Figure 2.

Unsupervised hierarchical complete linkage clustering was performed using 347 intrinsic genes (A). Gene-expression data for these 347 genes are represented in matrix format, with rows indicating genes and columns indicating samples. Overexpressed genes are colour-coded red and underexpressed genes are colour-coded blue. Colour saturation indicates the level of overexpression or repression. Branches of replicate samples are indicated using a blue rectangle. Cell-of-origin subtype classification for each breast tumour is indicated underneath the dendrogram. Cell-of-origin subtypes are colour-coded as follows: luminal A (yellow), luminal B (purple), normal-like (green), ErbB2-overexpressing (blue) and basal-like (red). Two sample clusters have been identified, one sample cluster containing most basal-like, ErbB2-overexpressing and luminal B samples (24 out of 27) and other sample cluster containing most luminal A and normal-like samples (30 out of 32; Pearson χ²; P<0.0001). Inflammatory breast cancer samples (16 out of 19) are contained within the first sample cluster and 30 out of 40 nIBC samples are contained within the second sample cluster (Pearson χ²; P<0.0001). Principal component analysis using the same gene set was performed (B) and similarly colour-coded. The distinct cell-of-origin subtype clusters are clearly visible, confirming our initial centroid-mediated classification. The ER+ cell-of-origin subtypes (luminal A, luminal B and normal-like) are separated from the ER− cell-of-origin subtypes (basal-like and ErbB2-overexpressing) along the X-axis (first principal component), implicating ER expression as the major discriminator within the intrinsic gene list. The classification was validated using IHC for ER, PR, ErbB2, EGFR and CK5/6 on a TMA containing core biopsies from the same patients as those used for the genome-wide gene-expression analysis. Microphotographs, visualising staining results are displayed in (C).