Detection of germline variants in Brazilian breast cancer patients using multigene panel testing

Genetic diversity of germline variants in breast cancer (BC) predisposition genes is unexplored in miscegenated populations, such those living in Latin America. We evaluated 1663 Brazilian BC patients, who underwent hereditary multigene panel testing (20–38 cancer susceptibility genes), to determine the spectrum and prevalence of pathogenic/likely pathogenic (P/LP) variants and variants of uncertain significance (VUS). Associations between P/LP variants and BC risk were estimated in a case–control analysis of BC patients and 18,919 Brazilian reference controls (RC). In total, 335 (20.1%) participants carried germline P/LP variants: 167 (10.0%) in BRCA1/2, 122 (7.3%) in BC actionable non-BRCA genes and 47 (2.8%) in candidate genes or other cancer predisposition genes. Overall, 354 distinctive P/LP variants were identified in 23 genes. The most commonly mutated genes were: BRCA1 (27.4%), BRCA2 (20.3%), TP53 (10.5%), monoallelic MUTYH (9.9%), ATM (8.8%), CHEK2 (6.2%) and PALB2 (5.1%). The Brazilian variant TP53 R337H (c.1010G>A, p.Arg337His), detected in 1.6% of BC patients and 0.1% of RC, was strongly associated with risk of BC, OR = 17.4 (95% CI: 9.4–32.1; p < 0.0001); monoallelic MUTYH variants c.1187G>A and c.536A>G, detected in 1.2% (0.9% RC) and 0.8% (0.4% RC) of the patients, respectively, were not associated with the odds of BC, the former with OR = 1.4 (95% CI: 0.8–2.4; p = 0.29) and the latter with OR = 1.9 (95% CI: 0.9–3.9; p = 0.09). The overall VUS rate was 46.1% for the entire patient population. Concluding, the use of multigene panel testing almost doubled the identification of germline P/LP variants in clinically actionable predisposition genes in BC patients. In Brazil, special attention should be given to TP53 P/LP variants.

www.nature.com/scientificreports/ large number of European, Asian and Middle Eastern countries. Hence, Brazilian people offer a unique opportunity to advance the understanding of cancer genetic features in a miscegenated population 8 . In Brazil, the majority of the inherited BC studies focused on the analyses of BRCA1/2 as well as TP53, given the relatively high population frequency of the TP53 R337H (also known as, c.1010G>A, p.Arg337His) variant in people from the South and Southeast regions of Brazil 9 . However, the likelihood of carrying P/LP variants in other BC susceptibility genes among BRCA1/2 and TP53-negative patients is largely unexplored.
Recent advances in next generation sequencing (NGS) technology has reduced the cost of massively parallel sequencing, provided to physicians and patients the option of sequencing multiple genes simultaneously and broadened our understanding of the genetic etiology of inherited cancers. Multigene panel testing has proved useful as a diagnostic tool for disorders where similar phenotypes can be influenced by multiple genes such as hereditary predisposition to BC, uncovering potentially actionable findings that may be missed by traditional testing paradigms. Several laboratories have released commercial multigene panel testing ranging from six to > 100 genes 10 . Panels are cheaper, faster and increase the yield of genetic findings, more than doubling the mutation detection rate in BRCA1/2-negative patients with suspected HBOC [11][12][13][14][15][16][17][18] . However, finding a mutation in a gene where the cancer risks and/or management strategies are not known, as well as the identification of higher numbers of variants of uncertain significance (VUS), can make the results cumbersome and challenging for a physician to interpret and guide treatment 10 .
Panels have been widely available in Brazil within the past 7 years, but no study has yet assessed the prevalence and mutational spectrum of germline variants in BC susceptibility genes other than BRCA1/2 and TP53 in a large cohort of individuals with BC, who were referred for genetic evaluation. Given the rapid uptake of multigene panel testing in clinical practice, these data are urgently needed to inform genetic counseling. In this study, we report the results from 1663 consecutive individuals with a history of BC who were referred for multigene panel testing.
Eighteen patients carried P/LP variants in two genes and one patient in three different genes. Of note, two patients presented P/LP variants in both BRCA1 and BRCA2, three patients in TP53 R337H in association with BRCA1 c.5266dupC or monoallelic MUTYH (n = 2). Additionally, mutated monoallelic MUTYH, particularly MUTYH c.1187G>A, was the most frequent partner of other mutated genes (such as, BRCA1, BRCA2, PALB2 and TP53), detected in seven patients (Supplementary Table S1).
