Diabetic retinopathy (DR) is one of the most common microvascular complications of diabetes and the leading cause of vision loss among the working-age population worldwide. With an insidious onset and severe consequences, it is known as the "silent killer" of vision. As the global burden of diabetes becomes increasingly severe, the prevalence of DR and the vision impairment it causes continues to climb. It is estimated that by 2045 there will be 270 million DR patients worldwide, and one in every three DR patients faces the risk of irreversible blindness. Although the root cause of DR is long-term persistent hyperglycemia, its exact pathogenesis remains unclear. In addition, existing screening technologies still have "blind spots" in the early identification of DR, so that many patients already suffer irreversible visual damage when DR is diagnosed. Given this severe situation, deeply exploring the molecular mechanisms of DR onset and developing new screening methods for early DR diagnosis and precise intervention has become an urgent need in the field of ophthalmology.

By comprehensively integrating ocular-omics, metabolomics and phenomics data from two multinational, cross-ethnic cohorts, and through untargeted metabolomics analysis, the researchers identified a series of DR metabolic molecular fingerprints. They further combined machine learning algorithms to construct a personalized long-term risk warning model for DR. The study leveraged large-scale data from the UK Biobank (UKB) and high-precision, multimodal data from the Guangzhou Diabetic Eye Study (GDES) for complementary fusion. The metabolomes of both large population cohorts were measured using the NMR method.

Figure 1. Schematic diagram of the research workflow

In the UKB discovery set, 118 of 168 metabolites were associated with the risk of developing DR, including 7 positively correlated metabolites and 111 negatively correlated metabolites. These fingerprints significantly improved the predictive performance for DR development, surpassing traditional indicators (Figure 2).

Figure 2. Correlation between NMR-detected substances in the UKB cohort and the occurrence of DR

Subsequently, the research findings were applied across ethnicities in the Guangzhou Diabetic Eye Study (GDES) cohort, further confirming the universality and reliability of these key metabolic fingerprints for DR prediction in an independent external population (Figure 3). To reduce the complexity of the testing process and lower the high testing costs, the team also developed a fast, low-cost and cost-effective intelligent DR prediction metabolite panel. Applied alone to the diabetic population, this panel already achieves considerable predictive performance for DR, demonstrating enormous cost-effectiveness advantages and clinical translation potential, and providing a practical tool for early screening of DR in large diabetic populations.

Excessive intracellular accumulation of lactate can activate G protein-coupled receptors, a process that downregulates protein kinase A and its downstream signaling, ultimately leading to structural and functional damage to the retinal endothelial cell barrier and thus causing microvascular dysfunction in DR. Citrate, a key intermediate of the tricarboxylic acid cycle, can cause retinal energy metabolism disorders when its levels change abnormally, promoting the progression of DR. Notably, although lactate was found to be clearly associated with ischemia in the medial nasal region of the optic disc, its effect on the blood supply to the inferomedial region was relatively limited. The specific molecular mechanisms behind this finding and how it acts in the pathological process of DR still require further research to clarify.

Based on NMR technology, this study, through metabolomics research and analysis of large-scale, long-term multinational longitudinal cohorts, mapped the metabolic fingerprint profile of DR, discovered key molecules such as lactate and citrate in the occurrence and progression of DR, and revealed the metabolic injury mechanisms of DR. Based on these key molecular changes, the team further constructed the first ultra-long-term personalized precision risk warning model for DR, using a low-cost and rapid method to screen out high-risk DR populations for early warning and precise intervention. These results provide a brand-new tool for early DR management in large diabetic populations, demonstrating enormous potential for clinical translation and application and significant social benefits.