The CBCS was approved by the Institutional Review Board at the University of North Carolina at Chapel Hill School of Medicine. The WCHS was approved by the Institutional Review Boards at the University of Medicine and Dentistry of New Jersey (presently Rutgers University), Mount Sinai School of Medicine, and Roswell Park Cancer Institute. The BWHS was approved by the Institutional Review Board at the Boston University School of Medicine. The MEC was approved by the Institutional Review Boards of the University of Hawaii and University of Southern California. Written informed consent was obtained from each participant.

Two case-control studies, the Carolina Breast Cancer Study (CBCS) and the Women’s Circle of Health Study (WCHS), and two prospective cohort studies, the Black Women’s Health Study (BWHS) and the Multi-Ethnic Cohort (MEC) comprise the AMBER Consortium (Palmer et al., 2014a,b; Bandera et al., 2015); the four studies are described below.

The CBCS is a population-based case control study of women ages 20–74, conducted in 44 counties in North Carolina that began in 1993 (Newman et al., 1995). Breast cancer cases were identified using Rapid Case Ascertainment in cooperation with the NC Central Cancer Registry. Controls were identified using Division of Motor Vehicles lists (for women under age 65) and Health Care Financing Administration lists (for women 65 and older). The age range of study participants was 20–74 years. CBCS contributed 1,408 African–American cases and 615 controls to this analysis.

The WCHS includes both European and African American women, 20–75 years of age, newly diagnosed with a first primary, and histologically confirmed breast cancer. Cases were identified from 10 counties in New Jersey through the New Jersey State Cancer Registry and through major metropolitan hospitals in New York City serving a large minority population. Controls, frequency matched to cases by 5-years age categories and self-reported race, were identified using random digit dialing in New York City and New Jersey, as well as from the community (Ambrosone et al., 2009; Bandera et al., 2013). Recruitment in New York City took place between 2003 and 2008. Recruitment at New Jersey started in 2006 and is ongoing. WCHS contributed 821 African American breast cancer cases and 834 African American controls to this analysis.

The BWHS is an ongoing prospective cohort study of 59,000 US women aged 21–69 years from across the United States who self-identify as “black” (Rosenberg et al., 1995). Beginning in 1995, women completed a 14-page postal health questionnaire. Biennial postal questionnaires are mailed to participants to update information. During 2004–2007, we obtained saliva samples as a source of DNA from 26,814 participants using the mouthwash-swish method (Cozier et al., 2004). BWHS contributed 901 cases and 2,249 controls to this analysis.

The MEC study is a prospective cohort that includes over 215,000 individuals from Hawaii and California (primarily Los Angeles) assembled between 1993 and 1996 (Kolonel et al., 2000; Pike et al., 2002). African Americans, Native Hawaiians, Japanese, Latinos and European Americans comprise the predominant population. Blood samples were collected from incident breast cancer cases starting in 1994 and identified by cohort linkage to Surveillance, Epidemiology and End Results (SEER) registries, as well as a random sample of MEC participants to serve as controls for genetic analyses. MEC contributed genetic data from 499 African American cases and 960 African American controls to this analysis.

Genotyping of constitutional DNA of CBCS, WCHS, and BWHS samples was performed by the Center for Inherited Disease Research (CIDR) using the Illumina Human Exome Beadchip v1.1 array with custom content. The standard content of this array includes more than 240,000 coding variants, as well as tag SNPs for GWAS hits, a grid of common variants, and ancestry informative markers (AIMs). A more complete description of the exome chip design and standard content is available from http://genome.sph.umich.edu/wiki/Exome_Chip_Design.

Genotype quality control for this study has been described previously (Laurie et al., 2010). Genotype data were available for 2,624 autosomal AIMs (Supplementary Table 1) included as standard content in the exome array and typed in AMBER participants, including 3,130 cases, 1,674 ER+ cases and 963 ER- cases (515 triple negative) and 3,698 controls. In addition, genetic data from 499 cases (130 ER- including 86 triple negative, 294 ER+, 75 ER unknown) and 960 controls in the MEC study were available from previous genotyping on an earlier version of the exome array that contained the same AIMs. Final sample size for the current admixture analyses was 8,287 participants: 3,629 cases (1,093 ER- including 601 triple negative, 1,968 ER+, and 568 ER status unknown) and 4,658 controls.

