Colorectal cancer (CRC) is a cancer of epithelial origin, localized to the large intestine and rectum. It is one of the most common cancers and will occur at some stage in approximately 5% of the population of the western world. The conventional prognostic factors for patient survival are histologic tumour grade (differentiation) and tumour stage (TNM, tumors/nodes/metastases, stages I–IV) [1,2], which is based on depth of tumour invasion, involvement of regional lymph nodes and metastatic spread to other organs [3]. If metastasis has occurred, patient 5-year survival after surgery falls dramatically from 90% to less than 10% [4]. It is therefore important to increase our understanding of the molecular changes leading to development, spread and metastasis of CRC and to identify potentially prognostic and predictive markers for the disease.

Despite the discovery of a range of intra- and extra-cellular protein biomarkers for CRC and continuing efforts to discover potentially prognostic and predictive markers using various approaches [5–12], the translation of this increasing volume of differential proteomic data into patient care remains a major challenge. This is partly due to the heterogeneous nature of CRC [13,14] and the complexity of processes involved in its development and spread, but also to challenges presented in detecting and characterising glycoproteins and low abundance proteins [15,16] in complex mixtures. In addition, validation of candidate proteins as prognostic markers is time-consuming and expensive, and recent critical reviews of differentially expressed proteins from comparative proteomic studies have suggested that some reported proteins may be associated with general cellular stress responses, rather than being disease-specific biomarkers [17,18]. The development of a reliable assay for determining prognosis and guiding post-surgical therapy therefore remains elusive.

The metastatic potential of CRC may be already encoded in the primary tumour [19], and molecular staging using a classifier based on 43 genes has identified patient prognosis more accurately than the traditional clinical staging, particularly for intermediate stage II and III patients [20]. It may be possible to identify disease signatures based on protein expression profiles in primary CRC tissues that reflect potential for disease progression and metastasis, and subsequently use these signatures to predict patient recovery and survival after surgical removal of the primary CRC tissue. With this goal in mind, we have developed a 122-antibody microarray (DotScan™ CRC microarray; Medsaic Pty Ltd, NSW, Australia) for immunophenotyping live cells from fresh disaggregated CRC tissues [21,22]. Microarrays prepared as 10 nl antibody dots on nitrocellulose-coated microscope slides (FAST; Grace Bio-labs, Bend, OR, USA) bind cells with corresponding surface molecules, producing a binding pattern that reflects the surface immunophenotype of the mixed cell population. Bound cells are then fixed to the microarray with formalin, and specific sub-populations are profiled by multiplexing with mixtures of soluble fluorescently-labelled antibodies. Hierarchical clustering of binding patterns for these sub-populations of cells yields patient clusters that may correlate with disease stage, tumour invasiveness, differentiation, drug susceptibility or patient outcome. This antibody-based multiple marker approach has already been validated for the classification of a range of common human leukaemias and lymphomas [23].

The evolution of CRC from an adenomatous polyp to metastatic disease is dependent not only on progressive accumulation of genetic and epigenetic abnormalities [24], but also on the complex interactions of different sub-populations of cells within the tumour microenvironment (cancer cells, normal stromal cells, infiltrating leukocytes and the soluble factors they produce) [25]. The importance of cell surface molecules for cell-cell adhesion and communication between cancer cells and non-malignant cells has been recently reviewed [26,27], and the significance of signalling pathways involved in CRC progression as a result of these interactions has also been described [28–32]. Since many drugs used in cancer therapy target cell membrane proteins [33], this review seeks to summarise current knowledge on cell surface molecules that may serve as therapeutic targets, have prognostic significance or provide important information about the invasive, metastatic and proliferative potential of cancers of the large intestine and rectum. Antibodies to some of these proteins have been incorporated into the current DotScan™ CRC antibody microarray; others could be added in the future. Readers are also referred to several previous reviews on markers of metastasis and prognosis in CRC [34,35].

Recently, the importance of the tumour microenvironment for the development, promotion and spread of cancer has become clear [36], and the potential for using tumour stroma-derived biomarkers in prognostication and therapy of cancer patients has been realised [37]. The stroma consists of the non-malignant cells of the tumour—fibroblasts, infiltrating leukocytes and cells of the blood vessels. However, despite experimental evidence for the importance of tumour stroma for cancer development and progression, clinical applications for stromal biomarkers remain limited. Down-regulation of TGF-beta receptor 2 (TGF-β-R2) in tumour-associated stroma correlates with a worse prognosis [38], while expression of platelet-derived growth factor receptor (PDGFR) on stromal cells correlates directly with metastatic potential and advanced-stage disease in CRC [39,40].

