Introduction

Breast cancer has a high prevalence worldwide, with approximately 2.1 million new cases diagnosed in 2018 with 630,000 deaths, representing 24.2% of all cancer cases in females [1]. In Brazil, 66,280 new breast cancer cases were estimated for 2020 [2]. A minor fraction of breast cancer cases presents strong hereditary components, and currently, more than 25 genes have been associated with breast cancer predisposition [3], conferring different cancer risks. However, only 4–5% of the hereditary cases are caused by highly penetrant mutations in genes such as BRCA1 and BRCA2 (BRCA1/2) [4], and genomic studies continue to uncover new genes supposedly related to this phenotype through a polygenic/oligogenic model [5].

Massive data derived from the use of the NGS technology have provided a better understanding of the spectrum of germline variants implicated in breast cancer oncogenesis, particularly in different populations. Of note, studies in highly ethnically admixed Latin American populations have identified geographic regions with exclusive mutations in the BRCA1/2, possibly due to founder effect [6]. In Brazil, the public health care system is responsible for covering the healthcare needs of about 70–80% of the population, but genetic testing and genetic counseling are not provided in general [7]. There are few studies on germline mutations predisposing to breast cancer in the Brazilian population, mainly focused in BRCA1/2 genes [8,9,10,11]. Nonetheless, interpretation of these genetic results is challenging, since the Brazilian population is not well represented in large human genetic variation databases.

Multigenic panels simultaneously analyze several genes, making it an affordable and more cost-effective alternative compared with Sanger sequencing. Only five studies have investigated in Brazilians the breast cancer susceptibility scenario beyond the BRCA1/2 genes [12,13,14,15,16].

In this study, we aim to evaluate the frequency of germline variants in up to 37 known or candidate genes for breast cancer susceptibility in a cohort of 105 Brazilian breast cancer patients referred to genetic testing due to the medical hypotheses of hereditary breast and ovarian cancer (HBOC).

Patients and methods

Study cohort

A total cohort of 105 unrelated women with breast cancer referred for molecular testing of hereditary cancer genes was studied between 2014 and 2019 in two centers: 74 from the Human Genome and Stem Cell Research Center (HUG-CELL) at the University of São Paulo, and 31 were evaluated at the Institute of Fetal Medicine and Human Genetics of São Paulo. All studied women were referred by clinical geneticists to molecular testing due to the risk of breast cancer. Of the total, 56.1% (59/105) women had a diagnosis age of 45 or less and 76.2% (80/105) had a positive family history, defined as one or more first or second-degree relatives diagnosed with breast or ovarian cancer. Clinical data were retrieved for 52 patients from the application form filled out by physicians; data from the remaining patients were not available due to differences in the admission protocols of the two participating centers.

All patients signed an informed consent form agreeing to have their data used for research, approved by the Research and Ethics Committee of the Institute of Biosciences of the University of São Paulo. All samples were de-identified for further analysis.

Multi-gene panel testing

Genetic testing was performed in different laboratories using custom panels containing phenotype-associated genes (Supplementary Table S1), followed by sequencing with Illumina MiSeq or HiSeq platforms, and alignment to the reference genome h19 (GRCh37) using BWA software [17] and variant calling with GATK best practices pipeline [18] and Unified Genotyper tool. Annotation of gnomaD frequencies and variant consequences was made using Annovar software [19]. The coding regions and splicing sites up to 37 known or candidate cancer susceptibility genes were sequenced (average coverage on target of 100–200x). The list of genes which was consensual in all panels is composed by ATM, AXIN2, BARD1, BRCA1, BRCA2, BRIP1, CDH1, CHEK2, MLH1, MRE11A, MSH2, MSH6, MUTYH, NF1, PALB2, PMS2, PTEN, RAD51, RAD51C, RAD51D, STK11 and TP53. The transcripts reference numbers are described at Supplementary Table 1. Additionally, 54 patients underwent investigation of large chromosomal rearrangements in the BRCA1/2 genes by Multiplex Ligation-dependent Probe Amplification (MLPA SALSA KITs P002-D1 and P045-B3). CNV analysis from NGS data was performed using the commercial software Nextgene (Softgenetics). Sequencing data were filtered for coding non-synonymous and splice site variants with < 0.5% frequency (ABraOM—https://abraom.ib.usp.br/; gnomAD exomes and genomes (v2.1.1)—https://gnomad.broadinstitute.org/; Kaviar (version 160,204-Public)—https://db.systemsbiology.net/kaviar/). Clinical databases were also consulted (LOVD, Breast Cancer Information Core (BIC), BRCA Exchange and ClinVar). Filtered variants were classified as P/LP and VUS using the ACMG 2015 criteria [20]. All variants described in this study were deposited in the public database LOVD (https://www.lovd.nl/3.0/home).

