Supplementary Figure 3 from A Monotonic and Prognostic Genomic Signature from Fibroblasts for Colorectal Cancer Initiation, Progression, and Metastasis

Abstract

<p>PDF file - 172KB, A: Strategy for filtering and selecting the FPG signature. We selected those probes that were common to the three types of fibroblasts with RMA-normalized values >4 (10,779 probes). By this means we eliminated many genes that either were not represented in any of the classes or whose expression values were so low that they could be confused with background noise. B: In a second step, we selected genes that fulfilled the condition: (logFC) NCF vs. CAF-PT >1.5 and CAF-PT vs. CAF-LM >1.5 (179 genes) and CAF-PT vs. NCF >1.5 and CAF-LM vs. CAF-PT >1.5 (98 genes), and with values of q<0.05. C: Unsupervised hierarchical standardized heatmap of the 277-gene signature in the 34 fibroblasts samples. Columns represent fibroblast samples and are arranged along the X axis; genes are arranged along the Y axis. The sequence of normal colonic fibroblasts (NCF) > carcinoma-associated fibroblasts from the primary tumour (CAF-PT) > carcinoma-associated fibroblasts from the liver metastasis (CAF-LM) had been assimilated to cancer progression. Blue and red indicate overexpression and underexpression, respectively. D: Signature optimization by PAM analysis in the training set (34 fibroblast samples), showing the minimum 25 genes providing the lowest misclassification error (Best 25 genes out off 277 gene signature). PAM selects a set of probes according to a threshold value, usually containing those probes that minimize the cross-validation error. The 25 genes are those selected by PAM analysis from the 277 probes common to the three types of fibroblasts but which were differentially expressed following the signature definition log n-fold change (logFC) values of NCF vs. CAF-PT >1.5 and CAF-PT vs. CAF-LM >1.5 (179 genes) and CAF-PT vs. NCF >1.5 and CAF-LM vs. CAF-PT >1.5 (98 genes), and with values of q<0.05. E: For the 25 genes described in C the misclassification error (confusion with the CAF-PT class) was zero for NCF and CAF-LM classes. F: Representative western blot of selected genes (SLC7A2, ARHGDIB, TFGB2) in four samples of each fibroblast class. Tubulin was used to normalize expression (loading control). As clearly depicted in the graph SLC7A2, ARHGDIB and TGFB2 were overexpressed in CAF-LM vs CAF-PT and CAF-PT vs NCF.</p>