• Cancer: breast
• Genetic epidemiology
• Genetics

Recently, Park and colleagues identified germline variants in x-ray repair cross-complementing gene-2 (XRCC2 (MIM 600375; NM_005431.1)) in a small number of breast cancer cases with a positive family history for the disease.1 The overall difference in the prevalence of protein-truncating and potentially deleterious rare missense variants between cases and controls was reported to be statistically significant. XRCC2 is involved in the repair of double-strand breaks via homologous recombination. In addition, a homozygous protein-truncating variant in XRCC2 has been detected in a Fanconi anaemia patient with consanguineous parents.2 Since most known high- and moderate-risk breast cancer genes have a function in DNA damage repair, and some are also Fanconi anaemia genes, it seems plausible that mutations in XRCC2 represent breast cancer susceptibility alleles. In order to evaluate the association between XRCC2 variants and breast cancer risk, we analysed the coding regions of XRCC2 in a cohort of 3548 non-BRCA1/2 familial breast cancer cases and 1435 healthy controls derived from various geographical locations. A more detailed description of the study population and mutations-scanning methods can be found in the online supplementary table S1.

We detected only one protein-truncating variant, a one-base-pair deletion, c.343T[8]>[7], present in a 41-year-old Italian control. Rare non-protein-truncating variants in XRCC2 were detected in 20 familial cases (0.6%) and nine healthy controls (0.6%). Polyphen2,3 SIFT4 and AlignGVGD5 were used to predict the effect of detected missense variants on XRCC2 protein function. Although the number of variants predicted to be damaging or neutral differed between the prediction algorithms (table 1), in all instances these categories were evenly represented among cases and controls. The only common variant in the coding region of XRCC2, c.563G>A (rs3218536), was found to have equal minor allele frequencies in familial cases (0.085) and healthy controls (0.086).

Thus, our data do not confirm an association between XRCC2 variants and breast cancer risk. It is possible that the study by Park et al represents a false-positive finding, or, alternatively, our data are a false-negative finding. The association reported by Park et al was based on six likely pathogenic variants in 1308 cases and zero in 1120 controls, which was significant in Fisher's exact test. However, it should be noted that of these six variants, four were considered possibly or probably damaging, based on in silico prediction. Moreover, the number of positives in cases and controls is extremely small, and the statistic is, therefore, likely to be very unstable. In the NHLBI ESP Exome Variant Server,6 a publicly available database describing variants found in exomes of patients with heart, lung and blood disease and healthy controls, rare variants in the coding region of XRCC2 are reported in 0.5% of all exomes, interestingly including two variants resulting in a gained stop codon. Whereas an allele frequency of 0.5% corresponds quite well with our findings, and with the frequency found by Park et al in their cases, it seems at odds with what they detect in controls. Even when including the innocuous missense change, Park et al found one rare XRCC2 variant in 1120 controls (0.09%). This frequency might have been an underestimation given that the study employed High Resolution Melting Curve analyses for mutation detection rather than direct sequencing. Indeed, when specifically testing two truncating variants in another 1436 controls, they found one of them to be positive, indicating that the frequency of pathogenic variants among controls is greater than what is suggested by their finding of zero out 1120.

Our study had 80% power to detect a relative risk of at least 2.1 (p<0.05) for a variant with an allele frequency of 0.5% (see online supplementary figure S1). Our results could, therefore, represent a false-negative finding if the relative risk associated with XRCC2 variants would be lower than two. We note that variants in a number of other DNA damage repair genes, such as CHEK2 and BRIP1, have been associated with breast cancer with such low risks.7 ,8 Park et al did not provide a quantitative estimate of the risk, but in the two families with an XRCC2 variant for which other family members were also available for DNA analysis, cosegregation of the variant with breast cancer was incomplete. This suggests that if an association between XRCC2 and breast cancer exists at all, it may not be very strong. A much larger sample than that studied by us here would need to be analysed to address this.

Another potential source of controversy is the selection criteria used for constituting the case- and control-series. Park and colleagues analysed 1308 breast cancer cases diagnosed before age 45 years and 1120 healthy controls recruited through population-based sampling by the Australian Breast Cancer Family Registry. In addition, they scanned 689 index cases from multicase breast cancer families and 150 male breast cancer cases. The current study included mostly clinic-based cohorts of cases that were forwarded for BRCA-mutation analysis because the prior probability of detecting a BRCA1 or BRCA2 mutation exceeded 10%.9 Hence, both studies attempted to enrich for ‘genetic’ familial breast cancer cases, but in slightly different ways, perhaps leading to different representations of certain case subgroups. Ethnic backgrounds of the cohorts may also differ proportionally between the studies, although European ancestry was represented in both, and variants were detected in comparable frequencies among individuals of Italian, Spanish, Dutch and US origin.

Neither group studied the effect of missense variants on protein function other than by in silico prediction algorithms. Whereas truncating variants are likely to cause reduction in activity of XRCC2 in homologous recombination, the effect of missense variants (if any) may be subtler. If and how loss of XRCC2 function translates into breast cancer risk will be difficult to assess, but the existence of an effect cannot be excluded at this moment. In mice, complete loss of Xrcc2 is embryonic lethal, and leads to increased genetic instability at the cellular level,10 ,11 a hallmark of many breast cancers.

In summary, our data do not confirm an association between XRCC2 variants and breast cancer risk, although a relative risk smaller than two could not be excluded. Our inability to reproduce the previously reported association might point at a more general obstacle in applying exome sequencing in order to find new genes involved in common complex diseases. Exome sequencing, typically, yields many rare candidate variants. Accordingly, the prior odds that any of these variants are truly associated with disease are small, even when such a variant has been detected in two independent exomes. In any case, our data suggest that variants in XRCC2 are unlikely to explain a substantial proportion of familial breast cancer.