Stromal CXCR4 (CD184) and CXCL12 expression is associated with distant recurrence and poor prognosis in rectal cancer after chemotherapy [41]. Patients whose colon tumours have high levels of stromal fibroblast activation protein (FAP) are more likely to have aggressive disease progression and potential development of metastases or recurrence [42]. Urokinase receptor (uPAR; CD87) expression in stromal fibroblasts has been implicated in the occurrence of haematogenous metastasis (dissemination via the blood) of CRC [43]. The presence of high levels of stroma and down-regulation of Deleted in Pancreatic Cancer 4 (SMAD4) predict worse survival for early-stage CRC patients [44]. Expression of CD10 (a cell surface metalloprotease) on stromal cells adjacent to cancer cells within the area of invasive growth is significantly higher in invasive CRC than in non-invasive CRC or adenomas [45]. The stromal expression of CD10 also appears to be significantly associated with abnormal accumulation of nuclear p53 in tumour cells and a larger tumour size, suggesting that CD10 expression may contribute to tumour invasion and possibly facilitate metastasis [45].

Thrombospondin-1 (TSP-1), mainly localised in fibroblasts of the CRC stroma [46], has been considered as an important negative-regulator of tumour angiogenesis. TSP-1 expression in primary CRC tumours has been correlated inversely with metastatic potential and prognosis [47–51]. In CRC liver metastases, however, TSP-1 expression has been linked to significantly poorer survival, suggesting that TSP-1 in liver metastases may have an alternative mode of action, facilitating tumour invasion rather than acting as an anti-angiogenic growth factor [51].

Inflammatory reactions in tumours can have a positive influence on survival [52,53] or alternatively can be associated with development of metastases and disease progression [54–56]. The role of tumour-associated macrophages (TAM) in tumourigenesis is complex because TAMs can either prevent or promote tumour development. While Forssell et al. [52] showed that a dense TAM infiltration at the tumour front positively influenced prognosis in colon cancer, strong pro-tumourigenic effects of TAMs in CRC have also been reported [54–57]. It has been suggested that the balance between pro- and anti-tumourigenic properties of TAMs may depend on their interaction with cancer cells, other stromal cells, and the tumour microenvironment [54] or be influenced by the degree of cell-cell contact [52].

The presence of high numbers of mature CD208+ infiltrating dendritic cells (DC) in the tumour epithelium has been associated with shorter overall patient survival, while patients with high numbers of CD1a+ DCs in the advancing margin of the tumour appear to have a shorter disease-free survival [58]. These findings are controversial, however, as a strong correlation has also been found between the presence of DCs together with high density T-cells and no detectable metastases [59] and better patient survival [60].

The presence of large numbers of tumour infiltrating lymphocytes (TILs) at the invasive tumour front has generally been associated with a good prognosis for CRC patients [61–64]. Patients with high levels of CD3+, CD8+, CD25+, CD45RO+ or CD95L(CD154)+ TILs appear to have a better clinical course [65–69] than those with low levels, particularly in early disease stages [70]. The importance of the intra-tumoural location, immunophenotype and density of these infiltrates has been demonstrated [71–73].

It has been shown that CD3+ TILs located in the tumour stroma are not as strong predictors of prognosis as T-cells located within cancer cell nests (clusters), consistent with the requirement for direct contact between target and effector cells [71–74]. In addition, although the presence of high density CD3+ TILS in primary tumours has been associated with good prognosis in patients with lymph node-negative CRC, there was no significant correlation between high density CD3+ TILs and absence of post-surgical metastases in patients with lymph node-positive CRC [75].

Patients with T-cells showing low CD4+/CD8+ ratios have a better clinical course than those with high ratios [59,76]. The infiltration of cancer cell nests by CD8+ T-cells has been associated with better survival [67], supporting the importance of activated cytotoxic CD8+ T-cells in the anti-tumour response to CRC [77,78].

