Characterisation of the oxysterol metabolising enzyme pathway in mismatch repair proficient and deficient colorectal cancer

Oxysterols are oxidised derivatives of cholesterol, formed by the enzymatic activity of several cytochrome P450 enzymes and tumour-derived oxysterols have been implicated in tumour growth and survival. The aim of this study was to profile the expression of oxysterol metabolising enzymes in primary colorectal cancer and assess the association between expression and prognosis. Immunohistochemistry was performed on a colorectal cancer tissue microarray containing 650 primary colorectal cancers using monoclonal antibodies to CYP2R1, CYP7B1, CYP8B1, CYP27A1, CYP39A1, CYP46A1 and CYP51A1, which we have developed. Unsupervised hierarchical cluster analysis was used to examine the overall relationship of oxysterol metabolising enzyme expression with outcome and based on this identify an oxysterol metabolising enzyme signature associated with prognosis. Cluster analysis of the whole patient cohort identified a good prognosis group (mean survival=146 months 95% CI 127-165 months) that had a significantly better survival (δ2=12.984, p<0.001, HR=1.983, 95% CI 1.341-2.799) than the poor prognosis group (mean survival=107 months, 95% CI 98-123 months). For the mismatch repair proficient cohort, the good prognosis group had a significantly better survival (δ2=8.985, p=0.003, HR=1.845, 95% CI 1.227-2.774) than the poor prognosis group. Multi-variate analysis showed that cluster group was independently prognostically significant in both the whole patient cohort (p=0.02, HR=1.554, 95% CI 1.072-2.252) and the mismatch repair proficient group (p=0.04, HR=1.530, 95% CI 1.014-2.310). Individual oxysterol metabolising enzymes are overexpressed in colorectal cancer and an oxysterol metabolising enzyme expression profile associated with prognosis has been identified in the whole patient cohort and in mismatch repair proficient colorectal cancers.


INTRODUCTION
Colorectal cancer is one of the most common types of malignancy affecting both men and women, with a worldwide annual incidence of greater than 1.2 million new cases [1,2]. The disease remains a leading cause of cancer-related mortality and, despite gradual improvements in prognosis, the 5-year survival remains relatively poor at approximately 55% [1]. Colorectal cancer develops slowly over several years and symptoms often only become apparent in the late stages, therefore many colorectal cancers present at an advanced stage. Patients presenting with distant metastatic disease have a 5-year survival of less than 10% [1].
Currently, colorectal cancer is commonly staged using the tumour, node, metastasis (TNM) staging system to guide treatment decisions and indicate prognosis. However, patients with the same stage of tumour often experience a wide range of different clinical outcomes. Despite the unequivocal value of current staging systems, Research Paper www.impactjournals.com/oncotarget there is a still need to develop reliable biomarkers to more accurately predict prognosis and risk stratify patients with colorectal cancer. Biomarkers can have a variety of roles in colorectal cancer including early detection, predicting prognosis, predicting response to therapy and aiding postoperative monitoring [3].
Oxysterols are oxidised derivatives of cholesterol, formed by the enzymatic activity of several cytochrome P450 enzymes [4,5]. Oxysterols function as key signalling molecules involved in the development and functioning of the immune system and the maintenance of cellular cholesterol homeostasis [6][7][8][9][10][11][12]. In addition to the established role of oxysterols in normal immune system functioning, it is increasingly acknowledged that the oxysterol pathway plays a role in tumourigenesis through altering host anti-tumour immunity. For example, oxysterols have been demonstrated to down-regulate the G-protein coupled receptor chemokine receptor 7 (CCR7) through activation of the ligand-activated transcription factor LXRα in dendritic cells [13]. CCR7 is involved in the migration of dendritic cells to draining lymph nodes, thus suppression of this chemokine receptor results in trapping of dendritic cells in the tumour and subsequent interference with antigen presentation to anti-tumour T-cells [7]. Through suppression of CCR7 in an LXRdependent manner, oxysterols impede host anti-tumour immunity. A further mechanism whereby oxysterols may promote tumour progression is via chemo-attraction of neutrophils [14,15]. Invading neutrophils may provide a critical growth and survival advantage in many solid tumours due to production of the pro-angiogenic factors prokineticin-2 and matrix metalloproteinase-9 [16].
Despite the recognition of the role of oxysterol signalling in tumourigenesis, the key cytochrome P450s involved in the oxysterol pathway have received very limited study in existing research with regard to their expression in tumours [17,18]. This study has profiled the expression of the cholesterol metabolising enzymes CYP2R1, CYP7B1, CYP8B1, CYP27A1, CYP39A1, CYP46A1 and CYP51A1 in primary colorectal cancer tissue using a well-characterised cohorts of colorectal cancers. The clinico-pathological significance of each of the cytochrome P450s studied was determined, including the relationship between expression and overall survival. An oxysterol metabolising enzyme expression profile associated with prognosis has been identified in the whole patient cohort and in mismatch repair proficient colorectal cancers.

