Heritable Susceptibility to Breast Cancer among African-American Women in the Detroit Research on Cancer Survivors Study

African-American women have higher rates of breast cancer associated with hereditary features such as the aggressive triple-negative (TN) subtype and young age of onset (1–5) compared with non-Hispanic Whites. The vast majority of studies evaluating inherited variants associated with breast cancer risk, however, are almost exclusively based on populations of non-Hispanic White women. Studies of familial breast cancers among African-American women have focused largely on BRCA1 and BRCA2 and have been primarily conducted in patients with breast cancer who meet National Comprehensive Cancer Network (NCCN) guidelines for genetic testing (diagnosed at age <50 years, having the TN subtype, or having a family history of a first-degree relative with young onset breast cancer or a relative with ovarian cancer; ref. 6). African-American women meeting NCCN guidelines for genetic testing have been found to be equally or more likely to harbor pathogenic BRCA1/2 variants compared with White women (7–9), yet there is evidence for a vastly different spectrum of variation among African-American women compared with White women, as represented by the identification of novel single base pair substitutions, an overrepresentation of rearrangements or copy number variations, and distinct African founder mutations (10).

Pathogenic variants in BRCA1 and BRCA2, however, account for only 25% of all hereditary breast cancers (11). Widespread implementation of massively parallel next-generation sequencing has substantially increased our understanding of heritable breast cancer genetics beyond BRCA1/2 through the systematic exploration and identification of at least 13 moderate-risk to high-risk breast cancer susceptibility genes (moderate risk: 20%–50% lifetime risk; high risk = 50%–85% lifetime risk; refs. 12–18). Pathogenic variants have been identified in the following breast cancer susceptibility genes (high risk: BRCA1, BRCA2, TP53, PTEN, STK11; moderate risk: BARD1, CHEK2, RAD50, RAD51C, RAD51D, PALB2, CDH1, ATM), and there is evidence for several additional moderate breast cancer risk genes, although these are less conclusive (BRIP1, PMS2, NF1, MSH6, MUTYH; refs. 11–20). Several studies have recently been published evaluating a subset of these genes in women of African ancestry, including a study of high-risk African-American women and population-based studies of patients with breast cancer in Nigeria, Uganda, and Cameroon (21–23). In the study of high-risk African Americans with breast cancer, 22% were found to have an unequivocally pathogenic variant in one of 10 known breast cancer genes (21). The population-based studies of Sub-Saharan African women identified pathogenic variants in 14.7% to 15.8% of women across 16 known breast cancer genes, and in particular, demonstrated high proportions of pathogenic BRCA1/2 variants (7%–11%) in these populations (22, 23). Further, Zheng and colleagues were able to quantify gene-level risks in their study of 1,136 Nigerian women with invasive breast cancer, where BRCA1, BRCA2, PALB2, and TP53 contributed the highest cancer risks (22). However, to our knowledge, no studies of the prevalence of inherited variants across all genes currently known to be associated with breast cancer risk have been conducted in African Americans affected with breast cancer unselected for high-risk characteristics in the United States.

Considering the underrepresentation of African-American women in genetic studies of breast cancer, the increased risk of second primaries and new primary cancers for women with hereditary cancers, and the clear clinical guidelines in terms of both treatment and follow-up for women with hereditary cancers, there is a critical need to identify and characterize disease-causing variants associated with breast cancer risk in the African-American population. The main goal of this analysis was to establish the prevalence of pathogenic variants in genes associated with breast cancer risk in a sample of African-American women with breast cancer in metropolitan Detroit unselected for age of onset, subtype, or family history. To do this, we performed targeted sequencing of germline DNA from African-American women diagnosed with breast cancer in metropolitan Detroit.

