Exploring genetic influences on adverse outcome pathways using heuristic simulation and graph data science

New approaches, like the Adverse Outcome Pathway (AOP) framework, have been developed to describe how chemicals cause toxicity by linking in vitro assays to adverse health outcomes. However, approaches, tools and resources for development of AOPs have not been well described. Here we review information resources for AOP development and define a streamlined process for linking a chemical to an existing AOP. We propose a four step process to facilitate AOP development: link the uncharacterized chemical directly to Molecular Initiating Events, Key Events, or Adverse Outcomes; identify analogs with toxicological information for the uncharacterized chemical; link the characterized chemical (initial chemical if characterized, a characterized analog if initial chemical is not) to Molecular Initiating Events, Key Events, or Adverse Outcomes; and identify AOPs that contain the Molecular Initiating Events, Key Events, or Adverse Outcomes that were found in Steps 1 and 3. The process and library of informational resources proposed and tested here served as the foundation for an informational online tool (AOPERA) that helps practitioners identify their current-state knowledge gaps, navigate the four-step process, and connect to relevant resources. AOPERA can be found at https://igbb.github.io/AOPERA_HTML. Additionally, we anticipate that by simplifying and standardizing the process of linking a chemical to a known AOP, we will lower the barrier to entry for this objective and increase its accessibility to new practitioners. In turn, this may increase the demand for new or improved AOPs to which practitioners can link chemicals, thereby contributing to the expansion of the library of known AOPs.

reproductive toxicity, voltage-gated sodium channel-mediated neural toxicity, hemolytic anemia induced by anilines, nephrotoxicity induced by 4-aminophenols (OECD 2011), skin sensitization (OECD 2012b), phototoxicity, aryl hydrocarbon receptor-mediated toxicity (Ankley et al. 2010), drug-induced liver fibrosis, and steatosis (Landesmann et al. 2012) as well as drug-induced cholestasis (Vinken et al. 2013). The advent of AOPs in toxicology has been acknowledged and is being brought forward by the OECD, which issued a guidance document and reporting template for the development of newly postulated AOPs and the assessment of its completeness (OECD 2012a). According to these OECD guidelines, AOPs rely on 2 anchors, namely the molecular initiating event and the adverse outcome, that are interconnected by a series of causally linked intermediate steps, of which some are to be considered as key events. Being designed to shed light onto the mechanisms that underlie pertinent toxicological effects, AOPs are typically represented as stand-alone processes that consist of a set of consecutive events (OECD 2012a). This portrayal of the toxicity landscape is not without criticism because of its inherent simplicity and reductionism. Indeed, toxicological processes in real life encompass entangled networks of serial, parallel, and intersecting pathways that are often difficult to decouple. The identification of the different building blocks of AOPs is mainly based upon a survey of the scientific literature to mine out relevant data, such as in chemico, in vitro, in vivo, and chemical structure information on reference chemicals associated with particular AOPs. Importantly, AOPs must be considered as open and flexible structures that should be continuously updated by feeding in established and newly generated data. Such iterative refinement exercises should ideally include the elaboration and quantification of the toxicodynamic relationships between neighboring events Toxicity testing typically focuses on generating data concerning the adverse effects of chemical substances on human health or the environment. However, there is an increasing desire to go beyond observation and prioritize the elucidation of the molecular machinery that drives the manifestation of toxicological apical end points (Andersen et al. 2012). This view was clearly embodied with the introduction of so-called toxicity pathways, which denote cellular response cascades that, when sufficiently perturbed, are expected to result in an adverse health outcome (NRC 2007; OECD 2012a). This designation has been criticized, as by definition it excludes true stress response pathways that only become activated by perturbation (Jennings 2013). A more comprehensive conceptual framework is based upon the adverse outcome pathway (AOP) approach. Having emerged initially in the field of ecotoxicology, an AOP refers to an analytical construct that portrays existing knowledge concerning the causal linkage between a direct molecular initiating event, i.e., the interaction of a chemical with a biological system, and an adverse outcome at a biological level of organization relevant to risk assessment (OECD 2012a). Over the last few years, the AOP tool has been proven equally applicable to describing toxicological processes underpinning human-relevant adverse effects, with examples including estrogen receptor-mediated