Monoclonal antibodies to oxysterol metabolising enzymes
The specificity of the monoclonal antibodies to CYP2R1, CYP7B1, CYP8B1, CYP27A1, CYP39A1, CYP46A1 and CYP51A1 was determined by ELISA using the immunogenic peptides and also by immunoblotting using whole cell lysates from cells overexpressing of each protein (Figure 1). A band migrating at the expected molecular weight was observed for each antibody in a lysate prepared from cells overexpressing the relevant protein while no bands were detected with the corresponding control lysate.

Expression of oxysterol metabolising enzymes in primary and metastatic colorectal cancer
Each antibody showed immunoreactivity in normal colonic epithelium, primary colorectal cancer and metastatic colorectal cancer. Where immunoreactivity was observed, immunostaining was localised to the tumour cell cytoplasm ( Figure 2). Nuclear or membranous staining was not observed. Whole section immunohistochemistry  of a sub-set of tumours showed no evidence of intratumour heterogeneity of expression of any of the oxysterol metabolising enzymes.
A general trend of increased expression in primary tumour cells compared to normal colonic mucosa was observed for all proteins studied ( Figure 3). Furthermore, expression of each oxysterol metabolising enzyme was slightly reduced in lymph node metastasis compared to primary tumour. Immunostaining for CYP7B1 in primary tumours showed the highest proportion of strong immunostaining, with 62.8% of primary colorectal cancers displaying strong immunoreactivity for this target protein.
Immunostaining for CYP46A1 showed 27.7% of primary tumours were strongly stained. Strong immunostaining for CYP51A1 was demonstrated in 21.9% of tumours and CYP8B1 immunoreactivity was classified as strong in 18.9% of tumours. The remaining target proteins had very low frequencies of strong immunostaining and no tumour showed strong CYP39A1 immunoreactivity.
When examining the difference in expression between all cases of primary colorectal cancer and lymph node metastasis CYP7B1 (p=0.035), CYP39A1 (p=0.001), CYP46A1 (p<0.001) and CYP51A1 (p<0.001) each showed a significant decrease in immunoreactivity in lymph node metastasis compared to primary tumour. There was no statistically significant difference in expression of CYP7B1 or CYP27A1 between the paired cases of Dukes C (stage 3) colorectal cancer and their corresponding lymph node metastasis. However, CYP2R1 (p=0.034), CYP8B1 (p=0.002), CYP39A1 (p<0.001), CYP46A1 (p<0.001) and CYP51A1 (p=0.001) each demonstrated significantly reduced expression in paired lymph node metastasis compared to Dukes C (stage 3) colorectal cancer.

Relationship of individual oxysterol metabolising enzymes with clinico-pathological parameters
The relationship between expression of each protein and relevant clinico-pathological parameters are summarised in Table 2. Expression of CYP2R1, CYP8B1, CYP27A1, CYP39A1 and CYP46A1 showed significant associations with Dukes stage. CYP39A1 and CYP46A1 were also significantly associated with tumour stage and lymph node stage. CYP7B1, CYP8B1, CYP39A1, CYP46A1 and CYP51A1 each displayed a statistically significant relationship with location of tumour in the colon versus the rectum. When the anatomical site of the tumour was stratified as proximal colon, distal colon or rectum, significant associations were found with expression of CYP8B1, CYP27A1 and CYP51A1. When investigating the relationship between expression of each protein and the presence of extramural venous invasion, CYP8B1, CYP27A1, CYP39A1 and CYP46A1 showed correlations with this pathological variable. In addition to extramural venous invasion, CYP27A1 was also associated with mismatch repair protein status. Expression of CYP51A1 was associated with tumour differentiation and mismatch repair protein status.

