Prediagnostic plasma proteomic patterns and protein signature for incident peripheral artery disease in type 2 diabetes

Peripheral artery disease (PAD) is a major vascular complication of type 2 diabetes (T2D), yet the circulating molecular alterations preceding its clinical onset remain incompletely characterized. We aimed to define the prediagnostic circulating proteomic architecture of incident PAD in T2D, reconstruct prediagnostic protein patterns before diagnosis, and explore the performance of a parsimonious protein panel for long-term risk stratification. We conducted a retrospective analysis of 1,407 UK Biobank participants with baseline T2D and plasma proteomic profiling using the Olink Explore 3072 platform. A total of 2,920 proteins were analyzed. Incident PAD was ascertained through linked hospital, primary care, procedure, death registry, and self-reported records. Associations between baseline proteins and incident PAD were evaluated using Cox proportional hazards models with false discovery rate correction to identify PAD-associated proteins. Functional enrichment, network, protein-trait integration, and exploratory mediation analyses were performed to characterize potential indirect statistical associations, without assuming temporal or causal ordering between clinical traits and protein levels. For prediagnostic pattern reconstruction, baseline protein levels in incident PAD cases were aligned by years before PAD diagnosis, standardized against matched controls, and modeled using LOESS. LASSO regression was first used to select candidate proteins, yielding eight proteins. Based on these eight candidate proteins, five survival prediction models, including LASSO-Cox, random survival forest, gradient boosting machine, survival support vector machine, and XGBoost, were developed in the training set and evaluated in the testing set. During follow-up, 121 participants developed PAD. In fully adjusted Cox models, 670 proteins were significantly associated with incident PAD, including 658 positive and 12 inverse associations. Additional adjustment for serum creatinine, urea, and urate attenuated but did not eliminate the protein–PAD associations, with 308 associations remaining significant in the extended fully adjusted model. The strongest positive associations included WFDC2, EDA2R, and NBL1, whereas UMOD, ENPP6, and NELL1 showed the strongest inverse associations. PAD-associated proteins were enriched in leukocyte adhesion, chemotaxis, cytokine-cytokine receptor interaction, humoral immune response, extracellular matrix organization, vascular remodeling, and endothelial homeostasis. Prediagnostic pattern reconstruction identified three reconstructed protein patterns: a mid-prediagnostic immune activation and remodeling pattern, a late metabolic and vascular stress pattern, and a progressive decline in chemokine-mediated migration signaling. Integrated protein-trait analyses showed strongest concordance with renal dysfunction, blood pressure-related traits, systemic inflammation, hematologic indices, and vascular comorbidities. Exploratory mediation analyses identified statistically compatible indirect associations involving selected clinical traits and proteins; however, these findings were not interpretable as evidence of causal biological mediation because clinical traits and proteins were measured concurrently. A compact seven-protein panel showed promising internal prognostic performance, but external validation is required. In individuals with T2D, incident PAD is preceded by a broad and biologically coherent circulating proteomic signature spanning immune activation, renal-vascular dysfunction, endothelial perturbation, and tissue remodeling. Prediagnostic pattern reconstruction suggests that multiple biological processes with distinct temporal patterns may precede clinical PAD onset. A compact seven-protein panel showed promising internal prognostic performance, but external validation is required. The mediation findings were exploratory and should not be interpreted as evidence of causal pathways.