The conditioned media (CM) from the different available CRC cell lines is the most analyzed secretome sample. In addition to being xenograft tumor models for CRC, human colon adenocarcinoma cell lines, such as CaCo-2, LIM1215, HCT-116 and HT-29, are used as in vitro CRC model systems to analyze intracellular and secreted proteins from intestinal epithelial cells [30,31,32,33,34,35,36,37,38,39,40,41,42].

CaCo-2 spontaneously differentiates into mature enterocytes, and at confluence, forms a polarized monolayer with the specific functional features of differentiated absorptive epithelial cells [42]. LIM1215 represents a moderately differentiated CRC cell line but a metastatic one unlike CaCo-2 [43]. On the contrary, HCT-116 differentiation is to be modulated; interestingly, undifferentiated HCT-116 cells can be compared with their metastatic derivative counterpart to study the tumor phenotype [38,40]. Similarly to HCT-116, HT-29 cells under standard culture conditions are also undifferentiated, but upon differentiation induction they acquire a polarized morphology [44].

The majority of secretome studies are focused on one or two CRC cell lines; however, there are few integrative studies on the secretome analysis of multiple CRC cell lines recapitulating the differences in differentiation status and thus reflecting the biological variability within CRC [45,46].

CM collection from selected CRC cell lines, upon 70–80% confluency, is performed from serum-free cultures. After 24–48 h, depending on the experimental design, the CM is centrifuged, filtered and further processed for secretome analysis. A critical step is to obtain CM samples without intracellular protein contaminants, which are released in the media under cellular stress conditions. In this regard, following the CM collection, it is recommended to check the cell viability and/or the lactate dehydrogenase (LDH) levels: in fact, counting viable cells via conventional assays and measuring low LDH levels (if not its absence) indicate negligible cell death and insignificant cytoplasmic contamination in secretome preparations [44].

Alternatively, the hollow fiber culture (HFC) system could be used to reduce cell death and detect low abundance proteins in the large volumes of collected CM [47]. This system, composed of small fibers sealed in a cartridge shell, provides an in vivo-like 3D environment able to hold many cells in a small volume, thereby concentrating the secreted proteins. The HCF system has been successfully used for secretome analyses of CRC [38] and other solid cancers [48].

In comparison with CRC CM analyses, to the best of our knowledge, there are few reports on CRC tissue secretomes [31,49,50,51]. As stated above, tumor tissue secretome and particularly TIFs are promising sources of CRC biomarkers, represented by proteins secreted in vivo in the tumor microenvironment. The advantage of TIFs, in fact, is the reduction of the large dynamic range characterizing human plasma/serum, and the simultaneous enrichment of tumor-secreted proteins. After collection and washing of CRC tissue specimen samples together with their normal counterparts (adjacent normal colon mucosa), they are sliced into 1–3 mm³ pieces and incubated in physiological medium, such as PBS, at 37 °C for 1 h. After high-speed centrifugation, the supernatant represents the tissue secretome to be further analyzed [49].

Otherwise, freshly isolated CRC tissue specimens can be cultured ex vivo as explants and their relative secretome preparations are mainly composed of TIFs proteins secreted by tumor cells [52]. In this context, samples widely employed in discovering CRC biomarkers, are the CM of cancer-associated fibroblasts (CAFs) isolated from fresh surgical CRC specimens and adjacent normal tissues [39,53,54]. Such a sample is suitable for analyzing the autocrine factors from CRC cells and paracrine factors from surrounding stroma cells, mostly formed by fibroblasts. Interestingly, the interaction of stroma cells with tumor cells, underlying the carcinogenesis and metastatic process, is mediated by soluble proteins released in the CM of cell co-cultures.

In addition to CM from co-culturing CRC cells with stroma cells, Bozzi and colleagues performed a secretome analysis of CRC organoid cultures of peritoneal metastatic lesions from one CRC patient. In serum-free cell culture conditions, CRC organoids are suitable materials for the identification of secreted biomarkers involved in cancer stem cell (CSC) self-renewal and cancer cell proliferation [55].

To study the mechanisms underlying tumorigenicity and/or drug resistance in CSCs, the proteins secreted from the CM of stem cells were identified after their isolation from human CRC tumors; their expansion in derived cell cultures and in vitro differentiation were also explored [56].

