Common Breast Cancer Susceptibility Variants in LSP1 and RAD51L1 Are Associated with Mammographic Density Measures that Predict Breast Cancer Risk

Introduction

Genetic factors play a major role in the pathogenesis of breast cancer (1-3). Recent multi-stage genome-wide association studies (GWAS) and candidate gene studies conducted by several groups, including the Breast Cancer Association Consortium (BCAC), have successfully identified and replicated associations between over 18 single nucleotide polymorphisms (SNPs) and risk of breast cancer in Caucasians (4-9).

Mammographic density, which reflects variations in the amounts of fat, stromal and epithelial tissues in the breast, is one of the strongest risk factors for breast cancer with risk being 4-6 fold higher for women in the highest relative to lowest density categories after adjusting for age and body mass index (BMI) (10, 11). The biology underlying the mammographic density and breast cancer association is essentially unknown, but twin and family studies suggest that additive genetic factors explain ~60% of variance in the density measures (12, 13). This raises the question of whether breast cancer susceptibility variants identified to date are associated with mammographic density measures. This could lead to new insights into the etiology of breast cancer by revealing the biological reasons for these associations with breast cancer risk (14).

Five studies have examined the association of breast cancer susceptibility SNPs with age and BMI adjusted measures of mammographic density (14-18). The most consistent finding was an association between [lymphocyte-specific protein-1, LSP-1]-rs3817198 and adjusted dense area and percent density, in the same direction as the association with breast cancer. The association was observed overall by Odefrey et al (17) but only in specific subgroups by others: in premenopausal women (14), current users of postmenopausal hormones (PMH),(15) or ER+/PR+ cases only (16). Other nominally significant reported SNP-density associations consistent with the association of these SNPs with breast cancer risk include associations of TOX3-rs12443621 (14, 15) and rs4666451 (14) with adjusted percent density, in pre-menopausal women only, and rs13281615 at 8q24 with both adjusted percent density and dense area (17). The largest study to date, a meta-analysis of five GWAS of mammographic density involving 4877 women with and without breast cancer, identified a genome-wide significant association between ZNF365- rs10995190, a known breast cancer susceptibility SNP, and adjusted percent density as well as weak evidence of possible associations with ESR1-rs2046210 (p=0.005) and LSP1-rs3817198 (p=0.04) (18).

Only one previous study (17), however, examined the SNP associations with the components that comprise the percent density phenotype, namely dense area and non-dense area. Dense area has been hypothesized to be the more relevant density phenotype for understanding the etiology of mammographic density (19) as tumors have been shown to arise within the radiodense tissue (20). Whether these SNPs influence dense and/or non-dense area could help to interpret the mechanism by which the loci influence density and possibly cancer.

We established an international collaboration - the DENSNP consortium - of studies with data on established breast cancer susceptibility variants and quantitative density measures from film mammography to conduct analyses of breast cancer susceptibility SNPs in relation to the three density phenotypes. This paper reports the findings for 15 breast cancer SNPs at 14 loci, identified through 2009 when the DENSNP consortium was established.

Materials and Methods

Results

There were 5,110 breast cancer cases and 11,785 non-cases of self-reported Caucasian race/ethnicity with available density phenotypes, risk factors and at least one of the 15 SNPs considered [Table 1]. The number of participants varied by SNP with the most comprehensive information for 2q35 (n=13,254), CASP8 (n=12,816) and FGFR2 (n=12,680), and least information for TGFB1 (n=3,099), RAD51L1 (n=7,610) and ESR1 (n=8,274).

The majority of the participants were aged ≥40 years (98%) and postmenopausal (77%), and approximately half of those aged ≥55 reported ever using PMH (48%) [Table 1]. In all, 44% of participants had a BMI<25 kg/m² [Table 1]. A small proportion was nulliparous (11%), precluding subgroup analyses by parity. The associations between these variables and the three density phenotypes are shown in Table 2, and were similar to those reported in the literature.

The results from our primary analyses of the 15 SNPs in 14 breast cancer loci with the three density phenotypes are shown in Figure 1 and described in Supplemental Tables 5a-c. Pictured are the parameter estimates from the mixed linear models corresponding to each genotype. There was strong evidence against the null hypothesis that none of the SNPs were associated with both the dense area (p<0.001) and percent density measures (p=0.001), but not with the non-dense area measure (p=0.5). This suggests that at least one of the 14 breast loci is associated with the density or dense area measures.

