Large-Scale Proteomics Uncovers Pre-Disease Inflammation-Lipid Subtypes to Refine Risk Stratification and Prediction...

Type 2 diabetes exhibits substantial heterogeneity prior to disease onset, posing challenges for early prevention. This study aimed to identify pre-disease subgroups with distinct risk profiles and develop subgroup-specific prediction models to facilitate early detection and precision intervention.

Proteomic data from 41 030 participants without diabetes at baseline in the UK Biobank were analysed. Proteins associated with incident type 2 diabetes were identified using multivariable-adjusted Cox regression and Least Absolute Shrinkage and Selection Operator (LASSO) regression. Finite Gaussian mixture model-based clustering was applied to define risk subgroups. Heterogeneity among subgroups was characterised according to proteomic patterns, clinical traits and disease risk. Subgroup-specific prediction models were subsequently developed using Cox regression.

A total of 113 protein markers were identified, stratifying the population into three subgroups: metabolically healthy group (MHG), mild inflammation group (MIG) and Dyslipidemia with inflammation group (DLIG). Compared with MHG, DLIG showed the highest risk of type 2 diabetes (HR = 2.58, 95% CI: 2.34-2.84), followed by MIG (HR = 1.71, 95% CI: 1.49-1.98). Enrichment analysis indicated dysregulation of immune-inflammatory and lipid metabolism pathways in DLIG and MIG. Clinically, DLIG exhibited higher BMI, waist circumference, triglycerides, and C-reactive protein levels, whereas MIG showed moderately elevated C-reactive protein. Subgroup-specific proteomic models outperformed traditional clinical models in predicting diabetes risk (AUC 0.820-0.889 vs. 0.709-0.784; C-index 0.787-0.847 vs. 0.687-0.747).

Proteomics-based clustering identified three pre-disease subtypes of type 2 diabetes characterized by inflammation-lipid profiles, improving risk prediction and supporting precision prevention.