A classical qualitative bottom-up proteomic gel-based workflow was used for identification of secreted proteins in the serum-free CM from different CRC cell lines and CAFs and in CRC tissues and their normal counterparts (Table 1). As for CM samples, after their collection, proteins are analyzed following a typical gel-based MS protocol (SDS-PAGE/in-gel tryptic digestion/nanoLC-MS/MS). As an example, by using this approach, Chen and colleagues analyzed two pairs of CAF/normal counterparts from two CRC patients, running each sample twice [54]. The combination of a limited number of biological/technical replicates coupled to a traditional gel-based proteomic approach, allowed to unambiguously identify only 230 proteins (at a false discovery rate of 1.3%). Moreover, the comparison of these data with a previously published dataset [53] produced poor overlapping, probably due to the biological variability among fibroblasts isolated from different CRC patients and also different experimental/technical conditions. On the contrary, a similar gel-based MS protocol performed on the secretome of CRC tissues and their normal counterparts led to the identification of 2703 unique proteins; the improvement in terms of uniquely identified species was related both to the sample type (tissue) and the availability of a suitable number of biological replicates (n = 4). Moreover, among the four biological replicates, the best overlapping in terms of commonly identified protein species was observed in the four CRC tissues rather than in the four patient-matched cancer-normal colon tissues, highlighting the importance of properly enrolling patients with matching clinicopathological features. Importantly, these identified proteins were then compared with those found in the secretomes of five typical CRC cell lines analyzed by the same authors and in the same experimental conditions of tissue samples [49]. This enlightened step clarified that secretory proteins identified in CRC tissue secretomes arose from neoplastic epithelial cells rather than the surrounding tumor environments [49].

In addition to the prefractionation by gel electrophoresis of CM samples, differential centrifugation ultrafiltration can be used to separate secretome components into molecular weight (MW)-based fractions (Table 1) [35,36]. Medium and high-MW (3–30 KDa and >30 KDa) secreted protein fractions were analyzed by the classical bottom-up proteomics gel-based workflows as described above. Interestingly, low-MW (1–3 KDa) soluble peptide fractions (peptidome) were subjected to a top-down proteomic approach [57]. All fractions from the bottom-up and top-down approaches were analyzed by nanoLC-MS/MS, and as expected, each approach showed distinct features in terms of identified protein/peptide species [36].

In addition to the reproducibility and robustness of secretome studies, label-free proteomic quantitative analysis guarantees elevated accuracy through spectral counting (SpC) and/or precursor ion intensity (total ion counts, TIC) methods (Table 1). To quantify all species present in secretome samples it is appropriate to perform both SpC and TIC methods as reported by Greening and colleagues; these approaches, in fact, complement each other because SpC and TIC procedures provide accurate estimations for high- and low abundance proteins, respectively [35].

As for reproducibility and accuracy of secretome analysis, the label-free LC-MS/MS shotgun quantitative proteomics can be considered a high-performing protocol; in particular, the CM from CRC cell lines is subjected to in-solution tryptic digestion and subsequently analyzed, at least in triplicate, using, in general, a bidimensional chromatography (cation exchange, SCX and reversed phase, RP) coupled to MS/MS.

As reported by Karagiannis and colleagues, this system led to 2979 unique proteins identified with a minimum of two peptides and FDR < 0.1%, and with a high reproducibility (>60%) among the triplicates [46]. Moreover, an innovative aspect of the quantitative approach performed by Karagiannis et al. regarded the introduction of an internal control, consisting of a panel of four known extracellular proteases. Their protein abundance was estimated both by ELISA and SpC: the statistically significant correlation among the two methods provided a validation of the label-free quantitative analysis [46].

In this context, label-free quantitative sequential windowed acquisition of all theoretical fragment ion (SWATH)-MS technology has been recently reported as a highly reproducible and sensitive method for CRC biomarker discovery [38,48]. SWATH-MS is a data independent acquisition (DIA) method that combines shotgun and selected reaction monitoring (SRM) analyses; it consists of the collection of high-resolution fragment ion maps of all detectable peptides present in the analytical sample. These peptides within specified precursor mass and retention time ranges are systematically unbiased fragmented [58]. SWATH-MS data analysis requires the in-advance creation of a spectral library in data dependent acquisition (DDA) of all the detectable peptides in the samples [59]. This technology was successfully applied to compare glycosecretomes of HCT-116 cell line with its metastatic derivative E1 [38]. Multilectin affinity chromatography (MLAC), in fact, is a suitable method to enrich glycoproteins into secretome samples, thereby enhancing the detection of low abundance CM-secreted proteins [38]. Recently, the study of glycosecretome has been gaining popularity because of the potential involvement of glycosylation modifications along the secretory pathway in intestinal epithelial cell differentiation. In this context, Link-Lenczowski and colleagues qualitatively and quantitatively analyzed chemical labeled-N-oligosaccharides from undifferentiated and differentiated CaCo-2 cells by using hydrophilic interaction liquid chromatography (HILIC)-high performance liquid chromatography (HPLC) and matrix-assisted laser desorption/ionization-time of flight-MS (MALDI-ToF-MS) approaches, respectively [42].

In addition to label-free approaches, metabolic (stable isotope labeling by amino acids, SILAC) [60,61] and chemical labeling (dimethyl labeling, iTRAQ) [62,63] were also applied for quantitative CRC secretome analysis (Table 2).

In the SILAC-based quantitative proteomics strategy, CRC cell lines are grown on heavy- (¹³C6-Lys/¹³C6-Arg) and light (¹²C6-Lys/¹²C6-Arg)-labeled medium under standard conditions until the confluency is reached. If two types or two different conditions of CRC cell lines are quantitatively compared, it is important to swap the labeled medium to avoid any bias in the labeling. As described above for CM collection, cells are washed and then cultured in the corresponding light- and heavy-labeled serum-free medium; CM are generally harvested after 48 h, concentrated and quantified. Protein extracts from heavy- and light-labeled cells are mixed in molar ratio of 1:1 and subjected to gel-based or shotgun proteomic analyses [32,33,40].

