Extracellular matrix gene expression profiling using microfluidics for colorectal carcinoma stratification.

In cancer, biomarkers have many potential applications including generation of a differential diagnosis, prediction of response to treatment, and monitoring disease progression. Many molecular biomarkers have been put forward for different diseases but most of them do not possess the required specificity and sensitivity. A biomarker with a high sensitivity has a low specificity and vice versa. The inaccuracy of the biomarkers currently in use has led to a compelling need to identify more accurate markers with diagnostic and prognostic significance. The aim of the present study was to use a novel, droplet-based, microfluidic platform to evaluate the prognostic value of a panel of thirty-four genes that regulate the composition of extracellular matrices in colorectal carcinoma. Our method is a novel approach as it uses using continuous-flowing Polymerase Chain Reaction for the sensitive detection and accurate quantitation of gene expression. We identified a panel of relevant extracellular matrix genes whose expression levels were measured by real-time quantitative polymerase chain reaction using Taqman® reagents in twenty-four pairs of matched colorectal cancer tumour and associated normal tissue. Differential expression patterns occurred between the normal and malignant tissue and correlated with histopathological parameters and overall surgical staging. The findings demonstrate that a droplet-based microfluidic quantitative PCR system enables biomarker classification. It was further possible to sub-classify colorectal cancer based on extracellular matrix protein expressing groups which in turn correlated with prognosis.


INTRODUCTION
Colorectal cancer is a significant cause of mortality worldwide with more than 746 000 cases in men and 614 000 cases in women diagnosed annually. 1 The biomarkers used in colorectal cancer mainly aim to either (a) aid in determining suitability for chemotherapy or (b) accurately determine prognosis. Biomarkers are sub classified into clinical (emergency versus elective presentation), serologic (tumour antigens such as carcinoembryonic antigen), or histologic (including nodal status, lymph node ratio, lymphovascular invasion, and perineural invasion). Nevertheless, these biomarkers have limited utility as they are not accurate at the individual level. To overcome this, genetic-based biomarkers have been characterised. These a) Author to whom correspondence should be addressed. Electronic mail: Tara.Dalton@ul.ie. Telephone: þ00-353-61202961.
include microRNAs, single nucleotide polymorphisms, and RNA-based determination of gene expression. The convergence of these efforts with the development of array-based technologies has led to the development of omics, or high-throughput based, biomarkers. However, these are hampered by lack of consensus between laboratories and are not in routine clinical use given costs and other factors. Currently, the consensus is that biomarkers have failed to yield a valid basis for personalised medicine and so the approach to their development must be refreshed. Since its introduction in 1992, 2 qPCR-based technologies have been widely exploited in the evaluation of biomarkers, [3][4][5] where potential markers are measured with greater accuracy to yield gene sets or signatures that can distinguish between a predefined defined subsets of tumours. 6 Micro total analysis systems, such as those utilising droplet microfluidics, provide clinical benefits in the setting of colorectal cancer by performing higher throughputs with enhanced data analysis capabilities. Furthermore, the quantity of available tissue has become a critical factor in -omnic studies as we are now detecting cancer in polyps. Polyps can easily be removed but we cannot accurately determine the probability that the tumour will have spread to nodes within the associated mesentery. This spread is thought to be a critical step in metastasis development. As the risk of spread is unknown, subjecting patients to resection of the segment of colon or mesentery can expose patients to increased and unnecessary risks of complications or death. Therefore, by scaling of the reaction from the microliter down to the nanolitre range dramatically reduces the required volume of reagents and samples, 7 reducing the overall cost per reaction and conserving the clinical samples. Histologically, cancer can be diagnosed in a few hundred cells, so developing an approach to determine nodal status accurately with minimal tissue allows that these technologies can be preferred to obtain improved biological and translatable results.
