Genome-wide association study identifies 32 novel breast cancer susceptibility loci from overall and subtype-specific analyses

Abstract

Breast cancer susceptibility variants frequently show heterogeneity in associations by tumor subtype^1-3. To identify novel loci, we performed a genome-wide association study including 133,384 breast cancer cases and 113,789 controls, plus 18,908 BRCA1 mutation carriers (9,414 with breast cancer) of European ancestry, using both standard and novel methodologies that account for underlying tumor heterogeneity by estrogen receptor, progesterone receptor and human epidermal growth factor receptor 2 status and tumor grade. We identified 32 novel susceptibility loci (P < 5.0 × 10^-8), 15 of which showed evidence for associations with at least one tumor feature (false discovery rate < 0.05). Five loci showed associations (P < 0.05) in opposite directions between luminal and non-luminal subtypes. In silico analyses showed that these five loci contained cell-specific enhancers that differed between normal luminal and basal mammary cells. The genetic correlations between five intrinsic-like subtypes ranged from 0.35 to 0.80. The proportion of genome-wide chip heritability explained by all known susceptibility loci was 54.2% for luminal A-like disease and 37.6% for triple-negative disease. The odds ratios of polygenic risk scores, which included 330 variants, for the highest 1% of quantiles compared with middle quantiles were 5.63 and 3.02 for luminal A-like and triple-negative disease, respectively. These findings provide an improved understanding of genetic predisposition to breast cancer subtypes and will inform the development of subtype-specific polygenic risk scores.

Extended Data Fig. 1. Overview of the analytic strategy and results from the investigation of breast cancer susceptibility variants in women of European descent.

Analyses included investigating for susceptibility variants for overall breast cancer (invasive, in-situ or unknown invasiveness) and for susceptibility variants accounting for tumor heterogeneity according to the estrogen receptor (ER), progesterone receptor (PR), human epidermal growth factor receptor 2 (HER2), and grade, and specifically investigating for variants that predispose for risk of the triple-negative subtype. 1) Genotyping data from two Illumina genome-wide custom arrays, the iCOGS and Oncoarray, and imputed to the 1000 Genomes Project (Phase 3). (2) Overall breast cancer (invasive, in-situ, or unknown invasiveness) analyses included 82 studies from the Breast Cancer Association Consortium (BCAC; 118,474 cases and 96,201 controls) and summary level data from 11 other breast cancer GWAS (14,910 cases and 17,588 controls; Supplementary Table 1). (3) Analyses accounting for tumor marker heterogeneity according to ER, PR, HER2 and grade included 81 studies from BCAC (106,278 invasive cases and 91,477 controls). (4) Analyses investigating triple-negative susceptibility variants included 91,477 controls and 8,602 triple-negative TN (effective sample, see Supplementary Note) cases from BCAC and 9,414 affected and 9,494 unaffected BRCA1/2 carriers from 60 studies from the Consortium of Investigators of Modifiers of BRCA1/2 (CIMBA; Supplementary Table 3). (5) Variants excluded following conditional analyses showing the identified variants to not be independent (P>1×10–6) of 178 known susceptibility variants (see Online Methods). (6) See Supplementary Figure 6 for results of country-specific sensitivity analyses. (7) See Supplementary Table 5 for the 22 independent susceptibility variants identified in overall breast cancer analyses. (8) See Supplementary Table 6 for the 16 independent susceptibility variants identified using two-stage polytomous regression, accounting for tumor markers heterogeneity according to ER, PR, HER2, and grade. Note that 8 of the 16 variants were also detected in the overall breast cancer analysis (9) See Supplementary Table 7 for the 3 independent susceptibility variants identified in the CIMBA / BCAC- triple-negative TN meta-analysis. Note that rs78378222 was detected in both the analyses using the two-stage polytomous regression and in CIMBA / BCAC- triple-negative TN.

Extended Data Fig. 2. Associations between three different polygenetic risk scores^,, and luminal A-like risk in the test dataset.

