Decoding Taxonomic Applicability in Adverse Outcome Pathways (AOPs): A FAIR Framework for Next-Generation Risk Assessment

Abstract

This article provides a comprehensive guide for researchers and drug development professionals on the critical role of taxonomic applicability within the Adverse Outcome Pathway (AOP) framework. It explores the foundational concepts of AOPs, detailing how defining the biological scope—the species, life stages, and sexes for which a pathway is relevant—is essential for building confidence in mechanistic toxicology and enabling regulatory use. The article outlines practical methodological approaches for defining and documenting applicability domains, drawing from recent case studies and the FAIR (Findable, Accessible, Interoperable, Reusable) principles roadmap. It addresses common challenges and optimization strategies for extrapolating AOP knowledge across taxa and discusses established frameworks for the weight-of-evidence validation and comparative assessment of AOPs. By synthesizing these elements, the article underscores how robust taxonomic applicability enhances the utility of AOPs in supporting New Approach Methodologies (NAMs) and transforming chemical safety and biomedical research.

What is AOP Taxonomic Applicability? Decoding the Core Concepts and Mechanistic Building Blocks

The Adverse Outcome Pathway (AOP) framework is a conceptual construct that organizes contemporary mechanistic toxicological knowledge into a structured sequence of causally linked events [1]. This framework begins with a Molecular Initiating Event (MIE), defined as the initial interaction between a chemical stressor and a specific biomolecular target within an organism [2]. This interaction triggers a sequential cascade of measurable Key Events (KEs) across increasing levels of biological organization—cellular, tissue, organ, and organismal—culminating in an Adverse Outcome (AO) relevant to risk assessment, such as impaired survival, development, or reproduction [1]. The primary utility of the AOP framework lies in its ability to translate data from high-throughput in vitro assays and in silico models into predictions of apical adverse effects, thereby addressing the critical need for efficient, mechanism-based safety assessments for the vast universe of untested chemicals [1].

This technical guide frames the AOP within the critical context of taxonomic applicability. An AOP developed in one species (e.g., a laboratory rat or a model fish) is often assumed to be relevant to others, but this extrapolation requires systematic evaluation [3]. Defining the Taxonomic Domain of Applicability (tDOA)—the range of species for which the pathway is biologically plausible—is therefore a foundational research question. The tDOA is determined by assessing the conservation of the essential biological components (e.g., proteins, receptors) and functional relationships that underpin each KE and Key Event Relationship (KER) across species [3]. This guide will detail the core principles of AOP construction, the methodologies for empirical and computational development, and the specific tools and approaches used to define and expand the tDOA, ensuring the robust application of AOPs in predictive toxicology and regulatory science.

Theoretical Foundation: Core Components and Constructs

An AOP is a simplified, linear representation of a toxicological pathway, deliberately constructed to facilitate knowledge organization and communication [1]. Its core architectural components are standardized and hierarchically organized.

The Hierarchical Chain of Events: MIE, KEs, and AO

The linear sequence of an AOP is its backbone, moving from a molecular perturbation to an organism- or population-level effect.

• Molecular Initiating Event (MIE): The MIE is the precise, initial point of chemical-biological interaction that triggers the pathway. A unified definition characterizes it as "the initial interaction between a molecule and a biomolecule or biosystem that can be causally linked to an outcome via a pathway" [2]. Examples include covalent binding to a specific protein, agonism/antagonism of a nuclear receptor, or inhibition of a critical enzyme.
• Key Events (KEs): KEs are measurable, essential changes in biological state that are both necessary and sufficient for the progression of the toxicity pathway. They bridge different levels of biological organization. An intermediate KE might be "increased oxidative stress in hepatocytes," while a later, organ-level KE could be "hepatic steatosis."
• Adverse Outcome (AO): The AO is the final, deleterious effect at a level of biological organization relevant to risk assessment (e.g., individual survival, growth, reproduction, or population sustainability). It must be clearly defined and of regulatory concern [1].

Key Event Relationships (KERs) and the Weight of Evidence

The connections between KEs are termed Key Event Relationships (KERs). A KER is not merely a temporal association but a causal linkage supported by empirical evidence and biological plausibility [1]. Establishing these causal links is critical and is formally evaluated using a Weight-of-Evidence (WoE) assessment. The Bradford Hill considerations (e.g., dose-response, temporal sequence, consistency) are adapted to evaluate the strength of evidence supporting each KER [4]. This WoE evaluation is what transforms a correlation-based pathway into a predictive AOP suitable for regulatory application.

Quantitative AOPs (qAOPs) and Networks

While the basic AOP is qualitative, the framework supports greater complexity. A Quantitative AOP (qAOP) incorporates mathematical models that describe the quantitative relationships between KEs (e.g., dose-response, time-course). This allows for prediction of the magnitude or probability of the AO based on the intensity of an earlier KE [1]. Furthermore, linear AOPs can intersect and merge to form AOP networks. These networks capture shared KEs (nodes) and pathway interactions, providing a more holistic view of how different stressors might converge on a common adverse outcome or modulate each other's effects [1].

Table 1: Core Components of an Adverse Outcome Pathway (AOP)

Component	Definition	Example	Evidence Required
Molecular Initiating Event (MIE)	The initial chemical-biological interaction that starts the pathway [2].	Covalent binding to skin protein (haptenization).	Direct molecular assay (e.g., peptide binding assay).
Key Event (KE)	A measurable change in biological state essential for pathway progression.	Dendritic cell activation and migration.	In vitro assay (e.g., CD86 expression in cell line).
Key Event Relationship (KER)	A scientifically supported causal link between two KEs [1].	Linking protein binding to dendritic cell activation.	WoE assessment using Bradford Hill criteria [4].
Adverse Outcome (AO)	The deleterious effect at the individual or population level.	Allergic contact dermatitis (skin sensitization).	Human or animal study confirming the clinical outcome.

Diagram 1: Linear structure of an AOP showing stressor, MIE, KEs, KERs, and AO.

Methodological Guide to AOP Development and Evaluation

Developing a scientifically credible AOP is a systematic process that integrates existing knowledge with targeted experimentation. The following workflow outlines the primary stages.

AOP Development Workflow

• AO Identification and Definition: The process begins by defining a specific AO of regulatory relevance (e.g., liver fibrosis, embryonic mortality). This clarifies the pathway's ultimate purpose.
• Backwards Pathway Elucidation: Using literature reviews and database mining, researchers work backwards from the AO to hypothesize the essential KEs and the MIE. Tools like the AOP-Wiki are invaluable for finding existing related KEs and avoiding duplication [1].
• KER and WoE Assessment: For each hypothesized linkage between KEs, a formal WoE assessment is conducted. This involves gathering empirical evidence (e.g., co-occurrence, dose-response alignment, temporal concordance) and assessing biological plausibility [4]. This step is critical for establishing causality.
• Experimental Confirmation and qAOP Modeling: Targeted experiments are designed to fill knowledge gaps, confirm essential KEs, and refine KERs. Data from these studies can be used to develop quantitative, predictive models (qAOPs) that define the mathematical relationships between events [1].
• Documentation and Peer Review: The finalized AOP is documented according to OECD templates and submitted for review to bodies like the OECD AOP Development Programme to ensure harmonization and scientific robustness [1].

Diagram 2: AOP development workflow from AO identification to peer review.

Experimental Protocols for Key AOP Components

Validating an AOP requires experiments tailored to its specific KEs. Below are generalized protocols for common assay types used in AOP development.

Table 2: Exemplary Experimental Protocols for AOP Development

AOP Component	Experimental Protocol	Key Measurements	Utility in AOP
MIE Confirmation	Direct Receptor Binding Assay: Incubate radiolabeled or fluorescent test chemical with purified target protein or cell membrane preparation. Use filtration or scintillation proximity to measure bound vs. free chemical.	Binding affinity (Kd), inhibition constant (Ki).	Provides direct evidence for the defined MIE [2].
Cellular KE	High-Content Screening (HCS): Expose relevant cell line to test chemical. Fix, stain for specific markers (e.g., phospho-proteins, oxidative stress, cell death). Use automated microscopy and image analysis for quantitation.	Fluorescence intensity, morphological changes, cell count.	Measures early, predictive KEs in a high-throughput format [1].
Organ/Tissue KE	Histopathology & Special Staining: Dose animals in vivo. At sacrifice, collect and fix target organ. Process, section, and stain with H&E or special stains (e.g., Masson's Trichrome for fibrosis). Blind scoring by pathologist.	Lesion incidence, severity grade, quantitative morphometry.	Confirms in vivo progression of pathway and links cellular KEs to tissue damage [4].
KER Support	Temporal/Dose-Response Study: Collect samples (e.g., blood, tissue) at multiple time points and doses post-exposure. Analyze biomarkers for upstream and downstream KEs (e.g., via ELISA, qPCR).	Correlation coefficients, time-lag, dose-response alignment.	Provides critical evidence for causality in WoE assessment [4].

Defining Taxonomic Applicability: The tDOA Challenge

A central thesis in AOP research is that pathways are not species-specific but are applicable across taxa where the underlying biology is conserved. Defining the Taxonomic Domain of Applicability (tDOA) systematically is therefore essential for credible extrapolation in ecological and human health risk assessment [3].

Principles of tDOA Determination

The tDOA is evaluated based on two pillars: structural conservation and functional conservation [3].

• Structural Conservation: The biological entities (e.g., genes, proteins, organelles) involved in each KE must be present and have conserved sequences or structures in the species of interest.
• Functional Conservation: These entities must perform the same biological function in the species of interest, and the causal relationships between KEs (KERs) must remain intact.

The empirical tDOA is limited to species explicitly tested in the studies supporting the AOP. The broader biologically plausible tDOA is inferred by extrapolating evidence of structural and functional conservation to untested species using computational and comparative biological tools [3].

Bioinformatics Workflow for tDOA Expansion

Bioinformatics tools are indispensable for efficiently evaluating structural conservation across the tree of life. The Sequence Alignment to Predict Across Species Susceptibility (SeqAPASS) tool is a prime example developed for this purpose [3].

SeqAPASS Analysis Protocol:

• Identify Query Proteins: For a given AOP, identify the specific protein targets involved in the MIE and critical KEs (e.g., a specific receptor, a metabolic enzyme, a transcription factor).
• Level 1 Analysis (Primary Sequence): Input the amino acid sequence of the query protein (from the source species). SeqAPASS performs BLAST searches against genomic/proteomic databases to identify putative orthologs in other species based on overall sequence similarity [3].
• Level 2 Analysis (Functional Domain): Evaluate conservation of known functional domains (e.g., ligand-binding domain, catalytic site) within the identified orthologs. Loss of a critical domain suggests loss of function.
• Level 3 Analysis (Critical Residues): Evaluate conservation of specific amino acid residues known to be essential for the protein's interaction with the chemical stressor or for its function (e.g., a key serine in an active site). Non-conservation at this level strongly suggests the species may not be susceptible via that MIE [3].

The results provide a line of evidence for the structural conservation of each KE's molecular basis across species, which can be combined with limited empirical data to define a scientifically justified tDOA.

Diagram 3: Workflow for defining tDOA using bioinformatic tools like SeqAPASS.

Application in Drug Development: From Hazard Identification to Risk Assessment

The AOP framework provides a powerful tool for improving the efficiency and mechanistic understanding of safety assessments in pharmaceutical development.

Use Cases in the Drug Development Pipeline

• Early Hazard Identification & Prioritization: Screening compound libraries against in vitro assays aligned with MIEs or early KEs (e.g., receptor binding, mitochondrial toxicity) can identify potential liabilities early, allowing for the prioritization of safer leads [1].
• Mechanism-Based Risk Assessment: When an adverse effect is observed in non-clinical studies, an AOP can be used to investigate its mechanism. Understanding whether the effect is on-target (related to the primary pharmacology) or off-target is critical for estimating human relevance and risk [1].
• Supporting Read-Across and Species Extrapolation: A well-defined AOP, with a clear tDOA, provides a scientific rationale for extrapolating toxicity findings from animal models to humans. It helps determine whether the biological pathway is conserved, informing the relevance of animal findings to human risk [3].
• Biomarker Qualification: KEs within an AOP represent candidate mechanistic biomarkers. The framework provides the biological context needed to qualify these biomarkers for use in non-clinical or clinical studies to monitor for specific pathway perturbations [1].

Table 3: Quantitative Benchmarks for AOP-Based Assessment

Application Area	Key Quantitative Metrics	Typical Benchmark for Confidence	Data Source
qAOP Modeling	Point-of-Departure (POD) ratio between KE and AO.	POD(KE) / POD(AO) < 10; suggests KE is a sensitive predictor.	Integrated in vitro to in vivo studies [1].
Assay Performance	Sensitivity, Specificity, Accuracy for predicting AO.	Balanced accuracy > 70-80% for use in screening.	Validation studies against reference chemicals [1].
tDOA Bioinformatics	Sequence identity/similarity for orthologs.	>60-70% identity often used as preliminary filter; Level 3 residue match is critical [3].	SeqAPASS or comparable BLAST analysis [3].
WoE Assessment	Number of supporting studies across species.	At least 2-3 independent studies in different models strengthens biological plausibility [4].	AOP-KB / AOP-Wiki repository [1].

The Scientist's Toolkit: Essential Research Reagents and Platforms

Developing and applying AOPs requires a combination of wet-lab, computational, and knowledge management tools.

Table 4: Essential Research Toolkit for AOP Development and Analysis

Tool Category	Specific Tool / Resource	Function in AOP Research	Key Provider / Source
Knowledge Management	AOP-Wiki (aopwiki.org)	Central repository for drafting, sharing, and browsing AOPs and KEs [1].	OECD
Bioinformatics	SeqAPASS Tool	Evaluates protein sequence conservation across species to inform tDOA [3].	US EPA
Bioinformatics	Comparative Toxicogenomics Database (CTD)	Identifies chemical-gene, gene-phenotype, and chemical-phenotype relationships to hypothesize KEs.	NCBI
In Vitro Assay Systems	Gene Reporter Assays (Luciferase, SEAP)	Measures transcriptional activation (e.g., nuclear receptor MIE) in high-throughput format.	Commercial vendors
In Vitro Assay Systems	Multiplex Cytokine/Apoptosis Assays	Measures multiple cellular KEs (e.g., inflammation, cell death) simultaneously.	Commercial vendors
Data Integration & Modeling	R/Bioconductor (qAOP packages)	Statistical analysis, dose-response modeling, and construction of quantitative network models.	Open Source
Chemical Tools	Pharmacological Agonists/Antagonists	Tool compounds to experimentally modulate specific KEs and test KER causality.	Commercial vendors
Reference Materials	OECD AOP Handbook & Templates	Provides standardized guidelines for AOP development and documentation [1].	OECD

The Adverse Outcome Pathway (AOP) framework provides a structured, mechanistic model for linking a molecular perturbation, through a series of measurable biological key events (KEs), to an adverse outcome (AO) of regulatory concern [5]. The utility of any AOP for decision-making, particularly within evolving paradigms like Integrated Approaches to Testing and Assessment (IATA), hinges on the confidence in its constituent relationships [6]. This confidence is formally evaluated through two interdependent pillars: Weight of Evidence (WoE) and Essentiality.

This technical guide explores these foundational concepts within the context of AOP taxonomic applicability. An AOP's "taxonomy" refers to the defined scope of its relevance across different species, life stages, sexes, and experimental models [5]. Determining this applicability is not a separate exercise but is directly dependent on the strength and nature of the evidence underpinning the pathway. A robust WoE assessment and rigorous demonstration of essentiality for each KE are prerequisites for reliably extrapolating an AOP beyond the specific conditions in which it was empirically observed. This document synthesizes current OECD-endorsed methodologies [5] and research perspectives [7] to provide a comprehensive manual for researchers and risk assessors building confidence in AOPs for targeted application.

The First Pillar: Weight of Evidence (WoE) Assessment

Weight of Evidence is a systematic approach to evaluating the collective body of evidence supporting the plausibility and causal linkages within an AOP. It moves beyond merely listing supporting studies to qualitatively and, where possible, quantitatively grading the strength of causal inference [7].

Methodological Framework for WoE Evaluation

The OECD AOP Developers' Handbook advocates for an expert-driven WoE assessment based on defined criteria [5]. The cornerstone of this evaluation is the adaptation of the Bradford Hill considerations for causal inference to the AOP framework, specifically applied to each Key Event Relationship (KER). The primary questions guiding the WoE for a KER are:

• Biological Plausibility: Is there a well-established biological basis supporting the causal linkage between the upstream and downstream KE?
• Essentiality: Is the upstream KE necessary for the downstream KE to occur? (This is explored in depth in Section 3).
• Empirical Support: What is the quantity, quality, and consistency of experimental data demonstrating that changes in the upstream KE lead to predictable changes in the downstream KE?

Table 1: Bradford Hill Considerations for AOP WoE Assessment [7] [5]

Consideration	Experimental Protocol & Data Type	Role in WoE
Strength & Consistency	Protocol: Multiple, independent in vitro and in vivo studies under varied conditions. Data: Dose-response and temporal concordance data for the paired KEs.	Establishes the reliability and reproducibility of the observed relationship.
Specificity	Protocol: Studies using selective modulators (agonists/antagonists, genetic knockouts) of the upstream KE. Data: Evidence that the downstream KE is altered only when the specific upstream KE is perturbed.	Supports a direct, non-coincidental linkage within the AOP network.
Temporality	Protocol: Time-course studies measuring both KEs in the same biological system. Data: Clear demonstration that the upstream KE precedes the downstream KE.	Validates the directionality of the proposed causal sequence.
Biological Gradient	Protocol: Dose-response experiments measuring the magnitude of both KEs. Data: A quantifiable, monotonic relationship between the magnitude of change in the upstream and downstream KE.	Enables quantitative prediction and supports a causal rather than threshold-based association.
Coherence & Plausibility	Protocol: Integration of data from omics, biochemical pathway analysis, and existing biological knowledge. Data: Evidence that the KER is consistent with the established understanding of biology.	Anchors the AOP in known biological theory, increasing its acceptability.
Experimental Analogy	Protocol: Comparative studies using prototypical stressors known to trigger the same KER. Data: Successful prediction of the downstream KE based on the upstream KE across different stressors.	Extends confidence to untested chemicals or stressors with similar properties.

Quantitative WoE Scoring: A Notional Approach

While WoE is often qualitative, quantitative frameworks enhance transparency and consistency. One approach involves scoring each Bradford Hill consideration for a given KER (e.g., on a scale of 0-3: None, Weak, Moderate, Strong) [7]. Scores can be aggregated, often with weighting for critical considerations like temporality and essentiality, to generate a quantitative confidence level for the KER and the overall AOP.

Diagram 1: The Weight of Evidence Assessment Framework (76 characters)

The Second Pillar: Establishing Essentiality

Essentiality is the property of a Key Event being a necessary component in the progression along the AOP [5]. It asks: If this specific KE is blocked or does not occur, will the subsequent downstream KEs and the Adverse Outcome also be prevented? Demonstrating essentiality is the most direct method for proving causal function beyond correlation.

Experimental Methodologies for Testing Essentiality

The OECD Handbook specifies that essentiality is best evaluated by examining the effects of preventing or modulating a KE on all downstream events [5]. The following table outlines core experimental strategies.

Table 2: Experimental Protocols for Demonstrating Key Event Essentiality [5]

Method	Detailed Experimental Protocol	Interpretation & Evidence Strength
Genetic Modulation (Knockout/Knockdown)	1. Select an in vivo or in vitro model system. 2. Using CRISPR/Cas9, RNAi, or other techniques, disrupt the gene responsible for the KE phenotype. 3. Expose the modified and wild-type models to the stressor. 4. Measure the targeted KE and all downstream KEs/AOs.	Strong evidence. If the downstream events are abolished or significantly attenuated in the modified model despite stressor exposure, it supports the KE as essential.
Pharmacological/ Biochemical Inhibition	1. Select a model system. 2. Administer a specific and potent chemical inhibitor that selectively blocks the activity or process defining the KE. 3. Co-expose the system to the inhibitor and the stressor. 4. Measure the targeted KE and all downstream KEs/AOs.	Moderate to Strong evidence. Strength depends on the inhibitor's specificity and potency. Successful blockade of the pathway supports essentiality.
Genetic/Experimental Disease Models	1. Utilize a model with a known mutation or condition that naturally impairs the biological process of the KE. 2. Expose this model and a healthy control to the stressor. 3. Measure the progression to downstream KEs/AOs.	Supportive evidence. Provides real-world analog of a blocked KE. Confounding factors in the disease model must be considered.
Dose-Response Concordance Analysis	1. Expose a model to a range of stressor doses/concentrations. 2. At multiple time points, quantitatively measure the magnitude of the upstream KE and a downstream KE/AO. 3. Perform statistical modeling (e.g., benchmark dose analysis) to compare response curves.	Supportive evidence. A high degree of concordance between the dose-response and temporal curves for the two events suggests a tight, possibly essential, linkage.

The Scientist's Toolkit: Reagents for Essentiality Testing

Table 3: Key Research Reagent Solutions for AOP Development

Reagent/Tool Category	Specific Examples	Function in AOP Research
Specific Chemical Inhibitors	Small molecule antagonists, enzyme inhibitors, blocking antibodies.	To pharmacologically inhibit a specific KE (e.g., receptor binding, enzyme activity) and test its essentiality for downstream effects [5].
Genetic Tools	CRISPR/Cas9 kits, siRNA/shRNA constructs, transgenic animal models.	To genetically knock out or knock down the expression of a target protein defining a KE, providing the strongest test of essentiality [5].
Validated Antibodies & Assay Kits	Phospho-specific antibodies, ELISA kits, activity-based probes.	To quantitatively measure the occurrence and magnitude of a KE (e.g., protein phosphorylation, cytokine release) in experimental systems [5].
'Omics Profiling Tools	RNA-Seq, targeted mass spectrometry panels, chromatin immunoprecipitation kits.	To provide unbiased data for identifying novel KEs, assessing biological plausibility, and establishing coherence within an AOP network [6].
Prototypical Stressors	Well-characterized reference chemicals (e.g., rotenone for mitochondrial inhibition).	To serve as positive controls to reliably trigger the AOP for method validation and comparative assessment of new substances [7].

Diagram 2: Logic Flow for Testing Key Event Essentiality (67 characters)

Synthesis: Integrating WoE and Essentiality for Taxonomic Applicability

The ultimate goal of rigorous WoE and essentiality assessments is to define the boundaries of an AOP's applicability. Confidence in the taxonomic scope is derived directly from the nature of the supporting evidence [5].

• High Confidence in Broad Applicability: An AOP supported by strong biological plausibility (conserved pathway across taxa), essentiality demonstrated in multiple models, and consistent empirical evidence across species can be applied more broadly, with careful consideration to taxonomic differences in kinetics and dynamics.
• Defined, Restricted Applicability: An AOP where essentiality is shown only in a specific rodent model, or where a critical KE is known to be species-specific, has a clearly constrained taxonomic scope. This is a scientifically valid and useful outcome for targeted safety assessment.

The process of AOP development is iterative. Gaps in essentiality evidence or weak WoE scores for specific KERs directly highlight research needs. Filling these gaps through targeted studies not only strengthens the AOP but also clarifies its taxonomic relevance, enabling more precise and efficient use in regulatory decision-making for chemical and nanomaterial safety [6].

Diagram 3: From AOP Confidence to Taxonomic Applicability (66 characters)

Theoretical Foundation: Taxonomic Applicability in the AOP Framework

The Adverse Outcome Pathway (AOP) framework organizes mechanistic knowledge into a sequence of causally linked Key Events (KEs), from a Molecular Initiating Event (MIE) to an Adverse Outcome (AO) relevant to risk assessment [8] [1]. A critical, yet often underexplored, component of an AOP is its Taxonomic Domain of Applicability (tDOA)—the range of species, life stages, and sexes for which the described pathway is biologically plausible and empirically supported [3] [5].

Defining the tDOA is not merely an academic exercise but a regulatory necessity. It determines the confidence with which data from tested surrogate species (e.g., lab rodents or model fish) can be extrapolated to protect untested species in the environment or to inform human health [3]. The tDOA is evaluated based on two pillars: structural conservation (the presence and similarity of the biological targets and mediators) and functional conservation (the preserved biological role and response of those elements across taxa) [3]. An AOP developed with data from a single species has a narrow empirical tDOA, but its biologically plausible tDOA may be much wider if evidence supports conservation [4].

Table 1: Foundational Concepts of Taxonomic Applicability in AOPs

Concept	Definition	Significance for Taxonomic Applicability
Taxonomic Domain of Applicability (tDOA)	The taxonomic group(s) (species, genera, families) for which an AOP, KE, or KER is valid [3].	Defines the boundaries for extrapolation and regulatory use of the AOP.
Empirical tDOA	The specific species for which experimental evidence for the KEs and KERs exists [3].	Represents the direct, observed support for the pathway. Often narrow.
Biologically Plausible tDOA	The broader taxonomic group for which the pathway is likely applicable, inferred from structural/functional conservation [3].	Enables hypothesis-driven extrapolation to untested species.
Structural Conservation	The preservation of the physical attributes of a biological entity (e.g., gene, protein, receptor) across taxa [3].	Assessed via bioinformatics (e.g., sequence, domain, and residue similarity).
Functional Conservation	The preservation of the biological role or activity of an entity across taxa [3].	Assessed through empirical in vitro or in vivo testing in new species.

The Multidimensional Scope of Taxonomic Applicability

Taxonomic applicability in AOPs is a multidimensional construct, extending beyond a simple list of species names.

Species and Higher Taxa: Defining the Domain

The primary dimension is the range of species to which an AOP applies. This is often framed at the level of higher taxonomic groups (e.g., "all vertebrates," "arthropods," "mammals"). The challenge lies in moving from the specific model organism used in development (e.g., Homo sapiens, Rattus norvegicus, Danio rerio) to defining these broader groups with confidence [5] [9]. This requires evaluating the evolutionary conservation of the essential biological components of the pathway [3].

Intraspecific Variability: Life Stage and Sex

AOPs frequently exhibit differential applicability within a species based on life stage and sex [5]. These factors can influence the presence, activity, or sensitivity of pathway components.

• Life Stage: Developmental stages may lack fully formed systems (e.g., a detoxification enzyme in early life stages) or exhibit unique susceptibilities (e.g., developmental neurotoxicity) [8] [5]. An AOP for thyroid hormone disruption leading to impaired neurodevelopment is inherently specific to early life stages [8].
• Sex: Sexual dimorphism in physiology, hormone regulation, and gene expression can render an AOP more applicable to one sex. For instance, pathways involving androgen or estrogen receptor signaling are intrinsically sex-specific, though they may have relevance for both sexes [8] [5].

Other Scope-Restricting Factors

The "scope" of applicability can be further restricted by ecological, morphological, or temporal factors analogous to those in taxonomic identification keys [10] [11]. For AOPs, this might include:

• Physiological State: Applicability only to reproducing individuals, migrating species, or organisms under specific metabolic stress.
• Health Status: Underlying disease or pre-existing conditions may be necessary for progression along the pathway.
• Temporal Dynamics: The pathway may only be triggered during specific seasons or times of day due to circadian biology.

Methodological Approaches for Assessment

Establishing taxonomic applicability requires a suite of complementary methods, from in silico predictions to empirical validation.

Bioinformatics for Assessing Structural Conservation

Bioinformatics tools provide efficient, scalable lines of evidence for structural conservation. The Sequence Alignment to Predict Across Species Susceptibility (SeqAPASS) tool is a prominent example [3]. SeqAPASS employs a tiered, hierarchical analysis:

• Level 1: Primary Sequence Similarity. Identifies potential orthologs of a query protein (e.g., a receptor that is the MIE) across species via global sequence alignment [3].
• Level 2: Functional Domain Conservation. Evaluates whether identified orthologs conserve known functional domains essential for the protein's activity in the AOP [3].
• Level 3: Critical Residue Conservation. Examines conservation of specific amino acid residues known to be critical for chemical binding, protein-protein interaction, or catalytic function [3].

Table 2: Hierarchical Analysis Levels of the SeqAPASS Tool for Evaluating Structural Conservation [3]

Analysis Level	Data Input & Method	Output & Interpretation	Utility for tDOA
Level 1	Query protein sequence; BLAST-based alignment against custom databases.	List of putative orthologs with percentage identity/similarity scores.	Identifies the broadest potential taxonomic range possessing a similar protein.
Level 2	Putative orthologs; mapping against Pfam/InterPro domain databases.	Assessment of whether essential functional domains are present/identical.	Narrows tDOA to species where the protein is likely functional.
Level 3	Protein structures or sequences; alignment focusing on known critical residues.	Evaluation of residue identity at sites crucial for the specific interaction in the AOP.	Provides high-confidence evidence for applicability to species where the molecular interaction is likely conserved.

Diagram 1: Hierarchical Bioinformatics Workflow for Assessing Structural Conservation (SeqAPASS Framework). This tiered approach sequentially filters species based on protein sequence, functional domain, and critical residue conservation to define a biologically plausible tDOA [3].

Integrating Toxicogenomics and Curation

Systematic annotation of KEs to specific genes and molecular pathways bridges AOPs with toxicogenomics data [9]. Curated knowledge bases link KE descriptions to:

• Gene Sets: Associated pathways (e.g., WikiPathways, Reactome), Gene Ontology terms, and phenotype ontologies.
• Biological Context: Cell types, tissues, and organs involved.

This curation allows researchers to interpret omics data from non-model species: if exposure in a new species alters genes belonging to a KE-annotated pathway, it provides supporting evidence for the functional activity of that AOP component in that species [9].

Empirical Validation of Functional Conservation

Computational predictions require empirical validation. Functional conservation is assessed through targeted in vitro or in vivo experiments.

• In Vitro Assays: Using cells or tissue fractions from a candidate species to test for the specific KE activity (e.g., receptor binding, enzyme inhibition, transcriptional activation) [1] [4].
• Short-Term In Vivo Tests: Exposing individuals of a candidate species to a stressor known to trigger the AOP and measuring intermediate KEs (e.g., biomarker expression, histological change) without needing to wait for the final AO [1] [4].

Table 3: Experimental Approaches for Assessing Functional Conservation of an AOP

Approach	Typical Methodology	Measured Endpoint (Example)	Information Gained for tDOA
In Vitro Bioassay	Expose cultured cells or subcellular fractions (e.g., microsomes) from target species to stressor.	Ligand binding affinity, enzyme inhibition potency (IC₅₀), reporter gene activation.	Confirms the functional activity of the MIE or an early KE in the species' specific biomatrix.
Toxicogenomics	Expose organisms (in vivo or in vitro) and conduct transcriptomic/proteomic analysis.	Differential expression of genes/proteins annotated to specific KEs [9].	Provides system-wide evidence for pathway perturbation in a new species.
Short-Term In Vivo Study	Limited-duration exposure of live organisms with focused tissue sampling.	Upstream/mid-pathway biomarker levels (e.g., plasma hormone, liver enzyme activity).	Demonstrates functional linkage between KEs in an intact organism of the candidate species.

Diagram 2: Integrated Workflow for Validating AOP Applicability to a New Species. Evidence flows from computational prediction to in vitro and targeted in vivo testing to confirm both structural and functional conservation [3] [4].

Practical Challenges and Current Limitations

Despite established methodologies, significant challenges remain in robustly defining taxonomic applicability.

• The "Species" Concept Itself: The definition of a species is not always unambiguous, with issues arising from taxonomic inflation, vandalism (poor-quality descriptions), and competing taxonomic lists [12] [13]. An AOP's tDOA described as "applicable to the genus X" becomes problematic if the constituent species of that genus are disputed.
• Data Gaps and Extrapolation Uncertainty: For most species, especially non-model organisms, empirical data for KEs are absent. Reliance on bioinformatics predictions alone carries uncertainty, as sequence conservation does not guarantee identical function in a different physiological context [3].
• Modeling Complex Scope Restrictions: Capturing nuanced applicability rules (e.g., "females in reproductive condition" or "larvae but not adults") in a machine-readable format is complex [11]. Current AOP-Wiki entries may lack this granularity.
• Integrating Evidence into the AOP-KB: While tools like SeqAPASS generate evidence, standardized formats and fields for integrating these computational tDOA assessments into the central AOP Knowledge Base (AOP-KB) are still evolving [3] [5].

Table 4: Key Research Reagent Solutions for Assessing Taxonomic Applicability

Resource / Tool	Type	Primary Function in Assessing Applicability
SeqAPASS Tool [3]	Bioinformatics Software	Provides tiered assessment of protein sequence, domain, and residue conservation across species to infer structural conservation for MIEs/KEs.
AOP-Wiki (aopwiki.org) [8] [5]	Knowledgebase	The central repository for AOPs, KEs, and KERs. Used to identify candidate AOPs and review existing empirical tDOA.
Unified Knowledge Space (UKS) / Curated Gene Sets [9]	Annotated Database	Links KEs to specific genes, pathways, and biological contexts. Enables interpretation of omics data from non-model species for functional evidence.
Ortholog-Specific Antibodies or PCR Primers	Wet-lab Reagent	Allows measurement of protein expression or gene transcription of a specific AOP target in tissues from a novel species.
Species-Specific In Vitro Assay Kits (e.g., luciferase reporter, EROD)	Wet-lab Assay	Measures functional activity of a conserved molecular target (e.g., receptor activation, enzyme inhibition) in cell lines or tissue samples from a candidate species.
Catalogue of Life / GBIF [10] [12]	Taxonomic Database	Provides reference taxonomic frameworks and occurrence data to clarify species identities and ranges, reducing nomenclatural confusion.

