Introduction

Women with a pathogenic variant in BRCA1 or BRCA2 are at substantially higher risk of developing breast, ovarian and other cancers1. For these women, risk reduction options include increased surveillance, chemoprevention and prophylactic surgery2. While the identification of a BRCA pathogenic variant constitutes a clear benefit for probands and their relatives3,4, a proportion of at-risk families do not present pathogenic variants in BRCA1 or BRCA2 and their genetic risk of breast cancer remains unexplained.

Significant progress has been made in the identification of inherited genetic factors underlying hereditary cancers. Pathogenic variants in PTEN, TP53, CHEK2, ATM, NBS1, RAD50, BRIP1 and PALB2, amongst others, have also been shown to confer moderate to high risk of breast cancer5,6. Recently, several commercial laboratories have replaced BRCA testing with extended panels of cancer susceptibility genes7. Yet, for many of the genes tested, there are no approved guidelines dictating the optimal prevention or treatment strategy should a loss of function variant be identified8.

The Hispanic/Latino population is the fastest growing ethnic minority in the US and is predicted to account for 35% by 20509. Although breast cancer (BC) incidence and mortality are lower compared to most ethnic groups in the US, the outcome and prognosis in Hispanic women are worse10. The prevalence and spectrum of BRCA variants varies significantly within Latin America and the Caribbean, with little overlap in the pathogenic variants most frequently reported in each country11. In Puerto Rico, our previous work estimated that a majority of the BRCA cases are explained by few recurrent founder mutations12,13. This study aimed at identifying hereditary cancer variants in BRCA-negative breast cancer cases from Puerto Rico using the BROCA panel of cancer predisposition genes14.

Methods

Study participants

The study protocol was approved by the Institutional Review Board of Ponce Research Institute (#120829-JD) and Moffitt Cancer Center (#MCC17404) and all methods were performed in accordance with relevant guidelines and regulations. Participants were recruited between September 2012 and September 2014 from a private surgery practice located in San Juan, Puerto Rico. Evaluation of the potential study participants was conducted by a single surgeon. All women were diagnosed with breast cancer and met the NCCN (National Comprehensive Cancer Network current at the time of the study design, version 1.2012) guidelines for BRCA testing, but had received a negative BRCA clinical test result. All study participants provided informed consent and received an orientation about genetics and hereditary cancers. Trained study personnel gathered a three-generation pedigree depicting family history of cancer. Participants whose grandparents had not been born in Puerto Rico were excluded. A complete surgery clinical pathology report was obtained for the tumor(s) of each participant. Not all tumors were evaluated by the same pathologist. A total of 97 women were interviewed, 56 (57.7%) met the inclusion criteria and agreed to participate. In addition, a sample containing a known BRCA2 c.3922G > T (p.Glu1308Ter) pathogenic variant was included as a blinded control. A cohort of 500 women from Puerto Rico that do not have a cancer diagnosis and that received a negative breast cancer screening mammography within 6 months of enrollment was used to assess frequency of variants of interests in the population of Puerto Rico.

BRCA1/2 large rearrangement screening

To ensure that all women included were true negative for BRCA1/2 pathogenic variants, comprehensive screening for large rearrangements was performed by multiplex ligation-dependent probe amplification (MLPA). The MLPA kits P002-D1 (BRCA1 NM_007294.3) and P045-B3 (BRCA2 NM_000059.3) and the protocol for the final assessment of allele dosage were used as described by the manufacturer (MRC-Holland, Amsterdam, the Netherlands)15.

Sequencing

Genomic DNA was extracted using Paxgene blood DNA Kit (Qiagen, Valencia, CA) and used for the preparation of paired-end libraries with 200 base pairs (bp) inserts. An oligonucleotide pool targeting exons, non-repeating introns and selected promoter regions of the genes of interest was used for enrichment (SureSelect). Following capture, samples were sequenced on an Illumina MiSeq. The total genomic region covered was 1.1 million bp (Mbp). The 53 genes of the BROCA (http://tests.labmed.washington.edu/BROCA_versions) panel are listed in Supplementary Fig. 1. This panel was developed for the evaluation of patients at risk of hereditary cancers, it is not restricted to BC susceptibility genes. Forty-eight samples achieved proper QA/QC standards and remained for further analyses.