To date, we have extensively characterised the role of several biomarkers in colorectal malignancy. Our research group has recently identified significant prognostic biomarkers in colorectal cancer 8,9 and established consensus profiles and associations with prognostic and predictive properties in colon and rectal adenocarcinomas. 10,11 Therefore, the hypothesis of this study is to use microfluidic droplets acting as distinct miniature reactors to identify differentially expressed extracellular-matrix (ECM) genes in colorectal cancer tissue and matched patient normal tissue to determine if differential expression could be correlated with an increased metastatic potential. The extracellular-matrix (ECM) has important roles in regulating normal cell and tissue development and function. 12,13 ECM components constantly interact with epithelia as ligands for receptors including integrins. In this manner, they instigate intracellular activities related to a vast array of biologic functions, including tissue development and homeostasis. 14 In fact, the ECM influences almost all cellular behaviour and is critical during major developmental processes. 15,16 However, in solid organ malignancy, ECM-based molecular and cellular interactions can be dramatically altered in a manner that potentiates metastasis development. 17 The metastatic process requires ECM remodelling, basement membrane degradation, and enhanced tumour cell motility. 18 Cell shape, function, proliferation, migration, and apoptosis are all influenced by cell-ECM interactions. These relationships are altered in solid organ malignancies. Therefore, the identification of prominent ECM tumour markers that derive biological insight into tumour development and progression would be of clinical value.

MATERIALS AND METHODS
Patients and Tissue Samples: Colorectal cancer tissue and paired adjacent physiologically normal mucosa tissue samples were collected (ethical approval number 73/11) from a heterogeneous group of 24 patients at different stages of malignancy, undergoing colorectal resection at University Hospital Limerick. Primarily, the tumours are adenocarcinomas. After removal of the colon and associated mesentery, the specimen was opened. Mucosal samples were taken from the tumour and normal mucosa at a distance of 10 cm from the tumour. Samples were immediately placed in Allprotect V R Tissue Reagent (Qiagen, Hilden, Germany) and stored in a À80 C freezer.
RNA extraction and cDNA Synthesis: Frozen tissue was immersed in liquid nitrogen and ground into powder. Lysis buffer was added to tissue and the sample transferred to tubes. Total ribonucleic acid (RNA) was extracted using RNeasy Plus Mini Kit V R (Qiagen, Hilden, Germany) according to manufacturer's instructions. RNA was evaluated for purity on the NanoDrop TM 1000 Spectrophotometer 3.7 (Thermo Fisher, Massachusetts, United States). RNA has its absorption maximum at 260 Nm and the ratio of the absorbance at 260 Nm and 280 Nm is used to assess the RNA purity of an RNA preparation. RNA quality was evaluated through visualization of the 28S:18S ribosomal RNA ratio on a 1% agarose gel. cDNA synthesis was performed using Superscript V R VILO cDNA Synthesis Kit (Thermo Fisher, Carlsbad, USA) according to the manufacturer's protocol. Reactions were incubated for 25 C for 10 min followed by 60 C for 45 min and 72 C for final denaturation for 15 min. Samples were stored at À20 C.
Endogenous reference controls: Relative quantification is based on the expression levels of a target gene versus one or more reference gene(s), and in many experiments it is adequate for investigating physiological changes in gene expression levels. 19 Peptidylprolyl Isomerase A (PPIA) and Hypoxanthine guanine phosphoribosyl transferase 1 (HPRT1) have been showed previously to be stably expressed in a large cohort of colorectal tissues, 20,21 and so were chosen to normalise expression values in the present study. The use of multiple reference genes increases the accuracy of quantitation compared to the use of a single reference gene, especially when the aim is to show relatively small fold-changes in RNA levels. 22 Further to this, our research group have characterised the expression level of nine candidate reference genes and determined their stability in malignant and normal colorectal tissue. 23 From this, Phosphoglycerate Kinase 1 (PGK1), Glucuronidase Beta (GUSB), and PPIA were ranked as the most stable genes between normal and cancerous tissue. Therefore, PGK1 and GUSB were used as additional reference genes. A list of selected reference genes can be seen in Table I.
