Predicting response to irinotecan/5-fluorouracil (5-FU) chemotherapy for advanced colorectal cancer based on gene expression in primary tumour

Background: Despite improvements in the treatment of advanced colorectal cancer (CRC), there remains a subset of patients who fail to benefit from chemotherapy. The identification of predictive response markers has been limited by the relative scarcity of metastatic tumour samples for molecular analysis, prompting a need to consider surrogate models for generation of predictive markers. Methods: We performed gene expression profiling of archived formalin fixed paraffin-embedded (FFPE) colorectal primary tumours (dating from 2001–2007) from 40 patients who received irinotecan/5-FU chemotherapy as first line treatment for advanced disease. Patients were classified as responders/non-responders based on radiological response. Gene expression profiling was performed using the Almac Diagnostics Colorectal Disease Specific Array (DSA). Data was analysed using Genespring GX v7.3. Principal Components Analysis (PCA) was used to separate responding and non-responding patients based on the tumour- derived expression data. Predictive classifiers were constructed using several class prediction methods. The performance of the classifiers was assessed by leave-one-out cross-validation. Results: 37 samples passed data QC assessments for inclusion in predictive analysis. PCA using genes passing a t-test and 1.5-fold change filter demonstrated clear separation between responding and non-responding patients. Predictive modelling using the k-nearest neighbour and Support Vector Machine (SVM) algorithms with Fishers exact test as feature selection method each generated seven different predictive classifiers containing 5 to 35 genes; these had an average predictive accuracy of 80%. Predictive modelling of this dataset is ongoing and these encouraging initial results will be extended to a larger patient cohort. Conclusions: DNA microarray profiling has been used to generate gene signatures predictive of response to irinotecan/5-FU therapy in advanced CRC; importantly, these predictive signatures have been generated from FFPE colorectal primary tumour facilitating their independent validation in large patient cohorts and potential clinical implementation in the event of successful validation.