The field is moving towards more dynamic and quantitative definitions of taxonomic applicability. Future directions include:

• Quantitative AOPs (qAOPs) for Interspecies Extrapolation: Developing models that incorporate quantitative differences in KE sensitivity or dynamics across species to predict effective stressor doses [1].
• Integrated Testing Strategies (ITS): Combining high-throughput in vitro data from human or model systems with targeted in vitro or in vivo tests in species of concern to build evidence for tDOA [1] [4].
• Enhanced Curation and Governance: Adopting principles from efforts like the Library of Identification Resources (which uses FAIR principles for taxonomic keys) and the push for a global governed species list to bring stability and clarity to the taxonomic entities referenced in AOPs [10] [12].

In conclusion, unpacking "taxonomic applicability" is fundamental to the credible use of AOPs in chemical safety assessment. It requires a multidimensional understanding of scope—spanning species, life stages, and sex—and demands a weight-of-evidence approach that integrates bioinformatics, curated knowledge, and empirical testing. As methodologies and knowledge bases mature, the confident extrapolation of pathway-based toxicity knowledge across the tree of life will become a cornerstone of predictive toxicology and next-generation risk assessment.

The field of toxicological risk assessment is undergoing a fundamental transformation, shifting from a reliance on descriptive, apical endpoint data in whole animals towards a predictive science grounded in mechanistic understanding [14]. This paradigm, often termed Toxicology in the 21st Century, aims to address the critical challenge that the number of chemicals requiring evaluation far exceeds the capacity of traditional testing methods [14]. Central to this evolution is the Adverse Outcome Pathway (AOP) framework, which provides a structured, modular representation of the sequence of biological events leading from a molecular perturbation to an adverse effect relevant to risk assessment [8].

An AOP is defined as a series of linked events at different levels of biological organization that lead to an adverse health or ecological effect following exposure to a stressor [8]. It is anchored by a Molecular Initiating Event (MIE), the initial interaction between a chemical and a biomolecule, and culminates in an Adverse Outcome (AO) relevant to risk assessment. These anchors are connected by a causal chain of measurable Key Events (KEs) [15]. However, the mere description of a pathway is insufficient for regulatory application. The true translational power of an AOP lies in its applicability—the defined boundaries within which its mechanistic predictions are reliable for specific chemical classes, species, life stages, and exposure scenarios [15]. This whitepaper explores the concept of taxonomic applicability within AOP research, detailing how establishing the domain of applicability transforms qualitative biological narratives into quantitative, decision-ready tools for safety assessment.

Table 1: Core Definitions in Adverse Outcome Pathway Framework [8] [15]

Term	Definition	Role in Risk Assessment
Molecular Initiating Event (MIE)	The initial interaction of a stressor with a biological target (e.g., receptor binding, protein inhibition).	Identifies the point of intervention and potential for high-throughput screening.
Key Event (KE)	A measurable change in biological state that is essential to the progression towards the AO.	Serves as a biomarker for monitoring pathway perturbation and building quantitative models.
Key Event Relationship (KER)	A scientifically based description of the causal or mechanistic linkage between two KEs.	Enables extrapolation (e.g., from in vitro to in vivo) and prediction of downstream effects.
Adverse Outcome (AO)	A biological change at the organism or population level considered relevant for regulatory decision-making.	Defines the regulatory endpoint of concern (e.g., liver fibrosis, population decline).
Applicability Domain	The defined boundaries (chemical, taxonomic, life stage) within which the AOP is considered valid.	Critical for determining the context of use and reliability of the AOP for a given assessment.

The Anatomy of an AOP: From Qualitative Description to Quantitative Tool

The foundational strength of the AOP framework is its standardized structure, which organizes fragmented mechanistic knowledge into a logical, testable sequence. This structure progresses from a qualitative AOP, which outlines the hypothesized causal linkages, to a quantitative AOP (qAOP), which encodes these relationships with mathematical models suitable for prediction.

Qualitative AOP Development follows established OECD guidelines and involves the identification of essential KEs and the description of KERs based on biological plausibility and empirical evidence [15]. Confidence in the pathway is assessed using modified Bradford-Hill criteria, evaluating factors such as dose-response concordance, temporal sequence, and consistency of evidence [15]. A parallel assessment addresses the taxonomic applicability of the KEs and KERs, questioning whether they are expected to be conserved across relevant species (e.g., from rat to human) [15].

Quantitative AOP (qAOP) Development is the critical step for regulatory use. It involves defining quantitative relationships for KERs, often through response-response modeling [16]. For example, a qAOP for liver carcinogenicity can quantify the relationship between the incidence of early proliferative lesions (a KE) and the eventual incidence of liver tumors (the AO). A study demonstrated that data from 90-day rodent studies could predict 2-year carcinogenicity outcomes, with predictive sensitivity greatly improved by incorporating a biomarker like BrdU labelling [16]. This quantification allows for the derivation of a Point of Departure (PoD) from an earlier, more easily measured KE, supporting faster and more efficient risk assessment [16].

Figure 1: The AOP framework links an MIE to an AO via a series of KEs connected by KERs. Its regulatory utility is unlocked by quantitative modeling (e.g., QKAR/qAOP) and the explicit definition of an applicability domain [8] [17] [16].

The Pillars of Taxonomic Applicability: Defining the Domain of Confidence

Taxonomic applicability asks a fundamental question: For which species is the described AOP valid? Establishing this domain is not a binary "yes/no" determination but a structured assessment of confidence based on the conservation of biological pathways. This assessment is a cornerstone for translating data from model organisms (e.g., rat) to humans or across ecological species.

3.1. Assessing Conservation of Key Events and KERs The primary scientific task is to evaluate the biological plausibility that each KE and the linkages between them are conserved in the target taxon. This involves examining the comparative biology of the relevant proteins, signaling pathways, and tissue-level responses. For example, an AOP for thyroid hormone disruption leading to developmental neurotoxicity relies on the conservation of the hypothalamic-pituitary-thyroid axis across mammals [8]. Regulatory agencies like the EPA actively investigate such cross-species applicability to ensure human relevance [8].

3.2. Integrating Chemical-Specific Properties: From AOP to Mode of Action (MOA) AOPs are intentionally chemical-agnostic, describing the biological pathway of toxicity independent of any specific stressor [14]. To assess a particular chemical, the AOP must be integrated with chemical-specific data, forming a Mode of Action (MOA) analysis [14]. The critical link is demonstrating that the chemical of concern can induce the MIE and subsequent KEs within the target tissue at relevant exposures. This integration considers Toxicokinetics (TK): does the chemical reach the target site? It also considers Metabolism: is it bioactivated or detoxified? This step effectively narrows the broad taxonomic applicability of the AOP to a specific, chemically defined applicability domain for the MOA.

3.3. The Role of New Approach Methodologies (NAMs) NAMs, including in vitro assays and in silico models, are essential for testing applicability. High-throughput screening can determine if a chemical triggers the MIE in human-derived cells. Advanced in vitro models, such as 3D human organotypic cultures, can test the progression of KEs in a human-relevant system. For instance, a study using 3D human bronchial epithelial cultures exposed to whole cigarette smoke successfully recapitulated KEs along an AOP for mucus hypersecretion, directly providing human-specific pathway data [18]. Computational models like Quantitative Knowledge-Activity Relationships (QKARs) further enhance applicability by using domain knowledge (e.g., drug metabolism, off-target effects) to predict toxicity, often outperforming traditional structure-only models (QSARs), especially for structurally similar compounds with divergent toxicities [17].

Table 2: Frameworks for Establishing and Applying AOP Applicability

Framework	Primary Focus	Role in Defining Applicability	Output for Risk Assessment
Adverse Outcome Pathway (AOP)	Chemical-agnostic biological pathway.	Defines the potential mechanistic sequence. Establishes initial confidence in cross-species KE conservation [15].	A generalized template for toxicity.
Mode of Action (MOA)	Chemical-specific application of an AOP.	Integrates TK/TD data to confirm the chemical induces the KEs in the target species. Defines the chemical-specific applicability domain [14].	A weight-of-evidence conclusion that the AOP operates for Chemical X in Species Y.
Integrated Approaches to Testing and Assessment (IATA)	A strategic, tiered testing workflow.	Guides the generation of new data (using NAMs) to address uncertainties in AOP/MOA applicability for a given decision context [14].	A tailored testing strategy and data package fit for a regulatory purpose.

Figure 2: Establishing taxonomic applicability is a multi-step process moving from assessing biological conservation, to integrating chemical-specific data to form an MOA, and finally to defining a precise domain for quantitative regulatory application.

Quantitative Frontiers: QKARs, qAOPs, and Predictive Modeling

The transition from qualitative pathway description to quantitative prediction is the bridge that connects mechanistic knowledge directly to risk assessment. This quantitative frontier is embodied in two advanced approaches.

4.1. Quantitative Knowledge-Activity Relationships (QKARs) QKARs represent a paradigm shift from traditional Quantitative Structure-Activity Relationships (QSARs). While QSARs predict toxicity based solely on chemical structure, QKARs leverage embedded domain knowledge—such as mechanisms of action, metabolic pathways, and off-target interactions—extracted from scientific literature using advanced AI [17]. In a direct comparison for predicting drug-induced liver injury (DILI) and cardiotoxicity (DICT), QKAR models consistently outperformed QSAR models [17]. Crucially, QKARs were better at differentiating between structurally similar drugs with different toxicities (e.g., ibuprofen vs. withdrawn ibufenac), a critical task where structural descriptors alone fail [17]. This knowledge-rich approach directly informs the biological applicability of a toxicity prediction.

4.2. Quantitative AOP (qAOP) Modeling A qAOP formalizes the relationships between KEs with mathematical functions. A seminal example is the development of a qAOP for non-genotoxic liver carcinogenicity [16]. Researchers used Bayesian logistic regression to model the probabilistic relationship between the incidence of early proliferative lesions (a KE) and the eventual incidence of liver tumors (the AO) using rodent data. The model showed that including a biomarker for cell proliferation (BrdU labelling) significantly improved the sensitivity and precision of predictions [16]. This allows a Point of Departure (PoD) for risk assessment to be derived from a 90-day study biomarker, potentially obviating the need for a 2-year carcinogenicity bioassay for certain chemicals within the model's applicability domain.

Experimental Protocols for AOP-Driven In Vitro Assessment

Translating an AOP into a testable hypothesis requires robust, human-relevant experimental models. The following protocol, based on a study investigating cigarette-smoke-induced airway mucus hypersecretion, exemplifies an AOP-based in vitro assessment strategy [18].

5.1. Experimental Objective To recapitulate a human-relevant AOP for chronic obstructive pulmonary disease (COPD)-like mucus hypersecretion by repeatedly exposing a sophisticated in vitro model to whole cigarette smoke (WCS) and measuring sequential KEs.

5.2. Materials and Methods

• Cell Model: Primary normal human bronchial epithelial cells (HBECs) from multiple donors, cultured at the air-liquid interface (ALI) to form a fully differentiated, 3D mucociliary epithelium [18].
• Coculture System: Differentiated 3D-HBECs are cocultured with M2-like macrophages (differentiated from U937 monocyte cell line) on the basolateral side to provide essential immune cell signals (e.g., IL-4/IL-13) that are part of the AOP [18].
• Exposure Regime: Tissues are repeatedly exposed to whole cigarette smoke (using a smoking machine) or clean air control over a duration of up to two weeks (e.g., 6 exposures) [18].
• Endpoint Analysis (Measuring KEs): Tissues are harvested at multiple time points to measure KEs along the hypothesized AOP:
  • Early KEs (Acute): Oxidative stress (ROS), EGFR activation, SP1 activation [18].
  • Later KEs (Chronic): Intracellular mucus production (MUC5AC staining), goblet cell metaplasia/hyperplasia (histology), and secreted mucus hypersecretion [18].

Table 3: Research Reagent Solutions for AOP-Based In Vitro Assessment (Example: Airway Toxicity) [18]

Reagent/Material	Function in the Experimental Protocol	Role in Informing AOP Applicability
Primary Human Bronchial Epithelial Cells (HBECs)	Forms the core 3D, differentiated tissue model at the air-liquid interface (ALI).	Provides human-specific biological context, ensuring KEs are measured in a relevant cellular system.
ALI Culture Medium (e.g., PneumaCult-ALI)	Supports the differentiation and long-term maintenance of a pseudostratified epithelium with functional cilia and mucus production.	Enables the expression of physiologically relevant phenotypes (e.g., mucus secretion) that are critical later KEs in the AOP.
Transwell Permeable Supports	Physical scaffold for ALI culture, allowing separate access to apical (air) and basolateral (medium) compartments.	Permits realistic aerosol exposure (apical side) and coculture with immune cells (basolateral side), modeling tissue complexity.
U937 Monocyte Cell Line	Differentiated into M2-like macrophages for basolateral coculture with 3D-HBECs.	Provides essential immune-derived signals (e.g., IL-4/IL-13) that are part of the AOP network, testing the necessity of cross-talk for the AO.
Whole Cigarette Smoke (WCS) Generation System	Provides a complex, realistic aerosol exposure rather than a simple solvent extract.	Tests the AOP under relevant exposure conditions, increasing confidence that the pathway is triggered by the actual mixture of concern.
Assays for KE Measurement (e.g., ROS kits, phospho-EGFR ELISA, qPCR, immunohistochemistry for MUC5AC)	Quantifies the magnitude of change at specific nodes in the AOP.	Generates quantitative, time-course data for KEs, enabling the development of response-response models and the assessment of inter-donor variability.

5.3. Data Interpretation and Applicability Insights This approach generates rich, time-course data mapping onto the AOP. Successful recapitulation of the KE sequence strengthens the biological plausibility of the AOP in a human system. Furthermore, observing inter-donor variability in the timing and amplitude of KEs mirrors known population variability in disease susceptibility, directly informing the human applicability and uncertainty of the pathway [18].

Figure 3: An AOP-guided experimental workflow for human-relevant risk assessment. The protocol involves developing a complex in vitro model, applying a repeated exposure, measuring sequential KEs, and using the data to refine the AOP's applicability domain for human health [18].

Implementation in Regulatory Decision-Making: The Path Forward

The ultimate test of AOP applicability is its successful integration into regulatory paradigms. This integration is actively underway, facilitated by several key developments.

6.1. Regulatory Frameworks Encouraging NAMs Initiatives like the FDA's Roadmap to Reducing Animal Testing explicitly encourage the adoption of NAMs, including AOP-informed approaches and AI models, for preclinical safety studies [17]. This creates a clear regulatory pathway for data derived from applicable, well-characterized AOPs.

6.2. AOP Knowledge Base (AOP-KB) and Collaborative Tools The AOP-KB, maintained by the OECD and partners, is a central repository that standardizes AOP development and sharing [8] [15]. Its modules, like the AOP Wiki, allow for the transparent documentation of an AOP's supporting evidence, confidence, and—critically—its applicability domain. This shared resource is vital for building global regulatory acceptance [15].

6.3. Integrated Approaches to Testing and Assessment (IATA) IATA are problem-solving frameworks that strategically combine AOPs with other data sources (in silico, in chemico, in vitro) within a defined applicability domain to address a specific regulatory question [14]. An AOP serves as the mechanistic backbone of an IATA, guiding which NAMs to use to measure specific KEs and how to interpret the resulting data stream to reach a safety decision.

In conclusion, the power of mechanistic knowledge in toxicology is unlocked not by its generation alone, but by the rigorous assessment of its applicability. Through structured evaluation of taxonomic conservation, integration with chemical-specific data, quantitative modeling via qAOPs and QKARs, and human-relevant experimental validation, the AOP framework provides the essential bridge. It transforms isolated biological insights into a structured, predictive, and decision-ready format, enabling a more efficient, humane, and human-relevant future for regulatory risk assessment.

The Adverse Outcome Pathway (AOP) framework is a structured representation linking a Molecular Initiating Event (MIE) through a series of intermediate Key Events (KEs) to an Adverse Outcome (AO) relevant to risk assessment [3]. A critical yet often inadequately defined component of an AOP is its Taxonomic Domain of Applicability (tDOA)—the range of species for which the pathway is biologically plausible [3]. Typically, an AOP is developed based on empirical data from a single or a handful of species, with assumptions about broader applicability that lack documented evidence [3]. This gap limits the confidence and utility of AOPs in regulatory decision-making, particularly for protecting untested species.

Concurrently, the volume and complexity of AOP-related data (e.g., omics, phenotypic, toxicological) are expanding rapidly. To maximize the scientific and regulatory value of this data, effective management is essential. The FAIR Guiding Principles—standing for Findable, Accessible, Interoperable, and Reusable—provide a foundational framework for enhancing the utility of digital assets by making them machine-actionable [19]. The integration of FAIR principles with AOP development, specifically to strengthen tDOA evidence, forms the core of a necessary 2025 roadmap. This integration enables systematic data collation, computational analysis, and transparent evidence assessment, transforming the tDOA from an assumption into a data-driven, well-characterized parameter.

Foundational Principles: FAIR and tDOA

The FAIR Guiding Principles

The FAIR principles aim to optimize the reuse of data and metadata by both humans and computational systems [19]. They are defined as follows:

• Findable: Metadata and data should be richly described and registered in searchable resources with persistent identifiers [19].
• Accessible: Data are retrievable using standardized, open protocols, with authentication where necessary [19].
• Interoperable: Data and metadata use formal, accessible, and broadly applicable languages and vocabularies to enable integration with other datasets [19].
• Reusable: Data and metadata are described with multiple relevant attributes, clear licenses, and provenance to meet community standards [19].

Defining the Taxonomic Domain of Applicability (tDOA)

The tDOA for an AOP is established by evaluating the conservation of its constituent KEs and KE Relationships (KERs) across species. Conservation is assessed along two primary lines [3]:

• Structural Conservation: The presence and conservation of the biological entities (e.g., proteins, genes, receptors) involved in the pathway.
• Functional Conservation: The consistent role and function of those entities across the taxa of interest.

Empirical toxicity studies traditionally provide evidence for functional conservation. In contrast, bioinformatics tools are increasingly critical for efficiently evaluating structural conservation across broad taxonomic groups, providing a scalable line of evidence for expanding the tDOA [3].

Quantitative Analysis: Bridging FAIR Data and tDOA Evidence

Applying FAIR-aligned bioinformatics tools generates quantitative evidence for tDOA assessment. The table below summarizes a case study analyzing an AOP for nicotinic acetylcholine receptor (nAChR) activation leading to colony failure in bees, using the SeqAPASS tool [3].

Table 1: SeqAPASS Analysis of Protein Conservation for an AOP (AOP 89: nAChR Activation to Colony Failure) [3]

Protein Target (Molecular Key Event)	Primary Taxa of Interest (Empirical tDOA)	SeqAPASS Level 1 (Primary Sequence)	SeqAPASS Level 2 (Functional Domain)	SeqAPASS Level 3 (Critical Residue)	Inferred Plausible tDOA
Nicotinic acetylcholine receptor (nAChR) subunit	Apis mellifera (Honey bee)	High similarity across Insecta	Domain architecture conserved in Insecta & Arachnida	Ligand-binding residues conserved in Insecta	Insecta (possibly Arachnida)
Voltage-gated sodium channel	Apis mellifera	High similarity across Animalia	Domain architecture conserved in Metazoa	Key functional residues broadly conserved	Metazoa
Acetylcholinesterase (AChE)	Apis mellifera	High similarity across Animalia	Catalytic domain conserved in Metazoa	Active site residues highly conserved	Metazoa
Gamma-aminobutyric acid (GABA) receptor	Apis mellifera	High similarity across Animalia	Ligand-gated ion channel domain conserved	Binding site residues conserved in Arthropoda & Vertebrata	Arthropoda, Vertebrata

The analysis demonstrates how structured, accessible data (protein sequences) enables computational tDOA extrapolation. For instance, while empirical data may be limited to honey bees, SeqAPASS analysis shows nAChR structural conservation likely extends across the Insecta class, and AChE conservation across Metazoa [3]. This directly informs and expands the plausible tDOA.

Table 2: Impact of FAIR Principles on tDOA Characterization Workflows

FAIR Principle	Traditional AOP Development (Low FAIR)	FAIR-Enhanced AOP Development (2025 Roadmap)	Impact on tDOA Assessment
Findable	Data in supplementary files; metadata sparse.	KEs/KERs linked to unique IDs in AOP-KB; datasets in indexed repositories.	Enables automated discovery of all relevant toxicity & omics data for cross-species analysis.
Accessible	Data behind paywalls or in proprietary formats.	Data retrievable via open APIs using standard protocols (e.g., SPARQL, REST).	Allows bioinformatics tools (e.g., SeqAPASS) to programmatically access needed sequence/structure data.
Interoperable	Inconsistent terminologies; custom data models.	Use of controlled ontologies (e.g., GO, ChEBI) & standard AOP-JSON schema.	Allows integration of AOP data with external biological databases (UniProt, Ensembl) for conservation analysis.
Reusable	Lack of provenance & clear licensing.	Rich metadata with experimental details, species info, and clear usage licenses.	Provides the context necessary to evaluate the quality and relevance of data for tDOA extrapolation.

Experimental Protocol: The SeqAPASS Bioinformatics Workflow

The following detailed protocol is based on the methodology used to generate the evidence in Table 1, exemplifying a FAIR-aligned computational experiment for tDOA definition [3].

Objective: To evaluate the structural conservation of protein targets associated with Key Events in an AOP across taxonomic groups, providing evidence for the Taxonomic Domain of Applicability (tDOA).

Materials & Input Data:

• Query Protein Sequences: FASTA format sequences for the protein(s) of interest from the reference species (e.g., Apis mellifera nAChR subunit). Source: Public databases (UniProt, NCBI).
• SeqAPASS Tool: The publicly available web-based Sequence Alignment to Predict Across Species Susceptibility tool developed by the US EPA [3].
• Taxonomic Database: Integrated taxonomic information for filtering and grouping results.

Procedure:

• Preparation and Submission:
  • Identify and retrieve the primary amino acid sequence(s) for the molecular target of each KE.
  • Access the SeqAPASS web tool. Submit each query sequence individually, specifying the reference species.

• Level 1 Analysis (Primary Sequence Similarity):

  • Action: The tool performs a BLAST-based alignment against a comprehensive protein sequence database.
  • Output: A list of orthologous sequences across species, with percent identity scores.
  • Interpretation: High sequence similarity suggests the protein is present and structurally related in other species. A threshold (e.g., ≥70% identity) can be used to infer initial taxonomic breadth.

• Level 2 Analysis (Functional Domain Conservation):

  • Action: For orthologs identified in Level 1, the tool maps conserved functional domains (e.g., Pfam domains) from the reference sequence to the target sequences.
  • Output: A matrix showing the presence/absence and integrity of specific domains across species.
  • Interpretation: Conservation of domain architecture is necessary for protein function. Loss of a critical domain narrows the plausible tDOA.

• Level 3 Analysis (Critical Residue Conservation):

  • Action: The tool aligns sequences around known critical amino acid residues (e.g., ligand-binding sites, catalytic triads) essential for the protein's role in the AOP.
  • Output: A detailed view of residue conservation at these specific positions.
  • Interpretation: Conservation of critical residues provides strong evidence for retained protein function (structural and functional conservation) across species. This is the strongest line of bioinformatics evidence for tDOA.

• Data Integration and tDOA Postulation:

  • Compile results from all three levels for all protein targets in the AOP.
  • The most conservative (narrowest) tDOA inferred from any critical KE protein defines the overall biologically plausible tDOA for the AOP.
  • Document all results as computational evidence weights in the AOP-Wiki or similar knowledge base.

Visualizing the Integrated Framework

The following diagrams, created using Graphviz DOT language, illustrate the logical relationship between FAIR data, computational analysis, and tDOA characterization, as well as the specific bioinformatics workflow.

Diagram: FAIR Data as the Foundation for AOP tDOA Characterization. This diagram illustrates how FAIR principles applied to underlying data support both empirical and computational evidence gathering, which are integrated to define the AOP's Taxonomic Domain of Applicability.

Diagram: SeqAPASS Three-Level Workflow for Structural Conservation Analysis. This workflow details the stepwise bioinformatics protocol for assessing protein conservation, from initial query to evidence upload.

The following toolkit is essential for researchers implementing FAIR-aligned tDOA expansion studies for AOPs.

Table 3: Research Reagent Solutions for tDOA and FAIR AOP Research

Category	Item/Resource	Function & Relevance	FAIR Alignment Example
Bioinformatics Tools	SeqAPASS Tool [3]	Provides hierarchical (sequence, domain, residue) assessment of protein structural conservation across species to inform tDOA.	Input/Output can be standardized (e.g., FASTA, JSON).
	BLAST/UniProt	Foundational for sequence retrieval and initial similarity searches.	Public, accessible databases with stable identifiers and APIs.
Data & Knowledge Bases	AOP-Wiki (aopwiki.org)	Central repository for AOP development, sharing, and collaborative annotation of KEs, KERs, and evidence [3].	Platform for making AOPs Findable and Reusable; tDOA evidence can be documented.
	UniProt, NCBI, Ensembl	Provide the essential, curated protein and genomic sequence data required for cross-species analyses.	Exemplify Findable, Accessible, and Interoperable resources.
Standardization Resources	AOP-JSON Schema	A standardized data exchange format for AOP information, enhancing interoperability.	Directly enables the I in FAIR for AOP data.
	Ontologies (GO, ChEBI, etc.)	Controlled vocabularies for describing biological processes, chemicals, and anatomical entities.	Critical for semantic Interoperability and data integration.
Experimental Reagents	Species-Specific Assay Kits (e.g., ELISA, qPCR)	To generate empirical functional conservation data for KEs in novel species identified via bioinformatics.	Data generated should be annotated with the species' taxonomic ID and detailed protocols (Reusable).
	Reference Chemicals (e.g., Neonicotinoids for AOP 89) [3]	Prototypical stressors used to empirically test the postulated AOP and its tDOA in vivo or in vitro.	Chemical should be identified via persistent ID (e.g., InChIKey, CAS) in metadata.

The 2025 Roadmap: Strategic Actions for Implementation

The path to fully realizing FAIR-enhanced AOP development requires coordinated action. The following roadmap outlines key strategic objectives for 2025:

• Mandate FAIR Metadata in AOP Submissions: AOP knowledge bases (e.g., AOP-Wiki) should require structured, ontology-anchored metadata for all submitted KEs and KERs, including explicit fields for empirical species and computational tDOA evidence [3].
• Develop Integrated Computational Workbenches: Create user environments that seamlessly connect AOP diagrams to bioinformatics tools (like SeqAPASS) and public data repositories, automating the generation of tDOA evidence reports.
• Establish Community tDOA Evidence Standards: Define consensus guidelines on weighing and integrating different lines of evidence (e.g., SeqAPASS Level 3 vs. in vitro functional data) for tDOA postulation.
• Promote the Citation of Data: Encourage and facilitate the citation of key datasets (omics, toxicity) used in AOP development via persistent identifiers, making the evidence base more Findable and Reusable [19].

The convergence of the AOP framework and FAIR data principles represents a transformative opportunity for predictive toxicology and regulatory science. By applying FAIR principles—ensuring AOP-relevant data is Findable, Accessible, Interoperable, and Reusable—we can systematically and efficiently address the critical challenge of defining the Taxonomic Domain of Applicability. The 2025 roadmap calls for integrating computational bioinformatics workflows, like the SeqAPASS protocol, directly into the AOP development lifecycle [3]. This creates a virtuous cycle where FAIR data enables robust tDOA characterization, and well-defined tDOAs enhance the reliability and reuse value of the AOP itself. The ultimate outcome is a more rigorous, transparent, and scalable knowledge base that strengthens confidence in using AOPs for species extrapolation and safety assessment.

How to Define and Apply AOP Taxonomic Domains: Practical Strategies and Case Studies

Within the broader thesis on adverse outcome pathway (AOP) taxonomic applicability, defining the boundaries and relevance of a pathway—its applicability—is a foundational scientific challenge. The AOP framework is a conceptual construct that organizes mechanistic knowledge, describing a sequential chain of causally linked events from a Molecular Initiating Event (MIE) to an Adverse Outcome (AO) at the organism or population level [8] [20]. It is chemically agnostic, meaning it describes biological response pathways that can be initiated by any stressor (chemical or otherwise) capable of triggering the initial molecular perturbation [1].

Determining whether a developed AOP is applicable to a specific chemical, species, life stage, or sex is critical for its reliable use in regulatory decision-making, chemical prioritization, and the development of New Approach Methodologies (NAMs) [8] [21]. Two philosophically and procedurally distinct methodological paradigms exist for defining this applicability: a priori reasoning and empirical data integration. This guide provides an in-depth technical comparison of these core methodologies, detailing their principles, workflows, strengths, and limitations within modern toxicological research.

Foundational Concepts: AOP Structure and the Applicability Domain

An AOP is composed of modular units: the MIE, intermediate Key Events (KEs), and the AO, connected by Key Event Relationships (KERs) [22] [20]. The applicability domain of an AOP defines the conditions under which its described causal sequence is expected to hold. Key dimensions of applicability include:

• Taxonomic Applicability: The species for which the pathway is relevant.
• Life Stage Applicability: Specific developmental or life stages susceptible to the perturbation.
• Sex Applicability: Relevance to male, female, or both sexes.
• Chemical Applicability: The classes of stressors (e.g., specific receptor agonists) capable of initiating the MIE.

A precisely defined applicability domain reduces uncertainty in extrapolation, a central hurdle in leveraging mechanistic data for risk assessment [21].

Methodology 1: A Priori Reasoning-Based Applicability Definition

A priori reasoning (deductive reasoning) establishes applicability based on existing foundational knowledge, biological plausibility, and theoretical principles before empirical testing for a specific case is conducted.

Core Principles and Workflow

This methodology relies on the expert-driven application of established biological principles. It begins with a well-characterized MIE (e.g., binding to a conserved nuclear receptor) and uses comparative biology to deduce taxonomic applicability. For instance, if the MIE involves a receptor with high sequence and functional homology across all mammals, applicability is inferred for the entire mammalian taxonomic class a priori. The workflow is linear and hypothesis-driven:

• Define the MIE and Critical KEs: Identify the essential, conserved biological targets and processes.
• Conduct Comparative Biological Analysis: Use existing genomic, proteomic, and physiological data to assess the conservation of these targets across taxa.
• Establish Plausible Boundaries: Deduce the applicability domain based on the presence/absence and functional conservation of the required biological components.
• Generate Testable Predictions: Formulate specific, falsifiable hypotheses about pathway activity in untested species or contexts.

Case Study Example: Ecdysone Receptor Agonism in Arthropods

The AOP for "Ecdysone receptor activation leading to mortality" in arthropods is a classic example built via a priori reasoning [22]. Ecdysone is a critical molting hormone in insects and crustaceans. The reasoning proceeds as follows:

• Premise 1: The ecdysone receptor (EcR) is essential for molting and development in arthropods.
• Premise 2: Certain synthetic chemicals (e.g., diacylhydrazines) are known to act as potent agonists of EcR.
• Premise 3: Ligation of EcR initiates a defined cascade of gene expression, disrupted molting, and eventual mortality.
• Conclusion (Applicability): Therefore, this AOP is a priori applicable to all arthropod species possessing a functionally conserved EcR and can be initiated by any chemical stressor capable of potently activating that receptor. Empirical testing then validates this deduced domain.

Strengths and Limitations

Table 1: Assessment of A Priori Reasoning Methodology

Aspect	Strengths	Limitations
Efficiency	Rapid, cost-effective for initial domain setting. No requirement for new, cross-species data generation.
Basis	Grounded in fundamental, conserved biology (e.g., receptor homology, essential pathways).	May overlook unique physiological, toxicokinetic, or compensatory mechanisms in specific taxa.
Use Case	Ideal for well-conserved pathways (e.g., endocrine signaling in vertebrates) and for prioritizing testing.
Risk		High risk of both false positives (over-extrapolation) and false negatives (missing applicable taxa) if biological complexity is oversimplified.
Output	Provides clear, testable hypotheses to guide targeted empirical work.	Lacks quantitative precision for predicting tipping points or thresholds in new species.

Methodology 2: Empirical Data Integration-Based Applicability Definition

Empirical data integration (inductive reasoning) defines applicability by aggregating, mining, and analyzing large-scale experimental and observational data to identify consistent patterns and associations that delineate the pathway’s operational domain.

Core Principles and Workflow

This data-driven approach treats applicability as an emergent property revealed through systematic analysis. It leverages high-throughput in vitro screening (HTS), toxicogenomics, curated disease databases, and computational modeling to map the functional space of an AOP [23] [9]. The workflow is iterative and discovery-oriented:

• Data Assembly: Aggregate heterogeneous data (e.g., ToxCast HTS assay data, gene-chemical-disease associations from the Comparative Toxicogenomics Database (CTD), transcriptomic responses).
• Computational Mining: Apply algorithms (e.g., Frequent Itemset Mining, natural language processing) to find non-random associations between MIEs/KEs and AOs across chemicals and systems [23] [24].
• Network Construction: Generate computationally-predicted AOP (cpAOP) networks where nodes (genes, diseases) and edges (associations) define the empirical applicability landscape [23].
• Domain Inference: The applicability domain is defined by the empirical footprint of the pathway—the chemicals, genes, and species contexts where the associative pattern is statistically robust.

