A non-invasive urinary diagnostic signature for diabetic kidney disease revealed by machine learning and single-cell analysis
Fig 2
Single-cell profiling of the human kidney in DKD.
(A) Dot plot displaying the expression of canonical marker genes used to identify the 11 major renal cell types. (B) UMAP projection of 20,220 high-quality cells from the integrated dataset (GSE131882; n = 3 controls, n = 3 DKD patients), color-coded by annotated cell type. Unsupervised clustering identified 11 distinct renal cell subpopulations, with key types abbreviated as follows: CD-PC (collecting duct principal cell); PTC (proximal tubule cell); Inj-PTC (injured PTC); DCT (distal convoluted tubule cell); LOH (loop of Henle); CD-ICA/CD-ICB (collecting duct intercalated cells); EC (endothelial cell); Podo (podocyte); Mes (mesenchymal cell); Imm (immune cell). (C) Stacked bar plot showing the proportional abundance of each cell type in control and DKD groups. (D) Stacked bar plot comparing the proportions of injured proximal tubule cells (Inj-PTC) and normal PTCs between control and DKD groups. The difference in proportion was assessed using a Chi-square test for categorical variables, and the p-value is unadjusted. (E-F) Violin plots showing the expression levels of injury markers (E) HAVCR1 and VCAM1, and functional genes (F) SLC34A1, CUBN, and SLC22A6 in Inj-PTC (n = 1794) versus normal PTC (n = 3794) clusters. Differential expression analysis for each gene was performed using the Wilcoxon rank-sum test; p-values are unadjusted. The black horizontal bar within each violin represents the median expression value.