COL1A2 and APOLD1 Define a Dual-Axis Molecular Framework for Diabetic Nephropathy-Retinopathy Comorbidity: An Integrative Multiomics and Machine Learning Study in Mouse Models.

Abstract

Background: Diabetic retinopathy (DR) and diabetic nephropathy (DN) are severe microvascular complications that frequently co-occur, suggesting shared pathogenic mechanisms. However, systematic identification of their common molecular drivers remains limited.

Methods: We performed an integrative multiomics analysis combining bulk transcriptomics (six datasets: GSE30528, GSE30529, GSE96804, GSE142025, GSE160306, and GSE221521), single-cell RNA sequencing (scRNA-seq; GSE216510 for DN, GSE178121 for DR), advanced computational modeling, and experimental validation. Differential expression analysis, weighted gene coexpression network analysis (WGCNA), and protein-protein interaction (PPI) network analysis were conducted, and 120 combinatorial machine learning models were constructed. Cell-cell communication and pseudotime trajectory analyses were performed. Western blotting was used to validate protein expression dynamics in db/db (type 2) diabetic mouse models at 1, 3, and 6 months after diabetes onset.

Results: Cross-tissue analysis revealed COL1A2 as a consistently upregulated gene and APOLD1 as a consistently downregulated gene in both DR and DN. These two genes define a dual-axis model: an early dysfunction axis marked by downregulated APOLD1 expression and a late structural remodeling axis driven by upregulated COL1A2 expression. Machine learning models built on comorbidity signatures achieved robust predictive performance in hold-out validation. scRNA-seq revealed that in DN, COL1A2 and APOLD1 were specifically expressed in fibroblasts; in DR, they were predominantly expressed in pericytes/vascular smooth muscle cells, with APOLD1 also detected in endothelial cells. Pseudotime analysis indicated that APOLD1 expression peaked early in disease trajectories, whereas COL1A2 expression accumulated at later stages. Western blotting confirmed the progressive upregulation of COL1A2 and downregulation of APOLD1 protein expression in both retinal and renal tissues over time in diabetic models, with consistent trends across the 1-, 3-, and 6-month timepoints.Cell-communication analysis revealed extensive network dysregulation: DN exhibited SPP1, RANKL, CD45, CSF, and FGF pathway activation and WNT, ARGN, and NOTCH pathway suppression, whereas DR exhibited MHC-I, EDN, LAMININ, and VEGF pathway suppression. In silico COL1A2 knockout in fibroblasts (DN) and smooth muscle cells (DR) induced distinct transcriptional responses-immune-related in DN versus vascular stress-related in DR.

Conclusion: On the basis of the study results, we propose a novel "dual-axis" model for diabetic microvascular comorbidity: an early dysfunction axis marked by downregulated APOLD1 expression and a late structural remodeling axis driven by upregulated COL1A2 expression. These findings provide a cohesive molecular framework and potential biomarkers for the concurrent pathogenesis of DR and DN.