Deep Learning-Assisted Nephrotoxicity Testing with Bioprinted Renal Spheroids

Abstract

We used arrays of bioprinted renal epithelial cell spheroids for toxicity testing with cisplatin. The concentration-dependent cell death rate was determined using a lactate dehydrogenase assay. Bioprinted spheroids showed enhanced sensitivity to the treatment in comparison to monolayers of the same cell type. The measured dose-response curves revealed an inhibitory concentration of the spheroids of IC $_{50}$ = 9 ± 3 μM in contrast to the monolayers with IC $_{50}$ = 17 ± 2 μM. Fluorescent labeling of a nephrotoxicity biomarker, kidney injury molecule 1 indicated an accumulation of the molecule in the central lumen of the spheroids. Finally, we tested an approach for an automatic readout of toxicity based on microscopic images with deep learning. Therefore, we created a dataset comprising images of single spheroids, with corresponding labels of the determined cell death rates for training. The algorithm was able to distinguish between three classes of no, mild, and severe treatment effects with a balanced accuracy of 78.7%.