Background

Inherited pathogenic variants (PVs) in the BRCA1/2 genes are the most common cause of hereditary breast cancer (BC). Associated lifetime risks for BC development were assessed in a variety of studies, recent estimates range from 60 to 75% for female BRCA1 and from 55 to 76% for female BRCA2 germline PV carriers [1–4]. Clinical management of individuals found to be at high risk for BC focuses on risk reduction and early diagnosis of cancer, and includes early age (25–30 years) BC screening via magnetic resonance imaging (MRI) and mammograms, or risk-reducing mastectomy. Preventive medical treatment, e.g., with tamoxifen or denosumab, in unaffected BRCA1/2 PV carriers is under discussion [5]. Accurate age-dependent estimates of BC penetrance in PV carriers are a crucial prerequisite in genetic counseling for making informed, individualized decisions about whether and when the risks associated with considered preventive measures are in relation to the expected BC risk.

BC incidence of BRCA1 PV carriers rises sharply in the 4th decade of life and then remains at a similar, constant level [2]. The majority of tumors in BRCA1 PV carriers are estrogen receptor-negative (ER-), but their proportion decreases with age [6, 7]. In BRCA2 PV carriers, BC incidence increases rapidly until ages of 40 to 50 years, and then remains almost constant [2]. The majority of tumors are ER-positive (ER+), but the proportion of ER- tumors increases with age [6, 7].

BC risks of female BRCA1/2 PV carriers were found to be affected by a variety of intrinsic and extrinsic factors, such as mammographic density [8, 9], family history of BC and ovarian cancer (OC), variant localization [2], birth cohort [10–12], reproductive history (e.g., age at menarche, first birth, and menopause, number of full-term pregnancies, breastfeeding) [13] and further modifiable factors such as body weight, physical activity, and alcohol consumption [14], among others.

Furthermore, numerous studies demonstrated that the effects of BC susceptibility loci, i.e., common single nucleotide polymorphisms (SNPs), which individually contribute only slightly to individual BC risks, but whose effects can be summed up to polygenic risk scores (PRSs) are able to achieve a clinically relevant degree of BC risk discrimination [8, 15–20]. However, it was also consistently shown that PRS risk stratification is reduced in BRCA1/2 PV carriers in comparison to the general population [21–23]. Moreover, a recent case-only study by Coignard et al. [24] comprising more than 60,000 unselected BC cases and 13,000 cases with BRCA1/2 mutations, provided evidence that several SNPs associated with BC risk in the general population, and are therefore considered in PRS calculations, are in fact associated with the BRCA1/2 mutation status, and hence do not have effects on BC risk in BRCA1/2 PV carriers.

In order to construct a highly informative study sample with respect to modifying factors of BC risk in female BRCA1 PV carriers, Sepahi et al. introduced the approach of investigating patients with “extreme phenotypes” due to age at primary BC diagnosis [25]. The authors analyzed a sample of 133 BRCA1-positive patients who were either diagnosed with primary BC at an age younger than 35 years (early diagnosis, ED) or remained cancer-free until the age of 60 years (late/no diagnosis, LD). The study concluded with the assumption that co-occurring truncating variants in further DNA repair genes associate with ED, although no statistical significance was reached to prove this hypothesis. All patients examined in the study of Sepahi et al. met the inclusion criteria of the German Consortium for Hereditary Breast and Ovarian Cancer (GC-HBOC) for germline testing (Additional file 1: Table S1), i.e. were highly selected for BC/OC family history and/or early BC/OC onset.

Here, we revisit the approach of Sepahi et al. to assess the utility of PRSs for the discrimination of extreme phenotypes with respect to age at BC onset in female BRCA1/2 PV carriers. We employed SNP sets for PRS computation as developed and evaluated by Kuchenbaecker et al. [22], which to our knowledge are the only published PRS SNP sets to date specifically adapted for BC risk stratification in women with a BRCA1/2 PV, namely an overall BC PRS comprising 88 loci, a PRS specific for ER+ BC comprising 87 SNPs and a PRS specific for ER- BC comprising 53 SNPs. In contrast to Sepahi et al., our sample is not only composed of patients selected by the GC-HBOC inclusion criteria for germline testing, but also by BRCA1/2 PV carriers recruited at the Suzanne Levy-Gertner Oncogenetics Unit at the Sheba Medical Center (Tel Aviv University, Israel). Due to a less restrictive access to an initial screening for Israeli BRCA1/2 founder mutations [26], these individuals are expected to be unselected for family history and/or early BC/OC onset.

