Metabolomic clustering of 24,638 UK Biobank participants with prediabetes yields three subtypes with progressively higher type 2 diabetes, cardiovascular, and chronic kidney disease risk

The study identified three metabolically distinct subtypes of prediabetes: a low-metabolic-risk group (3,891 participants, 31.59%), an intermediate-metabolic-risk group (5,598, 45.45%), and a high-metabolic-risk group (2,827, 22.96%). Prior stratification of prediabetes heterogeneity relied largely on clinical characteristics; this work instead uses circulating metabolites in a data-driven clustering, grounding stratification in biological pathways such as lipoproteins, fatty acids, amino acids, and inflammation markers. Clustering used 12,322 participants in the training set, with the elbow method selecting k=3 and bootstrap resampling (100 replicates) yielding mean Jaccard indices above 0.9; 12,316 participants in the validation set were assigned to predefined cluster centroids.

Risks of type 2 diabetes, cardiovascular disease, and chronic kidney disease increased progressively across the three subtypes. This provides metabolically grounded quantitative evidence that prediabetes is not a single homogeneous state, with risk differences persisting after adjustment for age, sex, BMI, HbA1c, family history, and other covariates. Over a median follow-up of 13.5 years, the validation set recorded 1,852 incident type 2 diabetes cases, 1,767 cardiovascular disease cases, and 683 chronic kidney disease cases; relative to the low-risk group, adjusted hazard ratios for type 2 diabetes were 1.41 (95% CI 1.22–1.63) for the intermediate group and 2.24 (95% CI 1.93–2.60) for the high group, for cardiovascular disease 1.14 (1.01–1.30) and 1.27 (1.10–1.46), and for chronic kidney disease 1.52 (1.23–1.88) and 1.82 (1.44–2.30).

Associations between diet quality and outcomes differed across metabolic subtypes, with healthier dietary patterns generally associated with lower risks of type 2 diabetes, cardiovascular disease, chronic kidney disease, and kidney failure within the intermediate-risk group. It was previously unknown whether dietary effects differ across metabolically defined subgroups of prediabetes; this work brings diet score and cluster into a joint analysis within one framework. Diet was assessed with a 10-item food frequency questionnaire, with healthy diet defined as above the median score of 4; the joint analysis was performed in the validation set and reported as hazard ratios in Figure 4.

Mendelian randomization suggested potential causal links between several nuclear magnetic resonance metabolomic markers and cardiovascular-kidney-metabolic outcomes. Beyond observational cluster associations, the study adds genetic-instrument evidence on the etiological direction for selected markers. Two-sample Mendelian randomization used genetic instruments from published UK Biobank GWAS data on 115,078 participants and FinnGen outcome summary statistics; large HDL cholesteryl esters were inversely associated with type 2 diabetes (IVW OR 0.898, 95% CI 0.853–0.946), medium HDL cholesteryl esters inversely with cardiovascular disease (IVW OR 0.937, 0.899–0.976), and creatinine with chronic kidney disease (IVW OR 1.613, 1.388–1.875), with consistent directions from weighted median and weighted mode methods.

The results are intended for risk stratification and precision prevention research in prediabetes: in cohorts with nuclear magnetic resonance metabolomic data, 16 metabolic markers can assign individuals to low-, intermediate-, or high-metabolic-risk groups, informing discussion of more intensive or targeted lifestyle and nutritional interventions. The study suggests the intermediate- and high-risk groups may benefit most from active dietary intervention and offers directional suggestions, such as a cholesterol-lowering cardioprotective pattern rich in unsaturated fats, soluble fiber, and whole grains for the intermediate group, and a more intensive low-glycemic-load, triglyceride-lowering, anti-inflammatory approach for the high group. The Mendelian randomization component adds genetic etiological clues for markers such as large HDL cholesteryl esters, medium HDL cholesteryl esters, and creatinine, which can inform subsequent mechanistic and intervention studies.

Several points warrant attention: the study population is primarily of European ancestry, so extrapolation to other populations should be cautious; the number of measured metabolites was limited and some biomarkers lack disease specificity; the cluster-outcome associations are observational, and although Mendelian randomization offers causal clues it still relies on instrumental variable assumptions. In addition, the specific effect sizes of diet-subtype interactions, how these subtypes could be translated into actionable stratification tools in practice, and whether these metabolic subtypes can be modified by intervention to improve outcomes remain open questions. This was a full-text reading, but some supplementary tables and figures (such as Supplementary Table 4 and Supplementary Figures 1 and 2) are not included in the main text, so those details require consulting the original supplementary files.