Unsupervised hierarchical cluster analysis and identification of prognostic signature
Unsupervised hierarchical cluster analysis was used as an unbiased exploratory statistical tool to examine the overall relationship of oxysterol metabolising enzyme expression with outcome and based on this identify an oxysterol metabolising enzyme signature associated with prognosis. A range of cluster solutions (number of clusters) was investigated to determine the optimum number of clusters that produced groups with different outcomes. Clustering the data into five clusters was identified as the optimum number of clusters for analysis in relation to the most prognostically significant groups ( Figure 4). These five clusters were then combined into two groups; a good prognosis group (group 1, cluster 1) and a poor prognosis group (group 2, cluster 2-5) (Figures 4 and 5). The relationship of each cluster group and pathological parameters is shown in Table 3 and the expression of each oxysterol metabolising enzyme in each cluster group is shown in Table 4. The good prognosis group showed contained with low expression of CYP2R1 (p=0.002), CYP8B1 (p<0.001), CYP27A1 (p=0.028) and CYP46A1 (p<0.001) relative to the poor prognosis group.

Multi-variate analysis
Multi-variate analysis showed that cluster group was independently prognostically significant in both the whole patient cohort (p=0.02, HR=1.554, 95% CI 1.072-2.252) and the mismatch repair proficient group (p=0.04, HR=1.530, 95% CI 1.014-2.310) in a model that included information about Dukes stage (Table 5, Model A). Multi-variate analysis also showed that cluster group was independently prognostically significant in the whole patient cohort (p=0.033, HR=1.497, 95% CI 1.032-2.172) and not the mismatch repair proficient cohort in a model that contained tumour stage and lymph node stage ( Table  6, Model B). Multi-variate analysis was also performed using only parameters that would be available at the time of a biopsy of colorectal cancer i.e. no pathological    information about tumour stage, lymph node involvement or extramural venous invasion. In this case oxysterol metabolising enzyme signature was highly significant in both the whole patient cohort (p=0.002, HR=1.791, 95% CI 1.236-2.595) and the mismatch repair proficient group (p=0.011, HR=1.703, 95% CI, 1.129-2.568) (Supplementary Table 9).

