cancer risk with respect to different prognostic factors. Specifically, we compared estrogen receptor–negative (ER−) case subjects with ER-positive (ER+) case subjects as the reference, and in a similar fashion progesterone receptor–negative (PR−) case subjects with PR-positive (PR+) case subjects, advanced case subjects with nonadvanced case subjects (advanced disease was defined as having regional or distant metastasis), and case subjects diagnosed before the age of 55 with case subjects diagnosed after the age of 55 years. A statistically significant association between an SNP and breast cancer subgroup in this analysis was interpreted as a statistically significant heterogeneous association of the SNP on the different disease characteristics. We also performed case–control analyses by subgroups according to ER+ or ER− status, PR+ or PR− status, advanced or nonadvanced disease, and age at diagnosis.
For all subjects, analyses were performed after adjusting for cohort, age, country within EPIC, study phase in NHS, and ethnicity for MEC and PLCO.
We calculated the interaction odds ratios (ie, ORs of the interaction term between each SNP and each established risk factor) that we could detect in our study with 80% or greater power. The power calculation was performed assuming a multiplicative model of interaction and taking into account multiple testing. For these calculations, we considered only the SNPs that showed a statistically significant association with breast cancer risk.
Mathematical models for all analyses are reported in the Supplementary Methods (available online). All statistical tests were two-sided, and all statistical analyses were performed with SAS version 9.2 (SAS Institute Inc, Cary, NC).
After exclusions, the analyses included 8576 breast cancer case subjects and 11892 control subjects from the six cohorts. The summary characteristics of the study population are shown in Table 1 and Supplementary Table 1 (available online). Case subjects and control subjects were predominantly white and were of European descent (7023 case subjects and 10065 control subjects; overall 83%) and peri- or postmenopausal (7097 case subjects and 9448 control subjects; overall 81%) at the time of enrollment. The mean age at diagnosis was 62.39 (SD = 9.08 years). All established risk factors were associated with breast cancer risk, as shown in Supplementary Table 2 (available online). None of the SNPs were excluded from further analyses because of deviation from fitness for Hardy–Weinberg proportion (P < 10−3).
For each SNP, the genotype frequencies in case subjects and control subjects and the association with breast cancer risk (Ptrend = 2.57 × 10−3–3.96 × 10−19) are shown in Table 2. Associations between SNPs and breast cancer risk did not differ materially from those reported previously (1–10), except for three SNPs that did not show evidence of association with breast cancer risk (LSP1-rs3817198, Ptrend = .89; COL1A1-rs2075555, Ptrend = .42; and RNF146-rs2180341, Ptrend = .11). Similar results were obtained when we corrected for multiple testing and adjusted for the established breast cancer risk factors; all SNPs remained statistically significantly associated with breast cancer risk (Ptrend = .023–1.02 × 10−11), except LSP1-rs3817198 (Ptrend = .33), COL1A1-rs2075555 (Ptrend = .74), and RNF146-rs2180341 (Ptrend = .60), as shown in Supplementary Table 3 (available online). Tests of heterogeneity for the associations between SNPs and breast cancer risk of SNPs across cohorts and different ethnic groups (Supplementary Table 4, available online) were not statistically significant, with the exception of TNRC9-rs3803662 (Pheterogeneity = 9.18 × 10−5) that showed statistically significant association with breast cancer risk in the white (P = 9.84 × 10−6) and African American (P = .004) subjects, an association of borderline statistical significance among Hispanic subjects (P = .04), and a non-statistically significant association in Asian American (P = .53) and Native Hawaiian (P = .65) subjects.
