Identification of candidate cancer predisposing variants by performing whole-exome sequencing on index patients from BRCA1 and BRCA2-negative breast cancer families

Breast cancer is the most common cancer and the leading cause of cancer deaths among women in the world [1]. About 10–20% of all BC patients occur in a familial context, with multiple family members affected across generations [2]. Familial BC susceptibility resulting from deleterious germline variations located on chromosome 17q21 was brought to light through linkage analysis for the first time in 1990 [3]. Since then, many highly penetrant rare variants (with a relative risk of above 10-fold) in BRCA1 (OMIM 113705), BRCA2 (OMIM 600185), TP53 (OMIM 191170), PTEN (OMIM 601728), STK11 (OMIM 602216) and CDH1 (OMIM 192090) to moderately penetrant rare variants (with a relative risk of 2 to 4-fold) in CHEK2 (OMIM 604373), PALB2 (OMIM 610355) [4], BARD1 (OMIM 601593), ATM (OMIM 607585), BRIP1 (OMIM 605882) have been reported. The exact penetrance associated to pathogenic variants in several of these genes is still under investigation. These genes were identified through linkage analysis, positional cloning and/or candidate gene sequencing [5‐17]. Furthermore, with the advent of DNA microarray technology, many low penetrant common variants (with a relative risk often much less than twofold) were unraveled through genome-wide association studies [18]. More recently, thanks to dramatic advances in the speed and scale of next-generation sequencing (NGS) technologies combined with sophisticated computation algorithms and a sharp decrease in sequencing cost, a path for the discovery of additional candidate BC predisposing variants has been opened. To name a few, variants in XRCC2 (OMIM 600375), FANCC (OMIM 613899), BLM (OMIM 604610) and PPM1D (OMIM 605100) have been more recently reported as candidate variants with a BC risk through NGS technologies [19‐21]. The aggregate currently known variants with high, moderate and low penetrance in familial BC susceptibility genes only account for up to 25–50% of all the high-risk BC families. This missing heritability in the remaining 50–75% of BC families [16, 22] reflects both the complexity of the BC genetic architecture and the challenges in identifying remaining BC predisposing variants for delivering timely screening, preventive intervention, and precision treatment.

Several studies have revealed that variants in familial BC susceptibility genes like BRCA1, BRCA2, TP53, PALB2, CDH1, PTEN (OMIM 601728), PIK3CA (OMIM 171834), STK11 (OMIM 602216), RINT1 (OMIM 610089) and NF1 (OMIM 613113) are not only associated with BC predisposition, but also with a number of other malignancies [7, 9, 23‐29]. In the current study, we hypothesized that in BRCA1 and BRCA2-negative families with elevated BC risk, the analysis of a large array of genes previously associated to (hereditary) cancer syndromes or cancer in general, could likely lead to the identification of additional candidate BC predisposing genes/variants. Thus, firstly, we identified all rare variants both exome-wide and cancer-associated gene panel-wide (492 genes) in 54 BC patients from BRCA1 and BRCA2-negative families with elevated BC risk and compared their relative incidence in 120 geographically matched controls. Secondly, all nonsense, frame-shift indels and splice-site variants detected in BC patients within the 492 genes of the panel, which we estimated to possess the highest probability to predispose to BC, were validated on an independent sequencing platform (Roche Junior).

It is expected that exome-wide NGS analysis of a germline DNA sample will reveal many variants when compared to a haploid reference genome, even when only rare variants (MAF ≤ 0.01) are taken into consideration. However, when comparing the total number of variants detected in two individuals of the same ethnicity we do not expect to find significant differences. We confirmed this assumption by (using the same wet bench and dry bench approaches) comparing the average number of variants found in persons belonging to two groups living in the same area (patients recruited in the same hospital): BC patients belonging to elevated risk BC families and controls not selected for personal or familial history of cancer but for cardiac arrhythmias. The ratio of average number of observed (rare) variants in both groups is very close to one for all types of variants (Fig. 1 (red) and Additional file 6) except for splice site and nonsense variants (0.86 and 0.88, respectively), probably because of the relatively small number of splice site and nonsense variants detected per BC patient and control. When focusing exclusively on the genes of the CAGP, similar observations were obtained (Fig. 1 (blue) and Additional file 6) except for the category of nonsense variants, where more than a two-fold excess of nonsense variants was detected in BC patients compared to controls (ratio = 2.39). Although a larger sample size is a minimal requirement to reach statistical significance, our data suggest that the nonsense variants found in excess in the genes of the CAGP among the BC patients (compared to controls) are implicated in the molecular mechanism modulating BC risk(about 50% of these nonsense variants). If the increased number of nonsense variants seen in BC patients is associated with increased cancer risk, one would expect that these nonsense variants will be more frequently identified in genes functionally correlated with the cancer predisposition process. To verify this assumption, the PANTHER over-representation Test (Released 20,171,205) [42] was used with a false discovery rate (FDR) < 0.05. This over-representation test compares a test gene list to a reference gene list and determines whether a particular class of genes (e.g. those associated to a specific biological process) is overrepresented or underrepresented. We found that genes involved in the DNA repair process, namely inter-strand cross-link repair (FDR: 4.94E-02), double-strand break (DSB) repair via nonhomologous end joining (FDR: 4.35E-02), non-recombinational repair (FDR: 4.42E-02), DSB repair (FDR: 8.35E-02) were overrepresented in BC patients while not in the controls. Indeed, four nonsense variants (out of 14) found in BC patients were found in genes involved in the DSB repair process while only one such variant (out of 13) was found among the controls. It remains unclear for us why the same phenomenon is not observed with the frameshift indels. It is possible that false positive indel calls masked a possible enrichment of the true positive frameshift indels.

