br Figure Overview of Cancer Secretome
Figure 1. Overview of Cancer Secretome Biomarker Consensus Scoring Approach and Results
(A) Schematic overview of the scoring method. Primary tumor transcriptomic (RNA-seq) profiles were compared to those of (1) paired-normal tissue of the same patients, (2) healthy tissue matching the tumor tissue of origin, and (3) all healthy tissues in the body for which data were available. Genes were ranked according to their relative Imipenem in tumor versus other samples; those with significantly elevated expression in the tumor were ranked highly, and these were combined into a single consensus rank score.
(B) For three representative cancer types, a t-SNE projection illustrates the separation of primary tumor (red), paired-normal tissue (yellow), and healthy tissue of origin (green) samples based on the abundance (log10TPM) of the top 10 consensus-ranked genes for that cancer type. t-SNE plots for the remaining 19 cancer types with at least one of each sample type are presented in Figure S1.
(C) Box and whisker plots showing the expression of CST1 and ANGPTL4 in all healthy tissue samples, as well as in tumor samples of the five cancer types with the highest median expression of the gene. The CST1 and ANGPTL4 genes are representative of multi-cancer and cancer-specific candidate markers, which ranked highly in the third comparison (tumor versus all healthy tissues) in multiple or only one cancer type, respectively.
(D) A heat-scatterplot presenting the results of the three comparisons for the top 10 consensus-ranked secretome biomarker candidates from each cancer type.
Only cancer types with sufficient tumor, paired-normal tissue, and healthy tissue of origin samples to conduct all three comparison types are shown. The color of the circles corresponds to the mean log2FC from the first two comparison methods (FC1 and FC2), while circle size is based on the extent to which a gene is expressed at higher levels in the tumor compared to all healthy tissues (quantified as p3). Stars next to gene names indicate those whose encoded proteins are present in the human plasma proteome (Schwenk et al., 2017), in which a filled star represents a canonical (confirmed) protein, an empty star indicates some
evidence but the status is uncertain, and proteins with no star are either undetected or not included in the database. Rows and columns were clustered based on Euclidean distance between mean log2FC values.
The elevated expression of top-scoring candidates in tumors compared to all normal tissues is illustrated in Figure 1C for two example genes, cystatin SN (CST1) and angiopoietin-like 4 (ANGPTL4). These genes are representative of two types of biomarker candidates: those with elevated expression in one cancer type (ANGPTL4) and markers with elevated expression in multiple cancer types (CST1). Previous studies have experi-mentally confirmed significantly elevated protein levels of ANGPTL4 in the serum of patients with renal cell carcinoma (Dong et al., 2017) and of CST1 in the serum and urine of colo-rectal cancer subjects (Yoneda et al., 2009) relative to non-can-cer controls.
Top-Scoring Biomarker Candidates
The results for the top-ranked genes across the 20 cancer types included in all three comparisons are illustrated in Figure 1D. There was a marked clustering of high ranks among many can-cers for the collagen (COL) and matrix metalloproteinase (MMP) genes. Many members of the MMP family have been detected at significantly elevated levels in the plasma, serum, and/or urine of patients with cancers such as bladder (Eissa et al., 2007), esophageal (Mroczko et al., 2008), colorectal (Dragutinovic et al., 2011), prostate (Roy et al., 2008), lung (Izbicka et al., 2012), breast (Patel et al., 2011), and renal (Sarkissian et al., 2008), compared to non-cancer controls. Collagens have simi-larly been validated as tumor biomarkers. Previous studies have, for example, measured a significant increased abundance of type IV collagens in the plasma of pancreatic cancer subjects (Ohlund et al., 2009), COL10A1 in the serum of colorectal cancer subjects (Sole´ et al., 2014), COL6A3 in the urine of bladder cancer subjects (Linde´n et al., 2012), or degradation products of types I, III, and IV collagens in the serum of ovarian and breast cancer patients (Bager et al., 2015) relative to controls.