Alignment, variation detection, annotation

Sequence reads were aligned to the reference human genome (hs37d5) with the Burrows-Wheeler Aligner (BWA)16, and duplicate identification, insertion/deletion realignment, quality score recalibration, and variant calling were performed with PICARD (http://picard.sourceforge.net/), the Genome Analysis ToolKit (GATK)17. Sequence variants were annotated using ANNOVAR18. Additional contextual information was incorporated, including allele frequency in other studies such as 1000 Genomes (Phase 3 20130502, downloaded 11/2015), NHLBI Exome Sequence Project (ESP6500 v2, downloaded 8/2013), and the Exome Aggregation Consortium (release 0.2, downloaded 1/2015). In silico functional impact predictions, and observed impacts were assessed from databases like ClinVar (http://www.ncbi.nlm.nih.gov/clinvar/) (release 9/2016), the Catalog Of Somatic Mutations In Cancer (COSMIC) (v68), RegulomeDB19, and FunSeq20. Analyses were performed using custom bash, Perl, R scripts, and VarSifter21. Variants distribution plots were realized using Lollipops (https://zenodo.org/badge/latestdoi/20224/pbnjay/lollipops). We obtained an average of 2.7 million reads per sample, 99% aligning to the human genome and an average depth of coverage of 83.9 (Supplementary Fig. 1). After filtering for genotype quality, read depth and 1 kG population frequencies, we identified a total of 2,270 single nucleotide variants and 490 indels with allelic frequency of less than 1% in the 1 kG Admixed American reference populations (Supplementary Fig. 2).

Variant classification

Each non-synonymous coding variant was assessed individually for classification into the following categories: (1) likely benign or benign, (2) likely pathogenic or pathogenic, (3) variant of unknown significance (VUS). This classification did not distinguish variants of high penetrance from those of moderate penetrance sometimes refered to as ‘risk factor’ in public databases. First, 16 variants that have been classified as benign or likely benign by expert panels (ENIGMA for BRCA1 and BRCA222 and Insight for Lynch Syndrome genes23) were not further investigated (n = 17). For the remaining variants, those that had a frequency >0.5% in any of the gnomAD24 populations (African, East Asian, non-Finnish European, Latino, South Asian) were filtered from the analysis unless they had been reported as likely pathogenic or pathogenic on ClinVar25.Finally, the remaining variants were classified based on the ACMG criteria26. The frequency of likely pathogenic/pathogenic variants and VUS identified in NCCN clinically actionable gene was assessed in a cohort of 500 women that do not have cancer history by multiplex PCR using a Sequenom analyzer. In Silico predictions of functional significance was obtained for each non-synonymous variant using the following algorithms: fathmm27, MutationAssessor28, PolyPhen-229 and SIFT30.

Ancestry

Global ancestry proportions were estimated using a panel of 106 ancestry informative SNPs that can discriminate indigenous American, African, and European ancestry31. Genotyping was done by multiplex PCR using a Sequenom analyzer as previously described31. For each individual, respective proportions of European, African and Native American ancestry were estimated using ADMIXTURE32. In pathogenic variant carriers, local ancestry at the site of the genes of interest was extracted from Affymetrix Axiom UK array genotypes that had been phased using Shape-it33 and for which locus-specific ancestry was determined using RFMix34.

CHK2 kinase assays

To determine the impact of CHEK2 missense variants we inserted p.Glu239Lys, p.Pro484Leu, and pArg519Leu variants in a wild type human CHEK2 sequence in pCMV2-FLAG by site directed mutagenesis using QuikChange II kit (Agilent, Santa Clara, CA) (primers available upon request). We also constructed two known pathogenic variants, p.His143Tyr and p.Ile157Thr, to serve as negative controls. Constructs were confirmed by Sanger sequencing.

