Introduction

Hereditary breast and ovarian cancer syndrome (HBOC) is caused by a growing list of genes conferring different predisposition risks. Germline heterozygous loss-of-function variants in BRCA1 and BRCA2 (BRCA1/2) genes confer high risk of breast and/or ovarian cancer (BC/OC), yet pathogenic variants in other genes such as TP53, PALB2, CDH1, STK11 and PTEN are also associated with a high risk of BC (Easton et al. 2015; Tung et al. 2016). ATM and CHEK2 are considered moderate BC risk genes with a relative risk ranging from 2.0 to 4.3 (Easton et al. 2015; Tung et al. 2016). In contrast, pathogenic variants in Lynch syndrome genes (MLH1, MSH2, MSH6, PMS2), together with those in RAD51C, RAD51D and BRIP1, contribute to a moderate increased risk of OC (Nakonechny and Gilks 2016). Finally, published genome-wide association studies provide evidence for approximately a 100 of common variants with low penetrance, conferring breast cancer risks below 1.5 times the risk in the general population (Easton et al. 2015; Fachal and Dunning 2015). Yet the contribution of variants in high, moderate and common low penetrance account for only 40–50% of the familial relative risk (Fachal and Dunning 2015). Therefore, it is very likely that there are other genes associated with HBOC.

Compared to the US (Susswein et al. 2016; Buys et al. 2017; Couch et al. 2017; Kurian et al. 2017), smaller European HBOC cohorts have been tested using massively parallel sequencing to estimate the prevalence of pathogenic variants in high- and moderate-risk genes (Castéra et al. 2014; Schroeder et al. 2015; Lhota et al. 2016; Eliade et al. 2017; Feliubadaló et al. 2017; Kraus et al. 2017; Tavera-Tapia et al. 2017; Tedaldi et al. 2017). Our aim was to identify deleterious variants in high and moderate cancer penetrance genes and describe their clinical actionability, as well as, to determine the genetic profile of potentially associated genes (the so-called candidate genes), in a cohort of HBOC BRCA1/2 negative Spanish families.

Patients and methods

Patients

The study included a total of 192 unrelated index cases from breast cancer high-risk Spanish families ascertained through the unit of familial cancer of Vall d’Hebron Hospital from Barcelona: 77 (40%) had BC at a young age (< 36 years), 60 (31%) had BC and belonged to a family with two or more relatives with BC or OC, 38 probands (20%) had OC (8 of them also had BC or endometrial cancer), seven (4%) had BC after 36 with one first/second degree relative with BC, OC or pancreatic cancer, six (3%) had two BC (bilateral or ipsilateral) regardless of BC family history and four (2%) had colon, endometrium, sarcoma or stomach cancer and a BC/OC family history. All index cases were previously screened for single nucleotide variants and large rearrangements in BRCA1 and BRCA2 genes and no pathogenic variant was identified. All were afterwards enrolled for panel testing between January 2013 and December 2015 and received genetic counselling and signed informed consent for the research study, approved by the Clinical Research Ethics Committee of the Hospital Vall d’Hebron. Clinical data including personal and family cancer histories, tumour histology and receptor status of breast tumours were collected through medical chart review. Confirmation of cancer among first- and/or second-degree relatives was obtained whenever possible.

Massively parallel sequencing and variant classification

Ninety-seven genes were included in our research panel. Thirty-four genes were well-known high/moderate-risk cancer genes (16 related to BC/OC and 18 associated to other cancers), and 63 were candidate genes with only initial evidence of cancer risk association and/or related to DNA repair (Supplementary Table 1 lists all the sequenced genes, also providing reference to publications that point to the potential link between each candidate gene and the familiar breast and ovarian cancer predisposition). The protocols and methodology for DNA extraction, capture design, library preparation, sequencing, data alignment, variant calling and variant prioritization are described in detail in Supplementary Methods and summarized in Fig. 1.

Fig. 1
figure 1

Sequencing platform, bioinformatics and variant prioritization pipeline

The frameshift, nonsense and exonic or intronic variants with a RNA analysis data that indicated a protein impact, were classified as pathogenic for the risk-associated genes but as deleterious for candidate genes. For a detailed RNA evaluation protocol, see qualitative and quantitative cDNA analysis section in Supplementary Methods. Missense variants with well-known reported clinical effect were also classified as pathogenic for the risk-associated genes. Variants were classified as variants of uncertain significance (VUS) if no functional data were available or the risk was not clearly established according literature or gene databases. These VUS were further clustered according to the in silico predictions of splicing or protein effect in VUS prioritized or no prioritized (details of the in silico analysis are in the Supplementary Methods). Finally, exonic or intronic variants in risk-associated genes were considered as benign, when they have been revised as such by an expert panel in ClinVar or by literature revision.

