Association of Novel Urine Biomarkers with Immune Checkpoint Inhibitor Nephrotoxicity

Background: Immune checkpoint inhibitor (ICI)-based cancer therapy saves lives, but ~20% of patients undergoing therapy develop acute kidney injury (AKI), and 2-5% face acute interstitial nephritis (AIN) with actionable consequences. Urine protein markers can correlate with specific kidney pathologies, yet exploring the urine proteome is challenging due to low concentrations of potential immunodiagnostic proteins. NUcleic acid Linked Immuno-Sandwich Assay (NULISA™) represents cutting-edge technology reported to be more sensitive than proximity extension assays like OLINK. We hypothesized that leveraging this ultrahigh sensitivity and parallel multiplexing proteomic technology could reveal novel urine or plasma biomarkers for non-invasive ICI-AIN detection. Methods: We used NULISA™ to assess 203 proteins in urine and plasma from AKI patients on ICI therapy. Cohorts were categorized as either ICI-AIN (n = 22) or non-AIN (n = 27) based on renal biopsy pathology and/or clinical diagnosis. Sample profiling was evaluated using log2 normalized count data, with P-values computed using Wilcoxon tests, and False Discovery Rate (FDR) adjustment. Results: Of the 203 targets, 149 (73.4%) were detectable in urine samples, and 193 (95.1%) in plasma samples. Clustering and principal component analysis (PCA) revealed distinct proteomic profiles and KEGG pathway analysis identified JAK-STAT and TNF signaling in urine compared to TLR and IL-17 signaling in plasma. Larger differences in protein levels between ICI-AIN and non-AIN were observed in urine. Top urine proteins with AUC > 0.8 (indicating a strong ability to discriminate ICI-AIN from non-AIN) and fold change > 8.0 (FDR 0.01) were associated with hypersensitivity, autoimmunity, immune checkpoint proteins, and immune cell activation. Statistical models, including L1 regularized logistic regression and CART, identified a novel 2-protein signature, as the most effective combination for identifying ICI-AIN. A logistic regression model of this ICI-AIN signature achieved an AUC of 0.94 in distinguishing ICI-AIN from non-AIN, outperforming any other marker(s) including CXCL9 (AUC 0.83). Conclusion: Our use of state-of-the-art NULISA technology demonstrates that urine and plasma proteomic profiles exhibit distinct characteristics. Statistical modeling identified a novel ICI-AIN signature for non-invasive immune nephritis detection. Funding: NIDDK Support