Table 3 shows the results of interaction tests between each of the 17 SNPs and the established risk factors, and the P values of likelihood ratio tests comparing models with or without interaction term(s) between SNPs and covariates are presented. The detailed results of tests for associations between SNPs and breast cancer risk within each stratum of established risk factors are shown in Supplementary Table 5 (available online). After correction for multiple testing, we observed no statistically significant interactions in any of the 153 (17 × 9 = 153) tests (adjusted P value threshold = .05/153 = 3 × 10−4). The strongest statistical significance was observed for interaction between 5p12-rs10941679 on chromosome 5 and use of estrogen-only HRT (P = .0072) (Table 3). This SNP was associated with increased breast cancer risk in users and nonusers of estrogen-only HRT but more strongly associated in the group of users, as measured by odds ratio of an increasing number of minor alleles (nonusers, ORallele = 1.10, 95% CI = 1.02 to 1.19; users, ORallele = 1.35, 95% CI = 1.19 to 1.53; ORinteraction = 1.22, 95% CI = 1.06 to 1.41). We did not observe any interaction between 5p12-rs10941679 and use of HRT overall (P = .97), or with combined estrogen plus progestin HRT (P = .80). Analyses taking into account the duration of HRT use showed that COX11-rs6504950 was associated more strongly with breast cancer risk in women who used HRT for more than 5 years than in women who used HRT for less than 5 years, although the interaction did not reach statistical significance when corrected for multiple testing (P = .0035) (data not shown). When we adjusted the statistical models for all established breast cancer risk factors, we did not observe any statistically significant interaction with the same adjusted P value threshold (Supplementary Table 6, available online). It was previously suggested that FGFR2-rs3750817 shows an interaction with HRT (13); however, we did not find any clear evidence of interaction with the use of HRT. Indeed, point estimates of risks associated with this SNP in HRT users (OR = 0.87, 95% CI = 0.81 to 0.94) and nonusers (OR = 0.79, 95% CI = 0.73 to 0.85) differed only at borderline level of statistical significance (P = .05; Table 3 and Supplementary Tables 5 and 7, available online).
To investigate whether the SNPs were associated with particular forms of breast cancer, we analyzed the associations between 17 SNPs and breast cancer risk by subgroups of advanced or nonadvanced disease, by ER or PR status, and by age at diagnosis (Table 4). We evaluated heterogeneity of associations between SNPs and breast cancer risk by case–case comparisons between case subjects grouped according to clinical variables. After correction for multiple testing for 17 SNPs (adjusted P value threshold = .05/17 = .0029), the results of the subgroup analyses showed that 5p12-rs10941679 was statistically significantly associated with greater risk of PR+ breast cancer than PR− breast cancer (Pheterogeneity = .0028), and FGFR2-rs2981582 was statistically significantly associated with greater risk of ER+ breast cancer than ER− breast cancer (Pheterogeneity = .0016). Additionally, if we considered associations showing P less than or equal to an arbitrary threshold of .01, another SNP on locus 5p12, rs4415084, also showed association with greater risk of PR+ breast cancer than PR− breast cancer (Pheterogeneity = .010). TNRC9-rs3803662 and the second SNP in the FGFR2 gene, rs3750817, were associated with greater risk of ER+ breast cancer than ER− breast cancer (Pheterogeneity = .0053 and .0063, respectively) (Table 4). FGFR2-rs2981582 was also associated with a higher risk of being diagnosed with breast cancer at younger age (Pheterogeneity = .0042), and similar was the observation for COL1A1-rs2075555 (Pheterogeneity = .0098) (Table 4). However, COL1A1-rs2075555 did not show an association with breast cancer risk overall in this study (Table 2). The SNP 8q24-rs13281615 was previously reported to be associated with risk of ER+ breast cancer, but not with risk of ER− breast cancer (30). However, we did not observe any evidence of a differential association between this SNP and breast cancer risk depending on ER status (Pheterogeneity = .13) (Table 4). The results of the subgroup analyses are shown in details in Supplementary Table 8 (available online).
In this article, we report findings from a consortium of large prospective studies of possible interactions between 17 polymorphisms that have been associated with breast cancer and established risk factors for the disease. Data were examined using a nested case–control design within the National Cancer Institute’s BPC3 and included 8576 case subjects with breast cancer and 11892 control subjects without breast cancer.