Methods

Study sample

As a PV, we defined class 4/5 protein-truncating variants with respect to 5-tier variant classification system suggested by the International Agency for Research on Cancer (IARC). Variant classification was performed according to the guidelines published by the Evidence-based Network for the Interpretation of Germline Mutant Alleles (ENIGMA) [27] and the American College of Medical Genetics and Genomics (ACMG) [28]. Only female BRCA1/2 PV carriers with a primary BC diagnosis before the age of 35 years (ED group) or cancer-free survival until the age of 60 years (LD group) were considered. No individual in the ED group was diagnosed with OC prior to BC, and no individual in the LD group underwent prophylactic mastectomy before the age of 60 years.

All patients in the study sample were identified through diagnostic germline testing and recruited between 1996 and 2017 by 13 centers of the GC-HBOC and at Suzanne Levy-Gertner Oncogenetics Unit at the Sheba Medical Center (Tel Aviv University). Individuals recruited by the GC-HBOC were selected by the corresponding inclusion criteria for germline testing (Additional file 1: Table S1), whereas the BRCA1/2 PV carriers recruited by the Sheba Medical Center met the broader Israeli inclusion criteria for BRCA1/2 screening [26], and are therefore considered unselected. All patients were tested for PVs in BRCA1/2 and the CHEK2:c.1100delC variant. Individuals with PVs both in BRCA1 and BRCA2 (n=1) and individuals with co-occurring variant CHEK2:c.1100delC (n=1) were excluded.

Regarding the classification of variant localizations within BRCA1/2, we applied – in concordance with Sepahi et al. [25] – the definitions of Rebbeck et al. [29], i.e., BRCA1 variants intersecting with regions c.179–505, c.4328–4945 or c.5261–5563 (with respect to transcript NM_007294.4) were considered to be localized within a breast cancer cluster region (BCCR) and variants intersecting with regions c.1380–4062 were considered to be localized within an ovarian cancer cluster region (OCCR). For BRCA2, PVs intersecting with c.1–596, c.772–1806 or c.7394–8904 (with respect to transcript NM_000059.4) were evaluated as being localized within a BCCR, and PVs intersecting with c.3249–5681 or c.6645-7471 as being localized within an OCCR.

SNP genotyping

For SNP genotyping, we used a customized 48.48 amplicon-based target enrichment panel (Access Array®, Fluidigm, San Francisco, CA, USA). Variants which could not be covered due to technical limitations, were replaced by adjacent SNPs in linkage disequilibrium (Additional file 2: Table S2). Subsequent parallel next-generation sequencing (NGS) of the barcoded amplicons of the samples was performed by employing an Illumina NextSeq500 sequencing device (Illumina, San Diego, CA, USA) as described previously [17]. All DNA samples were centrally analyzed at the Center for Familial Breast and Ovarian Cancer, University Hospital Cologne, Germany. Raw BCL files were demultiplexed using bcl2fastq2 Conversion Software v2.19 available at https://support.illumina.com Sequence reads were mapped to the human reference genome assembly GRCh37, including decoy sequences (hs37d5), using BWA-MEM of Burrows-Wheeler Aligner (BWA) v0.7.15 [30], and target-specific primer sequences were removed using BAMClipper v1.1 [31]. SNP calling was performed on a merged BAM file including all samples using FreeBayes v1.0.0 [32] under specification of common SNPs from dbSNP build 151 [33] as variant input, and under specification of -q 20, i.e., a minimum phred-scaled base quality of 20. Variant calls were filtered for a minimum read depth of 10 and converted to TPED format using PLINK v1.9 [34].

Quality control & PRS computation

Quality controls were performed as previously described [17]. SNP calls and samples were filtered for a minimum call rate of 0.95. One sample was identified as an outlier due to heterogeneity using the Bonferroni-corrected upper and lower 5 percentiles of the empricially estimated normal distribution and excluded. Samples with putative African or putative Asian ancestry were identified by multi-dimensional scaling based on a combined set of 342 overlapping SNPs in the sample and 1000 Genomes data. Four individuals were excluded because African or Asian ancestry could not be ruled out (Additional file 3: Figure S1).

