Abstract

Chronic Kidney Disease (CKD) affects about 10% of the world’s population but there is currently no treatment targeting its causes. The disease pathogenesis is complex and involves dysfunction of multiple cell types that can ultimately lead to kidney failure. To identify potential novel drug targets for CKD, we performed high-throughput CRISPR knock-out (KO) screens in two disease-relevant renal cell models using high-content imaging endpoints. 

First, we established gene editing strategies in 2 renal cell models, primary Human Glomerular Microvascular Endothelial Cells (HGMEC) and Renal Proximal Tubule Epithelial Cells (RPTEC). Using electroporation of gRNA/cas9 ribonucleoprotein (RNP), we achieved gene editing efficiencies over 95%.

Second, we established cytokine stress assays using IL1β and TNFα to mimic renal cell injury and dysfunction resulting in the loss of cell junction markers (CD31 in endothelial cells and ZO-1 in epithelial cells) which contributes to CKD pathogenesis. The TNFα and IL1β stress assays were analysed by high content confocal imaging of the cell junction markers CD31 and ZO-1.

The imaging assays were optimised for automation workflows and coupled to gene editing to enable CRISPR KO screening in an arrayed format.  

We screened a CKD-related gene list of 226 genes in presence or absence of cytokine stress with the aim of identifying genes that are able to prevent or induce the disease-like phenotype.

The imaging endpoints of the CRISPR screen were analysed using deep learning-based image analysis to separate the healthy and stressed phenotypes and to detect gRNAs leading to a significant effect compared to a neutral control gRNA. The arrayed CRISPR screens in primary renal cells showed excellent robustness and identified several hits, amongst them genes involved in the IL1β and TNFα signalling pathways. 

Moreover, an artificial intelligence (AI)-outlier detection tool was implemented to detect gene KOs that differ from the healthy phenotype without resembling the cytokine stressed phenotype. This allowed us to identify hits that induce different phenotypes beyond the tested IL1β and TNFα stress and so to unveil “Genotype-Phenotype” mechanism of action that could be relevant for the disease.