PROTEOMIC ANALYSIS OF SERUM OVER TIME TO FORECAST TREATMENT RESPONSE IN LUPUS NEPHRITIS

O061 / #515 Topic: AS15 - Lupus Nephritis-Clinical ABSTRACT CONCURRENT SESSION 10: INTEGRATING PROTEOMIC & TRANSCRIPTOMICS IN SLE 24-05-2025 10:40 AM - 11:40 AM Background/Purpose Lupus nephritis (LN) can cause severe complications and early mortality in SLE patients. Despite advancements, treatments for LN are not consistently effective and often have adverse effects. This study aims to develop a robust, noninvasive biomarker panel to predict treatment response, using high-throughput proteomics and machine learning to model serum protein expression, as part of the Accelerating Medicines Partnership RA/SLE Network. Methods Over 5,000 proteins were measured in the serum of 158 LN patients at the time of diagnostic kidney biopsy (baseline) and 12 weeks post-biopsy using Olink Explore HT. Clinical response was determined at 52 weeks as complete (CR; n=25), partial (PR; n=22), or no response (NR; n=49). Multivariate logistic regression adjusting for age, gender, and genetic ancestry and machine learning algorithm approaches (Extreme Gradient Boosting) were used to generate a robust prediction model of treatment responsiveness. Results At baseline, patients with NR exhibited 515 (p-value < 0.05; 160 upregulated) dysregulated proteins within the innate immune system, platelet activation, and pathways seen in neurodegeneration compared to CR. In addition, 1227 (p-value < 0.05; 1180 upregulated) proteins involving the TGFb, IL-10, Th1/Th2 differentiation, platelet activation, and leukocyte transendothelial migration pathways at 12 weeks post-biopsy were differentially expressed in patients with NR compared to CR. Proteins involved in T cell proliferation, Th17 cell differentiation, and TNF, Wnt, EGFR, and IL-10 signaling were persistently elevated at week 12 in NR compared to CR using paired analysis. The proteomic profiles of CR and NR are easily distinguishable at 52 weeks (AUC, 0.85 ± 0.10 with cross-validation accuracy of 76.1% ± 10.2%). While the ML models at baseline and 12 weeks showed less robust prediction performance with 65.9% ± 7.6% and 70.4% ± 7.3% (Figure 1A-C), the model that incorporates the baseline protein levels and the changes from baseline to 12 weeks post treatment showed the most robust prediction with AUC of 0.89 ± 0.06 and accuracy of 81.5% ± 6.2%. In particular, patients with a CR had a significant reduction in CD27, VEGF, HAVCR2, MEGF11, and VSIG4 from baseline to 12 weeks post-biopsy (Figure 1D-E). Furthermore, preliminary trajectory analyses have demonstrated the rapid decline of these proteins before treatment with the levels plateauing near the levels seen in healthy controls throughout the 52-week of study trial. Gene regulatory network analyses of the top predictors demonstrated significantly downregulated lymphocyte activation/differentiation (CD27, IL-7, IL-3, IL-16, CD83, and IL-10) and cellular migration pathways (NRP1, IL-16, PDGFB, CSF1, DDR1, FSTL1) in patients with a CR at week 12. Figure 1. Conclusions Early downregulation of specific immune pathways upon treatment precedes future clinical response in LN. Changes in serum protein expression, especially the soluble surface receptors shed upon cellular activation, at 12 weeks post-biopsy may serve as noninvasive biomarkers of 52-week treatment response.