For each person i, an indvidual PRS was derived via

$$\text{PRS}_{i} = \sum\limits_{j=1}^{N} \beta_{j} g_{ij}\ \text{with}\ g \in \{0,1,2\}, $$

where β is the per-allele log odds ratio and gij is the number of effect alleles in person i for locus j. Missing genotypes were imputed to the average observed genotype in the sample. Alleles with complementary alleles, i.e., either C/G or A/T, were excluded from PRS calculation due to ambiguity. This criterion also applied to SNP rs11571833, which is located within BRCA2. Quality filters resulted in effective SNP set sizes of N=77 for the overall and ER+ BC PRS, and of N=50 for the ER- BC PRS (Additional file 4: Table S3). Of the total 86 SNPs employed for PRS calculations, four showed statistically significant differences in observed allele frequencies when comparing samples recruited by GC-HBOC and by Sheba Medical Center (Bonferroni-corrected Fisher’s exact test p<0.05, Additional file 4: Table S3). PRSs for each SNP set were standardized to have mean 0 and variance 1.

Statistical analysis

Analyses were conducted using the GenABEL v1.8 utilities [35], R v3.6 and PLINK v1.9 [34]. All statistical tests were two-sided with p values below 0.05 considered statistically significant. The association of standardized PRSs with ED, respectively LD, of primary BC was assessed by employing a binomial logistic regression model (outcome 1: ED, outcome 0: LD) under adjustment for selection by the GC-HBOC inclusion criteria for germline testing and localization of BRCA1/2 PVs within the corresponding BCCRs and OCCRs.

A robust sandwich variance estimation approach [36], clustering observations based on family ID, was applied to account for related individuals in the sample.

Results

Quality controls resulted in a curated sample set comprising 295 BRCA1/2 PV carriers out of 293 families as input for the PRS computation (183 BRCA1; 112 BRCA2, Additional file 5: Table S4)). Of the total 295 BRCA1/2 PV carriers, 162 individuals developed BC before the age of 35 years and therefore belonged to the ED group. From the 133 individuals in the LD group (cancer-free survival until the age of 60 years), 41 were known to be diagnosed with BC after the age of 60 years. Key characteristics of the study sample are given in Table 1.

Table 1 Key characteristics of the study sample. BC: breast cancer; ER: estrogen receptor; OC: ovarian cancer; PV: pathogenic variant; (not) sICGT: (not) selected by the inclusion criteria for germline testing of the German Consortium for Hereditary Breast and Ovarian Cancer. *Diagnosis after age 60
Full size table

Standardized overall BC and ER+ BC PRSs showed a strong correlation at the individual level (BRCA1 PV carriers: Spearman’s correlation coefficient ρ=0.98 (95% confidence interval (CI): 0.97–0.98); BRCA2 PV carriers: ρ=0.99 (95% CI: 0.98–0.99), Fig. 1). Correlations between the individual standardized overall BC and ER- BC PRSs were considerably lower in comparison, but also statistically significant (BRCA1 PV carriers: ρ=0.54 (95% CI: 0.43–0.63); BRCA2 PV carriers: ρ=0.46 (95% CI: 0.30–0.60), Fig. 1).

Fig. 1
figure 1

Standardized polygenic risk score (PRS) values per individual and Spearman’s rank correlation coefficients ρ. (Upper left) Overall breast cancer (BC) PRS versus estrogen receptor-positive (ER+) BC PRS in 183 BRCA1 pathogenic variant (PV) carriers who were either diagnosed with BC before the age of 35 years (early diagnosis, ED) or were unaffected at least until the age of 60 years (late/no diagnosis, LD). (Upper right) Overall BC PRS versus estrogen receptor-negative (ER-) BC PRS in the same sample. (Lower left) Overall BC PRS versus ER+ BC PRS in 112 BRCA2 PV carriers who belonged to either the LD or ED group. (Lower right) Overall BC PRS versus ER- BC PRS in the same sample

Full size image

For both BRCA1 and BRCA2 PV carriers, and consistently across all PRSs under consideration, the average PRS was increased in the ED group in comparison to the average PRS in the LD group (Fig. 2). However, these differences reached statistical significance exclusively considering the overall or ER+ BC PRS of the BRCA1 PV carriers (two-sided Welch’s t-test p≤0.04), and the ER- BC PRS of the BRCA2 PV carriers in our study sample (two-sided Welch’s t-test p<0.001).

Fig. 2
figure 2

Boxplots and two-sided Welch’s t-test p values of standardized polygenic risk scores (overall breast cancer (BC), estrogen receptor (ER)-positive BC (ER+), ER-negative BC (ER-) PRS) of 183 BRCA1 and 112 BRCA2 pathogenic variant carriers, who were either diagnosed with BC before the age of 35 years (early diagnosis, ED) or were unaffected at least until the age of 60 years (late/no diagnosis, LD)

Full size image

Binomial logistic regression analyses were employed to assess the association of the standardized PRSs with ED of BC taking selection by the GC-HBOC inclusion criteria and localization of BRCA1/2 PVs, i.e., localization within the BCCRs or OCCRs, into account.