Odds ratios for different quantiles of the PRS against the middle quantile of the PRS. The odds ratios were estimated using the test dataset like (n = 7,325 Luminal-A like cases, n = 20,815 controls).

Extended Data Fig. 3. Associations between three different polygenetic risk scores^,, and luminal B/HER2-negative-like risk in the test dataset.

Odds ratios for different quantiles of the PRS against the middle quantile of the PRS. The odds ratios were estimated using the test dataset like (n= 1,779 Luminal B/HER2-negative-like cases, n = 20,815 controls).

Extended Data Fig. 4. Associations between three different polygenetic risk scores^,, and luminal B-like risk in the test dataset.

Odds ratios for different quantiles of the PRS against the middle quantile of the PRS. The odds ratios were estimated using the test dataset like (n = 1,682 Luminal B-like cases, n = 20,815 controls).

Extended Data Fig. 5. Associations between three different polygenetic risk scores^,, and HER2-enriched-like risk in the test dataset.

Odds ratios for different quantiles of the PRS against the middle quantile of the PRS. The odds ratios were estimated using the test dataset like (n = 718 HER2-enriched- like, n = 20,815 controls).

Extended Data Fig. 6. Associations between three different polygenetic risk scores^,, and triple-negative risk in the test dataset.

Odds ratios for different quantiles of the PRS against the middle quantile of the PRS. The odds ratios were estimated using the test dataset like (n = 2,006 triple-negative cases, n = 20,815 controls).

Figure 1.

Ideogram of all the independent genome-wide significant breast cancer susceptibility variants in overall, subtypes, BCAC triple-negative (TN) and CIMBA BRCA1 carriers meta-analysis. The 32 novel variants are labeled with arrows. The other significant variants are within +−500 or LD > 0.3 with previously reported variants.

Figure 2.

Heatmap and clustering of p-values from marker specific heterogeneity test for 32 breast cancer susceptibility loci (n = 106,278 invasive cases, n = 91,477 controls). P-values are for associations between the most significant variants marking each loci and estrogen receptor (ER), progesterone receptor (PR), human epidermal growth factor receptor 2 (HER2) or grade, adjusting for top ten principal components and age. P-values are raw p-values from two-tailed z-test statistics. Fifteen variants in red color were significant according to the global heterogeneity tests (FDR <0.05), of which 14 were identified by methods accounting for tumor heterogeneity. TN, triple negative.

Figure 3.

Susceptibility variants with associations in opposite direction across subtypes. The case-control odds ratios (OR) and 95% confidence intervals (95% CI) (left panel) are for associations of each of the five variants and risk for breast cancer intrinsic-like subtypes estimated from the first-stage of the two-stage polytomous regression fixed-effects model (n = 106,278 invasive cases, n = 91,477 controls). The case-case ORs 95%CI (right panel) are estimated from the second stage parameters of a fixed effect two-stage polytomous models testing for heterogeneity between the five variants and estrogen receptor (ER), progesterone receptor (PR), human epidermal growth factor receptor 2 (HER2) and grade, where ER, PR, HER2, and grade are mutually adjusted for each other. MAF, minor allele frequency.

Figure 4.

Heatmap of candidate causal variants (CCVs) overlapping with enhancer states in primary breast subpopulations for five variants with associations in opposite direction across subtypes. Three different breast subpopulations were considered: basal cells (BC), luminal progenitor (LP) and luminal cells mature (LM). Based on a combination of H3K4me1 and H3K27ac histone modification ChiP-seq signals, putative enhancers in BC, LP, and LM were characterized as “OFF”, “PRIMED” and “ACTIVE” (Online Methods). The CCVs overlapping with enhancers were colored as red, otherwise were white.

Figure 5.

Genetic correlation between the five intrinsic-like breast cancer subtypes and BRCA1 mutation carriers estimated through LD score regression. See Supplementary Table 16 for further details. Both the color and size of the circles reflect the strength of the genetic correlations.