qPCR of mRNA: The instrument that has been developed on which the PCR will be performed 24 is a droplet based microfluidic platform that performs qPCR in a continuous flowing process. The reaction droplets ($300gl) are surrounded by an immiscible oil to prevent contamination and carryover. Micro-scale polytetrafluoroethylene (PTFE) capillary tubing is integrated into dipping heads that are attached to modular robotic stages. The stages programmed sequence allows dipping into the solutions and aspirates predefined quantities of patient cDNA and nuclease free water (35gl:100gl), gene expression assay primer-probe mix (15gl) (TaqMan V R Gene Expression Assays, Thermo Fisher, Massachusetts, United States), and Mastermix (150gl) (TaqMan V R Gene Expression Master Mix V R -Thermo Fisher, Massachusetts, United States). The individual droplets are then combined at the liquid bridge which consists of capillaries where the droplets are suspended before merging in a free-balanced system. 25 The single PCR reactions are then conveyed through the system to the thermal zone. Thermal cycling is performed in a flowing serpentine cycler, which gives superior thermal performance because of the relative size of the droplet resulting in more sensitive experiments. The temperature blocks of 95 C (denaturisation) and 65 C (annealing) are isothermal to prevent a non-uniform temperature field which may lead to low amplification efficiency of nucleic acids and even non-specific PCR products due to insufficient annealing temperature of the PCR process. 26 Each cycle through a serpentine performs one cycle of PCR and there are 40 cycles in total. To detect the droplet signal a bio-fluorescence detection platform, comprising of reflectance probes, fibre optical cores, and high resolution cameras, analyse each microfluidic reactor droplet. From this data, the amplification s-curves for each sample are generated. A baseline is then set and the s-curve is post processed to output a threshold cycle (Ct) value or, according to the MIQE guidelines, 27 a quantification cycle (Cq) value, which is indicative of the number of RNA targets in the initial samples. The Cq is determined from a log-linear plot of PCR signal versus the cycle number. The background levels are normally determined from the first triplicate no-template-control (NTC) droplets in a run. This is subtracted from the subsequent ROX reference dye normalisation value. The DDCq method is currently the most commonly used method for studies that report changes in the expression of genes of interest relative to a reference gene. 28 For studies performed on the platform, the relative expression levels of the target genes are calculated using the Livak and Schmittgen method 29 using the formula 2 -(DDCt) where DDCq ¼ ½ðCq target gene normal sample Þ À ðCq reference gene normal sample Þ À ½ðCq target gene cancer sample Þ À ðCq reference gene cancer sample Þ: Statistical Analysis: Statistical analysis was carried out using IBM SPSS Statistics 21 (SPSS, Inc.,) The mean of triplicates was used for statistical analysis where the mean difference between the matched samples, the standard deviation, standard error, and 95% confidence interval of this mean difference is evaluated, respectively. The distribution of the data is tested for normality using histograms for both normal and tumour samples and also using the Shapiro-Wilk test (used where n < 50) for normality to determine whether the distribution of the differences in the dependent variable between the two related groups was approximately normally distributed. Whisker plots were generated and interpreted to indicate if there were any "outliers" in the data sets. Outliers were removed on the criteria of being two times the standard deviation of the dataset. Parametric tests (Paired-sample student t-tests) and nonparametric tests (Wilcoxon signed-rank) were used to assess related samples based on the distribution of the data. If homogeneity of variances was violated, then separate parametric and nonparametric tests were performed to assess effect variations. A two-tailed p-value <0.05 was considered statistically significant. Correlations between parameters and ECM gene expression levels in the samples analysed were determined by Pearson product-moment correlation coefficient, r. P values <0.05 were considered statistically significant.

Gene Selection criteria:
A detailed analysis of extracellular matrisome literature reviews and published colorectal cancer gene expression profiling studies was undertaken to identify key ECM genes involved in colorectal cancer formation and development. Genes that code for constituents such as integrins, collagens, proteases, and growth factors were selected to be included in the panel. Furthermore, an analysis of current ECM gene expression arrays such as the Taqman V R and Qiagen V R ECM and Adhesion Molecule arrays (Catalogue number: 4414133 and PAHS-013ZA) 30,31 was reviewed to support hypotheses developed for which genes could be most influential in the metastatic spread of the disease. A list of the selected genes for analysis in this study can be seen in Table II with references for studies which show specific genetic dysregulation.

