The overall response rate to fluoropyrimidine monotherapy in colorectal cancer (CRC) is limited. Transcriptomic datasets of CRC patients treated with 5-fluorouracil (5FU) could assist in the identification of clinically useful biomarkers. In this research, we aimed to analyze transcriptomic cohorts of 5FU-treated cell lines to uncover new predictive biomarker candidates and to validate the strongest hits in 5FU-treated human colorectal cancer samples with available clinical response data. We utilized an in vitro dataset of cancer cell lines treated with 5FU and used the reported area under the dose-response curve values to determine the therapeutic response to 5FU treatment. Mann-Whitney and ROC analyses were performed to identify significant genes. The strongest genes were combined into a single signature using a random forest classifier. The compound 5-fluorouracil was tested in 592 cell lines (294 nonresponders and 298 responders). The validation cohort consisted of 157 patient samples with 5FU monotherapy from three datasets. The three strongest associations with treatment outcome were observed in SHISA4 (AUC = 0.745, p-value = 5.5 × 10-25), SLC38A6 (AUC = 0.725, p-value = 3.1 × 10-21), and LAPTM4A (AUC = 0.723, p-value = 6.4 × 10-21). A random forest model utilizing the top genes reached an AUC value of 0.74 for predicting therapeutic sensitivity. The model correctly identified 83% of the nonresponder and 73% of the responder patients. The cell line cohort is available and the entire human colorectal cohort have been added to the ROCPlot analysis platform. Here, by using in vitro and in vivo data, we present a framework enabling the ranking of future biomarker candidates of 5FU resistance. A future option is to conduct an independent validation of the established predictors of resistance.
Keywords: RNAseq; gene expression; pharmacology; proliferation; receiver operator characteristics.