Identification of biomarkers for diabetic kidney disease progression by utilizing a nanobinder-enabled deep plasma proteomics platform

INTRODUCTION: Diabetes presents a global health challenge, with diabetic kidney disease (DKD) being a major complication that can progress to end-stage kidney disease. Predicting which patients will develop DKD and experience rapid progression is crucial, but current blood biomarker methods lack sensitivity and specificity. To address this, we developed the Proteonano™ platform for deep, untargeted plasma proteomic profiling that allowed discovery of protein biomarkers that effectively predicts DKD progression.

METHODS: Proteonano™ technology was developed as an affinity-selective mass spectrometry platform, including the usage of nanoparticle-based affinity protein binders (nanobinders) to enrich low abundance proteins and the employment of an automated pre-treatment workstation for parallel sample preparation. Patients were serially enrolled under approved ethical review. Serum samples were collected at the time of enrollment, kidney functions were assessed for up to five years, and patients were stratified as DKD progressors and non-progressors. Samples were processed through the Proteonano™ Lite pipeline and analyzed with a ThermoFisher Exlporis 480 mass spectrometer at data-independent acquisition mode. Data were processed using DIA-NN and a customized bioinformatics pipeline.

RESULTS: Serum samples from 67 DKD patients (DKD progressors=38; non-progressors=29) were analyzed. 1393 protein groups were identified, with an average of 951±11 (AVG±SE) per sample. 1185 proteins were mapped to the human plasma protein project plasma protein catalog, with concentrations spanning eight orders of magnitude, and the lowest at 3.0 pg/mL. Differential expression analysis identified 45 proteins between progressors and non-progressors, with VWF being the most upregulated and PASK the most downregulated protein. VWF was the best single differentiator but not superior to the urine albumin to serum creatinine ratio (UACR). Thus, multivariate analyses were subsequently conducted. Eight feature selection methods were employed, and the top features selected in each method were subjected to Akaiki information criteria (AIC) based model selection. The best model, selected by random forest method, included five proteins (ROC-AUC=0.97, 95 % confidence interval: 0.94-1.00), outperforming UACR (ROC-AUC=0.87). Adding UACR to this model marginally improved discrimination (ROC-AUC=0.99). These results demonstrate Proteonano™ nanobinder assisted sample preprocessing improves untargeted proteomic analysis of blood samples, which allowed identification of a novel protein panel that effectively identifies patients with increased risk for DKD progression.