In Response: Adverse outcome pathways--An industry perspective.

Abstract

The recent description of the adverse outcome pathway (AOP) approach has brought together 2 concepts that individually have a long history in chemical safety assessment. The use of biomarkers relating to mechanisms of action has long helped our mechanistic understanding of (eco)toxicological effects. In toxicology, such knowledge is crucial to determine the human relevance of mechanisms occurring in rodents or other surrogate test species. Biomarkers are routinely used in retrospective (environmental monitoring) and prospective hazard assessments. However, they have been limited by their simplicity and (perceived) lack of predictive power for whole-organism adverse effects that are the “currency” of hazard and risk assessments. The assessment of whole-organism effects is to a large extent in the form of endpoints proposed by the Organisation for Economic Co-operation and Development and other toxicity testing guidelines. The AOP concept promises to bring together mechanisms of action and toxicity outcomes by connecting and utilizing increasingly available mechanistic information and computational capacity to integrate markers at different levels of biological organization to conclude the “likely” adverse outcome (AO). This would be a major advancement, as it could then directly link to the needs of safety assessment—that is, predicting hazard-relevant endpoints. Adverse outcome pathways present a great opportunity to leverage these tools and to fully explore the possibilities. Like other stakeholders, industry has great interest in the potential benefits of an operational AOP approach. It could provide achievable methods to address the assessment of the ever-growing universe of chemicals covered by regulation globally. It addresses the need for the assessment of a greater number of substances at a scale for which it would be impractical to test from both capacity and resource standpoints. Adverse outcome pathways may allow for a sound science-based prioritization of substances and testing. In this regard it promises considerable benefit in the likely reduction of the number of in vivo tests and, thus, in the reduction of vertebrate animal testing. However, the scale of the challenge should not be underestimated. Not least in the challenge of commanding the new technologies and managing the large amounts of data generated but also in our understanding of the linkage to adverse effects. New approaches will be required to address extrapolation from in silico and in vitro systems to individual organisms and eventually to population-level responses—essentially the environmental protection goal. Of course, this needs to be operationalized not only within a scientific context but also in a workable regulatory framework in which transparent, predictable, and rapid decisions can be made (often by nonexperts) to allow for effective environmental protection. Therefore, it is likely that extrapolations will be perceived as adding uncertainty to assessments. However, for the approach to be workable in a regulatory setting, compounded conservatism (or “creeping safety”) 1 should be avoided. Otherwise, the benefits of the approach may be outweighed by an overly protective assessment, for example, multiplication of uncertainty factors, pushing registrants to conduct tests instead. One of the key challenges will inevitably be to determine the extent to which an AOP needs to be described. How much information is required for confidence in the “anchors” and key events underlying the pathway to support read-across between chemicals? Understandably, the focus to date has been on the AOP's left-hand side concerning mechanistic toxicity information. However, this question applies equally to the AOP's right-hand side in understanding what the individual- or population-level AO might be. Here, we must acknowledge that the effects we are able to measure (typically growth, development, and reproduction) are often surrogates for what we believe to be predictive of population-level consequences. However, these surrogates depend on the properties of the test system in which they are measured. Therefore, to make appropriate predictions and regulate individual substances equitably, we must understand what we are predicting and how this relates to the protection goal. Similarly, once an AOP is considered sufficiently robust, there needs to be a clear agreement on which tests and endpoints need to be performed for the “new” substance to confirm the appropriateness of the read-across. Hence, AOPs should find a compromise between simplification and completeness. Achieving this goal will be further complicated by the role of interacting toxicity pathways, which could vary among even chemically similar substances. Other important cases are those where different pathways may lead to the same perturbation of common key events (effect). For example, a case study for AOP development described impaired vitellogenesis (Vtg) in fish, leading to reproductive dysfunction 2. One of the potential mechanisms is via aromatase inhibition for which an AOP has been developed, based on data from the pharmaceutical azole, fadrozole 3. However, reduced Vtg in females could also be affected by mechanisms other than aromatase inhibition. For instance, some azole compounds are known liver toxicants 4, 5. Thus, reduced Vtg may also be caused by liver damage as the liver is the site of Vtg synthesis 6. Therefore, it is conceivable that, for some substances, liver effects could be the lead toxicity, even if aromatase inhibition occurs at higher dose levels. Hence, it should be reemphasized that the AOP development process should explicitly include an assessment for alternative mechanisms of action within a weight-of-evidence evaluation to avoid drawing the wrong conclusions on the underlying mechanism. Such evaluation typically includes comparison of relative potencies to initiate and propagate different pathways in an effort to discern what the “key” AOPs mediating the toxicity in question are likely to be and which can be effectively ignored because they would only be elicited at much greater concentrations. This may be especially important where the resulting AOP is used for hazard assessment. As hazards or intrinsic properties are increasingly used to regulate substances (e.g., endocrine disruption in the European Union), misassigning or falsely reading across a property may lead to severe regulatory action. As in the previous example, liver toxicity would be acceptable while aromatase inhibition might lead to a cutoff, even though the AO is identical. Thus, we emphasize the crucial importance of defining the minimal data requirements to characterize a pathway and consider alternative mechanisms before read-across to similar substances is concluded. Despite the factors discussed above, in the short term it is likely that the AOP concept will be most useful for hazard assessment. It is a particularly good fit for identifying potential endocrine disrupters as required by many chemical substance regulations. The widely accepted definitions of endocrine disruption require evidence that an adverse effect is consequent to changes in endocrine function—that is, both anchors of the AOP. Thereby, the concept may form a robust framework in which to organize such information to inform on endocrine potential and support regulatory decision making. Further, the read-across potential can be used to construct prioritized lists of substances for testing and even direct the most informative assay selection. Indeed, such a process is under way as part of the US Environmental Protection Agency's EDSP21 program 7. However, currently AOPs are well defined only for a few endocrine pathways. Use of the AOP concept for quantitative risk assessment will undoubtedly be a greater technical and regulatory challenge than implementation for hazard assessment. Predicting the dose or concentration at which an effect will be observed requires an understanding of the internal dose and explicitly that at the target of the molecular initiating event(s). Exposure alone is insufficient for read-across because various feedback mechanisms and compensatory responses may prevent the manifestation of the AO. An integration of exposure (in real life often highly variable), toxicokinetic aspects (absorption, distribution, metabolism, and excretion), and thresholds for toxicity will be required. Progress is being made along these lines (e.g., Stadnicka-Michalak et al. 8), but considerable work will be required to develop suitable models. These will require sufficient understanding of the most appropriate dose metrics and how they relate to variable exposures in the real world. Low exposure levels may progress through the first few steps in the AOP, but because of feedback and compensation would not lead to an AO (toxicity). Clearly, only those exposure levels that are sufficient to trigger the molecular initiating event and lead to an AO would be relevant for risk assessment. Here, taxonomic and species differences may play an important role as different species will likely differ in the similarity of receptors, metabolic capacity, sensitivity and compensatory feedback mechanisms, and so forth. The final challenge would be to integrate these individual-level effects to potential population-level consequences 9, aligned to the risk-assessment protection goals. In summary, industry is closely following the field of AOP development and sees areas of potential use (e.g., demonstrating human relevance, endocrine disruption, prioritization, and guiding testing strategies). However, quality criteria and minimal data requirements for AOPs need to be set along with appropriate exposure assessments (external and at the target site); all of these need to be integrated to prevent erroneous read-across between chemicals and species. James R Wheeler Dow AgroSciences Abingdon, United Kingdom Lennart Weltje BASF SE Limburgerhof, Germany