DISCUSSION
Colorectal cancer is one of the commonest types of solid tumour worldwide with an incidence that is still increasing especially in specific geographic areas [1]. While the molecular pathways involved in the initiation and the early stages of the development of colorectal cancer have been well defined this type of tumour still has a relatively poor prognosis with an overall survival of about 50-60%. The introduction of screening programs for its earlier detection and the development of targeted therapies for locally advanced and metastatic disease should impact on and improve the outcome from this disease [2,19]. However, there is still a clear requirement to identify biomarkers of colorectal cancer which can contribute to improved screening and earlier diagnosis and prognostic stratification [3,20].
This study has identified the expression profile of oxysterol metabolising enzymes in a well-characterised uniform cohort of primary colorectal cancers none of which had received pre-operative chemotherapy and/or radiotherapy. The expression of each enzyme was studied in primary colorectal cancer, corresponding lymph node metastasis and normal colonic mucosa. An oxysterol metabolising enzyme expression profile or signature associated with prognosis was identified.
Oxysterol metabolising enzymes are members of the cytochrome P450 superfamily of enzymes which catalyse NADPH-dependent mono-oxygenation reactions [21]. The cytochromes P450s are generally considered to belong to one of two distinct groups depending on whether they metabolise xenobiotics or endogenous substances and are classified into families, subfamilies and individual forms according to sequence homology and substrate specificity [22][23][24][25]. The major xenobiotic metabolising cytochrome P450s are members of the CYP1, CYP2 and CYP3 families. There is extensive evidence for the expression of xenobiotic metabolising enzymes in tumours [26][27][28][29]. There was no evidence of intra-tumour heterogeneity and this is consistent with our previous studies of other cytochrome P450 enzymes in tumours [29,30,31]. With some cytochrome P450s especially CYP1B1 showing increased expression in tumour cells and the tumour associated expression of individual cytochrome P450s has been exploited as therapeutic targets for P450 mediated pro-drug activation and as a cancer vaccine [29,[32][33][34]. The cytochrome P450s involved in the metabolism of a diverse range of endogenous compounds including eicosanoids, fatty acids, steroids and vitamins are the CYP4 family and higher numbered cytochrome P450 families.
Oxysterols are derived from cholesterol and can be produced by the hydroxylation of cholesterol by specific cytochrome P450 enzymes [35,36]. Tumourderived oxysterols are multifunctional lipid-signalling molecules and recent evidence indicates that they www.impactjournals.com/oncotarget have pleiotropic effects in tumours [9]. Individual oxysterols have been identified as having a variety of functions in tumours including influence on tumour cell proliferation and tumour growth, mediating the tumour microenvironment especially of immune cell function and inflammation, tumour invasion and metastasis via the matrix metalloproteinase system and also mediating tumour associated angiogenesis [7,12,14,16,37]. The overall influence of oxysterols on tumour biology will depend on the relative expression of individual oxysterol metabolising enzymes. Although this study has focused on oxysterol pathway and its influence in tumour progression and metastasis, it is worth noting that cytochrome P450 enzymes have pleotropic functions that might also impact on tumour progression by modulating other signalling pathways.
In this study monoclonal antibodies with specificity for individual oxysterol metabolising enzymes CYP2R1, CYP7B1, CYP8B1, CYP27A1, CYP39A1, CYP46A1 and CYP51A1, have been produced. Peptides to C-terminal amino acid sequences identified on the basis of sequence alignment and homology modelling of individual cytochrome P450s were used as immunogen as this approach has proved highly successful in the development of monoclonal antibodies selective for individual cytochrome P450s [26]. CYP2R1, CYP7B1, CYP8B1, CYP46A1 and CYP51A1 all showed significantly increased expression in primary colorectal cancer compared to normal colonic mucosa with CY7B1 demonstrating the highest proportion of strong immunoreactivity in colorectal cancer compared to all other enzymes studied. This is the first study to analyse the expression of CYP7B1 in colorectal cancer, with previous research focusing on expression levels in prostate and breast cancer due to the role of CYP7B1 in sex hormone metabolism [38,39]. CYP7B1 has been shown to be associated with survival in both breast cancer and prostate cancer [36,37]. The findings of increased expression of CYP8B1 and CYP46A1 in primary colorectal cancer are novel findings. This study found increased expression of CYP51A1 in primary colorectal cancer compared with normal colonic mucosa. This finding is consistent with a previous study that also found increased expression of CYP51A1 in primary colorectal cancer [28].