Results

The studied cohort presented a diagnosis age between 23 and 73 years (mean age = 44.9, SD = 11.7). When classified by age at breast cancer diagnosis, 12.4% (N = 13) were diagnosed aged 30 or younger, 23.8% (N = 25) between 31 and 40 years, 33.3% (N = 35) between 41 and 50 years, 30.5% (N = 32) older than 50 years (Supplementary Fig. 1). The mean age of patients with and without P/LP variants was 41.3 and 45.8 years, respectively, and five women without family history were diagnosed before age 35.

The vast majority of patients were diagnosed with breast cancer only (86.6%; 91/105); six (5.7%) also had ovarian cancer and nine had tumors in other sites besides breast: colorectal (N = 3), thyroid (N = 3), endometrium (N = 2) and liver (N = 1). Only one patient developed tumors in three sites (bilateral breast, ovary and liver).

Considering the patients with available data about breast cancer types (N = 49), 35 had invasive ductal carcinoma, seven intraductal carcinoma, five lobular, one papillary and one mucinous. The molecular subtypes based on the immunohistochemical evaluation of estrogen receptor, progesterone receptor, HER2 and Ki-67 status [21] were also available in 50 cases, with Luminal A being the most frequent (N = 23), followed by Luminal B (N = 12), triple-negative (N = 10), and HER2-enriched (N = 5).

In total, 22 patients (21.0%) were found to carry P/LP variants: 17 patients harboring pathogenic variants, and five with LP variants. Among the carriers of P/LP variants, 19 (86.4%) reported a positive family history (mean age = 42.5 years). The other three carrier patients without family history had a mean age of diagnosis of 36 years. We found 14 women carried P/LP variants affecting the BRCA1/2 genes (Fig. 1), and the remaining eight have variants mapped to ATM, BRIP1, CHEK2, MUTYH, RAD51D and TP53 genes. In the group of P/LP variants, non-synonymous single nucleotide (SNV), including stop codon mutations, were more frequent than splicing or frameshift (Fig. 2; Table 1). Considering the 22 P/LP variants, 10 were identified in the BRCA1 gene; among them, the c.5074+2T>C variant was identified in three patients, and was reported only in the Kaviar database (0.000006 allele frequency).

Fig. 1
figure 1

Circus plot showing the frequencies of pathogenic/likely pathogenic variants and VUS in 13 genes. P/LP variants are shown in the innermost ring, while VUS are shown in the outermost ring

Full size image
Fig. 2
figure 2

Distribution of identified pathogenic/likely pathogenic variants and VUS according to type of mutation

Full size image
Table 1 Description of pathogenic and likely pathogenic variants identified in the cohort (N = 22)
Full size table

A novel pathogenic ATM intragenic deletion covering exons 34 to 38 was identified in patient (P-38), diagnosed with breast cancer at the age of 39 years (Fig. 3). This ATM deletion was inherited from her mother, who also had breast cancer (52 years), and both developed the molecular subtype Luminal A. Additionally, the proposita’s maternal aunt was also diagnosed with breast cancer (at age 29 years); however, she was not available for genetic investigation.

