Urine proteomics identifies biomarkers for diabetic kidney disease at different stages

Abstract

Background: Type 2 diabetic kidney disease is the most common cause of chronic kidney diseases (CKD) and end-stage renal diseases (ESRD). Although kidney biopsy is considered as the 'gold standard' for diabetic kidney disease (DKD) diagnosis, it is an invasive procedure, and the diagnosis can be influenced by sampling bias and personal judgement. It is desirable to establish a non-invasive procedure that can complement kidney biopsy in diagnosis and tracking the DKD progress.

Methods: In this cross-sectional study, we collected 252 urine samples, including 134 uncomplicated diabetes, 65 DKD, 40 CKD without diabetes and 13 follow-up diabetic samples, and analyzed the urine proteomes with liquid chromatography coupled with tandem mass spectrometry (LC-MS/MS). We built logistic regression models to distinguish uncomplicated diabetes, DKD and other CKDs.

Results: We quantified 559 ± 202 gene products (GPs) (Mean ± SD) on a single sample and 2946 GPs in total. Based on logistic regression models, DKD patients could be differentiated from the uncomplicated diabetic patients with 2 urinary proteins (AUC = 0.928), and the stage 3 (DKD3) and stage 4 (DKD4) DKD patients with 3 urinary proteins (AUC = 0.949). These results were validated in an independent dataset. Finally, a 4-protein classifier identified putative pre-DKD3 patients, who showed DKD3 proteomic features but were not diagnosed by clinical standards. Follow-up studies on 11 patients indicated that 2 putative pre-DKD patients have progressed to DKD3.

Conclusions: Our study demonstrated the potential for urinary proteomics as a noninvasive method for DKD diagnosis and identifying high-risk patients for progression monitoring.

Keywords: DKD; Progression monitoring; Proteomics; Urine.

Fig. 1

Urine proteomic analysis of Diabetes, DKD, and CKD. a Clinical stages and the number of samples used in the discovery and validation datasets. b Principal component analysis of the proteomics data. Each dot represents an urinary sample. Blue: Diabetes, Yellow: DKD, Red: CKD. c Reactome pathway analysis of the disease-specific DEPs (323 DEPs for diabetes, 98 DEPs for DKD, 88 DEPs for CKD). d Pearson correlation coefficients between the abundance of the 2,946 urine proteins and the 13 routinely tested clinical or health indexes. e. Scatter plots of selected high abundance urine proteins and kidney function indexes with strong positive correlations

Fig. 2

A classifier for distinguishing DKD from Diabetes. a A bioinformatic analysis workflow to find candidate biomarkers between the Diabetes and the DKD group. n: number of samples used in the analyses. b Volcano plot displaying the differentially expressed proteins between Diabetes and DKD. Red and Blue indicated proteins that were significantly enriched in DKD and Diabetes, respectively (p values < 0.05, more than threefold change). Other proteins were colored in grey. c ROC curve of distinguishing the Diabetes and DKD samples predicted by the 2-protein classifier. d Boxplots showing the ALB and AFM abundance in Diabetes and DKD in the two datasets (center line: median, bounds of box: 25th and 75th percentiles, and whiskers: from Q1-1.5*IQR to Q3 + 1.5*IQR, p-value calculated by Mann–Whitney U test). e Reactome pathway analysis of the up-regulated (red) and down-regulated (blue) DEPs in DKD samples

Fig. 3

A classifier for distinguishing DKD4 from DKD3. a A bioinformatic analysis workflow to find candidate biomarkers between DKD3 and DKD4 samples. n: number of samples used in the analyses. b Hierarchical clustering of DKD3 and DKD4 DEPs using complete linkage. Protein expression values were normalized by z-scores. c Venn diagram indicating the overlap of DEPs between discovery and validation datasets. d ROC curve of distinguishing DKD3 and DKD4 predicted by a 3-protein classifier. e Reactome pathway analysis of the up-regulated DEPs in DKD4 samples. f Boxplots displaying the abundance of the complement component proteins at different stages of DKD and CKD in the two datasets

Fig. 4

A classifier for monitoring early transition to DKD. a A bioinformatic analysis workflow to find candidate biomarkers between DKD3 and Diabetes samples. b Venn diagram showing the overlap of DEPs in the discovery and the validation set. c ROC curve of distinguishing DKD3 and Diabetes predicted by a 4-protein classifier. d Dotplot indicating the Diabetes and DKD samples and their predicted stages by the classifier. Each point represented one urine sample. Blue and green dots presents Diabetes and DKD, respectively. The x-axis indicates the predicted DKD stage 3. Blue dots positioned on the right side of the Prediction line were predicted incorrectly as DKD3 by the model and were used as putative pre-DKD in the subsequent analysis. e Boxplots displaying the iFOT intensities of the 4 biomarkers (CP, TF, SERPINA5, VPS4A) in Diabetes, DKD3, DKD4, and CKD samples. f The risk scores(left panel) and the 4-marker expression (right panel) of the patient (clinical ID: 8073510). g A schematic summary of bioinformatic analysis workflow that derived the 3 prediction models in this study