Brain region mapping using global metabolomics

Abstract

Historically, studies of brain metabolism have been based on targeted analyses of a limited number of metabolites. Here we present an untargeted mass spectrometry-based metabolomic strategy that has successfully uncovered differences in a broad array of metabolites across anatomical regions of the mouse brain. The NSG immunodeficient mouse model was chosen because of its ability to undergo humanization leading to numerous applications in oncology and infectious disease research. Metabolic phenotyping by hydrophilic interaction liquid chromatography and nanostructure imaging mass spectrometry revealed both water-soluble and lipid metabolite patterns across brain regions. Neurochemical differences in metabolic phenotypes were mainly defined by various phospholipids and several intriguing metabolites including carnosine, cholesterol sulfate, lipoamino acids, uric acid, and sialic acid, whose physiological roles in brain metabolism are poorly understood. This study helps define regional homeostasis for the normal mouse brain to give context to the reaction to pathological events.

Figure 1. Global metabolomic approach for regional mapping of brain tissue

This workflow integrates two complementary technologies, untargeted LC/MS profiling using hydrophilic interaction chromatography (HILIC) coupled to electrospray ionization (ESI) mass spectrometry and Nanostructure Imaging Mass Spectrometry (NIMS). Following the heat fixation (FBMI), left brain hemisphere was dissected and each brain region was extracted separately for untargeted LC/MS profiling (mice specimens = 5). The intact, right brain hemisphere was imaged by laser desorption/ionization mass spectrometry to create maps of spatial distribution of metabolites of interest across the brain and within each sub-region.

Figure 2. Representation of global metabolomic data with a focus on significant differences in metabolite patterns across brain regions

A) Multi-group cloud plot showing differentially expressed metabolite features (bubbles) across different regions of brain (level of significance: p ≤ 0.01, Intensity > 20,000 ion counts). Metabolite features are projected depending on their m/z ratio and retention time. The color of the bubble indicates the level of significance (p-value), with darker color (in red tones) representing lower p-values. A Total Ion Chromatogram (TIC) is shown in the background. B) Variation patterns of two characteristic, lipid and water soluble brain metabolites across eight different regions of brain. Metabolites were putatively identified based on MS/MS data provided in Supplemental Figure S5 (NAAG) and in Table S2 (PS 22:6/22:6). ANOVA was used to calculate the statistical significance for n=5 in each group. Box and Whisker plots display the full range of variation (whiskers: median with minimum -maximum; boxes: interquartile range).

Figure 3. Heat map and associated dendograms representing the hierarchically clustered samples based on the similarity of metabolite patterns

Discriminating metabolites (ANOVA p ≤ 0.01, q ≤ 0.001, Intensity > 20,000) are shown on the right side. Isotopes, adducts, in-source fragments and multiply charged features were filtered out. MS/MS data for identified metabolites are provided in Supplemental Figure S4 and S5. MS/MS patterns for putative identifications of phospholipids are given in Table S2 and Table S3. Variation patterns of identified metabolites (with the exception of phospholipids) are presented by Box-Whisker plots in Figure S1 and S2.

Figure 4. Supervised pattern recognition OPLS-DA model for metabolomic profiles of brain regions

OPLS scores plot of HILIC-ESI-MS profiles (aligned by XCMS) shows discrimination among specific regions on the first and second component. The model brings out the specific variation of the metabolite composition according to the brain region (five biological replicates per region). Good separation was achieved for three different regions: midbrain, brain stem and cerebellum. All aligned metabolite features (15,835) were used to create the model and a total of five orthogonal components were calculated for cross-validation (R2Y(cum) = 0.68, Q2Y(cum) = 0.35).

Figure 5. Laser desorption/ionization MS images or maps of ions of interest

Extracted maps from the low mass part of total ion spectra (< 500 Da) show the spatial distribution of docosahexaenoic acid (DHA), glutamine, carnosine and docosanoyl taurine. Extracted maps from the higher mass part of total ion spectra (> 800 Da) show the spatial distribution of sulfatide and phosphatidylserine. Images were acquired from a 2 µm thin brain section that was mounted on etched silicon chip, coated with PFUA initiator, prior to imaging in negative ionization mode. Ions of interest were identified using the MS/MS data acquired by LC/MS profiling (Figure S3–S4, Table S3).