Introduction

It has been estimated that all known cancer susceptibility genes account for only 1% to 15% of familial cancers [1, 2]. Approximately 5% to 10% of hereditary breast and ovarian cancers result from dominant mutations in known single genes [3–6], particularly BRCA1/BRCA2. Therefore, the basis for a large fraction of genetic predisposition in families with breast and/or ovarian cancer remains to be uncovered.

Recent studies have highlighted DNA copy number variation (CNV) as the most prevalent type of structural variation in the human genome [7–9], and its role in normal development and disease has been demonstrated through its impact on gene expression and protein structure [10–13]. In particular, CNVs involving deletions have been reported as a cause of cancer susceptibility, occurring in up to 30% of highly penetrant cancer-predisposing genes, including BRCA1, BRCA2, APC, SMAD4 and TP53, as well as mismatch repair genes [14–16] (reviewed in [17, 18]).

Germline gains and losses of large DNA segments have recently been reported as factors predisposing individuals to neuroblastoma, prostate and colorectal cancer and BRCA1-associated ovarian cancer [19–24]. Nevertheless, whole-genome CNV profiling of patients fulfilling criteria for hereditary breast and ovarian cancer, but without BRCA1/BRCA2 mutations, has not been reported. In the present study, we investigated the germline CNV profiles of 68 unrelated familial and early-onset breast cancer patients who were negative for BRCA1/BRCA2 mutations, with the aim of detecting new genes contributing to breast and/or ovarian cancer predisposition.

Materials and methods

Study approval

The research protocol was approved by the ethics committee of the AC Camargo Cancer Hospital, São Paulo, Brazil (protocol 1175/08), and informed consent was obtained from the subjects.

Patients

Samples of peripheral blood cells for DNA extraction were collected after informed consent was obtained from 68 women attending the AC Camargo Cancer Hospital prior to any systemic treatment. They were selected for fulfilling at least one of the criteria for hereditary breast and ovarian cancer published in the National Comprehensive Cancer Network Practice Guidelines in Oncology version 1.2010 [25].

Confirmation of the family history of cancer was obtained whenever possible on the basis of pathology reports, medical records and/or death certificates. All women had previously tested negative for BRCA1/BRCA2 pathogenic mutations (based on Sanger sequencing of coding sequences). Because most of the affected relatives were already dead, were inaccessible or refused to participate, we were unable to investigate CNV segregation in the majority of the cases.

The criteria used to select patients, type of cancer, and age at cancer diagnosis are given in Additional file 1. Most of the patients (n = 48) were familial cases of hereditary breast and/or ovarian cancer in which at least one other family member was affected. The remaining 20 patients were considered hereditary breast and/or ovarian cancer patients for being isolated cases of early-onset cancer (≤ 45 years of age). Most of the tumors were invasive ductal breast carcinomas. Aside from two patients who had only ovarian cancer (patients 9 and 34), all of the other sixty-six patients had breast cancer (bilateral in patient 67, and patients 10 and 16 also had ovarian cancer).

Control sample

DNA samples were obtained from the peripheral blood cells of control participants after their informed consent was obtained. One hundred DNA samples (seventy-eight women and twenty-two men) were provided by the Genetic Center of the Institute of Biosciences, University of São Paulo, São Paulo, Brazil. They were obtained from noncarrier relatives of patients affected by mental impairment with clear genetic etiology unrelated to cancer predisposition (namely, fragile × syndrome or de novo chromosomal rearrangements). No information regarding their cancer history was available. Age-matching of controls and patients was considered unnecessary for this study since CNV frequency in blood is generally considered stable and unrelated to chronological age.

Comparative genomic hybridization based on microarray (array-CGH)

We performed comparative genomic hybridization based on microarray (array-CGH) using a 180 K whole-genome platform (design 22060; Agilent Technologies, Santa Clara, CA, USA), which has an average probe spacing of 18 kb. Briefly, samples were labeled with Cy3- and Cy5-deoxycytidine triphosphates by random priming. Purification, hybridization and washing were carried out as previously reported [26, 27]. Scanned images of the arrays were processed using Feature Extraction software (Agilent Technologies).

We applied the Genomic Workbench software (Agilent Technologies) for calling DNA CNV using the aberration detection method 2 statistical algorithm with a sensitivity threshold of 6.7. Poor quality hybridization (QC > 0.3) was disregarded. Duplication or deletion of genomic segments was considered when the log2 ratio of the Cy3/Cy5 intensities of a given region encompassing at least three probes was > 0.3 or < -0.3, respectively. All hybridizations were gender-matched and processed in reverse-labeling duplicates as described previously [28]. CNVs that were not detected in both experiments were disregarded.