Clinical actionability

We defined clinical actionability as a change in patient’s or family medical management according to current guidelines and uptake of cascade testing in relatives at risk (Balmaña et al. 2013; Llort et al. 2015; Paluch-Shimon et al. 2016). The clinical actionability was restricted to carriers of pathogenic variants in genes with well-established cancer risk. All these data were collected from the electronic medical records over the first year after testing.

Results

We used a multigene panel test of 34 cancer risk-associated genes and 63 candidate genes for HBOC syndrome. To determine the efficiency and accuracy of our sequencing platform, bioinformatics and variant classification pipeline, we analysed 13 samples with previously identified pathogenic variants in BRCA1, BRCA2, and CHEK2 (Supplementary Table 2). All the 12 del/dup/indel variants and one single nucleotide variant (SNV) were correctly detected. This validated pipeline was then applied to a cohort of 192 patients with HBOC and no BRCA1/2 pathogenic variants. The mean depth of coverage was of 141 reads and more than 81% of the targeted bases within the regions of interest were covered ≥ 30x.

A detailed overview of prioritized and classified variants found in the 192 analysed patients is shown in Fig. 2. We identified 1157 unique heterozygous exonic and intronic (in the 50 nt flanking splice sites) variants with a MAF ≤ 0.5%: 27 truncating, 17 splicing, 417 missense, one in-frame, 231 synonymous and 464 intronic. Nineteen and 18 unique loss-of-function variants were identified for the cancer risk-associated and candidate genes, respectively. Based on the database and literature revision, 25 variants in known cancer risk genes were classified as benign (Supplementary Data and Supplementary Table 3). The remaining 1095 variants were considered VUS: 104 in silico prioritized (predicted potentially pathogenic or deleterious) and 991 non-prioritized as deleterious (Fig. 2). The different genetic landscape obtained for each of the 192 patients, excluding non-prioritized and benign variants, is displayed in Supplementary Fig. 1.

Fig. 2
figure 2

Summary of prioritized and classified variants found in the 192 analysed patients. MAF minor allele frequency, NFE Non-Finnish European, ExAc Exome Aggregation Consortium, 1000G 1000 Genomes Project Data

Cancer risk-associated genes

Pathogenic variants

Twenty-four patients had one pathogenic variant in the following genes: MUTYH (7), PALB2 (4), ATM (3), APC (2), RAD51D (2), TP53 (2), BRIP1 (1), CHEK2 (1), PMS2 (1) or PTEN (1). The respective familial phenotype is detailed in Table 1 and complete in silico predictions as well as other relevant annotations are in Supplementary Table 4. The highest number of pathogenic variants, excluding heterozygous MUTYH variants, was in PALB2 (four variants, three of them novel) and ATM (three variants, one not previously described). These were all identified in families with BC and no OC cases. One out of PALB2 variants (c.3201+5G>T) alters the splicing process through an imbalanced expression of natural RNA isoforms (Table 1). Results obtained from RNA analysis confirmed a splicing alteration consisting of an imbalanced expression of several PALB2 alternative RNA isoforms (Duran-Lozano et al. 2018). The variant up-regulates isoforms Δ11,12 (in-frame) and Δ11 (frameshift), and down-regulates isoform Δ12 (frameshift). All transcripts are predicted to encode for non-functional proteins. Isoform Δ11,12 presumably contributes to variant pathogenicity by encoding a PALB2 protein lacking 79aa of the WD40 domain that mediates direct interactions between PALB2 and key proteins involved in homologous recombination. Semi-quantitative and quantitative analysis of PALB2 full-length transcript indicated haploinsufficiency in carriers (Duran-Lozano et al. 2018). One stop gain variant in PTEN was found in a patient with BC diagnosed at 46, a suggestive Cowden syndrome and a family history of BC/OC. Two TP53 pathogenic variants (c.587G>C, p.Arg196Pro, and c.783-1G>A) were found in probands with early onset BC (before age 30) and the absence of Li–Fraumeni family history. The missense variant is predicted to be deleterious by three bioinformatics in silico tools (Supplementary Table 4) and it is placed at the DNA-binding domain of TP53 (https://www.ncbi.nlm.nih.gov/Structure/cdd/wrpsb.cgi?seqinput=NP_000537.3). The effect of the splicing variant in TP53 was confirmed by RNA analysis, showing the retention of intron 7 (r.782+1_783-1ins) and deletion of the first 24 nucleotides of exon 8 (r.783_806del) as aberrant transcripts, which would encode a truncated and in frame proteins, respectively (Table 1; Fig. 3a). Semi-quantitative QIAxcel CE data showed > 0.5 reduction of FL transcript levels in the carrier compared to controls, suggesting that the variant allele is unable to produce FL transcript (Fig. 4b). Splicing fraction (SF) estimation showed that the isoform retaining intron 7 was the most expressed (54.4%), whereas the isoform lacking the first 24 nucleotides of exon 8 was present in a 14.5% (Fig. 4c). Both TP53 variants (c.587G>C, p.Arg196Pro, and c.783-1G>A) are listed in the IARC TP53 database (http://p53.iarc.fr/TP53GeneVariations.aspx, R18) (Bouaoun et al. 2016) reported in Li–Fraumeni and Li–Fraumeni like families, and in ClinVar as pathogenic or likely pathogenic.

