A Symposium Session, Adverse Outcome Pathways as an Integrative Framework for Predictive Toxicology: Combining Top-Down with Bottom-Up Thinking, was held on Monday, March 24, 2014 at the SOT 53rd Annual Meeting. A summary of this session is below.
Public health protection is the goal of regulatory safety or risk assessment, and entails answering this question: “What are the human health hazards that may result from exposure of X duration and Y concentration to agent Z?” Mechanistic data can guide this process at many levels, and with the explosion of high throughput screening (HTS) and system-wide association studies informing mechanistic considerations, the organization and analysis of these diverse datastreams has become a common focus of ongoing discussion and was the subject of this symposia.
Chris Corton, with the US Environmental Protection Agency, started the conversation by introducing the Adverse Outcome Pathway (AOP) as an analytical process, designed to aid in the organization and evaluation of the relationships among various molecular events and adverse effects of concern. Encompassing the biochemical and cellular endpoints also considered in Toxicity Pathway or Mode of Action analyses, the AOP framework guides the user to anchor Adverse Outcomes (AO) observed in individuals or populations to specific Molecular Initiating Events (MIE). The goal is to then link each MIE to a series of key events organized by increasing levels of biological complexity (e.g., from biochemical to organelle, then cellular, tissue, organ, system, individual, and population level effect), eventually culminating in a toxic response, or AO. As with key steps in a mode of action analysis, each event or key step is necessary but not sufficient to the eventual toxic response under consideration, and the interaction space between key steps is described as the nature of their molecular relationship, which could be described using weight of evidence (WOE) criteria. Dr. Corton later noted that genomic signatures could be used to both predict key events as well as support these WOE evaluations. Furthermore, while genomic-signatures could identify MIEs from in vitro systems, identification of downstream key events could only be currently achieved in vivo.
Maurice Whelan, with the European Union Reference Laboratory for Alternatives to Animal Testing, elaborated further on the relationships between individual MIEs and key steps, noting that a single AO could result from several MIEs, involving both common and unique key intermediates, and that each MIE could subsequently lead to a variety of AOs. Unlike other approaches, the AOP framework does not restrict users to generally linear sequences of key events and allows for as much interaction and complexity as required by our understanding of the pertinent physiology, and as supported by the available database. At its core, the AOP framework is a failure analysis system—a description of how and when a biological system can turn pathological, and the detailed, sequential conditions required for failure. Melvin Andersen, from The Hamner Institutes for Health Sciences, proposed that the key utility of AOPs would be in facilitating toxicological prediction, regardless of whether the AOP was constructed backwards from the AO to a postulated MIE (i.e., the “top-down” approach), or assembled as a progression from MIE through key events to a plausible health hazard (“bottom-up”).
Further analysis of relevant AOPs could guide development of designed-for-purpose cell-based assays. Dr. Whelan noted that the benefits of this AOP analytical approach, and the increased rational incorporation of ‘omics and other HTS datasets, could decrease (although likely not ablate) current requirements for animal toxicology testing, as well as speed up the chemical hazard evaluation process, by creating a system of agreed-upon AOPs that could be applied to a variety of chemical exposure scenarios with similar AOs. Current work in this field is being collected into an AOP guidebook and wiki-style knowledgebase, with an Effectopedia generated from the “scientific crowd-sourcing” of toxicological knowledge from interested participants. It is hoped that this will become a resource that serves as a “practical solution to the integration, curation, and dissemination of toxicological knowledge.” In addition to assembling data, other presenters discussed approaches to mining the overwhelming amount of data already available in curated chemical and biological effects databases. David Wild of Indiana University has been working on the development of semantic database assembly and querying tools, which view complex data as networks with “node-edge-node” relationships, and can identify drug and adverse effect associations based upon a probabilistic analysis of individual node linkages.
During the lively discussion that followed the individual presentations, participants noted the potential for miscommunication regarding what constituted an “adverse outcome” versus key events within an AOP. There was caution expressed that failure to establish an AOP in particular exposure circumstances should not be misconstrued as evidence for lack of potential hazard to human health.