We assessed population stratification by calculating principal components of genetic variation using the smartpca program in the EIGENSOFT package (Patterson et al., 2006). Supplementary Figure 1 shows a plot of the top two principal components of the participants stratified by study. No obvious differences in distribution of these principal components were observed across studies.

We used ADMIXMAP software version 3.8.3103 (Hoggart et al., 2004; Montana and Pritchard, 2004; Reich and Patterson, 2005) to estimate African locus-specific ancestry and determine excess sharing across the genome. ADMIXMAP uses a Markov chain Monte Carlo algorithm to generate a score test at each position in the genome that measures local ancestry associations with a phenotype (Hoggart et al., 2004). Admixture mapping is generally carried out using case-only and/or case-control analyses. The case-only approach compares locus-specific ancestry in each chromosomal position with average genome-wide ancestry. The case-control method tests for differences of locus-specific ancestry between cases and controls adjusting for individual admixture and other covariates. Case-only analysis has greater statistical power than case-control assuming that deviation in ancestry between cases and controls is not due to population stratification (Shriner, 2013). We therefore first performed a case-only admixture scan and then a case-control admixture analysis to confirm local ancestry associations from the case-only results. Case-control analyses were adjusted for global individual admixture, study, age (10-year groups), geographic region, and source of constitutional DNA (Oragene-saliva, blood, mouthwash-saliva). Statistical significance was assessed using Z statistics, with a threshold of |Z| > 4.0 considered genome-wide statistically significant. A negative Z-score indicates that higher African (lower European) ancestry at that particular locus is associated with higher breast cancer risk while a positive Z-score indicates that lower African (higher European) ancestry is associated with higher breast cancer risk. We performed analyses of breast cancer overall, and by ER and triple negative status. For each analysis, we ran ADMIXMAP using 2,000 burn-in iterations and 8,000 follow-on iterations. The percentage of African ancestry for each individual was determined using ADMIXMAP and compared across controls, all cases and by ER status as well as by site, using logistic regression adjusting for the covariates described above.

In order to identify SNPs that may help explain any observed association between local African ancestry and risk of breast cancer regions meeting the genome-wide significant criterion (|Z| > 4) in AMBER, fine-mapping analyses were performed using typed and imputed data from the African American Breast Cancer (AABC) consortium, which included participants from six epidemiological studies (Chen et al., 2011, 2013) after excluding MEC, WCHS, and CBCS subjects which were included in AMBER. Regional association analyses were conducted in 2,114 cases (1,116 ER+, 591 ER-, and 407 unknown ER status) and 1,911 controls typed on the Illumina Human1M-Duo BeadChip [34] and imputed using MACH with HAPMAP2 as a reference. Within each genome-wide significant admixture peak a Bonferroni correction for the number of directly genotyped markers tested was used to determine statistical significance (Palmer et al., 2016).

Table 2 shows the four genome-wide significant regions in case-only analysis (|Z| ≥ 4, P < 6 × 10^-5) that also had confirmatory results in case-control analysis (|Z|≥ 2.0, P < 0.05). We identified two novel genome-wide significant regions, showing evidence of excess African ancestry associated with ER+ breast cancer on 4p16.3 and 17q25.1. The 10q26 and 11q13.2 regions both contain SNPs previously identified in published GWAS of breast cancer in European ancestry women (Hunter et al., 2007; Turnbull et al., 2010), and fine-mapping efforts have identified independent signals in African American women (Chen et al., 2011).

Supplementary Table 2 contains admixture results for all 2,624 AIMs and breast cancer subtypes. Supplementary Table 3 shows admixture results with suggestive evidence of association, defined as 4 > |Z|≥ 3.5 in case-only and |Z|≥ 2.0 in case-control analyses. Of these 13 suggestive regions, seven contain at least one genome-wide significant marker previously associated with breast cancer (Hindorff et al., 2015).

