Analysis of hereditary cancer syndromes by using a panel of genes: novel and multiple pathogenic mutations

Individuals who were referred to our center for genetic testing with a hereditary cancer panel between June 2014 and March 2018 were evaluated. All samples were collected from the referring physicians during this study. As this study took place in a private diagnostic laboratory, subjects were not selected by strict inclusion criteria for genetic analysis. All individuals were informed about the significance of molecular testing, provided information about their personal and family history and have signed an informed consent form prior to molecular genetic testing and permission for the anonymous use of their data for research purposes and/or scientific publications. Information on demographics, clinical history, and family history of cancer was collected from test requisition forms, and pedigrees were provided by ordering clinicians at the time of testing.

NGS analysis of hereditary cancer susceptibility genes was performed using two different gene panels. The genes analyzed were selected based on their association to hereditary cancer predisposition. In the majority of cancer syndromes the mode of inheritance is dominant. Thus, a single pathogenic variant in heterozygosity in one of these genes may be the causative reason of cancer predisposition. Several of these genes also have autosomal recessive inheritance, or result in clinically distinct autosomal recessive conditions. BRCA2, BRIP1, PALB2, and RAD51C are associated with Fanconi anemia. ATM and MRE11A are associated with ataxia-telangiectasia and ataxia-telangiectasia-like disorder (ATLD), respectively. MLH1, MSH2, PMS2, and MSH6 are associated with constitutional mismatch repair deficiency (CMMR-D). MUTYH is associated with MUTYH-associated polyposis (MAP). NBN and RAD50 are associated with Nijmegen breakage syndrome and Nijmegen breakage syndrome-like disorder (NBSLD), respectively. The majority of patients who required hereditary cancer testing had a personal or family history of Breast and/or Ovarian cancer and therefore, the vast majority of genes analyzed in this study are associated with increased risk of Breast and/or Ovarian cancer. In addition, the genes were further classified as high, moderate/intermediate or low penetrance genes based on their relative risk for cancer development that they confer to pathogenic variant carriers. High penetrance (or high risk) genes are considered those which when mutated, confer a high Relative Risk of cancer development (greater than 4 times the risk of the general population). Moreover, they are included in guidelines for cancer predisposition testing and specific clinical management recommendations for patients carrying pathogenic variants have been formulated by large working groups [5‐7]. Pathogenic variants in moderate penetrance (or moderate risk) genes confer a 2–4 times risk of cancer development compared to the general population. Low penetrance/risk genes are those related to less than 2 times risk of cancer or those with limited or yet insufficient data available concerning their association and magnitude of cancer risk. Although this categorization is constantly altered in reflection to the accumulated clinical information, based on the latest published data [3, 5‐10], the genes analyzed are summarized in Table 1.

Genomic DNA was extracted from peripheral blood leukocytes using the QIAamp DNA Blood Mini Kit (QIAGEN) or MagCore® Genomic DNA Whole Blood Kit (RBC Bioscience) and was quantified using NanoDrop 2000c Spectrophotometer (Thermo Fisher Scientific).

The analysis of genes involved in hereditary cancer predisposition was performed using two different library reparation approaches. The first 451 individuals were analyzed using an amplicon-based method, while the following 746 individuals were analyzed using a solution-based capture approach.

Irrespective of the library preparation approach, products were subsequently analyzed by Next Generation Sequencing (NGS) using the Illumina platform, MiSeq. Briefly, NGS was performed using the MiSeq Reagent Kit v3 (600-cycle) (Illumina, San Diego, California, United States). Indexed DNA library concentrations were quantified as described above and normalized to 4 nM. The library was denatured using 5 μl of 4 nM library and 5 μl 0.2 N NaOH. The library was diluted using Pre-chilled HT1 buffer at a final concentration 10 pM. Finally, the 10 pM library was spiked in 6% of PhiX Control v3 (Illumina, San Diego, California, United States), which provides a quality control for cluster generation, sequencing, and alignment.