For BRCA1 PV carriers, among the three different PRS SNP sets employed, the standardized overall BC PRS displayed the strongest association with ED (odds ratio (OR) = 1.62; 95% CI: 1.16–2.31, p=0.007, Table 2) in our analysis. Additionally, a statistically significant association between selection by the GC-HBOC inclusion criteria for germline testing and ED with ORs ranging from 2.84 to 3.52 (p≤0.005) was observed. Localization of PVs within the BRCA1 OCCR was statistically significantly associated with LD with ORs ranging from 0.32 to 0.35 (p≤0.02), whereas no statistically significant effect of localization of PVs within the BRCA1 BCCRs was observed.

Table 2 Binomial logistic regression analysis results. Odds ratios (ORs), 95% confidence intervals (CIs) and p values based on two-tailed z-test for binomial logisitic regression analyses with extremely young age (<35 years) at breast cancer (BC) diagnosis (early diagnosis, ED) as the output considering standardized polygenic risk scores (PRSs), selection by the inclusion criteria for germline testing of the German Consortium for Hereditary Breast and Ovarian Cancer (sICGT) and localization of pathogenic variants (PVs) within the corresponding BC cluster regions (BCCRs) or OC cluster regions (OCCRs). Three PRS SNP sets were employed for PRS computation: overall BC, estrogen-receptor-positive (ER+) BC and ER-negative (ER-) BC PRS [22]. 183 BRCA1 and 112 BRCA2 PV carriers were considered who were either diagnosed with BC before the age of 35 years (ED) or unaffected until the age of 60 years
Full size table

Considering BRCA2 PV carriers, the strongest association with ED was observed for the ER- BC PRS (OR = 2.27; 95% CI: 1.45–3.78, p<0.001), whereas the overall and ER+ BC PRS did not show statistically significant association with ED (Table 2). Regardless of the PRS SNP set employed, no statistically signifcant effect of selection by the GC-HBOC inclusion criteria for germline testing and localization of PVs within the BRCA2 OCCRs or BCCRs was observed.

Discussion

In our study considering female BRCA1/2 PV carriers who either developed BC before the age of 35 years (ED) or remained cancer-free until the age of 60 years (LD), we assessed whether extreme age of BC onset due to the definition of Sepahi et al. [25] can be explained by PRSs, among other known factors. Three subtype-specific PRS SNP sets as proposed by Kuchenbaecker et al. [22] for BC risk discrimination in BRCA1/2 PV carriers were employed, namely an overall BC, an ER+ BC and an ER- BC PRS. All PRS SNP sets investigated consist of a maximum of 88 loci and therefore have the potential to be straightforwardly implemented in routine diagnostic multi-gene panel analyses.

Our findings indicate the highest ability of ED/LD discrimination in BRCA1 PV carriers if applying the overall BC PRS, and a slightly reduced but comparable performance of the ER+ PRS. For BRCA2 PV carriers, the ER- BC PRS showed the highest ability of ED/LD discrimination. The observed associations were statistically significant, allowing us to clearly demonstrate the contribution of PRSs to the development of extreme phenotypes of BRCA1/2 PV carriers with respect to age at primary BC diagnosis.

However, previous studies applying Cox regressions with years of life until first BC diagnosis as the outcome, reported the highest potential for BC risk stratification of the ER- BC PRS for BRCA1 PV carriers and of the overall BC PRS for BRCA2 PV carriers [22, 37]. Kuchenbaecker et al. analyzed data for 15,252 BRCA1 and 8,211 BRCA2 PV carriers from the Consortium of Investigators of Modifiers of BRCA1/2 (CIMBA) to develop and evaluate the PRS SNP sets employed here [22]. It is remarkable, that the authors focused primarily on distinguishing between carriers and noncarriers of a BRCA1/2 PV (which is essentially equivalent to distinguishing between carriers and noncarriers of a pathogenic alteration in a BC risk gene with high penetrance), but not between BRCA1 and BRCA2 PV carriers per se. Separate evaluation, however, concluded that the ER- BC PRS performed best for BRCA1 PV carriers, whereas for BRCA2 PV carriers, the overall BC PRS showed the strongest association. This conclusion was confirmed by Barnes et al. [37], who retrospectively evaluated the overall, ER+ and ER- version of a 313 PRS SNP set adapted to BC risk stratification in women unselected for germline mutation status [16]. This study was also performed on the basis of CIMBA data, and involved almost 19,000 female BRCA1 and more than 12,000 female BRCA2 PV carriers. Besides the different outcomes considered in regression analyses, the deviating conclusions regarding the best-performing PRSs could be due to artificial enrichment with LD cases in our study design. Only 6.1% (928/15,252) of the BRCA1 and 10.1% (833/8,211) of the BRCA2 PV carriers examined by Kuchenbaecker et al. did not receive a primary BC diagnosis until after an age of 60 years or were known to have no disease by that age, and thus, meeting the LD criteria; in the study by Barnes et al. this was true for 6.0% (1143/18,935) and 10.1% (1251/12,339), respectively.

