Stem Cell-Based Predictive Models

1 A hESC-Based Model After the protocol for generating hESC-derived HPTC-like cells had been established (described in section A hESC-Based Protocof ), a 2D in vitro model based on this cell type was developed (Li etal., 2014) (Table 23.1). This hESC-based model was tested with the same set of 41 compounds, which had been used for testing the HPTC-based model (see Section 23.3.2.1). Also the hESC-based model nsed drug-induced increases in IL6 and/or IL8 determined by qPCR as endpoint. When analyzed with the thresholding procedure, the values obtained for sensitivity and specificity were 0.68 and 0.84, respectively. Also for the hESC-based model case, all major performance metrics were determined and an AUC/ROC value of 0.8 was obtained. 

As similar compounds were screened with HPTC and hESC-derived HPTC-Uke cells, the results were directly comparable. In comparison to HPTC, the sensitivity of hESC-derived HPTC-like cells was relatively low. The mean and median values were 0.77 in case of HPTC (three batches), whereas a value of 0.68 was obtained with respect to hESC-HPTC-like cells. The lower sensitivity of hESC-derived HPTC-like cells was probably dne to lower expression levels of some transporters and receptors like MEG (Li et al., 2014), which are important for the nptake of PT-specific nephrotoxicants. The specificity was similar in case of HPTC and hESC-derived HPTC-like cells (0.84). Results obtained with a second independently differentiated batch of hESC-derived HPTC-like cells revealed that positive and negative results were reproducible (Li et al., 2014). 

In addition to the mRNA levels of 1L6 and 1L8, also established endpoints were measured with the best batch of HPTC (HPTCl, HPTC displayed interdonor variability) and hESC-derived HPTC-Uke cells. These endpoints included ceU viability, ATP depletion, GSH depletion, and LDH leakage. In aU cases the predictivity was substantially lower and the balanced accuracy ranged between 0.60 and 0.69, depending on the endpoint, hi contrast, the respective values for balanced accuracy were 0.90 (HPTCl) and 0.76 (hESC-derived HPTC-like ceUs) when 1L6/IL8 upregulation was used as endpoint. These results show that not only the selection of the ceU model is important, but also the endpoints must be carefully considered, hi the cases discussed here, good predictivity was only obtained when 1L6/IL8 expression was measured. 

STEM CELL-DERIVED RENAL CELLS AND PREDICTIVE RENAL IN VITRO MODELS 

2 Prediction of Drug-Induced PT Toxicity and Injury Mechanisms with an hiPSC-Based Model and Machine Learning Methods The weak points of the HPTC- and hESC-based models described previously (Sections 23.3.2.1 and 23.3.3.1) were the data analysis procedures. In order to improve result classification, the raw data obtained with three batches of HPTC and the 1L6/1L8-based model (Li et al., 2013) were reanalyzed by machine learning (Su et al., 2014). Random forest (RE), support vector machine (SVM), k-NN, and Naive Bayes classifiers were tested. Best results were obtained with the RF classifier and the mean values (three batches of HPTC) ranged between 0.99 and 1.00 with respect to sensitivity, specificity, balanced accuracy, and AUC/ROC (Su et al., 2014). Thus, excellent predictivity could be obtained by combining the lL6/lL8-based model with automated classification by machine learning. 

---

A race on the generation of stem cell-derived renal cells started in 2013. In January 2013 a protocol for the differentiation of hiPSC or hESC into IM was published (Mae et al., 2013). Odd-skipped related (OSR)l was used as the main IM marker, and up to 90% OSRO cells were obtained. More differentiated renal cell types were only obtained at low frequency, which was not sufficient for use of these cells in any application, including in vitro toxicology (Mae et al., 2013). However, briefly afterward the first protocol for the differentiation of human pluripotent stem cells (hPSC) into mature renal cells was published (Narayanan et al., 2013). This protocol was based on hESC, and the differentiated hESC-derived cells exhibited many features of HPTC and were therefore called HPTC-like cells. The hESC-derived HPTC-like cells were then directly used for the development of the first stem cell-based renal in vitro model for the prediction of DIN (Li et al., 2014). (This in vitro model will be described in more detail in the following.) Later in 2013 the race on the generation of stem cell-derived renal cells continued and various alternative protocols were developed (Lam et al., 2014 Xia et al., 2013 Kang and Han, 2014 Taguchi et al., 2014 Takasato et al., 2014) (Fig. 23.1). 

---