Alignment to the reference sequence (hg19), variant calling and interpretation were performed in the context of clinically relevant transcripts (listed in Table 1) using the optimized algorithms included in the SeqNext module of the commercial SeqPilot suite (JSI medical systems GmbH, Germany). For mapping an enhanced BWA algorithm is utilized by SeqNext. Only basecalls with quality score of 20 or above were considered for further processing. The Regions Of Interest (ROIs) were defined as exons ±50-bp intronic sequence for all genes included in the gene panels. An automatic search for homologous regions in the genome was performed for all ROIs to exclude “background reads”. Reads that match to active homologous sequences were filtered and not aligned to ROIs. Any potential target region or variant position with coverage of fewer than 50 reads was reviewed by analysts and analyzed by capillary sequencing if suspect. Target regions showed an average read coverage of 900x with a minimum depth of >50x for 99% of bases. Variants were called with a variant allele frequency (VAF) cutoff of 20% and each assessed for pathogenicity as described in the Variant classification and Bioinformatics analysis section.

All pathogenic, likely pathogenic variants and VUS were confirmed by Sanger Sequencing by performing a new DNA preparation from an alternative blood vial obtained from the tested individual (primer sequences and conditions available upon request). PCR products’ purification was performed using NucleoFast® 96 PCR Clean-up kit (Macherey-Nagel GmbH and Co., Düren, Germany), according to the manufacturer’s instructions. The sequencing reactions were carried out from 2 μl purified PCR product using the BigDye® Terminator v1.1 Cycle Sequencing kit (Applied Biosystems, Foster City, CA, USA). Sequencing reaction products were purified prior to electrophoresis using the Montage™ SEQ96 Sequencing Reaction kit (EMD Millipore Corp., Billerica, MA, USA). Electrophoresis of sequencing products was conducted on an Applied Biosystems 3130 Genetic Analyzer (Applied Biosystems).

Analysis of Large Genomic Rearrangements (LGR) for genes in which such mutational events have been previously described was carried out. Specifically, the following genes were analyzed in both panels: BRCA1, BRCA2, CHEK2, EPCAM (Exons 8, 9), MLH1, MSH2, MSH6, MUTYH, PALB2, RAD50 (Exons 1, 2, 4, 10, 14, 21, 23 and 25), RAD51C, RAD51D, and TP53.

For this purpose, Multiplex Ligation-dependent Probe Amplification (MLPA) analysis was performed for every sample analyzed by the amplicon-based method using the appropriate MLPA probe mix and according to manufacturer’s instructions: BRCA1: P002; BRCA2: P045; CHEK2: P190; EPCAM, MSH6; P072, MLH1, MSH2; P003; MUTYH: P378; PALB2, RAD50, RAD51C, RAD51D: P260; TP53: P056 (MRC Holland). An Applied Biosystems 3130 Genetic Analyzer was used for electrophoresis and the Coffalyser.Net software was used for the analysis.

In contrast to amplicon-based methods, capture-based approaches provide better uniformity of coverage [12]. Thus the capture-based approach allowed for computational analysis of LGRs from NGS data. For this purpose the CNV module of the software suite SeqPilot (JSI Medical Systems) and panelcn.MOPS [13] were used. Both algorithms are specifically developed for CNV analysis of sequencing data reporting 99–100% sensitivity and up to 100% specificity for the prediction of Large Genomic Rearrangements up to the level of a single gene exon. All LGRs detected with these algorithms were then verified experimentally using the MLPA technique as described above.

Peripheral blood mononuclear cells were separated from the other components of the blood by gradient centrifugation using Ficoll-Paque™ PLUS Media (Fischer Scientific). Total RNA was then extracted using Trizol reagent (Invitrogen, Paisley, UK), following standard protocol provided by the manufacturer. cDNA was synthesized using SuperScript™ VILO™ cDNA Synthesis Kit (Thermo Fisher Scientific) as described by the supplier. Primer sets used for PCR amplification are available upon request.

The annotation and interpretation of all identified variants was performed using an in-house local knowledge-base and a proprietary bioinformatics pipeline designed for the automation of the classification process. The clinical significance of all identified variants was examined using the standards and guidelines for the interpretation of sequence variants recommended by the American College of Medical Genetics and Genomics (ACMG Laboratory Quality Assurance Committee) and the Association for Molecular Pathology (AMP) [14]. Minor Allele Frequencies were examined through access to population databases and in specific to the Genome Aggregation Database (gnomAD) [15], the Exome Aggregation Consortium (ExAC) [15], the 1000 Genomes Project [16], the Kaviar [17], the NHLBI Exome Sequencing Project ESP6500 [18] and the Greater Middle East (GME) Variome Project [19] databases. Disease specific information for variants were retrieved from ClinVar [20], OMIM [21], and the Leiden Open Variation Database (LOVD) [22]. The impact of missense substitutions on protein structure and function was analyzed using the consensus predictive (in silico) algorithm MetaSVM [23] and Align GVGD (Grantham Variation, Grantham Deviation) [24]. The nucleotide conservation of all variants was examined through phyloP [25] and SiPhy [26]. Protein features and domain specific information were retrieved from the UniProt database [27]. The effect of variants on splicing was in silico examined using Human Splicing Finder [28]. All variant information and lines of evidence used for classification were stored, organized and continuously updated and upgraded in an in-house local knowledge-base. This also enabled a reproducible, rigorous and efficient reclassification process.