Case Study Example: Data-Driven AOP Network for Endocrine Disruption

A 2023 study developed a data-driven AOP network focused on Estrogen, Androgen, Thyroid, and Steroidogenesis (EATS) modalities [24]. The methodology was:

• Structured Search: A priori search terms from regulatory guidance were used to identify relevant AOPs in the AOP-Wiki.
• Automated Data Extraction & Curation: A computational workflow (R-script) extracted and filtered AOP components, followed by expert curation.
• Network Generation & Analysis: The script processed the data to visualize an AOP network, revealing shared KEs and connections between EATS-related pathways.
• Applicability Insight: The resulting network empirically mapped which specific MIEs (e.g., estrogen receptor activation) were linked through shared KEs to which AOs (e.g., reproductive dysfunction) across multiple taxa, defining a chemically and biologically granular applicability domain based on integrated evidence.

Strengths and Limitations

Table 2: Assessment of Empirical Data Integration Methodology

Aspect	Strengths	Limations
Basis	Grounded in observed, multi-dimensional data patterns; can reveal novel, unexpected associations.	Quality is entirely dependent on the scope, bias, and quality of underlying databases and assays.
Precision	Can provide quantitative associations and probabilistic applicability estimates (e.g., via qAOP models).	Computationally intensive and requires significant bioinformatic expertise.
Discovery	Powerful for AOP discovery and expanding networks, moving beyond linear pathways to complex systems [21] [24].	Identifies correlation; establishing causality still requires expert evaluation and experimental validation.
Scope	Capable of integrating diverse data (HTS, omics, pathology) for a holistic view.	Sparse or uneven data coverage across species/chemicals can leave significant gaps in the applicability map.

Comparative Analysis and Strategic Integration

The two methodologies are complementary rather than opposed. A priori reasoning is efficient and principle-based, ideal for initial framework development and hypothesis generation. Empirical integration is comprehensive and evidence-based, essential for validation, refinement, and quantitative assessment.

Table 3: Comparative Summary of Core Methodologies

Feature	A Priori Reasoning	Empirical Data Integration
Philosophical Basis	Deduction (theory-first)	Induction (data-first)
Primary Input	Established biological principles, expert knowledge	Experimental & observational data (HTS, omics, curated DBs)
Workflow Nature	Linear, hypothesis-driven	Iterative, discovery-driven
Key Output	Testable applicability hypotheses	Mapped associative networks, probabilistic domains
Best Suited For	Well-conserved pathways, rapid prioritization, guiding targeted testing	Complex/poorly understood pathways, quantitative model building, discovering cross-talk
Major Risk	Over-extrapolation based on oversimplified biology	Spurious correlations, gaps from data sparsity

The Integrated Strategy: The most robust approach to defining AOP applicability employs a synergistic loop:

• Use a priori reasoning to define an initial, plausible applicability domain based on conserved biology.
• Use empirical data integration to test, challenge, and refine that domain, identifying boundaries and exceptions.
• Use new empirical insights to update and improve the theoretical models, restarting the cycle.

This integration is central to developing quantitative AOPs (qAOPs) and reliable AOP networks, which are recognized as future priorities for the field [21].

The Scientist's Toolkit: Essential Research Reagent Solutions

Table 4: Key Research Reagents and Resources for AOP Applicability Research

Item / Resource	Function in Applicability Research	Example / Source
Curated Gene Annotation Databases	Provides systematic molecular annotation for Key Events, enabling cross-species homology mapping and omics data integration for empirical domain analysis [9].	Ensembl, NCBI Gene, annotation sets from studies like[citeation:6].
High-Throughput Screening (HTS) Assay Suites	Generates empirical data on chemical bioactivity across numerous molecular targets (potential MIEs), used to map chemical initiators for an AOP and its cpAOP networks [23].	US EPA ToxCast/Tox21 assay battery.
Comparative Toxicogenomics Database (CTD)	Integrates curated chemical-gene, chemical-disease, and gene-disease interaction data, serving as a primary source for mining empirical associations between MIEs/KEs and Adverse Outcomes [23].	http://ctdbase.org/
AOP-Knowledge Base (AOP-KB) / AOP-Wiki	Central repository for modular AOP components; essential for accessing existing KEs/KERs, conducting structured searches (e.g., for EATS modalities), and building networks [8] [24].	https://aopwiki.org/
Computational Mining Algorithms	Identifies frequent, non-random associations between items (e.g., chemicals, genes, outcomes) in large, sparse datasets to build cpAOP networks and infer applicability [23].	Frequent Itemset Mining (FIM) algorithms.
Ontologies (e.g., GO, DTO)	Structured, controlled vocabularies that standardize the description of biological processes, functions, and toxicological outcomes, enabling consistent annotation and computational reasoning across studies [20].	Gene Ontology (GO), Developmental Toxicity Ontology (DTO).
Species-Specific in vitro Models	Cell lines or primary cells from different taxa used to experimentally test the conservation of KE responses predicted by a priori reasoning or suggested by empirical data mining.	Rodent vs. human hepatocytes, zebrafish embryo models.
Weight-of-Evidence Evaluation Frameworks	Structured protocols (e.g., modified Bradford-Hill criteria) to assess the biological plausibility, essentiality, and empirical concordance of KERs, strengthening confidence in the deduced applicability domain [22] [25].	OECD Handbook[sciation:3].

Leveraging the AOP-Wiki and Computational Tools for Systematic Applicability Annotation

The Adverse Outcome Pathway (AOP) framework is a systematic, transparent approach for organizing toxicological knowledge into a structured sequence of causally linked biological events. It describes a logical progression from a Molecular Initiating Event (MIE), through intermediate Key Events (KEs), culminating in an Adverse Outcome (AO) of regulatory relevance [22]. This framework is central to modern toxicology, particularly for enabling the development of New Approach Methodologies (NAMs) aimed at reducing animal testing and supporting predictive risk assessment [9]. A critical aspect of this framework, and the focus of this technical guide, is taxonomic applicability—determining the specific species, life stages, and sexes for which a described AOP is valid [5]. This systematic annotation of applicability domains is essential for the credible extrapolation of mechanistic knowledge from model organisms to humans or across ecosystems, forming a core thesis for robust, fit-for-purpose use of AOPs in research and regulation [26].

This guide details the integration of the primary AOP repository, the AOP-Wiki, with advanced computational tools to systematize the extraction, annotation, and analysis of applicability information. As the AOP knowledge base grows, moving beyond manual curation to data-driven, automated workflows is paramount for efficiently leveraging this resource [27]. We explore practical strategies and provide detailed protocols for employing semantic web technologies, natural language processing, and graph-based analytics to enhance the consistency, interoperability, and utility of taxonomic applicability annotation within the AOP framework.

Foundational Knowledge: The AOP-Wiki as a Curated Repository

The AOP-Wiki serves as the central, crowd-sourced repository for AOP development and is a core component of the broader OECD AOP Knowledge Base (AOP-KB) [26]. It stores AOPs as modular constructs, where individual Key Events (KEs) and Key Event Relationships (KERs) are discrete units that can be reused across multiple pathways [5]. The development process follows a standardized workflow to ensure scientific rigor, as outlined in the OECD AOP Developers' Handbook [5].

Table: Core Components of an AOP as Structured in the AOP-Wiki

Component	Abbreviation	Definition	Role in Applicability
Molecular Initiating Event	MIE	The initial interaction of a stressor with a biomolecule within an organism that starts the AOP [5].	Often highly conserved across taxa; defines the initial point of perturbation.
Key Event	KE	A measurable change in biological state essential to the progression of the AOP [5].	Each KE has its own domain of applicability (taxon, life stage, sex).
Key Event Relationship	KER	A scientifically supported causal or predictive link between an upstream and a downstream KE [5].	Defines the biological plausibility of the sequence, which may vary by taxon.
Adverse Outcome	AO	An adverse effect at the individual or population level relevant to risk assessment [5].	Defines the regulatory or protection goal, which is taxon-specific.

The Wiki’s structure mandates the description of biological plausibility, essentiality, and empirical support for KERs, which are assessed using modified Bradford-Hill considerations [22]. A crucial part of a KE’s description is its domain of applicability, which includes annotations for taxonomy (e.g., NCBI Taxon ID), life stage, and sex [5] [28]. These annotations are critical for determining whether a pathway developed in one model species (e.g., zebrafish or rat) can be reliably extrapolated to another (e.g., human). However, this information has historically been embedded in free-text fields or simple links, making systematic computational querying and analysis challenging [28].

Computational Tool Ecosystem for AOP Data Extraction and Annotation

To overcome the limitations of free-text data, several computational tools and approaches have been developed to transform AOP-Wiki content into structured, machine-readable, and queryable formats. The following ecosystem of tools enables systematic applicability annotation.

Diagram Title: The AOP-Wiki Computational Ecosystem for Data Analysis

Semantic Conversion to Resource Description Framework (RDF)

A foundational advance is the conversion of the entire AOP-Wiki into a Resource Description Framework (RDF) model [28]. This process transforms wiki content into a network of semantic triples (subject-predicate-object), making the data machine-readable and interoperable.

• Protocol: The conversion uses a Python-based pipeline to parse the AOP-Wiki's XML download. Key entities (AOPs, KEs, KERs, Stressors) are mapped to RDF subjects. Their properties (titles, descriptions, taxonomic applicability) are semantically annotated using over 20 established ontologies, such as the Adverse Outcome Pathway Ontology (AOPO), NCBI Taxon, and Dublin Core [28].
• Outcome: The resulting RDF graph contains over 122,000 triples. Crucially, it adds more than 3,500 links to chemical databases and 7,500 links to gene/protein databases, embedding AOP components in a rich biological and chemical context [28].
• Utility for Applicability: Taxonomic identifiers (e.g., from NCBI Taxon) are explicitly linked to KE and AOP nodes. This allows for precise SPARQL queries to find all KEs applicable to a specific taxon or to identify AOPs where the domain of applicability differs between the MIE and the AO, highlighting potential extrapolation gaps.

Graph-Based Query Engines and Natural Language Interfaces

To simplify access to this complex knowledge, tools like AOPWIKI-EXPLORER have been developed [29]. This tool imports AOP-Wiki data into a Labeled Property Graph (LPG) database (e.g., Neo4j), which is optimized for traversing networks and relationships.

• Functionality: It provides a visual query interface and, innovatively, integrates a Large Language Model (LLM) to accept natural language queries (e.g., "Find AOPs related to liver fibrosis applicable to mammals") [29]. The LLM translates the query into a formal database query.
• Protocol for Applicability Filtering: Users can filter AOP networks based on granular metadata, including taxonomy, life stage, and sex. A workflow involves: (1) using a natural language or form-based query to specify an AO or MIE of interest, (2) applying taxonomic filters to isolate relevant pathways, and (3) visualizing the resulting sub-network to analyze conservation or divergence of KEs across taxa [29].

Data-Driven AOP Network Generation

For specific problem formulations (e.g., endocrine disruption), constructing an AOP Network (AOPN) from related individual AOPs is necessary. A data-driven workflow automates this [27].

• Protocol:
  • Structured Search: Define a list of search terms based on the problem (e.g., "estrogen receptor," "thyroid hormone"). Programmatically query AOP-Wiki page contents via its API or downloaded XML.
  • Relevance Filtering: Manually or semi-automatically curate the initial list, excluding irrelevant AOPs based on title and description.
  • Data Extraction & Processing: Use an R or Python script to download data for the relevant AOP IDs, extracting KE lists and KER linkages.
  • Network Assembly & Filtering: The script processes the data to identify shared KEs (network nodes) and builds the edge list. A final filter can be applied to include only KEs with specific taxonomic annotations (e.g., Homo sapiens).
• Outcome: This automated workflow generates a visual AOPN specific to the problem and desired applicability domain, enabling researchers to see how different MIEs converge on common KEs and AOs within a defined taxonomic context [27].

Experimental Protocols for Enhanced Applicability Annotation

Beyond extracting existing annotations, computational tools enable the enhancement of applicability data through integration with external biological resources.

Protocol: Curated Gene Annotation for Key Events

A major limitation is the sparse and inconsistent annotation of KEs with specific genes or proteins. A protocol combining Natural Language Processing (NLP) and manual curation addresses this [9].

Table: Multi-Step Protocol for Annotating Key Events with Genes and Systems

Step	Action	Tools/Resources	Output
1. Data Retrieval	Download AOP-Wiki data and gene set databases (WikiPathways, KEGG, Reactome, GO, HPO).	AOP-Wiki API, MSigDB [9].	Local database of KEs and gene sets.
2. NLP-Based Matching	Preprocess KE descriptions and gene set names (tokenization, lemmatization). Calculate weighted Jaccard Index to prioritize matches.	Python (nltk, pandas), custom dictionary [9].	Ranked list of top 5 potential gene set matches for each KE.
3. Manual Curation & Gap Filling	Expert evaluation of NLP matches. Add/remove annotations. Search databases for missing annotations for critical KEs.	Manual review, pathway databases [9].	Finalized, accurate list of gene sets (pathways, phenotypes, GO terms) per KE.
4. Biological System Annotation	Expand annotation of the biological context (cell type, tissue, organ) for each KE using controlled vocabularies.	Cell Ontology, Uber-anatomy ontology [28] [9].	Enhanced KE profile with molecular and anatomical context.

• Application to Applicability: The resulting curated gene annotations provide a molecular footprint for each KE. By comparing these gene sets across species (e.g., via orthology databases), one can systematically assess the evolutionary conservation of a KE, providing strong evidence for or against its taxonomic applicability [9].

Diagram Title: Workflow for Curating Gene and Biological System Annotations for Key Events

Protocol: Building Taxonomically Filtered AOP Networks for a Case Study

This protocol applies the data-driven AOPN generation method to a specific problem: mapping AOPs related to Estrogen, Androgen, Thyroid, and Steroidogenesis (EATS) modalities [27].

• Step 1 – Define Search Strategy: Search terms are derived from regulatory guidance documents (e.g., ECHA/EFSA). Example terms: "estrogen receptor agonist," "androgen synthesis inhibition," "thyroid hyperplasia" [27].
• Step 2 – Execute and Curate Search: Perform a full-text search on AOP-Wiki pages via its public interface or API. Initial results are screened by experts to exclude irrelevant pathways.
• Step 3 – Automated Network Generation: Run an R script that takes the list of relevant AOP IDs, fetches their structured data, and constructs a network where shared KEs become connection points.
• Step 4 – Apply Taxonomic Filter: The script filters the network nodes (KEs) to retain only those explicitly annotated with vertebrate (e.g., Vertebrata), mammalian, or human taxonomic identifiers.
• Result: A visual AOPN for EATS-mediated effects, specific to vertebrate or human applicability. This network reveals shared biological targets (e.g., specific nuclear receptors) and common adverse outcomes across different initiating stressors, all within a defined taxonomic domain [27].

Diagram Title: Data-Driven Workflow for Generating a Taxonomically Filtered AOP Network

From Qualitative to Quantitative: Integrating Data for Predictive Application

The ultimate goal of systematic applicability annotation is to support quantitative AOP (qAOP) development for predictive risk assessment [26]. This involves defining quantitative relationships between KEs (e.g., dose-response, time-course) that may be taxon-specific.

• Role of Computational Tools: The structured data from RDF graphs and annotated networks can be integrated with external toxicogenomics and chemical assay data [9]. For example, gene expression signatures from chemical exposures in a human cell line can be mapped to the curated gene sets of KEs in a human-applicable AOPN.
• Workflow for Building a qAOP:
  • Use the AOP-Wiki RDF or a graph query to select a well-supported, human-annotated AOP.
  • For each KE, retrieve its associated genes or pathways from the curated annotation database [9].
  • Integrate with external human in vitro or in silico data to derive quantitative parameters for KERs (e.g., the potency of an MIE required to trigger a specific change in a downstream KE gene signature).
  • The taxonomic specificity of the underlying KE annotations (human genes, human cell types) ensures the qAOP model is grounded in human biology, reducing uncertainties from cross-species extrapolation.

Diagram Title: Logical Progression from Qualitative to Quantitative, Taxonomically Defined AOPs

Table: Research Reagent Solutions and Computational Tools for Systematic Applicability Work

Tool / Resource Name	Type	Primary Function in Applicability Annotation	Access / Reference
AOP-Wiki RDF	Semantic Web Database	Provides a machine-readable, queryable version of the entire AOP-Wiki, enabling complex queries about taxonomic links and KE properties [28].	https://aopwiki.rdf.bigcat-bioinformatics.org [28]
SPARQL Endpoint	Query Language/Interface	Allows users to run custom queries on the AOP-Wiki RDF to extract, for example, all AOPs where the MIE is applicable to fish but the AO is defined for mammals [28].	Linked to the AOP-Wiki RDF site [28]
AOPWIKI-EXPLORER	Graph Database & LLM Interface	Facilitates intuitive exploration and filtering of AOP networks by taxonomic and other metadata using natural language or visual queries [29].	https://github.com/Crispae/AOPWiki_Explorer [29]
Curated KE-Gene Annotation Dataset	Curated Knowledge Base	Provides pre-computed, expert-validated associations between Key Events and specific genes/pathways, forming a molecular basis for assessing evolutionary conservation [9].	Integrated within the described Unified Knowledge Space; methodology published [9].
AOP Network Generation R Script	Computational Workflow	Automates the process of downloading AOP-Wiki data, constructing networks, and filtering based on user-defined criteria, including AOP ID lists and taxonomy [27].	Available as supplementary material in related publications [27].
NCBI Taxon Ontology	Controlled Vocabulary	The standard ontology for providing consistent taxonomic identifiers (e.g., NCBI:txid9606 for Homo sapiens) when annotating KEs and AOPs [28].	https://www.ncbi.nlm.nih.gov/taxonomy
Effectopedia	qAOP Modeling Module	A module of the AOP-KB designed for developing quantitative, computational models of AOPs, into which taxonomically defined parameters can be integrated [26].	https://www.effectopedia.org/ [26]

The systematic annotation of taxonomic applicability within the AOP framework is a critical endeavor for transforming mechanistic toxicology into a predictive, human-relevant, and regulatory-useful science. As demonstrated, leveraging the AOP-Wiki in conjunction with a growing ecosystem of computational tools—including semantic web technologies (RDF), graph databases, natural language interfaces, and curated biological annotations—enables researchers to move beyond manual, narrative reviews. These tools allow for the programmatic extraction, enhancement, and analysis of applicability data, facilitating the construction of fit-for-purpose AOP networks and the development of quantitative models grounded in specific biological domains. By adopting these practices, the toxicology community can strengthen the evidence basis for cross-species extrapolation, directly supporting the thesis that well-annotated AOPs are foundational for credible chemical safety assessment and the advancement of animal-free testing strategies.

The assessment of endocrine-disrupting chemicals (EDCs) represents a critical, yet complex, challenge in regulatory toxicology. EDCs are exogenous substances that interfere with the normal function of the endocrine system, leading to adverse health effects in an intact organism, its progeny, or populations [30]. A significant portion of regulatory focus is on chemicals interacting with the Estrogen, Androgen, Thyroid, and Steroidogenesis (EATS) modalities, due to the relatively advanced mechanistic understanding and availability of standardized tests for these pathways [31]. Historically, safety decisions for both human and ecological health have relied on data from animal toxicity tests [32]. However, there is a strong global regulatory and ethical drive to replace, reduce, and refine (3Rs) vertebrate animal testing [30] [33]. This shift necessitates robust frameworks for extrapolating hazard information across species.

The Adverse Outcome Pathway (AOP) framework is a conceptual construct that portrays existing knowledge linking a Molecular Initiating Event (MIE), such as the interaction of a chemical with a biomolecule, to an Adverse Outcome (AO) relevant to risk assessment, through a series of intermediate Key Events (KEs) and Key Event Relationships (KERs) [32]. AOPs are increasingly recognized as essential tools for supporting a mechanistic, pathway-based approach to toxicology. For EDCs, they can help establish whether an observed adverse effect can be plausibly linked to an endocrine mode of action, potentially reducing the need for animal testing by enabling predictions from in vitro or in silico data [34].

A core principle for leveraging AOPs in next-generation risk assessment is the taxonomic domain of applicability—defining how broadly across species the knowledge within an AOP is applicable based on the conservation of biological structure and function [32]. Extrapolating AOPs across vertebrates, therefore, requires a systematic evaluation of the conservation of MIEs (e.g., hormone receptor binding), KEs (e.g., altered gene transcription, cellular proliferation), and KERs (the causal linkages between events) from model organisms (e.g., zebrafish, rat) to other vertebrates, including humans. This case study analysis delves into the methodologies, data requirements, and challenges of performing such extrapolations for EATS-mediated endocrine disruption, contextualized within the broader research on AOP taxonomic applicability.

Conceptual and Regulatory Foundations

The AOP Framework and Taxonomic Applicability

The AOP framework is designed to break down silos between human and ecological risk assessment by organizing mechanistic knowledge in a way that highlights commonalities across species [32]. The utility of an AOP for cross-species extrapolation hinges on the conservation of the biological pathway. If the MIE and early KEs are structurally and functionally conserved across a taxonomic group (e.g., vertebrates), then data generated in one species can inform hazard identification for others within that group. Conversely, evidence of a lack of conservation can rationally limit the scope of extrapolation [32].

The formal evaluation of an AOP's taxonomic domain of applicability is a critical step in its development and use. It involves assessing evidence for the presence and functional consistency of each KE across different taxa. For EATS pathways, many core components—such as nuclear receptors (ER, AR), the hypothalamic-pituitary-gonadal (HPG) axis, and steroidogenic enzymes—are evolutionarily conserved among vertebrates, providing a strong basis for extrapolation [30]. However, critical differences exist in life-stage sensitivity, metabolism, and compensatory feedback mechanisms, which must be accounted for in quantitative extrapolations.

The Regulatory Landscape Driving the Use of NAMs and AOPs

Globally, regulations are evolving to mandate the reduction of animal testing and to facilitate the use of New Approach Methodologies (NAMs). NAMs is an umbrella term encompassing in silico, in chemico, in vitro, and targeted in vivo assays (like eleutheroembryo stages) that provide mechanistic data [32] [30].

• European Union: The REACH regulation mandates animal testing as a "last resort," and scientific criteria for identifying EDCs under biocides and plant protection product regulations require a mode-of-action analysis for which AOPs are ideally suited [30] [31].
• United States: The EPA has set a goal to eliminate mammalian testing by 2035 and is developing NAM-based alternatives for its Endocrine Disruptor Screening Program (EDSP) [30] [35].
• Canada: Amendments to the Canadian Environmental Protection Act (CEPA) formally recognize the need to replace, reduce, or refine vertebrate animal testing, with a strategy focused on implementing NAMs [33].
• International Coordination: The International Consortium to Advance Cross-Species Extrapolation in Regulation (ICACSER) was established to develop and promote bioinformatic and computational tools for this purpose [32].

The Organisation for Economic Co-operation and Development (OECD) provides a conceptual framework for ED testing that starts with existing data and in silico tools (Level 1), progresses through in vitro and mechanistic assays (Levels 2-3), and culminates in in vivo tests (Levels 4-5) [35]. AOPs serve as the organizing principle for designing Integrated Approaches to Testing and Assessment (IATA) within this framework, helping to target testing and integrate data from diverse NAMs [36] [31].

Table 1: Key Regulatory Frameworks and Their Approach to Endocrine Disruptor Assessment

Region/Program	Key Legislation/Initiative	Approach to ED Assessment	Role for NAMs/AOPs
European Union	REACH, Biocides/PPP Regulations, CLP [30] [31]	Scientific criteria requiring evidence of adverse effect, endocrine mode of action, and causal link.	AOPs central to mode-of-action analysis. In vitro assays accepted. Eleutheroembryo assays (e.g., XETA, EASZY) recognized [30].
United States	Endocrine Disruptor Screening Program (EDSP) [30] [35]	Tiered screening (Tier I) and testing (Tier II) battery.	"Pivot strategy" to replace Tier I assays with NAM batteries (e.g., ER/AR Pathway Models) [30] [35]. AOPs applied for evaluation [36].
Canada	Canadian Environmental Protection Act (CEPA) [33]	Risk assessment based on weight of evidence.	Strategy to identify and implement NAMs to replace, reduce, or refine animal testing under CEPA [33].
International	OECD Conceptual Framework [35]	Tiered framework from non-test data to in vivo studies.	Provides structure for IATA. AOP development is overseen by OECD advisory groups [31].

Methodological Framework for AOP Development and Cross-Species Analysis

Systematic AOP Network Development for EATS Modalities

A robust methodology for identifying and organizing existing AOP knowledge is the first step toward cross-species analysis. A case study on building an AOP network for estrogen-, androgen-, and steroidogenesis-mediated reproductive toxicity provides a clear protocol [31].

Experimental Protocol: Systematic AOP Network Assembly

• Define Key Terms: Derive a comprehensive list of search terms from regulatory guidance documents. This includes parameters for in vitro mechanisms, in vivo mechanistic parameters, EATS-mediated parameters, and parameters sensitive to EATS for mammals and non-mammalian vertebrates [31].
• Screen AOP Wiki: Manually screen all AOPs in the public AOP Wiki repository (https://aopwiki.org/). Each AOP is reviewed against the key terms by independent reviewers, with disagreements resolved by consensus [31].
• Apply Inclusion/Exclusion Criteria:
  • Include AOPs relevant to reproductive toxicity mediated by EAS modalities, describing effects related to key terms, relevant for vertebrate species, and listed in the AOP Wiki [31].
  • Exclude AOPs with no described KEs or KERs, or those deemed irrelevant by expert judgment [31].
• Network Visualization and Analysis: Import relevant AOPs into network analysis software (e.g., Cytoscape). Merge similar KEs and KERs to simplify the network. Analyze the network to identify central "core" KEs and KERs shared across multiple AOPs, which represent critical leverage points for testing and assessment [31].

This process, applied to mammalian reproductive toxicity, identified 42 relevant AOPs and 26 core KEs, 19 of which are already measured in OECD test guidelines [31]. This provides a direct link between AOP knowledge and existing standardized methods.

Integrating Transcriptomics with AOP Networks for Mechanistic Insight

Omics technologies, like transcriptomics, generate rich mechanistic data but linking them to adverse outcomes is challenging. A case study using zebrafish embryos exposed to cadmium or PCB-126 demonstrates a protocol for connecting transcriptomics data to an EATS-focused AOP network [37] [38].

Experimental Protocol: Transcriptomics-AOP Network Integration

• Exposure and RNA Sequencing: Expose zebrafish embryos to the test chemical (e.g., Cd or PCB126) for a defined period (e.g., 4 days). Perform RNA sequencing (RNA-Seq) on the exposed embryos and controls [37].
• Bioinformatic Analysis: Identify differentially expressed genes (DEGs). Perform Gene Ontology (GO) enrichment analysis to find overrepresented biological processes [37].
• AOP Network Mapping: Attempt a data-driven mapping by automatically linking standardized GO terms to Key Event titles in the AOP network. This often yields limited results due to terminology mismatches [37] [38].
• Expert-Driven Mapping: Manually map the enriched GO terms and pathways (supported by tools like Ingenuity Pathway Analysis) to relevant KEs in the AOP network based on biological knowledge. This typically reveals many more plausible connections [37] [38].
• Interpretation: Use the mapping to identify which MIEs and KEs in the EATS network are potentially perturbed by the chemical, forming a hypothesized mode of action that can be tested with targeted assays.

This study highlighted that while transcriptomics is powerful for revealing activity, a quantitative understanding of KERs is needed to infer adversity from molecular data alone [38].

Quantitative KER Quantification for Predictive Toxicology

A key frontier in AOP science is the development of quantitative AOPs (qAOPs) that define dose-response and time-course relationships for KERs. A protocol was developed to quantify the KER between "Decreased circulating testosterone" and "Decreased sperm count" [38].

Experimental Protocol: Literature-Based KER Quantification

• Structured Literature Search: Execute a systematic search of published studies reporting data on both circulating testosterone and sperm count in the same experiment [38].
• Reliability Assessment & Data Extraction: Assess the reliability of each study using a predefined tool. Extract data on species, strain, exposure, testosterone measures, and sperm count measures in a standardized format [38].
• Statistical Modeling: Develop a statistical model (e.g., a linear mixed-effects model) to describe the relationship between the change in testosterone and the change in sperm count, accounting for inter-species and inter-study variability [38].
• Application: The resulting quantitative model can be used to predict a decrease in sperm count (an adverse effect) based on a measured decrease in testosterone from a shorter-term in vivo study or even an in vitro steroidogenesis assay coupled with a pharmacokinetic model [38].

This approach directly supports the replacement of long-term, high-animal-burden reproductive studies with mechanistic, lower-burden tests.

In Silico Protocol for Endocrine Activity Profiling

Computational tools are essential for prioritizing chemicals for testing. A structured in silico protocol for assessing activity across EATS modalities integrates (Q)SAR predictions, read-across from similar chemicals, and existing experimental data within a weight-of-evidence framework [35].

Experimental Protocol: In Silico Endocrine Activity Assessment

• Define the Assessment Question: Specify the endpoint (e.g., ER agonist activity) and the chemical structure.
• Gather Existing Data: Collect any relevant in vitro (e.g., ToxCast ER Model AUC score) or in vivo (e.g., uterotrophic assay) data [35].
• Perform (Q)SAR Predictions: Run a battery of curated (Q)SAR models for the relevant endpoint. Evaluate the reliability and relevance of each prediction [35].
• Conduct Read-Across Analysis: Identify structurally similar compounds with reliable experimental data. Justify the analogue selection and predict the target chemical's activity [35].
• Integrate Evidence & Assign Confidence: Weigh all lines of evidence using a predefined scheme (e.g., the Hazard Assessment Framework). Assign an overall activity call (active/inactive) and a confidence level (low, medium, high) based on the consistency and quality of the evidence [35].

This protocol was demonstrated in a case study where metabolic uncertainty limited confidence in a negative prediction, highlighting that in silico assessments often trigger the need for targeted in vitro metabolic stability assays [35].

Case Studies in Cross-Species AOP Extrapolation

Case Study 1: PFOS and Thyroid-Mediated Neurodevelopmental Toxicity

This study compared a traditional ED assessment of perfluorooctane sulfonic acid (PFOS) with a mechanism-based assessment using an AOP network [38].

• Traditional Assessment: Concluded PFOS fulfills the scientific criteria as an ED for thyroid disruption leading to developmental neurotoxicity, based on in vivo data showing reduced thyroid hormone, altered brain development, and behavioral deficits [38].
• Mechanism-Based Assessment: Using a thyroid-related AOP network and only NAM data (in vitro assays, omics, literature), researchers could identify endocrine activity (e.g., binding to transport proteins, cellular effects) but could not conclusively demonstrate endocrine-mediated adversity. The critical gap was the lack of quantitative KERs linking the measured molecular changes to the adverse neurodevelopmental outcome in the absence of apical in vivo data [38].
• Implication for Extrapolation: This case underscores that while AOP networks organize knowledge and identify activity, quantitative, predictive understanding of KERs is essential for cross-species extrapolation that replaces apical endpoint studies. It also shows that well-established in vivo-based AOPs (like thyroid disruption leading to neurodevelopmental effects) provide the causal framework onto which NAM data can be mapped for new chemicals.

Case Study 2: Integrated Assessment of Tartrazine Using AOPs

A study on the food dye tartrazine employed an integrated strategy aligning in silico, in vitro, and in vivo evidence with AOPs [36].

• Method: In silico docking predicted binding to multiple nuclear receptors (AR, ERα, TR, etc.). ToxCast data showed activity for several of these targets. These predictions were mapped to existing AOPs for estrogen, androgen, and thyroid pathways. The in silico and in vitro evidence was then weighed alongside available in vivo studies within the OECD framework [36].
• Finding: The integrated, AOP-informed assessment provided a more comprehensive and mechanistic picture of tartrazine's potential endocrine activities than any single study, highlighting its multi-modal potential. It successfully organized fragmented data from different species and test systems into coherent hypothesized pathways [36].
• Implication for Extrapolation: This demonstrates the practical use of AOPs as an organizing framework for IATA. By mapping chemical-specific data to conserved KEs (e.g., "Antagonism of thyroid hormone receptor"), extrapolation from in silico predictions or fish assays to potential human health outcomes becomes more structured and transparent.

