Utility of the adverse outcome pathway concept in drug development

KEYWORDS:

• Toxicology for the 21st century
• mechanistic toxicology
• omics technologies
• hazard
• chemicals

1. Utility of the adverse outcome pathway concept in drug development

The safety evaluation of environmental and industrial chemicals is currently propagating the concept of adverse outcome pathways (AOPs). These developments have repercussions with drug development and safety assessments of drugs. Under the umbrella of Organisation for Economic Co-operation and Development (OECD) and closely linked to their chemical safety testing guideline program, AOPs are organizing in a crowdsourcing movement the mechanistic knowledge on how chemicals impact on human health and the environment. This happens, e.g. in the AOP-Wiki and for more quantitative AOPs in the Effectopedia platform. This is organizing existing knowledge on shared molecular initiating events and the subsequent key events along the chain from cellular to tissue, organ, organism, and population effects (Table 1).

Table 1. The adverse outcome pathway (AOP) concept and associated 21st century technologies.

Exposure	Kinetics (typically not included in AOP)	Molecular initiating event	Organelle effects	Cellular effects	Tissue effects	Organ response	Individual response	Population response
		Pathways of toxicity
		Mode of action
		AOP
Metabolomics of exogenous metabolites	PBPK	In silico Receptor HTS	Mechanistic Toxicology: Multi-omics HTS Microphysiological systems Imaging				Transcriptomics miRNomics Metabolomics of endogenous metabolites	Biomarker

The Organisation for Economic Co-operation and Development (OECD) scheme of AOP, to which kinetics were added, is depicted relative to the coverage of the AOP by pathways of toxicity or mode of action. The predominant Tox-21c technologies employed corresponding to the elements of the AOP are listed.

PBPK: physiology-based pharmacokinetic modeling; HTS: high-throughput screening.

This process is a key implementation approach for the toxicity testing in the 21st century paradigm suggested by the respective committee of the USA National Research Council/National Academy of Sciences in 2007 [1], next to high-throughput screening approaches piloted by the US Environmental Protection Agency’s ToxCast program and the Tox21 alliance of several US agencies as well as pathway of toxicity mapping [2,3]. Notably, the concept of pathways of toxicity (PoT) spearheaded by the Human Toxome project [4] differs somewhat from the AOP concept as compared in Table 2. These molecular and quantitative PoT are more closely what Effectopedia aims to curate than the AOP-Wiki platform. They are typically not a condensation of literature but the result of (multi-)omics measurements. It is important to understand that besides the narrative AOP quantitative PoT are needed, especially for modeling purposes (systems toxicology), i.e. to make predictions from virtual experiments to ‘validate’ our pathway understanding.

Table 2. Comparison of the adverse outcome pathway (AOP) and the pathways of toxicity (PoT) concepts.

AOPs	PoT
Narrative, low level of detail, existing information, largely linear series of events	Molecular, high level of detail, emerging information, network perturbation
Biased by existing knowledge	Untargeted identification typically by omics, experimental causality
Not quantitative, no flux, no dynamics	Aiming for quantitative relations and fluxes
No QA/validation yet	Causality (utility to be shown); suggested mechanistic validation
Repository: AOP-Wiki	Repository: Effectopedia (quantitative AOP)/Human Toxome Knowledgebase

In the end, AOP and PoT will serve slightly different purposes. AOPs represent an organized and (to an extent still to be developed) quality-controlled, organized repository of toxicological mechanistic knowledge. It is envisaged that the regulatory and the regulated community can use this as a point of reference when mechanistic knowledge is used as the basis for the development of testing strategies [5,6]. PoT, in contrast, promise to be more detailed (molecular), quantitative, and possibly even representative of the dynamic network perturbations. Thus, they qualify much for modeling and systems toxicology approaches [7]. Both the AOPs and the PoT concept have in common that a point of reference for regulatory toxicology and others shall be created. In simple terms, there should be no need for discussion when a registrant or a regulator refers in their argument to a mechanism represented by a quality-assured AOP.

Both approaches, AOPs and PoT have their challenges. Recent discussions around the lack of reproducibility of many scientific articles illustrate the problem of creating an AOP on this ground by cherry-picking the literature. The untargeted deduction of PoT is limited by the shortcomings of the model systems in vivo [8] and in vitro [9]. The omics technologies typically applied for PoT deduction do not make the situation easier as only transcriptomics is at this stage reasonably standardized for performance and reporting. Metabolomics – most promising as any effects at this level are close to phenotypic changes [10,11] – for example, lacks similar quality assurance [12]. All omics do have a clear signal-to-noise problem with (tens of) thousands parameters being measured with a handful of measurements and high noise. There are first attempts out of the Human Toxome project to reduce data dimensionality [13,14]. However, the most promising approach of combining orthogonal omics technologies is still in its infancy, rarely going beyond visualization and clustering analysis on the premature pathway maps available yet.

Bringing pathway concepts to drug development and safety on the first hand sounds like ‘bringing coal to Newcastle.’ Indeed, pathway-based drug development and excess pharmacology as a key concern in drug safety means that pathway thinking is actually further advanced in the drug field. Thus, actually, the AOP movement could first of all dramatically benefit of incorporating these pharmacological pathways of perturbing human biology. The pharmaceutical targets represent prototypic molecular initiating events and the resulting undesired effects key adverse outcomes. This could expand the AOP knowledge base quite quickly.