The better overall survival of CRC patients with high infiltrates of CD3+ (total) and CD8+ (cytotoxic) T-cells within cancer cell nests, was recently confirmed [70]. Interestingly, however, this correlation was not significant for the rectal cancer patient sub-group when it was analysed separately from the colon cancer group. Furthermore, in patients whose tumours were positive for microsatellite instability (MSI+), a hallmark of DNA mismatch repair (MMR) deficiency [79], no association was found between high levels of activated cytotoxic T-cells and a favourable prognosis. Although these findings are in agreement with those of several other groups, they are in conflict with a number of studies in which high infiltrates of CD8+ TILs correlated with good prognosis regardless of MSI-status [70].

Marked infiltration of CD8+ and CD57+ natural killer (NK) or NK-like T-cells in the advancing tumour margin has also been associated with longer disease-free survival, and this was most evident in MSI+ tumours [80]. Metastatic spread of CRC is associated with decreased NK-cell activity [81], but it remains speculative whether decreased NK-cell activity precedes the development of metastases and may therefore be prognostic.

Koch et al. [82] reported strong and significant enrichment for CD4+ T helper cells in CRC compared to corresponding normal mucosa, i.e., decreased proportions of CD8+ cells in total T-cell populations of CRC. However, a significantly higher proportion of the CD8+ TILs in CRC expressed markers of activation (CD69+) and cytotoxic activity (CD107a+) compared with normal mucosa. Significantly, the proportion of activated CD8+ TILs decreased with tumour stage [82].

If a tumour exhibits a high T-cell infiltrate, the prognosis for patients with strong tumour expression of major histocompatibility antigen class I (MHC 1) is better than for those who have weak MHC 1 expression [74], because the immune response mounted against the tumour cells is weaker in these patients.

TILs recovered from primary tumours showing no sign of metastasis differ immunophenotypically from those showing early signs of tumour metastasis (presence of vascular emboli, lymphatic invasion and perineural infiltration). Using flow cytometry, Pages et al. [68] found that TILs from tumours with no sign of metastasis showed higher expression of markers of cell adhesion (CD62L, CCR7, CD103, CD49d and CXCR3), activation (HLA-DR, CD98, CD80, CD86 and CD134) and differentiation (CD45RO, CD45RA, CD27, CD28, CCR7 and CD127). Primary CRC tumours without early signs of metastatic invasion such as vascular emboli (i.e., infiltration of cancer tissue into blood vessels) presented with significant increases in T-cell sub-populations from early memory (CD45RO+CCR7- CD28+CD27+) to effector memory CD8+ T-cells (CD45RO+CCR7-CD28+CD27-). Using immunohistochemistry-based tissue microarray analysis of 415 tumours, they confirmed that a high density of memory T-cells (CD3+CD45RO+) in the primary tumour correlated with the absence of early signs of metastatic invasion, confirming previous findings [83]. In a comparison of patients with high T-cell infiltration, primary tumours from patients with metastases had significantly decreased densities of CD8+ T-cells and effector memory T-cells (CD27-CD45RA-) than patients without metastases [59].

The surface antigen expression of total TILs differs from that of T-cells in peripheral blood, with markedly lower expression of CD29, CD49d and CD49f, and slightly higher expression of CD49a and CD49b [84]. CD11a and CD18 are reduced on CD8+, but not CD4+, sub-populations [84]. TILs also differ from lymphocytes within normal colon interstitium by lacking CD28, CD56, CD98 and CD154 [85] and showing altered expression of CD3, CD29, LFA-1 (CD11a/CD18) and LFA-3 (CD58) [86].

The function and relative importance of different CD4+ sub-populations in CRC is still only partially understood. Although CD4+ TILs are observed in CRC [59,80,87], the density of this subInt. population has shown little prognostic value in CRC to date, and is not significantly different in primary tumours of patients with metastatic or non-metastatic disease [59]. A high density of CD4+ TILs in CRC liver metastases has correlated with poorer patient outcome [63].