Pearson correlation analysis with ‘N-1’ correction ([29, 30]) was performed for the correlation between two parameters. Differences in the distribution of continuous variables between categories were analyzed by Mann–Whitney U test. All analyses were performed with R software version 3.4.4 ([31]). All statistical tests were 2-sided, and an adjusted P < 0.05 was considered statistically significant.

During the time period between June 2014 and February 2018, 1197 individuals were referred to our laboratory for genetic testing and specifically 631 individuals from Greece (52.7%), 408 from Romania (34.1%) and 158 from Turkey (13.2%). The median age at testing of individuals in our cohort was 45 years old (range 8 months – 87 years old). The majority of individuals tested were female (94%, 1126/1197) while only 6% (71/1197) were male.

Among the 1197 cases referred for testing, 77.6% (929/1197) had a personal history of cancer, 11.8% (141/1197) were unaffected at the time of testing, while for 10.6% (127/1197) of individuals no clinical data was available. The median time from diagnosis to testing was 1 year with approximately 73% (678/929) of affected individuals tested within 12 months of diagnosis. The majority (82.7%) (768/929) of affected individuals had a personal history of breast cancer. A family history of cancer constituted a major reason for referral, accounting for 84.0% (780/929) and 91.5% (129/141) of affected and unaffected individuals respectively. Among unaffected individuals with a family history of cancer, approximately 87% (112/129) had at least one first−/second-degree relative with a median of 3 relatives with history of any cancer. The subgroup of affected individuals with other cancers included patients with personal history of the following cancer types/sites reported: abdomen, adenocarcinoma, bile, brain, endocrine glands, endometrium, fallopian tubes, gastric, gynecological, Hodgkin’s, kidney, lymphoma, leukemia, liver, melanoma, pancreas, peritoneum, prostate, sarcoma, stomach, thyroid, unknown origin, uterine fibroma.

Individuals tested from Greece had the highest number of affected individuals (81.3%; 513/631) compared to individuals from Turkey (74.7%; 118/158) and Romania (73.0%; 298/408). In addition, among affected individuals from Greece, 87.3% (448/513) reported family history of cancer whereas in Turkey and Romania, 84.7% (100/118) and 78.2% (233/298) of affected individuals reported family history of cancer respectively. These differences underline the different criteria that the ordering physicians used for the selection of patients for genetic testing and the heterogeneous background of these populations. The detailed demographic and clinical features of the individuals are summarized in Table 2.

In the MUTYH gene, five variants c.536A > G p.(Tyr179Cys), c.734G > A p.(Arg245His), c.884C > T p.(Pro295Leu), c.1187G > A p.(Gly396Asp), c.1437_1439delGGA p.(Glu479_Glu480delinsGlu) accounted for all 24 MUTYH pathogenic cases identified in our cohort. In only 3 cases the MUTYH alterations were present in homozygosis or compound heterozygosis. In all three cases the patient had been referred for testing due to a relevant colorectal cancer (CRC) phenotype. One case was a 44-year old male with colorectal cancer and no other history of cancer in the family. The second case was a 43-year old male with colorectal cancer and no family history of cancer. The third case was a female patient diagnosed with small intestine cancer and polyps at ages 65 and 79 with her 75 year old brother having the same clinical features.

Nine of the individuals with a monoallelic pathogenic MUTYH variant, also carried pathogenic variants in another gene, more relevant to the reported phenotype, with the exception of a female individual with a diagnosis of CRC at age 61 in whom the second pathogenic variant was identified in the RET gene, without any thyroid cancer diagnosis reported in the family. Finally, a single MUTYH pathogenic variant was identified in 12 individuals. Two of the individuals were unaffected at the time of testing and had no reported CRC diagnosis in the family, while 7 of the patients were tested because of a breast cancer diagnosis and only one of them had a CRC diagnosis in the family (her father and his two sisters). The remaining 3 individuals in whom a single pathogenic MUTYH variant was identified were diagnosed with CRC at ages 61 (no other CRC in the family) and 51 (one SDR with CRC at 67 years), and FAP phenotype at age 45 with a strong family history of FAP phenotype.