The fact that the ToxCast high-throughput testing project, which is closely linked to the AOP concept, includes already compounds (plus associated human testing data) donated by six pharmaceutical companies (GSK, Hoffmann La Roche, Sanofi-Aventis, Pfizer, Merck, and Astellas) creates an interesting test case for the cross-resonance of Tox-21c with the pharmaceutical universe.

But there is also something in the box for the drug field from the Tox-21c concept [15]. This is a joint effort of the regulated and the regulating community in stock-taking what is a mechanism of human toxicity and what is not. It will help to define testing strategies, investigative toxicology, and submissions. At some point, such agreed mechanistic knowledge might also be turned against the traditional methods when we ask to which extent they represent our mechanistic understanding? How many promising molecules might we have sorted out in the past? Agreed mechanism opens up also to a mechanistic validation of new tests [16], i.e. evaluation of methods not by comparing to the faulty methods of the past and thus not really improving human predictivity but only lessening ethical concerns. By annotation of pathways to species, not necessarily only of toxicity pathways but also drug-able pathways, testing efforts can be streamlined, promising to improve drug profiling and selection and thereby reducing attrition. The strengths of the concept come in where we look into ecotoxicological effects of contaminating drugs in the ecosystem or off-target effects of substances. It is reasonable to assume that there are only a limited number of weak spots of the organism, the Achilles heels, that trigger toxicity pathways. Here, the mapping exercise of toxicity pathways becomes a common interest of the pharmaceutical and the (agro-)chemical/consumer product industry. Overlaps exist also, for example, with nutraceuticals and the food industry. The toolbox of AOP and the derived tools such as in silico tool for molecular initiating and key events can prove helpful for the evaluation of contaminants and contact materials, where actual assessments are difficult.

Ultimately, we need to move from individual pathways to networks they form or which underlie them. This would be a systems toxicology or pharmacology, possibly leading to a virtual patient, as attempted a few years ago in the respective European flagship project application finalist, which came close to being awarded a billion euro, which also illustrates the dimensions of the challenge.

While on a mid-term pathway-based approaches mainly promise to support drug development and safety, on a longer term there might also be clinical applications. A virtual patient is not far from the creation of a personal avatar for each patient, where the standard model is adapted to the genetic and pharmacokinetic parameters of the patients and where interventions can be modeled and optimized in virtual treatments. Certainly still largely science fiction, but these were any of the technologies of our current toolbox some decades ago too.

2. Expert opinion

The Tox-21c movement to pathway identification has clear benefits for the pharmaceutical industry: Such information supports species extrapolation (e.g. considering the fit of a given animal model), deviation from guideline studies, and weighing evidence. Pathways represent a key tool for the anticipation of off-target and ecotoxicological effects as well as investigative toxicology. Noteworthy, pathways can also be pathways for drug efficacy, i.e. any toxicity pathway can also be a pharmacological target.

A hallmark of many of the Tox-21c technologies is their holistic approach to information generation: ‘omics technologies are the prime examples in which the totality of genes, mRNA, microRNA, proteins, and metabolites is assessed. Similarly, high-content imaging (HCI) determines a multitude of structural features and functions in an automated fashion. Last, but not least, even though HTS approaches, for example with an in vitro battery of tests, provide minor mechanistic understanding, the outcome is still very useful to immediately exclude substances with obvious toxicological liabilities. HCI already plays a role in investigative toxicology in pharmaceutical industry, i.e. to clarify toxic effects in guideline-driven studies or clinical trials and their relevance to the patient population. It is clearly an information-rich technique, but not really employed yet for PoT identification.

The strong reliance of Tox-21c on human cell models is a strength and weakness from the perspective of pharmaceutical industry: The possible human relevance is evident, but there are limits to the predictive value of any cell model to predict human responses, especially if the effects concerned are not cellular/organ responses but systemic, metabolic, or even behavioral ones.

A lot of the toxicological approaches in pharmaceutical industry are cemented in International Conference on Harmonization-harmonized guidance. Any paradigm shift must be introduced gradually, i.e. accepting new methods not necessarily replacing a traditional one directly; most of the time, a new method will refine the strategy, which includes the abandoning of earlier components, and in the future, this may lead to a reduced need for animal data. The new technologies, including omics, computer modeling, and mechanistic approaches, are definitely underexploited for regulatory purposes, even though the majority of the novel methodologies are already applied in screening during drug discovery and as such avoid bad candidates making it to regulatory testing.

The data-poor chemical risk assessment process in chemical and consumer product industry often looks like the ugly little brother copying bits and pieces of the bigger brother of full assessment in drug toxicity testing. The additional problems of unclear exposure and enormous numbers of substances and mixtures to be considered, currently force the little brother to apply its own way of doing things. The crisis of attrition rates in pharmaceutical industry shows that the current development process makes us put our money too often on the wrong horse. Toxicology has its share in this. Perhaps it is worthwhile to explore some ideas of the little brother and in doing so help him to develop his ideas…

Declaration of interest

The work on the Human Toxome referred to was supported by an NIH Transformative Research Grant, ‘Mapping the Human Toxome by Systems Toxicology’ (RO1 ES 020750). The author has no other relevant affiliations or financial involvement with any organization or entity with a financial interest in or financial conflict with the subject matter or materials discussed in the manuscript apart from those disclosed.

Additional information

Funding

This paper was not funded.