Hierarchical cluster analysis allows the unbiased identification of groups of cases with similar expression profiles. Cluster analysis of the whole patient cohort identified five clusters which were mapped to two groups that were of prognostic significance; a good prognosis group which demonstrated low expression of CYP2R1, CYP8B1, CYP27A1, and CYP46A1 in comparison to their expression in the poor prognosis group. Similar prognostic potential of CYP27A1 and CYP7B1 in breast cancer has been reported [18,38]. That research highlighted the important roles played by the oxysterol metabolite 27-hydroxycholesterol which results from both anabolism and catabolism of CYP27A1 and CYP7B1 enzymes respectively in tumour pathophysiology. The fact that our study has examined all the key enzymes of oxysterol pathway in colorectal cancer will lead to a comprehensive understanding of roles played by such enzymes and their oxysterol metabolites in tumour. Multi-variate analysis confirmed independent prognostic significance. Of particular interest was the highly significant prognosis in a model containing only information available at the time of biopsy diagnosis of colorectal cancer. This is important as the concept is emerging of treating more patients diagnosed with colorectal with neoadjuvant therapy followed by either observational follow-up or salvage surgery and it will be essential to have prognostic or risk-stratification biomarkers for this scenario in which only tumour biopsies are available for study at the time of initial treatment decisions [40].
The mismatch repair pathway is one of the major pathways of colorectal cancer development [41,42]. Tumours that lack key mismatch repair proteins are classified as mismatch repair defective or deficient or unstable. Those patients with mismatch repair deficient tumours are already regarded as a distinct subgroup in selecting patients for adjuvant therapy and indicating prognosis, therefore the mismatch repair proficient group are of particular interest. For example, recent studies indicate that mismatch repair deficient tumours respond to immune checkpoint anti-programmed cell death 1 inhibitors in contrast to mismatch repair proficient tumours which showed no response [43]. This is due to the fact that mismatch repair deficient tumours have a high mutational load in coded proteins increasing the probability of recognition and elimination by the immune system. On the other hand, mismatch repair proficient tumours have low mutational load, proving effective in evading immune system. If this is the case then it is very plausible to speculate that most of existing and ongoing immunotherapies, including anti-CTLA4 and anti-CD20 will have limited therapeutic effect on mismatch repair proficient CRC patients. It is well-established that mismatch repair proficient tumours represent the majority of colorectal cancer patients and frequently has a worse prognosis compared to mismatch repair deficient tumours hence in most need of novel therapies. In the mismatch repair proficient group, a protein signature was identified that was associated with prognosis. This reflected the same relative expression of each cytochrome P450 as the whole patient cohort good prognosis group compared to the poor prognosis group. In the mismatch repair deficient group there was also a trend towards poorer prognosis.
This study also assessed the phenotypic expression of each P450 in both primary colorectal cancer and paired lymph node metastasis. When the primary tumours of lymph node positive cases (Dukes C, stage 3) were compared to the paired lymph node metastasis, expression of CYP2R1, CYP8B1, CYP39A1, CYP46A1 and CYP51A1 were each significantly reduced in the lymph node metastasis. This highlights the role of the tumour microenvironment in influencing the expression of the target proteins, a concept increasingly recognised in studies of the metastatic spread of malignancy [44][45][46]. The findings of this study have provided further evidence of the potential role of the tumour microenvironment in altering the phenotype of cancer cells. The spatial organization and hence interactions of individual types of immune cells within lymph nodes are distinct from the microenvironment of the primary tumour and thus contributes to a microenvironment that has a different structure and function to that of the primary tumour [47]. The variation in phenotype observed in metastasis compared to the primary malignancy also highlights the difficulty in effectively treating metastatic disease. Treatments for metastatic disease are often guided by assessment of the primary tumour which, as confirmed by this study, does not necessarily reflect phenotypic expression of disease at metastatic sites.
In conclusion this study has defined the expression of oxysterol metabolising P450s in a well-characterised cohort of colorectal cancers. An oxysterol metabolizing enzyme signature has been identified which is associated with prognosis in the whole patient cohort and the mismatch repair proficient cohort. This good prognosis group showed tumours with low expression of CYP2R1, CYP8B1, CYP27A1 and CYP46A1 relative to the poor prognosis group and a schematic model of the relationship of oxysterol metabolizing enzymes in good and poor prognosis colorectal cancers is outlined in Figure 6. This study also raises the possibility of therapeutic targeting of the oxysterol metabolising pathway as individual P450s are well-characterised actionable drug targets [48].