Age at BC diagnosis was significantly lower for BRCA1 P/LP variant carriers (38.0 ± 9.2 years) than in patients who were not P/LP germline carriers (43.5 ± 11.3; p < 0.001). Age at diagnosis was not associated with carriers of P/LP variants in any other genes when compared with non-carriers. Among 19 patients older than 75 years, four were P/LP variant carriers (21%), one in ATM, one in BRCA2 and two in CHEK2. Among 13 male patients, two (15.4%) were P/LP variant carriers, both in BRCA2.
Mutation spectrum of P and LP variants. Overall, 354 P/LP variants were identified in 335 patients.
Allelic heterogeneity among the patients was reflected in the appearance of 188 distinct P/LP variants in 23 genes (Supplementary Table S2). Although the mutational profile was heterogeneous, recurrent variants (detected in three or more individuals) were found in 8 genes:  Table 2). The most prevalent BRCA1 recurrent variants, which were the European founder variants c.5266dupC (n = 28) and c.3331_3334delCAAG (n = 13), accounted for 43.2% of all BRCA1 reported variants. The European founder CHEK2 recurrent variant c.349A>G (n = 7) accounted for 41.2% of all CHEK2 reported variants.
The TP53 R337H is of particular interest because it is widespread in Brazil due to a founder effect and is present in 0.3% of the southern and southeastern general populations 20 .
The Brazilian TP53 R337H variant. Overall, TP53 was the third most frequently mutated gene and contributed to 2.2% of BC cases in our cohort. TP53 P/LP variants were detected in 37 out of 1,663 BC patients and in 21 out of 18,919 reference controls (2.2% vs 0.1%; OR = 20.5; 95% CI: 11.6 -39.9; p < 0.001). It is noteworthy that the TP53 variants were concentrated in the South and Southeast (86.5%; Table 1) compared to the other regions of Brazil (32 vs 5; OR = 2.9; 95% CI: 1.1-7.4; p = 0.03).
The Brazilian TP53 R337H variant accounted for 70.3% of all TP53 reported P/LP variants and was also concentrated in patients from the South and Southeast regions of Brazil (Table 1)   . Among all the genes tested, the highest number of VUS was detected in ATM, followed by BRCA2 (Fig. 4). Approximately 90% of the VUS were missense variants (Supplementary Table S3).

Discussion
This is the largest nationwide cohort of Brazilian BC patients who underwent NGS mutigene panel testing reported to date. In this study, both allelic heterogeneity and founder mutations played a role in inherited BC. The most commonly mutated genes were BRCA1/2, which were identified in 10% of the entire cohort and accounted for almost 50% of all P/LP germline variants identified. In accordance with previous research from different countries, the use of a multigene panel test doubled the yield of P/LP variants detected, as well as increased in 12 www.nature.com/scientificreports/   Patients from all regions of the country were represented, mainly from the Southeast region of Brazil, which is the most densely populated, with more than 89 million people (or 42% of the Brazilian population). Patients from all other regions were also well represented, except for patients from the North region, which is the least densely populated with 8.8 million people in 3.87 million km 2 , covered mostly by the Amazon Rainforest.
The estimated frequency in the general population of P/LP BRCA1/2 mutations is 1:800-1:1000 per gene 21 ; however, the prevalence of pathogenic variants in BRCA1/2 varies considerably between different ethnic groups and geographic areas. In Brazil, there are no large population studies yet, so we do not have reliable estimates of its prevalence in this scenario. The prevalence of BRCA1/2 pathogenic variants in unselected, under the age of 35 or classified as high-risk BC patients was estimated to be 2.3% 22 , 16.5-20.4% and 3.4-22.5%, respectively [23][24][25][26][27][28][29][30][31][32] (Table 3). Our unselected cohort probably has the bias of comprehending mainly high-risk patients, as they were probably referred for genetic testing due to suspicion of the attending physician, identified a percentage of patients with a BRCA1/2 mutation of approximately 10%. The two most prevalent mutations are in accordance with the largest study of the Brazilian population reported to date: BRCA1 c.5266dupC and BRCA1 c.3331_3334delCAAG 33,34 . The BRCA1 c.5266dupC founder pathogenic variant is the most frequently reported in Brazil by several independent studies, but has not been observed elsewhere in South America, with the exception of an Ashkenazi community in Argentina. Notwithstanding, the BRCA1 c.3331_3334delCAAG was identified in BC patients in Spain and Portugal, as well as in Brazil, Chile, and Colombia. Despite a significant contribution of African ancestry to the genetic pool of some of the populations of Brazil, no recurrent pathogenic variants were traced back to the African continent in our cohort 35 .
