Selecting Breast Cancer Samples with Low TILs

To select for samples with relative low number of TILs, hierarchical clustering of expression data was used. This approach has largely been described in our previous work [18 ]. In short, the proportion of lymphocytic nuclei of 96 tumor samples was assessed on H&E-stained frozen sections. For subsequent mRNA analysis, the luminal and basal samples were processed separately. These samples were divided in two groups based on the TIL percentages (high and low TIL count, median split) on which subsequent ANOVA analysis was performed to find differentially expressed probe sets passing a FDR p-value <0.05. Finally, the overlapping probe-sets for the luminal and basal sample sets were determined to create the final mRNA immune signature, see Additional File 1. Following this approach, 14 out of a total of 17 CHEK2* 1100delC and 34 out of a total of 49 BRCAX breast cancers with relative low levels of this expression signature were selected for further supervised analyses.