Introduction

From the original publication of the two most widely known breast cancer (BC) susceptibility genes, namely BRCA1 and BRCA2 [1], [2], there has been an active pursuit of new genes contributing to the familial BC phenotype. Additional high penetrance genes have been identified for this disease, such as P53, PTEN or STK11 [3], [4], [5], however their number is not yet as high as originally expected. Fanconi Anemia (FA) pathway genes have also been found implicated in BC as moderate penetrance genes, given their incomplete segregation in affected families. Such is the case of PALB2, BRIP1, RAD51C and XRCC2 [6], [7], [8], [9], which present similar penetrance to other non-FA genes like ATM, CHEK2, RAD51D and RAD51B [10], [11], [12], [13]. Finally, 21 low risk alleles (odds ratio (OR) between 1.1 and 1.3) identified mainly through GWAS studies [14], [15], [16], [17], [18], [19] complete the picture of known BC susceptibility genes to date. However, all of the variants described so far account for less than 35 percent of familial risk of BC, leaving ample room for uncovering additional germline mutations that confer risk of this disease.

In this regard, linkage analysis has been one of the most widespread techniques for the identification of susceptibility genes. Our group presented one such study in 41 BC families, which revealed three genomic regions of interest [20]. This and other linkage studies have found evidence for more than 20 genomic regions in BC, but none have reached conventional levels of evidence to be validated as true BC susceptibility loci, suggesting a high degree of genetic heterogeneity as well as the need for new approaches to isolate single causal genes.

Massively parallel sequencing was demonstrated to be a good strategy for the identification of genes responsible for monogenic diseases [21], [22], [23] or diseases with a high degree of genetic heterogeneity [24]. The latter would apparently be the case for breast and ovarian cancer syndromes, where each of the identified genes belonging to the FA or other DNA repair pathways explain around 1% of the families with the disease. However, a third option in the form of a polygenic model of inheritance needs to be taken into consideration, especially since this scenario would represent a challenge for the identification of novel susceptibility genes using common massive resequencing strategies. Therefore, careful study design based in either specific phenotypic characteristics or differentially expressed tumoral findings needs to be implemented for efficacy in whole exome sequencing analysis.

In the present study we employed massively parallel sequencing technology to investigate 7 families with at least 6 women (between 6 and 10) bearing unilateral and/or bilateral BC, previously tested negative for BRCA1/BRCA2 mutations and younger than 60 y/o. Here, we found several rare variants that could putatively act as BC susceptibility genes but we also highlight the lack of evidence of novel high penetrance genes in this disease.

Materials and Methods

Results

Eleven individuals from 7 families and 7 HapMap cell lines, serving as internal and filtering controls, were analyzed by massively parallel sequencing in this study. Results including number of reads, sample coverage and sequencing depth have been summarized in Table S2.

The bioinformatics analysis and variant filtering pipeline was developed by our group and recently used in the identification of a novel familial pheocromocytoma gene [24]. The main steps are represented in Figure 1. After applying this filtering pipeline, an initial set of 67 (average of 10 per family) SNPs and 14 INDELs (average of 2 per family) passed all the filters (Tables 1 and 2).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Figure 1. Summary of the data analysis pipeline followed in the present study.

Raw sequencing data was screened for common artifacts prior to the alignment step in a first quality control phase (QC1). High quality (QC2) genome matches were analyzed for variants, in the form of departures from a consensus reference genome. Subsequently, variants were filtered by keeping those common to both members in each family and then discarding variants present in HapMap controls and dbSNP130. Further filtering by variant consequence, score and gene function (see material and methods for details) resulted in a list of 67 snps and 14 indel candidate variants.

https://doi.org/10.1371/journal.pone.0055681.g001

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 1. SNP Filtering Summary.

https://doi.org/10.1371/journal.pone.0055681.t001

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 2. INDEL Filtering Summary.

https://doi.org/10.1371/journal.pone.0055681.t002

First, we focused on a group of 25 high interest candidate variants including 12 indels, 3 stopgains, and 6 splicing variants. Additionally, 4 non-synonymous SNPs were included in this group because they either were present in two families (FANCM, CNTROB) or affected a P53-related gene, TP53I13 (Table S3). Validation of the final candidates was done by Sanger sequencing and segregation. Then, we performed segregation studies in other affected members of the same family from whom DNA was available (an average of 2–4 additional members/family except in family 694). Furthermore, those variants segregating in over 50% of the available individuals in each family were tested in a pool of up to 750 Spanish control women.

One of the variants identified in Family RUL153 was a deletion in CHEK2 (c.1100delC), a gene already described as a moderated susceptibility gene [11]. We extended the study to two additional affected members in this family and found that one of them (I-1) didn’t present the same mutation (Figure S1). The non-carrier showed an elevated diagnostic age (76 years old) when compared to the rest of the affected carrier individuals, so a haplotype analysis was performed and helped to confirm this individual as a phenocopy (data not shown).

