Date Published: May 28, 2019
Publisher: Public Library of Science
Author(s): Gregory S. Honda, Robert G. Pearce, Ly L. Pham, R. W. Setzer, Barbara A. Wetmore, Nisha S. Sipes, Jon Gilbert, Briana Franz, Russell S. Thomas, John F. Wambaugh, Jason A. Papin.
Linking in vitro bioactivity and in vivo toxicity on a dose basis enables the use of high-throughput in vitro assays as an alternative to traditional animal studies. In this study, we evaluated assumptions in the use of a high-throughput, physiologically based toxicokinetic (PBTK) model to relate in vitro bioactivity and rat in vivo toxicity data. The fraction unbound in plasma (fup) and intrinsic hepatic clearance (Clint) were measured for rats (for 67 and 77 chemicals, respectively), combined with fup and Clint literature data for 97 chemicals, and incorporated in the PBTK model. Of these chemicals, 84 had corresponding in vitro ToxCast bioactivity data and in vivo toxicity data. For each possible comparison of in vitro and in vivo endpoint, the concordance between the in vivo and in vitro data was evaluated by a regression analysis. For a base set of assumptions, the PBTK results were more frequently better associated than either the results from a “random” model parameterization or direct comparison of the “untransformed” values of AC50 and dose (performed best in 51%, 28%, and 21% of cases, respectively). We also investigated several assumptions in the application of PBTK for IVIVE, including clearance and internal dose selection. One of the better assumptions sets–restrictive clearance and comparing free in vivo venous plasma concentration with free in vitro concentration–outperformed the random and untransformed results in 71% of the in vitro-in vivo endpoint comparisons. These results demonstrate that applying PBTK improves our ability to observe the association between in vitro bioactivity and in vivo toxicity data in general. This suggests that potency values from in vitro screening should be transformed using in vitro-in vivo extrapolation (IVIVE) to build potentially better machine learning and other statistical models for predicting in vivo toxicity in humans.
Relatively few chemicals in commercial use have been fully evaluated for hazard, in part due to the resource intensive nature of in vivo animal testing [1–4]. To address concerns over the potential health effects of data-poor chemicals, new approach methodologies for chemical toxicity testing based on high-throughput in vitro and computational tools are being developed by researchers from government, industry, and academia . High-throughput screening assays, such as those used in the Tox21 and ToxCast programs, provide in vitro bioactivity data that may inform the potential hazard of a chemical [6, 7]. To link in vitro assays with particular in vivo endpoints, statistical and machine learning models have been developed that select and weigh the potency and hit call data from relevant assays [8–10]. Using toxicokinetics (TK) may potentially improve performance of such models and elucidate the general correlation between in vitro bioactivity and in vivo toxicity data [11–13]. Since the probability of a biochemical interaction is proportional to the chemical concentration of ligand at the receptor [14, 15], the 2007 National Academies of Sciences, Engineering, and Mathematics report “Toxicity Testing in the 21st Century” proposed that dose-response modeling using physiologically-based TK (PBTK) models is needed to use high-throughput screening data to estimate chemical risk . TK describes the mathematical relationship between external dose and internal concentrations, accounting for processes including absorption, distribution, metabolism, and excretion of a chemical . Utilizing TK, an in vitro bioactive concentration that is suggestive of potential hazard can be extrapolated to an administered equivalent dose (AED) on a mg/kg body mass/day basis, allowing for subsequent comparison to estimated exposure rates [2, 5, 17–21].
In this work, the parameters fup and Clint were measured in vitro for rat and incorporated in a PBTK model. These data were combined with previously published rat-specific in vitro TK data collected in the R package httk . A PBTK model was used to evaluate two dosimetry approaches for comparing high throughput screening in vitro bioactivity and rat in vivo toxicity data. A reverse dosimetry approach transformed in vitro concentrations to predicted administered equivalent doses. Conversely, a forward dosimetry approach transformed in vivo doses to predicted plasma concentrations. We restricted our evaluation to chemicals for which effects were observed for both in vitro bioactivity and in vivo toxicity data. For each combination of in vitro bioactivity and in vivo endpoint a regression analysis [24, 33, 34] was used to evaluate the performance of the PBTK model relative to the untransformed values and randomized PBTK results. Results were summarized based on the count of the number of times the PBTK model performed better than both a randomized result (y-randomized TK parameters) and comparison of the untransformed values (in vitro AC50 vs. in vivo dose). Processed data and models are provided in the R package httk  version 1.9 (https://cran.r-project.org/web/packages/httk/). All analyses were performed in R version 3.5.1. Input data and scripts for analysis are available in S1 File. A list of the abbreviations used in this work is included in S2 File.
Since the propensity for bioactivity is proportional to chemical concentration, TK frames the dose-response relationship and associated hazard characterization by linking external exposures (theoretical or relevant, depending on the scenario) to resultant internal concentrations. This work evaluates varying sets of IVIVE assumptions by examining the impact of a PBTK model on the association between in vitro bioactivity and in vivo toxicity data. In vitro HTTK parameters of fup and Clint were measured for 65 new chemicals and analyzed jointly with data from the literature for 97 chemicals. For the data considered, TK parameters, in vitro AC50, and in vivo doses were simultaneously available for 84 chemicals. Forward and reverse dosimetry results were determined using the PBTK model for each set of assumptions, and comparisons were made using a regression analysis. Two different in vivo data sets were used for the basis of this analysis: the endpoint level data containing 80 chemicals with doses corresponding to observed responses for 106 specific endpoints (68 pathological responses and 3 study types), and the POD level data containing 84 chemicals where effects had been aggregated into a single point of departure per chemical (the minimum LOEL-LOAEL taken across all available pathologies, studies, and study types). For the endpoint level data, different in vitro assay–in vivo endpoint combinations had different numbers of chemicals that were both active in vitro and that had a particular in vivo endpoint observed; 2787 endpoint combinations had at least 5 chemicals, while 48 combinations had at least 20. For the comparisons made between the POD level data and in vitro assay endpoints, there were 69 comparisons with at least 5 chemicals that were active in vitro and in vivo, while 17 comparisons had at least 20 chemicals.
Understanding our ability to use in vitro and in silico methods to quantitatively predict known doses exhibiting pathological effects in vivo is a prerequisite to estimating toxic doses for chemicals without in vivo toxicological data. In this work, the application of a PBTK model to clarify the association between in vitro bioactivity and in vivo toxicity data [11, 13, 27] was evaluated across a broad range of chemicals, in vitro assays, in vivo endpoints, and modeling assumptions. Evaluations were carried out for two analysis levels: 1) each specific in vivo endpoint was compared with each in vitro high-throughput screening assay component endpoint and 2) POD determined across pathologies and study types were compared with each in vitro high-throughput screening assay component endpoint. For both analysis levels (endpoint level and POD level, respectively), results were compared for forward dosimetry and reverse dosimetry. In the endpoint level analysis, both the forward and reverse dosimetry results demonstrated an improved performance when using the PBTK model. This strongly suggests that applying toxicokinetic models elucidates the association between in vitro bioactivity and in vivo toxicity data, particularly when study type and specific effect are considered.