Table 2: Factors Influencing Taxonomic Applicability of EATS AOPs

Biological Factor	Impact on AOP Extrapolation	Consideration for EATS Modalities
Conservation of MIE Target	High conservation supports extrapolation.	Nuclear receptors (ER, AR, TR) are highly conserved in sequence and function across vertebrates [30].
Toxicokinetics (ADME)	Differences in absorption, metabolism, and excretion can drastically alter internal dose and active metabolite profile.	Steroidogenesis pathways and hepatic metabolism enzymes (CYPs) can vary, affecting chemical activation/deactivation [32] [35].
Life Stage & Development	Sensitivity to EDCs is often life-stage specific. Developmental windows may not align across species.	Frog metamorphosis (T-screen) is a uniquely sensitive thyroid endpoint. Mammalian prenatal/perinatal stages are critical [30].
System Feedback & Redundancy	Compensatory mechanisms can buffer perturbations, varying in strength across species.	The HPG and HPT axes have robust feedback; effects on hormone levels may not linearly translate to organ toxicity in all species [38].
Presence of Alternative Pathways	A biological function may be regulated by different mechanisms in different taxa.	Some fish species have additional estrogen receptor subtypes (e.g., ERβ2) not present in mammals [30].

The Scientist's Toolkit: Key Reagents and Methods

Table 3: Research Reagent Solutions for AOP-Based ED Research

Tool/Reagent	Category	Primary Function in AOP Research
AOP Wiki (aopwiki.org)	Knowledge Repository	Central, publicly accessible database for published AOPs, KEs, and KERs. Essential for network assembly and literature review [31].
Cytoscape (with AOPWiki plugin)	Network Analysis Software	Visualizes and analyzes complex AOP networks, identifies central KEs, and facilitates mapping of experimental data [37] [31].
OECD Validated Test Guidelines	Standardized Assays	Provide reliable, reproducible methods for generating data on specific KEs (e.g., TG 455 for ER transactivation, TG 231 for fish embryo toxicity) [30] [31].
ToxCast/Tox21 Assay Battery	High-Throughput In Vitro Screening	Provides bioactivity profiles across hundreds of pathways. Data (e.g., ER/AR AUC scores) can be directly mapped to MIEs and early KEs [35] [36].
Zebrafish (Danio rerio) Embryo	Eleutheroembryo Model	A key in vivo NAM for screening developmental toxicity and endocrine activity (e.g., EASZY assay). Bridges in vitro and apical in vivo outcomes [37] [30].
Gene Ontology (GO) Databases	Bioinformatics Resource	Provides standardized terms for biological processes, molecular functions, and cellular components. Crucial for interpreting transcriptomics data and linking to AOP KEs [37].
Stable Cell Lines with Reporter Genes	In Vitro Reagent	Engineered cells (e.g., ERα CALUX, AR-EcoScreen) provide specific, sensitive readouts for receptor activation, a core MIE for EATS AOPs [30] [35].
qPCR Assays / RNA-Seq	Molecular Profiling	Measures changes in gene expression, a common KE downstream of receptor activation. Essential for building evidence for KERs and chemical-specific MoA [37] [38].
Molecular Docking Software (e.g., CB-Dock2, AutoDock Vina)	In Silico Tool	Predicts the binding affinity and mode of a chemical to a protein target (e.g., nuclear receptor), informing potential MIEs for prioritization [36].
Reference EDCs (e.g., 17α-ethinylestradiol, flutamide)	Chemical Standards	Positive control substances with well-characterized EATS activities. Critical for assay validation and as benchmarks for comparative analysis [30].

Extrapolating AOPs for EATS-mediated endocrine disruption across vertebrates is a viable and scientifically rigorous strategy that is central to the modern, 3Rs-aligned paradigm in toxicology. The case studies and methodologies reviewed demonstrate that the foundational conservation of EATS pathways provides a solid basis for qualitative extrapolation. The AOP framework successfully organizes fragmented data, supports the integration of NAMs into IATA, and focuses testing on the most informative KEs.

However, key challenges must be addressed to move from qualitative hazard identification to quantitative risk assessment across species:

• Quantifying KERs: There is a pressing need to develop more qAOPs with statistical models for critical KERs, as demonstrated for the testosterone-sperm count relationship [38].
• Accounting for Toxicokinetics: Cross-species extrapolation requires integrating in vitro bioactivity data with Physiologically Based Kinetic (PBK) models to predict target tissue doses in different species.
• Refining Taxonomic Domains: More empirical data on KE conservation and response in non-model vertebrates (e.g., birds, reptiles) is needed to define applicability domains more precisely.
• Harmonizing Ontologies: Improved alignment between AOP KE descriptions and standardized bioinformatics vocabularies (like GO terms) is required to enable automated, data-driven mapping of omics data [37].

The trajectory is clear: future ED assessment will be increasingly mechanism-based, relying on AOP networks to integrate data from batteries of in silico and in vitro NAMs. Success depends on continued collaboration within initiatives like ICACSER [32] to build the required toolkit of qAOPs, PBK models, and bioinformatic pipelines, ultimately enabling reliable predictions of adversity across vertebrates while fulfilling the mandate to reduce animal testing.

The Adverse Outcome Pathway (AOP) framework has emerged as a pivotal construct for organizing mechanistic toxicological knowledge, linking a Molecular Initiating Event (MIE) through a causally connected sequence of Key Events (KEs) to an Adverse Outcome (AO) relevant for risk assessment [8]. While individual AOPs serve as pragmatic units for development, they represent a simplification of biological complexity [39]. In reality, exposures to single or multiple stressors can engage multiple MIEs, leading to interconnected pathways that converge, diverge, and interact [40]. This complexity necessitates a shift from viewing toxicity as a linear pathway to understanding it as a network phenomenon.

An AOP Network (AOPN) is formally defined as an assembly of two or more AOPs that share one or more KEs [39]. These networks are recognized as the most probable units of prediction for real-world scenarios, offering a more accurate representation of pleiotropic and interactive effects [40]. Concurrently, the accurate prediction of chemical effects across species—a cornerstone of ecological risk assessment and translational toxicology—hinges on a clear understanding of the Taxonomic Domain of Applicability (tDOA). The tDOA defines the taxonomic range across which an AOP's KEs and KE Relationships (KERs) are biologically plausible [3].

This whitepaper synthesizes the convergent evolution of these two concepts: the development of complex AOP networks and the rigorous, bioinformatics-driven definition of their taxonomic applicability. We posit that the future of predictive toxicology lies in building complex, taxonomically informed AOP networks. These integrated frameworks are essential for advancing Integrated Approaches to Testing and Assessment (IATAs), improving the accuracy of chemical safety decisions, and reducing reliance on animal testing by ensuring New Approach Methodologies (NAMs) are anchored in conserved biology [41] [8].

Foundational Concepts: From Linear AOPs to Dynamic Networks

Core Components and Network Formation

An AOP is composed of modular units: Measurable Key Events (KEs) and the causal or correlative Key Event Relationships (KERs) that link them [8]. Networks emerge naturally when independently developed AOPs share common KEs (e.g., a shared MIE leading to different AOs, or distinct MIEs converging on a common intermediate KE) [39]. This shared-module architecture allows complex networks to be built from a repository of validated AOP building blocks.

Table 1: Comparative Attributes of Linear AOPs vs. AOP Networks

Attribute	Linear AOP	AOP Network (AOPN)
Representation	Single sequence from one MIE to one AO	Multiple intersecting and/or parallel pathways
Predictive Scope	One stressor, one primary outcome	Multiple stressors, interacting mechanisms, and pleiotropic outcomes
Biological Fidelity	Simplified model	Captures pathway crosstalk, feedback loops, and compensatory mechanisms
Regulatory Utility	Identifies assays for a specific pathway	Supports IATA development for complex endpoints like cholestasis or endocrine disruption [41] [42]
Taxonomic Consideration	Often developed for one model species	Can explicitly map taxonomic applicability across multiple nodes and pathways [3]

The Critical Role of Taxonomic Domain of Applicability (tDOA)

The tDOA is frequently narrowly defined in initial AOP development, based on the species used in the underlying empirical studies [3]. Confidently extrapolating an AOP beyond its initial tDOA requires evidence of structural and functional conservation of the essential biological elements (e.g., proteins, receptors) across species [3]. Explicitly defining the tDOA for each KE and KER within a network is critical for its reliable application in environmental and translational toxicology, ensuring predictions are biologically plausible for the species of concern.

Methodological Framework: Constructing and Informing AOP Networks

Strategies for AOP Network Development

Two primary strategies exist for AOPN development [39] [24]:

• Network-Guided Development: Intentionally developing new AOPs with shared KEs to build a network for a specific toxicological endpoint (e.g., chemical-induced cholestasis) [42].
• AOP Network Derivation: Programmatically or manually extracting and linking relevant existing AOPs from shared knowledge bases like the AOP-Wiki to address a specific problem formulation [24].

A data-driven derivation workflow, as demonstrated for Endocrine Disruptor (EATS modalities) AOPNs, involves [24]: * Structured Searching: Using controlled vocabularies and problem-formulated terms to query the AOP-Wiki. * Automated Data Extraction & Processing: Employing computational scripts (e.g., in R) to extract AOP, KE, and KER data. * Network Assembly & Filtering: Using graph-based tools (e.g., Cytoscape) to visualize the network, with filtering based on tDOA, sex, or life stage to tailor the network.

Diagram 1: Data-driven workflow for AOP network derivation (65 chars)

Bioinformatics for Defining Taxonomic Applicability

The Sequence Alignment to Predict Across Species Susceptibility (SeqAPASS) tool is a cornerstone bioinformatics approach for expanding tDOA evidence [3]. It provides lines of evidence for structural conservation through a three-tiered analysis:

• Level 1: Evaluates primary amino acid sequence similarity to identify potential orthologs.
• Level 2: Assesses conservation of known functional domains.
• Level 3: Compares specific amino acid residues critical for protein-ligand interaction or function.

Table 2: SeqAPASS Analysis for an AOP Case Study (nAChR Activation) [3]

Query Protein	Primary Role in AOP	SeqAPASS Level 1 (Sequence)	SeqAPASS Level 2 (Domain)	SeqAPASS Level 3 (Residue)	Inferred Taxonomic Breadth
Nicotinic acetylcholine receptor subunit alpha	Molecular Initiating Event (Target)	High similarity across insects	Functional ligand-binding domain conserved	Critical binding residues conserved in Hymenoptera	Plausible across bee families
Acetylcholinesterase	Compensatory KE	Orthologs identified broadly	Catalytic domain highly conserved	Active site serine conserved	Very wide (arthropods to vertebrates)
GABA-gated chloride channel	Linked KE (Crosstalk)	Moderate similarity	Neurotransmitter-gated ion channel domain present	Key residues variable	More restricted; requires empirical confirmation

Diagram 2: Expanding tDOA for a bee AOP using SeqAPASS (48 chars)

Advanced Integration: AI-Assisted Network Optimization

Emerging methodologies leverage artificial intelligence to systematize the curation and confidence assessment of complex AOPNs. A protocol for AI-assisted AOP network optimization involves [42]:

• Automated Literature Mining: Using platforms like Sysrev with predefined queries to continuously collect new relevant data for the network's KEs and KERs.
• Quantitative Confidence Scoring: Applying tailored Bradford-Hill criteria (Biological Plausibility, Empirical Evidence, Essentiality) to score each KER. Scores from individual criteria are integrated into a Total KER Confidence value.
• Dynamic Visualization: The optimized network is visualized with node (KE) size weighted by incidence in literature and edge (KER) thickness weighted by total confidence, creating an intuitive "mechanistic compass."

Table 3: Research Reagent Solutions for AOP Network Development

Tool/Resource	Type	Primary Function in AOPN Research	Example/Reference
AOP-Wiki	Knowledgebase	Central repository for accessing, sharing, and developing modular AOP components for network assembly.	aopwiki.org [8] [24]
SeqAPASS	Bioinformatics Tool	Provides evidence for structural conservation of proteins across species to define and expand tDOA.	US EPA Web Tool [3]
Cytoscape	Network Analysis Software	Visualizes, analyzes, and filters complex AOP networks; integrates with attribute data (e.g., confidence scores).	Open-source platform [42] [24]
Sysrev	AI-Assisted Data Platform	Facilitates systematic literature review and data extraction for updating and optimizing AOP networks.	sysrev.com [42]
GENOMARK / TGx-DDI	Transcriptomic Biomarker	NAMs that provide mechanistic data on genotoxicity for populating and validating KEs in relevant AOPNs.	Biomarker assays [41]
MultiFlow / ToxTracker	In Vitro Assay	Provides high-content data on specific DNA damage pathways for informing KE states in genotoxicity AOPNs.	Commercial assay kits [41]
OECD AOP Portfolio	Curated Guidance	Source of OECD-endorsed AOPs that provide high-confidence building blocks for network development.	OECD website [8]

Case Study in Integration: Genotoxicity Assessment via a Global AOP Network

Current genotoxicity testing faces challenges like misleading positives and limited mechanistic insight [41]. A proposed global AOP network for permanent DNA damage demonstrates the integrated application of these concepts [41]:

• Network Structure: Multiple MIEs (e.g., DNA adduct formation, topoisomerase inhibition, cross-linking) converge on shared intermediate KEs (e.g., DNA strand breaks, mutation). These pathways then diverge to various AOs (e.g., cancer, cell death, heritable mutation).
• Informing IATA: The network maps where specific NAMs, like ToxTracker (reporting on specific cellular stress pathways) or transcriptomic biomarkers (GENOMARK), can reliably measure critical KEs [41].
• Taxonomic Consideration: The tDOA for KEs like "Chromosomal Aberrations" is broad due to high conservation of DNA repair machinery, while the tDOA for the AO "Hepatocellular Carcinoma" may be restricted to susceptible species. This informs appropriate model selection.

This network approach moves beyond a battery of standalone tests to a mechanistically organized testing strategy, where results from NAMs are interpreted within their pathway context, improving prediction accuracy and reducing false positives [41].

Analytical Approaches for AOP Network Interrogation

Once constructed, AOPNs can be analyzed using concepts from graph theory to extract meaningful insights [40].

• Topological Analysis: Identifies network properties like highly connected "hub" KEs. A hub KE represents a critical point of pathway convergence, indicating a potential high-leverage target for assay development or risk mitigation.
• Critical Path Identification: Determines the most significant route(s) through a network based on criteria like strength of KER evidence, essentiality, or taxonomic relevance. This helps prioritize testing and research.
• Interaction Analysis: Qualitatively assesses where pathways may interact (e.g., at shared KEs), suggesting potential for additive, synergistic, or antagonistic effects from chemical mixtures [40].

The evolution from linear AOPs to taxonomically informed AOP networks represents a paradigm shift towards more predictive and mechanistic toxicology. The synergy is powerful: bioinformatics tools like SeqAPASS provide the evidentiary basis to confidently extend the tDOA of network components, while the network framework offers a realistic structure for applying this taxonomic knowledge to complex biological systems. This integrated approach directly supports the development of scientifically robust IATAs and Defined Approaches (DAs) that use NAMs efficiently, ultimately enhancing regulatory decision-making for environmental and human health while reducing reliance on animal testing [41] [8]. Future work must focus on standardizing methodologies for network derivation, expanding quantitative (q)AOP modeling within networks, and further integrating computational tDOA analysis into the AOP knowledgebase workflow.

Integrating Omics and Gene Annotation Data to Support and Refine Applicability Domains

The Adverse Outcome Pathway (AOP) framework has emerged as a central paradigm in modern toxicology and biomedical research, organizing mechanistic knowledge as a causally linked sequence of events from a Molecular Initiating Event (MIE) to an Adverse Outcome (AO) relevant to risk assessment [9]. A critical, yet often inadequately defined, component of an AOP is its Taxonomic Domain of Applicability (tDOA)—the range of species for which the pathway is biologically plausible [3]. Traditionally, tDOAs are narrowly defined based on the specific species used in empirical studies underpinning the AOP, limiting confidence in extrapolations essential for protecting human health and ecological systems [3] [43].

Concurrently, omics technologies (genomics, transcriptomics, proteomics, metabolomics) generate high-dimensional data that can reveal the molecular mechanisms of chemical exposures and disease [44] [45]. However, a significant gap exists between the rich detail of omics datasets and the structured, mechanistic knowledge captured in AOPs [9]. Integrating these domains is not merely additive; it is transformative. Systematically annotating AOP key events with gene sets and leveraging multi-omics data provide a robust, evidence-based methodology to refine and expand the tDOA. This integration moves beyond assumptions of taxonomic relatedness, offering a mechanistic foundation for defining the boundaries of AOP applicability and enhancing the use of AOPs in Next Generation Risk Assessment (NGRA) and New Approach Methodologies (NAMs) [46] [43].

Theoretical Foundation: From Empirical tDOA to Biologically Plausible tDOA

The tDOA of an AOP is established by evaluating the conservation of its components—the MIE, Key Events (KEs), and Key Event Relationships (KERs)—across species [3]. Two primary lines of evidence are considered:

• Structural Conservation: The presence and similarity of biological entities (e.g., genes, proteins, receptors).
• Functional Conservation: The preservation of the biological role or activity of those entities [3].

An AOP’s empirical tDOA is restricted to species for which direct experimental evidence exists. The goal of integration with omics and bioinformatics is to define a biologically plausible tDOA, which extrapolates applicability to a broader taxonomic space based on evidence of conserved biology [3]. This process is illustrated in the following conceptual workflow.

Methodological Framework: Integrating Omics and Annotating AOPs

Curated Annotation of Key Events to Gene Sets

A foundational step is linking the biological events described in AOPs to measurable molecular entities. This involves a systematic, multi-step curation process to annotate KEs with relevant gene sets (e.g., pathways, Gene Ontology terms) [9].

• Data Integration: AOP knowledge is extracted from the AOP-Wiki and integrated into a structured knowledge graph with gene sets from sources like Reactome, KEGG, and Gene Ontology [9].
• Computational Matching: Natural Language Processing (NLP) techniques pre-process KE descriptions and gene set names. A weighted Jaccard Index prioritizes matches based on the specificity of shared terms [9].
• Manual Curation and Refinement: Computational matches are manually evaluated by domain experts for biological accuracy and context. Irrelevant matches are removed, and gaps are filled by searching curated databases to ensure robust, meaningful annotations [9].

This process translates qualitative KE descriptions into quantifiable molecular signatures, enabling the interrogation of omics data for pathway-specific perturbations.

Bioinformatics Tools for Assessing Cross-Species Conservation

With KEs annotated to gene sets, bioinformatics tools assess the conservation of these molecular components across taxa.

• SeqAPASS (Sequence Alignment to Predict Across Species Susceptibility): This hierarchical tool evaluates protein conservation at three levels:
  • Primary Sequence Similarity: Identifies potential orthologs.
  • Functional Domain Conservation: Assesses conservation of known protein domains.
  • Critical Residue Conservation: Evaluates specific amino acids essential for chemical binding or protein function [3] [43].
• G2P-SCAN (Genes to Pathways - Species Conservation Analysis): This tool maps human gene sets to biological pathways (e.g., Reactome) and estimates the conservation of those entire pathways across a core set of model species (human, mouse, rat, zebrafish, fruit fly, worm) [43].

The combination of SeqAPASS (protein-centric) and G2P-SCAN (pathway-centric) provides complementary lines of evidence for structural and functional conservation, strengthening the weight of evidence for tDOA refinement [43].

Multi-Omics Data Integration Strategies

Omics data integration is crucial for moving from in silico predictions of conservation to empirical evidence of pathway perturbation. Several computational strategies are employed [44]:

• Conceptual Integration: Using shared biological concepts (e.g., gene names, pathway maps) to link disparate omics datasets.
• Statistical Integration: Applying multivariate analyses to identify co-expressed genes or correlated biomolecules across omics layers.
• Model-Based Integration: Employing network or pharmacokinetic models to simulate system behavior from multi-omics inputs.
• Pathway/Network Integration: Overlaying omics data onto biological pathways or interaction networks to visualize concerted molecular changes [44].

Table 1: Strategies for Multi-Omics Data Integration in AOP Context

Integration Approach	Core Methodology	Application in AOP/tDOA Refinement	Key Consideration
Conceptual	Linking data via shared entities (genes, pathways) using knowledge bases.	Annotating KEs; identifying conserved pathways across species.	Relies on existing, curated knowledge; may miss novel relationships.
Statistical	Multivariate analysis (correlation, clustering) of quantitative omics measures.	Identifying biomarker co-expression patterns that support a KER in new species.	Reveals associations but not necessarily causal relationships.
Model-Based	Mathematical modeling (e.g., PBPK, network inference) of system dynamics.	Predicting kinetic relationships between KEs; extrapolating dose-response.	Requires significant prior knowledge and validation.
Pathway/Network	Projecting data onto pathway maps or protein-protein interaction networks.	Visualizing multi-omics perturbation of an entire AOP; assessing pathway-level conservation.	Provides intuitive, mechanism-rich visualization of complex data.

The technical pipeline from raw data to tDOA refinement is summarized below.

Application and Validation: Case Studies and Quantitative Outcomes

Case Study: Refining tDOA for an AOP Involving Nicotinic Acetylcholine Receptors

AOP 89 (linking activation of the nicotinic acetylcholine receptor, nAChR, to colony death/failure in honey bees, Apis mellifera) was used to demonstrate tDOA refinement [3].

• Method: Nine proteins central to the AOP were used as queries in SeqAPASS. Analysis spanned three levels: primary sequence (Level 1), functional domain (Level 2), and key ligand-binding residues (Level 3) [3].
• Quantitative Outcome: The analysis provided granular conservation scores across insects. For example, while nAChR subunits were broadly conserved among bees, critical residues known to confer sensitivity to neonicotinoid insecticides showed more variable conservation, suggesting differential species susceptibility within the taxonomic domain [3].
• Impact: This bioinformatics evidence expanded the biologically plausible tDOA from primarily A. mellifera to other Apis and non-Apis bees, while also identifying where functional divergence might limit applicability [3].

Case Study: Combined Tool Approach for Pathway Conservation

A study combined SeqAPASS and G2P-SCAN to evaluate cross-species susceptibility for chemicals targeting PPARα, ESR1, and GABRA1 [43].

• Method: Molecular targets were identified via ToxCast and literature. G2P-SCAN mapped targets to Reactome pathways, and SeqAPASS evaluated the conservation of those pathway components.
• Quantitative Outcome: The integrated analysis showed that while a target protein (e.g., PPARα) might be conserved, key downstream pathways could be incomplete or divergent in some species. This pathway-centric view provided a more nuanced prediction of functional susceptibility than protein sequence analysis alone [43].
• Impact: The consensus data from both tools strengthened the weight of evidence for predicting chemical effects and helped define the tDOA for related AOPs [43].

Table 2: Quantitative Outcomes from Integrated Omics-AOP Analyses

Case Study Focus	Core Method	Key Quantitative Metric	Outcome for tDOA
nAChR AOP in Bees [3]	SeqAPASS protein conservation analysis.	Percent identity/similarity at primary sequence, domain, and residue levels.	Expanded plausible tDOA to non-Apis bees; identified specific residues for differential risk assessment.
Pathway Conservation for Chemical Targets [43]	G2P-SCAN + SeqAPASS.	Pathway completeness score and protein conservation score across 7 model species.	Provided nuanced prediction of functional susceptibility, refining tDOA boundaries based on pathway conservation.
Human Health AOP Annotation [9]	NLP + manual curation of KE-gene sets.	Number of KEs annotated; accuracy/recall of gene set matches.	Created a foundational resource linking >300 KEs to molecular data, enabling omics-based tDOA interrogation.

The Scientist's Toolkit: Essential Research Reagent Solutions

Table 3: Key Reagents, Tools, and Databases for Integrating Omics and AOPs

Tool/Resource Name	Type	Primary Function in tDOA Refinement	Access/Example
SeqAPASS	Bioinformatics Web Tool	Evaluates conservation of protein sequences, domains, and critical residues across species to infer structural susceptibility [3] [43].	https://seqapass.epa.gov
AOP-Wiki	Knowledge Repository	Central repository for AOPs, KEs, and KERs; provides initial empirical tDOA and biological context [3] [9].	https://aopwiki.org
Reactome / KEGG	Pathway Database	Provides curated biological pathways for annotating KEs and interpreting omics data in a pathway context [9] [43].	https://reactome.org; https://www.genome.jp/kegg/
Unified Knowledge Space (UKS)	Custom Knowledge Graph	Integrates AOP data with gene sets and ontologies to facilitate computational annotation and analysis [9].	(Requires custom implementation as described in [9])
G2P-SCAN Tool	Computational Pipeline	Maps human gene sets to pathways and estimates pathway conservation across core model species [43].	Tool described by Rivetti et al. (2023) [43]
RefChemDB / CompTox Dashboard	Chemical-Biological Database	Identifies molecular targets and bioactivity data for chemicals to inform MIE and KE annotation [43].	U.S. EPA CompTox Chemicals Dashboard
RNA-Seq / Mass Spectrometry Platforms	Omics Technologies	Generate transcriptomic and proteomic data to empirically test for perturbation of annotated KE gene sets in novel species.	Illumina, Nanopore (Genomics); LC-MS/MS (Proteomics)

The integration of omics and gene annotation data is fundamentally advancing the AOP framework from a qualitative to a quantitative, predictive science. Future directions include:

• Automated, Real-time Refinement: Leveraging Artificial Intelligence and machine learning to dynamically update tDOAs as new omics data and genome sequences become available [46] [47].
• Incorporating Epigenomics: Integrating epigenomic markers (e.g., DNA methylation) to understand how environmental exposures cause persistent, heritable changes that may affect susceptibility across life stages and generations, further refining applicability domains [45].
• Quantitative AOP (qAOP) Development: Using multi-omics data to parameterize computational models that describe the quantitative relationships between KEs, enabling precise prediction of effect thresholds across species [46] [43].
• Bridging to Higher-Order Outcomes: Extending the pathway concept beyond biological outcomes to include socio-economic impacts (e.g., Cost Outcome Pathways), where a refined tDOA is crucial for accurate economic burden estimation and policy decisions [48].

In conclusion, the systematic integration of omics data with curated gene annotations provides a powerful, evidence-based mechanism to support, refine, and justify the taxonomic applicability domains of AOPs. This convergence addresses a critical limitation in mechanistically based risk assessment, enabling more confident extrapolation across species, improving the development of NAMs, and ultimately supporting the protection of both human and ecological health.

Overcoming Common Pitfalls in AOP Taxonomic Extrapolation: Challenges and Optimization Tactics

Identifying and Addressing Knowledge Gaps in Cross-Species Conservation of Key Events

The Adverse Outcome Pathway (AOP) framework has emerged as a pivotal tool for organizing mechanistic toxicological knowledge, describing a sequential chain of causally linked events from a Molecular Initiating Event (MIE) to an Adverse Outcome (AO) at the organism or population level [8] [1]. This framework is chemically agnostic, focusing on the biological pathway perturbation itself. A critical, yet often inadequately defined, aspect of an AOP is its Taxonomic Domain of Applicability (tDOA)—the range of species for which the pathway is biologically plausible and operative [3].

Defining the tDOA is not merely an academic exercise; it is fundamental to the confident application of AOPs in regulatory decision-making and predictive toxicology. It directly informs the use of surrogate species in chemical safety assessments and enables the extrapolation of effects from data-rich to data-poor species [3]. Currently, the tDOA for most developed AOPs is narrowly and implicitly defined, limited to the specific species used in the foundational empirical studies [3]. This creates significant knowledge gaps regarding the conservation of Key Events (KEs) and their relationships (KERs) across the tree of life. Filling these gaps is essential for leveraging AOPs to protect ecosystem biodiversity and for translating findings from model organisms to human health assessments within a broader thesis on adverse outcome pathway taxonomic applicability. This guide outlines a systematic, evidence-driven approach to identify, evaluate, and address these cross-species conservation gaps.

Theoretical Foundation: Structural and Functional Conservation

The assessment of a pathway's tDOA rests on evaluating two core elements: structural conservation and functional conservation [3].

• Structural Conservation asks whether the biological entities (e.g., specific proteins, receptors, genes, organs) that comprise the KEs are present and measurable in the taxon of interest. Evidence includes the presence of orthologous gene sequences, conserved protein domains, and critical amino acid residues necessary for interactions [3].
• Functional Conservation asks whether these conserved structures perform the same biological role within the pathway in the taxon of interest. Evidence typically requires empirical data showing that perturbation leads to the expected downstream key event [3].

A hierarchical approach to evaluating structural conservation provides a scaffold for investigation. The Sequence Alignment to Predict Across Species Susceptibility (SeqAPASS) tool formalizes this into a multi-level bioinformatics workflow [3]:

• Level 1: Primary Sequence Similarity. Compares full-length protein sequences to identify orthologs.
• Level 2: Functional Domain Conservation. Evaluates the preservation of known functional domains and motifs.
• Level 3: Critical Residue Conservation. Assesses the conservation of specific amino acid residues known to be essential for protein-ligand binding, protein-protein interaction, or catalytic function.

Confidence in tDOA expands progressively with evidence gathered across these levels, which can then be integrated with functional data from tailored in vitro or in vivo assays.

Methodologies for Identifying and Evaluating Conservation Gaps

Computational Bioinformatics Workflow

The first line of evidence for expanding tDOA comes from in silico analysis. The workflow below details the protocol using SeqAPASS as a exemplar tool [3].

Experimental Protocol: SeqAPASS Analysis for tDOA Assessment

• AOP and Protein Identification: Select the AOP of interest and identify the specific protein(s) associated with each Molecular Initiating Event and Key Event (e.g., nicotinic acetylcholine receptor for an AOP on neuronal signaling) [3].
• Query Sequence Submission: Obtain the reference protein sequence (e.g., from Apis mellifera for a pollinator AOP) from a curated database like UniProt. Submit this as the query sequence to the SeqAPASS web tool.
• Level 1 Analysis: Run the Level 1 (primary sequence) comparison against the selected taxonomic database (e.g., invertebrates). Identify potential orthologs based on sequence similarity thresholds (e.g., ≥70-80% identity). Generate a list of species with putative orthologs.
• Level 2 Analysis: For species passing Level 1, perform Level 2 analysis focused on known functional domains (e.g., ligand-binding domains of the nAChR). Evaluate the percent similarity of these specific domains.
• Level 3 Analysis: For high-priority species, conduct Level 3 analysis. Input the known critical residues (e.g., amino acids critical for neonicotinoid binding in nAChR) and evaluate their absolute conservation across species.
• Data Synthesis: Compile results into an evidence table. Species demonstrating conservation at all three levels provide strong structural evidence for inclusion in the tDOA for that specific KE.

Empirical Assays for Functional Validation

Computational predictions of structural conservation must be paired with empirical tests of function. A tiered testing strategy optimizes resources.

Experimental Protocol: Tiered In Vitro Functional Assay

• Tier 1: High-Throughput Screening (HTS): Use established quantitative High-Throughput Screening (qHTS) platforms, such as those from the Tox21 consortium, to test chemical perturbations across a panel of orthologous proteins expressed in standardized cell lines [49] [50]. Assay endpoints should measure the immediate KE (e.g., receptor activation, enzyme inhibition).
• Tier 2: Mechanistic In Vitro Models: For species/proteins of high interest or uncertainty, employ more complex models. This may involve using primary cells, tissue explants, or stem-cell derived cell types from the target species to capture relevant cellular context and metabolizing capacity.
• Tier 3: Limited In Vivo Anchor Studies: In cases where in vitro to in vivo extrapolation is uncertain, conduct focused in vivo studies using the candidate species. These studies are designed not for routine testing, but to "anchor" and validate the functional responses predicted by the in vitro and in silico data for critical KEs.

Quantitative Integration via AOP Networks and Systems Toxicology

Addressing knowledge gaps ultimately requires moving from qualitative to quantitative predictions. Quantitative AOPs (qAOPs) and Quantitative Systems Toxicology (QST) models integrate computational and empirical data to predict the probability and severity of an AO across species [51].

• qAOP Development: This involves defining quantitative, dose-response relationships for each KER within a species. Cross-species scaling factors (e.g., based on metabolic rates, receptor densities) can then be applied to these relationships to extrapolate the qAOP [51].
• QST Model Application: Consortia-developed QST platforms, like DILIsym for drug-induced liver injury, provide a powerful framework [52]. These models incorporate species-specific physiological parameters, enzyme kinetics, and cellular response networks. By "swapping" these species-specific parameters, the same model platform can simulate outcomes across humans, rats, and dogs, directly addressing extrapolation gaps [51] [52].

Case Studies in Cross-Species Extrapolation

Pollinator Toxicity: nAChR Activation to Colony Failure

A seminal case study involves the AOP linking activation of the nicotinic acetylcholine receptor (nAChR) to colony death/failure in honey bees (Apis mellifera) [3]. While developed for A. mellifera, concerns extend to non-Apis bees.

• Knowledge Gap: Unknown whether the KEs (e.g., altered neuronal signalling, impaired foraging) are conserved across >20,000 bee species.
• Solution Applied: Researchers used SeqAPASS to evaluate structural conservation of nine proteins in the pathway (starting with nAChR subunits) across insect taxa [3]. Level 1 and 2 analyses confirmed broad conservation of nAChR in Hymenoptera and beyond. Level 3 analysis of critical binding site residues helped refine predictions of susceptibility, particularly for non-Apis bees.
• Outcome: The bioinformatics data provided a biologically plausible tDOA that extended beyond Apis, guiding prioritization for subsequent functional testing in bumblebees and solitary bees.

Liver Injury: Bile Acid Dysregulation

In pharmaceutical development, Drug-Induced Liver Injury (DILI) is a major challenge. A critical KE is the disruption of bile acid homeostasis.

