Type 2 diabetes is not a single-organ disease. It is a multisystem molecular syndrome.
A new Metabolism paper builds one of the most comprehensive UK Biobank multi-omics maps of T2D complications to date, integrating plasma proteomics, metabolomics, machine learning, and 19 clinical outcomes across cardiovascular, microvascular, renal, hepatic, pulmonary, neurological, and mortality endpoints.
The study established three T2D subcohorts:
Proteomics: 3,104 participants
Metabolomics: 28,834 participants
Multi-omics: 3,059 participants
The molecular layers were broad: 2,920 plasma proteins from Olink Explore and 168 NMR metabolic biomarkers from Nightingale Health. The clinical scope was equally broad, spanning ischemic heart disease, MI, stroke, PAD, heart failure, atrial fibrillation, diabetic kidney disease, retinopathy, neuropathy, MACE, mortality, dementia, COPD, chronic liver disease, chronic renal failure, gout, anxiety, and diverticular disease.
The key biological message: many T2D complications share a common molecular architecture, but each organ also carries outcome-specific signatures.
Across 9 cardiovascular and microvascular complications, 1,359 proteins were significantly associated with at least one complication. Shared risk proteins included NT-proBNP, GDF15, WFDC2, IGFBP4, TNFRSF10B, EDN1, TIMP1, ACTA2, and others. Protective or inverse signals included UMOD, EGFR, DCXR, CTSV, ASS1, and related markers.
Metabolomics showed a convergent risk landscape: creatinine, GlycA, glucose, acetate, MUFA, large HDL/VLDL-related measures, and HDL particle size appeared as recurrent risk signals, while albumin, DHA, omega-3 fatty acids, histidine, unsaturation, and small HDL-related measures tended to show inverse associations. These patterns point toward renal dysfunction, systemic inflammation, lipoprotein remodeling, impaired energy metabolism, and nutritional/inflammatory status as shared axes of diabetic complication biology.
Prediction was where the study became clinically provocative.
Protein-based models markedly outperformed clinical models, with a median ΔC-index of 0.108, while combined clinical protein models achieved the best performance, with median ΔC-index 0.109. Metabolites added only modest incremental value, median ΔC-index 0.027. A simplified 174-protein panel retained robust predictive performance, balancing accuracy and feasibility.
The interpretation is clear: in T2D, plasma proteomics may provide a practical molecular window into multisystem complication risk, while metabolomics adds complementary but smaller predictive information.
Caution: UK Biobank is predominantly European ancestry, and external validation is still needed.
But the direction is important: diabetes precision medicine may move from “glucose-centric risk” to multi-organ molecular risk stratification.
Reference: Zhang et al. Metabolism 2026. DOI: 10.1016/j.metabol.2026.156650.