Recent attention has focussed on CD4+CD25+ regulatory T-cells (Treg), which are characterised by expression of nuclear transcription factor forkhead box P3 (FoxP3). The CD3+/FOXP3+ cell ratio [88] and the CD8+/FOXP3+ cell ratio [89] are both predictive markers for disease-free survival time and overall survival time in patients with CRC, supporting the down-modulatory role of Treg in the presence of protective cytotoxic T-cells. Controversially, the presence of a high density of FoxP3+ Treg in CRC has also been associated with improved survival in stage II and stage III CRC patients [66,83] and in relapsed CRC patients undergoing chemotherapy or chemoimmunotherapy [90]. Although this result is unexpected and counterintuitive, it has been suggested that the increase in Tregs in tumour tissue may represent a feedback response to a pre-existing cancer cell-directed immune response [90]. Frey et al. [91] found that high FOXP3(+) Treg density was associated with early tumour stage and independently predicted improved disease-specific survival in MMR-proficient CRCs, but not in MMR-deficient CRCs, though high Treg density correlated with absence of lymph node involvement and vascular invasion in the latter. They suggested that the role of Treg may differ according to the clinical stage and the MMR status of CRC.

We have prepared live cell suspensions from enzymically disaggregated primary tumour tissue from CRC patients and corresponding normal mucosa. DotScan™ CRC antibody microarrays and fluorescence multiplexing were used to profile the cells, and the CD3+ subset showed differential expression of HLA-DR, TCRα/β, CD49d, CD52, CD49e, CD5, CD95, CD28, CD38 and CD71 in descending order of difference [21].

Cancer stem cells have been identified in several human malignancies, including CRC, and are believed to drive tumour initiation, progression and metastasis [233–238]. Tumour buds, discussed above, appear to have many features of CSC, and the possibility that a sub-population of tumour buds may represent malignant stem cells is an open question requiring further investigation [133].

The presence of a sub-population of CD26+ CSC in primary CRC tumours was found to predict distant metastases on follow-up [239]. These cells were associated with enhanced invasiveness and chemoresistance. High EpCAM expression was also associated with oncogenic potential in CSC [122], and CD166 (ALCAM) was found to be differentially expressed on EpCAM^high/CD44⁺ CSC [240]. Other CSC markers, CD133 and CD24, correlated with invasiveness and differentiation in CRC, while CD44 expression was related to tumour burden [241]. No correlation was found between the expression of these three markers and patient survival [241], but others have shown that the combined evaluation of CD133, CD166 and CD44 could be used to identify high and low survival among stage I and II CRC patients [242]. Although CD133+ cells have demonstrated CRC initiating potential [243], CD133 expression may not be restricted to cancer-initiating cells [244]. In addition, highly metastatic CD133- sub-populations of cancer-initiating cells have been identified in a human CRC xenograft model, suggesting that CD133+ tumour cells might give rise to a more aggressive CD133- subset during the metastatic transition [244].

CD200 (OX2) is a membrane glycoprotein that induces an immuno-regulatory signal through its receptor (CD200R), leading to the suppression of T-cell-mediated immune responses. CD200 has been associated with tumour progression and poor prognosis in several cancers [245,246] and is over-expressed in CRC with markers for CSC [247], suggesting that tumour-initiating CSC may evade immune system detection through CD200-induced suppression of T-cell-mediated immune responses.

A number of proteomic studies on CRC tissues have identified differentially expressed proteins from whole tissue samples of CRC and paired normal mucosa [8,16,248–252]. In other studies, enrichment for specific sub-populations of cells has been performed using techniques such as laser micro-dissection [253,254], fluorescence activated cell sorting (FACS) [240] or antibody-conjugated magnetic beads [240,255]. Interpretation of results is complicated by the fact that different methods have been employed for sample preparation, antigen detection and protein quantification. For antibody-based detection methods, the choice of monoclonal antibody is important, as differences in staining may reflect changes in epitope structure due to mutations, alternative splicing or post translational modifications, rather than up- or down-regulation of protein expression. In addition, some changes in protein activities during disease progression may correlate with altered sub-cellular localisation of proteins [120]. Thus the use of different methods of protein quantification may lead to conflicting results.

Methods used for membrane proteomics in cancer biomarker discovery have been previously reviewed [26]. Due to the hydrophobicity of membrane glycoproteins and their relatively low levels compared to soluble intracellular proteins [256], the identification and validation of novel differentially expressed surface markers in CRC has been limited. Membrane fractions from patient CRCs have been analysed using two-dimensional fluorescence difference gel electrophoresis (2D-DIGE) saturation labelling followed by MALDI-TOF-TOF mass spectrometry, with identification of 23 proteins, many classified as membrane or membrane-associated [5]. Others have been identified from membrane CRC fractions by iTRAQ [248]. Although approaches such as 2D-DIGE and mass spectrometry have led to the discovery of a number of differentially expressed proteins in CRC [5–8,10–12,257,258], the prognostic significance of these molecules and their contribution to cancer growth and metastasis has not always been easy to establish. Hence very few of these proteins have found clinical utility either as prognostic markers or therapeutic targets.

