In order to reliably assess the risk of adverse systemic effects of chemicals by using in vitro methods, there is a need to simulate their absorption, distribution, metabolism, and excretion (ADME) in vivo to determine the target organ bioavailable concentration, and to compare this predicted internal concentration with an effective internal concentration. The effective concentration derived from in vitro toxicity studies should ideally take into account the fate of chemicals in the in vitro test system, since there can be significant differences between the applied nominal concentration and the in vitro bioavailable concentration. Whereas PBK models have been developed to simulate ADME properties in vivo, the Virtual Cell Based Assay (VCBA) has been developed to simulate in vitro fate. In this project, the VCBA model in R code, was applied to better interpret previously obtained in vitro acute toxicity data and study how they can be compared to results from acute toxicity in vivo.
For 178 chemicals previously tested in vitro with the 3T3 BALB/c cell line using the Neutral Red Uptake cytotoxicity assay, physicochemical parameters were retrieved and curated. Of these chemicals, 83 were run in the VCBA to simulate a 96-well microplate set up with 5% serum supplementation, and their no effect concentration (NEC) and killing rate (Kr) optimized against the experimental data. Analyses of results of partitioning of the chemicals show a strong relation with their lipophilicity, expressed here as the logarithm of the octanol/water partitioning coefficient, with highly lipophilic chemicals binding mostly to medium lipid. Among the chemicals analysed, only benzene and xylene were modelled to evaporate by more than 10 %, and these were also the chemicals with highest degradation rates during the 48 hours assay. Chemical degradation is dependent not only on the air and water degradation rates but also on the extent of binding of the chemical.
Due to the strong binding of some chemicals to medium lipids and proteins we analysed the impact of different serum supplementations (0%, 5% and 10%) on the chemical dissolved concentrations. As expected, for the more lipophilic chemicals, different serum levels result in different dissolved concentrations, with lipid and protein binding reducing chemical loss by evaporation. Still the lack of saturation modelling might mislead the 0 % supplementation since the lipids coming solely from cells exudates are able to sequester chemical to a large extent, eg. after 48 hours, 63% (1.2E-5 M) of dimethyldioctadecylammonium chloride was bound to lipid from the cells. Although highly lipophilic chemicals have a very small bioavailable fraction, cellular uptake rate is also dependent on logKow, which compensates for this lack of bioavailability to some extent.
Based on the relevance of lipophilicity on in vitro chemical bioavailability, we have developed an alert system based on logKow, creating four classes of chemicals for the experimental condition with 10% serum supplementation: logKow 5- 10 (A), logKow <5 (B), logKow <2.5 (C), and logKow <2 (D). New chemicals from Classes A and B, which will in the future be tested in vitro, were run first on the VCBA, without considering toxicity (NEC and Kr set to 0). VCBA simulations indicated that these chemicals are more than 50% bound to medium proteins, lipids and plastic. Therefore, for chemicals with logKow falling in these classes, special care should be taken when extrapolating the obtained in vitro toxic concentrations to in vivo relevant doses.
A comparison of the VCBA-predicted dissolved concentrations corresponding to nominal IC50 values with the available rat oral LD50 values did not improve the previously obtained correlations. This is probably because other in vivo kinetic processes play an important role but were not considered in this in vitro-in vivo extrapolation.
The comparison of the VCBA predicted IC50 dissolved concentrations with the available rat oral LD50 values, did not improve the previously obtained correlations. Nevertheless, other in vivo kinetic processes that are not modelled may play an important role. They should be considered in the in vitro-in vivo extrapolations.
A local sensitivity analysis showed the relative low impact of Molar Volume and Molecular Diffusion Volume on the final dissolved concentration, supporting the use of approximated values obtained through the herein created QSARs. The logkow and Henry Law Constant showed, as expected, a high impact in partitioning. Killing rate was shown to also have a relative low impact in the final chemical concentration, indicating that although its optimization is important, finding the Kr that leads to the absolute best correlation between experimental and predicted concentration-viability curves, is not imperative. The VCBA can be applied to virtually any chemical as long as the physicochemical data (for the fate model) and the experimental toxicity data (that include cell growth/death) are available. However, being such a generic model, several assumptions had to be made: i) no distinction of chemical classes (inorganic, polar organic chemicals), ii) no consideration of metabolism, iii) saturation kinetics and iv) external in vitro conditions.
The advantages of having a generic model are that the VCBA can fit several experimental set ups and should be used in an exploratory manner, to help refinement of experimental conditions. The herein obtained VCBA results should be double checked experimentally the partition with a set of chemical compounds to better understand to what extent VCBA represents chemicals of different properties.
In future developments, it would be important to reduce the uncertainties of the model such as binding-saturation and consider inclusion of other endpoints such as metabolic activity.