Copy number validation by real-time PCR

Selected CNVs detected in the patient group were validated by real-time quantitative PCR (qPCR) [29] using the SYBR Green system (Roche Applied Science, Indianapolis, IN, USA) on a 7500 Fast Real-Time PCR System apparatus (Applied Biosystems, Foster City, CA, USA). As controls or for copy number calibration, we used three DNA samples obtained from healthy donors and the qPCR values for the GAPD and HPRT genes for normalization. All samples were run in duplicate, and the data were analyzed with Microsoft Excel software (Microsoft Corp, Redmond, WA, USA) using the comparative ΔΔCt cycle threshold method (Applied Biosystems), which assumes that the calibrator DNA has two copies of the control genes.

Data analysis

Detected CNVs were compared to CNV data from oligoarray studies documented in the Database of Genomic Variants (DGV) [30]. We classified the CNVs into "rare" and "common," with rare being those that encompassed coding sequences which had never been documented as variable in the general population (DGV). CNVs were evaluated regarding proportion of total and rare CNVs, frequency of deletions and duplications, length, and gene content using the Mann-Whitney U test and Fisher's exact test.

Gene annotation was performed using the University of California Santa Cruz Genome Browser (UCSC) [31], BioMart in the Ensembl Genome Browser [32] and Catalog of Somatic Mutations. We investigated the biological features of genes contained within the rare CNVs using GOTree Machine (GOTM) software [33] to measure the enrichment in the Gene Ontology (GO) and Kyoto Encyclopedia of Genes and Genomes (KEGG) categories. GOTM reports only those enrichments that are statistically significant as determined by the hypergeometric test [34].

Results

Full CNV data on the controls and patients can be found in Additional file 2. A whole extra copy of the × chromosome was identified in one patient. This chromosomal numerical alteration was not considered a CNV. The array-CGH results are summarized in Table 1. We found a total of 1,238 CNVs in 168 individuals. CNVs observed in both patient and control samples corresponded to 81.3% of the total, all of which overlapped common CNVs (DGV). CNVs detected exclusively in one of the groups corresponded to 110 events in patients and 121 in controls. The distribution of CNVs did not differ between patients and controls (P = 0.1724, Mann-Whitney U test).

Table 1 Summary of DNA copy number variation data from breast and/or ovarian cancer patients and controlsa
Full size table

Only 49 of the 1,238 CNVs could be classified as rare, and none of them were recurrent. Those CNVs classified as rare based on DGV data corresponded to 4% (49 of 1,238 CNVs) of all CNVs detected in our study. The log2 ratios of the rare CNVs (all outside the log2 -0.65 to log2 0.45 range) were not suggestive of mosaicism, indicating that these CNVs are likely constitutive. In the control group, 3.28% (23 of 702) of CNVs detected in 23 of 100 individuals (23%) were classified as rare, whereas 26 of 536 rare CNVs (4.85%) were found in 25 of 68 patients (37%). The median numbers of total and rare CNVs per genome did not differ between controls and patients. However, the proportion of rare CNVs in patients was higher than in the controls (P = 0.0311, Fisher's exact test). The relative frequencies of duplications and deletions among rare CNVs were similar between controls (14 duplications and 9 deletions) and patients (14 duplications and 12 deletions).

We also inspected the rare CNVs for length and gene content (Table 2). The length of rare CNVs in controls and patients did not differ significantly. The lengths of rare deletions or duplications did not differ between patients and controls or within the patient group alone. However, only in the control group were the deletions found to be significantly smaller than the duplications (P = 0.0089, Mann-Whitney U test).

Table 2 Size and gene content of rare copy number variationsa
Full size table

The 26 rare CNVs and the affected genes detected in patients are described in Table 3. The published literature regarding those genes already reported to be altered in cancer is listed in Additional file 3. The rare CNVs identified in the control group are listed in Additional file 4. To evaluate whether the rare CNVs detected among cancer patients represent common CNVs in the Brazilian population, we compiled CNV data obtained from independent samples studied in our laboratory using 180 K array-CGH. These individuals (120 females and 38 males) were selected on the basis of criteria other than cancer history, including 52 patients with dementia, 56 cases of nonsyndromic hearing loss and 51 cases of Müllerian anomalies (C Rosenberg; ACV Krepischi; unpublished data). None of the rare CNVs documented in this study were detected in these independent cohorts of patients.

Table 3 Genomic positions (build 36-Hg18), size, type and affected genes of the 26 rare copy number variations identified in patientsa
Full size table

In one family, the patient with early-onset breast cancer (patient 3), as well as her affected sister, who presented with breast cancer at 28 years of age, were found to carry a 540 kb 1p31.1 microdeletion (Figure 1A). The heterozygous deletion was confirmed in the two affected sisters by qPCR. In the second family, a 90 kb microdeletion at Xq25 was detected in two affected sisters, one of whom presented with bilateral breast cancer (patient 17).