Fig. 3
figure 3

Distribution of deletions already reported in the ATM gene in breast cancer patients. a Schematic presentation of the ATM protein and its functional domains. The mutations are represented by black lines and they are deletions or exon skipped, described by Renwick et al. [38] and Thorstenson et al. [33]. The deletion report in the present study is depicted with a red bar. b Distribution of probes along chromosome 11. In red probes with less coverage in exons 34–38 of the ATM gene (NextGene software image)

Full size image

We detected 16 variants of unknown significance (VUS) in fifteen patients (Table 2). The majority of the VUS were missense (14 out of 16), and two variants at splicing sites were identified (Fig. 2). Family history of breast and/or ovarian cancer was present in 75.0% (12/16) of patients carrying VUS. We also detected seven likely benign/benign variants in ATM (c.4388T>G; c.1810C>T and c.4709T>C), AXIN2 (c.1712G>A), BARD1 (c.1940A>G), BRCA2 (c.3786A>T) CDH1 (c.670C>T) genes (Supplementary Table S2).

Table 2 VUS identified in cohort (N = 16)
Full size table

Only three P/LP variants found in our study were reported in the Brazilian database of control individuals ABraOM: BRCA2 (c.8488-1G>A), CHEK2 (c.349A>G) and MUTYH (c.1187G>A). The absence of the other P/LP variants and VUS in the ABraOM database could be explained by the reduced cohort of 609 ABraOM exomes compared to the other databases analyzed (gnomAD and Kaviar). Three out of 16 VUS were not identified in any of the databases searched: two in ATM (c.8110T>C and c.6209A>G) and one in BRCA2 (c.6554C>T).

Discussion

In our cohort of 105 women, we detected P/LP germline mutations in breast cancer predisposition genes in 21.0% of them, being mostly BRCA1/2 variants (13,3%). There are few studies exploring the germline mutational burden in breast cancer predisposition genes in the Brazilian population [12,13,14,15,16], focusing mainly on BRCA1/2 analysis [8,9,10,11]. In these previous studies, conducted with non-uniform selection criteria, the frequencies of P/LP germline BRCA1/2 mutations ranged from 9.8 to 22.5%. Worldwide studies conducted in high-risk breast and/or ovarian cancer patients described mutation frequencies in these genes between 6.1 and 17.3%, depending upon the ascertainment criteria [3, 22,23,24].

In our study, the same pathogenic variant in the BRCA1 gene, c.5074+2T>C, was identified in three unrelated patients, all were evaluated at HUG-CELL. It is interesting to note that this variant is absent in Brazilians (ABraOM) and in worldwide variant frequency databases. In the study from Palmero et al. [11], which compiled the results of genetic testing of 649 individuals with P/LP BRCA1 (N = 441) and BRCA2 (N = 208) variants referred to 28 health services distributed in 11 Brazilian states, the c.5074+2T>C mutation represented 3.1% of the BRCA1 mutations, approximately 10 times less than in our cohort. In another study with 100 Brazilian patients diagnosed with ovarian cancer, the c.5074+2T>C mutation was detected in one patient [25]. Despite the small number of patients evaluated in our study, our data corroborate the observation made by Palmero et al. [11] that specific mutations occur with higher frequency in some Brazilian regions. This type of information is highly valuable to help elaborating cost-effective genetic screening strategies tailored to a given population.

In addition, seven P/LP variants were identified in ATM, BRIP1, CHEK2, MUTYH and RAD51D, all of them moderate and low penetrance genes, which represents approximately one-third of all deleterious variants detected in this cohort. This result supports the relevance of multigenic panels for the evaluation of patients with HBOC suspicion. Furthermore, the pathogenic TP53 p.R337H, which is associated with reduced cancer penetrance of Li-Fraumeni syndrome [26] was detected in a patient diagnosed at the age of 40. TP53 p.R337H is prevalent in the population of Southern/Southeastern Brazil, occurring at a frequency of approximately 0.3% in these regions [26, 27].

