Gene–gene interactions in breast cancer susceptibility

Abstract

There have been few definitive examples of gene–gene interactions in humans. Through mutational analyses in 7325 individuals, we report four interactions (defined as departures from a multiplicative model) between mutations in the breast cancer susceptibility genes ATM and CHEK2 with BRCA1 and BRCA2 (case-only interaction between ATM and BRCA1/BRCA2 combined, P = 5.9 × 10^–4; ATM and BRCA1, P= 0.01; ATM and BRCA2, P= 0.02; CHEK2 and BRCA1/BRCA2 combined, P = 2.1 × 10⁻⁴; CHEK2 and BRCA1, P= 0.01; CHEK2 and BRCA2, P= 0.01). The interactions are such that the resultant risk of breast cancer is lower than the multiplicative product of the constituent risks, and plausibly reflect the functional relationships of the encoded proteins in DNA repair. These findings have important implications for models of disease predisposition and clinical translation.

INTRODUCTION

Breast cancer is twice as common in women with an affected first degree relative and twin studies suggest that the majority of this familial aggregation is due to genetic factors. Three strata of breast cancer predisposition alleles are currently delineated on the basis of their prevalence and penetrance. High penetrance genes, such as BRCA1 and BRCA2, are characterized by multiple, individually rare, loss-of-function mutations that confer a high risk of breast cancer (RR > 10). Intermediate penetrance genes such as ATM, CHEK2, BRIP1 and PALB2 are characterized by rare, loss-of-function mutations that confer more modest risks (RR 2–4). Low penetrance variants, identified through genome-wide association studies (GWAS), are common but associated with smaller increases in risk (RR 1.08–1.26) (1–3).

While examples of gene–gene interactions are well recognized in model systems, there have been few definitive examples in humans. The identification by GWAS of multiple common variants predisposing to a given disease has fostered new opportunities, and renewed interest in gene–gene interactions in humans. For example, a gene–gene interaction was reported in a GWAS of psoriasis, whereby a variant in ERAP1 only confers risk of disease on certain human leukocyte antigen backgrounds (4).

In breast cancer susceptibility, the common alleles identified through GWAS exhibit independence of effects; the risk conferred by each combination of common variants is well modelled by the multiplicative product of the individual risks (2,3). Additionally, studies of the interplay of common variants with BRCA1 and BRCA2 mutations have likewise shown no evidence of interaction; the common variants that confer risk of estrogen receptor (ER)-positive breast cancer in the general population also confer a relative risk of similar magnitude in BRCA2 mutation carriers (in whom breast cancers are largely ER positive) (5,6). Similarly, variants at 19q that confer risk of ER-negative breast cancer in the general population also confer a relative risk of similar magnitude in BRCA1 mutation carriers (who typically develop ER-negative breast cancers) (7,8).

We previously identified rare, intermediate-penetrance genes that predispose to breast cancer (1,9,10). For one of these genes, CHEK2, we observed a lower mutation frequency in probands from BRCA1/BRCA2 mutation-positive pedigrees combined (termed ‘BRCA-positive’), than in probands from familial BRCA1/BRCA2 mutation-negative pedigrees (termed ‘BRCA-negative’) (9). This observation was supported in a subsequent study (11,12). However, it should be noted that the size of our previous study precluded evaluation of BRCA1 and BRCA2 separately, and combined data from populations that were subsequently found to have different CHEK2 mutation population frequencies (13). Nevertheless, the available data suggest that, in contrast to the common variants, the risks associated with mutations in intermediate-penetrance genes and the risks associated with mutations in BRCA1 and/or BRCA2 may not combine independently. Here, we have sought to further investigate this premise through analyses of the two most frequently mutated intermediate-penetrance genes, ATM and CHEK2, mutations of which are present in 2–3% of BRCA-negative familial breast cancer pedigrees (9,10). Since mutations in the high and intermediate-penetrance genes are rare in the general population, we performed a case-only analysis in breast cancer cases evaluating for departure from a baseline multiplicative (log-additive) combination of disease relative risks. The case-only analysis is a more powerful approach than a case–control analysis in this context, with the extra power being afforded through the assumption that the two factors under analysis are uncorrelated in the general population, which we believe to be valid as BRCA1, BRCA2, CHEK2 and ATM are all located on different chromosomes (14,15).

