Translational quantitative systems toxicology to improve the understanding of the safety of medicines
Adverse drug reactions (ADRs) remain a major complication of patient therapy. The fear of ADRs is a major impediment to the development of new, safe and effective therapies. Of particular importance in this respect are “off-target reactions” which usually cannot be predicted from the known pharmacological properties of the drug. The main organs of concern for such reactions are the liver, the kidney and the cardiovascular and gastrointestinal systems. It is widely acknowledged that the prediction of ADRs, particularly at early stages in the drug development process, is poor: animal tests fail to predict human ADRs in 30% of the cases. Thus it is essential to improve our understanding of the mechanistic basis of such reactions, and then our ability to predict ADRs in a quantitative rather than just a qualitative sense. Further development of systems modelling approaches is a key component in an overall scientific strategy to implement a step change in the drug development process with respect to drug safety. Pharmacokinetic-pharmacodynamic (PK-PD) modelling already plays an essential role in drug development and is used to help define dose in adults and in children, to predict drug-interactions (induction & inhibition of transporters and drug-metabolising enzymes), and to assess the effects of genetic variation and disease on drug disposition. The philosophy underlying this proposal is to build on existing PK-PD models that have a physiological basis and define systemic as well as specific organ/cell exposure to drug and metabolites in a holistic fashion, and to develop translational quantitative systems-based toxicological models for each of the four organs that are the focus of this call. Whilst acknowledging the cautionary words of George Box that “all models are wrong, but some are useful”, a guiding principle of this project will be to provide models with a defined purpose and, importantly, to understand the limitations of the application of these models within the context of each organ system.
Our principal objective is to assemble and curate, in a form amenable to quantitative systems modelling, the large pre-existing datasets generated in previous and ongoing EU and worldwide projects, alongside EFPIA legacy data, apply these data in our modelling tasks, and then set up a framework of open collaboration with these projects. This will avoid unnecessary generation of new data and will ensure that the doors of the data-vaults are firmly fixed open, both for the duration of this project, as well as for future IMI2 projects. This will necessitate seamless interaction with EFPIA partners to leverage in-house data. Where pre-existing data are incomplete (e.g. insufficiently time-resolved or at a single drug concentration), data sets will be back-filled through new experimentation. Common or organ-specific biological pathways and networks associated with drug-induced pathology will then be identified and differences and similarities in network responses across species described, enabling improved read-across from in vitro test systems. Innovative multi-level and multi-scale approaches will permit the quantitative evaluation of toxicity using conjointly molecular network models and PBPK approaches. In this context, the term pathway may at times refer to a logic-based representation of elements in a cause and effect relationship based on existing knowledge, e.g. a KEGG or REACTOME or an Adverse Outcome Pathway (AOP), while a network approach may include applications where elements may self-assemble in an unbiased fashion based on coalescent properties of the biological systems, e.g. co-expression. A key feature of both pathway and network identifications and model development will be the integration of a panel of mechanistic quantitative and translational biomarkers. This unique selling point of our approach, facilitated through leading roles of project partners in different related consortia, will test the accuracy of read-across between species and the human-relevance of the models generated.
The goal of the TransQST project is to develop a Quantitative Systems Toxicology approach, employing pre-existing data where possible, in order to yield new insight into drug-induced toxicity. Therefore, the project is in complete alignment with the call IMI2-2015-06-01: “Development of Quantitative System Toxicology (QST) approaches to improve the understanding of the safety of new medicines”.
A central tenet ofour programme will be to understand and define the degree of human physiological and pharmacological relevance of any test system that has been (or will be) used for generating the input data for modelling. By adopting this approach, one will be able to usefully interpret what happens when the test systems are perturbed by drug exposure, and ensure translatability of modelling tools. Translational biomarkers (in vitro, in vivo and across species) are a core aspect of our approach in terms of understanding how to develop, model and apply mechanistic biomarkers in a QST setting.
