Differential Burden of Rare and Common Variants on Tumor Characteristics, Survival, and Mode of Detection in Breast Cancer

Breast cancer is caused by complex interactions between a large number of genetic variants' reproductive history and lifestyle factors (1–3). The inherited component of breast cancer risk is due to a combination of rare variants in genes such as BRCA1 and BRCA2 that confer a high risk of the disease (more than 10-fold increase in risk compared with the general population), and common genetic variants that each confer only a small risk (less than 1.5-fold increase; refs. 1, 4, 5). Concerted effort has cumulated in the identification of 172 single-nucleotide polymorphisms (SNP) through multiple genome-wide association studies (6, 7). In contrast to rare protein-truncating variants (PTV) located in the coding regions of high or moderate penetrance genes, most SNPs are not found within genes, but rather within regions of the genome that regulate the activity of nearby genes (6). The risks conferred by each individual variant are modest; however, because they are common and their effects are likely multiplicative, the combined effect is considerable (1, 8). Michailidou and colleagues estimated that 1% of women with the highest genetic predisposition due to common variants have a risk of breast cancer that is more than three times greater than the population at large (6). Certain rare variants in other breast cancer predisposition genes, such as ATM, CHEK2, and PALB2, are estimated to confer 2- to 5-fold increased risk (1, 4, 9–11).

It is established that genetic traits contribute to both familial and sporadic breast cancer risks, but the impact of rare variants in breast cancer susceptibility loci on tumor characteristics and survival remains, to date, relatively limited (2). Because breast cancer is a heterogeneous disease, it is conceivable that distinct inherited genetic predispositions exist for different tumor subtypes based on pathologic features. The most frequently studied high-risk genes are BRCA1 and BRCA2. It has been observed by others that BRCA1-associated tumors more frequently exhibited less favorable prognostic factors compared with non-BRCA tumors (i.e., high mitotic count, high grade, triple-negative and basal-like subtypes, and hormone receptor negativity; refs. 12–14). BRCA2-associated tumors, on the contrary, are similar to non-BRCA tumors and are more likely to be hormone driven (13–15). Previous studies have shown that women at high risk for breast cancer based on low penetrance breast cancer susceptibility loci, both singly and collectively, were significantly more likely to be diagnosed with less aggressive tumor characteristics, such as ER-positive and low-grade tumors (16–18).

One important subgroup of breast cancer are tumors that develop within the time interval between screening examinations (interval cancers). These cancers are usually associated with more adverse biological features and poorer survival outcomes compared with breast cancers that are diagnosed through routine screening mammography (screen-detected cancers; refs. 18–20). Interval cancers in women with nondense breasts where masking does not influence sensitivity of the mammogram are associated with particularly unfavorable pathologic features (18, 19). We have previously shown that 77 common breast cancer genetic variants, summarized as an aggregate risk score, could discriminate between screen-detected and interval cancers (18). Here, we updated the analyses to include 162 SNPs to improve the discriminatory ability. Furthermore, no study to date has examined the impact of rare variants on mode of detection.

Using a large clinically representative breast cancer population (n = 5,099), we studied the associations with breast cancer tumor characteristics and survival for (i) rare PTVs in 31 known and suspected breast cancer predisposition genes currently included in commercial panels and (ii) combined genetic scores of all common breast cancer variants known to date. Most interval cancers have an aggressive phenotype and are, by definition, diagnosed in women who do not benefit from screening. We therefore studied the influence of rare and common breast cancer variants on mode of detection, taking mammographic density into account.