The novel pathogenic deletion mapped at ATM was identified in patient P-38 and her mother. Although not evaluated, one may speculate the presence of this same ATM deletion in her maternal aunt, who was also diagnosed with early breast cancer at 29 years of age (Fig. 3). ATM homozygous or compound heterozygous mutations are responsible for most cases of ataxia-telangiectasia (AT), an autosomal recessive disorder characterized by progressive difficulty in movement (ataxia) (OMIM #208900). It encodes a serine/threonine protein kinase with multiple functions in DNA repair, such as DNA double-strand break repair, chromatin remodeling and cell cycle checkpoints, through phosphorylating several downstream targets, including CHK2, NBS1, p53, and BRCA1 [28, 29]. Since the first report that female relatives of AT patients were predisposed to develop breast cancer [30], several studies have shown that ATM heterozygous mutations, especially loss of function variants, promote a moderate risk of cancer in breast, as well as in others sites [31, 32].

It is known that ATM is responsible for a significant portion of HBOC families without mutations in the BRCA1/2 genes. In a study of 270 HBOC BRCA1/2-negative families, Thorstenson et al. [33] identified seven ATM mutations in 10 families (3.7%), five of them causing protein truncation. Helgason et al. [34] identified that carriers of heterozygous loss of function variants in the ATM gene present an age of cancer diagnosis significantly earlier than non-carriers. This finding is corroborated in our study, since the average age of women without pathogenic variants was 45.8 years.

Considering the available molecular subtype information of patients with P/LP variants (N = 10), six were classified as Luminal A, one was Luminal B and three were triple-negative (Table 1). Among patients with P/LP variants in BRCA1, 25% (N = 1) had Luminal A subtype and 75% (N = 3) triple-negative subtype. Despite the small number of patients evaluated, it is possible to observe the well-known association between the presence of deleterious variants in the BRCA1 gene and the triple-negative breast cancer subtype [35]. Among patients with P/LP variants in BRCA2, one had the Luminal B subtype; and among patients with P/LP variants in other than BRCA1/2 (N = 5), all were Luminal A, the most common molecular subtype in breast cancer patients, representing approximately 70% of all cases [36].

In the present study, we found at least one VUS in 14.3% (15/105) of the patients. Among them, three missense VUS were not reported in any of the searched databases and were detected in patients with family history of breast and/or ovarian cancer (ATM c.8110T>C and c.6209A>G; BRCA2 c.6554C>T). These missense VUS might be the target of further investigation for a putative pathogenic relevance. A higher frequency of VUS has been reported in another recent study with Brazilian breast cancer patients [12]. After screening 21 DNA repair genes in 95 Brazilian HBOC patients, the authors described VUS and variants with conflicting data on pathogenicity in 76.6% of the patients. Differences in the classification criteria may explain the discrepant VUS frequency in both studies. We have applied rigorous criteria to classify variants, including consultation to several dedicated breast cancer databases.

Interpreting the clinical significance of rare coding non-synonymous SNVs is quite challenging, especially for cancer predisposition genes in admixed populations. In these cases, segregation studies can help to establish the clinical impact of the identified VUS and improve the clinical management of carriers.

The lower prices of NGS technologies has made it possible to popularize genetic testing for cancer predisposition [37], resulting in a growing detection of pathogenic variants and VUS in BRCA1/2 genes as well as in moderate penetrance genes. However, the Brazilian population exhibits great genetic heterogeneity due to its ethnic diversity, and is not well represented in international databases, which makes it difficult to evaluate the clinical relevance of germline variants found in multigenic panels [13].

This study presents limitations, such as the impossibility to obtain clinical data for the entire cohort, since the participating study centers had different inclusion criteria. In addition, the small number of evaluated women does not represent all the genetic diversity of the Brazilian population, with approximately 200 million inhabitants in a recognized ethnically admixed population. Nonetheless, our findings add to this scenario contributing to the characterization of the genetic background of breast cancer predisposition in the Brazilian population, as a useful resource to discriminate between deleterious variants and VUS.