A Metabolomic Signature of Ischemic Stroke Showing Acute Oxidative and Energetic Stress

Abstract

Metabolomics is a powerful data-driven tool for in-depth biological phenotyping that could help identify the specific metabolic profile of cryptogenic strokes, for which no precise cause has been identified. We performed a targeted quantitative metabolomics study in West African patients who had recently suffered an ischemic stroke, which was either cryptogenic (n = 40) or had a clearly identified cause (n = 39), compared to a healthy control group (n = 40). Four hundred fifty-six metabolites were accurately measured. Multivariate analyses failed to reveal any metabolic profile discriminating between cryptogenic ischemic strokes and those with an identified cause but did show superimposable metabolic profiles in both groups, which were clearly distinct from those of healthy controls. The blood concentrations of 234 metabolites were significantly affected in stroke patients compared to controls after the Benjamini-Hochberg correction. Increased methionine sulfoxide and homocysteine concentrations, as well as an overall increase in saturation of fatty acids, were indicative of acute oxidative stress. This signature also showed alterations in energetic metabolism, cell membrane integrity, monocarbon metabolism, and neurotransmission, with reduced concentrations of several metabolites known to be neuroprotective. Overall, our results show that cryptogenic strokes are not pathophysiologically distinct from ischemic strokes of established origin, and that stroke leads to intense metabolic remodeling with marked oxidative and energetic stresses.

Keywords: cryptogenic ischemic stroke; energetic metabolism; lipidomics; metabolomics; mitochondria; oxidative stress; stroke.

Figure 1

Study flowchart.

Figure 2

Principal component analysis (PCA) and orthogonal projection to latent structures discriminant analysis (OPLS-DA) scatter plot with samples represented as circles. (A) The first plan of the principal component analysis including samples from controls (green), stroke (violet), and pool samples. Pool samples analyzed in the first batch are colored in yellow and those analyzed in the second batch are colored in black. There is a neat distinction between samples from controls and stroke patients in the second principal component (PC2) whilst pool samples plot near the origin (0,0) with no clear batch effect. (B) OPLS-DA analysis of all samples aiming at discriminating between controls (green), cryptogenic strokes (light violet), and strokes of known origin (dark violet), which are considered as three different groups. This model had no good predictive capabilities mainly due to the lack of discrimination between cryptogenic strokes and strokes of a known etiology. (C) However, as expected from the PCA analysis, there is a clear distinction between control (green circles) and stroke (dark violet circles)samples taken as a single group. OPLS-DA scatter plot PCA and OPLS-DA axes are dimensionless. Legend: PC1, PC2: first and second principal components; pLV₁, pLV₂: first and second predictive latent variables; pLV: predictive latent variable; oLV: orthogonal latent variable.

Figure 3

Volcano plot combining uni- and multivariate analyses. The x-axis represents loadings from the OPLS-DA analysis discriminating between control and stroke patients, with positive/negative values indicating relatively elevated/diminished concentration in the plasma of stroke patients compared to that of controls. Log-transformed p-values obtained from the comparison between both groups form the y-axis, using 0.05 as the base for logarithm calculation. Only metabolites that retained their significance after the Benjamini–Hochberg correction are displayed as colored circles. Some significant metabolites are also labeled due to very low (i.e., very significant) p-values and were positively correlated (methionine sulfoxide (Met-SO), acetyl (C2), propionyl (C3) and oleoyl (C18:1) carnitines, diacyl phosphatidylcholine PC aa 40:2, and triacylglyceride TG 16:0/37:3) or negatively correlated (choline, arachidonic acid, docosahexaenoic acid, lysophosphatidylcholine 18:0, and acyl-alkyl phosphatidylcholine PC ae 34:3) with the predictive latent variable. Legend: P_corr: loading of each metabolite on the predictive latent variable represented as a correlation coefficient; 3-IPA: 3-indolepropionic acid; PUFAs: polyunsaturated fatty acids as acyl moieties in phosphatidylcholine species; DHEAS: dehydroepiandrosterone sulfate; GLCA: glycolithocholic acid.

Figure 4

Box plots for the ratios methionine sulfoxide over methionine (Met-SO/Met), acetyl plus propionyl carnitines over free carnitine ((C2 + C3)/C0), and between the sum of lysophosphatidylcholine and the sum of phosphatidylcholine species (LPC/PC). The Met-SO/Met ratio measures oxidative stress and was significantly higher in the plasma of stroke patients; fatty acid beta-oxidation as measured by (C2 + C3)/C0 ratio was higher in stroke patients; the opposite was observed for plasmatic phospholipase A2 activity with a decreased LPC/PC in the stroke group.