Below you find the results from the different workpackages within the ACuteTox project.

ACuteTox Executive summary
Executive summary 2008
Executive summary 2007
Executive summary 2006
Executive summary 2005
WP1: Generation of a database with toxicity data from animal tests and human accidents (in vivo data)
WP2: Generation of a database with toxicity data from animal-free tests (in vitro data)
WP3: Iterative amendment of the testing strategy
WP4: New cell systems and new endpoints
WP5: Improving in vitro/in vivo correlations – Absorption, Distribution, Excretion
WP6: Improving in vitro/in vivo correlations – Metabolism
WP7.1: Improving in vitro/in vivo correlations – Neurotoxicology
WP7.2: Improving in vitro/in vivo correlations – Nephrotoxicity
WP7.3: Improving in vitro/in vivo correlations – Hepatotoxicity
WP8: Development of the testing strategy
WP9: Pre-validation of the testing starategy
Main achievements and recommendations


ACuteTox Executive summary

The overall aim of the EU-funded project, ACuteTox was to develop and pre-validate a simple and robust in vitro testing strategy for prediction of human acute oral systemic toxicity. The extensive amount of work performed since the 1970s has led to the large number of existing in vitro models for acute toxicity testing. Many studies have shown good correlation between in vitro basal cytotoxicity data and rodent LD50 values, as well as human blood concentrations. However, a certain number of misclassifications will occur when the existing tests are used. ACuteTox aimed to identify factors that could optimise the in vitro - in vivo correlation for acute oral systemic toxicity.

Ninety seven reference chemicals were selected to represent different chemical classes for which reference acute oral lethality data were available from humans and rodents and were tested using 6 basal cytotoxicity assays. Moreover, human and animal in vivo data for these substances were collected from published literature. This led to identify the outliers from the in vitro-in vivo correlations, both using calculated human lethal concentrations (LC50) and rat oral LD50 data. Fifty seven reference chemicals, including the identified outliers and a balanced number of non-outliers from the reference set, were then tested in more than 70 tests methods covering oral absorption, distribution, clearance, metabolism and specific organ- and system-toxicity, such as haemato-, immuno-, neuro-, nephro- and hepatotoxicity. The overview of project activities and interaction between the different work packages is presented in Fig. 1.

The data generated in the project were stored in a novel internet-based database (AcutoxBase) developed within the project, and were used to assess the within-laboratory variability, the preliminary predictive capacity and in some cases also the between-laboratory variability of each in vitro assay.

Comprehensive statistical analyses of the collected in vivo LD50 data provided an evaluation of the variability and reliability of the animal tests, interspecies correlations, predictive capacities with regard to official acute oral toxicity categories, and deduction of performance criteria for in vitro methods.

A Partial Least Square Analysis (PLS) was first performed on all in vitro data generated with 57 chemicals tested in 75 assays. The in vitro IC50 values calculated in each laboratory were correlated with human LC50 values and rat oral LD50 values and the best combinations of in vitro tests that improved the existing correlation with in vivo (rat and human) data were identified.

A more extensive analysis was performed to assess the variability, repeatability and reproducibility of the single assays and to identify possible in vitro/in silico strategies that would allow classifying chemicals into the official acute oral toxicity categories (according to EU Classification, Labelling and Packaging (CLP) and GHS). For this purpose the Classification and Regression Trees (CART) were used as the classification algorithm of choice.

Based on statistical analysis eight test methods were assessed as promising for inclusion in the testing strategy and therefore selected for participation in the prevalidation study:
1. The neutral red uptake assay using the 3T3 fibroblast cell line (3T3/NRU);
2. The cytokine release assay using human whole blood (IL-1, IL-6, TNF-alpha);
3. Inhibition of colony forming unit efficiency in human cord blood-derived cells stimulated with CFU-GM (CBC/CFU-GM);
4. Gene expression (GFAP, HSP-32, MBP and NF-H) and Uridine incorporation measuring the total mRNA synthesis in primary rat brain aggregate cultures;
5. Cytomic panel measuring oxidative stress (intracellular peroxidative activity, intracellular levels of superoxide anion, oxidized DNA base 8-oxoguanine) and cytotoxicity screening (intracellular Ca2+ levels, mitochondrial membrane potential, plasma membrane potential) in HepG2, SH-SY5Y and A.704 cells;
6. The MTT assay using primary rat hepatocytes;
7. Kinetic parameters (volume of distribution, protein binding, clearance, and oral absorption using Caco-2 cells and neuronal networks) for the estimation of the oral dose from the effective concentration observed in vitro;
8. The estimation of compound passage through the blood-brain barrier using neuronal networks (for neurotoxicity assays).

In addition, the classification analysis performed using the training set (57 compounds) allowed the development of several strategies by assessing at least three different combinations of the selected assays that were challenged later with the new test set.

During the prevalidation study 32 blind-coded chemicals were evaluated in the selected assays, concentration-response analyses were performed and classifiers were challenged with the results of the concentration-response analysis obtained with the test set of chemicals.

In addition to the CART methodology, Random Forests were used for the classification task during the prevalidation study. Two classification approaches were evaluated, 1) a single step procedure only with Random Forest classification algorithm and 2) a two step tiered testing strategy. The results of the classification analysis carried out allowed identification of the best performing in vitro testing strategies.

The strategy that uses the Random Forest model consisted of 7 assays including 9 endpoints (CFU-GM/human cord blood cells assay, 3T3/NRU assay, primary rat brain aggregates including the following 3 endpoints: lowest gene expression, HSP-32 expression, NF-H expression, SH-SY5Y cell line/Lowest EC value obtained in cytomic panel, HepG2 cell line/MMP, primary rat hepatocytes/MTT assay, whole blood/IL-1 release, in a single step procedure, gave the best correct classification rate (69.26%) and resulted in only 2 compounds which toxicity was underestimated, as compared to the official toxicity classification (brucine, paraquat).

All outliers with neurotoxic potential except for thallium sulphate were identified by one or more assay in the optimised neurotoxicity test battery. The aggregated rat brain cell cultures constituted the most sensitive cell model and the multi-endpoint GENE assay detecting alterations in transcript markers for neurons, astrocytes, oligodendrocytes and cellular stress, together with a marker for total RNA synthesis seems to be the most complete assay for alerting neurotoxicity.

A number of potential hepatotoxic and bioactivated compounds were not alerted for by the heptoxicity/bioactivation cell models. The likely reason from this is that these toxic effects are chronic and not acute.

Although not included in the final testing strategy, all the known nephrotoxins had an IC50 value less than 50μM in the TER assay, hence providing an additional alert to the data from the basal cytotoxicity tests for chemicals with potential to cause nephrotoxicity.