We utilized DNA derived from saliva or peripheral blood lymphocytes from 192 African-American women with breast cancer at the Karmanos Cancer Institute (KCI) in Detroit, MI, enrolled in the Detroit Research on Cancer Survivors (ROCS) Study (24–26). Briefly, participants are eligible to join the cohort if they are African American; diagnosed with a first primary, invasive colorectal, lung, prostate, or female breast cancer on or after January 1, 2013; age 20 to 79 at diagnosis; and for this study were diagnosed and/or treated at KCI in Detroit, MI. Women were contacted on average at 22.7 months after diagnosis (SD = 10.3 months, range = 7–53 months). A total of 192 African-American women with breast cancer in Detroit ROCS were sequenced, consisting of all women with breast cancer enrolled in Detroit ROCS at the time of sequencing who provided questionnaire data and consented to use of their biospecimens in future studies. Detroit ROCS provided written informed consent upon entry to the study and consented to future genetic studies using their blood or saliva. This study was conducted in accordance with the ethical principles outlined in the U.S. Common Rule and was approved by the Wayne State University Institutional Review Board.

Clinical information for all Detroit ROCS participants was obtained from the Metropolitan Detroit Cancer Surveillance System registry, including histologic type; estrogen receptor (ER), progesterone receptor (PR), and HER2 status; and stage of disease at diagnosis. Age at diagnosis and family history of cancer were obtained from baseline questionnaires.

Samples were randomized into 96-sample pools based on TN status, family history, and age of onset. Custom libraries were prepared using 100 ng of DNA using the Illumina TruSeq Custom Amplicon Kit (Illumina) using standard protocols. The custom sequencing panel was designed using Illumina Design Studio (Illumina) to capture the coding sequence of 43 genes with estimated 99% coverage. Our analyses focused on the 18 genes selected on the basis of clinically defined contributions to familial breast cancer (well established: ATM, BARD1, BRCA1, BRCA2, CDH1, CHEK2, PALB2, PTEN, RAD50, RAD51C, RAD51D, STK11, TP53, BRIP1, MSH6, MUTYH, NF1, PMS2; refs. 11, 21, 27, 28). Sequencing was performed on the Illumina MiSeq platform with 500 cycles using V2 reagents, multiplexed for 100× mean depth of coverage. DNA and library preparations were quality checked using the Agilent Tape station to check for uniformity in intensity and fragment size range, and accordingly sample outliers were repeated or ultimately rejected. Library preparation and DNA sequencing was performed at the University of Michigan Advanced Genomics Core.

Raw and aligned data (using BWA-MEM; ref. 29) were checked with FastQC (30) and MultiQC (31). Samples that failed MultiQC quality control criteria were excluded (n = 10 failed per base sequence quality or per tile sequence quality metrics). The Broad Institute GATK best practices workflow was followed (32), using HaplotypeCaller multisample calling. T_s/T_v ratios were checked as a standard quality control measure. Variants were annotated using SnpEff to identify variants predicted to disrupt the primary structure of the protein. Variants were annotated using ANNOVAR (33), which includes the databases SIFT, Polyphen HDIV & HVAR, LRT, MutationTaster, MutationAssessor, FATHMM, RadialSVM, and LR to evaluate predicted pathogenicity for missense variants. Frequency of variants in African populations in the 1,000 genomes, ExAC, or NHLBI ESP6500SI databases were annotated. Variants were filtered to retain only coding or canonical (±2) splice site variants that met the following criteria: (i) missense, nonsense, start gain, stop loss, or indel; (ii) <1% minor allele frequency in African populations in the 1,000 genomes, ExAC, or NHLBI ESP6500SI databases; (iii) <1% frequency in the sequenced patient cohort; and (iv) not located in a simple repeat region as defined by the simple repeat track coordinates from the UCSC Genome Browser (https://genome.ucsc.edu/).

We followed the American College of Medical Genetics and Genomics guidelines for interpreting sequence variants (34), adapted on the basis of the availability of information for this cohort (Supplementary Tables S1 and S2). Individual criteria met for each variant are shown in Supplementary Table S3. Benign variants were filtered in the previous step given that frequency >5% in public databases is a stand-alone criterion to classify a variant as benign. Functional data were obtained from the Human Gene Mutation Database (HGMD; http://www.hgmd.cf.ac.uk). Any variant that did not meet the set criteria in Supplementary Tables S1 and S2 for establishing pathogenicity was considered a variant of uncertain significance. We cross-referenced our variant classifications with those in ClinVar (https://www.ncbi.nlm.nih.gov/clinvar/). Variants classified as pathogenic in ClinVar with no discrepancies were also considered pathogenic. Variants classified as either “pathogenic” or “likely pathogenic” were considered pathogenic in the current analysis.