• Knowledge Gap: Significant species differences in bile acid synthesis, transport, and regulation complicate extrapolation from rodent models to human risk [52].
• Solution Applied: The DILI-sim Initiative, a public-private consortium, developed a QST model that explicitly incorporates species-specific parameters [52]. The model simulated how compounds like bosentan disrupt bile acid transporters (e.g., NTCP). It correctly predicted human hepatotoxicity based on in vitro human transporter data, which was not evident from rat studies due to species differences in NTCP inhibition [52].
• Outcome: The QST model identified bile acid accumulation as a key mechanistic contributor to DILI and provided a quantitative framework for bridging the human-rodent knowledge gap, reducing reliance on animal testing that may be poorly predictive [52].

Table 1: Summary of Case Study Approaches and Outcomes

Case Study	AOP / Adverse Outcome	Core Knowledge Gap	Methodology Applied	Key Outcome for tDOA
Pollinator Toxicity [3]	nAChR activation → Colony failure	Applicability beyond honey bees (Apis mellifera)	Bioinformatics (SeqAPASS Levels 1-3)	Structural evidence supported plausible tDOA extension to non-Apis bees, informing testing priorities.
Liver Injury (DILI) [52]	Bile acid transporter inhibition → Human hepatotoxicity	Poor translatability of rodent in vivo data to human risk	Quantitative Systems Toxicology (QST) Modeling (DILIsym)	Model identified species differences in NTCP inhibition, enabling human-relevant risk prediction from in vitro data.

The Scientist's Toolkit: Essential Research Reagent Solutions

Addressing taxonomic applicability requires a suite of interoperable tools and data resources.

Table 2: Key Research Reagent Solutions for tDOA Investigations

Tool/Resource Name	Type	Primary Function in tDOA Assessment	Key Features/Utility
SeqAPASS [3]	Bioinformatics Tool	Evaluates structural conservation of proteins across species via sequence, domain, and residue analysis.	Provides hierarchical, evidence-based lines of evidence for protein conservation. Publicly available from US EPA.
AOP-Wiki [8] [1]	Knowledgebase	Central repository for AOP development and sharing. Stores and displays tDOA information for KEs and KERs.	Enables collaborative development and provides a structured format for documenting tDOA evidence.
EPA CompTox Chemicals Dashboard [49]	Data Integration Platform	Provides access to chemical properties, toxicity data (ToxCast/Tox21), and exposure information.	Allows linking of chemical stressors to MIEs and supports read-across for untested chemicals within a taxonomic group.
Tox21 10K Library & Data Browser [49] [50]	Assay Data & Platform	High-throughput in vitro screening data for ~10,000 chemicals across many pathways.	Functional data source for testing KE perturbation (MIE, early KEs) across orthologous proteins in standardized assays.
DILIsym / QST Platforms [51] [52]	Quantitative Systems Model	Mechanistic, mathematical models that integrate in silico, in vitro, and physiological data to predict toxicity.	Allows virtual "species swapping" via parameters to quantitatively extrapolate AOPs and address interspecies differences.
BioPlanet [49]	Pathway Analysis Tool	A curated knowledgebase of biological pathways. Supports pathway enrichment analysis of 'omics data.	Useful for identifying if broader pathway context surrounding a KE is conserved in a new species.

A Strategic Framework for Prioritization and Research

Systematically closing knowledge gaps requires a targeted strategy. The following framework prioritizes actions based on conservation evidence and assessment needs:

Table 3: Strategic Framework for Addressing Cross-Species Conservation Gaps

Conservation Evidence Level	Recommended Action	Tools & Methods	Goal
Low/No Structural Evidence (SeqAPASS L1 fails)	Exclude from tDOA for that specific KE. Consider alternative AOPs.	Bioinformatics screening (SeqAPASS Level 1).	Prevent erroneous extrapolation and focus resources on plausible taxa.
High Structural, Low Functional Evidence (SeqAPASS passes, little empirical data)	Priority for Tier 1 in vitro screening. Generate functional data using HTS.	Tox21-like assays, orthogonal in vitro assays using recombinant proteins/cells.	Validate computational predictions and establish functional potency.
High Structural, Conflicting Functional Evidence	Conduct Tier 2 mechanistic studies. Investigate compensatory pathways or metabolic differences.	Primary cells, 'omics (transcriptomics, metabolomics), pathway analysis (BioPlanet).	Diagnose the source of discordance (e.g., alternative signaling, differential metabolism).
High Structural & Functional Evidence for early KEs, unknown for late KEs	Develop/Apply qAOP or QST models. Focus on quantifying KERs and interspecies scaling.	QST platforms (e.g., DILIsym), pharmacokinetic/pharmacodynamic (PK/PD) modeling.	Enable quantitative prediction of the full AOP across species for risk assessment.

Confidently defining the Taxonomic Domain of Applicability is the cornerstone of robust, reliable application of the AOP framework in ecological and human health risk assessment. The process is iterative and evidence-based, moving from computational predictions of structural conservation to empirical validation of function, and culminating in quantitative integration via modeling. As illustrated by the case studies, tools like SeqAPASS and QST models are already enabling scientists to systematically bridge knowledge gaps.

The future of this field lies in enhanced integration and collaboration. Key directions include:

• Automated Curation: Linking tools like SeqAPASS directly to the AOP-Wiki to allow dynamic, evidence-driven updates to tDOA descriptions.
• Expanded In Vitro Taxa Coverage: Developing standardized in vitro models (e.g., cell lines, organoids) for a broader range of ecologically relevant species to facilitate functional testing.
• Consortium-Driven Model Development: Expanding the consortia model (exemplified by DILI-sim and Tox21) to tackle cross-species extrapolation for other priority endpoints like developmental neurotoxicity and endocrine disruption [50] [52].

By adopting the structured methodologies and tools outlined in this guide, researchers can transform the tDOA from a poorly defined assumption into a rigorously supported, predictive component of pathway-based toxicology.

The development and application of Adverse Outcome Pathways (AOPs) epitomize the 'Paradox of Complex Simplicity' in regulatory science: simplifying intricate biological phenomena into tractable, modular frameworks without sacrificing mechanistic fidelity [53] [54]. This whitepaper examines this paradox through the lens of AOP taxonomic applicability, a discipline focused on determining the relevance and boundaries of AOPs across species and biological contexts [55]. As AOPs transition from qualitative descriptions to quantitative, predictive networks (qAOPs), the challenge of balancing granular biological detail with utility for chemical risk assessment intensifies [56] [57]. We detail core methodologies, including ontology-based semantic analysis and Bayesian network modeling, that enable this balance [55] [56]. By providing structured experimental protocols, quantitative data summaries, and visualization tools, this guide equips researchers and drug development professionals with the strategies needed to navigate this paradox, thereby enhancing the reliability and regulatory acceptance of pathway-based approaches.

An Adverse Outcome Pathway (AOP) is a structured, modular sequence of biological events, from a Molecular Initiating Event (MIE) to an Adverse Outcome (AO), designed to support chemical risk assessment [5] [53]. The 'Paradox of Complex Simplicity' arises from the need to abstract and simplify vast, interconnected biological systems into linear, causal chains (simplicity) while retaining enough mechanistic depth to ensure predictive accuracy and biological plausibility (complexity) [54] [58].

This paradox is central to the challenge of taxonomic applicability—determining to which species, life stages, or biological contexts a given AOP reliably applies [55] [5]. Overgeneralization can lead to erroneous predictions, while excessive specificity limits utility. Resolving this requires objective, data-driven methods to evaluate the biological coherence and transferability of AOPs [55]. Semantic and quantitative analyses are now bridging this gap, transforming AOPs from static narratives into dynamic, computable knowledge frameworks aligned with the FAIR principles (Findable, Accessible, Interoperable, Reusable) for data [59] [56].

This paper synthesizes current methodologies for developing and evaluating taxonomically applicable AOPs, providing a technical guide for researchers navigating the intersection of mechanistic detail and decision-making utility.

Semantic Characterization: Establishing Coherence and Applicability

Ontology-based semantic analysis provides an objective, computational method to assess the internal coherence of an AOP and its potential taxonomic applicability by analyzing the biological relatedness of its components [55].

Core Methodology: Ontological Annotation and Similarity Measurement

The process translates descriptive AOP elements into computable logic statements using controlled biological vocabularies (ontologies) [55].

• Annotation with Logical Definitions: Each Key Event (KE) in an AOP is annotated using the Entity-Quality (EQ) syntax. This defines a phenotype by linking a biological entity (e.g., a gene, anatomy part) from a reference ontology with a quality describing its alteration (e.g., 'decreased', 'increased') [55].
• Profile Construction: All KEs within an AOP are aggregated to form a phenotypic profile. Similarly, profiles are constructed for genes, biological pathways, diseases, and chemicals from public databases [55].
• Semantic Similarity Calculation: Using a unified ontology graph (e.g., the Vertebrate Phenotype Ontology), semantic similarity metrics (e.g., Resnik, Lin) are computed. These metrics measure the shared information content between ontology terms, quantifying biological relatedness [55].
• Coherence and Mapping Assessment: The mean pairwise similarity of KEs within an AOP assesses its internal coherence. The similarity between an AOP's profile and pre-established biological profiles (e.g., for a specific disease pathway) evaluates its biological alignment and potential taxonomic scope [55].

Table 1: Semantic Analysis of AOP-Wiki Content (Representative Findings)

Analysis Level	Metric	Finding	Implication for Taxonomic Applicability
Key Event Relationship (KER)	Semantic Similarity Score	A substantial number of KERs showed significant semantic coherence [55].	Coherent KERs are more likely to represent conserved, transferable biological relationships.
Whole AOP	Mean Within-AOP Pairwise Similarity	Many AOPs exhibited high internal semantic similarity [55].	Coherent AOPs are biologically plausible units, increasing confidence in their structured logic for cross-species evaluation.
AOP-to-Pathway Mapping	Profile Similarity to Known Pathways	Coherent AOPs mapped to more known genes and pathways [55].	Strong mapping suggests the AOP captures a core, conserved biological process, informing its potential applicability across taxa.

Workflow for Semantic Evaluation

The following workflow outlines the steps for performing a semantic characterization of an AOP.

Diagram 1: Semantic Analysis Workflow for AOP Evaluation (6 steps)

From Qualitative to Quantitative AOPs (qAOPs)

Quantitative AOPs (qAOPs) embed mathematical relationships into the AOP framework, defining how changes in the magnitude or timing of an upstream KE predict changes in a downstream KE. This is critical for moving from hazard identification to dose-response prediction [56] [57].

Table 2: Methodologies for Quantitative AOP Development

Methodology	Core Approach	Data Requirements	Utility for Decision-Making
Systems Toxicology	Builds computational (e.g., ODE-based) models of the KE network based on prior knowledge of system dynamics [56].	Detailed kinetic & dynamic parameters from literature or experiments.	High-precision, mechanism-based prediction within defined bounds. Supports chemical-specific adjustment.
Regression Modeling	Fits statistical models (linear, logistic, power) to empirical data linking two or more KEs [56].	Concurrent or temporal in vivo or in vitro dose-response data for KEs.	Provides empirically derived, transparent relationships. Foundation for point-of-departure derivation.
Bayesian Network (BN) Modeling	Represents KEs and KERs as probabilistic networks, incorporating uncertainty and evidence updating [56] [57].	Qualitative/quantitative KE relationships; can integrate diverse data types and expert judgment.	Handles complexity and uncertainty. Ideal for probabilistic risk assessment and integrating new data.

Experimental Protocol for qAOP Development

A generalized protocol for generating data to develop a qAOP using a regression-based approach is outlined below.

• Objective: To establish a quantitative relationship between an upstream KE (e.g., Thyroxine (T4) in serum, Decreased) and a downstream KE (e.g., Altered retinal layer structure) [60].
• Experimental Design:
  • Stressor Exposure: Subjects (e.g., zebrafish embryos) are exposed to a concentration gradient of a stressor known to inhibit thyroperoxidase (e.g., propylthiouracil - PTU) [60]. Include vehicle and negative controls.
  • Temporal Sampling: Specimens are sampled at multiple, strategically timed intervals to capture the progression of effects.
  • KE Measurement:
    • KE1 (T4): Measure serum T4 levels via ELISA or LC-MS/MS at each time point and concentration.
    • KE2 (Retina): Quantify retinal layer structure via histomorphometric analysis (e.g., thickness of photoreceptor layer) or transcriptomic markers of retinal development.
  • Data Analysis: Perform a concentration-response analysis for each KE individually. Then, use regression modeling (e.g., linear, exponential) to relate the magnitude of KE1 to the magnitude of KE2, accounting for temporal lag if data permits.
• Output: A mathematical function (e.g., KE2 = a * log(KE1) + b) describing the KER, including confidence intervals. This forms a quantitative key event relationship (qKER).

Case Study in Action: AOP 363

AOP 363, "Thyroperoxidase inhibition leading to altered visual function via altered retinal layer structure," exemplifies the balance of detail and utility [60]. Its development followed a structured handbook [5], and it is under OECD review.

• The Simplified Chain: The AOP presents a linear sequence: MIE (Thyroperoxidase Inhibition) → KE1 (Decreased Thyroid Hormone Synthesis) → KE2 (Decreased T4) → KE3 (Decreased T3) → KE4 (Altered Retinal Structure) → AO (Altered Visual Function) [60].
• Underlying Complexity: The AOP summary references linkage to other AOP networks (e.g., on swim bladder inflation) and notes that the KEs are supported by data from diverse perturbations (chemical, genetic, surgical) [60]. This acknowledges complexity without overburdening the core narrative.
• Taxonomic Applicability Statement: While focused on fish due to data abundance, the authors hypothesize applicability to other vertebrates [60], creating a testable hypothesis for semantic or empirical evaluation.

The structure of this AOP and its place in a broader network can be visualized as follows.

Diagram 2: Structure of AOP 363 and Network Linkages (6 nodes)

Successfully navigating AOP development requires leveraging specific tools and resources that align with FAIR principles and support reproducible research [59] [5].

Table 3: Research Reagent Solutions for AOP Development

Tool/Resource Category	Specific Item / Example	Function & Relevance to the Paradox
Central Knowledge Repository	AOP-Wiki (aopwiki.org) [5] [60]	The primary, collaborative platform for developing and storing modular AOPs, KEs, and KERs according to a standardized handbook [5]. Enforces structured simplicity.
Semantic Analysis	Vertebrate Phenotype Ontology (VPO), Semantic similarity calculation tools [55]	Provides the controlled vocabulary and computational methods to objectively assess AOP coherence and biological alignment, addressing taxonomic uncertainty.
Quantitative Modeling	Bayesian Network Software (e.g., GeNIe, Netica), Statistical packages (R, Python libraries) [56]	Enables the development of qAOPs and probabilistic networks that incorporate complexity and uncertainty into predictive models for decision-makers.
Experimental Stressors	Propylthiouracil (PTU), Methimazole [60]	Well-characterized chemical stressors used to induce specific MIEs (e.g., thyroperoxidase inhibition) for generating empirical data to support and quantify AOPs.
Data FAIRification	FAIR AOP Implementation Profile, AOP-Wiki 3.0 Roadmap [59]	Guidelines and upcoming technical upgrades to ensure AOP data is Findable, Accessible, Interoperable, and Reusable, enhancing trust and utility for regulators.

Navigating the 'Paradox of Complex Simplicity' in AOP science is an iterative process of structured simplification followed by systematic re-complexification. The initial AOP provides a simplified, communicable storyline. Semantic analysis then validates and contextualizes this narrative within the broader, complex tapestry of biology [55]. Finally, quantitative modeling injects necessary complexity—in the form of dynamics, uncertainty, and probabilistic relationships—back into the framework to create tools fit for predictive risk assessment [56] [57].

The future of taxonomically applicable AOPs lies in their integration into connected, FAIR-compliant knowledgebases [59]. As shown in the diagram below, this creates a dynamic ecosystem where discrete AOPs, experimental data, and computational models interact, continuously refining the balance between biological detail and regulatory utility.

Diagram 3: The Integrated AOP Knowledge Ecosystem (5 components)

For researchers and drug development professionals, mastery of the methodologies outlined—semantic characterization, quantitative modeling, and structured case-building—is essential. By explicitly addressing taxonomic applicability and the inherent paradox of their work, scientists can build AOPs that are not only mechanistically robust but also decisively useful in shaping a safer environment and healthier future.

The global regulatory landscape for chemical and pharmaceutical safety assessment is undergoing a fundamental transformation. Driven by mandates such as the EU's REACH program, revisions to the U.S. Toxic Substances Control Act (TSCA), and global initiatives to reduce animal testing, regulatory bodies increasingly demand mechanistic, data-rich justifications for safety claims [1]. Concurrently, agencies like the U.S. FDA are embracing policies of "radical transparency," including the public release of Complete Response Letters (CRLs), making the clarity and defensibility of submitted evidence more critical than ever [61]. Within this context, the Adverse Outcome Pathway (AOP) framework has emerged as a central organizing principle for translating mechanistic data into predictions of adverse health effects relevant to regulatory decision-making [1] [8].

An AOP is a structured, sequential description of linked events at different biological levels—from a Molecular Initiating Event (MIE) to an Adverse Outcome (AO) at the organism or population level—that follows exposure to a stressor [8]. Its power lies in its stressor-agnostic nature; it describes biological perturbation pathways that can be triggered by any chemical or agent capable of interacting with the initial molecular target [1]. The core challenge, and the focus of this whitepaper, is that the regulatory utility of an AOP is contingent upon the transparency, credibility, and taxonomic applicability of its supporting evidence. Regulators do not adopt pathways; they adopt well-substantiated, transparently documented, and credible arguments for how mechanistic data predicts a relevant adverse outcome. This document provides a technical guide for researchers and development professionals to optimize their AOP-based claims for regulatory uptake.

Deconstructing the AOP Framework: Building Blocks for Regulatory Argumentation

The AOP framework decomposes complex toxicological processes into modular units, enabling systematic evidence assembly. Understanding these components is the first step in building a credible case for regulatory use.

• Molecular Initiating Event (MIE): The initial, specific interaction between a stressor (e.g., a drug candidate) and a biomolecule (e.g., receptor binding, enzyme inhibition, DNA binding) that triggers the pathway [8].
• Key Event (KE): A measurable, essential change in biological state at the cellular, tissue, or organ level that forms a necessary step toward the AO. KEs are the empirical anchors of the AOP [1] [62].
• Key Event Relationship (KER): A documented, causal link between two KEs (or an MIE and a KE). The KER provides the rationale for why a change in an upstream event is expected to lead to a change in a downstream event. The weight of evidence supporting each KER is the cornerstone of AOP credibility [62].
• Adverse Outcome (AO): A change at the organism level (e.g., organ dysfunction, cancer, reduced survival) or population level that is of direct regulatory concern [8] [63].

A common misconception is that AOPs are simple linear chains. In reality, linear AOPs can be assembled into AOP networks that capture shared KEs and biological interactions, providing a more holistic view of toxicity [1]. Furthermore, the development of Quantitative AOPs (qAOPs) that define mathematical relationships between KEs is essential for moving from qualitative hazard identification to quantitative risk assessment [1].

Table 1: Core Components of the AOP Framework and Their Role in Regulatory Submissions

AOP Component	Definition	Role in Building Regulatory Credibility
Molecular Initiating Event (MIE)	Initial chemical-biological interaction[e.g., binding, inhibition] [8].	Links specific chemistry to a biological perturbation; target for in silico or in vitro screening assays.
Key Event (KE)	Measurable, essential biological change at sub-organism level [62].	Serves as a reliable surrogate endpoint; enables use of New Approach Methodologies (NAMs) like high-throughput in vitro assays.
Key Event Relationship (KER)	Causal, evidence-based linkage between two KEs [62].	Forms the logical backbone of the AOP. The strength of evidence here determines the overall plausibility of the pathway.
Adverse Outcome (AO)	Regulatory-relevant effect at organism or population level [8].	Anchors the pathway to a regulatory endpoint (e.g., liver fibrosis, developmental neurotoxicity), demonstrating relevance.
AOP Network	Multiple linked AOPs sharing common KEs [1].	Accounts for compensatory mechanisms and mixture effects, increasing biological realism and predictive confidence.
Quantitative AOP (qAOP)	AOP with defined quantitative relationships between KEs [1].	Enables prediction of the dose-response and timing of the AO, bridging directly to quantitative risk assessment.

The Centrality of Taxonomic Applicability in AOP Claims

A claim of "applicability" is fundamentally a taxonomic claim. It asserts that the causal biological pathway described in an AOP is sufficiently conserved across the taxa separating the test system (e.g., in vitro human cell model, rodent in vivo study) and the target species of concern (e.g., human patients, an endangered fish species). Failures in transparently evaluating and documenting this applicability domain are a primary reason for regulatory skepticism.

Evidence Requirements for Taxonomic Extrapolation

The basis for extrapolation must be explicitly documented for each KE and KER:

• KE Conservation: Is the key event (e.g., receptor activation, caspase-mediated apoptosis) biologically plausible in the target species? Evidence includes sequence homology of molecular targets, functional similarity of physiological responses, and empirical data.
• KER Conservation: Is the causal relationship between the upstream and downstream KE expected to operate in the target species? This requires evidence of conserved signaling pathways, physiological feedback mechanisms, and tissue/organ system homology.
• Defining Boundaries: Explicitly state the taxonomic limits of the claim. An AOP for aryl hydrocarbon receptor activation may be applicable across most vertebrates but not invertebrates. This clarity is a hallmark of scientific and regulatory rigor.

Table 2: Framework for Assessing and Documenting Taxonomic Applicability of AOPs

Assessment Dimension	Key Questions for Researchers	Types of Supporting Evidence
Molecular & Cellular Conservation	Is the MIE target (e.g., receptor, enzyme) present and functionally similar? Are downstream cellular response pathways conserved?	Protein sequence homology, functional assay data across species, phylogenetic analysis of pathway components.
Physiological & Organ System Context	Are the tissues and organs involved anatomically and functionally comparable? Are systemic feedback loops (e.g., hormonal axes) similar?	Comparative anatomy/physiology studies, existence of analogous organ systems, conservation of homeostatic controls.
Empirical Supporting Data	Does empirical data from the target species, or a closely related surrogate, support the occurrence of the KEs and the sequence of events?	In vivo data from target species, case studies, epidemiological observations, cross-species in vitro model comparisons.
Boundary Identification	Under what conditions does the AOP not apply? What are the known modulating factors (life stage, disease state, genetic polymorphism)?	Data showing pathway divergence in certain taxa, evidence of modulating factors that alter response trajectories.

A Pragmatic Protocol for Building Credible, Transparent AOPs

The slow pace of formal AOP endorsement by bodies like the OECD—with only 17 endorsed as of 2021 amidst hundreds under development—highlights the resource intensity of comprehensive development [62]. A pragmatic, tiered approach focused on systematic KER development accelerates the creation of regulatory-useful knowledge.

Experimental Protocol: Systematic Review for Key Event Relationship (KER) Development

This protocol is tailored for developing a single KER, the core unit of causal inference in an AOP.

Objective: To assemble, evaluate, and transparently document all relevant evidence supporting a hypothesized causal relationship between an upstream Key Event (KEup) and a downstream Key Event (KEdown).

Materials: Access to scientific literature databases (e.g., PubMed, Web of Science, Scopus), reference management software, evidence tracking spreadsheet or database.

Procedure:

• Formulate the KER Hypothesis: Precisely define the KER in the format: "A change in [KEup] is anticipated to lead to a change in [KEdown] because...[proposed biological rationale]."
• Develop a Systematic Search Strategy:
  • Population/Subject: Define the biological system (e.g., mammalian hepatocytes, vertebrate cardiovascular system).
  • Intervention/Exposure: Define the perturbation related to KEup (e.g., "increase in oxidative stress," "inhibition of cyclooxygenase").
  • Comparator: Typically, a control state without the perturbation.
  • Outcome: The measurable change defined by KEdown.
  • Search Terms: Create a Boolean search string using terms for P-I-O. Include synonyms and controlled vocabulary (e.g., MeSH terms).
  • Databases & Limits: Specify databases to be searched and any filters (e.g., language, date).
• Execute Search & Screen Literature: Document the exact search date and results per database. Perform title/abstract screening against pre-defined inclusion/exclusion criteria, followed by full-text review. Use a PRISMA-style flow diagram to document the screening process.
• Evidence Extraction & Weight-of-Evidence Assessment: For each included study, extract data into a standardized table: study design (e.g., in vitro, in vivo), model system, test agent, dose/concentration, results for KEup and KEdown, measures of association (e.g., statistical significance, dose-response), and any noted confounding factors.
  • Assess evidence strength using the Bradford Hill considerations (e.g., temporal sequence, dose-response, consistency, biological plausibility, experimental coherence) [62].
  • Crucially, evaluate the taxonomic relevance of each study to the intended applicability claim of the AOP.
• Document "Canonical Knowledge" Where Appropriate: For KERs describing fundamental biology (e.g., "inhibition of mitochondrial complex I leads to decreased ATP production"), a full systematic review may be impractical. The OECD Handbook permits citation of authoritative reviews or textbooks in such cases, but the rationale for deeming knowledge "canonical" must be stated [62].
• Synthesize and Document: Summarize the overall weight of evidence supporting the KER. Explicitly state the level of confidence (e.g., High, Moderate, Low) and the taxonomic domain for which the evidence is strongest. All search strategies, extracted data, and assessment notes must be archived to ensure full transparency and reproducibility.

Visualizing Knowledge Structures and Workflows for Clarity

Visual diagrams are indispensable for communicating the logical structure of an AOP and the workflow for its development. Below are Graphviz DOT scripts that adhere to the specified style and contrast guidelines.

Diagram 1 Title: Structure of a Linear Adverse Outcome Pathway (AOP)

Diagram 2 Title: Workflow for Developing a Key Event Relationship (KER)

The Scientist's Toolkit for Credible AOP Development

Successfully navigating the intersection of cutting-edge science and regulatory expectations requires a specific suite of tools and resources.

Table 3: Research Reagent Solutions for AOP Development and Documentation

Tool/Resource	Category	Function & Relevance to Credibility	Source/Access
AOP Wiki	Knowledge Management Platform	The primary international repository for developing, sharing, and reviewing AOPs. Using its standardized templates ensures alignment with OECD formats and facilitates peer review [8] [64].	aopwiki.org
OECD Users' Handbook Supplement	Guidance Document	Provides the definitive technical guidance for AOP development, including evidence standards and review criteria. Adherence is de facto mandatory for regulatory submissions [62].	OECD Website
Systematic Review Software (e.g., Covidence, Rayyan)	Evidence Synthesis Tool	Streamlines the literature screening and data extraction process for KER development, ensuring a reproducible, auditable workflow and minimizing bias.	Commercial/Open Source
Evidence Tracking Database	Data Management	A customized spreadsheet or database (e.g., using REDCap or Airtable) is essential for logging extracted study data, Bradford Hill assessments, and taxonomic applicability notes for each KER.	Custom Implementation
Comparative Genomics/Bioinformatics Tools (e.g., Ensembl, BLAST)	Taxonomic Applicability Analysis	Enables direct comparison of gene/protein sequences (e.g., for an MIE target) across species, providing empirical data to support claims of pathway conservation.	Public Databases
qAOP Modeling Software (e.g., R Packages, Computational Systems Biology Platforms)	Quantitative Modeling	Allows the translation of a qualitative AOP into a quantitative, predictive model by defining mathematical relationships between KEs, bridging to risk assessment [1].	Various Platforms

Optimizing for regulatory uptake is not a final-step exercise in "polishing" a submission. It is a principled approach integrated into the entire research and development lifecycle. For AOP-based claims, this means:

• Starting with the Regulatory Endpoint in Mind: Define the AO and work backwards to identify measurable, causally linked KEs.
• Building Evidence Systematically: Treat each KER as a miniature research project requiring a transparent, defensible weight of evidence. Prioritize high-quality, taxonomically relevant data.
• Documenting for an Auditing Mindset: Assume every claim will be scrutinized. Archive search strategies, data, and decision rationales. Clearly delineate established fact from reasoned inference.
• Proactively Addressing Applicability: Do not hide uncertainties regarding taxonomic extrapolation. Define the known boundaries of the AOP and identify critical data gaps as research needs.

In an era defined by both radical regulatory transparency and a pressing need for efficient, human-relevant safety assessment, the credibility of mechanistic claims is paramount. By adopting the structured, transparent practices outlined in this guide, researchers and drug development professionals can construct AOP-based arguments that are not only scientifically robust but also explicitly designed for confident regulatory acceptance.

Strategies for Updating and Curating AOPs with Evolving Taxonomic Evidence

The Adverse Outcome Pathway (AOP) framework serves as a systematic, transparent approach to organize toxicological knowledge by describing a sequential chain of causally linked events from a molecular perturbation to an adverse outcome at the individual or population level [22]. A core scientific and regulatory challenge is defining and refining the taxonomic applicability of these pathways—clarifying which species, life stages, and biological systems an AOP is relevant to. As taxonomic evidence evolves through new genomic, phenotypic, and ecotoxicological data, AOPs must be dynamically curated to maintain their scientific accuracy and regulatory utility. This guide details a strategic, protocol-driven methodology for integrating advancing taxonomic knowledge into the AOP knowledgebase (AOP-KB), ensuring these living documents remain robust tools for predictive toxicology and chemical risk assessment [22].

An AOP is a modular construct beginning with a Molecular Initiating Event (MIE) and culminating in an Adverse Outcome (AO), linked by intermediate Key Events (KEs) and Key Event Relationships (KERs) [22]. AOPs are intentionally chemical-agnostic and designed for broad application; however, their biological plausibility is inherently tied to the conservation of molecular and physiological pathways across taxa. The taxonomic applicability of an AOP defines the boundaries of this conservation.

Evolving evidence—such as the discovery of species-specific receptor isoforms, differences in metabolic pathways, or unique compensatory mechanisms—can directly impact confidence in an AOP's predictions across species. Therefore, a static AOP becomes a liability. The process of updating and curating AOPs with taxonomic evidence is not merely administrative but a core scientific activity that strengthens the weight of evidence, reduces uncertainty in cross-species extrapolation, and ensures the AOP framework fulfills its promise in supporting next-generation risk assessment.

Foundational Concepts: AOP Structure and Taxonomic Applicability

The AOP framework's power lies in its structured, computable format. Understanding its components is essential for effective curation.

• Molecular Initiating Event (MIE): The initial interaction between a stressor and a biomolecule.
• Key Event (KE): Measurable changes in biological state at different levels of organization (cellular, tissue, organ, organism).
• Key Event Relationship (KER): A scientifically supported, causal link between two KEs.
• Adverse Outcome (AO): A regulatory-relevant effect at the individual or population level.
• Taxonomic Applicability: A metadata field associated with the AOP, its individual KEs, and KERs that specifies the taxa, life stages, and biological systems for which there is empirical evidence or strong biological plausibility [22].

The Central Curation Challenge: Taxonomic evidence can evolve at the level of the entire AOP network, a single KER, or an individual KE. For example, new research might show that a KER (e.g., "Increased intracellular calcium leads to Mitochondrial Dysfunction") is highly conserved across all vertebrates, while a specific KE (e.g., "Aryl hydrocarbon Receptor Activation") may be applicable only to a subset of species possessing that receptor [65]. Effective curation requires granular updates at the appropriate modular level.

A Strategic Workflow for Integrating Taxonomic Evidence

The following workflow provides a systematic, five-phase protocol for reviewing and integrating new taxonomic data into the AOP-KB.

Workflow for Taxonomic Evidence Integration in AOPs

Phase 1: Evidence Triage & Horizon Scanning

• Objective: Systematically identify new, taxonomically relevant data.
• Protocol:
  • Establish automated alerts for key terms (e.g., gene homologs, protein targets from the MIE) in major literature databases.
  • Monitor relevant chemical databases (e.g., EPA CompTox Dashboard) linked to AOP stressors [22].
  • Utilize the AOP-Wiki's integrated search, which now allows filtering by taxonomic applicability fields added in recent updates [66].
  • Prioritize data from systematic reviews, comparative genomics studies, and papers that explicitly test a KE or KER in novel species.

Phase 2: Targeted AOP Identification & Retrieval

• Objective: Locate all AOPs, KEs, and KERs in the AOP-KB potentially impacted by the new evidence.
• Protocol:
  • Use the AOP-Wiki search function to find AOPs by:
    • Prototypical Stressor: Search by specific chemical or PubChem ID [66].
    • Key Event Title: Identify modular KEs shared across multiple AOPs.
    • Biological Organization Level: Filter for AOPs operating at relevant biological scales (cellular, organ, etc.) [66].
  • Export relevant AOP network data in XML format for offline analysis, utilizing the nightly or quarterly download options [66].

Phase 3: Granular Evidence Assessment

• Objective: Evaluate the strength and implications of the new evidence for each affected AOP component.
• Protocol:
  • For each relevant KE and KER, apply the modified Bradford-Hill considerations (biological plausibility, essentiality, empirical support) [22].
  • Assess if the new evidence:
    • Expands applicability (e.g., confirms a KER in a new taxonomic class).
    • Restricts applicability (e.g., finds a protein target is absent in a phylum).
    • Modulates the relationship (e.g., identifies a species-specific modulating factor).
  • Document the assessment using a standardized template, citing the new evidence and its impact on the weight of evidence for taxonomic applicability.

