In order to study the secretome of undifferentiated and differentiated cells, Caco-2 cell cultures were set up and allowed to differentiate as described before [3]. This was done in several parallel flasks, and repeated three times. The main advantage of this model, when compared with direct plasma or serum analyses, is that here only what the malignant cell (undifferentiated) or its counterpart (differentiated) secretes to the medium is detected, while in serum/plasma studies many detected proteins may not have been secreted by the malignant cells. On the other hand, the main drawback of this system is that cultured cells are serum-starved for 24 h, which induces stress on cells that become prone to spontaneous autolysis, resulting in non-specific release of intracellular proteins [26]. This could obscure the interpretation of results, though this phenomenon should affect both culture types and thus the non-specifically released proteins may not be detected as differential between them.

In the three experiments, we counted an average of 7.4 ± 1.6 million cells in the undifferentiated flasks, and 10.6 ± 2.2 million cells in the differentiated ones. When the secretome of these cells was obtained, the amount of protein recovered was 196.4 ± 10.4 μg for the undifferentiated cells, and 262.9 ± 17.1 μg for the differentiated ones. The secretome was visualized for both types of cells by monodimensional electrophoresis (Figure 1a) which demonstrates some differences in the secreted proteins patterns between undifferentiated and differentiated Caco-2 cells. To corroborate the differentiation process, we measured the specific activities of the enterocytic enzymes alkaline phosphatase and maltase after differentiation in three independent experiments, observing an average 6-fold and 14-fold increase, respectively (Figure 1b).

The DIGE technique was used for the analysis of the secretome of Caco-2 cells, before and after differentiation, in the three independent experiments. Following the DIGE experimental design, we combined one undifferentiated and one differentiated sample, plus a pooled internal standard, per gel. Therefore, we analyzed nine samples in three gels. In Figure 2, we show a representative image from one such gel. The samples were randomly distributed as stated in Table 1.

After acquiring the three images of each gel, the protein spots were detected and aligned by comparison with the internal standard. The number of spots detected and matched in each gel is shown in Table 1. More than 1600 distinct protein spots were detected in each gel. This result is in agreement with those described for other CRC cell lines. As an example, more than 1000 distinct spots have been detected in the secretome of SW480 cells [24], whereas the secretome of the cell line LIM1215 allowed detection of approximately 2000 spots by DIGE [27].

For the quantitative comparison, the abundance of each spot was made relative to the total amount of protein secreted by the number of live cells in the corresponding sample, and only the 919 spots detected in all the samples were considered. Applying the Mann-Whitney U test, we found 96 spots with significantly different abundance (p ≤ 0.05) in the secretome of Caco-2 undifferentiated and differentiated cells. From those spots, we were able to locate and cut 34 out of a preparative gel, and those were analyzed by mass spectrometry. Eventually, we identified the 26 spots highlighted in Figures 2 and 3, and summarized in Table 2. Four of them were identified as potential contaminants, and therefore not included in Table 2 (spot 812: mixture of human cytokeratin 1, keratin 9 and type I keratin 16; spots 819 and 821: bovine albumin; spot 1606: bovine apolipoprotein A–I). As an example of the variation in abundance, some of the identified spots are shown in Figure S1, where 3-D images have been drawn on the basis of the spot abundance after normalization. Noticeably, some of the proteins were identified in several different spots, probably due to their bearing different post-translational modifications.

Finally, we applied the multivariate test PCA on the spot abundances. When PCA was applied to the 919 spots detected in all samples, it allowed the discrimination of the secretome of the undifferentiated cells from that of the differentiated cells. The first three principal components (PCs) explained 85.5% of the variability between groups; PC3 was able to distinguish day 6 and day 20 samples (p ≤ 0.05 by Mann-Whitney U test) (Figure 4a). Interestingly, both PC1 and PC2 could separate the samples per gel, that is, samples run on the same gel obtained similar values for those PCs. This highlights the importance of using an internal standard to allow matching between different gels, and thus correcting for technical variations. PC1 and PC2 accounted for 70% of the variance, whereas PC3, responsible for the difference between sample groups, accounted for 15.5% of the variance. When the PCA was repeated with only the 96 spots found significantly altered after differentiation, we found that the first PC was significant and explained by 64% of the variance of the data, giving a neat separation between the cell states, as shown in Figure 4b. Even with the 22 identified spots, the first PC was significant and was able to explain 65% of the variance, separating samples from both undifferentiated and differentiated secretome, as shown in Figure 4c.

