Investigating the effects of additional truncating variants in DNA-repair genes on breast cancer risk in BRCA1-positive women

Out of more than 30,000 HBOC index cases registered in the German Consortium for Hereditary Breast and/or Ovarian Cancer (GC-HBOC) biobank, 133 BRCA1-positive patients either with a personal history of breast cancer below the age of 35 years (early AAO onset) or without personal history of breast cancer at the age of 60 years (controls) were selected for this study. Patients who had undergone prophylactic mastectomy or prophylactic oophorectomy before the age of 45 years were excluded from the analysis [30]. Participants had signed a written informed consent and the study was approved by the local ethics committee (ethic vote number 053/2017BO2). Relevant information regarding age at menarche, number of pregnancies, and Oral contraceptive use was collected from the GC-HBOC database.

Reviewing published literature, genes were considered on the basis of a reported breast cancer association. In addition, all DNA-repair pathway genes were selected from KEGG GENES database (http://​www.​genome.​jp/​kegg/​genes.​html, last accessed: 26.11.2013; Additional file 1: Table S1). A target region of 895.2 kbp consisting of 311 genes was sequenced in total. The coding regions and exon-intron boundaries ±25 bps were targeted (using default parameters of Agilent SureDesign, except for Masking = Most Stringent) and enriched using Agilent SureSelect custom RNA probes (Agilent, Santa Clara, CA). Two hundred nanograms of genomic DNA were checked for quality and quantity by Qubit dsDNA Assay (Thermo Fischer Scientific, Waltham, MA, USA) and fragmented using a Covaris system (Covaris, Inc., Woburn, Massachusetts) to generate fragments of 120–150 base pairs length. Quality and fragment size of sheared DNA were checked using a TapeStation (Agilent, Santa Clara, CA). Sequencing libraries were constructed according to the Agilent SureSelectXT protocol. The pre-capture and post-capture libraries were quantified by a TapeStation. Libraries were sequenced either on a Miseq (Illumina, San Diego CA), NextSeq500 (Illumina, San Diego CA) or HiSeq2500 (Illumina, San Diego CA) platform using paired-end reads of 151 bps or 101 bps.

MegSAP, a free-to-use open-source bioinformatics pipeline was used for data analysis (version 0.1–379-gb459ce0, https://​github.​com/​imgag/​megSAP). In brief, adapter and quality trimming was applied using SeqPurge [31]; sequencing reads were mapped to the human genome version GRCh37 with BWA (v. 0.7.15) [32], and ABRA2 [33] (v. 2.05) was used for indel realignment; variant calling was performed by freebayes (v. 1.1.0) [34] and variant annotation was done using snpEff/SnpSift (v. 4.3i) [35]. Quality control was executed on three layers of information including raw reads, mapped reads and variants (Additional file 2: Table S2). We used Alamut batch (v. 1.5.1, Interactive Biosoftware) for splice site annotation.

Variants were automatically classified according to an algorithm based on a modified version of the American College of Medical Genetics and Genomics (ACMG) guidelines for variant classification [36]. According to this algorithm, splice variants at the position +/− 1 and +/− 2 are classified as likely pathogenic if the variant disrupts the function of the gene product unless the population frequency of the variant is not compatible for a pathogenic variant (minor allele frequency of 1% was used as a cutoff). For intronic variants located outside of the canonical splice sites including Cartegni splice sites [37] we referred to Alamut Visual (Interactive Biosoftware) incorporated prediction tools such as MaxEntScan, Splice Site Finder Like, and Human Splicing Finder. Variants were considered as pathogenic or likely pathogenic (collectively termed as pathogenic) if they led to a truncation, initiation loss or canonical splice site effect or if there was a relevant publication in favor of pathogenicity and if there was additional evidence in public database like ClinVar. In case there was no evidence such as functional assessment data available, missense, synonymous and intronic variants were classified as variants of unknown significance (VUS), benign or likely benign according to the Minor Allele Frequency (MAF > 1%) in the 1000 Genomes Project (1KGP), dbSNP, Exome Aggregation Consortium (ExAC) or ESP6500.