Proportions of pathogenic variant carriers and exact 95% confidence intervals (CI) were estimated using the binominal distribution using the “DescTools” package in R (https://cran.r-project.org/). Fisher exact tests were used to examine whether variant proportions vary across age groups or among patients meeting different numbers of eligibility criteria. Figures showing variant types and frequency by gene were generated using Mutplot (https://bioinformaticstools.shinyapps.io/lollipop/; ref. 35). All definitions of family history include first- and second-degree relatives. A cancer family history suggestive of a hereditary breast cancer syndrome was defined as having: (i) a first- or second-degree relative with breast cancer diagnosed under the age of 50 years, (ii) a first- or second-degree relative with ovarian cancer diagnosed at any age, or (iii) three or more first- or second-degree relatives with breast, prostate, colorectal, or kidney cancer diagnosed at any age. Associations between pathogenic variant status and high-risk characteristics were estimated using logistic regression.

A total of 192 women were eligible for this analysis, of which 182 had sufficient sequencing coverage and quality for analysis. The sequencing panel yielded 225-fold mean coverage for these 182 samples, with individual gene coverage ranging from 133- to 318-fold (Supplementary Table S4). The demographic and clinical data for our patient population are shown in Table 1. The majority of patients (65.7%) were 50 years or older at diagnosis, thus not considered early onset. About 80% patients had early-onset breast cancer (stage I or II) and either intermediate- or high-grade tumors. Slightly more than half of women had hormone receptor–positive (ER+ or PR+), HER2− breast cancer and ∼20% had TN breast cancer. About three in 10 participants had a family history of breast cancer.

Within well-established breast cancer susceptibility genes, we identified twenty-four pathogenic variants in 23 of the 182 women sequenced (12.6%, 95% CI, 8.2%–18.4%; Table 2; Supplementary Table S3). These variants were identified in five genes: BRCA2 (n = 8), BRCA1 (n = 6), ATM (n = 6), RAD50 (n = 3), and CDH1 (n = 1; Fig. 1). The overall frequency of pathogenic BRCA1/2 variants in this cohort was 7.7%. BRCA2 had the highest frequency of pathogenic variants (33.3% of pathogenic variants), which were mostly frameshift deletions (Fig. 2A), followed by ATM (25.0% of pathogenic variants), which also showed frameshift deletion/insertion or missense variants (Fig. 2C). BRCA1 accounted for 25.0% of all pathogenic variants, which were either frameshift deletion, missense, or nonsense variants (Fig. 2B). The remaining genes had lower frequencies of pathogenic variants (RAD50 13.0%, CDH1 8.7%). Time from diagnosis to enrollment did not differ between those with and without pathogenic mutations (pathogenic variant carrier: mean = 20.3 months, no pathogenic variant: mean = 23.0 months; t test P value = 0.10). Among genes with suggestive, although less definitive, evidence as breast cancer susceptibility genes, we identified an additional six women (3.5%) with pathogenic variants in four genes: NF1 (n = 2), MSH6 (n = 2), BRIP1 (n = 1), and MUTYH (n = 1; Supplementary Tables S3 and S5). We identified 109 variants of uncertain significance (VUS) in well-established breast cancer risk genes and an additional 34 VUS in probable breast cancer genes (Supplementary Table S6). One patient carried two pathogenic variants (Table 2). Patient 92333 had a frameshift deletion of BRCA2 (p.L1227fs) and a missense variant in ATM (p.R2443L). She was diagnosed at age 60 with a stage II, grade 3, hormone receptor–positive/HER2− breast cancer. She reported no family history of breast cancer, however she had one first-degree relative with prostate cancer.

We next evaluated the distributions of high-risk characteristics—defined as age of onset less than 50 years, having TN breast cancer, or a cancer family history suggestive of a hereditary breast cancer syndrome—by pathogenic variant status (Table 3). We did not observe significant differences in TN status, age of onset, or family history by pathogenic variant status. Women with pathogenic variants were 60% more likely to have a family history of breast cancer than women without pathogenic variants, although this was not statistically significant. These distributions did not change substantially when we included the eight pathogenic variants identified in suggestive breast cancer susceptibility genes (Supplementary Table S7). Given that there is known heterogeneity in the patient and/or tumor characteristics associated with specific genes (e.g., high rates of the TN subtype in BRCA1-associated tumors; ref. 36), we also examined the distribution of high-risk characteristics by individual gene. Among women with pathogenic variants, the three women with TN tumors had pathogenic variants in ATM (n = 1) and BRCA1 (n = 2). Women with early-onset breast cancer as a family history of breast cancer had mutations in each of the five genes.

