br INTRODUCTION br Early diagnosis is a major
Early diagnosis is a major factor contributing to cancer treatment success (Etzioni et al., 2003; World Health Organization, 2017). As such, there have been extensive efforts to identify with improved accuracy and sensitivity biomarkers that indicate the presence of cancerous MK-571 sodium salt hydrate in a subject (Belczacka et al., 2019; Sawyers, 2008). Recent work has focused on the analysis of markers in biofluids, such as urine, plasma, or cerebrospinal fluid, as they are non-invasive and can be tested with greater frequency than tissue biopsies (Crowley et al., 2013; Diaz and Bardelli, 2014; Webb, 2016). A class of proteins that are of partic-ular interest in this context is the secretome, which is the set of proteins secreted to the extracellular space, as they are gener-
ally more abundant in biological fluids than intracellular proteins (Kulasingam and Diamandis, 2008; Stastna and Van Eyk, 2012).
The secretome is considered a valuable reservoir of potential biomarkers for cancer and other diseases (Makridakis and Vla-hou, 2010; Xue et al., 2008), and a number of studies have aimed to explore this class of proteins in search of tumor biomarker candidates. For example, Welsh et al., 2003 used Gene Ontology (GO) terms associated with an extracellular location and protein sequence patterns to define the secretome to compare the mi-croarray gene expression profiles of 150 carcinomas spanning 10 tissues of origin to those of 46 healthy tissue samples. Biomarker candidates were validated via comparison with previ-ous studies that had measured increased expression of the gene or protein in cancer tissue or in the serum of cancer patients. Other bioinformatics-based approaches to predict secreted cancer biomarkers include those of Prassas et al. (2012) for co-lon, lung, pancreatic, and prostate cancers, and Vathipadiekal et al. (2015) for ovarian cancer. These and other, similar investi-gations demonstrate the validity of using a bioinformatics-based approach to predict proteomic biofluid markers and to identify many new, promising biomarker candidates. However, these studies were generally restricted to a limited number of samples, tissue types, and/or cancer types; were often based on microar-ray data rather than RNA sequencing (RNA-seq) data; provided only a single set of candidates rather than a complete ranked list; and conducted little or no exploration of the biological func-tions associated with the proposed biomarkers.
Proteomic approaches have often been used to profile the cancer secretome (Brandi et al., 2018; Geyer et al., 2017; Hanash et al., 2008; Makridakis and Vlahou, 2010; Papaleo et al., 2017; Schaaij-Visser et al., 2013; Xue et al., 2008). These studies generally involve in vitro analyses of cell-line conditioned media or analysis of tumor interstitial fluid (or a more distant fluid such as blood, plasma, urine, or saliva) (Papaleo et al., 2017). For example, Wu et al. (2010) used SDS-PAGE followed by liquid chromatography-tandem mass spectrometry (LC-MS/MS) to analyze the secretome of conditioned media for 23 human can-cer cell lines spanning 11 cancer types, which enabled the iden-tification of both cancer-specific and pan-cancer serological biomarker candidates. Four of the candidates were validated experimentally, showing significantly elevated levels in the
(legend on next page)
serum or plasma of liver, lung, or nasopharyngeal carcinoma pa-tients relative to healthy controls. Despite the extensive informa-tion gained from these experimental investigations, there still exist a number of challenges that result in high variability and conflicting results among studies. For example, the use of cell lines is not an ideal representation of the in vivo system, culturing conditions can affect cell physiology and protein detection, there is a bias toward high-abundance proteins, protein concentra-tions span a large dynamic range in plasma, studies differ in sample collection and storage methods, and artifactual proteins are often identified, despite little or no relation to the disease in question (Geyer et al., 2017; Hanash et al., 2008; Kulasingam and Diamandis, 2008; Papaleo et al., 2017).