Proteomic methodologies and their application in CRC research have been reviewed, and the limitations of traditional approaches for cancer biomarker marker discovery and future challenges have been discussed [259–261]. It has been suggested that multiple markers will be required to enable reliable sub-classification of cancers [262–264]. The use of immunohistochemistry for analysis of CRC tissue sections for multiple potentially prognostic/predictive markers [141,265,266] has the advantage of revealing antigen location; however the method is relatively low through-put, expensive, labour-intensive, and quantification is difficult. The use of tissue microarrays enables analysis of a large number of samples simultaneously, but the number of antigens that can be detected in a single assay is limited [267].

The DotScan™ CRC antibody microarray allows rapid immunophenotyping for 122 surface antigens on a population of disaggregated live tumour cells in suspension, requiring 4 × 10⁶ cells per assay. Density of cell binding per antibody dot depends on the proportion of positive cells in the population, level of antigen expression and affinity of the antibody. Figure 1 shows optical and fluorescent binding profiles for a CRC sample from a stage II patient. The optical binding pattern reflects the immunophenotype of the mixed cell population, while the CD3-PE and EpCAM-Alexa Fluor 647 staining detect T-cells and cancer/epithelial cells, respectively. The limit of detection is approximately 1 cell in 10 for the optical scans, while the sensitivity of fluorescent scans is four times higher. We have demonstrated sub-types of CRC based on hierarchical clustering of microarray patterns [22] that may provide prognostic information when data are retrospectively analysed with patient outcomes. This study is ongoing. With the addition of antibodies against newly discovered markers of CRC, this method could be used to correlate cell binding densities with disease stage, invasion, metastasis and patient outcome, and also for finding multiple-marker signatures allowing sub-classification of CRC.

Cell surface proteins are logical candidates as biomarkers and therapeutic targets; they enable a cell population to interact with other cells and soluble factors in the tumour microenvironment and reflect important genomic, epigenetic, transcriptional and translational changes during disease progression. As an alternative to the use of viable cells for analysis with the DotScan CRC antibody microarray, tumour cell-derived membrane fragments, microvesicles or microparticles could be used, as demonstrated with plasma-derived microparticles from patients with cardiovascular disease [268]. This alternative approach may facilitate the detection of potential prognostic biomarkers shed from cells located at the invasive front. Samples of the invasive front are required for routine pathology analysis and are not readily available for research purposes.

Zlobec and Lugli [133] have described the invasive front of CRC as a dynamic interface between pro- and anti-tumour factors. Tumour budding at the invasive front promotes progression and dissemination of CRC cells into the vasculature and lymph nodes, while infiltrating immune cells, particularly cytotoxic T-cells, mount an immune response. The CD8+ lymphocyte/tumour-budding index appears to be an independent prognostic factor in CRC [269]. E-cadherin expression is lower in tumour budding sites than in the centre of the tumour [270], possibly contributing to reduced cell-cell adhesion at the tumour front. It has been suggested that a comparison of the pro- and anti-tumour factors at the invasive front could contribute to development of a prognostic score for patients with CRC [133], with proteomic analysis providing new prognostic markers for CRC.

An alternative proteomic approach is the analysis of patient blood plasma or other biological fluids for CRC biomarkers [271], which include cancer-associated soluble factors, as well as tumour-derived microvesicles [272–275]. Prefractionation of biological fluids is required to detect low abundance proteins [276]. Lectin glycoarrays have been used to profile plasma for CRC biomarkers, with the identification of significant glycosylation changes associated with tumour progression [277–281].

In summary, the search continues for a reliable method of predicting metastatic spread and patient outcome after surgical resection of primary CRC. There is a general consensus that patterns of multiple biomarkers will be required; innovative approaches are needed to translate accumulating data into a useful method for sub-classifying CRC patients for disease stage and those likely to benefit from available treatments. Further progress in cell surface protein discovery will supplement new discoveries provided by other innovative technologies, such as metabolomics [282,283], study of oncogenic pathways [30,31], anti-sense therapeutic strategies [9,284], and combined proteomics/genomics strategies [285].