As for chemical labeling, the typical workflow includes the following steps similar to the metabolic one: (a) trypsin digestion of CM; (b) peptide labeling with heavy or light isotopes of a chemical compound; (c) mixing (1:1) of light/heavy peptide samples; and (d) proteomic analysis.

For all labeling methods, specific software, such as Proteome Discoverer and Maxquant, is used to calculate peptide ratios by comparing the intensities of the light- and heavy-labeled precursors at high resolution. In addition, light- and heavy-peptides unpaired in the mass spectra should be checked in the quantitative analysis [40].

In comparison to other classical quantitative proteomic approaches, SILAC-based strategy guarantees higher robustness and accuracy, but also flexibility with current experimental designs. In this case, however, an adequate number of experimental replicates improves the accuracy of quantitative data. The need for high reproducibility is a concept more and more crucial for secretome datasets compared to intracellular ones due to the lower abundance and complexity of secreted proteins to be quantified.

Given that the cancer secretome is a reservoir of tumor-associated antigens (TAAs), affinity strategy coupled to MS-based proteomics could be a promising approach for capturing immunogenic biomarkers directly from sera of CRC patients, but its intrinsic limit is related to the specificity of the antibody development. In fact, there has only been one recent report on the immune-affinity-MS approach for biomarker discovery in the CRC field [44]. Bukkhari and colleagues generated a polyclonal antibody repertoire against the secretome of HT-29 cell line. This affinity reagent (as immobilized anti-Sc antibody) was used for immunoprecipitate TAAs, whose identification in CRC patient sera compared to the normal sera was performed by a classical gel-based proteomic approach.

The most widely used bioinformatic tools for qualitative secretome analysis of MS/MS datasets are Mascot (http://www.matrixscience.com) [64], SEQUEST [65] and X!Tandem (https://www.thegpm.org/TANDEM/) [66].

In addition, to identify successfully degraded protein products, coming from cancer-related mutated species, and hence to provide a functional connection between genomic and proteomic data in cancer, specific bioinformatics resources, containing comprehensive variant proteins, need to be queried. For example, the Cancer Proteome Variation Database (CanProVar, http://canprovar2.zhang-lab.org) [67,68] contains more than 150,000 cancer-related variations and almost 10,000 cancer-related differentially expressed proteins associated with 26 cancers. Another freely-available tool is the Single Amino Acid Polymorphism Database (dbSAP, http://119.3.41.228:8080/dbSAP/index.html) [69] that integrates data from eight distinct databases and contains more than 16,000 unique variant peptides supported by more than 400,000 MS spectra. Finally, it is possible to query the database of differentially expressed proteins in human cancer (dbDEP 3.0, https://www.scbit.org/dbdepc3/) [70] based on MS data that contains information on amino acid variations, post-translational modifications and drugs.

Another major step during the experimental design of secretome studies is the biological and functional characterization of identified secreted proteins. Subsequent to MS analysis, secretome protein datasets are analyzed by using different bioinformatic tools to gain insights into their structural and functional annotations.

Firstly, secretory features of identified proteins are investigated by using SignalP (http://www.cbs.dtu.dk/services/SignalP/) [71], SecretomeP (http://www.cbs.dtu.dk/services/SecretomeP) [72] and TMHMM (http://www.cbs.dtu.dk/services/TMHMM) [73] tools; they are able to predict whether a protein is classically (SignalP) or non-classically (SecretomeP) secreted and presence of transmembrane helices (TMHMM). For each tool, specific cutoffs have to be considered in order to classify a protein as secreted species; otherwise, for those of exosomial origin, their putative protein annotations can be determined by querying databases such as ExoCarta (http://www.exocarta.org) [74] and Vesiclipedia DB (http://www.microvesicles.org/) [75].

In addition to determination of secretory pathways among the experimental dataset, clustering of secreted species into functional annotation terms is usually conducted in secretome studies. To accomplish this, over-represented proteins are identified in datasets through enrichment analysis in Gene Ontology (GO) annotations for molecular function and biological process. GO information is retrieved from different databases such as IPA database (Ingenuity^® Systems, www.ingenuity.com) [76] and UniProt DB (www.uniprot.org) [77] and from GO resources (http://geneontology.org) [78,79], and is then analyzed by using several bioinformatic tools such as DAVID (http://david.abcc.ncifcrf.gov/) [80,81], GeneMania (http://www.genemania.org/) [82], PANTHER (http://www.pantherdb.org) [83,84] and BiNGO (http://www.psb.ugent.be/cbd/papers/BiNGO/Home.html) [85]. Predicted protein–protein interactions are investigated using STRING (www.string-db.org) [86], whereas Cytoscape (www.cytoscape.org) [87] tool is used for network analysis.

The identification of over-/under-represented proteins in CRC secretome versus control by using bioinformatic tools, together with the overlap analysis with previously published similar datasets, allowed the selection of potential biomarkers for early CRC detection.