br When repeating the analysis with
When repeating the analysis with high-purity tumor samples (Figure S2), much of the enrichment of secretome expression increases in ECM-related functions were reduced or absent, indicating a potential contribution of non-tumor Actinomycin D to this behavior. However, significant coordinated expression de-creases among genes associated with adhesion and GAG binding were observed to an even greater extent when using high-purity tumor samples, suggesting a more tumor-specific behavior.
Another feature of interest was the significant decrease in expression associated with several gene sets that was unique to CHOL and LIHC. Even when ignoring the directionality of expression change, only CHOL and LIHC exhibited significant changes in these sets (Figure 3B). These sets included genes associated with normal liver function, including binding or activity related to lipids, alcohols, sterols, and lipoproteins. Thus, it appeared that CHOL and LIHC, which both originate from the liver, were decreasing the expression of their healthy, tissue-specific secretome components in favor of those related to malignant and invasive processes.
Figure 3. Gene Set Analysis of the Cancer Secretome
Heatmaps illustrate the (A) directional and (B) non-directional GSA results for secretome genes based on the tumor versus paired-normal fold changes and significance in 17 different cancer types. Only the GO molecular function gene set collection (MSigDB) was evaluated, and sets with <10 genes were excluded. In (A), the distinct directional gene set p values are calculated for coordinated increases (padj,dist-dir-up) and decreases (padj,dist-dir-down) in expression. The more significant (lower value) of the two directional p values for each gene set is shown in the heatmap as a log10-transformed value. The value is also ‘‘signed,’’ meaning that gene sets with a more significant decrease than increase (padj,dist-dir-down < padj,dist-dir-up) are made negative; otherwise, they are positive. Only gene sets with a padj,dist-dir % 0.01 (in either direction) in at least one cancer type are shown. A non-stacked bar plot to the left of the heatmap shows the sizes of the original gene sets (gray bars) and of the filtered gene sets containing only secretome genes (black bars). *The ammonium ion binding gene set was identical to the quaternary ammonium group binding set after removing non-secretome genes; thus, the latter set is not shown. **The chemokine activity gene set was identical to the chemokine receptor binding gene set after removing non-secretome genes; thus, the latter set is not shown. See also Figure S2.
Decreased Expression of Genes Specific to Tumor Tissue of Origin
Given that liver-derived cancers CHOL and LIHC exhibited significant and coordinated decreases in the expression of the secretome components specific to liver function, we investi-gated expression changes in the context of tissue specificity across all cancer types. In addition, to obtain a more compre-hensive picture of the secretory pathway clientele, we expanded the analysis to include any protein possessing a signal peptide, not only those that are destined for secretion (e.g., membrane proteins). This corresponded to a set of 3,491 signal-peptide genes, referred to hereafter as SP genes.
Tissue-specificity data from the Human Protein Atlas (HPA) (Uhle´n et al., 2015) was used to define the set of SP genes asso-ciated with each tissue (STAR Methods; Table S4). The DE anal-ysis (tumor versus paired normal) results for each cancer type were then evaluated in the context of the tissue-specific gene sets to determine whether any of the cancer types exhibited sig-nificant expression changes in the subset of SP genes that are typically associated with a particular healthy tissue. As in the pre-vious analyses, directionality of fold change was also taken into account to determine whether there were significantly coordi-nated expression increases or decreases.