Phase 4: Knowledgebase Update & Documentation

• Objective: Formally update the AOP-Wiki and create an audit trail.
• Protocol:
  • Edit Fields: Update the "Taxonomic Applicability" and "Evidence" sections on the relevant AOP, KE, and KER pages.
  • Link Evidence: Use the AOP-Wiki's enhanced functionality to upload Scientific Review Reports or add direct links to journal articles [66]. This increases transparency.
  • Annotate Changes: Clearly describe the rationale for the update in the page's history/log comments, referencing the assessment from Phase 3.
  • Flag for Review: If changes are substantial (e.g., narrowing the core applicability of a widely used KE), use the wiki's contributor system to notify other authors and coaches [66].

Phase 5: Peer Review & Versioning

• Objective: Ensure scientific quality and create stable reference points.
• Protocol:
  • Utilize the AOP-Wiki's coaching system to request a formal review by domain and taxonomic experts [66].
  • For OECD-endorsed or high-priority AOPs, follow the formal OECD review process. The wiki now supports direct links to OECD iLibrary published versions and journal-formatted articles [66] [65].
  • Once reviewed, generate a new versioned snapshot of the AOP. This creates a stable, citable record of the AOP at a point in time, which is crucial for regulatory use, while the main wiki page remains a living document [66].

Quantitative Landscape of AOP Development and Taxonomic Gaps

The growth and current state of the AOP-KB provide context for the curation challenge. The following table summarizes key metrics and highlights areas where taxonomic evidence is often lacking.

Table 1: AOP Knowledgebase Metrics and Taxonomic Characterization Gaps

Metric Category	Quantitative Data / Current State	Implication for Taxonomic Curation
AOP-KB Growth	Regular quarterly XML releases; Platform updated to v2.7 (2024) with improved search/filtering [66].	Increased volume requires efficient, targeted curation protocols. Enhanced filters aid in finding AOPs by status or content.
AOP Status	AOPs range from early development to OECD endorsed. The wiki includes filters for OECD status (e.g., Under Development, Review) [66].	High-priority targets for taxonomic refinement are OECD-reviewed AOPs used in regulatory contexts.
Evidence Strength	Evidence for KERs is assessed via Bradford-Hill criteria (Plausibility, Essentiality, Empirical Support) [22].	Empirical support is often the weakest criterion, especially for taxonomically broad claims. New data can directly strengthen this.
Common Taxonomic Gap	Many AOPs derived from mammalian models list applicability as "All vertebrates" or "All life stages" by default [22].	These are default assumptions requiring validation. Curation efforts should systematically challenge these broad claims with negative or specific evidence.
Modulating Factors	KER pages include sections for modulating factors (e.g., age, diet, sex) [66].	Taxonomy is a primary modulating factor. New species-specific data should be entered here to qualify a KER's applicability.

Experimental Protocols for Generating Taxonomic Applicability Evidence

To proactively address taxonomic gaps, researchers can design studies to test AOP components across species. The following protocol provides a generalizable template.

Protocol: Cross-Species In Vitro Assay for Key Event Relationship (KER) Validation

• Objective: Empirically test the dose-response and temporal concordance of a hypothesized KER in cell lines or primary cells from multiple species.
• AOP Context: This protocol is designed to strengthen the empirical support for a KER (e.g., "Activation of MEK-ERK1/2 leads to Increase, intracellular calcium") [65] across different taxa.

1. Experimental Design:

• Test System: Select representative cell types (e.g., hepatocytes, neurons) from at least 3 species spanning the claimed taxonomic range (e.g., human, zebrafish, chicken).
• Stressor: Use the prototypical stressor listed in the AOP (e.g., a specific heavy metal for AOP 500) [65] at a range of concentrations.
• Key Event Measurement:
  • Upstream KE (e.g., MEK-ERK1/2 activation): Quantify via Western blot for phosphorylated protein.
  • Downstream KE (e.g., Intracellular Calcium): Measure using fluorescent dye (e.g., Fluo-4 AM) in a plate reader or live-cell imaging.
• Controls: Include vehicle control and a positive control specific to each species' cellular machinery where possible.

2. Data Analysis & AOP Curation Integration:

• Dose-Response Concordance: For each species, model the relationship between upstream KE perturbation and downstream KE response. Similar EC50 values and curve shapes across species strengthen the KER's universal applicability.
• Temporal Concordance: Establish the sequence of events. The upstream KE must precede or occur concurrently with the downstream KE.
• Wiki Documentation: Upload dose-response curves and temporal data to the supporting information for the specific KER page. Update the "Taxonomic Applicability" field on the KER page to explicitly list the tested species, moving from a generic claim to an evidence-based list. Cite the study in the "Evidence" section.

Table 2: Research Reagent Solutions for Taxonomic AOP Curation

Tool / Resource	Primary Function	Role in Taxonomic AOP Curation
AOP-Wiki (aopwiki.org)	Central repository for AOP development, sharing, and review [66] [22].	The primary platform for editing taxonomic applicability fields, linking new evidence, and collaborating with coaches and contributors.
OECD AOP Developer's Handbook	Integrated electronic handbook within the AOP-Wiki providing official guidance [66].	Defines the standard methodology for describing AOPs, including evidence assessment, ensuring curation aligns with international best practices.
EPA CompTox Chemicals Dashboard	Database of chemicals with curated properties, assays, and exposure data [22].	Identifies chemical stressors related to an AOP's MIE and can reveal species-specific toxicity data to inform applicability.
Comparative Genomics Databases (e.g., Ensembl, NCBI Homologene)	Platforms for comparing gene sequences, homology, and functional annotation across species.	Validates biological plausibility by confirming the presence/absence and conservation of molecular targets (e.g., receptors, enzymes) involved in the MIE or early KEs.
Xenobiotic Metabolism Databases (e.g., BioTransformer)	Predicts and documents species-specific metabolic pathways for chemicals.	Identifies critical modulating factors; differences in metabolism can activate or deactivate a prototypical stressor, altering an AOP's applicability.
Systematic Review Management Software (e.g., Covidence, Rayyan)	Facilitates the screening and data extraction process for literature reviews.	Accelerates Phase 1 (Horizon Scanning) by enabling efficient, collaborative review of large volumes of literature to find taxonomically relevant studies.

The dynamic curation of AOPs with evolving taxonomic evidence is not an optional task but a fundamental requirement for the scientific credibility and regulatory acceptance of the framework. By adopting the structured workflow, experimental protocols, and tools outlined in this guide, the AOP community can transition from describing static pathways to managing living, taxonomically intelligent networks.

The ultimate goal is an AOP knowledgebase where the "Taxonomic Applicability" field is a rich, evidence-based, and frequently updated descriptor—not a placeholder assumption. This precision will empower regulators to make confident, species-specific predictions and guide researchers toward the most critical empirical gaps. As the AOP-Wiki's infrastructure continues to advance—with features like enhanced reporting and third-party tool integration—the community is now equipped with the technical capability to match this scientific ambition [66]. The path forward requires a concerted commitment to continuous, collaborative curation, ensuring the AOP framework remains a robust foundation for 21st-century toxicology and risk assessment.

Utilizing Data-Driven and AI-Based Approaches for Predictive Applicability Assessment

The establishment of a taxonomic domain of applicability (tDOA) is a critical, yet often inadequately defined, component in the application of the Adverse Outcome Pathway (AOP) framework for regulatory decision-making and chemical safety assessment [3]. Traditionally, tDOA is narrowly confined to species with existing empirical data, limiting confident extrapolation to untested organisms. This whitepaper elucidates a paradigm shift towards predictive applicability assessment, where data-driven and artificial intelligence (AI)-based methodologies are deployed to systematically extrapolate biological plausibility across taxa. By integrating bioinformatics tools, machine learning models, and multi-omics data, researchers can transcend empirical limitations, defining the tDOA through evidence of structural and functional conservation of key events (KEs) and their relationships (KERs) [3]. This technical guide details the core principles, computational workflows, and experimental protocols underpinning this integrative approach, providing a robust framework for enhancing the predictive power and regulatory utility of AOPs within a broader thesis on AOP taxonomic applicability.

The AOP framework is a structured representation of biologically sequential events, linking a Molecular Initiating Event (MIE) to an Adverse Outcome (AO) through measurable KEs [3]. While AOPs organize mechanistic knowledge, their utility in protecting ecosystems or predicting human health risks hinges on understanding their relevance across species—the tDOA. Conventionally, tDOA is inferred from the specific species used in cited studies, often with assumed, undocumented broader relevance [3].

The core challenge lies in the tension between specificity and extrapolation. An AOP developed for honey bees (Apis mellifera) may be biologically plausible in bumblebees, but without evidence, its application remains uncertain [3]. Resolving this requires moving from descriptive documentation to predictive assessment. This entails evaluating two pillars: structural conservation (the presence and similarity of biological entities like proteins) and functional conservation (the preservation of their biological role) [3]. Modern data-driven strategies provide the tools to assess these pillars computationally, creating a predictive model of AOP applicability before extensive in vivo testing.

Foundational Data-Driven Methodologies for tDOA Assessment

Bioinformatics-Driven Structural Conservation Analysis

Bioinformatics tools offer a first line of evidence for structural conservation by analyzing genetic and protein sequences across species.

• SeqAPASS (Sequence Alignment to Predict Across Species Susceptibility): A publicly available tool that employs a hierarchical, three-level evaluation [3].

  • Level 1: Primary Sequence Comparison. Identifies protein orthologs across species based on overall amino acid sequence similarity.
  • Level 2: Functional Domain Conservation. Evaluates the preservation of specific protein domains responsible for function (e.g., ligand-binding domains).
  • Level 3: Critical Amino Acid Residue Evaluation. Assesses conservation of individual residues known to be essential for protein-ligand interaction or protein function [3].

• Application Workflow: For an AOP, the proteins corresponding to each KE (e.g., a specific receptor for the MIE, enzymes for intermediate KEs) are identified as query sequences. SeqAPASS analyses across these three levels generate data on which taxa likely possess a functionally conserved version of that protein, thereby defining the plausible tDOA for that individual KE [3].

Integrating Empirical andIn SilicoData for Functional Assessment

Structural conservation suggests, but does not prove, functional conservation. A weight-of-evidence approach integrates multiple data streams:

• Empirical Toxicity Data: Existing in vivo and in vitro data from model and non-model species.
• In Vitro High-Throughput Screening (HTS) Data: Data from programs like ToxCast/Tox21, which can test chemical effects on conserved molecular targets across species.
• Literature Mining & Ontologies: Text mining and structured biological ontologies can help identify documented functional roles of orthologous proteins in different species.

The convergence of bioinformatics-predicted structural conservation with empirical or literature-based evidence of function strengthens the hypothesis of AOP applicability.

AI and Machine Learning for Predictive Modeling

AI and machine learning (ML) transform scattered data into predictive models of toxicity and applicability. These approaches can be categorized as top-down (data-driven, correlative) or bottom-up (mechanism-driven, simulative) [67].

Table 1: Comparison of Top-Down and Bottom-Up Computational Approaches for Predictive Assessment [67]

Method	Classification	Core Algorithms	Description & Role in tDOA Assessment	Key Reference Examples
Text Mining (TM)	Top-down	Latent Dirichlet Allocation (LDA), Named Entity Recognition (NER)	Extracts relationships between species, proteins, chemicals, and toxic outcomes from scientific literature. Can identify undocumented taxonomic connections or adverse events.	[67]
Quantitative Structure-Activity Relationship (QSAR)	Top-down	Random Forest (RF), Support Vector Machine (SVM), Artificial Neural Networks (ANNs)	Correlates chemical structural features with biological activity/toxicity. Can predict whether a chemical will perturb a KE in a new species if chemical-biological interaction data is available.	[67]
Association Rule Mining (ARM)	Top-down	Apriori, FP-Growth	Discovers co-occurrence patterns (e.g., between a specific phytochemical and liver toxicity). Useful for identifying potential KERs from large-scale datasets.	[67]
Random Walk with Restart (RWR)	Bottom-up	RWR, heNetRW	Simulates "diffusion" across biological network graphs (protein-protein, metabolic pathways). Predicts novel downstream KEs or affected pathways in a species, informing KER plausibility.	[67]
Molecular Docking (MD)	Bottom-up	Rigid-body/Flexible Docking algorithms	Predicts binding affinity and orientation of a chemical within a protein's 3D structure. Directly tests the MIE plausibility by modeling ligand-receptor interactions in species with modeled protein structures.	[67]
Physiologically Based Pharmacokinetic (PBPK) Modeling	Bottom-up	Nonlinear Mixed-Effects Modeling (NONMEM)	Simulates Absorption, Distribution, Metabolism, and Excretion (ADME) of chemicals across tissues. Critical for understanding internal dose and determining if a KE-concentration threshold can be reached in different species.	[67]

Practical Implementation: Protocols and Workflows

Case Study Protocol: Defining tDOA for an AOP Using SeqAPASS

This protocol, based on a published case study, details steps for computationally expanding the tDOA of an AOP linking nicotinic acetylcholine receptor (nAChR) activation to colony failure in bees [3].

• AOP Selection & KE Protein Identification: Select a defined AOP (e.g., AOP-Wiki ID 89). List all KEs and identify the specific proteins (or genes) primarily responsible for each KE. For AOP 89, this included nine proteins like nAChR subunits and detoxification enzymes [3].
• SeqAPASS Level 1-3 Analysis:
  • For each query protein, perform a Level 1 SeqAPASS analysis to identify orthologs across a broad taxonomic range.
  • Perform Level 2 analysis to assess conservation of key functional domains (e.g., the ligand-binding domain of nAChR).
  • Perform Level 3 analysis focusing on residues critical for the interaction of interest (e.g., neonicotinoid binding in nAChR).
• Data Synthesis & tDOA Assignment: Compile results. For each KE, assign a biologically plausible tDOA based on taxa where all three levels of analysis show high conservation. Compare this with the empirical tDOA from the AOP description.
• Integration with Empirical Evidence: Search for in vitro or in vivo data (e.g., toxicity studies, electrophysiology data) in species identified as structurally conserved. This combined evidence strengthens the tDOA definition.

Protocol: Developing an Explainable AI (XAI) Model for Biomarker-Based KE Assessment

This protocol outlines the development of an AI model to identify metabolomic biomarkers, which can serve as sensitive, early KEs. It is adapted from a study on breast cancer detection [68].

1. Study Design & Sample Collection:

• Cohort: Recruit defined case (e.g., chemically exposed, diseased) and control groups. The referenced study used 138 breast cancer patients and 76 healthy controls [68].
• Sample Type: Collect appropriate biospecimens (e.g., plasma, tissue, cell culture medium).
• Power Analysis: Conduct a statistical power analysis to determine minimum sample size. The referenced study used α=0.05, power=0.80, effect size=0.5 [68].

2. Metabolomics Profiling:

• Platforms: Use high-resolution mass spectrometry platforms like Liquid/Gas Chromatography Time-of-Flight MS (LC/GC-TOFMS) for broad metabolite coverage [68].
• Data Preprocessing: Perform peak alignment, normalization, and missing value imputation. Annotate metabolites using standard libraries.

3. Predictive Model Development & XAI Analysis:

• Algorithm Selection: Train and compare multiple ensemble ML algorithms (e.g., Random Forest, XGBoost, LightGBM) [68].
• Model Validation: Use k-fold cross-validation. Evaluate performance with metrics like Accuracy, Sensitivity, Specificity, and Area Under the ROC Curve (AUC).
• Explainability Analysis: Apply SHapley Additive exPlanations (SHAP) analysis to the best-performing model. SHAP quantifies the contribution of each metabolite (feature) to the model's prediction, identifying the most important discriminatory biomarkers [68]. In an AOP context, these top metabolites represent candidate biomarkers for a KE.

4. Biological Interpretation & KE Linkage:

• Map the top-ranked discriminatory metabolites identified by SHAP to known biochemical pathways.
• Hypothesize and experimentally validate their connection to a specific KE (e.g., oxidative stress, energy depletion).

Diagram 1: Integrated Workflow for Predictive tDOA Assessment (97 characters)

The Scientist's Toolkit: Essential Research Reagent Solutions

Table 2: Research Reagent Solutions for Predictive Applicability Assessment

Item / Solution	Function in Predictive Assessment	Example Application / Note
SeqAPASS Tool	Public web tool for hierarchical (sequence, domain, residue) cross-species protein conservation analysis.	Provides primary evidence for structural conservation of MIEs and KE proteins across taxa [3].
AOP-Wiki	Central repository for structured AOP knowledge, including KEs, KERs, and chemical stressors.	The starting point for identifying relevant KEs and their current, often limited, tDOA descriptions [69].
ToxCast/Tox21 Database	EPA/NIH databases containing high-throughput screening data for thousands of chemicals across hundreds of assay targets.	Provides empirical in vitro data on KE perturbation (MIE, early KEs) relevant to functional conservation assessment.
UNIPROT / NCBI Protein DB	Curated databases of protein sequence and functional information.	Sources for obtaining reference protein sequences for SeqAPASS queries and functional annotation.
Metabolomics Platforms (LC/GC-TOFMS)	High-resolution mass spectrometry systems for untargeted profiling of small-molecule metabolites.	Generates data for discovering novel KE biomarkers via AI/ML analysis, as demonstrated in breast cancer studies [68].
SHAP (SHapley Additive exPlanations)	A game-theoretic XAI method for interpreting output of any ML model.	Critical for interpreting AI models, identifying the most important metabolite or genomic features driving a prediction, and linking them to biology [68].
Python/R with ML Libraries (scikit-learn, TensorFlow, PyTorch)	Programming environments with extensive libraries for data preprocessing, statistical analysis, and machine learning.	The computational engine for developing custom QSAR, classification, and biomarker discovery models [70].
Molecular Docking Software (AutoDock, Glide)	Software for simulating and scoring the binding of a ligand to a protein target.	Provides bottom-up, mechanistic evidence for the plausibility of an MIE in a species with a homology-modeled protein structure [67].

Diagram 2: Data Integration Informs AOP Taxonomic Applicability (80 characters)

The integration of bioinformatics, artificial intelligence, and multi-omics data represents a transformative advancement for the AOP framework. It moves the assessment of taxonomic applicability from a descriptive, post-hoc summary of tested species to a predictive, hypothesis-driven science. By leveraging tools like SeqAPASS for structural analysis and XAI-enhanced ML models for functional biomarker discovery, researchers can define a biologically plausible tDOA with greater confidence and scientific rigor [3] [68].

Future progress depends on enhancing data accessibility and interoperability, developing standardized protocols for computational tDOA assessment, and fostering closer collaboration between computational biologists, toxicologists, and regulators. Ultimately, these predictive approaches will lead to more efficient, cost-effective, and protective chemical safety assessments, robustly grounded in mechanistic understanding across the tree of life.

Benchmarking and Validating AOP Taxonomic Scope: Assessment Frameworks and Future Directions

Weight-of-Evidence Frameworks for Evaluating Taxonomic Applicability Confidence

The Weight-of-Evidence framework is a systematic methodology for integrating and evaluating diverse data streams to support robust scientific conclusions. In toxicology and chemical risk assessment, WoE approaches are critical for interpreting complex biological information, particularly when data are derived from various sources with differing levels of reliability and relevance [71]. Concurrently, the Adverse Outcome Pathway framework has emerged as a foundational tool for organizing mechanistic knowledge. An AOP describes a sequential chain of causally linked events at different levels of biological organisation—from a molecular initiating event to an adverse outcome in an organism or population [72] [1].

This whitepaper frames WoE frameworks within the broader thesis of AOP taxonomic applicability research. A core challenge in applying AOPs for regulatory decision-making or drug development is determining their confidence and relevance across different species (taxa). AOPs are often developed using data from model organisms or in vitro systems, but their predictive utility for human health or other ecologically relevant species requires explicit evaluation [73] [1]. A structured WoE approach provides the necessary rigor to assess the strength and biological plausibility of an AOP's transferability across taxonomic groups, thereby defining the boundaries of its applicability and ensuring reliable extrapolation in safety assessments.

Foundational Principles of Weight-of-Evidence Frameworks

Weight-of-Evidence is not a single method but a philosophical and procedural approach to decision-making. It involves the transparent, systematic, and consistent integration of multiple lines of evidence, considering the strength, relevance, and consistency of each piece of data [71] [74]. In the context of toxicology and AOP evaluation, this is essential for moving beyond reliance on any single assay or study.

Core criteria employed in WoE assessments include:

• Consistency: The degree to which different lines of evidence point to the same conclusion.
• Biological Plausibility: The extent to which a hypothesized relationship (e.g., a Key Event Relationship in an AOP) is consistent with established biological knowledge.
• Reliability and Relevance: The inherent trustworthiness of the data source (reliability) and its pertinence to the specific question being asked (relevance).
• Dose-Response Concordance: The alignment of observed effects across different levels of biological organization with increasing stressor intensity.
• Temporal Sequence: Evidence that the proposed cause precedes the effect in time.

The European Food Safety Authority and other regulatory bodies have developed formal guidance for integrating evidence, which can be applied manually or through dedicated computational tools [71]. The output is a graded confidence conclusion—such as high, moderate, or low—regarding the question under assessment, which in this context is the taxonomic applicability of an AOP.

Table 1: Comparison of WoE Framework Applications in Different Fields

Field/Application	Primary Objective	Key Data Types Integrated	Reference
Freshwater Conservation	Guide management from existing biodiversity data.	Field monitoring data, statistical models, habitat variables.	[75]
Medical Device Biocompatibility	Provide realistic patient risk assessment by recalibrating worst-case data.	Chemical characterization, biological endpoint tests, clinical data.	[74]
Computational Toxicology (NAM Integration)	Build confidence for chemical risk assessment using non-test methods.	In silico model predictions, read-across data, QSAR results.	[71]
AOP Taxonomic Applicability	Evaluate confidence in pathway conservation across species.	Genomic homology, in vitro assay data, comparative toxicology studies.	[73] [1]

Structure and Components of Adverse Outcome Pathways

An AOP is a conceptual construct that links a measurable molecular perturbation to an adverse outcome of regulatory concern through a series of intermediate key events [8] [1]. Its core components are:

• Molecular Initiating Event (MIE): The initial interaction of a stressor (e.g., a chemical) with a biological target (e.g., receptor binding, protein inhibition) that triggers the pathway [73].
• Key Events (KEs): Measurable biological changes at the cellular, tissue, or organ level that are essential for the progression of toxicity.
• Key Event Relationships (KERs): Descriptions of the causal or mechanistic linkages between KEs, explaining how one event leads to the next.
• Adverse Outcome (AO): The detrimental effect at the organismal (e.g., reduced survival, cancer) or population level (e.g., decline) that is relevant for risk assessment [8].

AOPs are chemically agnostic; they describe biological response pathways that can be triggered by any stressor capable of modulating the MIE [1]. Their primary utility lies in organizing fragmented mechanistic data into a coherent narrative, which in turn identifies knowledge gaps, guides targeted testing, and supports the development of alternative testing strategies [73] [72].

AOP Structural Framework

The Challenge of Taxonomic Applicability for AOPs

A fundamental assumption in applying AOPs across species is the evolutionary conservation of the underlying biological pathways. However, the degree of conservation for specific MIEs, KEs, and KERs can vary significantly, creating uncertainty in extrapolation [1]. Assessing taxonomic applicability confidence is therefore a critical step before an AOP developed in rats, fish, or in vitro human cells can be reliably used for human health or ecological risk assessment.

Major challenges include:

• Species-Specific Differences: Variations in protein structure (affecting MIE), tissue physiology, metabolic pathways, and compensatory mechanisms can alter pathway progression.
• Data Completeness and Quality: The evidence supporting each KER is often uneven across species, with robust data for one taxon and limited or absent data for another.
• Defining Applicability Boundaries: Determining the specific taxonomic groups (e.g., all mammals, only rodents, specific fish families) for which the AOP is considered valid.

The U.S. EPA's AOP Database version 2 facilitates this evaluation by integrating AOP information with cross-species genomic homology data, allowing researchers to investigate the conservation of molecular targets (genes/proteins) across taxa [76]. Initiatives like the OECD's AOP Programme also emphasize the need for clear documentation of taxonomic applicability within AOP descriptions [72].

Table 2: Key Challenges and Data Sources for Assessing AOP Taxonomic Applicability

Challenge	Impact on Applicability Assessment	Exemplary Data Sources for WoE
Divergence in Molecular Target	An MIE may not occur if the target receptor or enzyme is absent or structurally different.	Genomic databases (Ensembl, HomoloGene), protein structure models.
Altered Tissue Response	Conservation of a cellular KE does not guarantee identical tissue or organ response.	Comparative physiology studies, tissue-specific 'omics data (e.g., from GTEx).
Differences in Metabolic Fate	A pro-toxin may not be activated, or a toxin may be detoxified more efficiently.	In vitro metabolism studies (e.g., Tox21 project on metabolic capability) [77], pharmacokinetic models.
Variability in Compensatory/Homeostatic Mechanisms	Resilient homeostatic networks may prevent progression to the AO in some species.	Comparative phenotyping studies, population variability data (e.g., from Diversity Outbred models) [77].

Methodologies for Integrating WoE to Assess Taxonomic Applicability

Applying a WoE framework to evaluate taxonomic applicability involves both qualitative and quantitative methodologies that systematically gather and weigh evidence related to the conservation of each AOP component.

Qualitative WoE Integration

This approach involves expert judgment based on predefined criteria. The Bradford Hill considerations (e.g., strength, consistency, specificity, temporality, biological gradient, plausibility) are often adapted for this purpose [1]. For each KER in the AOP, evidence is gathered from the literature for the source species (e.g., rat) and the target species (e.g., human). The quantity, quality, and congruence of this evidence are then scored (e.g., as strong, moderate, or weak) to reach an overall confidence rating for the KER's applicability. This process is repeated for all KERs to determine confidence in the entire AOP's applicability.

Quantitative & Semi-Quantitative Integration

These methods introduce more objectivity through scoring systems or computational models.

• Scoring Matrices: Evidence for each KER is categorized (e.g., in vivo, in vitro, in silico) and weighted based on predefined rules regarding study reliability and relevance. The scores are aggregated to produce a numeric confidence value [74].
• Bayesian Belief Networks: These probabilistic graphical models are powerful for integrating diverse and sometimes conflicting evidence. They can incorporate data on genomic homology, in vitro assay results across cell types, and apical endpoint data to calculate the probability that an AOP is operative in a new taxonomic group [1].
• Data-Driven Workflows: Projects like Tox21 generate high-throughput transcriptomic data for hundreds of chemicals. By analyzing whether chemicals known to activate a specific MIE induce conserved gene expression signatures across species-derived cell lines, one can generate empirical evidence for pathway conservation [77].

WoE Framework for Taxonomic Applicability

Case Studies in WoE for Taxonomic Applicability

Case Study 1: Skin Sensitization AOP (AOP 40)

The skin sensitization AOP is one of the most developed and internationally accepted pathways. Its MIE is covalent binding to skin proteins, leading to keratinocyte activation, dendritic cell activation, and ultimately T-cell proliferation (AO: allergic contact dermatitis) [1]. A defined approach using this AOP has allowed the replacement of traditional animal tests (like the guinea pig maximization test) for human hazard classification.

• WoE for Human Applicability: Confidence is high due to extensive data directly from human in vitro assays (e.g., Direct Peptide Reactivity Assay for MIE, KeratinoSens for keratinocyte response) and human patch test data. The biological pathway is well-conserved in humans.
• Assessment for Other Taxa: Evaluating this AOP for, say, aquatic organisms requires a new WoE assessment. The MIE (protein binding) may be conserved, but the inflammatory KEs involving specific cytokines and immune cells are likely not conserved in fish, leading to low confidence in direct applicability. This demonstrates the taxon-specific nature of AOP confidence.

Case Study 2: Endocrine Disruption and Pubertal Timing

The U.S. EPA's Endocrine Disruptor Screening Program and research projects within Tox21 aim to identify chemicals that alter hormonal signaling [8] [77]. An AOP network for altered pubertal timing might involve MIEs like activation of the Gonadotropin-Releasing Hormone Receptor (GnRHR).

• WoE for Cross-Species Extrapolation: A Tox21 project screened chemicals using a human cell line engineered with the human GnRHR [77]. To assess ecological risk, a WoE framework must evaluate the conservation of this receptor and its downstream signaling network in wildlife species (e.g., amphibians, fish). Genomic homology provides initial evidence, but functional assays in non-human systems are required to build moderate-to-high confidence for specific taxonomic groups.

Table 3: Summary of AOP Case Studies and WoE Application

Case Study	AOP Title / Endpoint	Key Molecular Initiating Event (MIE)	Primary Taxonomic Scope of Development	WoE Application for Taxonomic Applicability
Skin Sensitization	Skin Sensitization leading to Allergic Contact Dermatitis (AOP 40)	Covalent binding to skin proteins	Human (validated for replacement of animal tests)	High confidence for humans via defined in vitro assays. Low confidence for most aquatic taxa due to immune system divergence.
Endocrine Disruption	GnRHR Activation leading to Altered Pubertal Timing	Agonism of Gonadotropin-Releasing Hormone Receptor	Human (via engineered cell line screening)	Requires explicit assessment for each wildlife taxon. Genomics provide initial evidence; functional conservation studies are needed for higher confidence.
Retinoid Signaling	Disruption of Retinoic Acid Signaling leading to Developmental Defects	Antagonism of Retinoic Acid Receptor	Broad (highly conserved pathway across vertebrates) [77]	Potentially high confidence across vertebrates for core pathway, but susceptibility windows and phenotypes may vary by species.

Detailed Experimental Protocols for Generating WoE Data

Protocol: Systematic Review for KER Conservation

Objective: To qualitatively assess the weight of evidence supporting the conservation of a specific Key Event Relationship (KER) between two species.

• Define the KER: Precisely state the upstream Key Event (KEup) and downstream Key Event (KEdn) and their hypothesized relationship (e.g., "Increased oxidative stress leads to hepatic inflammation").
• Develop Search Strategy: Create a Boolean search string for biomedical databases (e.g., PubMed, Web of Science) incorporating terms for KEup, KEdn, the source species (e.g., "rat"), and the target species (e.g., "human," "rabbit").
• Screen and Select Studies: Two independent reviewers screen titles/abstracts and then full texts against pre-defined inclusion/exclusion criteria (e.g., primary research studies, relevant exposure/dose, direct measurement of both KEs).
• Extract Data & Assess Quality: Extract data on study design, dose-response, temporal sequence, and strength of association. Assess study reliability using a tool like Klimisch scoring.
• Synthesize Evidence and Score: Evaluate the body of evidence for consistency, biological plausibility, and concordance. Use a scoring matrix (e.g., adapted from [75]) to assign a confidence level (High, Moderate, Low) to the KER's conservation.

Protocol: High-Throughput Transcriptomics for Pathway Conservation

Objective: To generate empirical, quantitative data on AOP conservation using cross-species in vitro models [77].

• Cell Model Selection: Obtain relevant cell types (e.g., hepatocytes, neuronal progenitors) from the source (rat) and target (human) species. Use primary cells or well-characterized cell lines.
• Chemical Selection & Dosing: Select a set of reference chemicals known to trigger the MIE of interest in the source species. Include negative controls. Perform concentration-response treatments.
• RNA Sequencing & Bioanalysis: At appropriate timepoints, extract total RNA and perform RNA-Seq. Process reads through a standardized pipeline (alignment, quantification).
• Signature Development & Comparison: For the source species, identify a conserved gene expression signature associated with the MIE/KE using the reference chemicals. Apply this signature to the target species' expression data.
• Quantitative Assessment: Use metrics like Gene Set Enrichment Analysis (GSEA) to determine if the signature is significantly enriched in the target species after chemical treatment. The strength and consistency of enrichment across multiple chemicals provide a quantitative weight of evidence for pathway conservation.

Table 4: Key Research Reagent Solutions for AOP/WoE Studies

Reagent / Resource	Function in AOP/WoE Research	Example / Supplier Notes
Engineered Cell Lines for MIEs	Enable high-throughput screening for specific molecular initiating events (e.g., receptor binding, enzyme inhibition).	HEK293 cells engineered with human nuclear receptors (ER, AR, PPARγ); used in Tox21 [78] [77].
Metabolically Competent In Vitro Systems	Provide xenobiotic metabolism to better approximate in vivo bioactivation/detoxification, critical for accurate KE measurement.	Co-cultures with hepatocytes, S9 fractions, or engineered cells expressing CYP450 enzymes [77].
Diverse Population-Derived Cell Models	Capture human genetic variability in toxicodynamic response, informing susceptibility within a taxon.	iPSC-derived cells from diverse donors; Diversity Outbred rodent-derived cells [77].
High-Throughput Transcriptomic Assays	Generate systems-level data to identify conserved pathway-level signatures across species and chemicals.	TempO-Seq, targeted RNA-Seq platforms used for high-throughput screening in Tox21 projects [77].
Computational Prediction Tools (QSAR, Read-Across)	Provide in silico evidence for chemical properties and biological activity as one line of evidence in WoE.	Tools like PLATO (target fishing), TIRESIA (developmental toxicity prediction) [71].
AOP Knowledgebase (AOP-KB) Platforms	Central repositories for structured AOP information, essential for accessing and comparing existing pathways.	OECD's AOP Wiki, U.S. EPA's AOP Database (integrates gene, chemical, disease data) [72] [76].
Bioinformatic Homology & Ontology Tools	Assess conservation of molecular targets (genes/proteins) and biological processes across taxa.	Ensembl Compara, HomoloGene, Gene Ontology (GO) resources [76].