DIFFERENTIAL GENE EXPRESSION IN MATCHED SAMPLES
RNA expression levels of the panel of thirty-four ECM genes were measured by qPCR in the twenty-four malignant tissue samples and their corresponding normal tissue. The proportion of tissue represented by epithelium or stroma (i.e., the epithelial-stromal ratio) can affect the expression of genes. Expression levels of Cytokeratin-18 (KRT18) and Vimentin (VIM) were assessed by qPCR to approximate the epithelial and stromal tissue components of each sample 9,32 with average composition for the twenty four samples shown to be consistent and can be seen in Figure 1. Differential gene expression greater than two-fold difference between normal and malignant tissue was observed in twenty-two genes. The differences identified in expression were statistically significant in the case of fourteen genes ( Figure 2). These genes have varying biological functions and code for cell surface proteins, growth factors, and proteins involved in tissue remodelling and cell adhesion. The greatest difference in gene expression was observed in MMP7 (þ31 fold, p ¼ <0.001) between the matched tumour and normal mucosa in this study. IL-8 expression was significantly up-regulated (þ18-fold, p ¼ <0.001) and ITGBL1 expression was shown to be increased (þ11-fold, p ¼ 0.003) in malignant versus matched normal tissue. MMP9 expression was observed to be upregulated (þ7-fold, p ¼ 0.02) in the malignant tissue as was COL1A1 expression (þ6-fold, p ¼ 0.039) and ITGA2 (4-fold, p ¼ 0.007). CXCR2 (4-fold, p ¼ 0.022), SPARC (3.5-fold, p ¼ 0.033), and RACK1 (2.5 fold, p ¼ 0.019) were also identified as being upregulated in the matched comparison. Conversely, decreased expression levels were observed in six of the genes in the ECM panel. IGF-1 expression levels were lower (À8-fold, p ¼ 0.007) in cancer tissue than in the normal tissue. Decreases in expression levels for EGF (À5 fold, p ¼ 0.014) and EGFR (À3-fold, p ¼ 0.05) were also observed. CEACAM1 was also identified as being decreased (À4-fold, p ¼ 0.048). ITGA8 gene expression levels were shown to be downregulated (À3.6-fold, p ¼ 0.003) in the cohort. This information is tabulated in Table V.

ANALYSIS OF EXPRESSION COORDINATION AND GENE NETWORKING
After the initial gene expression levels were determined by experimental means, gene pairs which had similar expression patterns from the fourteen gene panel were sought. The expression levels in the 24 patient cohort were assessed using a Pearson's correlation coefficient test for significance (p-value < 0.05). The strongest positive relationships were observed between COL1A1 and SPARC expression (q ¼ 0.763, p-value ¼ <0.001) and COL1A1 and RACK1 (q ¼ 0.451, p-value ¼ 0.032). MMP7 and ITGA2 also showed a correlation (q ¼ 0.552, p-value ¼ 0.018). Correlations between ITGBL1 and both SPARC expression (q ¼ À0.720, p-value ¼ 0.001) and EGF (q ¼ À0.699, p-value ¼ <0.001) also emerged as being statistically significant. A strong positive correlation was observed between EGF and its receptor EGFR expression (q ¼ À0.470, p-value ¼ 0.032). The associations between the significant gene expression arrays are shown in Table III.
Subsequently, the patient cohort was stratified according to UICC staging to identify further differential expression patterns ( Figure 3). The cohort was categorised based on the histopathological diagnoses as follows; a) Group 1; benign disease samples including diverticulosis, hyperplasia, tubulovillous adenoma (TVA) polyps and Stage I samples (n ¼ 6) b) Group 2; stage II carcinomas (n ¼ 11) c) Group 3; stage III and stage IV advanced carcinomas (n ¼ 7). Gene expression was compared between malignant and matched normal tissue. The number of genes differentially expressed increased sequentially from Group 1 to Group 3.

ASSOCIATION BETWEEN MALIGNANT SAMPLE GENE EXPRESSION AND HISTOPATHOLOGICAL PARAMETERS
Tumour diameter: The gene expression of three ECM genes in the panel, IGF1(r ¼ À0.596, p ¼ 0.009), ITGBL1 (r ¼ 0.669, p ¼ 0.005), and VEGF (r ¼ À0.487, p ¼ 0.04), correlated with the diameter of the samples which were recorded during surgical examination. The tumour size at the largest point was recorded in millimetres (mm) with three categories of <30 mm, 30-40 mm, and >40 mm. IGF1 decreased expression (À13 fold) and ITGBL1 increased expression (26.4-fold) were observed to be present in tumours with a diameter greater than 40 mm.
