Figure 2

Characterization of transcriptional and genomic heterogeneity of breast cancer subtypes. A) Transcriptional diversity of cancers within each subtype from the TCGA gene expression data [4] captured by the mean dispersion distance metric. B) Distributions of within and between patient standard deviations of gene expression levels for each subtype estimated from the TCGA gene expression data. C) Genomic heterogeneity of DNA copy number within each subtype estimated by the mean pairwise Hamming distance in the DNA copy number profiles of cancers from the TCGA dataset. D) Mutational heterogeneity of subtypes estimated by the mean pairwise Hamming distance between the somatic mutation profiles of cancers from the TCGA dataset. E) Transcriptional diversity of cancers from the Affymetrix U133A datasets assessed using the mean dispersion distance metric. F) Transcriptional diversity based on mean dispersion distance of basal-like tumors that achieved pathological complete response (pCR; n = 96) or had partial or no response (RD; n = 159) to preoperative chemotherapy. G) Estimated distributions of within and between patient standard deviations of gene expression within the pCR and RD basal-like phenotypes. H) Patient-patient pairwise correlation plots clustered to show substructure within the pCR and RD basal-like tumors. The scale for the Pearson correlation coefficient ranges from 0 (white) to 1 (blue). While chemo-sensitive tumors (pCR) show less structure, resistant tumors (RD) show a greater number of subgroups with relatively uniform gene expression. Boxplots represent the distribution of the corresponding metric obtained from 500 bootstrap resampling iterations of 100 cases from each subtype or subgroup.