All women under the age of 80 and diagnosed with breast cancer from 2001 to 2008 in Stockholm, Sweden, were identified through the Stockholm–Gotland Regional Breast Cancer Quality Register (18, 19). Eligible women were invited to participate in the LIBRO1 study in 2009. In all, 5,715 women in the LIBRO1 study gave written informed consent to the retrieval of data from medical records and national registers, answered a detailed questionnaire on background and lifestyle risk factors, and provided a blood specimen for genetic analysis (18, 19). Of these women, 5,125 were successfully genotyped in a large-scale genotyping study on breast cancer risk (21). Of these women, 5,122 had enough DNA remaining for mutation testing using targeted sequencing. The final analytical data set comprised 5,099 samples that passed quality control. This study was approved by the Regional Ethical Review Board in Stockholm, Sweden (Karolinska Institutet, DNR2009/254-31/4), and conducted in accordance with the Declaration of Helsinki.

Information on tumor characteristics and treatment (adjuvant chemotherapy, endocrine therapy, and radiotherapy) was retrieved from the Stockholm–Gotland Regional Breast Cancer Quality Register (see details in refs. 17, 22, and 23). Tumor size was measured in millimeters. Lymph node involvement was dichotomized into positive or negative. Estrogen and progesterone receptor (ER and PR) status was recorded as negative or positive in the registers, determined by radioimmunoassay or IHC. The completeness of the registry data was 98% for tumor size and lymph node status and 80% for ER status. Information on grade (Nottingham histologic grade for invasive cancer and nuclear grade for cancer in situ) was available from 2004, with 93% completeness (17).

Data on molecular markers were retrieved in 2015 to 2016 from medical and pathology records at treating hospitals (previously described in ref. 22). HER2 status was dichotomized (positive/negative) in accordance with the Swedish Society of Pathology's guidelines: negative if protein expression showed 0 or 1+, or was higher with no confirmed gene amplification by FISH, and positive if FISH showed gene amplification (22). Proliferation marker Ki67 was measured according to contemporary guidelines and reported as percent staining (low if <20% and high otherwise; ref. 22). Molecular subtype based on age at diagnosis, ER, PR, HER2, and Ki67 status was assigned using a random forest algorithm (caret R package, v. 6.0.58) described in ref. 22.

Date of death was obtained via a data merger with the Swedish Cause of Death Register, which holds complete records of all deaths in Sweden since 1952 (linkage performed on June 27, 2017; ref. 24). The cause of death code “C50*” was used to identify breast cancer–specific death events. The correlation between last main discharge diagnosis and underlying course of death for breast neoplasm is high; the proportion reported as underlying cause of death was 95.9% (25).

Details on the identification of screen-detected and interval breast cancer for this study have been described previously (18, 19). All Stockholm women ages 50 to 69 have been invited to be screened at 24-month intervals since 1989, whereas women ages 40 to 49 were included from mid-2005 and screened at 18-month intervals. Screen-detected breast cancer was defined as a breast cancer diagnosis made after a positive screen finding but before the next visit or end of a normal screening interval. Interval breast cancer was defined as a breast cancer diagnosis made after a negative screen but before the next visit or end of a normal screening interval. Women diagnosed without a prior screening visit, women diagnosed after a normal screening interval had passed, and women with uncertain mode of detection were excluded from these particular analyses.

As described previously in Holm and colleagues, screen-film mammography images were collected from radiology departments and digitized with an Array 2905HD Laser Film Digitizer (Array Corp; ref. 19). Percent mammographic density was measured in prediagnostic mediolateral oblique view images of the cancer-free breast using a fully automated thresholding method (26). Briefly, the algorithm uses a machine learning approach to mimic percent mammographic measurements made using Cumulus, the gold-standard method for measuring mammographic density. A high correlation between the Cumulus measure and the automated measure has been observed (r = 0.884) in an external test set (26). The women were categorized into two categories of masking risk based on percent mammographic density (low risk <25%; high risk ≥25%).