Human embryonic kidney 293FT cells were transfected with empty pCMV2-FLAG vectors, or with expression vectors containing wild-type CHK2 or variants using Fugene HD (Promega, Madison, WI). After ~48 h cells were exposed to 6 Gy of ionizing radiation (IR) and lysates were collected after 1 h in RIPA buffer (150 mM NaCl, 10 mM Tris-Cl [pH 7.4], 5 mM EDTA, 0.1% sodium dodecyl sulfate, 1% Triton X-100, 0.1% sodium deoxycholate). Levels of total and phosphorylated CHK2 in lysates were analyzed by immunoblotting on a 10% SDS-PAGE gel. The following antibodies were used: anti-phospho-Serine 516 CHK2 (Cell signaling), anti-phospho-Threonine 68 CHK2 (Cell signaling), anti-CHK2 (EMD Millipore) and anti-β-actin (Santa Cruz Biotechnology).

For immunoprecipitation and kinase assays cell lysates were incubated at 4 °C for 2 h with anti-FLAG M2 affinity gel. Immunoprecipitates were washed three times with RIPA buffer and used for immunoblotting with anti-phospho-CHK2 and anti-CHK2. For kinase assay, the beads were incubated with Kinase Buffer (20 mM HEPES, pH 7.4, 10 mM MgCl2, 10 mM MnCl2, 1 mM PMSF, 40 μM cold ATP, and 13 μCi [γ-32P] ATP) and bacterially expressed GST-CDC25C peptide (aa 200–256; containing the phosphorylation site for CHK2) as an exogenous substrate. Kinase reactions were incubated for 30 min at 30 °C. Samples were resolved on a 12% SDS-PAGE gel.

Statistical analyses

Descriptive and statistical analyses were conducted in R Studio35.

Results

Characteristics of the study cohort

The description of demographic, hormonal and clinical characteristics of the study cohort is presented in Supplementary Table 1. Over half of the study participants (56.2%) had their breast cancer diagnosis before the age 50. The majority of the tumors were invasive ductal carcinomas (65.9%), of lower stages; grade I and II represented 77.4% of the invasive tumors, size below 2 cm (76.9%) and luminal A or B subtype (88.2%) (Supplementary Table 1).

Variants

There were 158 coding variants: 2 truncating, 2 deletions, 106 missense and 48 synonymous variants. Each affected proband had on average 0.8 non-synonymous coding VUS, ranging from 0 to 4.

The study participants were classified according to their carrier status of a variant affecting the genes coding regions (truncating, nonsense and missense variants): Group I (n = 5; 10.2%), patients harboring at least one coding variant classified as likely pathogenic or pathogenic (L/Pathogenic) in any gene. Group II (n = 12, 24.5%), patients with VUS in breast cancer susceptibility genes classified as “clinically actionable” by NCCN27. Group III (n = 11, 22.4%), patients with variants in any other gene in the panel. Group IV (n = 21, 42.9%), patients with no pathogenic variant or VUS found in the coding sequences (Fig. 1).

Figure 1
figure 1

(a) Proportion of unrelated BC patients by carrier status and (b) distribution after grouping by age of onset and family history characteristics. Carriers categories are: harboring at least one coding variant classified as likely pathogenic or pathogenic (L/Pathogenic), VUS in NCCN clinically actionable genes, VUS in other genes, or no L/Pathogenic variant or VUS identified. Early onset breast cancer refers to individuals for which diagnosis was at 45 years of age or less. Positive family history includes only individuals for which a first, or second degree relative was diagnosed with an early onset breast cancer. NCCN clinically actionable BC genes are ATM, BRCA1, BRCA2, CDH1, CHEK2, NBN, NF1, PALB2, PTEN, STK11, and TP53 (NCCN Guidelines for Genetic/Familial High-Risk assessment v3.201925). The complete list of genes included in the BROCA panel is available in the methods section. Coding variants included truncating, missense, deletions and duplications. Synonymous variants were excluded from this analysis. BC breast cancer; VUS variant of uncertain significance.