In contrast to the complexities of evaluating interactions among the multiple genotype combinations of common variants, there are three principal models of statistical genetic interplay between two genes in which mutations are rare. First, the mutations could act independently in which case the risks conferred would combine multiplicatively. Under this model, it would be anticipated that the frequency of ATM and/or CHEK2 mutations would be similar in BRCA1 and/or BRCA2-positive and BRCA-negative breast cancer cases. Secondly, the combination of mutations in the genes could be synergistic such that the resultant risk of breast cancer is higher than the product of risks. This would manifest as a higher frequency of ATM and/or CHEK2 mutations in BRCA1 and/or BRCA2-positive compared with BRCA-negative cases. Thirdly, there could be a negative interaction such that the resultant risk of breast cancer is lower than the multiplicative product of the constituent risks. This would manifest as a lower frequency of mutations in ATM and/or CHEK2 in BRCA1 and/or BRCA2-positive compared with BRCA-negative cases. From a biological perspective, any of these models are plausible and the interactions may differ for different pairs of genes.

To investigate these models, we undertook large-scale mutational analyses of ATM and CHEK2 in constitutional DNA from BRCA1-positive individuals, BRCA2-positive individuals, BRCA-negative individuals and population controls. We genotyped CHEK2_1100delC in 5280 unrelated women with breast cancer and a family history of the disease (358 BRCA1-positive, 357 BRCA2-positive, 4565 BRCA-negative) and in 2045 controls. Only ∼2% of these samples were included in our previous study (9). We also mutationally screened the full-coding sequence of ATM, in 71 assays, in 1411 of the 7325 samples (274 BRCA1-positive, 339 BRCA2-positive, 798 BRCA-negative) and in 515 controls. We used CHEK2_1100delC as a proxy for all mutations in the gene, because other pathogenic mutations in CHEK2 are extremely rare in the UK (16). For ATM, a large number of different mutations have been identified in the UK (10,17) and we therefore mutationally analysed the full gene and included all pathogenic mutations identified in our analyses. We compared the mutation frequencies using two-sided Fishers exact tests and also undertook multiple logistic regression to evaluate the interaction after adjusting for age of breast cancer and extent of family history.

RESULTS

The frequency of ATM mutations was 0/274 in BRCA1-positive cases, 1/339 (0.29%) in BRCA2-positive cases, 18/798 (2.26%) in BRCA-negative cases and 2/515 (0.39%) in controls (Table 1). Thus, the frequency of ATM mutations in BRCA-positive cases was similar to that in controls but significantly lower than that in BRCA-negative cases combined, and for each gene separately [interaction OR = 0.07 (95% CI: 0.002–0.45), P = 5.9 × 10⁻⁴ for BRCA1/BRCA2-combined, P= 0.01 for BRCA1 and P= 0.02 for BRCA2]. Case-only multiple logistic regression demonstrated that strength of family history was a predictor of ATM mutations but that age of breast cancer diagnosis was not. The reported interactions remained significant after adjustment for the effects of family history (Supplementary Material, Table S1).

Table 1.

Frequencies of ATM and CHEK2 mutations in BRCA-negative individuals, BRCA1-positive individuals, BRCA2-positive individuals and population controls

	ATM			CHEK2
	Total screened	ATM mutation (N)	ATM mutation (%)	Total screened	1100delC (n)	1100delC (%)
BRCA-negative	798	18	2.26	4565	94	2.06
BRCA1-positive	274	0	0	358	1	0.28
BRCA2-positive	339	1	0.29	357	1	0.28
Population controls	515	2	0.39	2045	8	0.39
Total	1926	21		7325	104

The frequency of CHEK2_1100delC was 1/358 (0.28%) in BRCA1-positive cases, 1/357 (0.28%) in BRCA2-positive cases, 94/4565 (2.06%) in BRCA-negative cases and 8/2045 (0.39%) in controls (Table 1). Thus, the frequency of CHEK2_1100delC in BRCA1-positive cases and in BRCA2-positive cases was each significantly lower than that in BRCA-negative cases [interaction OR = 0.13 (95% CI: 0.02–0.50), P = 2.1 × 10⁻⁴ for BRCA1/BRCA2 combined, P= 0.01 for BRCA1, P= 0.01 for BRCA2]. Again, case-only multiple logistic regression demonstrated that strength of family history was a predictor of CHEK2_1100delC mutations but that age of breast cancer diagnosis was not. The reported interactions remained significant after adjustment for the effects of family history (Supplementary Material, Table S2).

