FC080: A Systems Nephrology Framework for the Molecular Classification of Chronic Kidney Disease

Abstract

Abstract BACKGROUND Conventional stratification by clinical and histopathological phenotypes only approximates the heterogeneity of chronic kidney disease (CKD) and is insufficient to drive discovery of disease-modifying therapies or predict clinical outcomes. Recent advances in molecular stratification offer a mechanistic approach to disease classification but are limited by the availability of rich patient cohort datasets [1]. The National Unified Renal Translational Research Enterprise (NURTuRE) is a unique prospective cohort study for CKD and idiopathic nephrotic syndrome that comprises bio-banked patient samples from a broad range of clinical diagnoses and kidney functional states. Access to diagnostic biopsies and biofluids enables in-depth histological and molecular analysis and offers unique opportunities for a mechanistic disease understanding. Importantly, anonymized rich clinical data are available through the UK Renal Registry, allowing for a comprehensive view on CKD heterogeneity. Here we used unsupervised molecular clustering and dimensionality reduction to integrate and visualize the complex relationship between molecular, morphological and clinical data in this large cohort. We aim to generate mechanistic disease understanding for patient-centric, integrated target and biomarker discovery that will enable the development of novel precision treatments. METHODS Diagnostic kidney biopsies were obtained from the NURTuRE biobank. Samples from 286 patients comprising 16 primary diagnoses were RNA-sequenced and passed extended control of library quality and composition. Self-organizing maps (SOM) as implemented in the OposSOM R package were used for unsupervised clustering of transcriptomes, after adjusting for batch effects and sex. We used PHATE dimensionality reduction to visualize global data structure with the aim to project clinical, morphological and molecular data onto the resulting nonlinear disease trajectories. RESULTS Unsupervised clustering of kidney transcriptomes inferred five groups with distinct molecular landscapes (F, E, C, A and AB) that were generally consistent with molecular clusters previously described for CKD.1 Correlation of metagenes revealed a highly polarized global data structure with cluster C connecting F and E with the opposing clusters A and AB. Gene set enrichment analysis suggested that the polarization results from strong opposing metabolic and immune signatures, likely reflecting tubular atrophy, immune response and infiltration (Figure 1A). Consistently, cluster identity was independent of primary diagnosis but aligned with disease progression as reflected by kidney function decline (eGFR), increasing inflammation (CRP) and age range (Figure 1B). PHATE dimensionality reduction revealed complex nonlinear relations of transcriptomes aligned with disease progression and interpreted as molecular disease trajectories partitioned by SOM clusters (Figure 1C). We are now integrating real-world data, including longitudinal kidney function and free-text histopathology reports, and are expanding the molecular analysis by proteomics and genomics. Initial exploration of histopathology and eGFR series suggests a nonrandom distribution of morphological risk factors and disease progression rates across molecular clusters and trajectories, highlighting unique opportunities for drug discovery. CONCLUSIONS Molecular stratification aligns with disease progression irrespective of clinical diagnosis, reflecting common cellular and molecular mechanisms of disease associated with kidney function decline and substantial morphological changes. Further integration of complementary datasets, including clinical time series, histopathology and multiomics, will enable mechanistic interpretation of molecular clusters and trajectories with the goal of identifying novel, targetable drivers of CKD in well-defined patient subsets.