Full size image

To assess the prior probability of being carrier, each proband was grouped according to age of onset and/or family history of early onset breast cancer (at or before age 45). There was no significant difference in the distribution of carrier status after grouping probands by age of onset and/or family history of early onset breast cancer (P = 0.8, Fig. 1). It is noteworthy however that 3 out of 4 carriers of a L/pathogenic variant had a breast cancer diagnosis at or before 50 years of age.

Pathogenic variants

After excluding the BRCA2 p.Glu1308Ter carrier (blind control), which was properly identified, four probands in 48 (8.3%) were found to carry a pathogenic variant in a non-BRCA breast cancer gene (Table 1). CHEK2 p.Ile157Thr was observed in two unrelated individuals (Fig. 2). Family history for the first patient was characterized by multiple first-degree relatives with late onset breast cancer, prostate and gastric cancer. The other carrier was diagnosed with breast cancer at the age of 44 years with no known family history of cancer.

Table 1 Pathogenic and VUS non-synonymous variants in NCCN clinically actionable breast cancer genes.
Full size table
Figure 2
figure 2

Family history of cancer (site and age of diagnosis, dx) in relatives of carriers of a known or predicted pathogenic variant in CHEK2 (a), MUTYH (b), and RAD51B (c). Sites of carcinoma are Br breast cancer, Col colorectal cancer, Gas gastric cancer, MM multiple myeloma, Pr prostate cancer. Probands are designated by a red arrow.

Full size image

The patient harboring a pathogenic variant in MUTYH (p.Gly396Asp, PRI-117 III-1) was diagnosed with breast cancer at the age of 50 and had a brother with early onset colorectal cancer. A truncating variant, p.Arg47* was identified in RAD51B predicted to result in a protein lacking ~90% of the coding region. This patient (PRI-136 III-1) was diagnosed with an invasive ductal carcinoma at age 33 and family history included early onset breast and colorectal cancer. Pedigrees representing family histories of patients carrying known pathogenic variants are depicted in Fig. 2.

Variants of uncertain clinical significance

The location of coding variants in each gene is depicted in Supplementary Fig. 3. Tables 1 and 2 present all the non-synonymous variants classified as L/Pathogenic and VUS by ClinVar25 and/or using ACMG26 criteria. The highest number of VUS was observed in ATM (n = 6) (Table 1), a clinically actionable breast cancer gene. Despite having received a negative BRCA1/BRCA2 test result, two missense VUS were identified in BRCA2.

Table 2 Pathogenic and VUS non-synonymous variants in other hereditary cancer genes.
Full size table

All of the non-synonymous variants identified have previously been reported either in ClinVar25 or in the gnomAD24 databases. Despite a low observed prevalence in gnomAD, BLM p.Arg1407Lys and CDH1 p.Tyr523Cys were each present in two unrelated individuals in this cohort. Variants ATM p.Val2937Ala and BLM p.Ser854Asn, have been observed in the gnomAD Latino population exclusively (Supplementary Table 2). For variants that have been reported in more than one gnomAD populations, some exhibited significant differences in frequencies, notably BLM p.Arg1407Lys and CDH1 p.Tyr523Cys, which were 32 and 14 times more frequent in gnomAD Latinos than in non-Finnish European, respectively.

In Silico prediction analysis of functional significance led to conflicting interpretation for most of the non-synonymous variants, with the exception of ATM p.Arg2227His, MUTYH p.Glu396Asp, and SDHB p.Ser239Cys which were predicted damaging by all four models tested (Supplementary Table 3).

In 27.1% of the cohort (n = 13), no known pathogenic variant or coding VUS were identified. Among those, PRI-185 stands out for having strong family history of cancer including early onset breast cancer, ovarian cancer and colorectal cancer (data not shown). For this individual, the prior risk of being a BRCA carrier was estimated at 14.7% using the BRCAPRO model (data not shown).