Further, in our sample the proportion of the ER+ subtype among all BC cases of BRCA1 PV carriers in the ED group was 29.4% (20/68) and 55.6% (5/9) in the LD group, with missing data of receptor status for 52 tumors (ED: 48, LD: 4), whereas Mavaddat et al. reported in agreement with previous findings by Foulkes et al. an ER+ proportion of <20% among BC diagnoses before age 40 years, and of approximately 45% and >50% among BC diagnoses between ages 60 and 69 years, and >70 years, respectively [6, 7]. Regarding BRCA2 PV carriers in our sample, the ER+ proportion was 64.9% (24/37) in the ED group, with no data available for 9 tumors, and 59.1% (13/22) in the LD group, with missing data for 6 tumors. In contrast, Mavaddat found a proportion of >80% of ER+ tumors among BC diagnoses between the ages of 30 and 39 years in BRCA2 PV carriers, and of >70%, respectively >60%, for BC diagnoses between 60 and 69 years, respectively >70 years, of age [7]. Thus, the proportion of ER- tumors was lower than expected for BRCA1 PV carriers and higher as expected for BRCA2 PV carriers in our sample, which may also have led to divergent results regarding the comparison of performances between PRS sets compared with previous studies based on CIMBA data.

Our results, based on a sample independent of previous genome-wide association studies (GWASs) for PRS construction, also highlight the urgent need for a careful review of BC PRSs specific to BRCA1/2 PV carriers and their performance. The question is not only whether an ER-specific or overall BC PRS should be used, but also whether the PRS is equally appropriate for risk stratification in young and older women and whether it is even justified to use a common PRS for BRCA1 and BRCA2 PV carriers. Coignard et al. [24] suggested that several SNPs included in PRS computation are associated with BRCA1/2 mutation status rather than with BC risk in general. The authors presented 7 SNPs associated with BRCA1 carrier status and 3 SNPs associated with BRCA2 carrier status, of which 7 (BRCA1-associated: rs13281615, rs704010, rs4784227, rs616488 and rs17530068; BRCA2: rs10759243 and rs6001930) were identical or in linkage disequilibrium to SNPs included in the PRS calculations presented. In addition, Coignard et al. identified two, respectively three, loci that modify BC risks in BRCA1, respectively BRCA2, mutation carriers exclusively, which point towards the need for establishment of divergent PRSs for BRCA1 and BRCA2 PV carriers. However, our sample was too small to contribute to these efforts by investigating the associations of individual SNPs with BC with sufficient statistical power.

The study of Sepahi et al. [25] considering 133 BRCA1 PV carriers fulfilling either the ED or LD criteria, pointed towards an association of germline BRCA1 PVs within the OCCR with LD, although no statistical significance was reached. Our study now shows a statistically significant association based on a sample of 183 BRCA1 PV carriers. Consistent with Sepahi et al., there was no statistically significant effect of localization of PVs within the BRCA1 BCCRs. With regard to the localization of PVs in the BRCA2 gene, no statistically significant associations with ED were found, but this is probably due to the small sample size of only 112 PV carriers.

Besides limited sample size, this study has further limitations, as co-occuring truncating variants in DNA-repair genes were not considered as suggested for discrimination between extreme phenotypes with respect to age at BC onset by Sepahi et al. [25] and possible biases due to patient recruitment in two different countries can not be excluded. Further, adjustment of analyses for year of birth was not feasible, as considered individuals from the ED group were born in 1953 at the earliest, and individuals from the LD group no later than 1957.

Conclusions

Although the optimal PRS SNP set for this purpose remains to be found in future studies with larger sample size, our results show that the PRS is an essential prerequisite for the clinical management of unaffected BRCA1/2 PV carriers, as it helps to identify those women who should be offered prophylactic measures at a comparatively very young age. Furthermore, our results provide further evidence that the localization of pathogenic BRCA1/2 mutations should be considered in BC risk calculations.