Descriptive statistics such as medians, means and standard deviations for continuous data and proportion and 95%-CI for categorical data was used to characterize the study population and sequencing results. A multivariable logistic regression was carried out to control for the potential confounding effect of family history, age at menarche, parity, and use of oral contraceptives. Missing data was imputed using median or mode. The variable additional truncating DNA-repair variants was coded as yes if the patient carried a truncating DNA-repair variant and it was coded as no if the patient was not carrying a truncating DNA-repair variant. The outcome was considered the incidence of cancer. The regression analysis was performed in R 3.5.2. Using GraphPad Prism version 6.07 for Windows (GraphPad Software, La Jolla California USA), we performed Fisher’s exact test to compare the mutational location in each cohort. All p-values were two-tailed and p-values less than 0.05 were considered to be statistically significant. Maftools was applied to visualize BRCA1 pathogenic variants with a modified database [38].

Variants obtained from freebayes in VCF format (see above) were annotated using the eDiVA platform (https://​ediva.​crg.​eu/​) in order to obtain functional annotation (exonic, nonsynonymous, synonymous, splicing etc.), European population allele frequencies from 1KGP, Exome Variant Server (EVS) and ExAC databases, as well as functional impact scores from CADD. Variants not annotated as ‘exonic’ or ‘splicing’, as well as variants within segmental duplication (SegDup identity > = 0.9) were removed from further analysis. We performed sample quality control by screening for outliers in (a) number of variants per sample and (b) transition to transversion ratio per sample. Second, we calculated the first 10 PCA components of all samples using only synonymous SNVs that were not in linkage disequilibrium and had an allele frequency above 0.005 in EVS. Finally, we compared the rare variant load per gene between the early AAO cohort and controls. No outliers were detected in any QC test and early AAO patients and controls were clustering in a single group in the PCA. Following QC, we removed any variant with European AF higher than 0.01 in any of the three databases: EVS, 1KGP, and ExAC. Additionally, we excluded all variants annotated as synonymous or with a CADD score below 10 (considered neutral). Using the remaining rare, likely damaging variants we performed Burden and SKAT-O association tests implemented in the R package SKAT (https://​www.​hsph.​harvard.​edu/​skat/​download/​) version 1.3.0. The Null model for both tests was computed using the SKAT_Null_Model function with output set to dichotomous outcome (out_type = “D”) and no sample adjustment (Adjustment = FALSE). For the SKAT-O test we used the SKATBinary function with default parameters except for method that was set to “optimal.adj” (equivalent to SKAT-O method). Minor allele frequencies (MAF) of variants transformed with Get_Logistic_Weights were used as weights. The burden test was performed using the same function (SKATBinary) and parameters, except for method that was set to “Burden”.

In total, 133 BRCA1 positive women were screened for truncating variants in 311 DNA-repair genes. The cohort with early AAO consisted of 73 women who developed breast cancer at an age younger than 35 years (median age at onset, 27 years; interquartile range (IQR) 25–27 years). The controls consisted of 60 participants, cancer-free by the age of 60 years. Follow-up data showed that some developed breast cancer at an age older than 60 years (n = 25; 41.7%) with a median age at onset of 64 years (IQR, 62–67) or had no history of breast cancer (n = 35; 58.3% median age, 70 years; IQR, 63–75 years). The demographic characteristics of the participants are shown in Table 1.

In total, 117 patients from both cohorts carried a BRCA1 pathogenic single nucleotide variant (SNV), 13 patients carried a large deletion, and three patients carried a large duplication in BRCA1 (Fig. 1). In the early AAO cohort, 15.1% of all participants carried a frameshift founder mutation [39] in exon 20 of the BRCA1 gene (ENST00000357654: c.5266dupC:p.Gln1756fs). The European founder missense variant [40] in exon 4 (ENST00000357654: c.181 T > G: p.Cys61Gly) was the most frequent (10%) pathogenic variant found in the control cohort (Additional file 3: Table S3). All pathogenic variants in BRCA1 were confirmed by NGS.