We undertook a hospital-based study of heritable pathogenic variation in 18 known and probable breast cancer susceptibility genes among African-American women with breast cancer who were unselected for age of onset, subtype, or family history. Of the 182 women who were analyzed, 23 women (12%) had pathogenic variants in 5 genes (24 total pathogenic variants in BRCA1, BRCA2, ATM, RAD50, CDH1). BRCA2 had the highest proportion of variants, accounting for 33% of all pathogenic variants, followed by ATM and BRCA1 (25% of all pathogenic variants each). The total frequency of pathogenic BRCA1/2 variants in our unselected African-American cohort was 7.7%, which is lower than hospital-based studies of women with breast cancer in Nigeria (11.1%; ref. 22), Uganda (9.7%), and Cameroon (12.9%; ref. 23). Interestingly, we also found a relatively high pathogenic variation rate (3.3%) of ATM compared with the hospital-based African studies (0.4%–1.5%; refs. 22, 23), as well as to the study of African-American patients with breast cancer with high-risk characteristics (1.7%; ref. 21). Only two of the six ATM pathogenic variant carriers in our study met high-risk criteria and would have been included in a study using inclusion criteria similar to Churpek and colleagues (21). Differences in pathogenic variation frequencies also may reflect population admixture of European and African ancestry in African Americans (37).

Five genes in our sequencing panel (well established: RAD50; probable: MSH6, MUTYH, NF1, PMS2) were unique to our study compared with recent next-generation sequencing–based studies of breast cancer among women of African ancestry. We identified eight pathogenic variants in MSH6, MUTYH, NF1, and RAD50 for a prevalence of 4.4% in the overall cohort. We identified two women with germline pathogenic MSH6 variants, one with TN breast cancer diagnosed at age 61 with a family history of ovarian cancer and one with hormone receptor–positive, HER2− breast cancer diagnosed at age 45. Neither woman reported a family history of breast cancer. We included MSH6 in this study based on mounting evidence for breast cancer associated with Lynch syndrome genes. A recent report demonstrated that MSH6 pathogenic variant carriers are at two-fold increased risk of breast cancer compared with the general population (20). Women with breast cancer have also been shown to harbor pathogenic variants in MSH2, PMS2, and MLH1 (20, 38–40). Lifetime breast cancer risk for MLH1 and MSH2 pathogenic variant carriers has been estimated at 18.6% to 22% (39, 40). Further work on the association between pathogenic variation in mismatch repair genes and breast cancer risk, particularly in minority populations, is needed.

We also identified two women with pathogenic variants in NF1, three women with pathogenic variants in RAD50, and one woman with a pathogenic variant in MUTYH. NF1 was relatively recently established as a breast cancer susceptibility gene (41, 42). Breast cancers in women with NF1 pathogenic variants are typically diagnosed at younger ages, with the highest incidence in women under 40 years of age, and are associated with relatively poor breast cancer survival (42). One of the NF1 pathogenic variant carriers in our study was diagnosed under the age of 50. NCCN guidelines now recommend enhanced breast cancer screening for women with germline pathogenic NF1 variants (41). MUTYH, a gene typically associated with colorectal polyposis and colorectal cancer risk, is also associated with moderate breast cancer risk (43–50). The MUTYH pathogenic variant carrier identified in our study had no high-risk characteristics. RAD50 is also considered an intermediate-risk breast cancer gene, although the breast cancer phenotypes associated with pathogenic variants in this gene are still unclear (51–54). One of the women with a RAD50 frameshift variant in our study was diagnosed at the age of 35 with triple-positive breast cancer and also had a family history of early-onset breast cancer.