We replicated all of the previously reported associations between SNPs and breast cancer risk, except for LSP1-rs3817198, COL1A1-rs2075555, and RNF146-rs2180341, which did not show association with breast cancer risk. It is worth noting that the association with RNF146-rs2180341 was reported only in a small study focusing on Ashkenazi Jews (7), which did not include replication in samples of other populations. Likewise, the association with COL1A1-rs2075555 was reported by a single study with only 58 cases of breast cancer, nested in the Framingham Heart Study (4). In light of the lack of association between these two SNPs and breast cancer risk in our study, we think that they most likely represent false positives or are relevant only to specific populations, such as women of Ashkenazi Jewish ancestry. The association between LSP1-rs3817198 and breast cancer risk was investigated in several studies: A statistically significant association at genome-wide level (albeit with a rather low ORallele = 1.07) was reported by Easton et al. (1) but not confirmed in subsequent GWAS (8,9). Our results suggest that the association between this polymorphism and breast cancer risk is at best weak (ORallele = 1.00; 95% CI = 0.95 to 1.04; Ptrend = .89). For some of the other SNPs, whose associations with breast cancer risk are clearly replicated in our study, we found slightly lower odds ratios than reported in previous publications (1–10). However, the direction of associations was consistently the same, and our confidence intervals largely overlapped with those of the previous reports.
Previous studies have reported possible interactions between breast cancer susceptibility loci and established risk factors (13–17). These studies focused mainly on FGFR2 and MAP3K1 and hormonal and reproductive factors, particularly the use of HRT. A recent study within the Women’s Health Initiative (13) showed a possible interaction between SNPs in FGFR2 and HRT use. Another recent study (14) showed an interaction between SNP FGFR2-rs1219648 and use of combined HRT in women of European ancestry. These studies had smaller sample sizes than ours. FGFR2-rs1219648 is in strong linkage disequilibrium with FGFR2-rs2981582 (Pearson correlation coefficient r2 = 1) (28,29), which did not show any evidence of interaction with HRT overall or with subtypes of HRT in this study (Table 3 and Supplementary Table 5, available online). As shown in the article by Prentice et al. (13), the FGFR2 SNP showing the strongest interaction with HRT was rs3750817 (Pinteraction = .046 for use of estrogen-only HRT, and Pinteraction = .033 for use of estrogen–progestin HRT), which is only in modest linkage disequilibrium with rs2981582 (Pearson correlation coefficient r2 = 0.47). We genotyped rs3750817 in all the case subjects and control subjects in our analyses but did not observe any clear evidence of interaction with use of HRT. There was no evidence for interaction when we analyzed separately the use of estrogen-only HRT or combined estrogen plus progestin HRT.
A recent case–control study performed in a Japanese population (15) showed interactions between SNPs in FGFR2 and family history of breast cancer, age at menarche, and parity. We did not observe any statistically significant interactions between SNPs FGFR2-rs2981582 or FGFR2-rs3750817 and any of these risk factors. The study by Kawase et al. (15) had a much smaller sample than ours (456 case subjects and 912 control subjects), and statistical significance of the interactions reported was modest (the strongest result was observed for interaction with family history of breast cancer Pinteraction= .003); therefore, these could be chance findings.
Our results on interactions between SNPs and established risk factors are similar to those obtained in studies of comparable sample size, performed in the Million Women Study (18) and the Breast Cancer Association Consortium (19). Namely, no statistically significant interactions between SNPs and established breast cancer risk factors were detected in those studies, when multiple testing was taken into account (18,19).
Our study had greater than 80% power to detect interaction odds ratios (ie, ORs of the interaction term between each SNP and each established risk factor) ranging between 1.20 and 1.47 between the SNPs and the risk factors we considered. The power calculation was performed assuming a multiplicative model of interaction and taking into account multiple testing. Thus, we had a reasonably good chance to detect moderately large interactions between SNPs and established risk factors. This is also shown by the fact that 95% confidence intervals around interaction odds ratios were rather narrow for most SNP-established risk factor pairs (Supplementary Table 7, available online).