Among the proteins differently secreted by the Caco-2 cells before/after differentiation, we found enzymes involved in the glycoprotein metabolism (glucosidase II), carbohydrate metabolism (triose phosphate isomerase, NADP-dependent isocitrate dehydrogenase), energy metabolism (ATP synthase subunit beta, creatine kinase B), and detoxification (glutathione S-transferase omega-1), as well as lipid transport proteins (apolipoproteins A-I and A-IV, apolipoprotein E precursor) and proteins involved in hemostasis/tissue homeostasis (fibrinogen gamma chain). Considering the functions developed by those proteins, and the fact that alterations in their expression could be related to different physiological situations (as inflammation, starvation, exercise, etc.), they are likely to show low specificity as CRC biomarkers. Nevertheless, it is interesting to notice some of these proteins had been related before to the differentiation of Caco-2 cells, in particular by Stierum et al.[5] during the comparison of cellular lysates of Caco-2 cells before and after differentiation. In the case of the ATP synthase beta, these authors could not determine the direction of the change, while we found a higher amount (6.8 ratio) of the secreted form in undifferentiated cells. For the gluthathione S-transferase, Stierum et al. observed an up-regulation of the cytoplasmic form of the glutathione S-transferase A1, while we detected the form omega-1 with a 1.6 times higher amount in the secretome of undifferentiated cells. Higher secretion in the undifferentiated cells could be compatible with a smaller amount of protein retained in the cytoplasm in these cells, and therefore with increased intracellular expression in the differentiated cell lysates, as Stierum et al. reported. Another protein detected by these authors was the creatine kinase B, for which we found a higher amount of secretion in differentiated cells (ratio undifferentiated/differentiated = 0.3), while Stierum et al. found an upregulation followed by downregulation as the cells became differentiated. Finally, the triose phosphate isomerase was also identified in the lysates analyzed by Stierum et al., though its change was not determined. In our study, we found higher secretion of this enzyme in the differentiated cells.

In contrast to the proteins mentioned above, the translationally-controlled tumor protein, the transforming growth factor-beta-induced protein ig-h3, and the Niemann-Pick disease type C2 protein, seem to be the most interesting candidates as potential CRC biomarkers.

The translationally-controlled tumor protein (TCTP) was identified from one spot overexpressed in the secretome of undifferentiated cells, showing an average increase of 3-fold. This protein was first described in mouse tumor cells as a growth-related protein [28]. Despite its name, it is a highly conserved and ubiquitously expressed protein in all eukaryotes, highlighting its important role in the cell [29]. Although the TCTP functions are not well defined, previous studies demonstrated it is implicated in histamine release [30], calcium binding [31], and that it is anti-apoptotic [32]. Nowadays TCTP is described as a multifunctional protein that plays important roles in cell proliferation, immune response, tumorigenicity, and cell death, including apoptosis [29]. TCTP is secreted by different cell types, as macrophages [33] or human embryonic kidney cells [34]. Regarding CRC, the level of TCTP mRNA detected in three human colon carcinoma cell lines (SNUC2A, SNU-C4, and SNU-C5) suggests that a high TCTP mRNA expression in these cells could be related to the rapid cell growth and, therefore, a high potential of tumorigenesis [35]. Moreover, knockdown of TCTP inhibited proliferation, migration, and invasion activities in human colon adenocarcinoma LoVo cells [36]. Related to the cell line used on this work, Stierum et al.[5] observed the same trend on TCTP expression when comparing undifferentiated and differentiated Caco-2 cell lysates as we have now found in their secretome.

The transforming growth factor-beta-induced protein ig-h3 (TGFβIp) was identified from three different spots, all of them overexpressed in the undifferentiated Caco-2 cells. These different isoforms could be due to post-translational modifications, since at least 29 sites of vitamin K-dependent carboxylation have been described and annotated for this protein (http://www.uniprot.org/uniprot/Q15582), as well as several phosphorylation sites (http://www.phosphosite.org/proteinAction.do?id=3567117). TGFβIp, originally known as βig-h3, was first cloned as a TGF-β-induced gene on a human lung adenocarcinoma cell line [37]. It localizes to chromosome 5 (5q31) and consists of 17 exons that span around 34 kb [38]. The protein has 683 amino acids and contains a secretory signal. TGFβIp is not only induced by TGF-β, and other factors may be involved on its regulation. It appears in the extracellular matrix associated with collagen, fibronectin, laminin and glycosaminoglycans, and it supports the adhesion of many cell types by recruiting integrins [39]. This protein has been described both increased and decreased in tumors, but in particular it has shown an increased expression in CRC [40–42]. TGFβIp could be involved in multiple aspects of tumorigenesis, including tumor progression, angiogenesis and metastasis, though its role is not yet clear. In colon cancer it has been demonstrated that ectopic expression of TGFβIp enhanced the aggressiveness and altered the metastatic properties of colon cancer cells in vivo, whereas inhibition of its expression dramatically reduced metastasis. Mechanistically, it appears to promote extravasation, a critical step in the metastatic dissemination of cancer cells [43]. Besides, TGFβIp levels on serum have been found to be elevated in pancreatic cancer [44].

