Novel Metabolomics Serum Biomarkers for Pancreatic Ductal Adenocarcinoma by the Comparison of Pre-, Postoperative and Normal Samples

Abstract

Background: Pancreatic ductal adenocarcinoma (PDAC) is one of the most aggressive human malignancies. The metabolomic approaches are developed to discover the novel biomarkers of PDAC. Methods: 550 preoperative, postoperative PDAC and normal controls (NCs) serums were employed to characterize metabolic alterations in training and validation sets by LC-MS. Results: The results of PLS-DA analysis indicated that three groups could be distinguished clearly and the post-PDAC group is adjacent to a normal group as compared with pre-PDAC group. Further results showed that histidinyl-lysine significantly increased whereas docosahexaenoic acid and LysoPC (14:0) decreased in pre-PDAC patients as compared with NCs. And these three markers had a significant tendency to recover after tumor resection. The validation set results revealed that for CA19-9 negative patients, 92.3% (12/13) of them can be screened using these three metabolites. The combination of these markers could significantly improve the diagnostic performance for PDAC, with higher sensitivity (0.93), specificity (0.92) and AUC (0.97). Moreover, network and pathways analyses explored the latent relationship among differential metabolites. The glycerolipid metabolism and primary bile acid synthesis showed variation in network and pathway analysis. Conclusions: These three markers combined with CA199 displayed high sensitivity and specificity for detecting PDAC patients from NCs. The results indicated that these three metabolites could be regarded as potential biomarkers to distinguish PDAC from NCs.

Keywords: Biomarkers; Metabolomics; Multivariate analysis; Pancreatic Ductal Adenocarcinoma.

Figure 1

The workflow of the metabolomics data analysis.

Figure 2

Metabolic profiling analysis among NC, pre-PDAC and post-PDAC groups. The score plot for PLS-DA (A) to discriminate pre-PDAC (n=185) and NC(n=146); and cross-validation plot obtained from 100 permutation tests (B); The score plot for OPLS-DA (C) to discriminate pair-wise pre-PDAC (n=87) and post-PDAC (n=87); and cross-validation plot obtained from 100 permutation tests (D); Three-dimensional score plot for PLS-DA (E) to discriminate pre-PDAC (n=87), post-PDAC (n=87) and NC (n=146); and cross-validation plot obtained from 100 permutation tests (F).

Figure 3

HCA-heatmap and the expression of discriminative metabolites. (A) HCA-heatmap plot indicating relative levels of differential metabolites in samples of the training set. (B) Box plots for comparing concentration levels of the three discriminative metabolites in different groups in the training set. (C) Box plots for comparing concentration levels of the three discriminative metabolites in different groups in validation set. * p< 0.05 ** p<0.01 *** p< 0.001 **** p< 0.0001.

Figure 4

Evaluation of potential biomarkers. Classiers are the biomarker signature generated in the training set and presented here for the validation set. (A) Scatter plot for graphical representation of the biomarker signature score. Y-axis score of biomarker signature with the cut-off ≥ 0.41 and CA19-9 on the X axis with the cut-off ≥ 37 U/mL (< 37µg/mL as CA19-9-negative). (B) Blue circles are pancreatic cancer (n=34).

Figure 5

Analyzed of the correlation network and pathways altered in PDAC. (A) Correlation network constructed with 18 differential metabolites (Pearson correlation analysis, |r| > 0.6). Blue sub-network constructed with glycerophospholipids (LysoPCs and LysoPEs). Red sub-network constructed with steroids (ST) and bile acid (BA). Nodes in red and blue represent the metabolites down-regulated and up-regulated in PDAC, respectively. (B) Significantly changed pathways. Disordered pathways in PDAC group; small p value and big pathway impact factor indicate that the pathway is greatly influenced.