20190908 - Chemical risk assessment: How well do in vitro and in silico data predict the in vivo situation? (EuroTox)

Modern high-throughput screening technologies (HTS) (Houck et al., 2013) are enabling the use of use of new approach methodologies (NAMs) that can provide information about chemical hazard and risk assessment without using whole animals (ICCVAM, 2018). Gaining broader acceptance of NAMs in risk assessment will require comparisons between the point of departure (POD) from HTS and from traditional animal testing studies (Kavlock et al., 2018; Thomas et al., 2019). High-content imaging (HCI), a type of HTS technology, is widely used to evaluate drug safety (O'Brien et al., 2006) and to screen chemicals for toxicity (O'Brien and Edvardsson, 2017). By using single-cell data, "sublethal" cellular alterations identified by HCI such as, mitochondrial activity, lysosomal mass, oxidative stress and apoptosis are found to be important for translating in vitro effects to in vivo safety (O'Brien, 2014). Here we describe a case study in which HCI was used to estimate in vitro potency values in rat hepatocytes and that were quantitatively extrapolated to in vivo oral administered equivalent doses (AEDs) using high-throughput toxicokinetic (HTTK) modeling (Pearce et al., 2017) . The AEDs were then compared with in vivo rat liver POD values from repeat-dose subchronic (90d) and chronic (3y) exposures. We first selected 51 chemicals from ToxRefDB v2.0 that produced hepatic effects following repeat-dose subchronic (90d) and chronic (3y) exposures. Second, we treated rat primary hepatocytes with 10 concentrations (0.2 to 100µM) of these chemicals. Third, we used high-content imaging (HCI) to measure endoplasmic reticulum stress, mitochondrial function, lysosomal mass, steatosis, apoptosis, DNA texture, nuclear size and cell number at three time points (24, 48 and 72 h). Fourth, we used the HCI data to estimate AC50 values for all in vitro endpoints and time points using curve-fitting (Filer et al., 2016). Finally, we extrapolated the AC50 values to administered equivalent doses (AED) by toxicokinetic modeling with multiple dose metrics and compared them with hepatic lowest observed adverse effect levels (LOAELs). On average, AED derived from HCI in rat primary hepatocytes were 7 folds lower than rat liver chronic and subchronic LOAELs. While the ratio between the LOAEL and AED values varied by the HCI endpoint, in vitro exposure duration and choice of dose metric in toxicokinetic modeling, 60% of the ratios were within 10 folds. In contrast, AED based on diverse in vitro assays from ToxCast were 58 folds lower than LOAELs. Our findings demonstrate the feasibility of NAMs using HTS and HTTK for screening level assessments and provide a systematic appraisal of multiple in vitro factors on the predicted margin of safety. We believe that more sophisticated computational approaches will further improve the performance of NAMs for predicting health effects. This abstract does not reflect US EPA policy.