Mucinous Component: From the cohort, four of the twenty tumours were documented as having a mucinous component and all were Stage III carcinomas. Colorectal mucinous carcinomas present at a more advanced stage, predominantly in men, with higher right colon location FIG. 2. Gene expression fold change levels for the patients included in the study showing statistically significant differences between normal mucosa and matched tumour samples from our patient cohort. We took the average fold change relative to normal tissue sample and normalised to the geometric mean of the reference genes.   Differentiation: The expression levels of three genes showed a moderate correlation with the level of differentiation which was recorded during surgical resection. Upregulation of COL3A1 was significantly associated with poorer differentiation (r ¼ 0.469, p ¼ 0.024) along with upregulation of MMP9 (r ¼ 0.423, p ¼ 0.05). It has been shown 34 that increased levels of MMP-9 correlate with poorer differentiation. A slight increase in IGF-1R (r ¼ 0.533, p ¼ 0.049) levels also correlated with poorer differentiation.
Recurrence: Three of the samples in the cohort were taken from patients who later presented with a recurrence of the carcinoma. PTK2 (r ¼ 0.435, p ¼ 0.043) levels in these patients were shown to be elevated in comparison with patients whose cancers had not recurred.
Age: Patients were categorised into two groups of being up to and including sixty-four years of age and sixty-five and older. Increased MMP7 expression correlated with increased age Gender: IGFBP3 expression levels were noted to correlate (r ¼ 0.424, p ¼ 0.044) with patient gender with male patients having an increased expression (þ4-fold) compared to females Lymph Node Status: In nonmetastatic colorectal cancer, lymph node status is the strongest pathologic predictor of patient outcome and is used for determining the most appropriate adjuvant treatment for patients. In the cohort, patient's lymph nodes were examined for abnormal cellular behaviour and inflammation. The nodal status was determined using the American Joint Committee on Cancer (AJCC) TNM system where the "N" portion of this classification system refers to the nodal status indicating the presence, if any, of cancer spread to other lymph nodes. The system is: NX: Regional lymph nodes cannot be assessed, N0: Cancer has not spread to regional lymph nodes, N1: Cancer has spread to 1 to 3 lymph nodes, and N2: Cancer has spread to 4 to 9 lymph nodes. Four patients were classified as NX, twelve patients were classified as N0, six were classified as N1, and 2 were classified as N2. The expression of PTK2, or Focal Adhesion Kinase (FAK) as it is more commonly known, showed a step-wise increase from N0 to N1 to N2 (r ¼ 0.587, p ¼ 0.006). A stepwise decrease was also observed for ITGA8 for N0 to N2 (r ¼ À0.523, p ¼ 0.018).
Invasiveness: The invasiveness of the tumour is closely linked to the size and the grade of the tumour and is determined by whether or not the tumour penetrates the basement membrane and/or muscle layers in the lumen of the colon. For this study, a tumour was categorised as being invasive or not. In the twenty-four patient cohort, nineteen samples are documented to be invasive malignancies of varying degree. The depth of invasion was not recorded. These samples were analysed to determine if changes in expression for the ECM genes correlated with the invasiveness of a cancer. SPARC and COL1A1 expression levels were shown to be increased significantly in samples which were invasive (r ¼ 0.465, p ¼ 0.022) (r ¼ 0.494, p ¼ 0.017).
Perineural Invasion (PNI): PNI is grossly underreported in CRC and could serve as an independent prognostic factor of outcomes in these patients. 35 The PNI absence was reported in six of the 20 tumour samples. Perineural invasion was not stated for fourteen others. MMP-9 levels were significantly decreased in patients with no perineural invasion (r ¼ 0.481, p ¼ 0.037). Increases in MMP-9 aid a cancer cell's progression to metastasis and have been shown 36 to correlate with increases in perineural invasion in the literature.
Stage: The clinical staging of the patients disease was also determined by the AJCC TNM system. Stage grouping can be seen in Table IV. ITGBL1 expression (r ¼ 0.588, p ¼ 0.013) was identified as showing a stepwise increase in expression which correlated with the staging of the patient cohort with an $8 fold increase in expression for patients with Stage II and $22 fold increase for Stage III in comparison to Stage I. LAMA1 expression showed an increase in Stage II (þ20-fold) but a significant decrease in expression in Stage III (À40 fold) (r ¼ 0.509, p ¼ 0.037).