The strongest associations were seen with rs3817198 (LSP1) and the dense area (p=0.00005) and percent density (p=0.001) phenotypes with little evidence for between-study heterogeneity [Figure 2]. The adjusted mean dense area was 23.7cm² for T/T carriers, 25.1cm² for T/C carriers and 26.0cm² for C/C carriers (Supplemental Table 5a-b). The adjusted mean percent density for T/T carriers was 19.4% compared to 20.1% for T/C and 20.5% for C/C carriers, respectively. These associations were consistent across studies [Figure 2] and persisted after exclusion of studies that had previously reported on LSP1 and density, namely NHS, AMDTSS, LIFE, MEC, EPIC-Norfolk I and SASBAC(14-18) (e.g. p=0.004 for dense area). There was also evidence of an inverse association between rs10483813 (RAD51L1) and adjusted percent density (p=0.003), but not with adjusted dense area (p=0.07) [Figure 1]. These associations were consistent across studies [Figure 2] with the adjusted mean percent density for T/T genotype being 21.1%, compared to 20.5% for T/A and 19.0% for A/A.

There were nominal associations of adjusted percent density and dense area with rs2046210 (ESR1), rs1045485/rs17468277 (CASP8), rs4973768 (SLC4A7/NEK10) and rs3803662 (TOX3) [Supplemental Tables 5a-b] which were in the direction of the published corresponding breast cancer associations but not statistically significant after taking into account multiple testing [Figure 1]. None of the investigated SNPs were associated with non-dense area [Figure 1; Supplemental Table 5c].

The genetic associations above did not diminish after further adjustment for parity or view (data not shown) and, in general, did not appear to differ by case status, BMI, menopausal status, or PMH use [Supplemental Tables 6a-c] but the study had low power to examine interactions.

We also examined the association of these SNPs with breast cancer risk before and after adjustment for the density measures by pooling data from studies that recruited both cases and non-cases [identified in Supplemental Table 1]. Using 3,175 cases and 6,504 non-cases from eight studies, the per C-allele odds ratio (OR) for rs3817198 (LSP1) was 1.04 (95% CI 0.97, 1.12) without adjustment for either density measure. When including dense area as a covariate, the OR was 1.03 (95% CI 0.96, 1.10), and after adjustment for percent density instead, the OR was 1.02 (95% CI 0.95, 1.11). Similarly, using 2,765 cases and 3,022 non-cases from four studies, the per A-allele OR for rs10483813 (RAD51L1) was 0.92 (95% CI 0.84, 1.00) without adjustment for either density measure, 0.93 (95% CI 0.85, 1.01) after adjustment for dense area, and 0.94 (95% CI 0.86, 1.03) after adjustment for percent density.

Discussion

There is wide inter-individual variability in mammographic density measures, but known epidemiologic risk factors account for only 20-30% variability in percent density (13, 24, 25). We hypothesized that common low-penetrance breast cancer susceptibility variants contribute to the remaining inter-individual differences in the density phenotypes and examined this within a large international consortium (DENSNP). Here, we report the first findings from this collaborative effort and identify associations between adjusted measures of density and two breast cancer susceptibility SNPs, rs3817198 (LSP1) and rs10483813 (RAD51L1), which were in the same direction as the corresponding SNP associations with cancer risk.

The most marked association with density was with rs3817198 (LSP1). We also confirmed this association using the 10 studies that had not previously published on the LSP1 variant and density association, providing consistent evidence for this mammographic density locus. The mechanisms through which this SNP (or more likely the causal allele(s) it tags) may affect density and cancer risk are unclear. The LSP1 gene encodes an intracellular F-actin binding protein, which is expressed in lymphocytes, neutrophils, and endothelium and might regulate neutrophil motility, adhesion to fibrinogen matrix proteins, and transendothelial migration (26).

