Serum Metabolite Profiling Of The Hepatitis B Virus Related Cirrhosis

Objectives New and effective diagnostic and monitoring methods have been in pressing need as for the early diagnosis and therapy of hepatitis B virus related cirrhosis. With the development of metabolomics, more and more focuses have been put on how to imply this method in medical studies and practices.this study, through analyzing the hepatitis B virus related cirrhosis patient serum metabolite profiles tested by the high performance liquid chromatography combined with a LTQ Orbitrap XL mass spectrometry (HPLC-LTQ Orbitrap XL MS), is to explore how to use the metabolomics method in investigating the mechanism of diseases development through comparing different clinical disease states and discovering feature metabolites for monitoring and assessing cirrhosis development and constructing discriminate models based on the selected feature metabolites using pattern recognition methods and machine learning methods.Methods A HPLC-LTQ Orbitrap XL MS platform was used to analyze the serum specimens from 19 healthy people and 81 hepatitis B virus related cirrhosis, in which the number of patients at Child-Pugh A, B or C grade is 37,33 and 11 respectively. 20% samples were selected randomly form each group as a testing group not involved in the data analyzing procedures to test the models constructed in the study. Based on the left 80% samples called training group samples, principle component analysis (PCA) and orthogonal partial least square (OPLS) methods were first used to analyze the variation of the metabolite profiles data, then 3 two-class orthogonal partial least square discriminate analysis (OPLS-DA) models were constructed to discriminate cirrhosis Child-Pugh grade B and A, C and B, A and healthy control respectively and another four-class OPLS-DA model was also constructed by the four groups to observe the cluster patterns of all training samples, and meanwhile these models were all tested by testing group samples. After that, the VIP (variable importance in projection) values, variable confidence intervals in the VIP plot and coefficient plot and non parameter test were used to select feature ions which can separate different Child-Pugh grades in the two-class OPLS-DA models, and these ions were then identified and assigned to corresponding metabolite. At last,3 two-class OPLS-DA models and 3 binary SVM (support vector machine) classifiers were constructed based on the feature ions to discriminate cirrhosis Child-Pugh grade B and A, C and B, A and healthy control respectively, and the performances of the models were tested by n-fold cross validation.Results The performance of the HPLC-LTQ Orbitrap XL MS platform is drifted with time, which is the greatest disturbance factor of the experiment, whose effects however are counteracted between different sample groups due to the random injection of samples. The 4 OPLS-DA models based on metabolite profiles data all have good prediction abilities (Q2 for the 4 models is 89.5%,66.4%,90.3%,80.8% respectively), and the correct classification rates of the 4 models on testing group samples are all 100%. In total,197 feature ions were selected, of which 68 metabolites were identified including amino acid, bile acid, hormone and lipid. The products of lipid metabolism is the biggest part accounting for 30, of which 26 are phospholipids including 15 Lysophosphatidylcholine having different combinations of fatty acids of varying lengths and saturation, which have a decreasing trend with the severity of cirrhosis. The 3 two-class OPLS-DA models based on the 197 feature ions all have better prediction ability (Q2=90.4%,72.8%,93.4%), and the sensitivity, specificity and correct classification rate of 3 binary SVM classifier can all achieve 100%.Conclusions (1) The study verified that reasonable experiment design and proper usage of the random principle can counteract the influences from no-interested factors on metabolite profiling studies. (2) A feature metabolites discovery method by OPLS-DA model is reported. (3) Testing the changes of lysophospholipid profile by HPLC-MS platform based metabolomics method may be a promising tool in monitoring the development of cirrhosis. (4) The HPLC-MS based feature ions OPLS-DA models and SVM classifiers can be promising new approaches in assessing and monitoring cirrhosis development.