Monoclonal antibodies
Monoclonal antibodies to individual oxysterol metabolising enzymes (CYP2R1, CYP7B1, CYP8B1, CYP27A1, CYP39A1, CYP46A1, and CYP51A1) were developed in collaboration with Vertebrate Antibodies Ltd (Aberdeen, UK) using synthetic peptides. Peptides within the protein sequences were selected from regions which were antigenic, exposed on the surface and unique to the target protein [26,[49][50][51]. The amino acid sequences and location on the proteins are indicated in Table 7. The peptides were obtained from Almac Sciences Ltd, (Edinburgh, UK) and conjugated individually to ovalbumin for the immunisations and to bovine serum albumin for the ELISA test [50,52]. The immunisation of mice, production of hybridoma cells and ELISA screening were carried out essentially as described previously except that the antigen was given by subcutaneous route [47]. The hybridomas were cloned by limiting dilution until a single ELISA positive colony was grown in a 96 well plate. Individual cell lines were then grown at high cell density for the preparation of the antibody stock which was used subsequently for their characterisation by immunoblotting and immunohistochemistry. All the antibodies are now commercially available from Vertebrate Antibodies.

Immunoblotting
Whole cell lysates from cells (human embryonic kidney cells-HEK 293) overexpressing CYP2R1, CYP7B1, CYP8B1, CYP27A1, CYP39A1, CYP46A1 and CYP51A1 respectively were used as positive controls for immunoblotting while lysates from cells containing vector only were used as negative controls. The cell lysates and their corresponding controls were obtained from (Novus Biologicals, Cambridge, UK). Cell lysates (5 μg protein/ lane) were resolved by electrophoresis on NuPAGE 4-12% Bis-Tris gels (Fisher Scientific, Loughborough, UK). Following protein transfer to nitrocellulose membrane the membranes were washed for 45 minutes at room temperature in phosphate buffered saline-Tween-20 (PBST) containing 3 % (w/v) skimmed milk powder to block non-specific protein binding. Membranes were incubated overnight at 4°C with individual monoclonal antibodies diluted in PBST (1/2 dilution) and then washed 6 times for a total of 60 minutes in 1% skimmed milk. The membranes were subsequently probed for 60 minutes with a secondary antibody conjugated horseradish-peroxidaseconjugated anti-mouse IgG (1/2000, Sigma-Aldrich, Dorset, UK). Membranes were then washed (6 times) for a total of 60 minutes in 1% skimmed milk and protein bands visualized using the enhanced chemiluminescence detection system (Fisher Scientific) [26,49].

Colorectal cancer tissue microarray
The patient cohorts of mismatch repair proficient and deficient colorectal cancers included 650 patients with a histologically confirmed diagnosis of primary colorectal cancer. All patients had undergone elective surgery for primary colorectal cancer, at Aberdeen Royal Infirmary (Aberdeen, UK), between the years of 1994 and 2009. Any patients who had received neoadjuvant chemotherapy and/or radiotherapy were excluded. Tissue was obtained retrospectively from the Aberdeen Colorectal Tumour Bank the Grampian Biorepository (www.biorepository.nhsgrampian.org), Aberdeen, UK provides governance for this tissue bank, see ethics statement). Survival information, in the form of all-cause mortality (i.e. overall survival), was available for each patient. At the time of censoring patient outcome data there had been 309 (47.5%) deaths. The mean patient survival was 115 months (95% CI 108-123 months). The study was conducted according to REMARK criteria and clinico-pathological characteristics of the patients and their tumours and relationship with survival are detailed in Table 8 and Supplementary Table 1.
The histopathological reporting of the tumours was conducted in line with The Royal College of Pathologists UK guidelines for the histopathological reporting of colorectal cancer resection specimens and incorporating guidance from TNM5 [53]. The histopathological processing of the colorectal cancer excision specimens is detailed in Supplementary information Materials and Methods S1.
A tissue microarray was constructed as described from blocks of formalin fixed, paraffin embedded tissue specimens and included 650 primary colorectal cancers, 285 lymph node metastasis and 50 samples of morphologically normal colonic mucosa obtained from resection specimens at a site at least 10cm distant from the tumour [26,54,55]. Two cores each measuring 1mm in diameter were examined per primary tumour, lymph node and normal tissue sample as detailed in the Supplementary information Materials and Methods S1.