Figure 1
figure 1

Rare germline copy number variations detected in three unrelated breast cancer patients. (A) to (C) Top panels exhibit the array comparative genomic hybridization profiles of the chromosome regions (images based on the Genomic Workbench software) containing the selected copy number variation (CNV) (left) and the copy number of the corresponding sequences assessed by quantitative PCR in patients 3, 13 and 27 and control individuals (right). Bottom panels show the CNV positions in the chromosomes (small vertical red bars in the ideograms) and the corresponding genomic segments. The encompassed genes according to the Reference Sequence (RefSeq) collection are indicated by blue lines in the RefSeq genes track, and the CNVs loci reported in the general population are indicated by the blue (gain) and red (loss) bars in the Database of Genomic Variants (DGV) Struct Var track (data retrieved from the DGV) (images derived from the UCSC Genome Browser, freeze October 2011). Part (A) shows microdeletion at 1p31.1 of a 540 kb genomic segment containing the sequences of three genes. This alteration cosegregated with early-onset breast cancer in patient 3 and her sister. Part (B) shows microdeletion at 9p21.3 of a 137 kb genomic segment encompassing the gene KIA1797 and the microRNA MIR491 (patient 13). Part (C) shows microduplication at Xq13.1 of a 640 kb genomic segment affecting the FAM155B and EDA genes (patient 27).

Full size image

Figure 1 depicts two additional rare CNVs identified by array-CGH and validated by qPCR in unrelated patients. Figure 1B shows a 137 kb deletion at 9p21.3 detected in patient 13. Figure 1C illustrates a 640-kb duplication mapped at Xq13.1 (patient 27).

We characterized the function of genes located in rare CNVs in both patients and controls by GO term and KEGG pathway analysis using the Gene Ontology Tree Machine. We did not detect any significant difference in gene content in either patients or controls.

Discussion

We used array-CGH to investigate the role of rare germline CNVs in probands of individuals with a familial history of breast and ovarian cancer. Because evaluation of CNV profiles depends on ethnic background, array platform and method of analysis [35–38], all experiments were performed using the same platform, the same analytical parameters and a Brazilian control group. We disregarded the possibility that the CNVs were somatically acquired because none of the results were suggestive of mosaicism. Furthermore, the limited data available indicate that CNV profiles are rather stable in adult tissues (reviewed in [39]).

Common CNVs often contain cancer-related genes and likely play a role in carcinogenesis [40]. However, only a minority of CNVs, those with low population frequencies (rare CNVs), would be likely to contain genes that are highly penetrant genetic factors for disease susceptibility, including cancer [7, 41]. In our study, the median numbers of total and rare CNVs per genome were quite similar in patients and controls, reflecting a lack of genomic instability in this cohort of patients. These results are in agreement with data derived from a study of BRCA1-associated ovarian cancer patients [42]. Nevertheless, the patients did present a higher proportion of rare CNVs compared to controls. None of these rare CNVs were present in an independent cohort of more than 150 individuals, providing support for their nonpolymorphic nature in the Brazilian population. Assuming that some of these rare CNVs are cancer-related, the patients would carry an increased cancer risk proportionate to the number of rare genomic imbalances. The reason why we found a greater proportion of rare CNVs in patients than in controls is not clear. We could speculate that deleterious CNVs tend to be eliminated and, for some reason, conceivably less efficient apoptosis or DNA repair mechanisms, this selection would be less stringent in these patients. Whatever the reason may be, the connection between this finding and the patients' phenotypes deserves investigation.

Part of the rare genomic imbalances harbors genes that could potentially affect cancer susceptibility (see Table 3). For example, a 540 kb microdeletion at 1p31.1 was detected in two affected sisters (Figure 1A). Among the genes mapped to this deleted segment, the most relevant to cancer progression is probably ST6GALNAC5, a sialyltransferase recently reported to mediate breast cancer metastasis to the brain [43]. Another interesting alteration detected in a patient, an approximately 137 kb deletion at 9p21.3, encompassed the KIA1797 and the MIR491 genes (Figure 1B). A germline CNV affecting this genomic segment was recently reported in a colorectal cancer cohort [23]. The finding of a similar 9p21.3 deletion in independent cohorts of cancer patients strengthens their pathogenic role in cancer predisposition.

Our data support the hypothesis that germline DNA CNV is a genetic factor contributing to breast cancer predisposition, which is in accord with the findings of other studies indicating CNVs as risk factors in cancer, including neuroblastoma [19], colorectal cancer [22, 23], hepatocellular carcinoma [44], aggressive prostate cancer [20], nasopharyngeal carcinoma[45]and BRCA1-associated ovarian cancer[38]. Our findings of a possible association of a cancer predisposition phenotype with rare CNVs affecting different genes are in line with the genetic heterogeneity reported in breast cancer. This picture is different from most of the aforementioned studies, which detected recurrent common CNVs associated with cancer risk, except for prostate, colorectal and ovarian cancer CNV studies, which exhibited high CNV heterogeneity.

Conclusions

Our analysis of rare CNVs in a small cohort of BRCA1/BRCA2 mutation-negative breast and/or ovarian cancer families suggests an intriguing excess in the proportion of rare CNVs compared to controls. The future challenge will be to expand sample sizes and to follow cosegregation of given CNVs with cancer phenotype within families to identify which of the genes involved in the rare CNVs might contribute to familial breast cancer predisposition.