#3119 DEEP LEARNING IDENTIFIES SUBPHENOTYPES OF DIABETIC KIDNEY DISEASE DRIVEN BY GENETIC VARIATIONS IN THE RHO PATHWAY

Abstract

Abstract Background and Aims Diabetic kidney disease (DKD) is a leading cause of end-stage kidney disease (ESKD), however therapies targeting causal pathways have been limited by disease heterogeneity. Integrating electronic health record (EHR) data and genomics may uncover hidden subphenotypes in DKD. In this study, we use deep learning to identify a novel genetic variant of ARHGEF18 associated with significantly higher risk of DKD and ESKD (Figure 1A). We further employed quantitative microscopy techniques and biochemical assays to elucidate the mechanistic role of ARHGEF18 and its variant in podocytes. Method DKD patients from the Mount Sinai BioMe Biobank were used in this study. Unsupervised clustering and accounting for population structure in a deep learning framework identified two clusters: Cluster M (mild) and S (severe). We then performed a genome wide association study (GWAS) of patients within each cluster compared with healthy controls. For mechanistic studies of the novel variant, cytoplasmic, focal adhesion, cytoskeletal morphometrics as well as live-cell motility, Rho GTPase activity, and protein degradation experiments were performed using confocal and total internal reflection fluorescence (TIRF) microscopy as well as cell-free biochemical assays using immortalized human podocytes expressing ARHGEF18 wild-type (WT) and mutant transcripts. Results We employed autoencoders and unsupervised clustering of EHR data on 1,372 DKD patients to establish two clusters with differential prevalence of ESKD. There was a greater prevalence of proteinuria in Cluster S compared to Cluster M (Figure 1B). Further exome sequencing study in these patients identified a novel variant in ARHGEF18, a Rho guanine exchange factor highly enriched in podocytes (Figure 1A). Nephroseq database showed an increased ARHGEF18 expression in chronic kidney disease (CKD) kidney biopsy samples compared to healthy controls (Figure 1C). Overexpression of ARHGEF18 mutant transcripts in human immortalized podocytes led to impairments in cell adhesion, focal adhesion architecture, and cell motility. Live TIRF microscopy experiments showed preferential subcellular localization of GEF18 mutant to the periphery of migrating podocytes whereas GEF18 WT localized at the perinuclear/cytoplasmic region (Figure 2A, B). GEF18 mutant cells also displayed an increased RhoA activation (Figure 2C). Upon inhibition of protein synthesis using cycloheximide (CHX), we observed a significantly slower degradation of GEF18 mutant protein over a 12h period indicating increased protein stability (Figure 2D). GEF18 mutant also showed resistance to ubiquitin mediated degradation leading to pathologically increased protein levels. Conclusions We report a novel gain of function variant of ARHGEF18 that drives podocyte dysfunction through impaired protein localization and degradation. Targeting this pathway could help regulate RhoA activation and cytoskeletal rearrangements preventing podocyte effacement in DKD.