Pathogenic variants in the TP53 gene are very relevant for the Brazilian population. In general, the global prevalence estimates of P/LP TP53 variants are within the range of one carrier in 3,555-5,476 individuals 36 . In Brazil, the TP53 R337H variant is estimated to occur in about 2.7 per 1,000 individuals born in southern Brazil 20 . In the 2000s, Brazilian researchers associated the TP53 R337H variant, which affects the oligomerization domain, with an increased risk of developing adrenocortical carcinomas. Subsequent studies have shown that the same variant could also increase the risk of other cancers, such as BC, but the penetrance was different [37][38][39][40][41] . The TP53 R337H variant confers a lifetime cancer risk by age 60 years of 80% in females and 47% in males. In comparison, in classic LFS, those with mutation located in typical DNA-binding domain, the cancer risk is 90% in women and 73% in men 42 . The reasons concerning the reduced penetrance of this variant is still controversial www.nature.com/scientificreports/ and usually associated with its location in the gene and biochemistry stability, which is pH dependent. A recent study showed that an extended haplotype cosegregating the TP53 R337H and XAF1 E134* alleles may lead to a more aggressive cancer phenotype than TP53 R337H alone, acting as a functional modifier by attenuating the transactivation of wild-type and hypomorphic TP53 variants, such as R337H. Carriers harboring the extended haplotype were more likely to be diagnosed with sarcomas and multiple tumors, nevertheless this association was not observed in BC patients. Further studies are needed to validate these findings and evaluate their implications on genetic counseling and clinical management of TP53 R337H carriers 43 . BC is the most common malignancy diagnosed in LFS. In Brazil, in high-risk BC patients, the prevalence of TP53 R337H ranged from 3.4-7.1% in the South/Southeast 44,45 and 0.9% in the Northeast region 30,46 (Table 4). In a cohort of 815 women affected by BC in southern Brazil who developed the disease before age 45 years, the prevalence of the TP53 R337H variant was 12.1% 45  Of note, it should also be emphasized that almost 30% of TP53 P/LP variants occurred on sites other than R337H. Some of these pathogenic variants were already reported in Brazilian BC patients, such as c.733G>A 46 and c.818G>A 39 , while one was detected in patients with Spanish ancestry (c.743G>A) 49 .
Thus, these results confirm that inheritance of TP53 R337H contribute to a significant number of BC cases in Brazil. These findings reaffirm the need for differentiated guidelines for monitoring and risk reduction strategies in patients with hereditary BC in Brazil. The investigation of the TP53 R337H variant in the Brazilian premenopausal patients diagnosed with BC is essential. These patients and their relatives who carry the same variant should receive intensive surveillance which includes at least whole-body magnetic resonance imaging (MRI) and central nervous system MRI, according to Toronto protocol 50 . In addition, breast MRI should be offered annually from age 20 years and mammography annually after age 30 years. For these patients, risk-reducing bilateral (adeno)mastectomy should be discussed. For BC patients, mastectomy should be the preferred option in an attempt to avoid radiotherapy. Nonetheless, radiotherapy should be considered when the risk of locoregional recurrence is high.
In the present study, germline variants in BC susceptibility genes other than BRCA1/2 and TP53 were also found in approximately 8% of the BC patients. Among BC clinically actionable genes, ATM, CHEK2 and PALB2 www.nature.com/scientificreports/ were the most frequently mutated. This finding is in accordance with reports from a recent study analyzing BC predisposition genes in a large cohort of patients 47 . In this work, the cited genes were associated with high or moderate BC risk with similar effect sizes in European and Asian patients which are ancestries well represented in certain regions of Brazil.
ATM was the fifth gene with the highest number of P/LP alterations; no founder mutation was found, but it had the highest number of VUS. The most common variant found in CHEK2 was c.349A>G, representing almost 1/3 of the P/LP variants in this gene. The protein encoded by this allele was found to be defective in functional tests and is likely to be pathogenic 51 . It was found in men with prostate cancer in Portugal and in women with BC in Europe and Brazil 46,52,53 .
Pathogenic variants in other genes, such as BARD1 and RAD51C were also detected. The c.176_177delAG in BARD1, was quite common (0.24%) in the present series and, interestingly, it was also detected in other BC Brazilian patients, as well as in Spanish patients, but was not reported in a recent literature review of studies analyzing BARD1 as a cancer predisposing gene, mainly comprehending French or white people 46,54,55 .
Biallelic MUTYH P/LP variants are associated with an autosomal recessive disorder, characterized by polyposis and increased risk of colorectal carcinoma. However, the cancer risk associated with germline variants in individuals carrying only one MUTYH defective allele is controversial. Studies have shown that risks of colorectal cancer for carriers of monoallelic variants in MUTYH with a first-degree relative with colorectal cancer are sufficiently high to warrant more intensive screening than for the general population, as a consequence NCCN guidelines propose colonoscopy every five years beginning at age 40 years 56 . Nevertheless, there is no strong evidence of the association of increased BC risk and carriers of monoallelic variants in MUTYH 47 . In our cohort, the fourth most commonly mutated gene was MUTYH due to the high prevalence of two monoallelic variants: MUTYH c.1187G>A and MUTYH c.536A>G. Our study, in accordance with the majority of previous studies, confirmed that those variants were not associated with increased BC risk. Thus, although it is a frequent finding in patients undergoing multigene panel testing, a monoallelic MUTYH variant should not prompt increased surveillance or risk-reducing strategies for BC 57 .