Table 1 Pathogenic variants in genes associated with cancer risk
Fig. 3
figure 3

Results of in vitro mRNA analysis of five variants leading to aberrant splicing. Sequencing results are shown for the patient carrying the variant and a negative control. A red X illustrates the position of the variant. Nucleotide and exon numbering for cDNA is based on NCBI entries NM_000546.4 (TP53), NM_000535.5 (PMS2), NM_020937.3 (FANCM), NM_000123.3 (ERCC5) and NM_024596.3 (MCPH1)

Fig. 4
figure 4

Qualitative and semi-quantitative cDNA analyses of the RNA effect caused by the five variants leading to aberrant splicing. a Capillary electrophoresis in a QIAxcel instrument of RT-PCR products from one carrier and controls shows different isoforms for all of the genes, except for FANCM, where the difference between the aberrant transcript and the full-length is only 1 bp and QIAxcel has not enough resolution to differentiate. For the sake of simplicity, only one out of the eight controls analysed is shown. PMS2 c.989-2 A>G variant analysis showed a common band in carrier and controls. After Sanger sequencing of whole amplified sample, we could not find out the identity of this presumed transcript (data not shown). Consequently, we cannot confirm if it is an alternative RNA form of PMS2 or an unspecific transcript. b Full-length (FL) relative amounts measured in the carrier (red dots) and in eight healthy controls (blue dots). Healthy controls mean is indicated in dotted line. Error bars indicate standard error of the mean (SEM). c Bar graphs showing the mean relative abundance (splicing fraction) of each transcript detected, indicated by different colours

Regarding OC-associated genes, two RAD51D variants were identified in two OC patients (c.94_95delGT, c.694C>T), and one pathogenic variant (c.1702_1703delAA) in BRIP1 was identified in a young onset (32y) BC patient without OC family history.

The PMS2 variant c.989-2A>G was identified in a proband with personal and family history of BC who did not meet Amsterdam or Bethesda criteria. The splicing effect, confirmed by RNA analysis (Table 1; Fig. 3b), results in an in-frame exon 10 skipping (r.989_1144del), with the loss of part of the PMS2 dimerisation domain. Semi-quantitative analysis showed a 0.5 reduction of the FL transcript in the carrier (Fig. 4b) compared to controls. Splicing fraction estimation showed that the ∆10 isoform was higher expressed (58%) than the FL (41.8%) in the carrier allele (Fig. 4c). The same splicing alteration was also obtained by Borràs et al. (Borràs et al. 2013) and it appears as likely pathogenic in InSIGHT (The International Society for Gastrointestinal Hereditary Tumours) database. The same patient carries a frameshift variant (c.580_581del) in BARD1 (Supplementary Fig. 1), a BC candidate gene whose protein interacts with BRCA1. The proband’s mother, diagnosed with bilateral BC, is an obligate carrier of both BARD1 and PMS2 variants (data not shown).

