LAUSR

To the Editor — The slow progress of intensive international efforts over the past two decades to identify the remaining genetic causes of familial breast cancer suggests that numerous predisposition genes exist, which are likely to be much rarer than BRCA1 and BRCA2, and to each account for a small proportion of families. Searching for very rare risk alleles in the context of population diversity further increases the challenges of identifying and validating new breast cancer–associated genes, as evidenced by the inability of subsequent studies to validate primary reports of new genes1–3. Here we argue that the previously reported breast cancer–susceptibility gene RECQL is not associated with breast cancer risk in an Australian population. Two loss-of-function (LoF) founder mutations in RECQL have been genotyped in Polish and French-Canadian subjects, and a greater-than-fivefold-increased risk has been reported for c.1667_1667+ 3delAGTA carriers in the Polish women (0.23% in 13,136 unselected breast cancer cases versus 0.04% in 4,702 controls)4,5. Likewise, a 16-fold-increased risk has been reported for c.634C> T carriers in FrenchCanadian subjects with early-onset breast cancer compared with population newborn controls (0.69% in 1,013 cases versus 0.014% in 7,136 controls). At that time, a smaller study of familial breast cancer cases and controls from Han Chinese individuals from Northern China (448 cases and 1,588 controls) supported a role for RECQL in breast cancer predisposition but was underpowered compared with the study reported in Polish and FrenchCanadian populations6. Another study has reported a 0.54% carrier frequency of RECQL LoF variants in 1,110 highrisk females from Hong Kong with breast cancer; however, very few controls were sequenced (88 controls)7. In contrast, a null result has been reported for the c.1667_1667+ 3delAGTA variant (reported to be enriched more than fivefold in Polish individuals with breast cancer) among 2,596 breast cancer cases and 2,132 controls (0.35% versus 0.28%, odds ratio 1.23, P = 0.69) from Central European individuals who were likely to have ancestry similar to that of the Polish population8. Because some groups are now suggesting that RECQL be included in germline breast cancer–predisposition testing panels, it is crucial to be certain that this gene has genuine relevance to breast cancer predisposition, because adverse outcomes could potentially arise if the conclusions of the original publications are incorrect or overestimated. To address this important clinical question, we sequenced all exons and at least 10 bp of the exon–intron flanking regions of RECQL in 9,112 subjects from Australia (Supplementary Note). The case subjects were female index patients diagnosed with breast cancer (> 95% of cases) or ovarian cancer from 4,536 families with breast cancer with a negative result after BRCA1 and BRCA2 mutation testing, and were ascertained from the Variants in Practice Study from the combined Victorian and Tasmanian Familial Cancer Centres, and Pathology North, NSW Health Pathology, Newcastle, Australia. The controls were 4,576 women in the LifePool cohort (see URLs) who were above 40 years of age and were cancer free as of May 2016. All exons and exon–intron boundaries of RECQL were sequenced with high and consistent coverage in cases and controls (average sequencing depths of 189.0 and 195.4, respectively). Overall, 96.7% of the bases among the cases and 98.4% of the bases among the controls were sequenced to a depth greater than tenfold. We identified 13 LoF mutations (defined as nonsense, frameshift or essential splice-site mutations) in the cases and 25 in the controls (0.29% versus 0.55%, odds ratio 0.52, 95% confidence interval 0.25–1.06, P = 0.072 by two-tailed Fisher’s exact test here and subsequently) (Table 1 and Supplementary Tables 1 and 2). One variant, c.1859C> G (p.Ser620Ter), located in the last exon and resulting in a stop codon 29 amino acids upstream of the normal stop codon, accounted for most of the LoF mutations (6 in the cases and 16 in the controls). No carriers of the Quebec founder mutation (c.634C> T) were identified in either the cases or controls, and only two cases and one control were carriers of the Polish founder mutation c.1667_1667+ 3delAGTA. The LoF mutation frequency in our controls was similar to that reported in non-Finnish Europeans in the Exome Aggregation Consortium (ExAC) browser (0.65% in ~27,000 non-Finnish European participants and 0.51% in ~53,000 all participants; ExAC version 0.3, excluding TCGA data, released 13 January 2015)9. The personal and family histories of patients with breast cancer and carrying RECQL LoF mutations are shown in Supplementary Table 3. A marginally higher frequency of rare (minor allele frequency ≤ 0.5%) missense variants was observed in the cases compared with the controls, but the difference was not statistically significant (54 cases, 1.19%, versus 37 controls, 0.81%, P = 0.073) (Supplementary Table 4). In addition, no significant difference was observed between the cases and controls when the variants were filtered for likely pathogenicity with the commonly used in silico tools Condel, PolyPhen-2, SIFT, CADD and REVEL (Supplementary Table 5). The variants c.644G>A (p.Arg215Gln), c.1363C>T (p.Arg455Cys), c.1373T>A (p.Met458Lys) and c.1685C>T (p.Thr562Ile), previously reported by Sun et al.6 to lack helicase activity and thus to potentially represent pathogenic alleles, were not detected in either our case or control cohorts, in agreement with the rarity of these variants reported in nonFinnish Europeans in ExAC. Given the rarity of the LoF alleles in RECQL, genetic heterogeneity among study subjects may distort the relativerisk estimation in case–control studies. Principal-component analysis with 74 ancestry-informative markers showed that the study subjects were predominantly of European ancestry (> 96%), and there was a significant overlap in ancestry distributions between cases and controls (Supplementary Fig. 1), thus strongly suggesting that our findings were not due to major differences in ancestry between our cohorts. It is generally accepted that a predisposition gene is considered actionable only if the 90% confidence limit of the estimated relative risk is greater than four10; thus, RECQL would be excluded, because our study had 80% power to detect an odds ratio of at least 1.89 (P < 0.05) for a variant with a population frequency of 0.65% (RECQL LoF variant frequency in ExAC