DNA-methylation signature accurately differentiates pancreatic cancer from chronic pancreatitis in tissue and plasma

Abstract

Objective: Pancreatic ductal adenocarcinoma (PDAC) is a lethal malignancy. Differentiation from chronic pancreatitis (CP) is currently inaccurate in about one-third of cases. Misdiagnoses in both directions, however, have severe consequences for patients. We set out to identify molecular markers for a clear distinction between PDAC and CP.

Design: Genome-wide variations of DNA-methylation, messenger RNA and microRNA level as well as combinations thereof were analysed in 345 tissue samples for marker identification. To improve diagnostic performance, we established a random-forest machine-learning approach. Results were validated on another 48 samples and further corroborated in 16 liquid biopsy samples.

Results: Machine-learning succeeded in defining markers to differentiate between patients with PDAC and CP, while low-dimensional embedding and cluster analysis failed to do so. DNA-methylation yielded the best diagnostic accuracy by far, dwarfing the importance of transcript levels. Identified changes were confirmed with data taken from public repositories and validated in independent sample sets. A signature of six DNA-methylation sites in a CpG-island of the protein kinase C beta type gene achieved a validated diagnostic accuracy of 100% in tissue and in circulating free DNA isolated from patient plasma.

Conclusion: The success of machine-learning to identify an effective marker signature documents the power of this approach. The high diagnostic accuracy of discriminating PDAC from CP could have tremendous consequences for treatment success, once the result from still a limited number of liquid biopsy samples would be confirmed in a larger cohort of patients with suspected pancreatic cancer.

Keywords: chronic pancreatitis; pancreatic cancer.

Figure 1

Low-dimensional embedding of cohort samples using UMAP dimensionality reduction of different ‘omic’ datasets. Results are shown that were obtained by applying DNA-methylation, mRNA expression or miRNA expression data, respectively, or by using combined mRNA expression and DNA-methylation or combined mRNA and miRNA expression data. Individual tissue samples are colour-coded as indicated: PDAC, red; CP, blue; N, green. CP, chronic pancreatitis; mRNA, messenger RNA; miRNA, microRNA; N, healthy individuals; PDAC, pancreatic ductal adenocarcinoma; UMAP, Uniform Manifold Approximation and Projection.

Figure 2

Schematic workflow of biomarker selection process. Single-‘omic’ and multi-‘omic’ data from PDAC, CP and N tissue samples were collected and analysed. The workflow consists of four main steps: feature selection, model training, internal validation and validation on independent datasets. AUC, area under the curve; cfDNA, cell-free DNA; CP, chronic pancreatitis; mRNA, messenger RNA; miRNA, microRNA; N, healthy individuals; PDAC, pancreatic ductal adenocarcinoma.

Figure 3

Evaluating the machine-learning performance on PDAC and N samples. (A) Comparison of 10-fold cross-validated model performances. For each ‘omic’ dataset, the relationship between the model performance measured by AUC value and the feature number is indicated. For each analysis, the highest AUC value achieved with the smallest possible number of features is marked by a rectangle. (B) Unsupervised hierarchical clustering of the two CpG marker sites cg02964172 and cg17184704 as defined by the best performing DNA-methylation model. (C) Left, the ROC curve of the diagnostic prediction model is shown as calculated with the two CpG sites in the public validation dataset GSE49149. On the right, the methylation levels of the two CpGs are shown as determined in PDAC and N samples from the public validation dataset. Mean methylation values are indicated by horizontal lines. Differential analysis was performed by t-test (***p<0.001, ****p<0.0001). AUC, area under the curve; mRNA, messenger RNA; miRNA, microRNA; N, healthy individuals; PDAC, pancreatic ductal adenocarcinoma; ROC, receiver operating characteristic.

Figure 4

Identification and validation of biomarkers for the differentiation of PDAC and CP. (A) Comparison of 10-fold cross-validated model performances. For each ‘omic’ dataset, the relationship between the model performance measured by AUC value and the feature number is indicated. The highest AUC value with the smallest possible number of features is marked by a rectangle. (B) Unsupervised hierarchical clustering of five methylation markers selected from the best predictive models in the DNA-methylation and the combined DNA-methylation and mRNA expression datasets. (C) Left, the normalised methylation index is shown of five selected methylation markers in 24 PDAC and 24 CP samples. Experimental validation was by MethyLight qPCR. The mean values of normalised methylation indexes are indicated by horizontal lines. Differential analysis was performed by t-test (****p<0.0001; ns, not significant). In the right panel, ROC curves are shown of single and combined (red) predictive markers cg03306374 (green) and cg15506157 (blue) as calculated from the independent validation dataset. AUC, area under the curve; CP, chronic pancreatitis; mRNA, messenger RNA; miRNA, microRNA; PDAC, pancreatic ductal adenocarcinoma.

Figure 5

Regional plot of an epigenome-wide association analysis of cg03306374 and cg15506157 in PDAC and CP tissues. Co-methylation patterns were found in the PRKCB (A) and KLRG2 gene regions (B), respectively. At the very top, the gene position within the respective chromosome is shown. Below, each dot stands for a particular CpG site. Its genomic position is indicated along the X-axis. The negative log-transformed p value reflects the association of a CpG site with the disease status. The panels in the middle show the ENSEMBL annotation tracks including genes/transcripts and the direction of transcription. Also, the positions of the CpG sites (vertical lines) are indicated. The lower panels present Spearman’s correlation coefficients of DNA-methylation levels between selected CpG sites in the two genomic regions. The colour scheme of the heatmap is reflected also in the association panel at the top with respect to correlation to the reference CpG sites cg03306374 and cg15506157. Blue colouring of a dot stands for low correlation values; red colouring indicates high correlation values. CP, chronic pancreatitis; PDAC, pancreatic ductal adenocarcinoma.

Figure 6

Differentiation between PDAC and CP based on DNA-methylation. ROC curves are shown that were calculated on the basis of differential DNA-methylation at cg15506157 and the three other CpGs in the KLRG2 gene region (red line), cg03306374 and the five surrounding DNA-methylation sites in the PRKCB gene region (green line) or a combination of all 10 sites (blue line). AUC, area under the curve; CP, chronic pancreatitis; PDAC, pancreatic ductal adenocarcinoma.

Figure 7

Diagnostic accuracy of DNA-methylation in liquid biopsy samples. Variation in DNA-methylation in cell-free DNA isolated from patient plasma is presented as box plot for the six CPG sites in gene PRKCB and the four CpG sites in KLRG2. The M-value was chosen as a measure of methylation. At the top, the respective false discovery rate adjusted p values are given. Only the first three numbers of the CpG identifiers are indicated: cg033: cg03306374; cg031: cg03156893; cg032: cg03217795; cg054: cg05436658; cg095: cg09507526; cg213: cg21370856; cg155: cg15506157; cg006: cg00699934; cg009: cg00919016; cg052: cg05224190. CP, chronic pancreatitis; PDAC, pancreatic ductal adenocarcinoma.