Integration into risk assessment of open source human omics data from in vitro studies
Douglas Connect GmbH, Switzerland
Background: A milestone in toxicity research was the emergence of toxicogenomics, resulting from the application of knowledge gained from genomics science into conventional toxicology. Toxicogenomics specifically tackles the complex interactions between toxic effects and the structure and activity of the genome. Thus, predictive toxicology is undergoing a paradigm shift, from phenomenological to mechanistic (e.g. -omics data based) models that may represent an important alternative to the classical in vivo approach applied in chronic and systemic toxicity testing. Along those lines, numerous international projects have recently generated and gathered valuable data and created open data warehouses that are required to build these predictive models.
Objective: Within the SEURAT-1 - ToxBank project, case studies have been proposed in which the toxicity of different compounds is assessed using only available human in vitro data. In one of these, ab initio risk assessment is modelled, i.e. it is assumed that no information on the toxicity of the compound is available. An approach using the combination of omics data with information extracted from adverse outcome pathways (AOPs) to identify areas of concern and support an evidence-driven risk assessment was proposed and performed on piperonyl butoxide (PBO).
Methodology: To quickly identify areas of concern, omics data represents a good starting point, since they show general adaptations of the cell to the exposure. We have identified and used a set of transcriptomics measurements performed on three different human liver in vitro models (HepaRG, HepG2 and hES-DE-Hep). The methodology applied included (i) identification of relevant pathways (the transcriptomics data was analyzed using the program InCroMAP), (ii) correlation between pathways and diseases (the top pathways were selected and analyzed further using the services of the Comparative Toxicogenomics Database), and (iii) verification of adverse effects versus specific AOP Key Events (using the information from the Adverse Outcome Pathway Knowledge Base).
Results: We showed that transcriptomics data is able to identify fibrosis as one major adverse outcome of treatment with PBO and that HepaRG was the most appropriate cell model to be used for testing this specific adverse effect. Using information from AOPs, we were able to verify key events and, in this way, strengthening the evidence for this specific adverse effect. Finally, we proposed some additional testing on some key events for which we could not identify available data.