DNA from a total of 1197 individuals was analyzed by one of the two panels described in Table 1. The first 451 individuals were analyzed using the 26 gene panel, while the following 746 individuals were analyzed using the 33 gene panel. Both panels included genes associated with high, intermediate and low cancer risk. At least one clinically significant variant was identified in 264 of samples (22.1%) (Fig. 1a) including individuals with no available information about their personal or family history of cancer. The mutation frequency among the individuals of Greek, Romanian and Turkish ethnicity was 20.4% (129/631), 27.0% (110/408) and 15.8% (25/158), respectively (Table 3). Clinically significant alterations were identified in 27 of the 36 genes analyzed (See Additional file 2: Table S1), with a total of 161 unique pathogenic and likely pathogenic variants being detected among the 264 carriers (See Additional file 3 Table S2, Additional file 4: Table S3, and Fig. 1b). Notably, 28 of the 161 unique variants identified in our cohort had never been reported before in variant databases [20]. Frameshift mutations were the most prevalent mutation type, accounting for the 34.2% (55/161) of the variants identified, followed by nonsense, missense, splicing mutations, large rearrangements and in frame insertion/deletion mutations (26.7% (43/161), 16.8% (27/161), 14.3% (23/161), 6.8% (11/161) and 1.2% (2/161) respectively) (See Additional file 1: Figure S1A).

The most prevalent variant was the frameshift c.5266dupC p.(Gln1756Profs*74) in the BRCA1 gene which was detected in 26 probands in all three nationalities tested (See Additional file 3: Table S2). The second most common BRCA1 variant was the nonsense mutation c.3607C > T p.(Arg1203*), which was absent in the Greek population tested, but was detected in 9 Romanians and 1 individual of Turkish origin. Additionally, the low penetrance missense CHEK2 variant c.470 T > C p.(Ile157Thr) was identified in 16 cases, which is the second most common variant detected. This variant has been shown to increase the risk of breast and colorectal cancer [32‐34].

Of notice is also the relatively high percentage of large genomic rearrangements identified (See Additional file 4: Table S3). Of the 16 LGRs detected, 10 occurred in the BRCA1/2 genes, 2 in MSH2, 2 in CHEK2, while a single LGR was detected in EPCAM, MLH1 and PMS2.

Among the individuals with a pathogenic finding, 43.6% (126/289) of the alterations identified cases, the alteration identified occurred in the BRCA1 or BRCA2 genes, indicating the significant contribution of these two genes in hereditary cancer predisposition. The BRCA1 gene was found to be mutated in 90 individuals (with a mutation frequency of 7.5% (90/1197) among the entire cohort), while BRCA2 was mutated in 36 cases (percentage of 3.0% (36/1197) of the cases tested). This finding was expected given the prevalence of breast and ovarian cancer personal and family history in our cohort. Concerning the mutation distribution among individuals with positive findings, 56.4% of the alterations detected were located in one of the other hereditary cancer related genes (21.6, 19.9, and 15.0% in high, moderate and low risk genes respectively) (Fig. 1c). If molecular analysis was restricted to BRCA1/2 only the hereditary etiology of cancer would have been identified in 10.5% (126/264) of the cases. The analysis of the other high penetrance genes of the panel increases the percentage of alterations detected by 4.3%, while the analysis of the moderate/low penetrance genes leads to the identification of an additional 7.4% pathogenic variants (4.1 and 3.3% for moderate and low penetrance genes, respectively) (Table 3). Apart from the BRCA1/2 genes other highly mutated genes are CHEK2 (2.5%), MUTYH (1.8% monoallelic variants, 0.3% biallelic variants), PALB2 (1.7%), ATM (1.2%) and RAD50 (0.9%).

Among individuals with personal history of breast and/or ovarian cancer we observed a statistically significant difference in the mutation frequencies between high risk and moderate risk genes (16.5% versus 6.2%, p < 0.0001). There was also significant difference between the positive rates in high risk of medium risk genes based on the affection status, especially for breast cancer. For example, we observed a significantly increased mutation rate in high risk genes in individuals with personal history of breast/ovarian cancer compared to unaffected individuals with family history of breast/ovarian cancer (16.5% versus 8.8%, p = 0.0338).