Grade: ITGA2, ITGB1, and CD9 gene expression levels were correlated with the grade of the malignant samples analysed albeit with a small sample size. ITGA2 was observed be lower in T3 and T4 tumour samples in comparison to T1 and T2 tumour samples (r¼ À0.657, p ¼ 0.004), while the levels of ITGB1 replicated this pattern (r ¼ À0.501, p ¼ 0.041). CD9 expression levels also correlated with the grade of the cancer samples with a stepwise decrease in expression (r ¼ À0.613, p ¼ 0.007) from T1 (þ10 fold), T2 (þ2.14 fold), T3 (þ0.05 fold) to T4 (À11.5 fold). In addition, the T1 sample and the two T4 samples were removed from the dataset and T2 and T3 correlations were evaluated independently. When this was done, the correlation between ITGA2 and ITGB1 expression with the grade of the tumour remained statistically significant. However, the correlation between CD9 expression and grade did not remain statistically significant.

DISCUSSION
Identifying differentially expressed genes acts as an additional clinical tool that may be used to diagnose patients and may influence the potential treatment decisions. The global dysregulation observed in this gene expression study of the ECM indicates that substantial dysregulation and remodelling of the ECM occur during colorectal pathogenesis and this can lead to increased metastatic potential. The differential expression of fourteen genes which were statistically significant was identified. The largest dysregulation was observed in MMP7. MMP7 cleaves cell surface proteins, promotes adhesion of cancer cells, and increases the potential of tumour metastasis 37 and has been widely shown to be significantly higher than that of normal cells in numerous studies both for colorectal carcinoma [38][39][40] and other cancers. 41-43 IL-8 was also dysregulated and its overexpression has been detected in many carcinomas, including colorectal cancer 44 and is associated with poor prognosis. 45 It has been suggested that tumour cells produce IL-8 as an autocrine growth factor that promote tumour growth, tissue invasion, and metastatic spread. 46 ITGBL1 is a gene that is not frequently highlighted in the literature but has been shown to have a role in breast cancer bone metastasis by activating TGFb signalling pathway. 47 MMP9 is a second protease from the initial panel that emerged as having differential expression in malignant tissue in comparison to the matched normal tissue. Its overexpression has been observed in colon cancer studies where it is largely associated with metastatic progression of colorectal cancer and overall poor prognosis. 48,49 COL1A1 was observed to be increased which may be contribute to abnormal tissue remodelling 50 and increased tumour invasion potential. ITGA2 has been linked to having a role in cell migration and metastasis by increasing cofilin phosphorylation 51,52 and was observed to be upregulated in the cohort, while CXCR2, which is the receptor for LAMA1 and also a critical mediator of cellular senescence, was observed to be overexpressed. Upregulated expression of SPARC was shown which conforms with literature. 53 RACK1 overexpression was also evident between samples and our research group has shown it interacts with PTK2 and IGF1R to promote cellular migration. 54 IGF1 gene expression has been reported to be higher in colorectal cancer that that of normal mucosa. 55,56 However, decreased levels were observed in this study in the malignant tissue. CEACAM1, a gene shown to be upregulated in gastric cancer and has been shown to promote angiogenesis, 57 was also observed to be overall decreased between normal and cancerous tissue. Restoration of CEACAM1 expression in tumour cell lines often abolishes their oncogenicity in vivo, and therefore, this adhesion molecule has been regarded as a tumour suppressor. 58 Decreased levels of ITGA8 levels were observed in the cohort. ITGA8 has been shown to be one of a panel of genes on an enriched molecular pathway in colorectal cancer. 59 EGF and its receptor EGFR were both downregulated in the study and both have been shown to be implicated in the growth and invasion of colon cancer. 60,61 The overexpression of EGFR in colon cancers without a KRAS mutation has led to targeted therapies by blocking EGFR which has led to increased patient benefits. 62 When analysed with regard to UICC classification of the disease, the pattern of dysregulation became more pronounced and allowed for potential molecular markers to be identified. By showing the differential expression with regard to UICC staging (Figure 3), a number of genes emerge as being potential biomarkers for Stage I, Stage II, and Stage III/IV colorectal cancer. These genes could become targets for specific treatment in patients presenting with these stages of colorectal cancer or become an influencing factor in determining a patients' diagnosis. The stages at which genes become more influential in disease development and progression is shown (Figure 4). Once expression levels were determined by experimental means, the panel of fourteen genes was assessed for co-expressed genes (genes with similar expression patterns) using Pearson's Correlation test. Strength of relationship is determined by closeness to 1 (direct correlation) or À1 (inverse correlation). COL1A1 was identified to be co-expressed with both SPARC and RACK1. COL1A1 and SPARC genes participate in ECM and cytoskeletal control and are also involved in the carcinogenesis of many malignancies and both have been shown to be relevant for tumour prognosis. 