Of the 5,125 women who were successfully genotyped within the iCOGS project (27), 5,122 had enough DNA remaining for mutation testing by targeted sequencing. Target-enriched sequencing libraries of germline DNA from 5,122 patients with breast cancer were prepared at the Centre for Cancer Genetic Epidemiology (University of Cambridge) as part of a larger effort that included samples from other cohorts. The gene panel included coding sequences and intron/exon boundaries for a total of 31 genes for which there are prior evidence of association with breast cancer risk, including all genes being currently offered on commercial panels for breast cancer. Assay design has been previously described (10). See eMethods and Supplementary Fig. S1 in Data Supplement 1 and Supplementary Tables S1 to S3 in Data Supplement 2 for details on amplicon design, sequencing experiment, and data processing. A total of 5,099 samples with valid variants called formed the final analytical data set.

For BRCA1 and BRCA2, all frameshift and nonsense variants with the exception of the BRCA2 c.9976A>T (BIC: K3326X) and other variants located 3′ thereof (n = 105), as well as all consensus splice acceptor or donor sequence sites (see Borg and colleagues; ref. 28), were defined as PTVs. Public data on pathogenic BRCA1/2 variants (includes frameshift insertion/deletions, nonsense, splice sites, and missense variants conclusively demonstrated to be pathogenic) that have been curated and classified by an international expert panel, the ENIGMA consortium, was also downloaded from http://brcaexchange.org/(access date: February 22, 2017) and used for the identification of BRCA1/2 PTVs. In the remaining 29 non-BRCA1/2 genes, variants predicted to introduce a premature stop codon or to disrupt a transcript's reading frame or to lead to aberrant splicing were considered as candidate PTVs. PTV carriership was coded as a binary variable (carrier/noncarrier), with the assumption that tested PTVs were all associated with the trait in the same direction and with the same magnitude of effect. The maftools package in R (downloaded on October 26, 2017) was used to summarize, analyze, annotate, and visualize variants studied (29).

To evaluate the collective genetic effect of breast cancer loci identified through genome-wide association studies, we constructed three separate polygenic risk score (PRS) for each individual using methodology described in refs. 8 and 18. Briefly, genotypes for more than 11 million SNPs were generated by imputation using the 1000 Genomes Project reference panel for 5,125 women in the LIBRO1 study (described in ref. 27). The three PRS were then computed by summing associated alleles of 162 of 172 previously reported SNPs (10 newly identified SNPs based on different custom SNP array in ref. 6 were not successfully imputed in our data, see list in Supplementary Table S4; Data Supplement 2) weighted by the beta estimates associated with overall, ER-positive and ER-negative breast cancer risk in an independent, large-scale genotyping experiment comprising 61,282 female cases with breast cancer and 45,494 female controls of European ancestry (OncoArray; ref. 6), respectively. LIBRO1 samples were not among those included in the generation of the OncoArray results (i.e., weights were independent). Association between a trait and a PRS implies that a genetic signal is present among the selected markers (30).

Multinomial logistic regression (multinom function in the nnet R package), which allows for more than two categories of the outcome variable, was used to assess the associations between genetic risk load (PTV carriership and PRS; per standard deviation increase) and breast cancer tumor characteristics as well as mode of detection. One of the categories of each tumor characteristic (outcome variable) is designated as the reference category, and each of the other levels is compared with this reference level. Odds ratios and corresponding 95% confidence intervals were estimated. Multivariable Cox proportional hazard regression models (using the R “survival” package) were used to estimate hazard ratios (HR) for 10-year breast cancer–specific survival, with time since diagnosis as the underlying time scale. Time at risk was calculated from the date of study entry (i.e., blood draw and left truncation) until the date of death due to breast cancer, date of death from any cause, or end of follow-up (truncated at 10 years), whichever came first. Kaplan–Meier plots (univariate, log-rank test) were visualized using the ggsurvplot function in the R package "survminer."

Analyses examining tumor characteristics and survival were repeated for subsets of patients based on age at diagnosis (<50 and ≥50 years). All PTV analyses in non-BRCA carriers were also repeated for a subset of known hereditary breast and ovarian cancer (HBOC) genes (ATM, PALB2, CHD1, CHEK2, PTEN, and TP53).