Immunohistochemistry
Immunohistochemistry for each antibody was carried out using the Dako EnVision™ system (Dako, Ely, UK) using a Dako autostainer as previously described [26,49,54]. Sections were soaked in xylene to remove paraffin then rehydrated in alcohol prior to immunohistochemistry. When required (CYP7B1, CYP27A1, CYP39A1, CYP46A1 and CYP51A1), antigen retrieval was  performed by microwave oven heating in a citrate buffer solution. Slides were fully immersed in pre-heated citrate buffer (pH 6) then heated in an 800W microwave at full power for 20 minutes. The slides were then allowed to cool at room temperature and placed in cold running water to complete the cooling process. Antigen retrieval was not required for monoclonal antibodies to CYP2R1 and CYP8B1. The initial step in the automated staining protocol was a wash buffer rinse (Dako). Next, slides were incubated with the primary antibody for 60 minutes then washed with buffer. Each primary antibody was applied as undiluted tissue culture supernatant. Slides were then washed in buffer and peroxidase enzyme block was applied for a period of 7 minutes and slides were then again rinsed with wash buffer. Subsequently, the prediluted peroxidase labelled polymer coupled to goat anti-mouse/rabbit secondary antibody was applied for 30 minutes then rinsed with buffer to remove any unbound antibody. The diaminobenzidine (DAB) substrate was then applied for 7 minutes to demonstrate sites of peroxidase activity before a final wash with buffer then water. Slides were immersed in 0.5% copper sulphate for 2 minutes to intensify the DAB stain then washed with running water. Finally, slides were immersed in filtered Harris haematoxylin to lightly counterstain the nuclei before being dehydrated in alcohol and xylene and mounted. Omitting the primary monoclonal antibody from the immunohistochemical procedure and replacing it with antibody diluent (Dako) acted as a negative control. Normal liver was used as a positive control for CYP2R1, CYP7B1, CYP8B1, CYP39A1, CYP46A1 and CYP51A1. Grade 3 breast cancer of no special type was used a positive control for CYP27A1.
Immunohistochemistry was also performed on whole sections of a sub-set of tumours to investigate possible intra-tumour heterogeneity. Sections from the same tissue blocks from which tissue cores were obtained were used.
Following the completion of the immunohistochemistry protocol, the slides were examined by light microscopy using an Olympus BX 51 light microscope (Olympus, Southend-on-Sea, Essex, UK) equipped with an Olympus C4040 camera (Olympus). The intensity of immunostaining was quantified using a semi-quantitative scoring method as previously described [26,49,54]. The intensity of immunostaining (negative, weak, moderate, strong) and its localisation (cytoplasmic, nuclear, membranous) was assessed in the first instance by one investigator (RS). Following this primary scoring, a second investigator (GIM) independently scored each pair of cores. In the case of any discrepancies (less than 5 % of cases, kappa=0.931), both investigators simultaneously re-assessed the core in order to reach an agreed score. The highest scoring core for each individual tissue sample was recorded.

Assessment of mismatch repair protein status
Mismatch repair protein status had previously been assessed by immunohistochemistry using antibodies to MLH1 and MSH2 [26]. Mismatch repair protein status was recorded as either proficient or defective.

Statistics
Statistical analysis of the data including the Mann-Whitney U test, Wilcoxon signed rank test, chi-squared test, Kaplan-Meier survival analysis, log-rank test and Cox multi-variate analysis (variables entered as categorical variables) including the calculation of hazard ratios and 95% confidence intervals was performed using IBM SPSS version 22 for Windows 7™ (IBM, Portsmouth, UK). The log rank test was used to determine survival differences between individual groups. A probability value of p≤0.05 was regarded as significant. The influence of different cut-off points in relation to survival was investigated by dichotomising the immunohistochemistry intensity score for each marker. The groups that were analysed were negative staining versus any positive staining, negative and weak staining versus moderate and strong staining and negative, weak and moderate staining versus strong staining.
Unsupervised hierarchical cluster analysis was carried out using the within-group average linkage method with Pearson correlation as the cluster measure and cluster analysis was performed without any transformation of the data or imputation of missing values.

Ethics
The colorectal cancer tissue microarray is held under the auspices of the Grampian Biorepository which has delegated research ethics authority (11/NS/0015) from The North of Scotland research ethics committee to approve research projects involving human tissue and data. This project was approved by the Grampian Biorepository scientific access group committee (Tissue request No. 0002). Written consent for the formalin fixed wax embedded tissue samples included in the colorectal cancer tissue microarray was not required.

ACKNOWLEDGMENTS
The immunohistochemistry was performed with the support of the Grampian Biorepository.

CONFLICTS OF INTEREST
Abdo Alnabulsi is a PhD student supported by Vertebrate Antibodies, Beatriz Cash and Ayham Alnabulsi are employees of Vertebrate Antibodies (www.vertebrateantibodies.com) and Graeme Murray is a scientific advisor to Vertebrate Antibodies. Rebecca Swan has no conflicts of interest to declare.

GRANT SUPPORT
Rebecca Swan was supported by the Jean Shanks Foundation. This study was supported by funding from Friends of Anchor and the Encompass kick start and SMART:Scotland award schemes of Scottish Enterprise.