We did not find an association between age at diagnosis, subtype, and cancer family history and pathogenic variant carrier status, although we were underpowered to detect significant associations between pathogenic variant status in individual genes and patient characteristics. Specifically, we did not find an association between BRCA1 and TN subtype, although we did note that TN breast cancer was more frequent in BRCA1 compared with BRCA2 pathogenic variant carriers (33% vs. 0%, respectively). This is in contrast to previous findings where ∼70% of breast tumors arising in BRCA1 pathogenic variant carriers are TN (4, 36, 55). In 2018, a large international clinic-based TN breast cancer consortium further demonstrated that pathogenic variants in BARD1, BRCA1, BRCA2, PALB2, and RAD51D were strongly associated with high risk of TN breast cancer (56). Pathogenic variants in BRCA1 were also associated with a 3-fold increased risk of TN breast cancer compared with other subtypes in the hospital-based cohort of Nigerian women (22). We also found that 21% of our patients with pathogenic variants had a family history suggestive of a hereditary breast cancer syndrome, and 29% of these women had early-onset breast cancer (<50 years). Adedokun and colleagues (16) have previously reported that patients with pathogenic variants in BRCA1/2, did not have a family history of breast cancer. The authors emphasized that the absence of a family history does not preclude the presence of pathogenic BRCA1/2 variants and family history provided by the patient population is limited in most published reports.

One of the major strengths of our study is that it examines a population of African-American women in metropolitan Detroit previously unstudied for a wide spectrum of breast cancer risk genes, compared with currently published studies, not selected for age of onset, subtype, or family history of breast cancer within the catchment area of the Karmanos Cancer Institute in metropolitan Detroit. We also included a broader gene panel than has been reported in any single study of Africans Americans to date. Patient information was collected for the Detroit ROCS study participants using a standardized questionnaire. At the same time, we were limited in analyses of family history because these data were based on self-report without confirmation or multiple source report. Our study was also underpowered to detect associations between patient characteristics and pathogenic variant status in individual genes, which is important due to the known phenotypic heterogeneity of hereditary breast cancers. We were also unable to evaluate loss of heterozygosity in tumors for women with novel pathogenic variants, which is important for establishing causality, particularly for missense variants. This limits our ability to make causal inferences about these variants, and future tumor-based studies will be important in definitively determining their pathogenicity. We also relied on self-reported race to classify women as African American given that we were unable to estimate ancestry using data from our targeted sequencing panel. This is important given that, as we describe above, pathogenic variation frequencies could be influenced by population admixture.

This study adds substantially to the current literature describing the spectrum of pathogenic variants among an unselected cohort of African-American women with breast cancer, thus helping to define the genomic landscape of breast cancer susceptibility in African-American women with hereditary breast cancers. This will help identify high-risk individuals who would benefit from enhanced screening, surveillance, and personalized treatment in the future.

K.S. Purrington reports grants from NCI during the conduct of the study, as well as grants from NCI outside the submitted work. G. Dyson reports grants from NIH during the conduct of the study. A.G. Schwartz reports grants from NCI during the conduct of the study. No potential conflicts of interest were disclosed by the other authors.

K.S. Purrington: Conceptualization, formal analysis, funding acquisition, methodology, writing–original draft. S. Raychaudhuri: Data curation, formal analysis, writing–review and editing. M.S. Simon: Conceptualization, writing–review and editing. J. Clark: Resources, writing–review and editing. V. Ratliff: Resources, writing–review and editing. G. Dyson: Formal analysis, writing–review and editing. D.B. Craig: Formal analysis, writing–review and editing. J.L. Boerner: Resources, writing–review and editing. J.L. Beebe-Dimmer: Resources, funding acquisition, writing–review and editing. A.G. Schwartz: Resources, funding acquisition, writing–review and editing.

This work was supported, in part, by the Epidemiology Research Core and the Biobanking and Correlative Sciences Core and NIH Center Grant No. P30CA022453 to the Karmanos Cancer Institute at Wayne State University for conduct of the study (to A.G. Schwartz, J.L. Beebe-Dimmer, G. Dyson, J.L. Boerner, J. Clark, D.B. Craig). This work was also supported in part by MDCSS (HHSN261201300011I, to A.G. Schwartz, J.L. Beebe-Dimmer) and NIH (U01CA199240, to A.G. Schwartz, J.L. Beebe-Dimmer, K.S. Purrington).

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.