The SNP rs3817198 in RAD51L1, a gene on chromosome 14q24.1 involved in the double-strand DNA-repair and homologous-recombination pathway, may also be associated with the adjusted density measures, although the evidence is less compelling than for rs3817198 (LSP1). The biological mechanisms underlying the possible association of this variant with density and cancer risk are unknown. RAD51L1 interacts with RAD51, and a SNP in the 5’UTR of RAD51 has been found to be associated with breast cancer risk for BRCA2 mutation carriers (27). However, mutations in BRCA1 and BRCA2 have not been found to be associated with the density phenotypes (28, 29).

Several breast cancer GWAS have consistently identified polymorphisms in intron 2 of fibroblast growth factor receptor 2 (FGFR2), with each copy of the T allele of rs2981582 being associated with about a 26% increased breast cancer risk (30). Our study had 90% power to detect an average difference in percent density of less than 1% between homozygote carriers and non-carriers of this SNP, if such a difference truly exists, and therefore the lack of finding an association suggests that density is unlikely to mediate the association between FGFR2 and breast cancer risk. Similar considerations apply to SNPs in several other breast cancer loci, including TOX3-rs3803662, 2q35-rs13387042 and MAP3K1-rs889312. These loci are likely to contribute independently of density to risk prediction. In fact, when we added LSP1-rs3817198 and RAD51L1- rs10483813 to a risk model with age, BMI, menopause, study and percent density the inclusion of these two SNPS did not affect the AUC whereas the addition of the remaining 12 SNPs increased the AUC from 0.62 to 0.65 (p<0.001).

Previous studies were based on smaller sample sizes (ranging from 578 (16) to 4,877 (18)), which could have precluded the detection of small effects. Our study is the largest conducted so far with sample sizes greater than 6,000 for all but one SNP and greater than 10,000 for all but 5 SNPs. We had over 90% power to detect per-allele differences in adjusted percent density of 1% or less for all but three SNPs (rs17468277, rs10483813 and rs4415084), and even for these SNPs, we were similarly powered to detect per-allele differences of less than 2%. However, limited power precluded a more detailed examination of interactions with BMI (e.g. differential SNP effects in BMI-defined quartiles) and PMH use (e.g. different SNP effects by type of PMH, recency of use). The study also had low power to assess the mediation of the SNP and breast cancer associations by density.

The mammographic density readings were performed in different sets of films (e.g. left, right or both breasts; CC or MLO views), but it is unlikely that this may have affected substantially our findings because there is a high correlation between a woman’s density measurements taken from the various breast-view combinations(31). For cases, both pre-diagnostic films and films from the unaffected breast at the time of diagnosis, but prior to treatment, were used - an approach used by others (10); furthermore, our findings were not modified by case status. One small study (PNS) used digitized copies of digital mammograms, but its exclusion did not affect the results shown here. Although mammographic density readings were not standardized, all studies used a similar interactive-threshold approach and had very high within- and between-observer repeatability (typically >90%) (32). Also, all analyses were adjusted for study hence minimizing the impact of any between-study differences on density measurements which would have likely reduced our power to detect real associations. Reassuringly, we were able to reproduce the well-established influences of age, BMI, parity, menopausal status and PMH on density phenotypes within each one of the participating studies as well as in joint analyses.

Our findings suggest that two of 14 well-established breast cancer loci may contribute to the large between-woman differences in risk-predicting density phenotypes, consistent with estimates of 5-10% genetic overlap between this biomarker and breast cancer (33). The two common variants in LSP1 and RAD51L1 explained 0.2% (combined, 0.1% for each) of the variance in adjusted percent density and dense area, although the overall contribution could be larger if the true causal variants are more strongly associated with density than the tagging SNPs we examined here. At the individual level, these SNPs were associated with a 0.6% absolute increase in percent density per allele for LSP1 and 0.8% absolute decrease in percent density per allele for RAD51L1. These magnitudes can be compared with, for example, the change in density measures of 1% decrease per year of ageing (34), 2% increase with use of PMH and 2% decrease over the menopausal transition (35). Our findings are consistent with the hypothesis that mammographic density is likely a polygenic trait, influenced by many common low-penetrance variants, and/or rarer variants with larger effects which cannot be identified through current GWAS. Identification of such variants, and clarification of their role and function, is likely to improve our understanding of the biology of mammographic density and how this phenotype is associated with breast cancer risk.

Supplementary Material