Three monoallelic pathogenic variants in MUTYH were found in seven patients, the novel c.1101dup, and the recurrent c.536A>G (p.Tyr179Cys) and c.1187G>A (p.Gly396Asp), also known as p.Tyr165Cys and p.Gly382Asp (Lipton and Tomlinson 2004), respectively. The expected carrier rate for MUTYH monoallelic pathogenic variants in healthy controls of 1.5–2% (Nielsen et al. 2011) is lower than that observed in our series of 192 HBOC patients (3.6%).

The APC moderate risk variant for colorectal cancer in people of Ashkenazi Jewish decent c.3920T>A (p.Ile1307Lys) (Liang et al. 2013) was found in two BC patients. This variant is present in 0.3% of the Spanish population (CIBERER Spanish Variant Server, http://csvs.babelomics.org/) and has recently been associated with BC (Leshno et al. 2016). The CHEK2 variant c.470T>C (p.Ile157Thr), a founder variant in Northern European populations and considered a low penetrance variant for BC (Han et al. 2013), was found in one BC family originally from East Germany.

In the 34 risk-associated cancer genes, from a total of 427 unique variants we categorized 383 as VUS (Fig. 2). After an in silico analysis, literature and database revision, only 8% (35/427) were prioritized as deleterious (Table 2, Supplementary Table 5). The genetic characteristics, familial phenotype, as well as published data for these prioritized VUS are described in Supplementary Data. The remaining 82% in silico non-prioritized VUS (348/427) may be considered as either simply VUS or likely benign variants (Supplementary Data and Supplementary Table 6). The number of VUS prioritized and non-prioritized for each known-predisposition cancer gene is shown in Fig. 5, where the genes are ordered according the quantity of nucleotides sequenced and analysed.

Table 2 Variants of unknown significance prioritized by in silico analysis and the database or literature revision, in risk-associated cancer genes
Fig. 5
figure 5

Distribution of unique variants according to sequenced and analysed nucleotides of genes associated with cancer risk other than BRCA1/2. Genes are ordered for analysed region size showing pathogenic variants, VUS “in silico” prioritized and VUS “in silico” no prioritized

Actionability of pathogenic variants

Pathogenic variants in ATM, BRIP1, PALB2, PMS2, PTEN, RAD51D, TP53 and APC genes (16 patients) conferring a well-established high or moderate cancer risk were considered actionable for carriers. The findings prompted increased surveillance or prevention options in 75% (12/16) of these probands. In addition, genetic testing results prompted cascade testing in 53/71 (77%) relatives at risk in 12 families during first-year post disclosure (Table 3). Two deleterious variants in the ATM gene identified in probands with BC and strong family history prompted breast cancer surveillance with annual magnetic resonance imaging (MRI) in probands and relatives, while non-carriers were recommended yearly mammograms based on their family history. In families with BRIP1 and RAD51D pathogenic variants, risk-reducing salpingo-oophorectomy at age 50 was discussed as an option to reduce their ovarian cancer risk. The two patients with early BC and a TP53 germline mutation were recommended surveillance for other TP53-associated cancers (Villani et al. 2016). In addition, in one of the families, the genetic information was used for reproductive decision-making. For PMS2, we offered eight predictive tests and both the proband and her relatives who carried the pathogenic variant started Lynch syndrome surveillance with annual colonoscopy, and gynaecological screening in females. The patient with the PTEN deleterious variant underwent a prophylactic mastectomy and hysterectomy to maximally reduce her cancer risk. PALB2 carriers were offered screening with annual mammogram and MRI, and one of them underwent bilateral mastectomy. APC moderate penetrance carriers were recommended colonoscopies every 5 years from age 40. Yet, colorectal cancer (CRC) risk-associated with monoallelic MUTYH carriers is still under debate, recommendations for MUTYH monoallelic variant carriers would also include colonoscopies every 5 years from age 40. The CHEK2 low penetrance variant did not have an impact in clinical management, since cancer screening was based mainly on family history (Table 3).