Functional impact of the CHEK2 variants

In addition to the pathogenic CHK2 variant found in two unrelated probands (Table 1), we identified three rare missense variants of uncertain significance (p.Glu239Lys, p.Pro484Leu, and p.Arg519Leu) (Table 2). To assess their impact on CHK2 stability and activity we expressed epitope-tagged CHK2 constructs containing the variants in 293T HEK cells and compared them to the wild-type CHK2 construct (positive control) and with known CHK2 pathogenic variants p.His143Tyr and p.Ile157Thr.

Variant p.Pro484Leu, located at the end of the kinase domain showed dramatically decreased levels of expression when compared with the wild type (Fig. 3C,D). Levels of expression for p.Pro484Leu were lower than for both pathogenic variants p.His143Tyr and p.Ile157Thr, but differently from p.His143Tyr it still retains the ability to auto-phosphorylate at Serine 516 and to be phosphorylated by ATM at Threonine 68 (Fig. 3C,D). Variants p.Glu239Lys and p.Arg519Leu showed levels comparable to the wild type CHK2 (Fig. 3C,D).

Figure 3
figure 3

(A) Diagram human CHK2 protein domains and variants tested in kinase assays in the current study (purple) or known pathogenic variants (red). (B) Ribbon and Surface view of human CHK2 dimer (aa 89–501) with variants tested in this study (purple) and known pathogenic variants (red) (PDB ID: 3I6U). (C) Western blot analysis of CHK2 expression and activation in HEK293T expressing wild-type CHK2 or the indicated missense variants. Cells were exposed to 6 Gy IR and WCL (whole cell lysates) were extracted 1 h post-exposure. (D) Immunoprecipitated FLAG-CHK2 complexes from HEK293T lysates were immunoblotted for CHK2 and phospho-CHK2. E. In vitro kinase assays from HEK293T cell lysates. FLAG-CHK2 immunoprecipitates were incubated with a purified GST-CDC25C (aa 200–256) peptide. Both auto-phosphorylation of CHK2 and phosphorylation of CDC25C were used as a measure of CHK2 kinase activity.

Full size image

In an in vitro kinase assay, p.Glu239Lys and p.Pro484Leu showed markedly decreased kinase activity comparable to the known pathogenic variant while variant p.Arg519Leu resulted in substrate phosphorylation levels similar to wild type CHK2 (Fig. 3E). In conclusion, functional analysis suggest that two CHEK2 variants found in our cohort (p.Glu239Lys and p.Pro484Leu) are likely to affect CHK2 activity in vivo and be associated with increased risk of breast cancer. While the evidence is less clear for p.Arg519Leu, this variant in the nuclear localization sequence and proximal to the Serine 516 site seems to auto-phosphorylate less efficiently (Fig. 3C,D).

Non coding variants

To determine whether our patients harbored potentially functional variants in addition to protein alteration, we examined variants with the two highest functional score using two annotation tools: RegulomeDB (classes: 1a/f, 2a/b/c) and FunSeq (classes: 5, 4). Of the six variants most likely to be functional by both methods, two were near CTNNA1: one was considered to be in an ultra-sensitive region by FunSeq (11 probands), and the other was predicted to be motif breaking for HDAC2 (3 probands). An MSH6 intronic variant was predicted to be motif breaking for ETS1, GABPA, and EGR1 (3 probands). A variant in the upstream promoter region of STK11 was predicted to be motif breaking for SP1 (1 proband). The promoter region of ATM harbored a variant predicted to be motif breaking for BCL11A and BCL3 (1 proband). Finally, a PALB2 promoter region variant is predicted to be motif breaking for MAX, MYC, and USF1 (1 proband).

Global and local ancestry in carriers of pathogenic variants

In the study cohort, the contribution of each ancestral population to the genome was 74.3% (SD 16.2), 14.9% (SD 12.2) and 10.8% (SD 7.4) for European, African and Native American ancestry, respectively (Supplementary Fig. 4,A). Carriers of a pathogenic variant were distributed throughout the spectrum of global ancestries. On the CHEK2 locus, both carriers of the p.Ile157Thr pathogenic variant were homozygous for European ancestry (Supplementary Fig. 4,B). The MUTYH p.Gly396Asp pathogenic variant was also located in a European ancestral context (Supplementary Fig. 4,C). In contrast, the RAD51B p.Arg47* variant was located on a chromosome segment of African origin (Supplementary Fig. 4,D).