Table 1 describes the characteristics of the 5,099 LIBRO1 patients with breast cancer included in this study. The mean age at diagnosis for this unselected breast cancer population was 58.4 years (range, 25–79 years). Four in 5 of the patients were postmenopausal at breast cancer diagnosis (82.5%). One in 5 patients (18.7%) reported a first-degree family history of breast cancer.

In total, 649 rare PTVs (255 unique variants, including 271 stop gain, 1 stop loss, 282 frameshift deletion, 35 frameshift insertion, 3 missense and 57 splice-site variants) were identified in 597 unique individuals (Fig. 1; Supplementary Fig. S2 in Data Supplement 1; Supplementary Tables S5 and S6 in Data Supplement 2). Of these 597 individuals, 92 had PTVs in either BRCA1 (n = 50) or BRCA2 (n = 42). Mean PRS did not differ significantly between PTV carriers and noncarriers (two-sample t test; Supplementary Fig. S3 in Data Supplement 1).

Carriers of at least one PTV in any of the 31 genes studied (including BRCA1 and BRCA2) were younger and more likely to present with more aggressive tumor characteristics compared with noncarriers (Table 2). After adjusting for age and year of diagnosis, tumors of PTV carriers were more likely to be ER-negative (vs. ER-positive: 1.28; 1.00–1.62; Table 2). Tumors of PTV carriers were also more likely of larger size (mm, ≥20 vs. <20: 1.24; 1.03–1.48), higher grade (poorly vs. well-differentiated: 2.14; 1.51–3.04), more proliferative (Ki67 ≥ 20% vs. Ki67 < 20%: 1.36; 1.03–1.81), and more frequently of luminal B (vs. luminal A: 1.74; 1.13–2.69) and basal-like (vs. luminal A: 3.08; 1.94–4.90) subtypes than tumors of noncarriers (Table 2). After removing 92 BRCA1/2 carriers, carriers of at least one rare variant in any of the 29 other genes studied remained associated with higher grade (poorly vs. well-differentiated: 1.71; 1.17–2.49) and luminal B subtype (vs. luminal A: 1.85; 1.17–2.91) compared with noncarriers (Table 2).

The analyses were repeated for subset of women diagnosed with breast cancer at below age 50 (Supplementary Table S7 in Data Supplement 1) and at age 50 and above (Supplementary Table S8 in Data Supplement 1). Among women diagnosed with breast cancer at a younger age, tumors in PTV carriers (BRCA1/2 inclusive) were likely ER-negative (vs. ER-positive: 1.75; 1.11–2.75), PR-negative (vs. PR-positive: 1.82; 1.18–2.80), of larger size (mm, ≥20 vs. <20: 1.51; 1.02–2.23), of high grade (poorly vs. well-differentiated: 3.45; 1.31–9.06), highly proliferative (Ki67 ≥ 20% vs. Ki67 < 20%: 2.67; 1.37–5.22), and of basal-like subtype (vs. luminal A: 5.24; 2.35–11.66; Supplementary Table S7 in Data Supplement 1). Among younger women, no tumor characteristic remained significantly associated with PTV carriership after the exclusion of BRCA1/2 carriers (Supplementary Table S7 in Data Supplement 1). Among women diagnosed with breast cancer at an older age, tumors in PTV carriers (BRCA1/2 inclusive) were more likely of higher grade (poorly vs. well-differentiated: 1.92; 1.31–2.82) and luminal B (vs. luminal A: 2.03; 1.26–3.26) and basal-like (vs. luminal A: 2.33; 1.29–4.21) subtypes. (Supplementary Table S8 in Data Supplement 1). Grade and luminal B subtype remained significantly associated with PTV carriership after the exclusion of BRCA1/2 carriers (poorly vs. well-differentiated: 1.65; 1.10–2.48 and 2.21; 1.36–3.59; Supplementary Table S8 in Data Supplement 1).

