Main

The protein-truncating mutation CHEK2*1100delC (checkpoint kinase 2) is a moderate-penetrance breast cancer risk variant with an estimated relative risk of two- to threefold.1,2 However, several studies have shown that the cumulative lifetime risk of breast cancer for CHEK2*1100delC carriers is markedly higher in women with a family history than in those without,3,4,5 and that CHEK2*1100delC carriers have a higher probability of developing bilateral breast cancer.6 These observations are quantitatively consistent with a simple polygenic model suggesting that CHEK2*1100delC combines multiplicatively with other genetic loci. However, this has not yet been established empirically.

Genome-wide association studies have identified common genetic variants that are associated with increased risk of breast cancer. A polygenic risk score (PRS) based on 77 low-penetrance variants may explain approximately 12–14% of the excess familial risk and has been shown to identify individuals at high risk at the population level.7,8 Some of these variants predispose predominantly to either estrogen receptor–positive (ER+) or estrogen receptor–negative (ER−) disease, the two main etiological subclasses of breast cancer.9 CHEK2*1100delC carriers are more strongly predisposed to ER+ disease: approximately 90% of carrier tumors are ER+ in comparison to 77–78% of noncarrier tumors.10

Here, we report the synergistic risk effects attributable to CHEK2*1100delC and the common breast cancer susceptibility variants individually and summarized in terms of the PRS.7,8

Materials and Methods

Study participants

Female patients with invasive breast cancer and healthy controls of European ancestry from studies participating in the Breast Cancer Association Consortium (BCAC) were included (Supplementary Table S1 online). Data from a study were included if the study provided genotype data of the common variants from at least one breast cancer patient carrying the 1100delC variant. This selection yielded data from 32 studies and a total of 79,202 study subjects, including 848 CHEK2*1100delC carriers (Supplementary Table S2 online) for pairwise interaction analyses. Complete quality-controlled7,10 genotype data for all common variants and CHEK2*1100delC were available from 33,624 study subjects (369 CHEK2*1100delC carriers, Supplementary Table S2 online). These data were used in the analyses involving the PRS.

All participating studies were approved by their institutional review committees. Each study followed national guidelines for participant inclusion and informed-consent procedures.

Genotyping

All variants except CHEK2*1100delC were genotyped centrally using a custom Illumina iSelect genotyping array (iCOGS, Illumina, San Diego, CA) as part of the COGS consortium studies as described.7,8 CHEK2*1100delC was genotyped primarily using a custom-made TaqMan assay (Applied Biosystems, Foster City, CA); a small minority was genotyped using iPLEX.10 In addition to the 38,549 study subjects genotyped using the iCOGS array, 40,653 BCAC study subjects were genotyped for up to 25 of the common risk variants. These data were used in the pairwise interaction analysis (Supplementary Tables S2 and S3 online). These samples were genotyped by independent studies following BCAC genotyping standards as described previously.11,12

Statistical analyses

Statistical analyses were performed using Stata SE 10 (StataCorp, College Station, TX) and R version 2.15.2 (R Foundation for Statistical Computing, Vienna, Austria). For the common variants, a log-additive model was assumed; i.e., the risk was analyzed in terms of the number of disease-associated alleles (0, 1, 2) carried. CHEK2*1100delC was assumed to follow a dominant inheritance model because the number of rare homozygotes was small (n = 19). All analyses adjusted for the study and seven principal components defined on the basis of the genome-wide data from the iCOGS project as described previously.7 All reported tests were two-sided.

