A serum proteomics model identified 6 proteins that independently predicted kidney function decline in patients with autosomal dominant polycystic kidney disease. It outperformed standard clinical and imaging-based prediction tools.
Autosomal dominant polycystic kidney disease (ADPKD) is the most common inherited kidney disorder and a leading genetic cause of end-stage renal disease. Predicting decline in estimated glomerular filtration rate (eGFR) is essential to determining timing of treatment, including the use of tolvaptan—the only approved disease-modifying therapy.
Lead author Hande Ö. Aydogan Balaban, of the University Hospital Cologne, and fellow researchers developed a linear regression model using 6 proteins measured from blood samples: Endothelial Plasminogen Activator Inhibitor (SERPINF1), Glutathione Peroxidase 3 (GPX3), Afamin (AFM), FERM Domain Containing Kindlin-3 (FERMT3), Complement Factor H Related 1 (CFHR1), and Retinoic Acid Receptor Responder 2 (RARRES2). The model, derived from the German AD(H)PKD registry, explained 31.4% of the variance in annual eGFR decline (adjusted R² = 0.314).
By comparison, a clinical model that incorporated age, sex, baseline eGFR, and Mayo Imaging Classification (MIC) explained 23.3% of variance (adjusted R² = 0.233). Combining clinical and proteomic data increased predictive power to 34% (adjusted R² = 0.340).
All 6 proteins showed statistically significant associations with eGFR slope (P < .05). SERPINF1, FERMT3, CFHR1, and RARRES2 were negatively associated with eGFR slope, which indicated faster decline with lower levels. GPX3 and AFM were positively associated. RARRES2 had 22.2% missing values, yet even the presence or absence of detection showed a significant negative correlation with eGFR slope.
Root mean square error (RMSE) for the proteome model was 2.03. The combined model had a lower RMSE of 1.83, compared with 2.11 for the MIC model. In both early- and later-stage chronic kidney disease, the proteome-based models demonstrated higher predictive accuracy.
The model was validated using 2 independent cohorts: 305 patients from the same registry and 173 patients from the Dutch DIPAK study. Although DIPAK used EDTA plasma rather than serum, the proteome model retained comparable performance to MIC with an RMSE of 11 vs 12.4 when predicting future eGFR.
Protein selection was based on linear modeling and weighted least absolute shrinkage and selection operator (wLASSO) methods. Only proteins with consistent associations across both methods were included in the final model.
To test disease specificity, the proteins were also evaluated in patients with IgA nephropathy. Only 9 of 29 candidate proteins showed significant correlations with eGFR in both cohorts, which suggested that the 6-protein panel may be more specific to ADPKD.
The authors noted that the model could assist in predicting progression in ADPKD when imaging or genetic data are unavailable. Further studies are needed to validate the findings using targeted assays for clinical application.
The authors reported no conflicts of interest.
Source: nature communications