63 Co-expression was also significant between MMP7 and ITGA2, ITGBL1 and SPARC, ITGBL1 and EGF, and EGF and EGFR (Table III). A similar pattern of expression can be used to identify co-expression networks which are of biological interest as co-expressed genes can be (a) controlled by the same transcriptional regulatory program (b) functionally related or (c) members of the same pathway or protein complex. 64 Furthermore, expression coordination between genes involved in different functional categories suggests that the functional pathways can modulate each other. Associations between the gene expression patterns of all genes and histopathological parameters recorded was also assessed. Strong associations between these parameters were observed highlighting ECM genes which potentially could be used in aiding tumour phenotype predictions. Graphical representation of the association of expression analysis with histopathological parameters can be seen in Figure 5 where the genes identified could be used alone or as complement to existing CRC diagnostic tools for cancer development and/or progression.

CONCLUSION
This study highlights fourteen ECM genes, which show differential expression between tumour tissue and matched normal colon mucosa. The genes which emerged as being different have been previously shown in the literature to play key roles in the carcinoma process for a range of individual cancers and in particular, colorectal carcinoma. A pairwise correlation  Table IV.
pattern emerged also which showed that certain genes were co-expressed. Further to this, correlations emerged between the gene expression pattern and clinic-pathological parameters which were obtainable from pathology reports such as patient gender, tumour diameter, lymphovascular invasion, and nodal status. The genes highlighted in this study may represent important targets in colorectal carcinogenesis and could provide useful clinicopathological tools in the management and treatment of colorectal cancer. Expanding the patient cohort and genetic panel could yield further potential biomarkers for metastasis and other treatment targets in colorectal cancer. The successful utilisation of novel technologies such as microfluidics to solve biological questions potentially allows us to maximise biomarker discovery rates and improve cancer patient prognosis. qPCR integrated microfluidics provides numerous advantages for gene expression: economies of scale, parallelisation and automation, and increased sensitivity and precision that comes from small volume reactions. The microfluidic droplet technology used in this study has the ability to analyse the expression levels of hundreds of genes across hundreds of samples in a continuous flow regime while using reduced volumes of both sample and reagents. The instrumentation will assist in the development of further microliter and nanoliter microfluidic platforms which will progress molecular diagnostic advances in the biological research field. Using this technology, we have shown that differential gene expression levels greatly influence many biological conditions in both normal and abnormal circumstances and can provide valuable information into the tissue and cell developmental behaviour in response to environmental stimuli. The three-dimensional matrix, which supports, protects, and stabilises all structures and cells, has diverse functions in the body. Accumulating evidence supports the theory that the ECM is mechanically, biochemically, and functionally influential in the metastatic behaviour of cancer cells. All structures within the matrix environment are inherently linked. By identifying ECM related genes that influence the metastatic potential of cells, we could potentially use these genes to predict prognosis. Comparing these findings with current array-based studies supports the emerging panel of ECM genes as being dysregulated in colorectal cancer. This highlights the potential of using microfluidic-based approaches for candidate biomarker studies and is an example of one of the first applications of microfluidic technology to colorectal cancer studies. The results of this gene expression profiling could stimulate an interest in the direct use of expression profiles in the routine clinical setting. However, the data outlined is preliminary. Further studies involving larger patient populations (including different tissue types) and confirmed biomolecular networks are merited giving the potential for more informed diagnosis and early identification of metastasis in malignancies.