Fatty liver disease (FLD) affects over 25% of the global population and may lead to liver-related mortality due to cirrhosis and liver cancer. FLD caused by occupational and environmental chemical exposures is termed "toxicant-associated steatohepatitis" (TASH). The current review addresses the scientific progress made in the mechanistic understanding of TASH since its initial description in 2010. Recently discovered modes of actions for volatile organic compounds and persistent organic pollutants include the following: (i) the endocrine-, metabolism-, and signaling-disrupting chemical hypotheses; (ii) chemical-nutrient interactions and the "two-hit" hypothesis. These key hypotheses were then reviewed in the context of the steatosis adverse outcome pathway (AOP) proposed by the US Environmental Protection Agency. The conceptual understanding of the contribution of environmental exposures to FLD has progressed significantly. However, because this is a new research area, more studies including mechanistic human data are required to address current knowledge gaps.

Aquatic herbicides are commonly used to control a wide variety of algae and plants, but they also have the potential to contaminate and affect nontarget organisms. However, the impacts of low-level 2,4-dichlorophenoxyacetic acid (2,4-D) herbicide exposure on larval fish are not well understood. We conducted a series of experiments to determine the effects of low concentrations (0.05, 0.50, and 2.00 ppm) of 2 commercial 2,4-D amine salt herbicide formulations (Weedestroy® AM40 [WAM40] and DMA® 4 IVM [DMA4]) and pure 2,4-D on the development and survival of fathead minnows (Pimephales promelas) at various life cycle stages. Larval survival (30 d post hatch [dph]) was decreased following exposure of eggs and larvae to pure 2,4-D (0.50 ppm; p ≤ 0.001), as well as to WAM40 (0.50 and 2.00 ppm; p ≤ 0.001, p ≤ 0.001) and DMA4 (0.50 and 2.00 ppm; p ≤ 0.001, p ≤ 0.001). The results also narrowed the critical window of exposure for effects on survival to the period between fertilization and 14 dph. Development was not negatively altered by any of the compounds tested, although the commercial formulations increased larval total length and mass at 2.00 ppm. Altogether, the results indicate that the use of 2,4-D herbicides for weed control in aquatic ecosystems at current recommended concentrations (<2 ppm whole lake; <4 ppm spot treatment) could present risks to fathead minnow larval survival. Environ Toxicol Chem 2018;37:2550–2559. © 2018 SETAC

The Adverse Outcome Pathway (AOP) framework provides a template that facilitates understanding of complex biological systems and the pathways of toxicity that result in adverse outcomes (AOs). The AOP starts with an molecular initiating event (MIE) in which a chemical interacts with a biological target(s), followed by a sequential series of KEs, which are cellular, anatomical, and/or functional changes in biological processes, that ultimately result in an AO manifest in individual organisms and populations. It has been developed as a tool for a knowledge-based safety assessment that relies on understanding mechanisms of toxicity, rather than simply observing its adverse outcome. A large number of cellular and molecular processes are known to be crucial to proper development and function of the central (CNS) and peripheral nervous systems (PNS). However, there are relatively few examples of well-documented pathways that include causally linked MIEs and KEs that result in adverse outcomes in the CNS or PNS. As a first step in applying the AOP framework to adverse health outcomes associated with exposure to exogenous neurotoxic substances, the EU Reference Laboratory for Alternatives to Animal Testing (EURL ECVAM) organized a workshop (March 2013, Ispra, Italy) to identify potential AOPs relevant to neurotoxic and developmental neurotoxic outcomes. Although the AOPs outlined during the workshop are not fully described, they could serve as a basis for further, more detailed AOP development and evaluation that could be useful to support human health risk assessment in a variety of ways.