DISCUSSION

We have demonstrated a significantly lower frequency of mutations in each of CHEK2 and ATM in each of BRCA1-positive and BRCA2-positive breast cancer cases compared with BRCA-negative breast cancer cases. This observation is consistent with a model in which in the presence of a mutation in one gene, little or no additional risk is conferred by a second mutation in another gene, and thus differs from the model proposed for common, low risk variants in BRCA1 and BRCA2 mutation carriers (5–7).

Our data would also be consistent with models in which the risks conferred by mutations combine sub-multiplicatively, such as a model in which mutation-associated risks combine additively (on a penetrance scale). Experiments to estimate directly the risk of breast cancer to individuals carrying mutations in two genes would be informative; however, as such individuals are extremely rare, the required size of such studies currently renders them infeasible. It would also be of interest to evaluate the interplay of BRCA1 and BRCA2 mutations with mutations in other rare, intermediate-penetrance genes such as PALB2 and BRIP1 (18,19).

The interactions we have detected may reflect the biological relationships between the proteins encoded by these cancer susceptibility genes. BRCA1 has a central role in several pathways coordinating various cellular processes in response to DNA damage, including DNA repair and preservation of genomic integrity (20). BRCA2 acts downstream of BRCA1 and is involved in DNA repair through homologous recombination (20). ATM and CHK2 (which is encoded by CHEK2) have roles upstream of BRCA1. ATM phosphorylates CHK2 following DNA damage by ionizing radiation, which prevents entry of the cell into mitosis. CHK2 then associates with, phosphorylates and activates functions of BRCA1 (21). The evidence for a direct functional relationship between BRCA2 and either CHK2 or ATM is less robust. A parsimonious biological interpretation of our mutational interaction is a model whereby in the presence of functional abrogation of the downstream members of this pathway, i.e. BRCA1 and BRCA2, impairment of function of upstream proteins such as ATM and CHK2 confers little, or no additional risk of breast cancer.

Our findings have important implications for clinical risk estimation in breast cancer and other conditions. Increasingly, risk prediction algorithms are incorporating additional parameters, such as genotypic data from predisposition factors, to enhance predictive value. Typically, risk factors are assumed to act independently and are combined multiplicatively in these algorithms, but if the actual interaction deviates from this model, substantial inaccuracy in risk estimation may result. It is, therefore, critically important that the nature of interactions between risk factors is elucidated, to both improve our understanding of the genetic architecture of disease and for accurate risk estimation in the clinic.

MATERIALS AND METHODS

Samples

We included constitutional DNA extracted from whole blood from 5053 cases of breast cancer collected via the Genetics of Familial Breast Cancer Study (FBCS) (488 BRCA1 or BRCA2 mutation carriers and 4565 BRCA-negative cases). The samples were from unrelated Caucasian women ≥18 years from 23 genetics centres in the UK who had breast cancer and a family history of breast cancer. All the cases were screened for germline mutations, including large rearrangements, in BRCA1 and BRCA2. We included DNA from 227 BRCA1 or BRCA2 mutation carriers provided by the EMBRACE study (Epidemiological study of BRCA1 and BRCA2 mutation carriers; http://www.srl.cam.ac.uk/genepi/embrace/). The EMBRACE study is a cohort of individuals from BRCA1 and BRCA2 mutation-positive families and has been established through clinical genetics centres in the UK and Eire. The samples included in this experiment were from women aged ≥18 years who carry a mutation in BRCA1 or BRCA2.

The extent of family history was quantified using a Family History Score, defined by the number of relatives with breast cancer and weighted by their degree of relatedness to the index case. A score of 1.0 was assigned to the index case, with an additional 0.5 for each affected first degree relative, and an additional 0.25 for each affected second degree relative. The score of an individual with bilateral cancer was doubled.

We used 2045 controls from the 1958 Birth Cohort Collection, an ongoing follow-up of persons born in Great Britain in 1 week in 1958 which included a biomedical assessment during 2002–2004 at which blood samples and informed consent were obtained for creation of a genetic resource. At least 97% of these controls are of white ethnicity (http://www.b58cgene.sgul.ac.uk).