Polygenic risk score

To investigate the combined effects of common variants and CHEK2*1100delC, a polygenic risk score (PRS) based on the main effects of the common variants was calculated using the formula , where n is the number of loci included in the model, a is the number of susceptibility alleles in locus i, and OR is the per-allele odds ratio for breast cancer, estimated separately for each variant in the entire data set (Supplementary Table S4a online, “All” column). Results using a PRS based on previously reported ORs7,8 were essentially identical (data not shown). The PRS was approximately normally distributed in all study subgroups and it was standardized by the mean and standard deviation of the PRS among healthy individuals.8 For pairs of linked variants with r2>0.75, we included in the PRS only the lead variant (rs2981579, not rs2981582; rs12662670, not rs3757318; rs554219, not rs614367). We excluded two variants (rs78540526 and rs75915166) included in the PRS used by Mavaddat et al.8 (which were not genotyped on the iCOGS array) as well as rs17879961, the CHEK2 missense variant I157T, because the number of study subjects carrying both 1100delC and I157T was very low (n = 5). Thus, the resulting PRS included 74 variants. The interaction between PRS and CHEK2*1100delC was assessed by comparing the following nested logistic regression models: a model including the PRS and 1100delC genotype and a model supplemented with an interaction term coded as the product of the PRS and 1100delC. In the analyses of the PRS and positive family history of breast cancer, positive family history was defined as at least one first-degree relative with breast cancer.

The cumulative lifetime breast cancer risk of CHEK2*1100delC carriers in different PRS percentiles was derived by assuming an average lifetime risk of 22% for CHEK2*1100delC carriers13 and previously published relative risk estimates associated with the PRS.8

Pairwise interaction analyses

We tested for pairwise interaction between each common variant and CHEK2*1100delC as described for the interaction between the PRS and 1100delC. P values were corrected for 77 parallel tests using the Benjamini-Hochberg method.14 The OR for breast cancer was estimated separately for each of the common variants for the whole dataset and for the subgroup of 1100delC carriers. These analyses were also performed separately using a subgroup of breast cancer patients with ER+ disease because 1100delC is associated with ER+ breast cancer.10 We tested for heterogeneity in the ORs among different BCAC studies by including (separately for each variant) an interaction term between the variant and the study. No significant heterogeneity was found for any variant (data not shown). Statistical power was estimated as previously suggested for risk interaction analyses.15

Results

We analyzed the combined effects of CHEK2*1100delC and common low-penetrance breast cancer risk variants using data from the international Breast Cancer Association Consortium (Supplementary Table S2 online). The PRS summarizing the individual effects of 74 common variants was strongly associated with breast cancer risk among CHEK2*1100delC carriers (OR per unit of standard deviation 1.59 (1.21–2.09), P = 0.0008) and the OR was similar to that of noncarriers (1.58 (1.55–1.62), Pinteraction = 0.93). ORs for the highest and lowest quintiles of the PRS distribution were 2.03 (0.86–4.78) and 0.52 (0.16–1.74) for CHEK2*1100delC carriers, respectively, when compared to the middle quintile ( Table 1 ). Both estimates were similar to those among noncarriers.

Table 1 Breast cancer risk associated with the PRS for noncarriers and the carriers of CHEK2*1100delC
Full size table

The OR associated with CHEK2*1100delC in the analysis data set (2.99 (2.32–3.85)) was attenuated when the model was adjusted for positive family history of breast cancer. The OR associated with the PRS was also slightly attenuated ( Table 2 ). No significant interaction between risk effects associated with 1100delC, PRS, and positive family history was found. However, in a case-only analysis, there was a significant association between the PRS and family history of breast cancer among both CHEK2*1100delC carriers (OR 1.29 (1.01–1.65), P = 0.04) and noncarriers (OR 1.17 (1.12–1.21), P = 4E-16) (Supplementary Figure S1 online).

Table 2 Relative breast cancer risk associated with CHEK2*1100delC, PRS, and positive family history of breast cancer in the analysis data set
Full size table

When 77 common variants were considered individually, we found nominally significant interactions between five variants and CHEK2*1100delC for overall breast cancer (rs11249433, rs11780156, rs204247, rs2981582, and rs704010; Supplementary Table S4a online). Two of these represented synergistic (more than multiplicative) and three antagonistic interactions (the estimated effect in 1100delC carriers being in the direction opposite to that in noncarriers). However, none of the interactions was significant after correction for multiple testing. Nine variants showed a nominally significant interaction for ER-positive breast cancer (Supplementary Table S4b online).