Table 3 Overview of surveillance and/or prevention options carried out in the patients with high and moderate pathogenic variants

Candidate genes

Deleterious variants

Eighteen loss-of-function variants were detected in candidate BC/OC genes in 17 patients (1 BARD1, 1 ERCC3, 1 ERCC5, 2 FANCE, 1 FANCI, 2 FANCL, 1 FANCM, 1 MCPH1, 1 PPM1D, 2 RBBP8, 3 RECQL4 and 1 with SLX4 and XRCC2), 3 of which also carry pathogenic variants in known cancer genes (Table 4, Supplementary Table 7). The analysis of the FANCI variant (c.989_991del) in a second degree relative of the proband with bilateral BC at 43 years was negative (data not shown). Two of the three RECQL4 truncating variants were identified in OC patients. One OC patient with the RECQL4 variant c.2272C>T (p.Arg758*) also carried a VUS in BRIP1 (c.3695_3698del), and the other patient with a RECQL4 variant (c.2636dup) carried a VUS in POLE (c.1309G>A, p.Val437Met) (Table 4, Supplementary Fig. 1). Two RBBP8 truncating variants were found in two early onset BC patients without family history of BC. Interestingly, six different deleterious variants were found in the Fanconi Anemia pathway genes (FANCE, FANCL, FANCI and FANCM) and two in the nucleotide excision repair family genes (ERCC3 and ERCC5). Two were located at canonical splice sites (FANCM c.2161-1G>A and ERCC5 c.2678+1G>A) and their impact on splicing was confirmed by cDNA analysis (Table 4; Fig. 3c, d). The effect of the splicing variant in FANCM (c.2161-1 G>A) showed a deletion of the first nucleotide of exon 13, as aberrant transcript (r.2161del) (Fig. 3c). This variation creates a new reading frame ending in a STOP codon (p.Ala721Leufs*39). Capillary electrophoresis using the QIAxcel instrument showed no difference between carrier and controls due to the limited resolution of the method (Fig. 4a). However, carrier FL transcript levels were lower than in controls, suggesting that the aberrant transcript is degraded by the nonsense-mediated decay pathway (Fig. 4b). RNA analysis of the c.2678+1 G>A splicing variant in ERCC5 showed three transcripts (FL, ∆12 and ∆11,12), all of them confirmed by Sanger sequencing (Figs. 3d, 4a). Both ∆12 and ∆11,12 transcripts are predicted to introduce a premature termination codon (p.Gly845Glufs*13 and p.Glu774Argfs*5, respectively). Semi-quantitative QIAxcel data showed  > 0.5 reduction of FL transcript levels in the carrier compared to controls, suggesting that the variant allele does not produce FL transcript (Fig. 4b). Splicing fraction estimation showed that isoform ∆12 was present in a 27% and the ∆11,12 isoform in a 18.9% (Fig. 4c).

Table 4 Deleterious variants in the candidate genes for predisposing familial BC and OC

We also detected a loss-of-function variant in the p53-inducible protein phosphatase gene PPM1D in an OC patient diagnosed at the age of 67 (after chemotherapy) with a family history of prostate cancer (Table 4). Somatic mosaic deleterious variants in PPM1D in peripheral blood mononuclear cells have been associated with BC/OC susceptibility (Ruark et al. 2013). Moreover, missense variants predicted deleterious without mosaicism were also found in prostate cancer families (Cardoso et al. 2016). However, last published data suggest that PPM1D somatic mosaic variants in OC cases are primarily caused by treatment (Pharoah et al. 2016). Our patient did not show any evidence of mosaicism in a Sanger sequence (Supplementary Fig. 2) although the alternate allele frequency was 31% in the massive sequencing output. Further studies, such as cosegregation in relatives and genotyping in other tissues (buccal cells or tumour), are needed to verify the mosaicism of this variant.

We found a deleterious splicing variant (c.322-1G>C) in MCPH1 in a patient diagnosed with BC at 48. This gene encodes a DNA damage response protein and has been associated with BC susceptibility and an increased spontaneous genomic instability (chromosomal rearrangements) in peripheral blood lymphocytes (Mantere et al. 2016). Its splicing effect was confirmed by cDNA analysis, showing the presence of two transcripts corresponding to the FL and the ∆5 (r.322_436del) (Table 4; Fig. 3e). Full-length transcript levels were reduced by 0.5 in the carrier compared to controls (Fig. 4b) and the SF estimation revealed that the ∆5 isoform was expressed as the main transcript (65%) by the carrier allele (Fig. 4c). However, the variant did not cosegregate in the proband’s sister (affected with BC), and genomic instability analysed by inducing chromosomal fragility by diepoxybutane (DEB) in the proband’s lymphocytes was not demonstrated (data not shown).

