Development and analysis of lipidomics procedures for the causal investigation of Alzheimer's disease

Abstract

Uncovering sets of molecular features which cause a healthy metabolic state to transition to one of disease, requires extensive experimentation and often presents a difficult analysis. In the case of neurodegenerative diseases, such as Alzheimer's Disease, simply obtaining suitable samples can be a challenging endeavor. Many current 'Omics' techniques excel at profiling a vast array of molecules, such as water-soluble metabolites, lipids, and proteins, in order to compare groups of samples from healthy and diseased organisms. Such approaches primarily use various associations between molecules and disease to identify biomarkers. However, these 'omics' experiments frequently result in intriguing biological hypotheses, but to date have rarely provided mechanistic explanations. How then, can mechanistic explanations be recovered from metabolite or lipid profile data? In our work, we applied these methods to 6 Alzheimer's diseased brain samples and 6 age matched controls. When analyzed via mass spectrometry, lipids which differed significantly between control and disease were identified, but this information was not able to'provide mechanistic insight. The beginning of any 'omics' based experiment starts with the extraction of the desired molecules. In order to assess the efficiency of three different lipid extraction methods, a lipid standard was extracted from a matrix composed of rat liver tissue and analyzed by mass spectrometry. The classic Folch extraction was found to be best at reproducibly extracting a wide range of lipids. Several of the lipids identified from human brains showed oxidative damage. Lastly, 5 statistical measures of dependence and 3 network algorithms were investigated for their ability to reconstruct mechanistic relationships in a dynamic model of arachidonic acid metabolism. Many of the metabolites of arachidonic acid are oxidation products. Under conditions of high noise and relatively few samples, standard measures of correlation, such as Pearson's correlation, Spearman's correlation and Kendall's Tau were found to perform the best. Metrics which incorporate nonlinear metabolic relations and network algorithms were found to be applicable, when sample size is large and the signal to noise ratio is close to l.