The integration of Weight-of-Evidence frameworks with Adverse Outcome Pathway science provides a powerful, systematic approach for evaluating the confidence with which mechanistic toxicological knowledge can be extrapolated across taxonomic boundaries. This process transforms AOPs from hypothetical constructs into qualified tools for predictive toxicology in drug development and chemical risk assessment. By explicitly documenting and scoring the evidence for the conservation of each key event relationship, researchers and regulators can define the applicability domain of an AOP, enabling its reliable use in species-specific safety decisions while avoiding inappropriate extrapolations.

Future advancements in this field will be driven by:

• Increased Data Integration: Leveraging large-scale databases like the EPA's AOP-DB v.2, which links AOPs to cross-species genomics and population variability data [76].
• Standardized Reporting: Wider adoption of formal WoE assessment templates and confidence scoring systems within the OECD AOP development programme [72].
• Complex Model Systems: Greater use of phylogenetically broad in vitro models (e.g., cell lines from multiple species) and computational models to generate quantitative, predictive data on pathway perturbation.
• Artificial Intelligence: Application of machine learning to automate evidence extraction from literature and to integrate multimodal data streams for more robust and dynamic confidence assessments.

The ultimate goal is an iterative, evidence-driven cycle where WoE assessments of taxonomic applicability guide targeted research to fill critical data gaps, which in turn refine the AOP and increase confidence in its application, leading to more efficient and reliable safety evaluations across species [75] [1].

This whitepaper presents a comparative analysis of Adverse Outcome Pathways (AOPs) within the context of a broader thesis on AOP taxonomic applicability. AOPs are conceptual frameworks that map the mechanistic sequence of causally linked events from a molecular initiating event (MIE) to an adverse outcome (AO) of regulatory relevance [9]. A critical challenge in AOP utility is defining their taxonomic domain of applicability (tDOA)—the range of species for which the pathway is biologically plausible [79]. The analysis herein evaluates the strengths and limitations of AOPs when applied across diverse taxonomic groups (e.g., mammals, fish, invertebrates), focusing on the evolution of cross-species AOP networks (AOPNs) and the computational tools enabling taxonomic extrapolation. The integration of FAIR (Findable, Accessible, Interoperable, Reusable) data principles and New Approach Methodologies (NAMs) is emphasized as foundational for advancing reliable, animal-sparing chemical risk assessments [59].

The AOP framework has become a central organizing principle in modern toxicology and ecotoxicology [9]. It provides a structured representation of toxicological mechanisms, facilitating the use of mechanistic data for hazard identification and risk assessment. A core, yet often complex, component of an AOP is its taxonomic applicability [80]. An AOP developed in a model organism like the rat (Rattus norvegicus) or the nematode (Caenorhabditis elegans) is most reliable for that species. Its relevance to other species—a process termed read-across—must be critically evaluated based on the conservation of the underlying biological pathway [79].

This evaluation is not binary but exists on a continuum. Some Key Events (KEs), particularly upstream MIEs involving fundamental cellular processes (e.g., reactive oxygen species formation, receptor binding), may have broad taxonomic conservation. In contrast, downstream KEs and the final AO, which may involve organ-specific physiology or complex life-stage development, often have a narrower taxonomic domain [81]. For instance, an AOP network for impaired androgen signaling leading to shortened anogenital distance (AGD) has upstream events applicable to all mammals, but the specific AO (shortened AGD in male offspring) is directly relevant only to mammalian species [81]. Therefore, a comparative analysis of AOPs across taxa is essential to define the boundaries of their application, identify knowledge gaps, and build confidence in their use for protecting ecosystems and human health under a One Health perspective [79].

Foundations for Comparison: AOP Structure and Taxonomic Annotation

Core AOP Components and Evidence

An AOP is composed of causally linked Key Events (KEs) spanning biological organization. The strength of these Key Event Relationships (KERs) is evaluated using the Bradford-Hill criteria, leading to a Weight of Evidence (WoE) assessment [4]. Confidence in an AOP is rated as High, Moderate, or Low based on the empirical support for both the KEs and the KERs. Essential structured metadata for taxonomic analysis includes:

• Biological Context: The organ, cell, and life stage for each KE [80].
• Taxonomic Applicability: The species for which evidence for a KE or KER exists [80].
• Stressors: The prototypical chemicals or stressors known to trigger the pathway [82].

Tools for Taxonomic Annotation and Analysis

Advancements in bioinformatics have produced tools specifically designed to assess taxonomic applicability:

• SeqAPASS (Sequence Alignment to Predict Across Species Susceptibility): A web-based tool that compares protein sequence similarity across species to predict the potential for a chemical to interact with a molecular target (e.g., an enzyme or receptor) in untested species [79].
• G2P-SCAN (Genes-to-Pathways Species Conservation Analysis): An R package that evaluates the conservation of entire biological pathways or gene sets across a broad range of species, moving beyond single protein comparisons [79].
• Biovista Vizit: A knowledge graph platform that mines the AOP Knowledge Base (AOPKB). Its January 2025 release enhances the ability to tag all biomedical entities and extract experimental methods from the corpus, creating a cross-reference index that can reveal taxonomic information embedded in free text [80].

Methodologies for Assessing Taxonomic Applicability

Protocol for Cross-Species AOP Network Development

A methodology for extending an AOP's tDOA, as demonstrated for a reproductive toxicity AOP for silver nanoparticles (AgNPs) in C. elegans (AOP 207), involves a multi-step integrative approach [79]:

• Data Collection and KE Matching: Assemble data from in vivo ecotoxicology studies, in vitro human toxicology models, and existing AOPs. Systematically match reported endpoints (e.g., gene expression, oxidative stress, fertility metrics) to standardized KE terms in the AOP-Wiki.
• AOP Network Assembly and Quantitative Assessment: Construct a putative cross-species AOP network linking the matched KEs. Evaluate the strength and uncertainty of the KERs using Bayesian Network (BN) modeling, a probabilistic approach well-suited for complex biological data [79].
• Taxonomic Domain Extension via In Silico Tools:
  • Perform a Genes-to-Pathways Species Conservation Analysis (G2P-SCAN) to identify the core gene set associated with the AOPN's KEs and assess its conservation across species.
  • Conduct a SeqAPASS analysis on critical molecular targets (e.g., the NADPH oxidase complex in AOP 207) to evaluate protein sequence conservation.
  • Synthesize results to propose a biologically plausible extended tDOA. This approach successfully extended the tDOA of AOP 207 to over 100 taxonomic groups [79].

Protocol for Molecular Annotation of AOP Key Events

To link AOPs to molecular data (e.g., transcriptomics) and facilitate cross-species comparison, a rigorous curation protocol was established [9]:

• Knowledge Graph Foundation: Use a Unified Knowledge Space (UKS) built on Neo4j, integrating AOP data from the AOP-Wiki with gene sets from pathways (WikiPathways, KEGG, Reactome), Gene Ontology (GO), and phenotypes (HPO).
• NLP-Based Matching: Preprocess KE descriptions and gene set names (tokenization, lemmatization). Calculate a weighted Jaccard Index (JIW) based on token matches, weighting rare, informative terms more heavily.
• Manual Curation and Consolidation: Experts manually evaluate the top computational matches, remove irrelevant ones, and fill gaps by searching curated databases. The goal is to annotate each KE with a specific, biologically relevant set of genes that can be measured via omics technologies, thereby creating a bridge between toxicogenomics and the AOP framework.

Table 1: Comparative Analysis of AOP Characteristics Across Major Taxonomic Groups

Taxonomic Group	Exemplar AOP / Network	Conserved Upstream Events (Broad tDOA)	Taxon-Specific Downstream Events (Narrow tDOA)	Primary Evidence Sources
Mammals	AOPN for Androgen Inhibition & Shortened AGD [81]	Androgen receptor antagonism; altered steroidogenesis.	Development of male reproductive tract; anogenital distance.	Mouse, rat, human (epidemiological).
Fish (Teleosts)	Various AOPs for Embryonic Development.	AHR activation; oxidative stress.	Yolk sac absorption; swim bladder inflation.	Zebrafish, fathead minnow, medaka.
Invertebrates (Nematodes)	AOP 207: AgNP-induced Reproductive Failure [79] [82]	NADPH oxidase activation; MAPK signaling; ROS formation.	Germline apoptosis; reduced brood size.	Caenorhabditis elegans.
Birds	Putative via tDOA extension [79].	Core oxidative stress pathway genes (via G2P-SCAN).	Avian-specific reproductive or developmental outcomes.	In silico extrapolation from conserved pathways.

Diagram 1: Workflow for Assessing AOP Taxonomic Applicability (76 characters)

Comparative Analysis of Strengths and Limitations

Strengths of the AOP Framework Across Taxa

• Mechanistic Transparency: AOPs make toxicological mechanisms explicit, allowing scientists to evaluate which KEs are likely conserved based on fundamental biology. This is more reliable than black-box correlative approaches.
• Facilitation of Read-Across: The framework provides a structured, hypothesis-driven basis for extrapolating chemical hazards from data-rich to data-poor species, a critical need in ecotoxicology [79].
• Integration of Diverse Data: AOPs can unify data from in silico, in vitro, and in vivo studies across different species into a coherent narrative, as demonstrated in the AgNP case study [79].
• Foundation for NAMs: By identifying conserved, measurable KEs, AOPs directly guide the development of non-animal testing strategies (e.g., high-throughput in vitro assays, biomarkers) for use in regulatory decision-making [9] [59].

Limitations and Challenges in Cross-Taxa Application

• Inconsistent Annotation: Taxonomic metadata in the AOP-Wiki is often incomplete or buried in free text, hampering automated analysis. Tools like Biovista Vizit are addressing this by scanning full text [80].
• Knowledge Gaps: For many taxonomic groups, especially non-model organisms, empirical data for KERs is sparse, making quantitative confidence assessment (qAOP development) difficult [4].
• Complexity of Biological Conservation: A conserved MIE does not guarantee a conserved AO. Divergent physiology, life history, and compensatory mechanisms can alter pathway outcomes in different species.
• Technical Barrier: Effective use of bioinformatics tools (SeqAPASS, G2P-SCAN, BN modeling) requires specialized expertise, potentially limiting their adoption.

Table 2: Evaluation Metrics for AOP Taxonomic Applicability Assessment

Metric Category	Specific Metric	Interpretation for Taxonomic Applicability
Empirical Evidence	Number of supporting species per KE/KER.	Direct evidence breadth. Sparse data limits confidence.
In Silico Conservation	SeqAPASS similarity score (0-100%) for MIE target.	High score suggests molecular susceptibility is plausible.
In Silico Conservation	G2P-SCAN pathway conservation p-value.	Low p-value indicates pathway is evolutionarily conserved.
Quantitative Confidence	Bayesian Network confidence for KERs.	High probabilistic confidence supports network robustness across data sources.
FAIRness	Completeness of taxonomic metadata (structured fields).	Enables machine-actionable search and integration [59].

Table 3: Key Research Reagent Solutions for AOP Development & Analysis

Tool / Resource	Type	Primary Function in AOP Research
AOP-Wiki	Knowledge Repository	The central, crowd-sourced repository for publishing and browsing AOPs, KEs, and KERs [9] [82].
Biovista Vizit	Knowledge Graph Tool	Mines the AOPKB and biomedical literature to visualize relationships, tag entities, and extract hidden taxonomic and methodological context [80].
SeqAPASS	Bioinformatics Tool	Predicts protein target conservation across species to support tDOA expansion for MIEs [79].
G2P-SCAN (R package)	Bioinformatics Tool	Evaluates the conservation of biological pathways and gene sets across taxa to support tDOA expansion for series of KEs [79].
OBI (Ontology for Biomedical Investigations)	Controlled Vocabulary	Provides standardized terms for annotating experimental methods used to generate AOP evidence, facilitating integration and search [80].
Unified Knowledge Space (UKS)	Data Integration Framework	A Neo4j-based knowledge graph that systematically links AOP components to genes, pathways, and phenotypes, enabling computational annotation [9].

Diagram 2: Modular AOP Network Structure with Taxon-Specific Outcomes (83 characters)

The future of robust cross-taxa AOP analysis hinges on enhancing the FAIRness of AOP data. The 2025 FAIR AOP Roadmap outlines coordinated efforts to standardize annotations, including taxonomic applicability, making AOPs more machine-actionable and interoperable with other biological databases [59]. This will accelerate the use of artificial intelligence and natural language processing to mine existing literature and populate knowledge gaps. Furthermore, the development of quantitative AOPs (qAOPs) with probabilistic predictions, often using Bayesian approaches, will allow for more nuanced risk estimations across species [79]. Emerging frameworks like Cost Outcome Pathways (COPs), which link AOs to socio-economic burdens, may also benefit from clear taxonomic definitions to accurately assess impacts on ecosystem services and human populations [48].

In conclusion, a comparative analysis of AOPs reveals a powerful but maturing framework. Its strength lies in providing a common language and mechanistic logic for integrating diverse data across the tree of life. Its primary limitation is the patchy and inconsistent annotation of taxonomic applicability. By leveraging emerging bioinformatics tools, adhering to FAIR data principles, and focusing on the development of modular AOP networks with clearly defined conserved and taxon-specific modules, the scientific community can significantly advance the reliable application of AOPs for predictive toxicology and risk assessment in a multi-species context.

The Adverse Outcome Pathway (AOP) framework is a conceptual construct that organizes mechanistic biological knowledge from a molecular initiating event (MIE) to an adverse outcome (AO) relevant to risk assessment [8]. An AOP describes a sequential chain of causally linked key events (KEs) at different biological levels (e.g., molecular, cellular, organ, organism) that lead to an adverse effect following exposure to a stressor [83] [8]. This framework is central to the paradigm shift in toxicology towards mechanistic, human-relevant, and predictive science, supporting the development and use of New Approach Methodologies (NAMs) [83] [84].

The utility of AOPs diverges significantly between human health risk assessment (HHRA) and ecological risk assessment (ERA). In HHRA, the AO is typically an adverse health effect in an individual human (e.g., cancer, organ toxicity), which informs the derivation of human health reference values (HHRVs) [85]. In ERA, the ultimate AO often pertains to impacts on population viability, community structure, or ecosystem function [63] [86]. This fundamental difference in endpoint selection drives variations in AOP development, validation strategies, and application. Understanding these differences is critical within the broader thesis of AOP taxonomic applicability, which examines how AOP knowledge is structured, evaluated, and applied across species and biological contexts to enable predictive toxicology and next-generation risk assessment [87] [83].

Comparative Analysis: Scope, Endpoints, and Application

The development and application of AOPs are tailored to the distinct goals of human health and ecological protection. The following table summarizes the core differences.

Table 1: Core Differences Between Human Health and Ecological AOP Applications

Aspect	Human Health Risk Assessment (HHRA)	Ecological Risk Assessment (ERA)
Primary Adverse Outcome (AO)	Disease, organ dysfunction, impaired development or reproduction in an individual human [8].	Effects on population survival, growth, reproduction, or community/ecosystem structure and function [63] [86].
Key Taxonomic Focus	Homo sapiens. Extrapolation from model organisms (e.g., rat, zebrafish) requires bridging data [83].	Multiple ecologically relevant species (fish, invertebrates, amphibians, plants). Species sensitivity distributions are key [63] [86].
Regulatory Context	Derivation of Human Health Reference Values (HHRVs) [85], chemical safety under laws like TSCA, drug development safety pharmacology.	Protection of natural resources under laws like the Clean Water Act; requires evaluating impacts on listed species and habitats [86].
Validation Emphasis	Establishing human biological relevance of KEs, often using human cell-based NAMs and translational biomarkers [8] [84].	Establishing quantitative linkages from molecular KEs to population-relevant effects (e.g., survival, fecundity) via modeling [63].
Data Integration	Often integrated into IATA (Integrated Approaches to Testing and Assessment) for chemical classification and HHRV derivation [83] [85].	Used in population models to extrapolate individual-level effects to population-level consequences [63].
Major Challenge	Capturing systemic complexity (e.g., endocrine, nervous, immune systems) and inter-individual variability in in vitro systems [83] [8].	Accounting for ecological complexity: species interactions, environmental stressors, and exposure variability in the field [63] [86].

A recent comprehensive mapping of the AOP-Wiki database reveals distinct thematic priorities in developed AOPs [83]. As of May 2023, the database contained 403 unique AOPs, with only 29 formally endorsed by the OECD [83]. The analysis shows a strong research focus on diseases of the genitourinary system, neoplasms (cancer), and developmental anomalies for human health [83]. For ecological contexts, AOPs frequently center on reproductive impairment, growth inhibition, and mortality in keystone species. Current international efforts, such as the EU's PARC (Partnership for the Assessment of Risks from Chemicals) initiative, prioritize case studies in three key areas that span both human and ecological concerns: immunotoxicity/non-genotoxic carcinogenesis, endocrine/metabolic disruption, and developmental/adult neurotoxicity [83]. This mapping helps identify both well-defined biological areas and critical research gaps for future AOP development.

Case Examples and Experimental Protocols

Human Health Case: Thyroid Disruption Leading to Developmental Neurotoxicity

This AOP describes how chemical inhibition of thyroid hormone synthesis (MIE) can lead to impaired brain development and cognitive deficits (AO) in children [8].

Experimental Protocol for In Vitro KE Measurement (Thyroperoxidase Inhibition):

• System: Use a rat thyroid follicular cell line (FRTL-5) or human primary thyrocytes cultured in hormone-depleted medium.
• Exposure: Treat cells with the test chemical (e.g., a perchlorate, disulfiram, or pesticide) across a concentration range for 24-72 hours. Include a positive control (e.g., methimazole) and vehicle control.
• Key Event Measurement (Iodide Organification):
  • Method: Use a modified Campbell's method. Briefly, incubate cells with Na¹²⁵I for a defined period.
  • Processing: Wash cells to remove free iodide. Precipitate cellular protein using cold trichloroacetic acid (TCA).
  • Quantification: Measure the radioactivity in the TCA-precipitated fraction (organified iodine) using a gamma counter.
  • Endpoint: Calculate the percentage of iodide organification inhibition relative to the vehicle control. Concentration-response modeling yields an IC₅₀ value.
• Linkage to AOP: The quantified in vitro KE data (IC₅₀) serves as a point of departure for predicting the in vivo dose required to reduce thyroid hormone levels, which is then integrated into physiologically based kinetic (PBK) models to estimate a human-relevant hazard dose [8].

Ecological Case: Aromatase Inhibition Leading to Population Decline in Fish

This AOP links the inhibition of the cytochrome P450 aromatase enzyme (MIE), which converts androgens to estrogens, to population collapse via impacts on sexual differentiation and fecundity [63].

Experimental Protocol for Population-Level Modeling:

• Individual-Level Effect Assessment:
  • Test System: Conduct a Fish Sexual Development Test (FSDT - OECD TG 234) with zebrafish or fathead minnows exposed from embryo to ~60 days post-fertilization.
  • Endpoint Measurement: At test termination, histologically sex individuals. The key metric is the percentage of phenotypic males in the treated groups versus controls.
  • Data Output: Generate a concentration-response relationship for the incidence of male-skewed sex ratio.
• Population Model Translation (as described by Kramer et al., 2011 [63]):
  • Model Selection: Use an age- or stage-structured population matrix model (e.g., a Leslie matrix) for the studied species.
  • Parameterization: The primary vital rate affected is fecundity (F). The sex ratio data from the FSDT is translated into a reduction in effective fecundity. For example, if a treatment results in 90% males, the egg production for that cohort is modeled as 10% of the baseline.
  • Model Simulation: Run the model over multiple generations with and without the chemical stressor.
  • Population Endpoint: Key outputs include the intrinsic population growth rate (λ), time to extinction, or minimum viable population size. A λ < 1 indicates a declining population.
• Linkage to AOP: The individual-level KE (altered sex ratio) is quantitatively bridged to the population-level AO (growth rate decline) through the model, providing a direct line of evidence for ecological risk characterization [63].

Visualizing AOP Development and Application

Core AOP Structure and Taxonomic Context

Diagram 1: Generalized AOP structure and taxonomic context.

AOP Validation and Integration Workflow for Risk Assessment

Diagram 2: AOP validation workflow for risk assessment.

The Scientist's Toolkit: Essential Research Reagent Solutions

The development and application of AOPs require a suite of specialized tools, reagents, and models. This toolkit bridges molecular biology, computational science, and ecology.

Table 2: Essential Research Toolkit for AOP Development and Application

Tool/Reagent Category	Specific Examples	Primary Function in AOP Context
Bioinformatic & Curation Tools	AOP-Wiki [83], AOP-KB [8], AOP-helpFinder [83], DisGeNET/Gene Ontology [83]	Knowledge Assembly & Gap Analysis: Crowdsourced AOP development, literature mining for KE evidence, mapping AOPs to disease or pathway ontologies.
In Vitro Bioassay Systems	Stable reporter gene assays (ER/AR/TR); differentiated human iPSC-derived neurons/ hepatocytes; zebrafish embryo (FET test).	KE Measurement & NAMs: Provide human- or species-relevant systems for quantifying specific MIEs and cellular KEs (e.g., receptor binding, gene activation, cytotoxicity).
OMICs Profiling Kits	RNA-Seq libraries; targeted mass spectrometry panels for phosphoproteins or metabolites; multiplex ELISA for cytokines.	Mechanistic Discovery & Pathway Identification: Enable unbiased identification of novel KEs and characterization of key event relationships following stressor exposure.
Reference Databases	PubChem; ChEMBL; Comparative Toxicogenomics Database (CTD); US EPA CompTox Chemicals Dashboard.	Stressor Characterization & Read-Across: Provide chemical structure, property, and bioactivity data to inform MIE and support grouping for AOP applicability.
Computational Modeling Software	Population modeling (RAMAS, Vortex); PBPK modeling (GastroPlus, Simcyp); AOP network analysis (Cytoscape).	Quantitative Extrapolation: Translate in vitro or individual-level effect data to human pharmacokinetic doses or population/ecosystem-level risks [63].
Standardized Test Guidelines	OECD TG 458 (ER TA), TG 455 (AR TA), TG 234 (FSDT), TG 443 (ER/ND EVV), EPA OCSPP 890.2200 (T3).	Regulatory Confidence: Provide internationally accepted, validated protocols for generating reliable KE data that can support regulatory acceptance of AOP-informed assessments [84].

The ongoing FAIRification (Findable, Accessible, Interoperable, Reusable) of AOP data and metadata is critical to maximizing the utility of this toolkit [87]. This involves implementing standardized formats, controlled vocabularies, and computational workflows to ensure AOP knowledge is structured for machine readability and integration, thereby accelerating its use in next-generation risk assessment [87] [83].

The Role of Quantitative AOPs (qAOPs) in Strengthening Extrapolation Predictions

The Adverse Outcome Pathway (AOP) framework has emerged as a pivotal conceptual tool in modern toxicology, providing a structured representation of the mechanistic sequence of events from a molecular initiating event (MIE) to an adverse outcome of regulatory concern [54]. This framework organizes biological knowledge into a series of measurable key events (KEs), linked by key event relationships (KERs), effectively creating a "series of biological dominos" [54]. While qualitative AOPs have significantly enhanced hazard identification, their application in quantitative risk assessment and extrapolation has been limited due to their descriptive nature [88] [89].

The transition to Quantitative AOPs (qAOPs) represents a critical evolution, embedding mathematical descriptions of dose-response and response-response relationships within the AOP structure [88] [90]. This quantification is fundamental for moving from hazard identification to predictive risk assessment, enabling the extrapolation of effects across doses, from in vitro to in vivo systems, and across different species [88]. Framed within a broader thesis on AOP taxonomic applicability, this guide explores how qAOPs strengthen extrapolation predictions by providing a formal, computational bridge between mechanistic toxicology and regulatory decision-making. The development and application of qAOPs address a core challenge in 21st-century toxicology: reliably predicting adverse effects in humans and ecological species using non-animal New Approach Methodologies (NAMs) [91].

Foundational Concepts: From Qualitative AOPs to Quantitative Predictive Models

Core Definitions and Principles

An AOP is a chemically agnostic, modular framework that describes a biologically plausible sequence of events [54]. It originates with a Molecular Initiating Event (MIE), defined as the initial interaction between a stressor and a biomolecule. This triggers a cascade of Key Events (KEs), which are measurable changes at different levels of biological organization (cellular, tissue, organ, organism). The logical and causal links between KEs are described as Key Event Relationships (KERs) [54]. The adverse outcome is an effect relevant to risk assessment, such as organ dysfunction or reduced population fitness [88].

The AOP framework is built on several key principles. It is not stressor-specific, meaning one AOP can describe the pathway for a class of chemicals sharing an MIE. It is modular, allowing KEs and KERs to be shared and connected into larger AOP networks, which better represent biological complexity. Finally, AOPs are living documents that evolve with new scientific evidence [54].

The Imperative for Quantification: qAOPs

A qualitative AOP establishes a hypothetical causal chain. A qAOP transforms this chain into a predictive model by mathematically formalizing the KERs [88]. According to recent workshops, this quantification is essential for leveraging AOPs in higher-tier chemical safety assessments within regulatory frameworks like the EU's [91].

Quantification can be applied at different levels of completeness [88]:

• Quantitative KER: A model defining the dose-response or response-response relationship for a single pair of linked KEs.
• Partial qAOP: A model quantifying relationships for more than one, but not all, KERs in a pathway.
• Full qAOP: A mathematical construct that models the dose-response or response-response relationships for all KERs described in an AOP.

This quantification allows for the prediction of the magnitude or probability of a downstream adverse outcome based on the measurement or prediction of an upstream event, which is the cornerstone of reliable extrapolation [88].

Quantitative Extrapolation: Core Applications of qAOPs

qAOPs are specifically designed to address major extrapolation challenges in toxicology. Their predictive power is derived from the mathematical formalization of biological pathways, enabling quantitative translations that are not possible with qualitative frameworks.

• In Vitro to In Vivo Extrapolation (IVIVE): qAOPs provide a biological context for translating effects measured in cell-based systems (often at the MIE or early KE level) to outcomes at the tissue, organ, or whole-organism level. This requires integrating toxicokinetic (TK) models to bridge the gap between the external concentration applied in vitro and the internal dose at the target site in vivo [88] [89]. A qAOP defines what happens biologically, while a TK model defines how much of the stressor reaches the target to initiate the sequence.

• Cross-Species Extrapolation: A primary uncertainty in ecological and human health risk assessment is extrapolating toxicity data from tested to untested species [54]. qAOPs facilitate this by focusing on the conservation of biological pathways. If the sequence of KEs and KERs is evolutionarily conserved, the qAOP structure can be transferred across species. The quantitative differences—such as varying sensitivity thresholds, reaction rates, or physiological rates—are then captured in the parameter values of the model. For example, a model linking estrogen receptor activation to reproductive impairment can be applied across fish species, with species-specific parameters adjusting the quantitative output [54].

• Dose-Response Extrapolation: qAOPs enable prediction of effects at low, environmentally relevant doses based on data from higher-dose studies. By modeling the progression of effects along the pathway, qAOPs can identify non-linearities, thresholds, and compensatory feedback mechanisms that simple empirical curve-fitting might miss [88].

• Chemical and Mixture Extrapolation: Since AOPs are not chemical-specific, a developed qAOP for a given MIE can be applied to any chemical triggering that same MIE. The chemical-specific properties are introduced via toxicokinetic and physicochemical data, allowing the model to predict different potencies and outcomes for different stressors sharing a common mechanism [89]. For chemical mixtures, AOP networks can identify convergent KEs, helping to predict additive or synergistic effects [54].

The following table summarizes the primary types of quantitative relationships that are formalized within qAOPs to enable these extrapolations.

Table 1: Types of Quantitative Key Event Relationships (KERs) in qAOPs [88]

KER Type	Description	Common Mathematical Form	Extrapolation Utility
Dose-Response	Relates the concentration/dose of a stressor to the magnitude of a specific Key Event (KE).	Sigmoidal (Hill), Power, Linear functions.	Links external exposure to initial biological perturbation; foundational for IVIVE.
Response-Response	Relates the magnitude of an upstream KE to the magnitude of a downstream KE.	Linear regression, Power laws, Ordinary Differential Equations (ODEs).	Predicts progression of effect through the pathway; core of cross-species and dose extrapolation.
Temporal	Defines the time delay or dynamic sequence between the occurrence of two KEs.	Time-lag functions, Systems of ODEs.	Critical for modeling chronic effects and understanding recovery potential.

Building a qAOP: Methodologies and Experimental Protocols

The qAOP Development Workflow

Developing a qAOP is an iterative process that begins with a well-defined qualitative AOP. The workflow, as outlined by experts, involves several key stages [88] [90]:

• Problem Formulation: Precisely define the risk assessment question, the required prediction (e.g., a point of departure for a specific adverse outcome), and the applicability domain (species, life stage).
• AOP Selection/Development: Identify an existing qualitative AOP from the AOP-Wiki or develop a new one, ensuring sufficient weight of evidence for the KERs.
• Data Curation & Gap Analysis: Gather all existing quantitative data for the KERs (dose-response, temporal, response-response). Identify critical data gaps that prevent quantification.
• Model Assembly & Parameterization: Select appropriate modeling approaches (see Section 4.2) for each KER. Integrate them into a coherent model and parameterize them using curated data.
• Model Evaluation & Validation: Test the qAOP's predictive performance against independent datasets not used in its construction. Evaluate uncertainty and sensitivity.
• Documentation & Communication: Transparently document all assumptions, data sources, model code, and performance metrics to facilitate review and regulatory acceptance [88].

Diagram 1: The iterative qAOP development workflow (Max width: 760px).

Modeling Approaches for qAOPs

The choice of modeling technique depends on the biological complexity of the KER, the available data, and the assessment question [88]. Approaches range from empirical to mechanistic:

• Statistical & Regression Models: Used for direct, empirical quantification of KERs where rich data exists (e.g., linear regression between plasma vitellogenin and egg production).
• Bayesian Networks: Graphically represent probabilistic dependencies among KEs. They are powerful for integrating diverse data types and quantifying uncertainty, useful when data is sparse or heterogeneous.
• Ordinary Differential Equations (ODEs): Capture the dynamic, time-dependent interactions between biological components (e.g., signaling cascades, feedback loops). They offer high biological fidelity but require significant data for parameterization.
• Toxicokinetic-Toxicodynamic (TKTD) & Physiologically Based Kinetic (PBK) Models: These are not alternatives but essential integrations. TK/PBK models simulate the absorption, distribution, metabolism, and excretion (ADME) of a chemical to predict internal target site concentrations. This output becomes the input dose for the qAOP's MIE, enabling true IVIVE [89].

Detailed Experimental Protocol: A Case Study in Fish

The following protocol is derived from a seminal study that developed a qAOP network for the synthetic glucocorticoid beclomethasone dipropionate (BDP) in fathead minnows, explicitly investigating the role of internal exposure dynamics [89].

Table 2: Key Experimental Data from BDP qAOP Case Study [89]

Exposure Regime	Target Water Conc.	Measured Plasma Conc. (BDP+17-BMP)	Area Under Curve (AUC) Ratio (Exp2/Exp1)	Key Adverse Outcome Observed
Experiment 1: Oscillating	1000 ng/L (peak)	4.0 ± 2.0 ng/mL	1.0 (Baseline)	Mild gluconeogenesis disruption
Experiment 2: Sustained	1000 ng/L (constant)	60 ± 35 ng/mL	~15	Severe immunodepression, skin androgenisation

Objective: To quantify the relationships between waterborne BDP exposure, internal plasma concentrations of its active metabolites, and a network of adverse outcomes (immunodepression, skin androgenisation, gluconeogenesis disruption).

Materials & Organisms:

• Test Organism: Sexually mature fathead minnows (Pimephales promelas).
• Test Chemical: Beclomethasone dipropionate (BDP).
• Exposure System: Flow-through or semi-static aquarium systems with temperature, photoperiod, and water quality control.
• Analytical Equipment: Liquid Chromatography with Tandem Mass Spectrometry (LC-MS/MS) for quantifying BDP and its metabolites (17-BMP, BOH) in water and fish plasma.

