We used different FDR cutoffs for breast and stroma to obtain lists of differentially expressed genes of comparable size. The fact that we had to use a higher FDR in the case of prostate LEE011 1211441-98-3 cancer confirms that the overall stromal response is weaker than in breast cancer. Comparison to datasets from studies on human breast and pancreatic and murine prostate cancer revealed a high degree of similarity between upregulated genes in our breast cancer patient stroma and upregulated genes in the Ma et al. and Bauer et al. study of breast tumors as well as in the Binkley et al. study of the stromal response to pancreatic cancer . Significant similarity was also found with the mouse stromal response to neuroendocrine prostate cancer growth. The prostate cancer stromal signature was also significantly related to these four datasets, albeit to a lesser degree than the breast cancer signature. As expected, our breast cancer stromal signature was more closely related to the two breast signatures than our prostate cancer stromal signature. In addition, both our breast and prostate cancer stromal signatures displayed similarity with pancreatic cancer and mouse neuroendocrine prostate cancer stroma signatures. Closer examination of the signatures, however, revealed that the similarity resided primarily among genes implicated in tissue remodeling. Periostin, found to be upregulated in both breast and prostate cancer stroma, was selected for immunohistochemical validation in a panel of human tumors known to be associated with a prominent stromal reaction. Representative images shown in Figure 2 confirm the increase of POSTN expression in the stromal compartment of breast and prostate tumor samples, compared to their normal counterparts. Intense POSTN expression was also observed in the stroma of ovarian carcinoma, as well as in lung and colon carcinoma where it was concentrated at the interface between the tumor epithelial cells and the stromal compartment that presented a robust inflammatory reaction. It is noteworthy that POSTN was not expressed in the tumor cells of the samples analyzed. Kaplan-Meier survival analysis indicated that the overall lists of breast and prostate cancer stromal genes had high prognostic value in human breast and prostate cancer datasets, respectively, but did not allow the identification of genes whose expression level is most strongly associated with patient survival. To address this issue, univariate Cox analysis was GDC-0199 in vivo performed to correlate the level of gene expression with patient survival. For each gene, a z value was obtained, indicating the strength of the correlation between the level of gene expression and patient survival. Positive z values indicated that the expression level of a gene was associated with poor prognosis,