In a subset of 5,007 patients without BRCA1/2 PTVs (all ages), higher PRS weighted by overall breast cancer risk estimates were associated with less aggressive breast cancer tumor characteristics [ER-negative vs. ER-positive: 0.79; 0.72–0.86; PR-negative vs. PR-positive: 0.85; 0.80–0.91; smaller size (mm, ≥20 vs. <20: 0.94; 0.88–1.00), lower grade (P trend = 0.046); HER2-enriched vs. luminal A: 0.79; 0.68–0.91 and basal-like vs. luminal A: 0.73; 0.59–0.91; Table 2]. The odds ratios and corresponding 95% confidence intervals observed for the associations between a collective genetic score computed by summing associated alleles of 162 of 172 previously reported common breast cancer variants (PRS) weighted by ER-positive breast cancer risk estimates and tumor characteristics were not appreciably different from that of the PRS weighted by overall breast cancer risk estimates (Table 2). PRS weighted by ER-negative breast cancer risk estimates were not significantly associated with any of the breast cancer tumor characteristics studied, with the exception of ER status (ER-negative vs. ER-positive: 1.10; 1.01–1.19; Table 2).

In Cox regression models adjusting for year and age of diagnosis (left-truncated for time to blood draw), PTV carriers in any of the 31 genes studied were associated with worse 10-year breast cancer–specific survival than noncarriers (HR; 95% confidence intervals (CI): 1.65; 1.16–2.36; Table 3). The significant association between PTV carriership and worse survival remained in a subset of 5,007 women who were not BRCA1/2 carriers (1.76; 1.21–2.56; log-rank test P = 0.0033, see Kaplan–Meier plot in Fig. 2). The worse survival was more pronounced in a subset of non-BRCA carriers who were diagnosed with breast cancer at below age 50 (2.61; 1.25–5.47; Table 3). No significant survival differences were observed with PRS. Adjusting for tumor characteristics or treatment did not appreciably change the hazard ratios (Table 3).

Table 4 shows associations between breast cancer genetic loci and mode of detection. Compared with noncarriers, PTV carriers were 2-fold more likely to be diagnosed with interval breast cancer than screen-detected breast cancer (1.96; 1.19–3.24) in women with low mammographic density (Table 4). The increased risk of being diagnosed with an interval cancer than a screen-detected cancer persisted after the exclusion of BRCA1/2 carriers (1.89; 1.12–3.21; Table 4). In contrast, higher PRS weighted by overall and ER-positive breast cancer risks were associated with lower risks of developing interval breast cancers compared with screen-detected cancers (0.77; 0.64–0.93 and 0.75; 0.62–0.90, respectively) in women with low mammographic density.

In a subset analysis, the results from analyses based on PTVs in known HBOC genes (ATM, PALB2, CHD1, CHEK2, PTEN, and TP53) were not appreciably different from those of the analyses based on all 31 known and suspected breast cancer genes on the panel (Supplementary Figs. S4–S6 in Data Supplement 1).

In this breast cancer case-only study, we simultaneously examined rare and common breast cancer susceptibility variants and (i) their impact on tumor characteristics, (ii) their influence on breast cancer survival, and (iii) their effect on whether or not women will benefit from routine mammography screening. Our results showed that patients who were carriers of rare PTVs in any of the 31 genes studied, with or without the exclusion of BRCA1/2 variants, were more likely to develop high-grade tumors, have worse survival, and be diagnosed with interval breast cancer when compared with noncarriers. In contrast, patients with higher genetic predisposition to breast cancer based on a collective PRS computed from common variants had less aggressive tumors and were more frequently diagnosed at mammography screening.

