In Response: The path forward for the adverse outcome pathway framework--A regulatory perspective.

Abstract

The adverse outcome pathway (AOP) framework, at its core, is a way of organizing biological and toxicological knowledge in a manner that facilitates inference and extrapolation. Much of the interest in AOPs is driven by the need to enhance the use of mechanistic or pathway-based data and predictive approaches as a basis for regulatory decision making 1-4. Although AOP description was initially rather ad hoc, the international community, through coordination with the Organisation for Economic Co-operation and Development's (OECD's) Extended Advisory Group on Molecular Screening and Toxicogenomics, has transformed adverse outcome pathway description into a formalized yet flexible process, employing standardized templates, terminology, and conventions (3-5. Both the efficiency of AOP development and transparency in AOP application have been advanced through development of a common internationally harmonized and publicly accessible knowledge base as a central repository for AOP descriptions 6. Together, these developments have fostered more consistent, systematic, and transparent AOP description, which facilitates regulatory application. Concepts of weight-of-evidence assembly and evaluation employing Bradford Hill considerations 7 have been incorporated as a fundamental aspect of AOP description. This is critical in that the amount of support for the relationships depicted in an AOP and the precision and confidence with which one can extrapolate from one element in the AOP to another ultimately dictate the type(s) of regulatory decision making it can reasonably be applied to. The development of AOPs and their representation in the AOP knowledge base employ a modular structure in which the basic elements of an AOP description, key events and key event relationships, are assembled into AOPs that describe a single trajectory of biological failure connecting a molecular initiating event to an adverse outcome (AO) considered relevant to regulatory decision making 3. Because key events and key event relationships can be shared by multiple AOPs, construction of simple AOP descriptions in the AOP knowledge base leads to de facto creation of networks of AOPs that capture the intrinsic complexity of potential biological and toxicological interactions that can modulate toxicological outcomes. Thus, the framework employs an elegant structure in which more sophisticated and complex understanding can be assembled from simpler units. Although the AOP framework has evolved considerably, further development is needed for it to play a central role in the 21st-century practice of regulatory toxicology. Advances are needed in 4 critical areas. First, the AOP knowledge base needs to be populated with AOP descriptions. As of January 2015, the AOP knowledge base contained 8 AOP descriptions that had been formally developed according to OECD guidance. Other AOP descriptions (n ≈ 40) had been initiated but were in early stages of development that did not include a transparent presentation of supporting evidence. The initial contents represent a mere fraction of the universe of toxicologically relevant pathways. There are many plausible AOPs which could be added to the AOP knowledge base and supported with existing knowledge. In an effort to address the need for improved coverage in the AOP knowledge base and to enhance the utility of data being generated through pioneering high-throughput screening programs (e.g., ToxCast, Tox21 8), the US Environmental Protection Agency's Chemical Safety for Sustainability program has initiated an effort to develop putative AOPs related to more than 300 unique molecular targets. Additional AOP development efforts are under way under the auspices of the OECD AOP development program 9. Because these AOPs are being developed in the publicly available AOP knowledge base 6, the ongoing AOP development efforts will be transparent. This serves both to minimize redundant efforts and to provide an open and collaborative environment in which a wide range of scientists can contribute their expertise. Consequently, there are many opportunities for the scientific community to engage and address the need for more AOP descriptions. Population of the AOP knowledge base with a robust set of AOPs and properly linking their descriptions through the use of shared key events and key event relationships is a critical prerequisite for a second important advance in the application of the AOP framework—development of approaches for extracting and analyzing AOP networks. Relevant human and ecological exposures almost universally involve exposure to multiple chemicals. Furthermore, many, if not the majority of, chemicals in the environment have the potential to interact with and perturb multiple biological targets. Consequently, to reasonably predict consequences of real-world exposures, it will be critical to contextualize how interactions among multiple AOPs within complex spatial and temporal domains of an organism will influence their cumulative outcome(s). The challenge of mixture toxicology is long-standing and well recognized. Although AOP networks alone do not provide all the tools needed to address that challenge, they do provide a mode of action–based road map that can aid evaluation of whether cumulative effects of chemicals may be independent, additive, reinforcing, or counteractive. Even if only through qualitative evaluation, consideration of points of interaction/intersection among AOPs, represented as shared key events in an AOP network 3, 4, facilitate hypothesis-driven assessments of potential mixture effects. Such approaches are needed in an environment where comprehensive testing of all individual chemicals, let alone all relevant mixtures, is unrealistic. Third, there is a need to develop computational frameworks that can take chemical-specific property information and/or biological effects measurements and translate them into a predicted probability or severity of an AO (generally for a specific exposure scenario). This has been termed the development of quantitative AOPs. A critical part of this process is developing a quantitative understanding of “points of departure” that define the magnitude and/or duration of change in an upstream key event needed to elicit a state change in a downstream key event. It involves understanding the adaptive mechanisms through which organisms respond to and compensate for different types of stressors and what their limitations are. This requires different kinds of toxicology research from those commonly employed in the past. In many cases, more intensive characterization of dose–response time-course surfaces across a range of biological scales and endpoints will be required to develop this quantitative understanding (e.g., Ankley and Villeneuve 10). Although such experiments may be technically challenging and resource-intensive, the ultimate goal is to unveil generalizable quantitative relationships that can be applied across a broad diversity of chemical, toxicological, and taxonomic space. Effective analysis of AOP networks may help reveal critical convergent nodes for which quantitative characterization of state-change conditions would provide the most broadly applicable improvements in AOP–based quantitative predictions. A final challenge to the AOP framework is that of regulatory acceptance. For the AOP framework to be broadly applied in regulatory (eco)toxicology, it is critical to establish predictive utility by developing AOP–derived hypotheses and testing them. An AOP is in essence a conceptual model based on biological understanding and experience. From it, a specific set of expectations regarding the response to a defined biological perturbation can be derived. Because the key events used to define an AOP are, by definition, measurable and causally related to one another, those expectations can be explicitly tested. In addition, by elucidating the cause(s) of significant deviations from expected results, AOPs can be refined toward greater predictive sophistication. Although we may never achieve perfect prediction of toxicological outcomes for all chemicals in all scenarios, the AOP framework is positioned to help us more effectively leverage new data, in the context of past experience and existing knowledge, to support regulatory decision making. The contents of this article neither constitute nor necessarily reflect official US Environmental Protection Agency policy. Daniel L. Villeneuve US Environmental Protection Agency Mid-Continent Ecology Division, Duluth, MN