With respect to family history, the majority of patients in the control cohort had at least one first-degree relative with breast and/or ovarian cancer as compared to the early AAO patients (56.2% versus 98.4%). Women with larger families who reached older ages are expected to have more relatives with breast and/or ovarian cancer on average in comparison to those whose families are smaller and younger. This can explain the difference between family history of early AAO cohort and control cohort (Table 1).

To compare allelic variation in type and location of pathogenic variants across the BRCA1 protein between the early age at onset and the control cohort, we compared the pathogenic variant accumulation in different regions of BRCA1. Whereas no differences were detected for the Breast Cancer Cluster Regions (BCCRs), which are associated with increased risk of breast cancer (Additional file 4: Figure S1a), differences were found for the Ovarian Cancer Cluster Region (OCCR). 22 (45.3%) patients in the control cohort (Fig. 1, Table 1) carried a pathogenic variant within the OCCR compared to 15 (25.9%) of patients in the early AAO cohort, though the statistical significance was not reached (p-value = 0.07). Patients with large deletions or insertions and splice site variants were excluded from this analysis since they either span more than one region or their impact on protein function is not certain, respectively. In the early AAO cohort, 56 patients (76.7%; 95%-CI 65.4 to 85.3%) of BRCA1 mutation carriers carried a truncating variant while 6 patients (8.2%; 95%-CI 3.1 to 13.3%) carried a missense pathogenic variant (ENST00000357654: c.181 T > G: p.Cys61Gly) and 11 patients (15.1%; 95%-CI 7.8 -25.4%) carried a copy number variation (CNV). In contrast, 47 patients (78.3%; 95%-CI.65.8% to 87.9) carried a truncating variant in controls, 8 patients (13.3%; 95%-CI 5.9 to 24.6%) carried a missense pathogenic variant (Additional file 4: Figure S1b) including ENST00000357654: c.181 T > G: p.Cys61Gly, and c.5096G > A: p.Arg1699Gln and 5 patients (8.3%; 95% CI 2.8 to 18.4%) carried a CNV.

We evaluated 311 genes that maintain genome integrity and/or have been associated with HBOC. The mean sequencing depth was 456x ± 197.3 SD. Additional file 2: Table S2 shows the detailed results and quality parameters of sequencing. A total of 3703 variants was identified and of those 43 (1.2%) truncating variants (Additional file 5: Table S4) were detected in 36 DNA-repair genes. The affected genes were mainly Single Strand Break Repair genes (SSBR, 30.6%), Double Strand Break Repair genes (DSBR, 30.6%), and check-point factor genes (11.1%). The remaining truncating variants were identified in genes with other functions such as BRCA1/2 interactors, centrosome formation and signal transduction. In overall, 42 women had at least one additional DNA-repair truncating variant. In the early AAO cohort, 26 out of 73 patients (35.6%; 95%-CI 24.7 - 47.7%) carried at least one additional truncating variant and two cases carried two additional truncating variants in DNA-repair genes (Additional file 6: Figure S2a). Among controls, 16 out of 60 participants carried an additional DNA-repair germline truncating variant (26.7%; 95%-CI 16.1 to 39.7%). In this cohort, three participants carried two germline DNA-repair truncating variants; at least one of them affected a DSBR pathway gene (Additional file 6: Figure S2b).

We investigated the effect of additional DNA-repair truncating variants on the risk of developing breast cancer among BRCA1 mutation carriers, adjusted for age at menarche, oral contraceptive use, parity and family history. Despite the fact that it did not reach the conventionally accepted p-value of 0.05, the odds ratio is in favor of increased breast cancer risk for double heterozygote patients (OR: 3.1; 95% CI 0.92 to 11.5, p-value = 0.07). To confirm the validity of our model, the same analysis was carried out on a subset of subjects who were matched for family history (early AAO cohort; n = 41 and control cohort; n = 59) adjusted for age at menarche, oral contraceptive use and parity (OR: 3.3; 95%-CI 0.92 to 13.3; p-value = 0.07). Consistent results were obtained for this subset of cohorts.