Experimental Procedure:

• Exposure Design: Conduct two parallel 21-day exposures.
  • Regime A (Oscillating): Expose fish to a pulsed pattern where BDP water concentration peaks at a high level (e.g., 1000 ng/L) for 10 hours every 4 days, with low background concentration between peaks.
  • Regime B (Sustained): Expose fish to a constant nominal concentration of BDP (e.g., 1000 ng/L) for the entire 21 days.
• Water Sampling & Analysis: Regularly collect water samples from each exposure tank throughout the study. Analyze via LC-MS/MS to confirm measured concentrations and characterize the exposure dynamics (peak, trough, time-integrated AUC).
• Biological Sampling: At termination (Day 21), euthanize fish and collect blood via caudal puncture using heparinized syringes. Centrifuge blood immediately to separate plasma. Snap-freeze plasma in liquid nitrogen and store at -80°C.
• Plasma Analysis: Extract steroids from plasma samples. Use LC-MS/MS to quantify concentrations of BDP and its metabolites (17-BMP, BOH). This provides the internal dose metric.
• Endpoint Assessment: Perform a suite of analyses on the exposed fish to measure KEs and adverse outcomes:
  • Immunodepression: Perform a functional immune assay (e.g., lymphocyte proliferation assay) or measure biomarkers of immune function.
  • Skin Androgenisation: In males, quantify the development of nuptial tubercles, a secondary sex characteristic mediated by androgens.
  • Gluconeogenesis Disruption: Measure plasma glucose levels and/or hepatic expression of key gluconeogenic enzymes (e.g., PEPCK).
  • Histopathology: Examine tissue sections (e.g., liver, skin) for pathological changes.

Data Integration & Modeling:

• Construct a pharmacokinetic (PK) model to describe the uptake, metabolism, and elimination of BDP, linking water AUC to plasma AUC of active metabolites.
• Develop response-response models (e.g., using regression or ODEs) linking the plasma AUC of active metabolites to the magnitude of each measured KE (e.g., plasma glucose level, immune cell count).
• Integrate the PK and pharmacodynamic (PD) models to create a predictive qAOP network. This model can then predict the severity of multiple adverse outcomes based on any given water exposure scenario.

Developing and applying qAOPs requires a multidisciplinary toolkit. The following table details essential resources, drawing from the case study and general qAOP development needs.

Table 3: Research Reagent Solutions for qAOP Development [88] [90] [89]

Category	Item / Resource	Function in qAOP Research
Biological Resources	Relevant In Vitro Assays (e.g., receptor binding, gene reporter, high-content imaging).	Measure Molecular Initiating Events (MIEs) and early cellular Key Events (KEs) for hazard identification and dose-response data.
	Well-Characterized In Vivo Models (e.g., fathead minnow, zebrafish, rat).	Provide in vivo data for quantifying response-response KERs and validating integrated qAOP predictions.
Analytical Tools	LC-MS/MS Systems	Quantify stressor and metabolite concentrations in exposure media and biological matrices (e.g., plasma) for toxicokinetic analysis.
	Transcriptomic & Proteomic Platforms (RNA-Seq, mass spectrometry).	Identify novel KEs, provide mechanistic support for KERs, and generate quantitative molecular data for model parameterization.
Computational Resources	AOP-Wiki (aopwiki.org)	Central repository for qualitative AOP knowledge; starting point for qAOP development.
	TK/TD & PBK Modeling Software (e.g., GNU MCSim, Simcyp, Berkeley Madonna).	Simulate internal exposure dynamics and integrate them with qAOP models for IVIVE.
	Statistical & ODE Modeling Environments (e.g., R, Python with SciPy/NumPy).	Develop and compute the mathematical functions describing quantitative KERs.
Reference Chemicals	Protypical Stressors (e.g., Beclomethasone dipropionate, 17-BMP [89]).	Well-studied chemicals used as positive controls to perturb specific pathways and generate high-quality data for qAOP parameterization.

Current Challenges and Future Directions

Despite their significant promise, the widespread regulatory adoption of qAOPs faces several hurdles, as highlighted in recent scientific workshops [91].

• Incomplete Coverage and Quantification: Existing AOPs qualitatively cover only a fraction of known toxicities, and far fewer have been quantitatively elaborated. Critical areas like developmental neurotoxicity lack robust pathways [91].
• Lack of Standardized Evaluation: There are currently no published, universally accepted standards for characterizing, validating, and reporting qAOPs. This creates uncertainty for regulators regarding model reliability and appropriateness for decision-making [91].
• Data Gaps and Integration Complexity: Building high-fidelity qAOPs often requires data types (e.g., temporal response-response) that are not routinely generated in standard toxicity tests. Integrating diverse data streams (omics, in vitro, in vivo) into a coherent model remains technically challenging.
• Knowledge and Training Gaps: The interdisciplinary nature of qAOPs—requiring toxicology, biology, mathematics, and computational skills—creates a training need for both developers and assessors [91].

The future of the field hinges on addressing these challenges. Key recommendations from the scientific community include [91]:

• Developing community-endorsed best practice guidelines for qAOP model development, evaluation, and documentation.
• Prioritizing the creation and regulatory review of demonstrative case studies to build confidence in qAOP predictions.
• Establishing curated data repositories with high-quality, standardized datasets suitable for quantifying KERs.
• Implementing education and training programs to build capacity in quantitative and computational toxicology.

Quantitative Adverse Outcome Pathways represent a transformative advancement in predictive toxicology. By embedding mathematical rigor into the AOP framework, qAOPs transform conceptual pathway diagrams into computational tools capable of strengthening extrapolation predictions—from in vitro to in vivo, across species, and between exposure scenarios. As evidenced by case studies, the integration of toxicokinetics and dynamic biological modeling is critical for accurate prediction [89]. The ongoing work to standardize development practices, coupled with the generation of robust case studies, is essential for fulfilling the promise of qAOPs. Ultimately, their successful implementation will enable more efficient, mechanism-based, and animal-sparing chemical safety assessments, solidifying their role within the broader thesis of taxonomic applicability in environmental and human health protection.

The Adverse Outcome Pathway (AOP) framework has revolutionized toxicology by providing a structured, mechanistic model linking molecular perturbations to adverse health outcomes, facilitating the use of New Approach Methodologies (NAMs) and supporting regulatory decision-making [8] [1] [54]. However, a critical gap exists in its capacity to translate these biological outcomes into socio-economic consequences, which are vital for holistic risk management and policy prioritization [48] [92]. This whitepaper introduces and elaborates on the novel Cost Outcome Pathway (COP) framework, a direct extension of the AOP paradigm designed to bridge this gap. A COP formally connects an Adverse Outcome (AO) to downstream Cost Outcomes (COs), quantifying the societal and economic burden of environmentally-induced diseases [48] [93]. Framed within the imperative to define the taxonomic applicability of AOPs, this document provides an in-depth technical guide on constructing and utilizing COPs. It details foundational concepts, integrates bioinformatics tools for taxonomic domain validation, outlines experimental and in silico protocols for COP development, and presents a curated toolkit for researchers. The integration of COP into the AOP workflow enables a more comprehensive assessment from molecular initiation to societal impact, offering a powerful tool for researchers, risk assessors, and policymakers to prioritize interventions based on both human health and economic evidence.

The Adverse Outcome Pathway (AOP) is a conceptual construct that organizes existing knowledge into a sequential chain of causally linked events, from a Molecular Initiating Event (MIE) to an Adverse Outcome (AO) relevant to risk assessment [8] [22]. It serves as a translational tool, allowing data from high-throughput in vitro assays and other NAMs to be reliably used to predict apical outcomes traditionally derived from animal testing [1] [94]. This framework is modular, chemically agnostic, and structured around measurable Key Events (KEs) linked by Key Event Relationships (KERs) [22] [54]. Its development and curation, overseen by organizations like the OECD and supported by repositories like the AOP-Wiki, have made it a cornerstone of modern, mechanistic toxicology [1] [54].

A pivotal, yet historically underexplored, aspect of AOP development is the Taxonomic Domain of Applicability (tDOA). The tDOA defines the range of species for which an AOP is biologically plausible and is critical for extrapolating findings from model organisms to humans or untested wildlife species [3]. Confidence in tDOA is established through evidence of structural and functional conservation of the KEs and KERs across taxa [3]. This taxonomic precision forms a foundational thesis for robust AOP application, ensuring that pathways are not just mechanistic descriptions but are accurately scoped for reliable extrapolation.

Despite its strengths, the AOP framework stops at the biological adverse outcome. It does not account for the subsequent socio-economic impacts—such as healthcare costs, lost productivity, and diminished quality of life—that result from population-level health effects [48] [93]. For policy makers tasked with resource allocation and chemical prioritization, this economic dimension is crucial [92]. The emerging Cost Outcome Pathway (COP) framework directly addresses this limitation. A COP extends an AOP by linking the AO to one or more Cost Outcomes (COs), thereby creating an integrated model that maps a toxicological cascade all the way to its societal cost [48]. This whitepaper positions the COP as the next logical frontier in pathway-based risk assessment, grounded in the rigorous taxonomic and mechanistic understanding mandated by advanced AOP science.

Foundational Concepts: From AOP to COP

Core Components of an Adverse Outcome Pathway (AOP)

An AOP is a linear, simplified representation of a toxicological process, structured as follows [8] [22] [54]:

• Molecular Initiating Event (MIE): The initial interaction between a stressor (e.g., a chemical) and a biomolecular target within an organism (e.g., receptor binding, protein inhibition).
• Key Events (KEs): Measurable, essential biological changes at various levels of organization (cellular, tissue, organ, organism) that occur following the MIE.
• Key Event Relationships (KERs): Descriptions of the causal linkages between KEs, supported by evidence of biological plausibility, empirical data, and quantitative understanding.
• Adverse Outcome (AO): An adverse effect at the individual or population level that is of regulatory concern (e.g., organ failure, cancer, population decline).

AOPs are designed to be modular, allowing KEs to be shared across different pathways, forming AOP networks that better reflect biological complexity [22] [54].

Defining the Cost Outcome Pathway (COP) Framework

The Cost Outcome Pathway (COP) framework extends the logic of the AOP beyond biological adversity into the socio-economic domain [48] [93].

• Definition: A COP is a conceptual sequence that links an Adverse Outcome (AO) of an AOP to one or more Cost Outcomes (COs). It formalizes the causal chain through which a population-level health effect generates a socio-economic burden.
• Core Component – Cost Outcome (CO): A measurable socio-economic consequence linked to the AO. This is an umbrella term encompassing various metrics such as direct healthcare costs, lost productivity (indirect costs), and composite measures like Disability-Adjusted Life Years (DALYs) [48] [92].
• Linkage: The connection between the AO and CO is established through evidence-based relationships, often informed by epidemiological data and health economic models. For example, an AO of "decreased IQ in children" can be linked to a CO of "increased socio-economic burden" via models that estimate lifetime earning loss attributable to cognitive deficit [48] [93].

Table 1: Comparative Overview of AOP and COP Frameworks

Feature	Adverse Outcome Pathway (AOP)	Cost Outcome Pathway (COP)
Primary Objective	Organize mechanistic toxicological knowledge to predict health hazards [1].	Quantify the socio-economic burden of health hazards to inform policy [48].
Initiating Event	Molecular Initiating Event (MIE) – chemical-biological interaction [8].	Adverse Outcome (AO) – the health effect of regulatory concern.
Sequential Events	Key Events (KEs) – biological changes at increasing levels of organization [22].	Cost Events / Relationships – steps linking health outcome to economic cost (e.g., healthcare utilization, productivity loss).
Final Outcome	Adverse Outcome (AO) – individual or population-level health effect [54].	Cost Outcome (CO) – quantifiable socio-economic impact (e.g., monetary cost, DALYs) [92].
Key Evidence	Biological plausibility, empirical toxicity data, essentiality of KEs [22].	Epidemiological association, health economic data, burden of disease studies [93].
Primary Utility	Hazard identification, support for NAMs, chemical prioritization for testing [8] [54].	Risk management, cost-benefit analysis, policy prioritization and justification [48].

Taxonomic Applicability: A Foundational Thesis for Reliable Pathways

A scientifically robust AOP, and by extension a credible COP, must explicitly define its taxonomic domain of applicability (tDOA) [3].

• Concept of tDOA: The tDOA is the range of taxa for which there is credible scientific evidence that the described pathway (its KEs and KERs) is conserved and operative. It addresses a central uncertainty in risk assessment: extrapolating findings across species [3].
• Evidence for Conservation: Establishing tDOA relies on assessing both structural conservation (e.g., presence and similarity of a protein target) and functional conservation (e.g., the protein performs the same role in the pathway) [3].
• Bioinformatics Tool – SeqAPASS: The Sequence Alignment to Predict Across Species Susceptibility (SeqAPASS) tool is a critical resource for evaluating structural conservation [3]. It performs a tiered analysis:
  • Level 1: Compares primary amino acid sequence similarity to identify potential orthologs.
  • Level 2: Evaluates conservation of specific functional domains.
  • Level 3: Assesses conservation of individual amino acid residues critical for chemical binding or protein function [3].
• Workflow for Defining tDOA: The process involves identifying the proteins central to each KE in an AOP, using SeqAPASS to evaluate their conservation across species of interest, and integrating these computational results with available empirical toxicity data to define a biologically plausible tDOA [3].

Integrated Experimental &In SilicoProtocol for COP Development

This protocol outlines a systematic approach for developing a COP, anchored in a well-characterized AOP and its tDOA.

Phase 1: AOP Foundation and Taxonomic Scoping

• Select and Characterize a Relevant AOP: Choose an AOP with a strong weight of evidence and a human-relevant AO (e.g., developmental neurotoxicity leading to decreased IQ) [48]. Extract all KEs and KERs from the AOP-Wiki.
• Define the tDOA for the AOP: a. For each KE involving a specific molecular target (e.g., receptor, enzyme), obtain the reference protein sequence (e.g., human UniProt ID). b. Input each sequence into the SeqAPASS web tool. Perform Level 1, 2, and 3 analyses for a broad range of vertebrate taxa, focusing on mammals and specifically Homo sapiens for human health COPs [3]. c. Document the degree of sequence, domain, and residue conservation. High conservation across mammals supports the plausibility of the AOP in humans.

Phase 2: Quantitative Bridging from AO to CO

This phase uses the neurodevelopmental COP (AO: Decreased IQ → CO: Increased Socio-economic Burden) as a case study [48] [93].

• Define the Quantitative Relationship between Exposure and AO: a. Data Source: Utilize epidemiological meta-analyses that provide a pooled effect estimate. For example, a study might report that a specific increase in prenatal exposure to a chemical (e.g., μg/L of a PFAS in blood) is associated with a β-coefficient of -0.5 IQ points (95% CI: -0.8, -0.2) in children [48]. b. Model: Establish a linear or log-linear concentration-response function: ΔIQ = β * ΔExposure.
• Define the Quantitative Relationship between AO and CO: a. Costing Metric: Select an appropriate CO metric. For IQ loss, a common approach is lifetime earning potential [48]. b. Economic Model: Apply a health economic formula. A simplified model is: Economic Loss = (ΔIQ) * (Earning Loss per IQ point) * (Affected Population Size). c. Parameterization: * ΔIQ: Derived from Step 1. * Earning Loss per IQ point: Obtain from longitudinal economic studies (e.g., a value of $X,XXX per IQ point lost, discounted to present value). * Affected Population Size: Estimated from exposure assessment data (number of births in an area with exposure above a certain threshold).

Phase 3:In SilicoImplementation and Uncertainty Analysis

• Build a Computational Model: Implement the quantitative relationships from Phase 2 in a computational environment (e.g., R, Python). The model should take exposure estimates as input and output a distribution of possible economic costs.
• Propagate Uncertainty: Use probabilistic methods (e.g., Monte Carlo simulation) to incorporate uncertainty from all parameters:
  • Uncertainty in the exposure-response coefficient (β).
  • Uncertainty in the economic valuation parameter (Earning Loss per IQ point).
  • Variability in exposure across the population.
• Sensitivity Analysis: Perform sensitivity analyses (e.g., using Sobol indices) to identify which input parameters contribute most to variance in the final cost estimate, guiding future research priorities.

Table 2: Key Research Reagent Solutions for AOP/COP Development

Tool / Resource Name	Type	Primary Function in COP Development	Access/Source
AOP-Wiki	Knowledge Repository	The central hub for qualitative AOP information. Used to identify and analyze the foundational AOP for COP extension [1] [54].	https://aopwiki.org
SeqAPASS	Bioinformatics Tool	Evaluates structural conservation of protein targets across species to define the Taxonomic Domain of Applicability (tDOA), ensuring the AOP is relevant to the species of interest (e.g., human) [3].	https://seqapass.epa.gov
Curated Gene & System Annotations [9]	Annotated Database	Provides manually curated associations between AOP Key Events and specific genes/pathways. Bridges mechanistic KEs to measurable molecular signals (e.g., from transcriptomics), supporting the development of NAMs for KEs [9].	Integrated into knowledge bases (e.g., via the publication's supplementary data).
OECD AOP Knowledge Base	Authoritative Database	Source for OECD-endorsed AOPs, which have undergone rigorous peer review and represent high-confidence pathways suitable as COP foundations [8] [94].	OECD AOP Portal
Health Economic Valuation Literature	Data Source	Provides critical parameters for linking AOs to COs (e.g., cost per case of disease, value of a statistical life year, productivity loss estimates) [48] [93].	Published literature, WHO Global Health Estimates.

Data Synthesis and Quantitative Analysis for COPs

The transition from a qualitative AOP to a quantitative COP requires synthesis of data from toxicology, epidemiology, and economics.

Table 3: Quantitative Data Synthesis for a Neurodevelopmental COP (Case Study)

Data Layer	Parameter	Example Value / Source	Role in COP
Toxicological/Epidemiological	Exposure-Response Coefficient (β)	-0.5 IQ points per unit exposure increase (e.g., log10 μg/L PFAS in serum) [48].	Quantifies the magnitude of the AO (IQ loss) per unit of population exposure.
Population Health	Baseline Incidence of AO	Prevalence of sub-clinical cognitive deficit in reference population.	Establishes the background rate of the outcome.
Exposure Assessment	Affected Population Size (N)	Estimated number of children with exposure above a threshold level in a given region/year.	Scales the individual-level effect to the population level.
Health Economic	Economic Cost per Unit AO	$18,000 - $24,000 in lifetime earning loss per IQ point lost (discounted present value) [48].	Converts the health outcome (ΔIQ) into a monetary Cost Outcome (CO).
Integrated Output	Total Attributable Economic Burden	ΔIQ * Cost per IQ point * N = Total Population Cost (e.g., $X Billion annually for a specific exposure scenario).	The final CO metric for policy consideration.

The integration of socio-economic impacts through the Cost Outcome Pathway (COP) framework represents a significant evolution of the AOP concept. By bridging the gap between mechanistic toxicology and health economics, COPs provide a more complete narrative for decision-making, from the initial molecular trigger to the ultimate societal cost. This alignment is essential for justifying regulatory actions and prioritizing research on chemicals with the greatest potential for public health and economic harm [48] [92].

Future advancements in this field will depend on several key developments:

• Standardization of COP Templates: Developing OECD-like guidance for COP development, including standardized taxonomies for Cost Outcomes and best practices for quantitative bridging [92].
• Advanced In Silico Tools: Creating integrated software platforms that seamlessly combine AOP networks (with defined tDOA), exposure models, and economic valuation functions to run automated COP-based impact assessments.
• Expansion to Other Disease Endpoints: Applying the COP framework to AOPs for other high-burden conditions such as cancer, metabolic disorders, and cardiovascular disease, where robust epidemiological and economic data exist [93].
• Dynamic and Probabilistic COPs: Moving beyond static point estimates to develop dynamic COPs that can project future costs under different regulatory or exposure scenarios, fully incorporating probabilistic uncertainty and variability analyses.

By embracing the COP framework within the rigorous context of taxonomically-applicable AOPs, the scientific and regulatory community can powerfully translate mechanistic insights into actionable economic evidence, ultimately fostering a more protective and sustainable approach to chemical risk management.

Conclusion

The systematic definition and validation of taxonomic applicability are not merely technical details but foundational to the credibility and utility of the AOP framework. By adhering to structured methodologies, embracing FAIR principles for data management, and rigorously applying weight-of-evidence assessments, researchers can transform AOPs from conceptual models into powerful, trusted tools for next-generation risk assessment. The future of AOPs lies in their expansion into complex, quantitatively defined networks that are explicitly annotated for biological scope, seamlessly integrated with toxicogenomics and AI-driven discovery, and potentially extended to evaluate socio-economic outcomes. This evolution will be crucial for reducing animal testing through confident application of New Approach Methodologies (NAMs) across biomedical and environmental toxicology, ultimately leading to more efficient and protective safety decisions.

References

• 1. The Adverse Outcome Pathway: A Multifaceted Framework ... [https://pmc.ncbi.nlm.nih.gov/articles/PMC5906804/]
• 2. Defining molecular initiating events in the adverse outcome ... [https://pubmed.ncbi.nlm.nih.gov/25354311/]
• 3. Defining the Biologically Plausible Taxonomic Domain of ... [https://pmc.ncbi.nlm.nih.gov/articles/PMC10100214/]
• 4. Practical approaches to adverse outcome pathway ... [https://academic.oup.com/etc/article/36/6/1429/7740225]
• 5. Developers' Handbook 2.5 [https://aopwiki.org/handbooks/3]
• 6. Translating Scientific Advances in the AOP Framework to ... [https://www.mdpi.com/2079-4991/10/6/1229]
• 7. A weight of evidence assessment approach for adverse ... [https://pubmed.ncbi.nlm.nih.gov/26724267/]
• 8. Adverse Outcome Pathways | US EPA [https://www.epa.gov/chemical-research/adverse-outcome-pathways]
• 9. A curated gene and biological system annotation of ... [https://www.nature.com/articles/s41597-023-02321-w]
• 10. Library of Identification Resources: a FAIR overview of ... [https://pmc.ncbi.nlm.nih.gov/articles/PMC12405937/]
• 11. On the modelling and application of the taxonomic coverage of ... [https://rogue-scholar.org/records/hqhgm-s4w93]
• 12. Towards a global list of accepted species II. Consequences ... [https://link.springer.com/article/10.1007/s13127-021-00518-8]
• 13. Role of Genomic Typing in Taxonomy, Evolutionary Genetics ... [https://pmc.ncbi.nlm.nih.gov/articles/PMC88989/]
• 14. Adverse Outcome Pathways—Organizing Toxicological ... [https://www.sciencedirect.com/science/article/abs/pii/S0022356524191926]
• 15. Adverse outcome pathways as tools to assess drug-induced ... [https://pmc.ncbi.nlm.nih.gov/articles/PMC5436615/]
• 16. Towards a quantitative adverse outcome pathway for liver ... [https://www.sciencedirect.com/science/article/pii/S2468111325000192]
• 17. Beyond QSARs: Quantitative Knowledge–Activity ... [https://pmc.ncbi.nlm.nih.gov/articles/PMC12646586/]
• 18. adverse-outcome-pathway-based in vitro assessment with ... [https://www.frontiersin.org/journals/toxicology/articles/10.3389/ftox.2025.1564857/full]
• 19. FAIR Principles [https://www.go-fair.org/fair-principles/]
• 20. Definitions [https://www.ecetoc.org/special-report-19/preface/definitions/]
• 21. Advancing the Adverse Outcome Pathway Framework [https://pmc.ncbi.nlm.nih.gov/articles/PMC6156781/]
• 22. PRACTICAL APPROACHES TO ADVERSE OUTCOME ... [https://pmc.ncbi.nlm.nih.gov/articles/PMC6058314/]
• 23. An integrative data mining approach to identifying adverse ... [https://www.sciencedirect.com/science/article/abs/pii/S0300483X16300348]
• 24. Development of a data-driven approach to Adverse ... [https://www.frontiersin.org/journals/toxicology/articles/10.3389/ftox.2023.1183824/full]
• 25. Application of the Adverse Outcome Pathway Framework to ... [https://pmc.ncbi.nlm.nih.gov/articles/PMC6197844/]
• 26. Adverse Outcome Pathways: Moving from a Scientific Concept ... [https://academic.oup.com/etc/article/38/6/1152/7737077]
• 27. Development of a data-driven approach to Adverse Outcome ... [https://pmc.ncbi.nlm.nih.gov/articles/PMC10203404/]
• 28. Providing Adverse Outcome Pathways from the AOP-Wiki in a ... [https://pmc.ncbi.nlm.nih.gov/articles/PMC8978481/]
• 29. AOPWIKI-EXPLORER: An interactive graph-based query ... [https://www.sciencedirect.com/science/article/pii/S2468111324000100?dgcid=rss_sd_all]
• 30. New Approach Methodologies for the Endocrine Activity Toolbox [https://pmc.ncbi.nlm.nih.gov/articles/PMC10258674/]
• 31. Building an adverse outcome pathway network for ... [https://www.frontiersin.org/journals/toxicology/articles/10.3389/ftox.2024.1357717/full]
• 32. International Consortium to Advance Cross‐Species ... [https://pmc.ncbi.nlm.nih.gov/articles/PMC9161440/]
• 33. Strategy to replace, reduce or refine vertebrate animal ... [https://www.canada.ca/en/environment-climate-change/services/canadian-environmental-protection-act-registry/implementing-modernized-cepa/strategy-replace-reduce-refine-vertebrate-animal-testing.html]
• 34. Adverse Outcome Pathways (AOPs) focusing on endocrine ... [https://www.ecetoc.org/task-force/adverse-outcome-pathways-aops-focusing-on-endocrine-active-chemicals/]
• 35. An in silico protocol for endocrine activity assessment [https://www.sciencedirect.com/science/article/abs/pii/S2468111325000246]
• 36. Integrated Approaches to Testing and Assessment for the ... [https://www.sciencedirect.com/science/article/abs/pii/S0278691525006350]
• 37. Using transcriptomics data and Adverse Outcome Pathway ... [https://www.sciencedirect.com/science/article/pii/S0160412025001035]
• 38. Advancing assessments of endocrine disruptors using ... [https://openarchive.ki.se/articles/thesis/Advancing_assessments_of_endocrine_disruptors_using_adverse_outcome_pathways_and_novel_methodologies/29197022]
• 39. Adverse Outcome Pathway Networks I: Development and ... [https://pmc.ncbi.nlm.nih.gov/articles/PMC6004608/]
• 40. Adverse Outcome Pathway Networks II: Network Analytics - PMC [https://pmc.ncbi.nlm.nih.gov/articles/PMC6010347/]
• 41. Advancing Genotoxicity Assessment by Building a Global ... [https://pubmed.ncbi.nlm.nih.gov/40785417/]
• 42. Optimization of an adverse outcome pathway network on ... [https://www.sciencedirect.com/science/article/pii/S1532046423001867]
• 43. Combination of Computational New Approach Methodologies ... [https://pmc.ncbi.nlm.nih.gov/articles/PMC10926110/]
• 44. Multi-Omics Integration for the Design of Novel Therapies and ... [https://pmc.ncbi.nlm.nih.gov/articles/PMC10594525/]
• 45. Integration of “Omics”-Based Approaches in Environmental ... [https://www.mdpi.com/2305-6304/13/9/714]
• 46. Advancing human health risk assessment: the role of new ... [https://www.frontiersin.org/journals/toxicology/articles/10.3389/ftox.2025.1632941/full]
• 47. Future of Drug Development: How Omics Data Platforms ... [https://www.kandasoft.com/blog/future-of-drug-development]
• 48. The Cost Outcome Pathway framework: Integrating socio ... [https://www.sciencedirect.com/science/article/pii/S0300483X25001842]
• 49. Data and Tools [https://tox21.gov/data-and-tools/]
• 50. Tox21 Operational Model [http://ncats.nih.gov/research/research-activities/Tox21/operational-model]
• 51. Quantitative systems toxicology - PMC - PubMed Central - NIH [https://pmc.ncbi.nlm.nih.gov/articles/PMC5754001/]
• 52. Development of quantitative systems pharmacology and ... [https://www.sciencedirect.com/science/article/abs/pii/S1740675717300166]
• 53. Adverse Outcome Pathway (AOP) Development I [https://pmc.ncbi.nlm.nih.gov/articles/PMC4318923/]
• 54. Adverse Outcome Pathway (AOP) Research Brief [https://www.epa.gov/chemical-research/adverse-outcome-pathway-aop-research-brief]
• 55. Semantic Characterization of Adverse Outcome Pathways [https://pmc.ncbi.nlm.nih.gov/articles/PMC7393770/]
• 56. Advancing the quantitative understanding of adverse outcome ... [https://academic.oup.com/etc/article-abstract/44/3/614/7942974]
• 57. current status, methodologies, and future directions [https://pubmed.ncbi.nlm.nih.gov/39864436/]
• 58. Adverse Outcome Pathway Development II: Best Practices [https://pmc.ncbi.nlm.nih.gov/articles/PMC4318924/]
• 59. The FAIR AOP Roadmap for 2025: Advancing Findability ... [https://assessments.epa.gov/risk/document/&deid%3D366972]
• 60. AOP 363, Thyroperoxidase inhibition leading to altered ... [https://aopwiki.org/aops/363]
• 61. Reconciling FDA 'Radical Transparency' and SEC ... [https://businesslawtoday.org/2025/11/reconciling-fda-radical-transparency-sec-disclosure-requirements/]
• 62. A Pragmatic Approach to Adverse Outcome Pathway ... [https://pmc.ncbi.nlm.nih.gov/articles/PMC8633887/]
• 63. Adverse outcome pathways and ecological risk assessment [https://academic.oup.com/etc/article/30/1/64/7762515]
• 64. Adverse Outcome Pathways [https://www.setac.org/explore-science/methods-and-approaches.html]
• 65. Aop:500 - Activation of MEK-ERK1/2 leads to deficits in ... [https://aopwiki.org/aops/500]
• 66. about the AOP-Wiki [https://aopwiki.org/info_pages/3]
• 67. Computational approaches to predict the toxicity of ... [https://pmc.ncbi.nlm.nih.gov/articles/PMC11811359/]
• 68. PMC [https://pmc.ncbi.nlm.nih.gov/articles/PMC12029019/]
• 69. Increase, Regenerative cell proliferation (forestomach ... [https://aopwiki.org/relationships/782]
• 70. Análise Preditiva: Introdução e Casos de Uso - data avenue [https://dataavenue.com.br/en/blog/analise-preditiva-introducao-e-casos-de-uso/]
• 71. Weight of Evidence: Criteria and Applications [https://pubmed.ncbi.nlm.nih.gov/39312160/]
• 72. Adverse Outcome Pathways [https://www.oecd.org/en/topics/sub-issues/testing-of-chemicals/adverse-outcome-pathways.html]
• 73. Adverse Outcome Pathways - National Toxicology Program [https://ntp.niehs.nih.gov/whatwestudy/niceatm/comptox/ct-aop/aop]
• 74. Taring the scales: Weight-of-evidence framework for ... [https://www.sciencedirect.com/science/article/pii/S027323002400031X]
• 75. A framework tool that applies weight-of-evidence ... [https://www.frontiersin.org/journals/freshwater-science/articles/10.3389/ffwsc.2025.1520312/full]
• 76. The 2021 update of the EPA's adverse outcome pathway ... [https://www.nature.com/articles/s41597-021-00962-3]
• 77. Active Projects - Tox21 [https://tox21.gov/projects/]
• 78. Tox21 and adverse outcome pathways [https://www.sciencedirect.com/science/article/abs/pii/B9780128136027000405]
• 79. Environmental Toxicology and Chemistry | Oxford Academic [https://academic.oup.com/etc/article/44/9/2511/7943011]
• 80. Biovista Vizit January 2025 Release - AOP Features [https://www.biovista.com/biovista-vizit-january-2025-release-aop-features/]
• 81. AOP report: Adverse Outcome Pathway Network for ... [https://pubmed.ncbi.nlm.nih.gov/40888748/]
• 82. Reactive Oxygen Species (ROS) formation leads to cancer ... [https://aopwiki.org/aops/505]
• 83. Comprehensive mapping of the AOP-Wiki database [https://pmc.ncbi.nlm.nih.gov/articles/PMC10958381/]
• 84. A call to action: Advancing new approach methodologies ... [https://www.sciencedirect.com/science/article/pii/S0273230025001345]
• 85. Comparative review of human health reference values ... [https://pubmed.ncbi.nlm.nih.gov/40714797/]
• 86. Ecological Risk Assessment | US EPA [https://www.epa.gov/risk/ecological-risk-assessment]
• 87. The FAIR AOP roadmap for 2025: Advancing findability, ... [https://pubmed.ncbi.nlm.nih.gov/40950714/]
• 88. Building and Applying Quantitative Adverse Outcome ... [https://pmc.ncbi.nlm.nih.gov/articles/PMC6771761/]
• 89. Internal exposure dynamics drive the Adverse Outcome ... [https://www.nature.com/articles/srep21978]
• 90. Towards a qAOP framework for predictive toxicology [https://www.sciencedirect.com/science/article/pii/S2468111321000414]
• 91. Scientists discuss how Quantitative Adverse Outcome ... [https://www.ecetoc.org/news/scientists-discuss-how-quantitative-adverse-outcome-pathways-can-be-used-more-effectively-in-eu-chemical-safety-assessments/]
• 92. The Cost Outcome Pathway framework: Integrating socio ... [https://pubmed.ncbi.nlm.nih.gov/40571180/]
• 93. The Cost Outcome Pathway Framework: Integrating socio ... [https://www.medrxiv.org/content/10.1101/2024.02.20.24303098v1]
• 94. Adverse outcome pathways [https://foodpackagingforum.org/resources/background-articles/adverse-outcome-pathways]