In this work, new in vitro and molecular techniques were applied to establish a new, early test system for toxicological research. A wide range of alternative approaches are currently being developed to gain mechanistic information, to speed up the process of early screening in drug development, to improve the toxicological testing procedure itself and, of course, to reduce the number of animals used for toxicity testing. At the same time, new technical developments and options are being adopted into toxicology laboratories and tested for their suitability and robustness. One promising approach is the analysis of gene expression changes by microarrays (Amin et al., 2002). The combination of both of these basic approaches, in vitro experiments and modern technology, will help to answer some of the key questions faced by toxicology.
Primarily, the applicability of two commercially available gene expression platforms was examined by a thorough comparative study of data gained from in vitro as well as in vivo experiments. Our results demonstrated that the high quality and correlation of generated data on a technical level lead to a high concordance in terms of the biological interpretations, making both platforms applicable for use in toxicological studies. This result was supported by the high correlation with TaqMan gene expression data. Recently, the FDA initiated a microarray “control” study (MAQC), which clearly showed the intra- and interlaboratory comparability of microarray results as well as the consistent results obtained from different microarray platforms (Guo et al., 2006; Shi et al., 2006).
The comparison of several in vitro culture systems, each with their own advantages and disadvantages in terms of throughput, viability and metabolic activity (Table 2), on both morphological and functional levels, as well as the global gene expression level permitted insights into basal mechanisms which take place during cell culture. The combination of both global gene expression and primary hepatocytes has been performed before in smaller studies covering only limited, more specific questions, when compared to the data presented here in this thesis (Baker et al., 2001; Boess et al., 2003; Braeuning et al., 2006). This PhD work was an important step towards the understanding of how varying culture conditions affect hepatocellular differentiation and function. At the same time, this comparison and subsequent optimizations lead to the establishment of a standardized and robust long-term hepatocyte culture system with clearly characterized morphological, functional and gene expression functions.
All of this data was necessary to allow for good data interpretation based on the background level of gene expression during culturing and to define the horizon of expectation to ensure the reliability of this test system.
The main problem of all primary hepatocyte cultures is the reduction of metabolic activity over time in culture. While this is true for short term cultures like suspension cultures, liver slices and ML cultures, our data showed a deceleration of this process by culturing the hepatocytes in the SW conformation without FCS. Not only the basal gene expression of several CYPs was found to be higher in SW- cultures, but also the treatment with well known inducers resulted in an improved inducibility of the four CYPs tested. These findings are supported by published data on both the functional level as well as in terms of gene expression (Elaut et al., 2006a; LeCluyse et al., 2000; Richert et al., 2002; Rogiers & Vercruysse, 1998; Coecke et al., 2005).
These results provided us with confidence to go forward with this in vitro culture system for a toxicogenomics study using several well known hepatotoxicants to show compound dependent gene expression changes and to compare different mechanisms of action. This data was not only used for mechanistic analyses but also to successfully develop a computer based discrimination model for hepatotoxicity. Up to now, studies employing such predictive models are based on in vivo data and are mainly focused on acute toxicity (Hamadeh et al., 2002b; Zidek et al., 2007; Ellinger-Ziegelbauer et al., 2008; Ruepp et al., 2005). This model is the first study combining in vitro toxicology and toxicogenomics to test the possibility of using primary hepatocytes dosed for 9 d to depict sub-chronic toxicity.
Surprisingly, even though a relatively small database was used, the classification of the compounds used was successfull, with a misclassification rate of only 7.5% after 9 days. Knowing the fact that multiple gene expression changes are caused by the perfusion itself and the adaption to the culture conditions, this is a high-quality result and reflects the robustness of this in vitro system to predict the in vivo outcome.