Our exome wide analysis revealed only seven genes (ASP, C17orf80, FAM111B, GRAMD2, SP100, USP45 and ZNF534) with high impact mutations (PDAVs) in three (or more) BC patients while comparable mutations were not found (or only once) among the control samples. None of these genes was reported to possess cancer predisposing properties and therefore not included in the CAGP. Gene Ontology (GO) annotation [43] for molecular function, biological processes and Reactome Pathways indicated that SP100 and USP45 are involved in DNA repair while ZNF534 is involved in DNA-templated regulation of transcription, making them good candidate cancer predisposing genes. No molecular function or biological process was annotated to C17orf80 and FAM11B, whereas ASPH and GRAMD2 were reported to be involved in calcium homeostasis and transport (Additional file 8).

When restricting our variant analyses performed on BC patients to the 492 genes of the CAGP, we found novel as well as known PDAVs in several genes known or suspected to be BC predisposing (Table 2 and Additional file 10). Genes participating in DNA DSB repair process e.g. PALB2, BARD1, CHEK2 and RAD51C are particularly intriguing as DSB repair process defective tumors can be selectively targeted by PARP (poly (ADP-ribose) polymerase) inhibitors resulting in synthetic lethality [44‐46]. We also found PDAVs in genes linked to DNA repair, FA or occurring in some types of cancers but not well studied in the context of familial BC (Table 2 and Additional file 10). These candidate BC predisposing genes are also interesting to scrutinize further in familial BC setting as it is known that familial BC susceptibility genes can also predispose to multiple cancers [30]. Furthermore, we detected PDAVs in genes associated with other hereditary syndromes but not clearly related to cancer (Table 2 and Additional file 10). These genes are mostly derived from the FaCD panel, which is uncurated. PDAVs detected in the CAGP from control samples but not present in the BC patients are listed in Additional file 11.

Only about 44% of the BC patients were found to harbor a PDAV in one (and exceptionally in two or three) gene(s) of the CAGP in this study. Eleven out of 31 PDAVs detected were not reported in dbSNP147 and therefore considered novel. We should keep in mind that these PDAVs are not necessarily BC predisposing. Therefore, their cancer predisposing attributes should be further investigated in much larger cohort /case-control studies or by performing co-segregation analyses in positive families (if sufficient families are available). Although in this study we mainly focused on candidate PDAVs found in genes of the CAGP (which only accounts for 2% of the full exome), we must remain aware that missense variants in genes of the CAGP but also PDAVs and missense variants outside this gene panel may also predispose to BC. For instance, we identified 7 genes not represented in the CAGP in which PDAVs were over-represented in the BC patient cohort. Moreover, BC predisposition may not necessarily rely solely on the presence of one particular variant in the family but may result from combinatorial interactions between several variants. Indeed, it has been proposed that in the majority of BC families, BC predisposition could be polygenic in nature and the contribution of several variants located in genes associated to moderate or low risk could be responsible for the increased susceptibility to BC [47]. The mechanism how these different variants cooperate at the molecular level to create an increased BC risk is a matter of further investigation [48].

On average, twice more nonsense variants were found in BC patients than in controls when analyzing the genes from the CAGP. Moreover, GO analysis (biological process) of the genes accumulating those nonsense variants indicated that genes involved in the DSB repair process were overrepresented in the BC patients but not in controls. Comparable observations were not made for the other variant types in the CAGP, nor when considering the whole exome. Taken together, our observations might indicate that a nonsense variant found in the CAGP of a BC patient has more than 50% chance to be associated with BC risk while similar conclusions cannot be drawn for “probably/possibly damaging” missense or frameshift mutations. Larger case-control studies should be performed to confirm these assumptions and validate our candidates. This preliminary study in 54 BC patients from BRCA1 and BRCA2-negative BC families with elevated cancer risk identified candidate BC predisposing PDAVs (known as well as unknown) in 30 genes; PALB2, BARD1, CHEK2, RAD51C, FANCA, RINT1, EXO1, RECQL4, CCNH, MUS81, TDP1, DCLRE1A, DCLRE1C, CD96, CYP1A1, DHCR7, DNAH11, ESCO2, FLT4, HPS6, MYH8, NME8, TTC8, ASPH, C17orf80, FAM111B,GRAMD2, ZNF534, SP100 and USP45. The seven last genes of this list were not connected to the cancer process so far. These novel candidate variants and their associated genes should be further investigated with other methods to confirm their role in BC predisposition.