Figure 2: Methylation profile of candidate DMRs in the TCGA breast cancer cohort.

(a) A heatmap showing methylation profile of TCGA breast cancer samples across 4,987 HM450K probes overlapping hypermethylated DMRs identified in the discovery cohort. Rows are probes and columns are TCGA breast cancer samples profiled on HM450K—83 normal, 105 ER− tumour and 354 ER+ tumour samples. (b) A classifier (Partial Least Squares model) based on methylation values of 282 TNBC-specific probes assigns TCGA HM450K tumour samples into TNBC and non-TNBC with high accuracy; ROC analysis yields AUC of 0.9; assigning samples to highest scoring class (TNBC or non-TNBC) yields sensitivity of 0.72 and specificity of 0.94. The TCGA HM450K cohort was randomly split into training set (TNBC n=37; non-TNBC n=193) and testing set (TNBC n=36; non-TNBC n=193). The model was trained on training set and prediction accuracy assessed on testing set. (c) Box plots showing the distribution of methylation levels for two adjacent regions on chromosome 19 in TCGA normal (n=83), TNBC tumour (n=73), and other breast tumour samples (n=354). These two regions which span the promoters of ZNF154 and the adjacent ZNF671 gene, are hypermethylated in the discovery cohort and exhibit regional TNBC-specific hypermethylation in TCGA cohort, i.e. they are more heavily methylated in TNBC tumours as compared with normal and other tumour subtypes, as shown in the box plots (t-test; mean diff >0.1; P value <0.05). (d) Box plots showing the distribution of expression levels of ZNF154 and ZNF671 genes in TCGA normal (n=92), TNBC tumour (n=119) and ER+ tumour (n=588) samples. The difference in expression of TNBC tumours versus ER+ is significant for both genes (t-test; ZNF154 mean diff=−0.51; P value=9.04e−10; ZNF671 mean diff=−0.85; P value=5.50e−21).