A mechanistic framework for integrating chemical structure and high-throughput screening results to improve toxicity predictions

Semantic characterization of adverse outcome pathways

Basic toxicological information is lacking for the majority of industrial chemicals. In addition to increasing empirical toxicity data through additional testing, prospective computational approaches to drug development aim to serve as a rational basis for the design of chemicals with reduced toxicity. Recent work has resulted in the derivation of a "rule of 2," wherein chemicals with an octanol-water partition coefficient (log P) less than 2 and a difference between the lowest unoccupied molecular orbital and the highest occupied molecular orbital (ΔE) greater than 9 (log P<2 and ΔE >9 eV) are predicted to be 4 to 5 times less likely to elicit acute or chronic toxicity to model aquatic organisms. The present study examines potential reduction of aquatic toxicity hazards from industrial chemicals if these 2 molecular design guidelines were employed. Probabilistic hazard assessment approaches were used to model the likelihood of encountering industrial chemicals exceeding toxicological categories of concern both with and without the rule of 2. Modeling predicted that utilization of these molecular design guidelines for log P and ΔE would appreciably decrease the number of chemicals that would be designated to be of "high" and "very high" concern for acute and chronic toxicity to standard model aquatic organisms and end points as defined by the US Environmental Protection Agency. For example, 14.5% of chemicals were categorized as having high and very high acute toxicity to the fathead minnow model, whereas only 3.3% of chemicals conforming to the design guidelines were predicted to be in these categories. Considerations of specific chemical classes (e.g., aldehydes), chemical attributes (e.g., ionization), and adverse outcome pathways in representative species (e.g., receptor-mediated responses) could be used to derive future property guidelines for broader classes of contaminants.

Phenones and hydroxy benzophenones are widely used as UV radiation filters, and in the manufacturing of insecticides and pharmaceuticals. Understanding the estrogenic potential these chemicals is of interest to the US Environmental Protection Agency and other international environmental organizations. The current study sequentially combined complementary in vitro rainbow trout estrogen receptor (rtER) binding and liver slice vitellogenin (Vtg) mRNA induction assays in the context of a defined ER-mediated adverse outcome pathway (AOP). Cyclic phenones, branched phenones, and hydroxybenzophenones bound to rtER with relative potency ranging from no affinity to high binding affinity of 0.11%, and many induced Vtg gene expression in rt liver slices. In addition, cyclohexylphenylketone which did not bind rtER binding in cytosol was biotransformed within liver tissue to a chemical that induced Vtg expression.

A systematic review and BMD modeling approach to develop an AOP for humidifier disinfectant-induced pulmonary fibrosis and cell death

The following article represents a mini-review of an intensive 10-year progression of genome-to-phenome (G2P) discovery guided by the adverse outcome pathway (AOP) concept. This example is presented as a means to stimulate crossover of this toxicological concept to enhance G2P discovery within the broader biological sciences community. The case study demonstrates the benefits of the AOP approach for establishing causal linkages across multiple levels of biological organization ultimately linking molecular initiation (often at the genomic scale) to organism-level phenotypes of interest. The case study summarizes a US military effort to identify the mechanism(s) underlying toxicological phenotypes of lethargy and weight loss in response to nitroaromatic munitions exposures, such as 2,4,6-trinitrotoluene. Initial key discoveries are described including the toxicogenomic results that nitrotoluene exposures inhibited expression within the peroxisome proliferator activated receptor α (PPARα) pathway. We channeled the AOP concept to test the hypothesis that inhibition of PPARα signaling in nitrotoluene exposures impacted lipid metabolic processes, thus affecting systemic energy budgets, ultimately resulting in body weight loss. Results from a series of transcriptomic, proteomic, lipidomic, in vitro PPARα nuclear signaling, and PPARα knock-out investigations ultimately supported various facets of this hypothesis. Given these results, we next proceeded to develop a formalized AOP description of PPARα antagonism leading to body weight loss. This AOP was refined through intensive literature review and polished through multiple rounds of peer-review leading to final international acceptance as an Organisation for Economic Cooperation and Development-approved AOP. Briefly, that AOP identifies PPARα antagonist binding as the molecular initiating event (MIE) leading to a series of key events including inhibition of nuclear transactivation for genes controlling lipid metabolism and ketogenesis, inhibition of fatty acid beta-oxidation and ketogenesis dynamics, negative energy budget, and ultimately the adverse outcome (AO) of body-weight loss. Given that the PPARα antagonism MIE represented a reliable indicator of AO progression within the pathway, a phylogenetic analysis was conducted which indicated that PPARα amino acid relatedness generally tracked species relatedness. Additionally, PPARα amino acid relatedness analysis using the Sequence Alignment to Predict Across Species Susceptibility predicted susceptibility to the MIE across vertebrates providing context for AOP extrapolation across species. Overall, we hope this illustrative example of how the AOP concept has benefited toxicology sows a seed within the broader biological sciences community to repurpose the concept to facilitate enhanced G2P discovery in biology.