The Niemann-Pick disease type C2 protein (NPC2) appeared as one spot with a 6-fold average elevation in the secretome of undifferentiated cells. NPC2 is a cholesterol-binding glycoprotein whose encoding gene is located in chromosome 14 [45]. It consists of a 151-amino acid sequence, and the mature protein has a molecular weight between 17 kDa and 20 kDa. NPC2 is highly conserved among major species, and it is present in fluids such as milk, epididymal fluid, bile, and plasma [46–48]. NPC2 binds cholesterol with high affinity and was implicated in mediating intracellular cholesterol trafficking through the late endosomal/lysosomal compartment [49]. In addition, it has been shown that this protein plays an important role in the regulation of hematopoiesis [50] and immunity [51]. Recently, its implication in adipocyte differentiation [52] and fibroblast activation in disease pathogenesis [53] has been also described. To date there are no reports about any relationship with CRC, but an increased expression of the mRNA of this protein has been described in melanoma [54].

Noticeably, the proteins shown in Table 2 are either annotated in curated databases with a signal peptide, or can be predicted by computational methods to be secreted through non-classical pathways. The only exceptions are two proteins which nonetheless have been described in secretions by other authors. In particular, the creatine kinase B has been described in exosome-like vesicles released by intestinal epithelial cells [55] and in the conditioned medium of prostate cancer cell lines [56], whereas the glutathione S-transferase omega-1 has been described in the secretome of different cell lines including the colon cancer cells SW480 and SW620 [25].

Overall, the strategy we describe in the present study serves as a method for identifying proteins specifically altered in relation to colon tumorigenesis, and further studies are needed to validate their utility in serum as biomarkers for the disease.

The Caco-2 human colon adenocarcinoma cell line was purchased from the European Collection of Cell Cultures (ECACC No: 86010202). For routine growth we cultured them in 25 or 75 cm² plastic flasks containing Dulbecco’s modified Eagle’s medium (DMEM) supplemented with 4.5 g/L glucose, 20% fetal calf serum (FCS), 100 units/mL of penicillin, 100 μg/mL of streptomycin, 2 mM l-glutamine, and 1% non-essential amino acids at 37 °C, in a 5%-CO₂ humidified atmosphere. Cells were sub-cultured after reaching 90% confluence by treating them with a solution of 0.05% Trypsin/0.02% EDTA in PBS. The number of cells per flask was measured by counting them on a Neubauer chamber. Trypan blue was used as a stain to distinguish between dead and living cells.

For differentiation experiments, pre-confluent cultures were disrupted with trypsin, cells were seeded on 75 cm² flasks at 12,000 cells/cm². Medium was changed every day after 48 h of seeding. Considering the descriptions of Pinto et al.[3] and Rousset et al.[4], and the values of differentiation marker enzymes, we selected day 6 as the undifferentiated state and day 20 as the differentiated one.

The process of enterocytic differentiation was evaluated through the assay of marker enzymes as the alkaline phosphatase and the maltase. For these assays, cells were harvested by scraping, washed with PBS, pelleted and re-suspended in ice-cold ultrapure water with protease inhibitors (Complete Mini tablets, Roche). Re-suspended pellets were placed on ice for 5 min, homogenized with a Potter-Elvehjem, and then frozen at −80 °C.

The alkaline phosphatase activity (EC 3.1.3.1) was measured with the method described by Engstrom [57]. As substrate we used 16 mM para-nitrophenyl phosphate in pH 9.8 50 mM sodium borax buffer plus 1 M MgCl₂; 200 μL of substrate were mixed with 50 μL of sample and incubated at 37 °C for 30 min. The reaction was stopped by adding 0.6 mL of 0.25 M NaOH. The mixture was centrifuged at 700 g for 5 min and the optical density (OD) at 410 nm was measured.

The maltase activity (EC 3.2.1.20) was measured following the method of Dahlqvist [58], where maltose served as the substrate. For the assay, 40 μL of sample where incubated with 40 μL of 55 mM maltose in 0.1 M maleate buffer pH 6.0 for 1 h at 37 °C. The glucose generated from the maltose was measured using a commercial enzymatic test (Spinreact) based on the method described by Trinder [59,60].

Protein concentrations were determined using the Bradford microassay [61] and results were expressed as specific activity.

