From islet to blood: macrophage remodeling signatures for diagnosis and risk stratification in type 1 diabetes

Type 1 diabetes (T1D) is an autoimmune disease characterized by progressive β-cell destruction, yet current risk stratification tools, which rely mainly on genetic susceptibility and autoantibody profiles, remain insufficient for accurately predicting disease progression. We aimed to characterize macrophage-related inflammatory transcriptional activity in T1D and to develop peripheral blood-based biomarkers for diagnosis and risk stratification.

We integrated bulk RNA-seq, single-cell RNA-seq, and spatial transcriptomic data from human islets with public and in-house peripheral blood transcriptomic datasets. Macrophage heterogeneity and remodeling trajectories were analyzed in the islet microenvironment, and machine learning was used to derive tissue- and blood-based proinflammatory macrophage-related genes (PMRG). Diagnostic and prognostic models were then constructed and validated in peripheral blood cohorts, including a longitudinal islet autoimmunity (IA) cohort. SHAP analysis was applied to improve model interpretability. Independent PBMC RT-qPCR and mouse pancreatic immunofluorescence were performed to validate selected PMRG-related genes.

T1D islets showed marked immune remodeling with myeloid enrichment and five distinct macrophage subtypes. Pseudotime analysis identified a pro-inflammatory macrophage trajectory and 265 PMRGs, from which a 9-gene islet-derived PMRG (iPMRG) was obtained. Spatial transcriptomics supported the association of iPMRG-high macrophage signals with disrupted β-cell regions, and CellChat analysis inferred altered inflammatory communication programs. In peripheral blood mononuclear cells (PBMCs), the iPMRG-based diagnostic classifier distinguished T1D from healthy controls with an optimism-corrected AUC of 0.736. For prognosis, a 15-gene prognostic PMRGs was used to construct a risk score that, when integrated with clinical variables, predicted progression from IA to clinical T1D with time-dependent AUCs of 0.825, 0.814, and 0.860 at 12, 36, and 60 months, respectively. SHAP analysis identified the PMRG risk score as the dominant predictor and highlighted six core driver genes (PID1, TFPI2, SERPINB2, SOX4, DUSP2, and MT1X). The computational findings were further supported by independent validation in PBMCs and mouse pancreatic tissues.

Our study highlights the heterogeneous and dynamic nature of macrophage remodeling in the T1D islet microenvironment, which is translated into accessible peripheral blood signatures. The resulting diagnostic and prognostic models provide an interpretable framework for T1D risk stratification and may support future strategies for earlier detection and precision prevention.