FANCM was mutated in families 694 and 531 (Figure S1). The former family carried the p.Arg1931 stop mutation (c.5791C>T) annotated as rs144567652 while the latter carried a non-synonymous mutation (c.4392A>T), both of them validated by Sanger sequencing. However, a segregation analysis discarded the variant in family 531 as a plausible BC related allele (Table S3). We then centered our study on the stopgain variant in FANCM, since some genes from the Fanconi pathway have already been described as BC susceptibility genes [31]. Given that segregation studies were not feasible in this family, in order to analyze the possible effect of this variant we performed a case control association study by TaqMan, including 3409 BRCAX cases and 3896 controls from Italy, Netherlands, Australia and Spain (Table S1). We found 10 positive cases and 5 positive controls, with an estimated OR = 2.29 (95%CI = 0.71–8.54), p = 0.13.

Other candidate variants segregating in more that 50% of the individuals in our families such as SOSTDC1 (c.664delA) (family 49, involved in cellular proliferation, differentiation, and programmed cell death); CNTROB (c.2081A>G, rs139292572) (family 3311, a centrosomal BRCA2 interacting protein) and HSP90B1 (c.2362_2364delGAA, rs5800607) (family 3311, codifying for a heat shock protein), were discarded for being present in control population with a MAF >1% (Table S3). On the other hand, SLBP (c.697G>A) (a stem-loop binding protein) and CNTROB (c.1819G>A), both in family RUL036, were not found in the same set of controls. A splice site variant in the oncogenesis-involved WNT8A (c.103G>T) gene was found in family 694. We were unable to confirm experimentally whether the variant altered the splicing, due to a very low expression level of the gene in fibroblasts (data not shown). However, after performing a case-control study by TaqMan in 2043 BRCAx cases and 2186 controls, none of the 4229 samples were found to carry the same variant. Ultimately, four rare variants were selected from the group of 25 high interest variants: FANCM (c.5791C>T, stopgain), SLBP (c.697G>A, splicing), CNTROB (c.1819G>A, missense) and WNT8A (c.103G>T, splicing).

We then focused on the analysis of the remaining 56 missense variants of unknown significance, not previously described in ensemble database and predicted to be deleterious or probably damaging in at least one of the two predictive programs, SIFT [29] and Polyphen [30]. In order to evaluate their role in cancer susceptibility, the variants were genotyped in 2693 BRCAx cases and 2544 controls using an Open Array platform (Applied Biosystems). Technical problems during custom design, as well as discrimination issues in the analysis of specific probes reduced the number of variants finally evaluated to 39. Table S4 includes genotyping-related information on genes, frequencies, OR and p values.

From the previous group of 39 variants, we selected for further validation those variants with an estimated OR higher than two or not present in controls. Thirteen variants fulfilling these criteria were Sanger-validated and analyzed for segregation in their original families (except for variants in family 694) (Table S5). Five variants (in addition to two from family 694) were validated and showed segregation in at least 50% of the available individuals. We consider these seven variants as well as the previous four candidates our potential rare variants candidate to play a role in familial BC development (Table 3).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 3. Final Candidate Variants.

https://doi.org/10.1371/journal.pone.0055681.t003

Discussion

The present study was designed to identify new high susceptibility genes in familial BC. Family selection was based in the highest possible number BC cases per family under the age of 60 (6 to 10 cases through generations), while having no other tumor types segregating in the family. As proposing candidates relying on next generation sequencing results alone is not sufficient evidence, segregation, control population and case-control studies were performed to gain further insight on the relevance of the proposed candidates in BC.

A total number of 81 variants, 67 snps and 14 indels, were identified as potential candidates after applying our filtering protocols. These were subdivided into two groups to facilitate the analysis, a first group containing mainly putative protein truncating variants or variants shared between several families and a second group of rare missense mutations. Among those in the first group, we detected a previously identified moderate susceptibility indel variant in CHEK2 (c.1100delC) in family RUL153 (Figure S1), which served to confirm the adequacy of our filtering strategy for the detection of known BC-related variants.

Additional variants were also analyzed in this first group. FANCM variant c.5791C>T generates a premature stop codon, originating the loss of 118 aminoacids from the c-terminal end of the main transcript and putatively influencing FANCD2 monoubiquitination. A case control study in a cohort of over 7300 samples revealed an OR = 2.29 and a non-significant Fisher’s Exact Test p value (p = 0.13). However, given the low frequency of this variant (0.0011 in cases and 0.00077 in controls), its variable prevalence in the available populations (0.6% in the Italian, 0.3% in the Spanish, but not found in the Netherland samples and found once (0.14%) in the Australian controls, data not shown) and FANCM’s important role in DNA repair, it would be interesting to analyze this gene in a higher number of samples to fully understand its contribution to BC. Also, hints of a possible link with BC were found for BRCA2 interacting CNTROB (c.1819G>A), SLBP (c.697G>A) and WNT8A (c.103G>T). Those presented a high level of familiar segregation in the available individuals (greater than 50%) and absence of the variant in over 700 controls (CNTROB and SLBP) or in over 4000 samples (WNT8A), similarly to what could be expected for rare susceptibility variants (Table S5).

