Metabolomics profiling of plaque enhancement in ischemic stroke patients with large artery intracranial atherosclerosis.

Background: Vessel wall plaque high enhancement (HE) may be regarded as an unstable and vulnerable feature of atherosclerosis. We aimed to find differential serum metabolites in ischemic stroke patients with and without HE plaques and establish models to predict plaque enhancement.

Methods: Sixty participants were recruited and divided into HE and non-HE groups. Non-targeted metabolomics was performed to analyse the differential serum metabolites in the two groups. Partial least squares discriminant analysis was used to combine the parameter selection and classification into one operation. We used the least absolute shrinkage and selection operator, random forest machine learning models, and five binary classification models with 10-fold cross-validation approaches for deeper variable reduction and selection. Finally, logistic regression was used to evaluate the metabolites' effects on intracranial atherosclerotic plaque contrast enhancement. For accuracy analysis, we compared the areas under the receiver operating characteristic curves and used internal validation for nomogram.

Results: A total of 279 differential metabolites were detected in HE patients compared with the non-HE controls. Machine learning determined a combination of three decreased metabolites, 5α-dihydrotestosterone, D-fructose 6-phosphate, and phosphatidylcholine (32:2), was associated with HE and may serve as potential biomarkers.

Conclusions: Decrease of three metabolites is useful for predicting HE of intracranial atherosclerosis plaques in patients with ischemic stroke.