Adverse outcome pathway (AOP) has become a key component in next-generation risk assessment (NGRA), as AOPs provide mechanistic and causal basis for linking toxicological responses across multiple levels of biological organization, thus allowing utilization of new approach methodologies (NAMs) to support prioritization, classification, and hazard assessment of chemicals. Quantitative AOPs (qAOPs) further allow estimation of points of departure (PODs) along a cascade of toxicological events and prediction of final adverse outcome(s) based on early key events. The present study developed a zebrafish-based qAOP model for an OECD-endorsed AOP linking aromatase inhibition to reproductive dysfunction (AOPWiki, AOP no. 25). The model was built based on newly generated laboratory data from a prototypical aromatase inhibitor propiconazole (PRO) and further challenged by another chemical tebuconazole (TEB) to evaluate the model's performance and applicability. We found that female and male zebrafish have different effects to establish AOP after exposure to PRO and qAOP was expanded and applied to cross-species prediction with fathead minnow. During the environmentally relevant concentrations of PRO and TEB for 21 days, aromatase CYP19, vitellogenin, and steroidogenesis were affected significantly, and the reproduction of zebrafish significantly decreased. The present study aimed to expand the mechanism of aromatase and assembled quantitative AOP in zebrafish and further linked with in vitro, which extended aromatase inhibitor-mediated reproductive dysfunction as a demonstrative case of published qAOP model in fathead minnow Pimephales promelas (AOPWiki, AOP no. 25).

The adverse outcome pathway concept: A pragmatic tool in toxicology

Linking AhR and Nrf2 activation to neurotoxicity through adverse outcome pathway-based machine learning.

Cosmetic products are used daily on a global scale. Therefore, it is necessary to ensure that these products, and their ingredients, do not cause any adverse human health effects under normal usage; to ensure this, risk assessment must be performed. Traditionally, risk assessments are performed in vivo, i.e. conducting tests on animals using the chemical(s) of interest. However, over the past decade there has been an increase in research into the use of alternative toxicity testing methods, such as in vitro, in chemico and in silico. Whilst there are a number of alternative techniques that may be employed, no one method can be used in isolation as a full replacement for an in vivo test. Therefore, the Adverse Outcome Pathway (AOP) concept is an emerging method by which information provided by the in vitro, in chemico, and in silico approaches can be utilised in an integrated testing strategy. The AOP concept links an upstream molecular initiating event to a downstream adverse outcome, via a number of testable key events.In silico approaches utilise computers in order to develop predictive models. Within the AOP paradigm in silico method work to identify the key features of a chemical (structural alerts) that induce a molecular initiating event (MIE). A collection of structural alerts that induce the same MIE are considered to be an in silico profiler. Typically, these in silico profilers are supported by associated toxicity, or mechanistic, information pertaining to the ability to induce a specific MIE. The overall aim of the work presented in this thesis was the development of an in silico profiler, based upon the hypothesis that the induction of mitochondrial toxicity is a key driver of organ-level toxicity. The research presented herein demonstrates the ability to identify, and develop, two types of structural alert; mechanism- and chemistry-based; that pertain to mitochondrial toxicity. Due to the differences inherent in these two types of alert they should be utilised for different purposes. As such, the main usage of the mechanism-based alerts should be in the formation of chemical categories and subsequent data gap filling via read-across. In comparison, the chemistry-based alerts should be utilised for the purposes of prioritising chemicals, within an inventory, that should undergo additional testing in in vitro and/or in chemico assays. It is envisaged that these two types of structural alerts could be used to profile chemical inventories as part of a tiered testing strategy.Therefore, the future work discussed in detail the need to expand the chemical space covered by the alerts. Additional future work involves utilising experimental information from in vitro/in chemico assays to verify the mechanism-based alerts and to refine the chemistry-based alerts by discerning mechanistic information associated with them. Furthermore, it is envisaged that these alerts could be incorporated into predictive tools, such as the OECD QSAR Toolbox, to enable their use for screening and prioritisation purposes.