The resulting discrimination model was challenged with two blinded compounds to prove its ability do detect hepatotoxicity based on global gene expression. EMD X is a former Merck compound which was stopped in development and is known to be hepatotoxic. Using our model it was clearly predicted to be hepatotoxic. AAP has been reported to lose toxic potency in primary hepatocytes over time in culture (Jemnitz et
mechanistic processes taking place in culture and the insensitivity of primary hepatocytes to AAP toxicity.
In the last few decades, a new paradigm has emerged based on the assumption that knowing the mechanism of action of a toxic compound would enable the development of predictive models which would help new, safer compounds to be brought quicker onto the market. The search for adaptive changes in gene expression has resulted in many genes being proposed as predictive biomarkers, although only a few of them have been shown to be really decisive. Currently, new techniques in bioinformatic analysis has lead to the identification of gene signatures and networks which seem to contain more information and therefore to be more reliable than single gene biomarkers (Khor et al., 2006).
The ultimate goal of these in vitro toxicogenomic studies is the establishment of a predictive screening model which is easy to use and which delivers reliable, high quality results. The results presented here are very promising, but this study is just the starting point for a more thorough classification process. As mentioned before, the size of the database used for classification is crucial for the validity of the system. This is highlighted by the fact that the best results were obtained with the whole dataset (low and high dose together). Is it really beneficial to combine two dosing schemes, or is the improvement due to the increasing size of the dataset? The high dose was chosen due to the reduction of cell viability, but changes in gene expression resulting from low dose treatment were seen as well. These low-dose effects may also contain important information for the prediction model.
Another important point to consider is the dosing-scheme itself. Always controversially discussed (Monro, 1990; Campbell & Ings, 1988) and of central importance to the outcome of any in vitro experiment, there are currently no specific guidelines available. To avoid false positive or negative results, a list of general criteria would be helpful to exclude unsuitable samples due to incorrect dosing or differences in the culturing conditions. In toxicology testing, doses greatly in excess of pharmacologically active doses are used to induce adverse effects, therefore there might be effects obtained also for (in vivo) non toxic compounds, leading to false results. On the other hand, if a threshold value is not achieved, even toxic compounds may be classified as non toxic. A potential solution would be the application of a minimum number of deregulated genes according to t-test statistic and/or fold-change. A minimum set of deregulated genes might be adequate for discrimination. Whereas for non toxic compounds the genes affected should either be involved in non-damaging processes or random, toxic compounds should generate gene profiles clearly connected to adverse cellular fate
and viability. The conduction of these tests with multiple doses, which is enabled by in vitro experiments, is also a possibility to increase data quality.
The compound selection allowed a proof of concept for the constructed prediction model, although it was too small to cover all of the various potential mechanisms of hepatotoxicity. The gene set of 724 genes was capable of discriminating the compounds used to build the model, as well as to correctly classify newly added compounds with a misclassification rate of 7.5%. These results need to be further validated and refined, by including more compounds with specific modes of action or to focus a certain compound classes. This will increase the robustness of the predictive system and facilitate improved data interpretation.
Finally, the insecurity of extrapolating the results in between species, especially to men, may be overcome by the possibility to conduct these experiments with human hepatocytes. Also human hepatocytes can be successfully cultured in either ML- or SW-conformation, there is still the need to optimize the culture conditions. Because of the difficulties and the costs of getting high quality human hepatocytes in a sufficient amount, there might also be other options like the new HepaRG cell line which may be considered. Yet, the data obtained during this work is promising but not sufficient to attest the qualification of either possibility.
To conclude, screening tests alone do not allow for a final estimation of the hazard and risk of a compound, but molecular toxicology can contribute by improving the mechanistic understanding, refining the predictivity of toxicological outcomes and to significantly reduce animal usage in toxicology and, more generally, in drug discovery. We have now a robust, semi-validated long-term cell culture system that can be used in drug discovery for predicting hepatotoxicity as well as helping the toxicologist to understand a compounds mechanism of action. Therefore, the development of this predictive in vitro test system can be seen as a contribution to the efforts to implement the principles of 3R into the daily toxicological work.