Considering the type of gene altered in relation to the cancer type or the cancer family history of the patient, we observed a high prevalence of BRCA1 and BRCA2 pathogenic variants in both patients with personal and family history of breast cancer (12.6% (97/768) and 8.7% (9/103), respectively). A pathogenic variant in other high penetrance genes was detected in 3.9% (30/768) of breast cancer patients and in 1.0% (1/103) of the patients with breast cancer family history, while the moderate/low penetrance gene mutation rate was 9.9% (76/768) and 4.8% (5/103), respectively. In total, at least one pathogenic variant was detected in 24.7% (190/768) of breast cancer affected individuals and 14.6% (15/103) of those with family history of breast cancer. Among individuals with positive findings and a personal history of breast cancer, the mutation distribution for high, moderate and low risk genes was 62.1, 24.1 and 13.8%, respectively. In addition to BRCA1/2 other highly mutated genes in those patients were PALB2, CHEK2, MUTYH and ATM (See Additional file 1: Figure S2).

Among patients with colorectal cancer, 27.9% (19/68) of the affected patients and 21.9% (7/32) of the individuals with colorectal cancer family history presented a pathogenic variant in at least one of the genes tested. The presence of a pathogenic variant in the MMR genes MLH1, MSH2 and MSH6 was the most common finding in these patients followed by monoallelic alterations in MUTYH and APC alterations. In the limited number of individuals tested because of a personal or family history of ovarian cancer (42 and 30 cases, respectively), a pathogenic/likely pathogenic variant was identified in 19.0% (8/42) and 23.3% (7/30) of cases. In patients with a personal history of cancer other than breast, ovarian and colorectal, a pathogenic/likely pathogenic variant was identified in 21.3% (13/61) of the cases. In this group of individuals the percentage of BRCA1/2 mutated cases was as expected lower (4.9% (3/61)). Finally, in tested individuals with no information about personal/family history of cancer, a clinically significant variant was identified in 14.2% (18/127) of the cases with 9.4% (12/127) in BRCA1/2 genes and 4.7% (6/127) in other high, moderate and low risk genes (Table 3).

Splicing variants are considered to be a common cause of cancer susceptibility. A total of 23 splicing variants were identified at 5’or 3’of the exon and one variant occurred at the last exonic nucleotide. Of those 19 were already described in international bibliography or in variation databases and were classified as pathogenic. For the remaining 5 cases, mRNA was requested for better classification of the splicing variant. In three cases mRNA was not available, while for two individuals, carrying splicing variants in PALB2 and BMPR1A, mRNA analysis was carried out.

The first variant was PALB2 c.49-1G > A, which was identified in a 54 year-old female, who was affected by breast cancer at the age of 49, with a family history of diverse cancers. This alteration was a replacement of the last nucleotide base of intron 1 of the PALB2 gene. Since this particular location is strictly conserved in human and other genomes, it was expected that incorrect mRNA splicing occurred with subsequent production of a truncated and non-functional protein. This alteration had not been described in variant databases. We therefore undertook mRNA analysis in order to better classify it by determining its impact at the RNA level. This analysis indicated that the variant leads to the elimination of two amino acid residues p.(Leu17_p.Lys18) (Fig. 2), which are known to be in a functionally relevant region of the protein [35‐37]. Specifically, a conserved coiled-coil motif is present at the N-terminus of the PALB2 protein (aa 9–42) and mediates the PALB2-BRCA1 protein-protein interaction through a similar motif in the BRCA1 protein (aa 1393–1424) [35‐37]. This coiled-coil domain has also been reported to mediate PALB2 dimerization or oligomerization [38, 39] suggesting a possible competition between the PALB2-PALB2 self-interaction and the PALB2-BRCA1 complex formation. Using site directed mutagenesis several residues within the coiled-coil motif have been shown as important for the hetero-oligomeric interaction between PALB2 and BRCA1 as well as the PALB2 dimerization or oligomerization leading to reduced HR activity [40]. Interestingly the p.Lys18Ala variant of PALB2, although not affecting the BRCA1-PALB2 complex formation, exhibited reduced PALB2 HR activity, suggesting that the variant may affect the integrity of the coiled-coil motif and that even a modest distortion of the structure could result in reduced HR activity, even if the binding of BRCA1 is not affected [40].