To explore the role of the 56 identified missense mutations, we performed a case-control association study in over 5200 samples (BRCAx cases and controls), a large enough cohort to obtain a preliminary idea of association with BC. Since we discarded variants with a MAF higher than 1% and only rare variants were considered after this point in the analysis, we did not have enough statistical power or we could not calculate it in those variants present in neither cases nor controls. Therefore, we selected variants with a high probability of being linked with cancer by prioritizing variants with OR higher than 2 or absent in controls, a damaging predicted functional impact and segregating in over 50% of the available individuals in each family. Our final candidates include 7 variants in genes related with cell survival, proliferation, angiogenesis, cell cycle progression, cell adhesion and other roles plausibly related to cancer development and progression (Table 3). These genes are candidates to be studied by targeted resequencing in a large series of cases (BRCAx) and controls in order to assess their role in familial cancer susceptibility. In this regard, based on the current evidence, we cannot conclude that these variants are pathogenic and they should not be added to mutation databases as pathogenic variants based solely on the data presented here.

As our families were selected for the high number of affected women across generations and their seemingly dominant model of inheritance, our results were unexpected. However, two recent whole exome studies were also unsuccessful in finding novel high penetrance genes in BC, pointing our result towards a more common scenario in this disease. While selecting their families with less strict criteria, Snape et al [32] sequenced 50 BRCAx probands to find no high susceptibility genes. More recently, Thompson et al [33] found two new genes by whole exome sequencing, FANCC and BLM, which would be mutated in approximately 0.5% of BRCAx families. Both studies also suggested a number of rare variants to be analyzed further in future studies.

A number of limitations in the present study need to be taken into consideration. The number of individuals/families selected for this re-sequencing study is smaller than that of similar studies. While this approach might limit the number of potential candidate variants detected as compared to other less stringent family selection strategies [32], it is a reflection of the small number of non-BRCA1/2 families presenting a high number (from 6 to 10) of BC events per family, which in turn represents a highly desirable starting point for the identification of novel high penetrance genes in any disease. Additionally, the observed phenotype in our families could be related to undetectable alterations using the current approach: mutations in regions outside the captured genomic pool, undetected alterations due to alignment/variant calling errors or those presenting insufficient evidence to pass the filtering criteria. When possible, this and other potential pitfalls in this study, such as rejecting mutations for being in less well known regions (e.g. introns), discarding causal variants for being present in the databases or choosing a phenocopy for the initial re-sequencing in one of the families, were taken into consideration. In this regard, the filtering strategy was conceived to prioritize sensitivity while using filtering principles widely applied in successful massively parallel sequencing studies [24], [32], [34], [35]. The fact that low stringency filtering thresholds paired to the use of alternative analysis to assess the efficacy of most of our filtering steps (see methods) yielded no additional potential candidates further supports our results.

Finally we cannot rule out that some of these families shift a putative explanation towards a polygenic model where moderate and low penetrance alleles would play a predominant role. This hypothesis has been suggested by different authors [36], [37] mainly based in linkage analysis, modeling or targeted resequencing [38] due to the lack of success in finding high susceptibility genes even in regions with statistically significant lod scores. Nowadays, new technologies like massively parallel sequencing in stringently selected families, as in our case, support it from a practical point of view. An example of this is a recent paper by Sawyer et al [39] showing how women affected by familial BC had a highly significant excess of risk alleles compared with controls. Unfortunately, due to the small number of cases in our study we could not analyze this issue.

According to the presented data, families fulfilling the criteria of high risk for BC and including several affected women across generations show evidence of a high genetic heterogeneity and complexity. In this context, some families would be explained based in a polygenic model where rare variants would play an interesting role in BC susceptibility.

Further studies in a larger number of families, using a combined strategy of whole exome/whole genome sequencing, case-control association studies including thousands of samples and targeted resequencing of genomic regions containing rare variants is needed to confirm these preliminary results and to define the genetic architecture of non-BRCA1/2 BC families.

Supporting Information

Acknowledgments

We would like to thank Ms. Alicia Barroso, Mr. Daniel Herrero and the rest of the members of the Human Genotyping Unit (CEGEN) for their collaboration in this project, as well as Roger Milne for his comments and help in the statistical analysis. We also thank Ms. I. Van Leeuwen for supplying patient blood samples and Margaret McCredie for her key role in the establishment and leadership of the ABCFR in Sydney, Australia. We are grateful to Bernard Peissel and Daniela Zaffaroni of Fondazione IRCCS Istituto Nazionale dei Tumori, Monica Barile and Bernardo Bonanni of Istituto Europeo di Oncologia and the personnel of the CGT laboratory at the IFOM-IEO Campus, Milan, Italy, as well as to all the families who donated their time, information and biospecimens for this project.