Discussion

This study aimed at identifying inherited pathogenic variants in BRCA-negative breast cancer cases from Puerto Rico. Pathogenic variants were identified in CHEK2, MUTYH, and RAD51B in four cancer patients, which represented 8.3% of the cohort. In addition, ~30% of the VUS identified were found in NCCN actionable genes. Our study reveals an unmet clinical need of genetic testing that could benefit a significant proportion of at-risk women.

We identified four carriers with pathogenic or likely pathogenic CHEK2 variants (p.Ile157Thr in two unrelated individuals, p.Glu239Lys and p.Pro484Leu), which represents over 8% of our cohort. In ClinVar, the p.Ile157Thr variant (VCV000005591.3) currently has conflicting interpretations with most submissions representing Likely pathogenic (8) or Pathogenic (17/20). The Ile157 residue is central in the FHA-kinase domain interface and functionally analyses of this variant by several independent groups indicates that the substitution generates a partially defective protein36,37,38,39,40. The p.Ile157Thr variant may confer a modest increase in unselected and familial BC risk (OR ~ 1.5) and a high risk (OR > 4) for lobular breast cancer, although risk estimates are not available for Puerto Rican women41.

Two missense VUS (p.Glu239Lys and p.Pro484Leu), located in the N- and C-terminal lobes of the kinase domain36,38, respectively, were shown to have marked decreases in kinase activity, and may constitute pathogenic variants. Although not central to its stabilization, the p.Glu239Lys is also located at the FHA-Kinase dimerization interface36. A change from a negatively to a positively charged amino acid might have an impact on CHK2 dimerization. Pro484 faces away from the dimerization axes but replacing proline may be reflected in a less rigid positioning of the alpha helix adjacent to the C-terminal end of the kinase domain36. Of note, although p.Arg519Leu, located in the NLS region, only displayed mild effects, our assay did not evaluate its cellular distribution. It is currently a variant of uncertain significance.

Significant differences in the prevalence of CHEK2 variants across ethnic groups have been observed in the literature42. The cancer spectrum in families of carriers of pathogenic or likely pathogenic CHEK2 variants based on functional assays, was consistent with the literature including breast, early onset breast, prostate, and colorectal cancers. Remarkably, two out of four families reported multiple cases of second primary or bilateral breast cancers (PRI-177 III-2 and III-3; PRI-193 III-1 and IV-1), consistent with published meta-analysis reporting the association of CHEK2 with second primary breast cancers43. Potentially pathogenic CHEK2 variants were overrepresented in this cohort, suggesting that CHEK2 variants may affect the Puerto Rico population disproportionately. We have previously described a founder effect for a recurrent BRCA2 variant in Puerto Rico (p.Glu1308Ter)13, a phenomenon that is likely contributing to increased variant frequencies in this population.

One patient was heterozygous for a known pathogenic variant in MUTYH, with a family history of early onset colorectal cancer consistent with the reported cancer risk for this gene. MUTYH-associated polyposis is inherited as an autosomal recessive trait44, with bi-allelic carriers demonstrating up to 28-fold increase in colorectal cancer risk45. Mono-allelic carriers have modest increases in risk of colorectal cancer, which was shown to differ in magnitude based on the variant46,47 and family history of colorectal cancer47. There is also evidence supporting a role of MUTYH in extra colonic tumor risk48, including emerging associations with the risk of BC, as two missense variants in MUTYH were shown to confer a significant increase in breast cancer risk in Sephardic Jews, a population in which the MUTYH mutation prevalence was high49. As of now, the data supporting the association between monoallelic MUTYH pathogenic variants and BC risk remains sporadic and additional evidence will be required before management guidelines are modified for this gene.