Discussion

Our analyses of the synergistic effects of CHEK2*1100delC and 77 common low-penetrance variants on breast cancer risk strongly support the predicted multiplicative polygenic model.8,16,17 Although this has previously been shown for combinations of low-penetrance variants8 and for variants in combination with BRCA1 and BRCA2 mutations,18 this is the first direct demonstration of a “moderate” risk gene; therefore, it has important implications for risk prediction. The PRS was a significant risk factor for CHEK2*1100delC carriers, and the estimated OR per unit of standard deviation was very similar in CHEK2*1100delC carriers and in noncarriers, consistent with the hypothesis that the common susceptibility variants combine with the rare CHEK2*1100delC variant in an approximately multiplicative fashion. Similarly, the PRS risk estimates for the highest and lowest quintiles did not differ between the CHEK2*1100delC carriers and noncarriers. These two estimates for the CHEK2*1100delC carriers alone did not reach statistical significance ( Table 1 ), possibly reflecting limited statistical power due to the relatively low number of healthy variant carriers (Supplementary Table S2 online). However, this is the largest study genotyped for CHEK2*1100delC and these common variants. Even though some of the point estimates are not significant, they are consistent with previous reports. Most importantly, we did not find evidence for deviation from the multiplicative model, suggesting that the PRS could be used in risk stratification of 1100delC carriers in a manner similar to that for noncarriers.

The unadjusted OR for the CHEK2*110delC variants ( Table 2 ) was higher in our analysis data set than that in previous reports.2,13 Adjusting for positive family history markedly attenuated the CHEK2*1100delC-associated OR, suggestive of some oversampling of familial cases. The PRS OR was also slightly attenuated after the adjustment. However, CHEK2*1100delC, PRS, and family history remained significant risk factors in the combined model ( Table 2 ), suggesting that the common variants together explain part of the excess familial risk as previously suggested,16 but that the PRS also has predictive value in breast cancer families segregating CHEK2*1100delC.

Recently, a large study estimating the risk associated with CHEK2*1100delC in relation to age, tumor subtype, and family history reported that the cumulative lifetime risk for 1100delC carriers was approximately 22%.13 Assuming that the relative effect of the PRS is the same in carriers and noncarriers (OR >1.48 (1.39–1.57) or <0.65 (0.60–0.70) for percentiles >80% or <20%, respectively),8 20% of the 1100delC carriers with the highest PRS would have a lifetime risk higher than 32.6% (30.6–34.5%), thus exceeding the threshold for the high-risk category (>30%) according to the UK NICE guidelines for familial breast cancer.19 Similarly, for the 20% of 1100delC carriers with the lowest PRS, the lifetime risk would be less than 14.3% (13.2%-15.4%), i.e., close to the average population risk. These observations imply that if CHEK2*1100delC is to be used for risk prediction, then it can be made more effective by including the PRS, which represents the risk-modifying effects of common variants, in the prediction.

CHEK2*1100delC carrier cancers do not represent a phenotypically distinct subgroup of breast carcinomas. Instead, the phenotypic diversity of CHEK2*1100delC-associated cancers resembles that of breast tumors in general.10 Thus, it was not surprising that the relative risks conferred by the common variants were similar for the CHEK2*1100delC carriers and noncarriers, and that no significant pairwise interaction was found. We estimated that we had sufficient statistical power (80%, at P < 0.05) to detect a pairwise interaction between CHEK2*1100delC and any of the common variants if the interaction OR was 2.5 or more, but not enough power to detect interactions comparable in magnitude to the risk effects associated with the low-penetrance variants (OR 1.1–1.5). Thus, it remains possible that more modest departures from a multiplicative model may exist. However, if so, then much larger case-control studies, perhaps combined with pedigree analyses, will be required to detect them.

In conclusion, our analyses confirm the predicted multiplicative relationship between CHEK2*1100delC and the common low-penetrance variants. Hence, the PRS could be applied to risk prediction of the variant carriers in a manner similar to that used for the general population. Most importantly, the PRS could help identify the high-risk group of the CHEK2*1100delC carriers who would benefit most from clinical intervention.

Disclosure

The authors declare no conflict of interest.