To test the effect of additional truncating variants in specific DNA-repair pathways, we compared the mutational load in DSBR and SSBR genes between the two cohorts. Among the early AAO cohort, 8/73 women (11.0%; 95%-CI 4.9 -20.5%) carried an additional truncating variant in DSBR compared to 5/60 women (8.3%; 95%-CI 2.8 -18.4%) in the control cohort. Regarding the SSBR genes, we found 8/73 women (11.0% %; 95%-CI 4.9 -20.5%) in the early AAO cohort carrying additional SSBR truncating variants as compared to 5/60 women (8.3%; 95%-CI 2. %-20.5) in the control cohort. The mutational load in DSBR and SSBR did not differ between both cohorts (Fig. 2). Further comparison has been carried out between SSBR- and DSBR- mutation carriers with non-carriers (Additional file 7: Figure S3; Additional file 8: Table S5). In none of the cases differences were statistically significant.

Among control cohort, 25 (41.7%) patients developed breast cancer at a median age of 64. For these patients the tumor characteristics were compared with the tumor characteristics of the early AAO patients. The immunohistochemical staining of estrogen and progesterone receptors did not differ significantly with respect to the AAO, though the ER and PR negativity was more frequently found in the early AAO cohort compared to affected control patients (p-value = 0.28 and 0.76 respectively, Table 2). Tumors of the early AAO group tended to show a higher histological grade compared to the tumors of the affected control patients (Table 2) although the difference failed to reach the significant level (p-value = 0.24). Expression of estrogen and progesterone receptors, grading of tumors and histological types of tumors were not significantly different between patients with additional truncating variants in DNA-repair genes and patients without additional DNA-repair truncating variants (Additional file 9: Table S6).

To assess the load of rare missense (VUS + pathogenic variants) variants in DNA-repair genes on the AAO of breast cancer in BRCA1-positive patients we performed a Burden test and a SNP-set (sequence) Kernel Association Test (SKAT-O). To this end, a comprehensive quality control of early AAO cohort and controls was done (see Methods). No differences were observed between early AAO cohort and controls in (a) variants per sample, (b) rare variant load per gene, (c) transition-transversion ratio, and (d) top 10 PCA components. Next, we removed all common variants (MAF > 1% in EVS, 1KGP, or ExAc) as well as all synonymous variants from both early AAO and control cohort. To search for genes conveying an increased risk, we used patients of the early AAO cohort as cases and patients of the late AAO cohort as controls (Additional file 10: Table S7). Although there was no significant gene identified after FDR correction, several genes showed significant un-corrected p-values in at least one of the two RVAS tests, requiring more investigation in independent larger cohorts. These candidate genes include MYBBP1A (early AAO: 13, controls: 3), MRE11 (7:0), TDG (5:0), WRN (7:1), TP53BP1 (10:3) and REV1 (8:2) as well as one potential risk reducing factor, PTCH1 (early AAO: 1, controls: 8).

Interestingly, two cases carrying pathogenic variants in both BRCA genes were found in either cohort. Case 1 was a patient affected with breast cancer at the age of 26 yrs. She had two first-degree relatives with breast cancer. There was no ovarian cancer and no second-degree relative with any type of cancer. She carried a BRCA1 pathogenic variant (ENST00000357654: c.1016dupA) and an additional BRCA2 pathogenic variant (ENST00000544455.1: c.3585_3686delAAAT). Unfortunately, tumor characteristics were not available for this patient. Case 2 was diagnosed with breast cancer at the age of 63.9 years. Her family history was indicative for HBOC: A first-degree relative with breast cancer and three first-degree relatives with ovarian cancer. Also, there was a second-degree relative with breast cancer. She carried a nonsense variant in BRCA1 (ENST00000357654: c.1687C > T) and a nonsense variant in BRCA2 (ENST00000544455.1: c.8875G > T). An additional truncating variant was found in EME2, (ENST00000568449: c.541_544delGCTG) a DSBR gene. The immunohistochemical staining showed a triple negative tumor.