In the 63 candidate genes, from a total of 730 unique variants, we categorized 712 as VUS (Fig. 2). After an in silico analysis, only 10% (69/730) of the VUS were prioritized as potentially damaging (Supplementary Data and Supplementary Table 8). The remaining 88% of the VUS non-prioritized (643/730, Fig. 2) may be considered as VUS or likely benign variants (Supplementary Data and Supplementary Table 9).

Discussion

Overall, 16 BRCA1/2 negative patients (8%) with HBOC were found to carry deleterious variants in high and moderate cancer genes (excluding carriers of MUTYH mono-allelic variants and CHEK2 low-risk variant, Fig. 6). Although our analysis did not include identification of copy number variants, our detection rate of 8% of pathogenic variants in genes other than BRCA1/2 in high-risk breast cancer families is within the range of other published studies with similar inclusion criteria, ranging from 2.9 to 9.3% (Eliade et al. 2017; Slavin et al. 2017). The highest number of pathogenic variants was found in the PALB2 and ATM genes, both previously associated with BC, and reinforces the role of these two genes as essentially BC risk genes (Easton et al. 2015; Tung et al. 2016). Our results add clinical evidence of the benefit of sequencing through panel testing of several BC/OC susceptibility genes compared to the strategy of sequential testing. Our study also demonstrates the utility of multigene panels in patients who previously underwent non-informative BRCA1/2 genetic screening. The panel approach provides a high number of variants of unknown significance that require the use of a prioritization system to select those with the highest probability of being associated with risk. In our study, we used in silico and database information to prioritize 35 variants in known cancer genes that merit additional studies to unequivocally define them as pathogenic. Excluding the two patients who carried concurrently a VUS and a pathogenic variant in high- or moderate-risk cancer genes, 17% of the patients (33/192) harboured a prioritized VUS (Fig. 6).

Fig. 6
figure 6

Patient distribution according to the variant classification. *Not included patients with pathogenic variants. **Not included patients with potentially pathogenic VUS in risk cancer-associated genes

Identification of deleterious variants in ATM, BRIP1, PALB2, PMS2, PTEN, RAD51D, TP53 and APC genes had a clinical impact, resulting in a change in the medical management of the probands and/or cascade testing in relatives. Overall, 12 out of 16 (75%) of the variants identified in high and moderate penetrance genes were clinically actionable. Other studies have reported an actionability of pathogenic variants in new genes between 69 and 91%, depending on the criteria used to define clinical actionability (Eliade et al. 2017; Frey et al. 2017). In this regard, we did not consider actionable the CHEK2 c.470T>C low penetrance variant, and the heterozygous MUTYH variants had not been disclosed at the time of submission. There has been much debate over whether MUTYH heterozygotes also have an increased risk of developing colorectal cancer (CRC) or other types of tumours such as BC (Nielsen et al. 2011; Win et al. 2016). Starting colonoscopy at age 40 every 5 years has been proposed and there are no recommendations for BC screening.

Some of the variants we detected would have not been identified without the multiplex testing. The PMS2 c.989-2A>G variant was found in a family with mainly BC, who did not meet Amsterdam or Bethesda criteria. Recent studies have found pathogenic variants in PMS2 in BC families undergoing panel testing, suggesting that the use of this methodology might expand the PMS2-associated cancer risks (Ten Broeke et al. 2015; Eliade et al. 2017; Espenschied et al. 2017). A patient with early onset BC was found to have the BRIP1 c.1702_1703delAA variant. Large studies have established BRIP1 as being associated with a moderately increased risk of epithelial OC, but association with BC risk is not robust (Ramus et al. 2015; Easton et al. 2016; Couch et al. 2017). However, the BRIP1 c.1702_1703delAA variant had previously been associated with significant risk of both OC and BC in Spanish patients and it was also identified in one individual with lung cancer (LC) out of 2,758 Spanish individuals with other cancer types (Rafnar et al. 2011). Additionally, this variant was also identified in one out of 40 unrelated Spanish CRC patients with strong CRC familial aggregation (Esteban-Jurado et al. 2016). In our study, this variant was identified in the proband’s father diagnosed with LC and one paternal aunt with CRC. Overall, c.1702_1703delAA might be a Spanish founder allele that deserves further research on its association with different cancer types. Neither of the two families with TP53 mutations fulfilled the 2009 Chompret criteria (Tinat et al. 2009). When these criteria were revised in 2015 it was suggested to consider women diagnosed with BC before the age of 31 to be eligible for TP53 testing. Our results reinforce the application of these criteria, feasible through panel testing, especially if it impacts the patient’s medical management. The identification of secondary findings is a challenge for health care professionals in two different ways: one is the difficulty in determining the best screening for a patient with a pathogenic variant in the absence of the classical phenotype (Rana et al. 2018), and the other one is helping the patient understand and adapt to the implications of this findings during the pre-test and post-test genetic counselling. New prospective studies are warranted to update the cancer spectrum and cancer risk of mutation carriers identified in settings that do not resemble the classical phenotype, as well as the psychological impact associated with these findings. In addition, the health care professional needs to discuss the possibility of finding VUS or moderate penetrance variants, which have been associated with increased uncertainty and distress in patients (Esteban et al. 2018). For this reason, we advocate the clinical relevance of genetic counselling in this context.