A heritable component has been found not just for breast cancer risk but also for breast cancer survival (31, 32). Breast cancer survival is largely dependent on tumor characteristics that determine risk of recurrence and influence choice of adjuvant therapy (33). The inheritance pattern for breast cancers of different tumor characteristics may vary, resulting in differential prognosis and survival. As supported by our data, ER-receptor negativity and basal-like subtype are intrinsic properties of BRCA1-related cancers, particularly among patients diagnosed at a younger age (15, 34). The exclusion of carriers of PTVs in BRCA1/2 completely attenuated the differential association between ER status and PTV carriership. However, one or more germline PTV in any of the non-BRCA1/2 genes remained significantly associated with high-grade (poorly differentiated) and luminal B cancers, suggesting that women who are PTV carriers in non-BRCA1/2 genes may not only be at an increased risk of developing breast cancer, but are also at an increased risk of developing more aggressive breast cancers, which often tend to grow quickly and have a worse outlook.

Carriers of PTVs were more likely to die from breast cancer than noncarriers. Although BRCA1/2-associated cancers are often associated with worse tumor phenotypes, it has been documented that patients with breast cancer who are BRCA1/2 carriers do not necessarily exhibit worse survival patterns than BRCA1/2 noncarriers (35). In agreement, a slightly stronger effect for worse survival was observed in carriers of non-BRCA1/2 PTV carriers after the removal of BRCA1/2 carriers. The association between carriership of non-BRCA1/2 genes with worse survival in spite of grade and luminal B subtypes being the only tumor characteristics that remained differentially associated with carriership is also supported by our earlier work on the inheritance of breast cancer prognosis among sisters, where we showed that breast cancer survival is largely inherited independently of tumor characteristics (31).

Although it is important to know who is at high genetic risk of developing breast cancer, it is also important to know who is at high risk of developing tumors that do not currently benefit from screening mammography. As shown in our study, PTV carriers are at 2-fold higher risk of developing interval breast cancer, which are tumors that routine mammography screening fails to detect. Notably, this effect was observed only when we restricted analyses to “true” interval cancers, that is, interval cancers in women with nondense breasts where masking does not influence sensitivity of the mammogram. Interval cancers have been associated with particularly unfavorable pathologic features (18, 19). These results forward the hypothesis that women who are PTV carriers and subsequently diagnosed with breast cancer are more likely to be diagnosed with fast-growing tumors than patients who are noncarriers. For these women who are at higher-than-average risk of developing breast cancer, and more often than not, more aggressive interval breast cancer, surveillance at a younger age and shorter screening intervals complemented by more sensitive screening methods such as magnetic resonance imaging and ultrasound may be warranted (36).

The opposite result (more favorable tumor characteristics) was seen for women with high PRS. These women are more likely to develop breast cancer, but the mode of detection is likely to be routine screening. A possible explanation for more favorable tumor characteristics and increased probability of detection at screening being associated with higher PRS may stem from the study population base in which the common variants were identified (potential enrichment for screen-detected cancers and breast cancer survivors). In addition, of the 90,969 breast cancers with known ER status in the landmark work by the Breast Cancer Association Consortium and OncoArray Consortium, 69,501 (∼76%) were ER-positive (6). Due to increased statistical power to detect SNPs associated with ER-positive breast cancer, a large proportion of the 172 previously reported SNPs may be more strongly associated with ER-positive breast cancers, which are in turn more often associated with favorable tumor characteristics (37).

In the post-genome era, information on genetic differences in breast cancer risk may have clinical relevance, such as in how women at different risks should be screened (6). Large prospective studies are already exploring risk-based screening and prevention recommendations for breast cancer based on a combination of genetic and other hormonal and lifestyle factors (38, 39). Beyond risk and screening, genetic differences predisposing to specific tumor characteristics have the potential to improve management and treatment (40).

