Comparison of physiologically based pharmacokinetic modeling platforms for developmental neurotoxicity in vitro to in vivo extrapolation

This study describes an in vitro to in vivo extrapolation (IVIVE) approach to derive human-relevant administered equivalent doses based on chemical partitioning into developmental neurotoxicity (DNT) target organs during the critical period of brain development. Three physiologically based pharmacokinetic (PBPK) modeling platforms were evaluated for their suitability for this DNT-IVIVE approach. This dataset includes the model code for the two open-source platforms used for this approach- the U.S. EPA's R package, httk, and PK-Sim from Open Systems Pharmacology.

Methods
PBPK modeling was performed in batch mode in each platform using a pregnancy model at 15 gestation weeks (GW) and 24GW. For modeling of infants, a standard PBPK model was run at 2 weeks and 6 months of age for each platform; for httk, this was the preliminary brain—adipose model described in Unnikrishnan et al. (2024) (Unnikrishnan. et al. in prep). The individual for simulation was created based on age for PK-Sim, which accounts for ontogeny. As httk does not consider ontogeny, age was set in httk by body weight using the GastroPlus (Simulations Plus) default weights for a 2 week and 6 month-old male, which were 3.7 kg and 8.29 kg, respectively. Chemical exposure was modeled as a single oral bolus of 1 mg/kg body weight and was simulated for a 24-hour period. As the maximum number of integration steps that GastroPlus can run in batch mode is 500, the integration time interval was divided into 500 steps for all platforms to ensure equivalent granularity across platforms for estimation of the maximal concentration (Cmax).

To implement this as a mid-throughput approach, PBPK modeling was conducted in batched mode. For httk, the R code was written to allow for batched modeling using both the pregnancy and brain—adipose models. Several chemicals would not run due to falling outside httk’s applicability domain. Therefore, these chemicals were forced to run using the command, “physchem.exclude = FALSE” or “class.exclude = FALSE.” While PK-Sim is typically run using a GUI, the GUI does not allow for batched modeling. Therefore, code was written in R to perform batched modeling for PK-Sim. The R code for both httk and PK-Sim is freely available here on Dryad, as well as on GitHub (https://github.com/esqLABS/pregnancy-neonates-batch-run/tree/master).

Values for Cmax were derived for DNT-relevant compartments—fetal, fetal venous, and maternal plasma at 15GW and 24GW, brain and plasma at 2w and 6m, and fetal brain at 15GW and 24GW in httk. To allow for comparison between models, code was written in Python to compute a single effective fetal compartment concentration by volume-averaging the concentrations of the fetal sub compartments (code is available through this Dryad repository and at https://github.com/esqLABS/pregnancy-neonates-batch-run/tree/master).