| Pancreatic cancer (PC) is a fatal disease and ranks as the fourth leading cause of death from cancer in the western world. Nearly 80% of PC patients have impaired glucose metabolism or diabetes mellitus (DM), so that the association between diabetes and pancreatic cancers is exciting opportunity as identification of modifiable prognostic factors can lead to better understanding of the biology of cancer, thus achieving an earlier diagnosis and more satisfactory therapies in this disease.Metabonomics is a branch of science concerned with the study of systems biology. It concerns the study of low molecular weight compounds (typically < 1000 Da) in biofluids and other complex matrixes. Many metabolites are the final downstream products of genome and reflect best the operation of the biological system. To date, global metabolic profiling of human biofluids has been increasingly used as an effective tool for disease diagnosis to elucidate significant changes in tumor metabolism, to explore candidate“biomarkers”from variance within a huge number of endogenous metabolites and to characterize the biological pathways. With the aid of the recent advent of technologies for quantitative and comprehensive metabolites analyses, which provides a more sensitive and specific diagnosis than single biomarker approaches, and reveal distinct differences in metabolism between diseased individuals and healthy ones in a novel systems-levelIn the present study, pressurized capillary electrochromatography (pCEC) was used in the plasma metabolomics study, verifying the high performance of the system in separation of endogenous metabolites and its ability as an early diagnostic method for pancreatic cancer. Moreover, through studying the metabolic variations in serums of pancreatic cancer patients, type 2 diabetes mellitus patients and healthy volunteers with UPLC-MS and GC-MS metabolomic technology, more disease information could be gain to help finding the common and specific biomarkers among the diseases, which gave new eyes on the understanding of the mechanism between the two diseases. Meanwhile, the classification performance of nonlinear artificial neural network (ANN) on the three sample groups was studied. Main methods and results are:1. The application of pCEC coupled with Ultra-violet (UV) detection has been investigated for the production of global metabolite profiles from human plasma, and its capabilities of classification pancreatic cancer patients. The pCEC separation of plasma samples was performed on a reversed phase (RP) column with gradient elution. The applied voltage, detection wavelength and type of acidic modifiers on separation of plasma samples were optimized with pooled quality control (QC) sample. The stability and repeatability of the methodology were also determined by the repeat analysis of QC sample. The effects of different scaling methods on the results of orthogonal partial least square discrimination analysis (OPLS-DA) based on pCEC-UV data set were also investigated. The results of current study clearly showed the different phenotypes of metabolites of pancreatic cancer patients and healthy controls based on pCEC-UV plasma profiles. OPLS-DA data are shown to provide a valuable means of convenient classification. This work indicated that pCEC-UV method can be used as a cost-effective and information-rich, while relatively simple and inexpensive approach for plasma profiling on disease metabolomics studies.2. A UPLC Q-TOF MS system was employed to distinguish the serum global profiles of 20 pancreatic cancer patients, 19 type 2 diabetes mellitus patients and 25 heatlty volunteers. OPLS-DA was used for group differentiation and potential biomarkers selection. As shown in the scores plot, the distinct clustering among pancreatic cancer, diabetes mellitus patients and health controls was observed. Bile acids, free fatty acid and lysoPC were tentatively identified based on accurate mass, isotopic pattern and MS/MS information. Especially, isomeric lysoPCs were distinguishen based on retention time and peak intensity ratio of product ions, and 12 pairs of lysoPCs regioisomers were identified in human serum. Bile acids were specially increased in pancreatic cancer patients compared with diabetes and health control. Though the levels of some lysoPCs decreased both in pancreatic cancer and diabetes patients, the ratio variation of lysoPC18:1 isomers differed between the two groups of pacients. This study indicated the potential of metabolomic strategies for explanation of the mechanism of diseases and detection of specifical biomarkers.3. The GC-MS based serum metabolomics approach was also used to investigate the pathophysiological variation among pancreatic cancer, type 2 diabetes mellitus patients compared to health control. The derivatization condition was optimized. Pattern recognition was carried out with PCA and OPLS-DA to get the similar results as UPLC/MS metabolomics study, clearly classify the three groups of serum samples. The metabolites got from this test were different from those from UPLC/MS, maily including amino acid, glucose, cholesterol and organic acid. Compared with healthy controls, the sucrose metabolism,glycolysis,and amino acid metabolism of pancreatic cancer and diabetes mellitus patients were disordered to some degree. Significantly altered serum metabolites in pancreatic cancer subjects include increased leucine, phenylalanine, and hexadecanioc acid. Glycine, glutamine, serine, proline, and critic acid decreased in diabetes subjects. The disordered metabolites including increased cholesterol, fructose, and glucose and decreased lactate were detected in both serum groups. It indicated that both common and specific metabolic pathways of pancreatic cancer and diabetes mellitus were disordered. The relationship between the two diseases could be further studied according to the results of the present test.4. Based on the 26 metabolites obtained from UPLC/MS study, the classification models of various samples were established with nonlinear ANN. The influence of different number of hidden layers and different number of nodes in hidden layers on the convergence of ANN and classification results was compared. It was found that the three-layer ANN with less hidden layers was not sufficient to classify the samples of pancreatic cancer patients, type 2 diabetes mellitus patients and heatlty volunteers, with which the percent of accuracy was only 86.7%, while the percent of accuracy with the four-layer ANN with the optimal structure {26,9,7,3} was up to 96.7%. With sufficient samples, the supervised ANN may be a powerful classification tool in the metabolomics study. |
PDF Full Text Request