The RMA (Robust Multichip Average) algorithm was first applied to the microarray raw data to obtain gene expression data. All statistical analyses were performed using R and the Bioconductor suite (http://www.r-project.org/ ).
PCA was performed using the prcomp R function with default parameters.
Hierarchical cluster analysis was based on Pearson correlation between the samples. Differentially expressed genes between tumor and normal samples were identified with the limma package of Bioconductor, which applies empirical-based methods to a moderated t-statistic and takes multiple testing into account by providing an estimate of the false discovery rate (FDR). This analysis was performed in a paired way, .i.e. comparing tumor and normal samples from the same patient.
For the pairwise correlation analysis, the Pearson correlation was calculated in the ExpO breast and prostate subsets. Gene expression and annotation data from the ExpO consortium (http://www.intgen.org/expo/ ) were downloaded from GEO (GSE2109) in December 2008, including batches 1–16. The breast and prostate cancer subsets (354, respectively 83 samples) were extracted and processed separately with the RMA procedure (quantile normalization at probe-level data).
For comparison with published stromal signatures, multiple testing correction was done with the Bonferroni procedure. We re-analyzed the expression data of Ma et al.[13] (link) to obtain a list of differentially expressed genes comparing invasive breast ductal carcinoma stroma versus normal stroma. For that we used the expression data deposited in GEO (series GSE14548) and performed a paired analysis of differential expression using limma. The probesets with FDR<1% were then selected and used for the comparison. We compared our upregulated stromal genes with the ones found upregulated in breast carcinoma-associated fibroblasts compared to normal mammary fibroblasts in Bauer et al.[22] We compared our data with the pancreatic cancer stroma genes set identified in Binkley et al.[15] (link) For the comparison with the mouse study from Bacac et al.[16] (link) we considered the list containing the mouse genes found to be upregulated in invasive compared to pre-invasive prostate tumor stroma. These genes were converted into human genes using HomoloGene (build 62) and taking into account only the mouse genes with a unique homologene human ortholog.
PCA was performed using the prcomp R function with default parameters.
Hierarchical cluster analysis was based on Pearson correlation between the samples. Differentially expressed genes between tumor and normal samples were identified with the limma package of Bioconductor, which applies empirical-based methods to a moderated t-statistic and takes multiple testing into account by providing an estimate of the false discovery rate (FDR). This analysis was performed in a paired way, .i.e. comparing tumor and normal samples from the same patient.
For the pairwise correlation analysis, the Pearson correlation was calculated in the ExpO breast and prostate subsets. Gene expression and annotation data from the ExpO consortium (
For comparison with published stromal signatures, multiple testing correction was done with the Bonferroni procedure. We re-analyzed the expression data of Ma et al.[13] (link) to obtain a list of differentially expressed genes comparing invasive breast ductal carcinoma stroma versus normal stroma. For that we used the expression data deposited in GEO (series GSE14548) and performed a paired analysis of differential expression using limma. The probesets with FDR<1% were then selected and used for the comparison. We compared our upregulated stromal genes with the ones found upregulated in breast carcinoma-associated fibroblasts compared to normal mammary fibroblasts in Bauer et al.[22] We compared our data with the pancreatic cancer stroma genes set identified in Binkley et al.[15] (link) For the comparison with the mouse study from Bacac et al.[16] (link) we considered the list containing the mouse genes found to be upregulated in invasive compared to pre-invasive prostate tumor stroma. These genes were converted into human genes using HomoloGene (build 62) and taking into account only the mouse genes with a unique homologene human ortholog.
Full text: Click here