The second case was a 38 year-old breast cancer female who presented the c.1166G > T, p.(Ser389Ile) in the BMPR1A gene. This alteration was located in the last nucleotide of exon 10, a strictly conserved region in human and other genomes. Algorithms developed to predict the effect of single nucleotide changes on mRNA processing, predicted that this change may alter splicing of the resultant mRNA but this prediction had not been confirmed experimentally, thus an RNA analysis of this area was undertaken. The RT-PCR analysis did not reveal any change in the length of the RNA produced (Fig. 3). However, Sanger sequencing analysis revealed the absence of the altered nucleotide c.1166 T at the RNA level, with only the wild type allele c.1166G detected. The most probable explanation for this finding is that the altered c.1166 T allele produces an incorrectly spliced and unstable transcript. Even if this finding is an indication of the pathogenicity of this variant, additional analysis at the protein level is required in order to classify it as pathogenic.

Two cancer predisposition causing variants were identified in 25 individuals (Additional file 5: Table S4). In the majority of the cases (15 individuals), a personal history of breast cancer was reported. In 2 cases both pathogenic variants were located in high penetrance genes, while in the remaining cases at least one of the genes mutated belonged to the moderate or low penetrance group. Furthermore, in two cases the individuals carried two pathogenic variants in the same gene (MUTYH and CHEK2 respectively). The most common alteration identified in this group was the low penetrance CHEK2 variant p.(Ile157Thr), which was detected in 4 cases.

The overall VUS rate among the 1197 individuals in our cohort was 34.8% (417/1197) with a maximum of 4 VUS per individual (Table 4). In particular, 72.4% (302/417) of individuals had 1 VUS detected whereas in 23.5% (98/417), 3.1% (13/417) and 1.0% (4/417) of individuals, 2, 3 or 4 VUS were identified, respectively (See Additional file 1: Figure S4A). There was a statistically significant difference (p = 0.0122) in the VUS rate between individuals from Southeastern Europe (Greece and Romania) and Western Asia (Turkey) with 33.5% (348/1039) of individuals from the Balkan Peninsula receiving a VUS as a result compared to 43.7% (69/158) of individuals tested from Turkey. Similar VUS rates were observed among affected and unaffected individuals and their subcategories (Table 4). The highest VUS rate was observed among unaffected individuals with family history of Breast cancer (41.7% (43/103)) and among individuals with personal history of cancer other than Breast, Ovarian or Colorectal (41.0% (25/61)) cancer. The lowest percentages of VUS were observed in individuals with personal history of Ovarian cancer (31.0% (13/42)) and among unaffected individuals with family history of colorectal cancer (31.3% (10/32)). At least one VUS was identified in all genes tested by the two versions of the hereditary cancer panel (See Additional file 6: Table S5), with most variants detected in the ATM gene where 59 unique variants were found in 88 individuals (7.3%). Approximately, 43.0% of VUS were detected in high-risk genes whereas 31.9 and 25.1% were detected in moderate- and low-risk genes respectively (See Additional file 1: Figure S4B and S4C). The vast majority of the 424 unique variants of uncertain significance detected were missense mutations (See Additional file 1: Figure S1B) resulting in conservative (54.5% (231/424)), non-conservative (22.4% (95/424)) and radical (23.1% (98/424)) amino acid substitutions predicted at the protein level (See Additional file 6: Table S5).

In the group of individuals with a personal history of breast cancer, 34.9% (268/768) received at least one VUS in their report with 2.6% (20/768) having a VUS in the BRCA1/2 genes, 15.2% (117/768) in other high-risk genes for breast cancer (CDH1, PTEN, STK11, TP53, and PALB2), 10.7% (82/768) in moderate-risk genes for breast cancer (ATM, CHEK2 and NBN) and 6.4% (49/768) in low-risk genes or with limited information. The gene with the highest VUS rate among the entire cohort was ATM (7.4% (88/1197)), followed by RAD50 (2.8% (34/1197)), CHEK2 (2.4% (29/1197)) and APC (2.0% (24/1197)). Among high-risk genes for breast cancer BRCA2 (1.9% (23/1197)) and PALB2 (1.7% (20/1197)) had the highest VUS rates. ATM, APC and BRCA2 have the largest total coding region and could therefore be susceptible to more variation as observed before [41].

Approximately 19 variants (4.5% of VUS reported) have been reclassified since they were reported (affecting 7.7% (32/417) of individuals with a VUS). Variant reclassification in our cohort resulted in a 2.5% decrease of the VUS rate (See Additional file 1: Figure S5). All revised reports included a variant being downgraded from uncertain significance to likely benign classification mostly because of increasing observations in the tested population and/or new information from variant databases or large datasets.