APC moderate penetrance variants can also be considered as an unexpected finding, as none of the patients had known polyps and were tested because they had breast cancer. However, one of the proband´s parents had five polyps. The finding prompted colon screening recommendation to age 40 in carriers.

Overall, our initial data suggests that panel testing may identify unexpected deleterious variants in genes not associated with the proband’s clinical phenotype, and it may be especially useful when family history is limited or unknown. A larger study is warranted to investigate the prevalence of secondary genetic findings, at least in the most prevalent cancer genes, i.e. BRCA1/2 and the MMR genes (MLH1, MSH2, and MSH6).

Some of the pathogenic or prioritized variants identified in cancer risk genes have been previously reported in BC/OC or colorectal Spanish families suggesting that they may be recurrent in our population (Tables 1, 2): PALB2 c.2257C>T, ATM c.3754_3756delinsCA, ATM c.4388T>G and RAD51D c.94_95delGT in Tavera-Tapia et al. (Tavera-Tapia et al. 2017), RAD51D c.694C>T, p.(Arg232*) in Gutiérrez-Enríquez et al. (Gutiérrez-Enríquez et al. 2014), BRIP1 c.1702_1703delAA in Rafnar et al. and Esteban-Jurado et al. (Rafnar et al. 2011; Esteban-Jurado et al. 2016) and Borràs et al. PMS2 c.989-2A>G (Borràs et al. 2013).

Excluding the patients harbouring pathogenic variants in associated cancer genes, 5% of the probands (Fig. 6) presented loss-of-function variants in our proposed candidate HBOC genes. The highest number of deleterious variants was found in RECQL4 (three variants, Table 4), corresponding to a carrier frequency of 1.5%, higher than the 0.11% estimated in ESP (Exome Sequencing Project) and 1000G databases (Fu et al. 2017). RECQL4 together with BLM, RECQL, RECQL5 and WRN encode helicase proteins (RecQL helicase family), which play a number of important roles in DNA metabolism to preserve genome integrity (Suhasini and Brosh 2013). Additionally, two different RBBP8 deleterious variants were detected. The protein coded by this gene is associated with BRCA1 and is thought to modulate the functions of BRCA1 in transcriptional regulation, DNA repair and/or cell cycle checkpoint control. Moreover, it is suggested that this gene may itself be a tumour suppressor acting in the same pathway as BRCA1 (Chinnadurai 2006). Further case–control studies are required to unequivocally confirm whether deleterious variants in RECQL4 or RBBP8 confer BC or OC susceptibility.

A deleterious variant or VUS prioritized in cancer genes was not identified in 67% of the probands with clinical suspicion of genetic susceptibility to breast or ovarian cancer (Fig. 6). This suggests that there is missing heritability which might be explained by the multiplicative effect of low-risk alleles (Fachal and Dunning 2015) or alternatively by new rare alleles. In addition, three of the probands (2%) harboured two loss-of-function variants in two different genes (Supplementary Fig. 1), raising the possibility of a potentially more complicated underlying susceptibility mechanism in these families.

In conclusion, our study demonstrates the utility of multigene panels in patients who previously underwent non-informative genetic screening of BRCA1/2. Further research including large case–control studies are warranted to validate the clinical spectrum of pathogenic variants in new genes as PALB2, ATM, CHEK2, RAD51C, RAD15D and BRIP1, their cancer risk and the role of the candidate genes in the susceptibility to HBOC.