Objective 1: Develop an open (for consortium members), focused and sustainable knowledge database to build system toxicology models.
We will develop data harmonization and mining tools to provide single-point access to ’omics type data, molecular network information and any models used or developed in the project. The knowledge management platform will be based on open-source tools, previous IMI investments and existing web services will be used wherever possible. Human and pre-clinical data that assists an expanded knowledge of mechanistic toxicity will be identified and curated and existing ontologies will be used to support harmonisation of data across public and private resources. Tool development will be associated with long-term maintained resources and platforms whenever possible to ensure sustainability.
Objective 2: Provide a clearer understanding of the translational confidence from non-clinical species to human.
Transfer of knowledge along the different phases of drug development is fundamental to improve the understanding of the safety of new medicines. In particular, cross-species extrapolation between different laboratory animals and translation to first-in-human trials is challenging because of the uncertain comparability of physiological processes. To improve the situation, the applicants will use established and novel approaches of physiologically-based pharmacokinetic (PBPK) modelling that allows translation of mechanistic knowledge from one species to another by specifically considering physiological and biochemical interspecies differences, based on model parameter domains of a target species into the PBPK model of a reference species, in both in vitro and in vivo systems. These simulations consider activities of relevant enzymes and transporters, species-specific physiology, plasma protein binding, enzyme and transport kinetics. Using our already established database of mouse and human pharmacokinetics and toxicity data, the accuracy of cross-species modelling will be determined. In exactly the same manner, we will take account of species differences in susceptibility to drug-induced disease. Moreover, drug toxicity networks (developed by identifying the relevant modules in protein interaction networks involving drug targets and toxicity-associated proteins) will be integrated into the knowledge platform to support the development of quantitative models of pathways related to the development of adverse events. Pathway similarity across species will be assessed using network alignment algorithms to evaluate divergences to toxicological insults. All work will be underpinned by knowledge of the physiological phenotype of test systems so that results can be explained in a robust translational context.
Objective 3:Support key risk assessment decisions, such as safety margin, clinical monitoring and reversibility, using mechanistic and quantitative modelling.
An important and as yet unaccomplished task in drug safety research is to predict concentrations of pharmaceutical compounds in tissues. For this purpose, the applicants will develop and apply integrated PBPK, spatio-temporal and metabolic models. These techniques will, for example, predict if a compound accumulates in specific cell types of a human tissue and whether the tissue concentrations exceed levels where adverse effects can be expected. We will utilise the DDMoRe interoperability framework (where possible) and other approaches (e.g. Lua scripting) to access existing PBPK models available within consortium partners. Plasma and organ/cell concentration data, as well as quantitative data from mechanism-based biomarkers, will then be linked to organ-specific systems biology models and logic-based modelling will be used to model the complex processes of drug toxicity and recovery following cessation of drug exposure. This will allow improved understanding of both reversible and irreversible clinical outcomes.
Objective 4: Provide improved methods for visualising and analysing complex high content data to support drug safety assessment.
The experience of some of the partners (e.g. EMBL-EBI, IMIM, UM, SIMCYP) in developing recognized user-friendly software in the field is a guarantee of the accomplishment of this objective. For example, partner UM has extensive experience with the use of the ConsensusPathDB interaction database. In particular, we will implement visualization tools for networks based on open source software (e.g. Cytoscape and R libraries). These tools allow integration of different types of data onto networks and we can leverage a variety of network analysis tools available in the community.
Objective 5: Help inform regulatory decisions by providing evidence supporting the usefulness of QST modelling to support safety risk assessment.
We will test the human relevance of the network associations and models that are developed through the use of normal and drug-perturbed clinical samples and correlation with translatable biomarker data. It is essential that there is a common understanding and endorsement of the translational models that will be developed, and the purpose they are fit for, by academics, industry and drug regulators. The MRC Centre for Drug Safety Science (ULIV) has a well-established format of workshops and subsequent peer reviewed authoritative reviews to inform policy that can be enacted.