Figure 2

Differential CpG methylation in breast tumors according to clinical or tumor factors. Generalized linear regression models were used to compare methylation at each of 935 CpG sites in breast tumors according to clinical or prognostic factors while controlling for age, race, menopausal status, and stage (except in analyses of tumor size or nodal status; tumor size was adjusted for in the analysis of nodal status, and vice versa). (A) Venn diagram showing overlap of significantly differentially methylated sites (false discovery rate (FDR) P <0.05) according to hormone receptor (HR) status, intrinsic subtype (basal-like versus luminal A), and p53 status. Full lists of differentially methylated CpG loci are given in Additional files 9, 10, and 11: Tables S6-S8. (B) Bar graph summarizing the numbers of differentially methylated CpG loci that were relatively hypermethylated or hypomethylated in association with clinical or tumor characteristics. For analysis of stage, methylation varied between stages 1 to 4. (C) Volcano plots showing global patterns of differential methylation across all 935 CpGs. All multivariate models were adjusted for age, race, menopausal status, stage, except for stage (adjusted for age, race, and menopausal status only), and tumor size (adjusted for age, race, menopausal status, and lymph node status). Probes significantly differentially methylated at the P <0.05 level in multivariate analysis fall above the solid line and at P <0.1 above the broken line. (D) Box-and-whisker plots showing the top five CpGs exhibiting significant differential methylation according to clinical staging or tumor characteristics. Each box plot shows the median β-value (dark bar within box) and the interquartile range (IQR = Q3-Q1) (outer boundaries of box). The whiskers (broken line) cover (Q1 − 1.5IQR, Q3 + 1.5IQR). Multivariate and FDR-adjusted P values are shown for each boxplot. No CpGs were differentially methylated according to lymph node status.