The study was conducted as part of our ongoing research to identify and characterize breast cancer susceptibility factors and was approved by the London Multicentre Research Ethics Committee (MREC/01/2/18).

Laboratory methods

Genomic DNA was extracted from whole blood using standard methods. We successfully genotyped CHEK2_1100delC in 7325 samples using either Taqman methodology (Applied Biosytems) or direct sequencing (primers available on request). Products were sequenced by capillary sequencing using the BigDye Terminator Cycle Sequencing Kit and an ABI 3730 Genetic Analyzer (Applied Biosystems, Foster City, CA, USA). Sequences were analysed using Mutation Surveyor software v3.20 (SoftGenetics, State College, PA, USA) and by visual inspection. All mutations were confirmed by bidirectional sequencing of a second, independently amplified polymerase chain reaction product.

We mutationally analysed 1926 samples through the full coding sequence and intron/exon boundaries of ATM in 71 fragments via Conformation Sensitive Gel Electrophoresis, as previously described (10). We used Taqman methodology to genotype the intronic mutation IVS40-1050A>G (which results in, 5762ins137, the most common ATM mutation in UK ataxia–telangiectasia families).

Statistical methods

Statistical analyses were performed using STATA10 (StataCorp, TX, USA). The frequencies of mutations in ATM and of CHEK2_1100delC were evaluated using two-sided Fisher's exact tests. The interaction odds ratios presented are those for the respective logistic models in which ATM mutation status or CHEK2_1100delC genotype is the response variable and BRCA1 and/or BRCA2 mutation status is the explanatory variable (14,15). In order to ensure that the interactions observed in the univariate case-only analysis were not influenced by the potential confounding effects of age of diagnosis or family history score, case-only multiple logistic regression was performed including ATM mutation status or CHEK2_1100delC genotype as the response variable and BRCA1 and/or BRCA2 mutation status, age of onset of breast cancer and family history score as explanatory variables. In each case, the modelling was performed evaluating BRCA1/BRCA2 mutations combined and BRCA1 mutations and BRCA2 mutations individually. Only samples for which all covariate data were available were included in these analyses (213/1411 samples excluded from the ATM analysis and 224/5280 samples excluded from the CHEK2_1100delC analysis). Each model was compared with a null model (one with no explanatory variables) via a likelihood ratio test. The effect of each explanatory variable was evaluated using a likelihood ratio test, comparing the saturated model including and excluding the explanatory variable of interest.

SUPPLEMENTARY MATERIAL

Supplementary Material is available at HMG online.

AUTHOR CONTRIBUTIONS

N.R., M.R.S. and C.T. designed the experiment. M.W.-P., N.R. and C.T. coordinated recruitment to FBCS. J.Ba., J.Be., A.F.B., C.B., G.B., C.Ch., J.C., R.D., A.D., F.D., D.G.E., D.E., L.G., A.H., L.I., A.K., F.L., Z.M., P.J.M., J.P., M.P., M.R., S.Sh. and L.W. coordinated FBCS sample recruitment from their respective Genetics centres. S.P. and D.F.E. contributed samples from the EMBRACE study. A.R., S.S., D.H., A.E., D.P. and C.T. undertook mutational screening and data management. C.T. performed statistical analyses with advice from D.F.E and P.D., C.T. and N.R. drafted the manuscript with substantial input from D.F.E., M.R.S., D.G.E., D.E., and P.D., N.R. and C.T. oversaw and managed all aspects of the study.

Conflict of Interest statement. None declared.

FUNDING

This work was supported by Cancer Research UK (C8620/A8372 and C8620/A8857); US Military Acquisition (ACQ) Activity, Era of Hope Award (W81XWH-05-1-0204), and the Institute of Cancer Research (UK). We acknowledge National Health Service funding to the Royal Marsden/Institute of Cancer Research National Institute of Health Research Specialist Biomedical Research Centre for Cancer. C.T. is a Medical Research Council funded Clinical Research Fellow. P.D. is supported by a Wolfson-Royal Society Merit Award. EMBRACE sample collection was supported by grants C1287/A8874 and C1287/A10118 from Cancer Research UK, together with funds raised in memory of Jill Birch. We acknowledge use of DNA from the British 1958 Birth Cohort collection, funded by the Medical Research Council grant G0000934 and the Wellcome Trust grant 068545/Z/02.