Accurate Prediction of Human Organ Toxicity

NEPHROTOXICITY PREDICTION

This high-throughput method (Su et al., 2016) was awarded the Lush Science Price 2016 and combines cell-based assays with bioinformatics. The method predicts specifically toxicity to renal proximal tubular cells (PTC), which are one of the main targets for chemical-induced toxicity in the kidney. The method can be used with human primary renal PTC, human kidney-2 (HK-2) cells (established proximal tubular cell line) or induced pluripotent stem cell (iPSC)-derived PTC-like cells (Kandasamy et al., 2015) cultivated in 384-well plates. Cells are imaged by high-content imaging after compound exposure. Chemical-induced toxicity to renal PTC in humans is predicted based on chemical-induced changes of predictive cellular phenotypic features, which were identified by phenotypic profiling and machine learning. With human primary PTC the method has a test sensitivity of 87% and a test specificity of 90%. Machine learning and 10 trials of 10-fold cross validation were applied for the predictive performance analysis. The method was applied in a case study on triazole fungicides initiated by the National Institute for Public Health and the Environment (Netherlands) (van der Ven et al., 2020), and for the prediction of the nephrotoxicity of >300 ToxCast chemicals in collaboration with the US Environmental Protection Agency.

PREDICTION OF HEPATOCYTE TOXICITY IN HUMANS

This high-throughput method (Hussain et al., 2020) works with HepaRG cells in 384-well plates. Similar to the method for nephrotoxicity prediction, it is based on a combination of high-content imaging, phenotypic profiling and machine learning. The method has been validated with 69 reference chemicals that were known to be toxic or not toxic for hepatocytes in humans. It has a test sensitivity of 73% and test specificity of 92%. Machine learning and 10 trials of 10-fold cross validation were applied for the predictive performance analysis.

PREDICTION OF VASCULAR ENDOTHELIAL TOXICITY IN HUMANS

This method based on high-content imaging has been validated with a reference set of 32 chemicals that are known to be toxic or not toxic for vascular endothelial cells in humans. It has a sensitivity and specificity of 89% and 96%, respectively. The method was established with human umbilical vein endothelial cells (HUVEC), and similar results were obtained with 3 other types of human primary vascular endothelial cells.

References

1. F. Hussain, S. Basu, J. J. H. Heng, L.-H. Loo and D. Zink, “Predicting Direct Hepatocyte Toxicity in Humans by Combining High-Throughput Imaging of HepaRG Cells and Machine Learning-Based Phenotypic Profiling,” Archives of Toxicology, 94[8] (2020) 2749-2767. DOI:10.1007/s00204-020-02778-3 Link

2. K. Kandasamy, J. K. C. Chuah, R. Su, P. Huang, K. G. Eng, S. Xiong, Y. Li, C. S. Chia, L.-H. Loo and D. Zink, “Prediction of Drug-Induced Nephrotoxicity and Injury Mechanisms with Human Induced Pluripotent Stem Cell-Derived Cells and Machine Learning Methods,” Scientific Reports, 5 (2015) 12337. DOI:10.1038/srep12337 Link

3. L. T. M. van der Ven, E. Rorije, R. C. Sprong, D. Zink, R. Derr, G. Hendriks, L.-H. Loo and M. Lujiten, “A Case Study with Triazole Fungicides to Explore Practical Application of Next-Generation Hazard Assessment Methods for Human Health,” Chemical Research in Toxicology, 33[3] (2020) 834-848. DOI:10.1021/acs.chemrestox.9b00484 Link

4. R. Su, S. Xiong, D. Zink and L. -H. Loo, “High-Throughput Imaging-Based Nephrotoxicity Prediction for Xenobiotics with Diverse Chemical Structures,” Archives of Toxicology, 90[11] (2016) 2793-2808. DOI:10.1007/s00204-015-1638-y Link

KEY HIGHLIGHTS

• High accuracy

• High throughput

• Toxicity prediction by machine learning

• Not restricted to specific classes of chemicals

• Award winning

• Applied in collaborations with regulators