Methylation Profiling Reveals a Distinct Subgroup of BRCAx Tumors
(A) Unsupervised heirarchical clustering of the 14 BRCAx tumor samples across all 16237 genes using multiscale bootstrap resampling to generate p values for clusters (pvclust). Clusters with p < 0.0001 are boxed in red. Samples are labeled with their intrinsic subtype to show that the groupings are not related to the tumor phenotype.
(B) Heatmap and clustering of 156 significant (p.fdr < 0.05) differences between the BRCAx-a subgroup (dark blue) compared to the remaining tumors (pale blue, termed BRCAx-b) shows that the majority of differences (136/156) are increases in methylation in BRCAx-b. The full list of genes is presented in Table S2.
(C) Bisulphite sequencing validation of GEMIN8, HTR6 and LHCGR comparing BRCAx-a subgroup to the BRCAx-b subgroup. The median methylation levels are indicated by the red line. Genomic locations for the regions presented are from Human Mar. 2006 Assembly (hg18). Stars indicate statistically significant differences between groups at p < 0.05 and p < 0.01 as indicated, wilcoxon signed rank sum test.
(D) Validation of methylation differences in GEMIN8, HTR6 and LHCGR in 47 formalin fixed paraffin embedded breast tumor DNA samples. Distribution histograms are presented below to show the bimodal distribution of methylation for these genes.

Autocorrelation between Frequently Methylated Genes in Familial Breast Cancer
(A) Histogram of the frequency of methylation in tumors (percentage of tumors with MeDIP t-value > 0.5) across all unique genes, showing 1581 genes methylated in > 30% of tumors.
(B) All genes were ordered by their chromosome location, such that each gene was next to its nearest gene. Only one representative transcription start site was used for genes with multiple start sites. Autocorrelation analysis of the frequencies of methylation for each gene shows increased autocorrelation (>95% confidence interval, blue dotted line) between any gene and its next three neighboring genes, showing evidence for LRES in clusters up to four genes.

DNA-Methylation Profiles of BRCA1, BRCA2, and BRCAx Tumors
(A) Percentage of genes that pass the FDR-corrected p value cutoff of p < 0.05 in an F-test testing for differences among mutation subgroups (BRCA1 versus BRCA2 versus BRCAx) or for intrinsic subtypes (basal versus luminal A) across all genes. DNA-methylation data (MeDIP) was compared to gene-expression microarray data (GEM) for all genes or was broken down into LCPs, ICPs, and HCPs (see Figure S2). For mutation-status groups, the methylation profiles identified 822 significant (FDR-corrected p < 0.05) genes (112, 124, 586 for LCP, ICP, and HCP, respectively) compared to 47 significant (FDR-corrected p < 0.05) genes (11,13, 23) in the gene-expression profiles. The significant genes in the MeDIP profiles are biased toward the HCPs (p = 7.15 × 10⁻³⁷, chi-square test), whereas the GEM profiles are not (p = 0.517, chi-square test). For intrinsic subtypes, the gene-expression profiles identify 2811 significant (FDR-corrected p < 0.05) genes (606, 634, 1571 for LCP, ICP, and HCP, respectively), whereas the methylation profiles did not identify any significant (FDR-corrected p < 0.05) differences.
(B) Heatmap and clustering of 822 significant (FDR-corrected p < 0.05) differences between 33 tumors representing mutation subgroups indicated on the right index—BRCA1 (red, n = 11), BRCA2 (green, n = 8), and BRCAx (blue, n = 14)—reveal a distinct cluster of BRCA1 tumors compared to BRCA2 and BRCAx tumors. BRAC2 and BRCAx tumors do not divide into separate clusters. The index on the left shows the intrinsic subtypes basal (red, n = 14), luminal A (blue, n = 10), luminal B (green, n = 4), HER2 (yellow, n = 4), and normal-like (gray, n = 1).
(C) Bisulphite-sequencing analysis of RASSF1A, SGK1, LRRC55, LHCGR, and PKD2 in 33 fresh frozen tumors. The median methylation levels are indicated by the red line. Genomic locations for the regions presented are from UCSC human March 2006 assembly (hg18). Stars indicate statistically significant differences between groups at p < 0.05 and p < 0.01 as indicated (Wilcoxon signed-rank sum test).
(D) Validation of methylation differences in RASSF1A, SGK1, LRRC55, LHCGR, and PKD2 in the 47 FFPE breast tumor DNA samples.

Validation of Previously Identified Hypermethylated Genes in Breast Cancer
(A) GSEA comparing a list of genes that are commonly hypermethylated in breast cancers (n = 72) to the 16237 genes preranked on MeDIP t-values averaged across all tumors (black line), or separated into BRCA1 (red), BRCA2 (green), or BRCAx (blue) tumors. The nominal p value estimates the statistical significance of the enrichment score for a single gene set using a random permutation of the gene list.
(B) GSEA of all breast tumors (black line in Figure 4A) showing each of the 72 genes represented by black bars. This analysis shows enrichment (p = 0.042) of the hypermethylated genes across the whole data set, with a core set of 38 genes significantly enriched (p < 0.0001) among the frequently methylated genes (core hypermethylated genes; Table S3).
(C) t test for intergroup analyses comparing BRCA1 to BRCA2, BRCA2 to BRCAx, or BRCAx to BRCA1, presented as −log(10) p values. Bars above 1.3 (p = 0.05) indicate significant differences in methylation among the groups for 28 of these genes.
(D) Representative examples of genes that differentiate tumors on the basis of mutation status. MeDIP t-values are presented as box and whisker plots representing median (center line), interquartile range (box), and 95th percentiles (whisker), and samples outside this range are represented as points.
(E) Epityper validation of GREM1 comparing 33 familial breast tumors representing mutation subgroups BRCA1 (red, n = 11), BRCA2 (green, n = 8), and BRCAx (blue, n = 14). Presented is the mean methylation at each CpG site (or CpG cluster) across the amplicon (± SEM). The line between sites indicates contiguous CpG sites. Genomic location for the region presented is from the UCSC human March 2006 assembly (hg18).

Copy Number versus Methylation Reveals Frequent Methylation in Tumors with LOH or Copy-Number Gains
(A) Heatmap of median methylation levels (MeDIP t-values) for each copy-number group. For each gene (n = 16237), tumors were broken down into copy-number groups (y axis) of tumors showing homozygous deletion (HD, n = 653), loss of heterozygosity (LOH, n = 4921), normal diploidy (n = 16237), copy-neutral LOH (n = 15213), or gains (n = 9070). This shows 607 (12.3%) genes with increased methylation in the tumors that show LOH compared to the normal diploid tumors and 1032 (11.4%) genes with increased methylation in the tumors that show copy-number gains compared to diploid tumors. The methylation of genes in copy-neutral LOH regions was not significantly different from that of diploid tumors, with only 123 genes with increased methylation (0.8%). Blue, increased methylation; yellow, decreased methylation.
(B) Box plot of median methylation levels (y axis, MeDIP t-values) for the genes that have LOH and increased methylation (compared to diploid tumors), copy-neutral LOH and increased methylation, or copy-number gains and increased methylation. This analysis shows a wider range of increased methylation in the LOH and GAIN tumors compared to the neutral LOH genes.
(C) Gene-expression analysis (log-fold change) of the same genes shown as a box plot of difference in median expression levels of diploid tumors compared to LOH tumors, copy-neutral tumors, or copy-gain tumors. This shows that tumors containing LOH and methylation more often have decreased expression and that copy-gain tumors often maintain higher expression.