To study the extracellular medium, cells were kept in FCS-free medium for 24 h before collection [62]. After 24 h, medium was collected and centrifuged at 600 g for 5 min, so that particles on suspension were discarded with the pellet. The medium was then dialyzed for 24 h against ultrapure water (cut-off: 12–14 kDa), freeze-dried on a Christ Alpha 2–4 equipment and kept at −80 °C.

For monodimensional electrophoresis, 10 μg of total protein were separated in 10% (v/v) polyacrylamide (30% T, 2.6% C) denaturing minigels [63]. Gels were stained with Coomassie brilliant blue for protein visualization.

Dried medium was reconstituted on ultrapure water and cleaned using 2-D CleanUp Kit (GE-Healthcare). Pellets were diluted on lysis buffer (8 M urea, 2 M thiourea, 4% (w/v) CHAPS and 30 mM Tris) and the protein concentration was measured using a modified Bradford method [64]. An internal standard was prepared by pooling equivalent amounts of protein from each of the six samples included in the experiment. Stock solutions for the cyanine dyes were reconstituted in N,N′-dimethylformamide (DMF) keeping the proportion CyDye:DMF at 1:1.5 to obtain the working solution. Cy2 was used for labeling the internal standard, while Cy3 and Cy5 were randomized between undifferentiated and differentiated samples (Table 1). For each sample, 50 μg of protein were labeled with 200 pmol of the correspondent Cy dye in ice for 30 min. Labeling was terminated by adding 1 μL of 10 mM lysine and incubating 10 min in ice.

For the 2-DE separation, a mixture of the Cy2-, Cy3- and Cy5-labeled samples (150 μg), was applied on pH 3–11, 24-cm non-linear Immobiline Dry Strips (GE Healthcare) by cup loading. Isoelectric focusing (IEF) followed this program: 1 h at 120 V, 2 h at 500 V, 2 h at 1000 V, 16 h at 5000 V, and a final holding step of 500 V for no longer than 10 h using the Ettan IPGphor (GE Healthcare). After IEF, strips were equilibrated for 12 min with 6 M urea, 100 mM Tris, 30% (w/v) glycerol, 2% (w/v) SDS and 0.5% (w/v) DTT; then for 6 min with 6 M urea, 100 mM Tris, 30% (w/v) glycerol, 2% (w/v) SDS and 4.5% (w/v) IAA. The second dimension was run on 10% polyacrylamide gels (30% T; 2.6% C) for 17 h at 2 W/gel.

Three images per gel were acquired with the Typhoon 4900 Imager (GE Healthcare) with the appropriate excitation and emission wavelengths for each of the three dyes. Protein patterns for each sample were then analyzed with the software SameSpots (GE Healthcare). Abundances were given as normalized intensity values corrected by a factor consistent in the total protein quantity of the secretome (μg) divided by the number of cells (in millions) in it, which allowed us to compare the amount of protein secreted by each cell instead of just its proportion in the total pool of proteins.

In order to choose the spots which expression showed significant abundance variation, we applied the non-parametric Mann-Whitney U test. Multivariate studies were done through principal component analysis (PCA) as described before [65]. All analyses were performed with SPSS (release 16) and p values ≤ 0.05 were considered statistically significant.

For the proteins identified, we investigated if they had a classical signal peptide annotated in www.uniprot.org. Besides, we employed the Secretome 2.0 Server to predict secretion through a non-classical route [66].

Selected spots were cut out from Coomassie-stained gels, reduced, alkylated and in-gel digested with 10 μg/mL trypsin in 25 mM ammonium bicarbonate at 37 °C overnight. Peptides were eluted with 5% (v/v) trifluoroacetic acid and 75% (v/v) acetonitrile. Samples were mixed with a matrix of 5 mg/mL α-cyano-4-hydroxycinnamic acid in 50% (v/v) acetonitrile. The mixture was analyzed by matrix-assisted laser desorption/ionization-time of flight (MALDI-TOF), in a 4800 Proteomics Analyzer (Applied Biosystems) by peptide mass fingerprinting (PMF). When proteins were not identified by this method, peptides were further fragmented (MALDI-TOF/TOF) and analyzed by sequence query (SQ). Database searching was done with the MASCOT Daemon search engine (Matrix Science) against the human nrNCBI database. Search parameters were stated as follows: enzyme trypsin; allowance of 1 missed cleavage; propionamide as fixed modification; methionine oxidation as variable modification; 50 ppm of peptide mass tolerance for PMF; 100 ppm peptide mass tolerance and 0.3 Da fragment mass tolerance for SQ. Protein hits were considered significant when the protein score was above the p ≤ 0.05 threshold, with at least two significant peptides.