Mass spectrometry-based metabolomic and lipidomic profiling stratifies stages of diabetic retinopathy

Abstract

Aims

Diabetic retinopathy is a metabolic complication of diabetes. This study aimed to elucidate metabolic and lipidomic alterations associated with diabetic retinopathy progression and identify biomarkers for its diagnosis and staging.

Methods

We analyzed serum profiles of 962 molecules, including 653 lipids and 309 polar small metabolites, from 167 individuals: 45 without retinopathy, 69 with non-proliferative diabetic retinopathy of varying severity, and 53 with proliferative diabetic retinopathy. Statistical models were applied to single-omics datasets, and machine learning algorithms were used to integrate metabolomic and lipidomic data for identifying features that best discriminate disease stages.

Results

Purine and sphingolipid metabolism were significantly altered in both non-proliferative and proliferative stages. Tyrosine metabolism was disrupted in non-proliferative disease, while glycine-serine-threonine metabolism was prominent in proliferative disease. Progression was associated with reduced sphingomyelins, phosphatidylcholines, and lysophosphatidylcholines. Among ten machine learning models, the K-nearest neighbor algorithm achieved the highest classification performance. The lipid ST 24:1;O4 and metabolite beta-hydroxyisovaleric acid were the top contributors.

Conclusions

Mass spectrometry-based metabolomics/lipidomics integrated with machine learning accurately stratify diabetic retinopathy and identify novel molecular markers, and may support early diagnosis and targeted intervention of diabetic retinopathy.