The main strength of our study is that it is a large, clinically representative breast cancer cohort with rich data collected on patient (genetic, lifestyle, and hormonal factors) and tumor characteristics. The mammography screening program in Sweden is mainly government-funded and decentralized (19, 41). Hence, all women essentially have the same access to healthcare and screening services, making our study population suitable for studying the effects of genetic variants on mode of detection. The screening participation rate in Sweden is high at 70% to 80% (41, 42). In addition, the availability of mammographic density data made it possible to study the effects of “true” interval cancers (i.e., tumors not masked by dense breast tissue; ref. 19).

Limitations associated with the LIBRO1 study have been discussed in previous works (18, 19). In addition, the definition of screen-detected or interval breast cancer is dependent on local screening guidelines, where the screening interval can range between 1 and 3 years (43, 44). However, biennial mammography screening for breast cancer is implemented in most of the countries worldwide (42). It should be noted that amplicon sequencing may be subject to contamination (45). CHEK2 1100delC (n = 92), ATM Q2825X (n = 67), and FANCM Q1701X (n = 54) constituted a third of the PTVs identified. Although the carrier frequency of CHEK2 1100delC (1.8%) and FANCM Q1701X (1.1%) is consistent with prior publications in European or Finnish populations (∼2% for CHEK2 1100delC and ∼1% for FANCM Q1701X; ref. 46).

In conclusion, we found that genetic variants in both rare and common susceptibility loci are associated with breast cancer tumor subtypes, survival, and mode of detection. A deeper understanding of how inherited genetic variants affect breast cancer biology could have profound implications for clinical management and future screening practices (47).

S.H. Teo has received honoraria from speakers bureau of AstraZeneca. No potential conflicts of interest were disclosed by the other authors.

Conception and design: J. Li, B. Decker, A. Torstensson, H.N. Christensen, A.M. Dunning, D.F. Easton, K. Czene

Development of methodology: W.X. Wen, B. Decker, K.A. Pooley, D.F. Easton

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): B. Decker, A.M. Dunning, C. Luccarini, D.F. Easton, P. Hall, K. Czene

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): J. Li, E. Ugalde-Morales, W.X. Wen, B. Decker, J. Allen, K.A. Pooley, L. Dorling, D.F. Easton, S.H. Teo, P. Hall, K. Czene

Writing, review, and/or revision of the manuscript: J. Li, E. Ugalde-Morales, W.X. Wen, B. Decker, M. Eriksson, A. Torstensson, H.N. Christensen, A.M. Dunning, J. Allen, K.A. Pooley, J. Simard, D.F. Easton, S.H. Teo, P. Hall, K. Czene

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): J. Allen, K. Czene

Study supervision: D.F. Easton, K. Czene

We thank Don Conroy, Caroline Baynes, Patricia Harrington, Martine Dumont, and Stéphane Dubois for assistance with the sequencing experiments. We also thank Hanis Mariyah Mohd Ishak for careful reading of the manuscript.

This work was supported by AstraZeneca (to K. Czene), the Swedish Research Council (grant no. 2014-2271 to K. Czene), Swedish Cancer Society (grant no. CAN 2016/684 to K. Czene), FORTE (grant no 2016-00081 to K. Czene), ALF Medicine (grant no. 20170088 to K. Czene), and the Cancer Risk Prediction Center (CRisP; www.crispcenter.org), a Linnaeus Centre grant (no. 70867902 to P. Hall and K. Czene) financed by the Swedish Research Council. Targeted sequencing was supported by Cancer Research UK grants C1287/A16563 to D.F. Easton and C8197/A16565 to A.M. Dunning, and the PERSPECTIVE project, funded from the Government of Canada through Genome Canada and the Canadian Institutes of Health Research, the Ministère de l'Économie, de la Science et de l'Innovation du Québec through Genome Québec, and the Quebec Breast Cancer Foundation (to J. Simard). B. Decker was supported by the Intramural Research Program of the National Human Genome Research Institute. J. Li is a recipient of a National Research Foundation Singapore Fellowship (NRF-NRFF2017-02).

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.