MicroRNAs as Bile-based biomarkers in pancreaticobiliary cancers (MIRABILE): a cohort study

Abstract

Background: Biliary obstruction can be due to both malignant and benign pancreaticobiliary disease. Currently, there are no biomarkers that can accurately help make this distinction. MicroRNAs (miRNAs) are stable molecules in tissue and biofluids that are commonly deregulated in cancer. The MIRABILE study aimed to identify miRNAs in bile that can differentiate malignant from benign pancreaticobiliary disease.

Materials and methods: There were 111 patients recruited prospectively at endoscopic retrograde cholangiopancreatography (ERCP) or percutaneous transhepatic cholangiography (PTC) for obstructive jaundice, and bile was aspirated for cell-free RNA (cfRNA) extraction and analysis. In a discovery cohort of 78 patients (27 with pancreatic ductal adenocarcinoma (PDAC), 14 cholangiocarcinoma (CCA), 37 benign disease), cfRNA was subjected to small-RNA sequencing. LASSO regression was used to define bile miRNA signatures, and NormFinder to identify endogenous controls. In a second cohort of 87 patients (34 PDAC, 14 CCA, 39 benign disease), RT-qPCR was used for validation.

Results: LASSO regression identified 14 differentially-expressed bile miRNAs of which 6 were selected for validation. When comparing malignant and benign pancreaticobiliary disease, bile miR-340 and miR-182 were validated and significantly differentially expressed ( P <0.05 and P <0.001, respectively). This generated an AUC of 0.79 (95% CI: 0.70-0.88, sensitivity 65%; specificity 82%) in predicting malignant disease.

Conclusion: Bile collected during biliary drainage contains miRNAs able to differentiate benign from malignant pancreaticobiliary diseases in patients with obstructive jaundice. These bile miRNAs have the potential to increase diagnostic accuracy.

Graphical abstract

Figure 1

Study design of the MIRABILE study. (A) A graphical figure demonstrating the overall study design together with relevant pancreaticobiliary diseases. (B) Schematic diagram showing the workflow: RNA was isolated from bile samples, followed by quantification and then either RNA sequencing analysis or RT-qPCR. (C) Venn diagram showing the distribution of patients between the discovery and validation cohorts used in the analysis of bile cfRNA. RT-qPCR, quantitative reverse transcription polymerase chain reaction.

Figure 2

Bile cell-free RNA (cfRNA) demonstrated differences in RNA quantity and expression. (A) Relative distribution of small RNA reads for each bile sample from the discovery cohort (n=78). (B) Principal component analysis (PCA) plot of bile miRNA expression showing clustering of samples. (C) Heatmap with unsupervised hierarchical clustering showing malignant samples grouping to the left and including all miRNAs with adjusted P-values <0.01. CCA, cholangiocarcinoma; dCCA, distal CCA; ERCP, endoscopic retrograde cholangiopancreatography; mRNA, messenger RNA; N/A, not applicable; PC, principal component; PDAC, pancreatic ductal adenocarcinoma; PTC, percutaneous transhepatic cholangiography; rRNA, ribosomal RNA; snRNA, small nuclear RNA; tRNA, transfer RNA.

Figure 3

Differential expression analysis identified differentially expressed miRNAs in bile samples. Venn diagrams for (A) the total number of differentially expressed miRNAs (adjusted P-value <0.05), and (B) the total number of upregulated miRNAs only when comparing PDAC, CCA, and benign cohorts. Volcano plots demonstrate differentially expressed miRNAs with the top 10 most statistically significant miRNAs annotated (upregulated and downregulated) for the pairwise comparisons: (C) malignant (PDAC and CCA) vs. benign disease; (D) PDAC vs. benign; (E) CCA vs. benign and (F) PDAC vs. CCA. Red illustrates miRNAs with FDR<0.05. Vertical dotted lines indicate log2(fold change)=+/− 1. CCA, cholangiocarcinoma; hsa, homo sapiens; PDAC, pancreatic ductal adenocarcinoma.

Figure 4

RT-qPCR validation of candidate miRNAs identified by LASSO regression analysis. (A) Expression of candidate miRNAs in benign and malignant pancreaticobiliary disease (left to right): miR-196a, miR-182, miR-335 and miR-340. *P<0.05, **P<0.005, ***P<0.0005. (B) ROC curve with corresponding AUC value for the 2-miRNA signature (miR-182 and miR-340) to predict malignant disease vs. benign disease (left), or to predict PDAC vs. benign disease (right). AUC, area under the curve; PDAC, pancreatic ductal adenocarcinoma.