We used an independent study (African American Breast Cancer Consortium, AABC) to test for association of SNPs in the admixture associated regions identified for ER+ and overall breast cancer. We defined significant association at P < 5.8 × 10^-6 after Bonferroni correction for multiple testing based on adjustment for the 8,653 typed markers included. AABC does not have information about triple negative breast cancer, thus we did not follow up the three regions showing suggestive association with this subtype. We report on the most significant associations in each genomic region identified in the admixture mapping (Table 3; Supplementary Figure 2).

This region contains no previously reported genome wide significant breast cancer loci, but loci for bladder cancer, as well as other traits and diseases include triglycerides and cholesterol levels, bone mineral density, type 2 diabetes and Parkinson’s disease (Supplementary Table 4). The most strongly associated SNP in 4p16, rs112545418, in ZFYVE28, was in moderate linkage disequilibrium (r² = 0.48) with a nearby variant rs17132398, associated with ER-positive breast cancer at P = 1.8 × 10^-4. Analysis of rs17132398 using RegulomeDB database (Boyle et al., 2012) showed that this SNP is located inside an enhancer region across multiple cell lines, and is likely to affect binding of transcription factors and expression of ZFYVE28, which is a negative regulator of epidermal growth factor receptor (EGFR) signaling (Mosesson et al., 2009). The low frequency of the rs112545418 risk A-allele (1%) and the corresponding risk T-allele of rs17132398 (0.4%) in European ancestry populations may explain why it has not been identified in previous GWAS of breast cancer.

The 10q26 showed excess of local African ancestry associated with risk of overall breast cancer and was our strongest admixture association with overall breast cancer. This region contains well established SNPs in FGFR2 associated with increased risk of breast cancer (Chen et al., 2011). A recent fine-mapping of FGFR2 (Meyer et al., 2013) identified three independent signals in Europeans and East Asians: the first signal represented by rs35054928, the second signal by rs45631563, and the third one by rs2981578 (Meyer et al., 2013). The rs2981578 was previously identified as the SNP with the strongest association with overall and ER+ breast cancer in FGFR2 in African Americans (Chen et al., 2011), and it was also shown recently by us using AMBER consortium data (Ruiz-Narvaez et al., 2016). Interestingly, the excess of African ancestry in breast cancer cases extends over both FGFR2 and a second region spanning from 131 to 134 Mb on chromosome 10. This second region contains our significant variant, rs12244041 (Table 3; Supplementary Figure 2), as well as a non-synonymous variant, rs148966337, previously shown to be associated with both overall and ER- breast cancer (Haiman et al., 2013). This is a less common variant (1.9%) that was only present in the African American women from the MEC study, but completely absent in women of other continental ancestries (Haiman et al., 2013).

Although the chromosome 11q13 region has been previously shown to contain breast cancer SNPs (Michailidou et al., 2013; Ahsan et al., 2014), all three of the SNPs we found within the admixture-identified region were independent of the published GWAS-identified SNPs. One of the three SNPs in the chromosome 11 region, rs117564384, resides in SCGB1D2. SCGB1D2 expression is high in and almost exclusive to mammary tissue (Culleton et al., 2007), significantly differs between histologically normal breast tissue and breast cancer tissue (Zubor et al., 2015), and it also has been used as a marker to detect disseminated tumor cells (DTC) in breast cancer (Brown et al., 2006; Lacroix, 2006).

In summary, this is the largest admixture analysis of African American breast cancer, both in number of subjects and markers, to date. One limitation of this report is the lack of characterization of tumors by grade and stage that may show differential association with locus-specific African ancestry. Nevertheless, the current study includes more than twice the number cases and a higher density of AIMs throughout the genome compared to the previous admixture mapping of breast cancer in AA women (Fejerman et al., 2009). In addition, the use of controls in the present study allowed us to confirm significant results from the case-only scan. Fine-mapping of the identified regions showed that multiple genomic variants may be within admixture associated regions. Our results suggest that several loci with small to moderate effects rather than a few with strong effects are responsible for the differences of incidence of ER subtypes among US white and black women.