Interpreting transcriptional signatures and their relevance to disease biology, prognosis and treatment resistance in colorectal cancer

Abstract

Introduction Colourectal cancer (CRC) is a common heterogeneous malignancy with a number of different disease subtypes influenced by distinct biology. The clinical implications of this disease heterogeneity are likely to influence treatment strategies further in the future. Therefore, patient stratification is paramount in order to improve patient response to treatment, improve survival outcome and reduce the risk of relapse. Computational algorithms using a bioinformatics approach have enabled the stratification of patients using a variety of different classifiers. These classifiers have been applied to patient clinical data with matching transcriptomics and have dichotomised patient cohorts into different subgroups with different clinical markers, biological features and prognostic outcomes. Since its utility was first demonstrated in breast cancer, bioinformatics and patient stratification using gene expression datasets have been extended to study diseases such as CRC. Combining transcriptional and mutational profiles may help to elucidate the underlining biology of different CRC subtypes, with subsequent application in treatment selection. The identification of biomarkers will permit clinicians to more accurately stratify patients into different groups that are more likely to benefit from conventional chemotherapy, radiotherapy, targeted therapy or immunotherapy treatment. Methodology Differential gene expression analysis (DGEA) was applied to drug resistant cell lines comparing sensitive and resistant cells. DGEA was also employed to compare BRAFMT and BRAFWT samples followed by a 1-way ANOVA analysis to extract a leukocyte-derived signature. Classifiers including support vector machines (SVM) and linear discriminate analysis (LDA) were employed to predict different mutational and molecular profiles in CRC. DGEA was also performed to determine the role of GREM1 expression in drug resistance. Hierarchical clustering was used throughout the course of this project using the Partek Genomics Suite software. Additionally, Kaplan-Meier curves were also employed for survival analysis and boxplots were used for determining the level of gene expression in different samples using the Prism software. Additionally, gene set enrichment analysis (GSEA) was utilised. Results In Chapter 3, we employed a series of drug resistance CRC cell line signatures in order to identify the key molecular contributors involved in drug resistance. However, the molecular determinants of drug resistance in independent drug resistant cell lines did not reveal similar molecular pathways when tested in data from patient tumour samples. This indicates that our drug resistant signatures identified in vitro could not recapitulate drug resistance in vivo. This we attributed to clonal diversification across cell lines and importantly the role of the tumour microenvironment (TME). In Chapter 4, we identified a leukocyte derived signature with potential for classifying both BRAFMT status and MSI status. Exploring the underling biology of this signature added further knowledge to the role of the TME in influencing the molecular classification of tumour samples. Previous investigators have published molecular signatures for the classification of both BRAF and MSI status, but have not highlighted in detail the role of the TME in molecular classification. In Chapter 5, a gene called GREM1 was interrogated to determine its role in drug resistance. High expression of GREM1 was found to increase the risk of patient relapse in CRC. Moreover, it was noted that high expression of GREM1 was associated with cancer associated fibroblasts which may explain why GREM1 is associated with a poorer patient prognosis. We also established a link between GREM1 expression and drug resistance. Therefore, GREM1 may act as a potential biomarker and could conceivably be used for patient stratification in the future. However, further verification and validation is required. Once again the role of the TME is paramount to consider when attempting to classify patients as GREM1 originated from cells of the stromal compartment. Conclusions Our data suggest that the TME greatly influences drug resistance in CRC. Overexpression of GREM1, a gene identified from our bioinformatic analysis of drug resistance was correlated with a poorer prognosis. Moreover, we determined that high GREM1 expression was also associated with cancer associated fibroblasts, making it a potential candidate as a biomarker of drug resitance in CRC when considering FOLFIRI treatment.