A RAD51B truncating variant p.Arg47* was observed in a family characterized by early onset breast cancer and colorectal cancer. Although a segregation analysis would be required to assess the carrier status of relatives with colorectal cancer, to our knowledge, there has been no report of the association of RAD51B variant with the familial risk of colorectal cancer. In a French study, RAD51B likely deleterious variants were observed in both breast-ovarian and breast-only cancer families50. GWAS have consistently identified the 14q24.1 genomic interval locus as a breast cancer susceptibility locus, which can be broken down into 2 signals mapping within or close to RAD51B. The first has been shown to be associated with triple negative phenotype and mammographic density51, and the second with male breast cancer risk52. When combined with GWAS evidence, our results suggest that RAD51B may be a candidate that warrants reconsideration for inclusion in breast cancer susceptibility gene panels. The local ancestry at the RAD51B locus in the carrier of p.Arg47* was predicted to be of African origin, which highlights the complexity of Hispanic populations in studying genetic basis to disease as pathogenic factor may originate from any of the ancestral populations that make up their genetic backgrounds. The identification of pathogenic variants in genes not commonly included in hereditary cancer testing is a reminder that negative results, especially in the context of strong family history, should be interpreted with care.

In the current study, over 46% of the study participants were found to carry a non-synonymous coding VUS. In the BRCA1/2 genes, previous studies have reported higher rates of novel mutations or VUS in historically understudied populations53,54,55. In the clinic, VUS are associated with additional challenges. In U.S community-based practices, clinicians reported less confidence in interpreting and counseling multigene panel testing for cancer risk assessment when compared to single-gene testing56. In a survey among members of the National Society of Genetic Counselors, only 63.2% felt their patients understood the meaning of a VUS result57. For counselees, VUS results are less understood58 and have been shown to be associated with psychological distress59. The BROCA panel was developed to provide comprehensive testing for hereditary cancers, and includes several genes that have not been associated with BC risk. While standardized panel simplifies genetic testing, it also results in a higher probability of identifying VUS, which remain challenging in the clinic.

In 42.9% of the cohort, no potential causative variant was identified. Familial aggregation and twin studies have shown that even in non-hereditary breast cancer cases, a significant proportion of risk is inherited60,61. To date, genome-wide association studies (GWAS) have identified over 100 loci associated with breast cancer risk62, accounting for up to 14% of the familial risk attributed to common variants63. Although the risk associated with individual GWAS loci is not elevated enough to inform clinical decisions, polygenic risk scores were proposed as risk stratification tools in population screening programs and targeted prevention64,65,66,67. It is possible that high-risk families for which no potential causative variant is identified share an excess of low penetrance variants.

Although this is the first study addressing specifically hereditary cancer risks in an entirely Puerto Rican population, the sample size was limited. In addition, because the inclusion criteria required a BRCA1/2 negative clinical test result, patients of higher socio-economic status with better access to health care are likely to be overrepresented in this cohort. While the clinic where recruitment took place serves patients from all over the island, its location in the San Juan metropolitan area probably favored the recruitment of patients from the metropolitan area. The European contribution to the global genetic ancestry of the cohort was estimated at 74.3%, which is higher than the previously reported average of 63.7% for the Island in a census based population68. European ancestry in Puerto Rico has been associated with higher socio-economic status68. Consequently, having access to BRCA genetic testing, which was an inclusion criterion in this study, is likely to be correlated with a higher socio-economic status and might explain this enrichment in European ancestry.

Our understanding of the genetic factors driving disease risk is biased towards populations of European descent69. This is especially concerning, as it has been documented that rarer variants are more likely to show geographic clustering. In this study, we provide the first description of the pathogenic variants underlying BC genetic risk in Puerto Rican Hispanic women who tested negative for a BRCA mutation. Over 12% of the study participants had a known pathogenic variant or a variant that was predicted pathogenic by functional assay. For a large proportion of the genes included in hereditary cancer gene panels, clinical guidelines for management of risk have yet to be established pending the availability of accurate estimates of their respective cancer spectrum and the magnitude of associated risk.