The additional pathogenic variants uncovered by multigene panel testing appears clinically relevant, albeit it is also unveiling a large number of variants that we are still not able to clearly define and classify, the VUS. We have found 767 distinctive VUS in 46.1% of our patients and 88.5% were missense variants. Studies have found that particularly among racial/ethnic minorities there is an increased likelihood of VUS results compared to women of European ancestry due to limited understanding of the normal spectrum of genetic variation in understudied groups 58 . At present, VUS management in the clinical context is challenging. Although it is typically recommended that patients with VUS are managed based on their personal and family history, rather than on the test result, communicating uncertainty has been shown to have the potential to overwhelm patients and increase their worries. In addition, a higher rate of risk reducing surgery among patients with VUS than among patients with negative results has been reported 59 . In order to overcome the challenge of VUS reclassification, the development and improvement of well represented clinical variants databases, predictive algorithms and in vitro functional assays are urgently needed.

Conclusion
In summary, the largest nationwide cohort of Brazilian BC patients who underwent multigene panel testing identified that BRCA1/2 accounted for almost 50% of all P/LP germline variants. The use of a multigene panel test almost doubled the identification of P/LP germline variants in BC predisposition genes other than BRCA1/2, as well as increased in 12 times the chance of finding a VUS. In general, the spectrum and frequencies of germline variants in non-BRCA1/2 genes mirrored those described in the literature, except for TP53 variants. In our cohort, the third most frequently gene mutated was the TP53 due to the high number of TP53 R337H carriers in the South and Southeast region of Brazil. As a consequence, the high prevalence of this TP53 variant has a significant impact in screening and risk-reducing strategies in Brazil. Sequencing and variant interpretation. Genomic DNA was obtained from a buccal swab or peripheral blood sample using standard methods. DNA Sequencing was performed by high-end Illumina platforms (HiSeq 2500 and HiSeq 4000). Base calling was performed using original Illumina tools (bcl2fastq). Bioinformatics pipeline followed Broad Institute best practices (https:// gatk. broad insti tute. org/ hc/ en-us/ secti ons/ 36000 72266 51-Best-Pract ices-Workfl ows). After alignment to the reference genome GRCh37 / UCSC hg19, low quality and duplicate readings were removed, and variants (SNPs/indels) were detected with GATK HaplotypeCaller. Enrichment and analysis concentrated on the coding sequences, flanking intronic regions (± 20 bp) and other specific genomic regions previously identified to harbor causing variants. Promoters, untranslated regions and other non-coding regions were not analyzed. Exonic deletions and duplications (CNV) were identified using ExomeDepth, an R package that estimates the number of copies by comparing the reading depth for each target with the mean reading depth for the same target from samples genotyped from the same sequenced library. If a CNV was identified, multiplex ligation-dependent probe amplification (MLPA) assay was employed to confirm the finding. The variants were classified according to algorithms based on machine learning developed by Mendelics Análise Genômica S.A and described with a nomenclature compatible with the norms and guidelines of the American College of Medical Genetics and Genomics (ACMG)/Human Genome Variation Society (HGVS). Variants interpreted as pathogenic (P) and likely pathogenic (LP) were considered positive. All variants were evaluated by a medical geneticist or pathologist or certified oncologist. Frequencies were calculated according to the total number of patients tested.

Methods
Brazilian genomic database. Reference control data were obtained from the Mendelics Análise Genômica S.A. database, which contains panel and exome sequencing data from 18,919 Brazilian individuals, sequenced as part of various disease-specific genetic tests, excluding samples from cancer cases. Case-control analysis was performed by variant or pooling P/LP variants to the gene level and comparing the frequency in BC patients relative to Brazilian reference controls.
Statistical analysis. Patients characteristics and sequencing results were tabulated, with descriptive statistics including medians, means, and standard deviations for continuous data and proportions with 95% confidence interval (CI) for categorical data are presented. A χ 2 test or Fisher exact test was used to compare proportions among cohorts and P values less than 0.05 were considered significant. Continuous variables were compared by t tests or Anova, followed by Bonferroni post-test, as necessary. Odds ratios (OR) and 95% CI were calculated by established methods. Statistical analysis was performed using SPSS Version 16.