We cannot exclude the existence of real interactions of smaller magnitude (including interactions between SNPs and established risk factors that did not show a statistically significant association with breast cancer risk or had a relatively small association with breast cancer risk), which our study was not sufficiently powered to detect. If such interactions exist, they may shed light on poorly understood biological mechanisms, including the hitherto unknown function of most SNPs studied here. However, the relevance of such small interactions in terms of risk assessment and prevention would be limited.
Results from subgroup analyses on clinical characteristics of tumors were generally in agreement with previous reports (3,6,30–32), including a meta-analysis of all published data (32). Findings from previous studies suggested that several SNPs are predominantly associated with ER+ breast cancer: TNRC9-rs3803662 (3,30–32), 5p12-rs4415084 (6), 5p12-rs10941679 (6), FGFR2-rs2981582 (6,30–32), 8q24-rs13281615 (30). In addition, FGFR2-rs2981582 was also reported to be more strongly associated with PR+ cancers than with PR− cancers (30). The SNP 2q35-rs13387042 was reported to be associated exclusively with ER+ and PR+ cancers (3), although later reports have shown that it is associated with both receptor-positive and receptor-negative cancers (31,32). In our data, SNPs on chromosome 5p12, FGFR2 and TNRC9 were preferentially associated with ER+ and/or PR+ breast cancer. In addition, SNP 2q35-rs13387042 showed a strongly statistically significant association with risk in ER+ and PR+ cases but not with ER− and PR− cases, although the heterogeneity was not statistically significant in our data, in agreement with previous studies (31,32). Because ER and PR status are the major markers of breast cancer subtypes, these observations suggest that inherited risk variants of these subtypes may vary. Contrary to one previous report (30) but consistent with results from a second study (32), we did not observe any evidence that SNP 8q24-rs13281615 had a stronger association with breast cancer risk depending on ER or PR status.
Our study has a few limitations. The vast majority of white subjects in the study are of European descent, and statistical power for analyses in other ethnicities is limited. In addition, many statistical tests were performed and, given that there were no a priori hypotheses about the possible interactions of SNPs and established risk factors, our findings should be taken with caution. Nevertheless, this is one of the largest cohort studies to systematically investigate possible interactions between major established risk factors for breast cancer and polymorphisms in the known susceptibility regions. It is very unlikely that we had nondifferential measurement error to the extent that could be a serious flaw in our study. Genotyping quality was monitored by a series of intra- and interlaboratory measures, including blind duplicated samples and measures of deviation from Hardy–Weinberg equilibrium. With respect to the established risk factors we included in our analyses, it is known that they are reliably measured in prospective cohorts, as documented by specific validation studies performed in some of the BPC3 cohorts (33–37).
Our study provides evidence against the hypothesis that common polymorphisms associated with breast cancer risk strongly modify the association of established factors with breast cancer risk. Our null findings are important given the size, prospective design, and the comprehensive approach of our study. However, our results do not rule out small departures from a multiplicative odds model for the joint association of pairs of individual SNPs and risk factors, nor does absence of departure from a multiplicative odds model necessarily imply that these genetic loci and risk factors do not interact in some causal mechanism. Moreover, absence of interaction as we have defined it here does not imply absence of a “public health interaction,” where the benefit from reducing a risk factor in terms of absolute risk reduction differs across genotypes (38).
In conclusion, we studied almost 9000 women with breast cancer and 12000 control subjects without breast cancer and showed that the 17 low-penetrance breast cancer susceptibility polymorphisms studied here do not strongly interact with established risk factors.
US National Institutes of Health, National Cancer Institute (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 authors are solely responsible for the design of the study, the analysis and interpretation of the data, the writing of the article, and the decision to submit the article for publication. The writing team of the article included D. Campa, R. Kaaks, L. L. Marchand, C. A. Haiman, R. Travis, R. G. Ziegler, D. J. Hunter, S. Lindström, and F. Canzian.