Analysis and Simulation of Glioblastoma Cell Lines-Derived Extracellular Vesicles Metabolome

Abstract

Glioblastoma (GBM) is one of the most aggressive cancers of the central nervous system. Despite current advances in non-invasive imaging and the advent of novel therapeutic modalities, patient survival remains very low. There is a critical need for the development of effective biomarkers for GBM diagnosis and therapeutic monitoring. Extracellular vesicles (EVs) produced by GBM tumors have been shown to play an important role in cellular communication and modulation of the tumor microenvironment. As GBM-derived EVs contain specific "molecular signatures" of their parental cells and are able to transmigrate across the blood-brain barrier into biofluids such as the blood and cerebrospinal fluid (CSF), they are considered as a valuable source of potential diagnostic biomarkers. Given the relatively harsh extracellular environment of blood and CSF, EVs have to endure and adapt to different conditions. The ability of EVs to adjust and function depends on their lipid bilayer, metabolic content and enzymes and transport proteins. The knowledge of EVs metabolic characteristics and adaptability is essential for their utilization as diagnostic and therapeutic tools. The main aim of this study was to determine the metabolome of small EVs or exosomes derived from different GBM cells and compare to the metabolic profile of their parental cells using NMR spectroscopy. In addition, a possible flux of metabolic processes in GBM-derived EVs was simulated using constraint-based modeling from published proteomics information. Our results showed a clear difference between the metabolic profiles of GBM cells, EVs and media. Machine learning analysis of EV metabolomics, as well as flux simulation, supports the notion of active metabolism within EVs, including enzymatic reactions and the transfer of metabolites through the EV membrane. These results are discussed in the context of novel GBM diagnostics and therapeutic monitoring.

Keywords: extracellular vesicles; glioblastoma; machine learning; metabolism modeling; metabolomics.

Figure 1

Cell and extracellular vesicles (EV) pellets were re-suspended in a Radioimmunoprecipitation assay (RIPA) buffer and subjected to SDS-PAGE and Western blot analysis. Samples were transferred to nitrocellulose membranes and probed with anti-CD9 antibody (1:2000, Abcam) followed by detection using goat anti-rabbit IgG-HRP conjugated secondary antibody. The signals were detected using an enhanced chemiluminescent (ECL) kit.

Figure 2

Analysis of sample differences from 1D Nuclear Overhauser Effect Spectroscopy (NOESY 1D) ¹H NMR spectra of cells, media and EVs for glioblastoma (GBM) cell lines (LN18, A172 and U118) and normal human astrocytes (NHA). (A) Principal component analysis (PCA); (B) t-distributed stochastic neighbor embedding (t-SNE). Grouping of sample source (cells, EV, media) as well as cell type is indicated.

Figure 3

Relative concentrations of metabolites determined and quantified from NMR spectra. Metabolites are ordered using hierarchical cluster analysis across all samples. Values are scaled to mean of 0 and standard deviation of 1 across all samples and metabolites. AC—Adenosylhomocysteine; 2HG —2-hydroxyglutarate; GSSG—Glutathione (oxidized); GSH—Glutathione (reduced); G6P—Glucose 6-phosphate; PC—Phosphorylcholine.

Figure 4

ANOVA selection of the most different metabolites between cell and EV extracts for GBM cell lines. Shown are metabolites with ANOVA > 5.

Figure 5

(A) ANOVA selection of the most different metabolites between cell four cell types in cells and EV extracts. Shown are metabolites with ANOVA > 5.G6P—Glucose 6-phosphate; PC—Phosphorylcholine; GSSG—Glutathione (oxidized); GSH—Glutathione (reduced). (B) PCA representation of sample groups obtained from the most significantly different metabolites in four sample groups for cells and EVs, presented in A.

Figure 6

(A) ANOVA analysis of major metabolic differences between EVs, media and cells for LN18 and U118 lines. EVs show the biggest difference between two cell lines with the number of metabolites having an ANOVA value over 5. For cells and media, differences are much more subtle, with only a small number of metabolite showing an ANOVA value over 3. (B) The difference between correlation coefficients of metabolites in EV and media of U118 and LN18 cells, where a positive value (red) indicates a higher correlation in U118 cell lines and a negative value (green) shows a higher correlation in LN18 cells.

Figure 7

Metabolic pathways with the most significant enrichment for proteins found in EVs from LN18 but not EVs from U118 cells and metabolites showing different concentration between these two groups of EVs (Figure 6).

Figure 8

Part of the TCA-cycle-related metabolic processes. GIMME calculated fluxes in LN18 and U118 EVs made possible based on previously determined proteins in these vesicles. GIMME analysis provides a flux model with a minimized use of low-expression reactions while maximizing the objective reaction, in this case biomass preservation. Reactions are shown using the modeldraw.rxns routine in COBRA running under Matlab. Excluded from the representation are the cofactors including CO₂, H₂O, ATP, ADP, NAD, NADH, NADPH, NADP, H, Pi. In the figure, rectangles represent reactions with rates of fluxes in parentheses; ellipses represent metabolites; the red ellipses represent dead-end metabolites; gray arrows represent zero-rate fluxes; green arrows represent positive-rate (forward) fluxes; and blue arrows represent negative-rate (backward) fluxes. Reactions and metabolites notation is based on the Recon 3 metabolic network and is: akg—oxoglutarate (a-ketoglutarate), icit—isocitrate, cit—citrate, succ—succinate, mal—malate, fum—fumarate, coa—coenzyme A; gtp—guanosine triphosphate; fadh2—Flavin adenine dinucleotide.

Figure 9

Part of the Glutathione metabolism flux GIMME optimization of flux through remaining Recon 3 reactions based on the present proteins in EVs. Reactions are shown using modeldraw.rxns routine in COBRA running under Matlab. Excluded from the representation are the cofactors including CO₂, H₂O, ATP, ADP, NAD, NADH, NADPH, NADP, H, Pi. Rectangles represent reactions with rates of fluxes in parentheses; the red rectangles represent reactions with only one metabolite; ellipses represent metabolites; the red ellipses represent dead-end metabolites; gray arrows represent zero-rate fluxes; green arrows represent positive-rate (forward) fluxes; and blue arrows represent negative-rate (backward) fluxes. Reactions and metabolites notation is based on the Recon 3 metabolic network and is: gthrd—reduced glutathione; gthox—oxidized glutathione; glu—glutamine; gly—glycine; ala—alanine; cys—cysteine; 5oxpro—5-oxoprolinate; cgly—carbamoyl glycine.

Figure 10

Experimental workflow of sample preparation for cells, media and EV analysis by NMR spectroscopy. Metabolomics, as well as computational flux simulation, allowed the investigation of possibly active pathways in EVs. Shown in red are the number of biological replicates measured for each cell and sample type.