Nov 30, 2014 - Maurice Whelan. Received: 14 November 2014 ...... 11, Geneva, Switzerland. Wu S, Blackburn K, Amburgey J, Jaworska J, Federle T (2010) A.
Arch Toxicol (2015) 89:15–23 DOI 10.1007/s00204-014-1421-5
RESEARCH ARTICLE
SEURAT: Safety Evaluation Ultimately Replacing Animal Testing—Recommendations for future research in the field of predictive toxicology George Daston · Derek J. Knight · Michael Schwarz · Tilman Gocht · Russell S. Thomas · Catherine Mahony · Maurice Whelan
Received: 14 November 2014 / Accepted: 20 November 2014 / Published online: 30 November 2014 © Springer-Verlag Berlin Heidelberg 2014
Abstract The development of non-animal methodology to evaluate the potential for a chemical to cause systemic toxicity is one of the grand challenges of modern science. The European research programme SEURAT is active in this field and will conclude its first phase, SEURAT-1, in December 2015. Drawing on the experience gained in SEURAT-1 and appreciating international advancement in both basic and regulatory science, we reflect here on how SEURAT should evolve and propose that further research and development should be directed along two complementary and interconnecting work streams. The first work stream would focus on developing new ‘paradigm’ approaches for regulatory science. The goal here is the identification of ‘critical biological targets’ relevant for toxicity and to test their suitability to be used as anchors for predicting toxicity. The second work stream would focus
The views expressed in this article are those of the authors and do not necessarily reflect the views or policies of the US Environmental Protection Agency, the European Commission and the European Chemicals Agency. Mention of trade names or commercial products does not constitute endorsement or recommendation for use. G. Daston Central Product Safety Department, The Procter and Gamble Company, 8700 Mason Montgomery Road, Mason, OH 45040, USA D. J. Knight European Chemicals Agency, Annankatu 18, P.O. Box 400, 00121 Helsinki, Finland M. Schwarz (*) · T. Gocht Department of Toxicology, Institute of Experimental and Clinical Pharmacology and Toxicology, Eberhard Karls University of Tübingen, Wilhelmstr. 56, 72074 Tübingen, Germany e-mail: Michael.schwarz@uni‑tuebingen.de
on integration and application of new approach methods for hazard (and risk) assessment within the current regulatory ‘paradigm’, aiming for acceptance of animal-free testing strategies by regulatory authorities (i.e. translating scientific achievements into regulation). Components for both work streams are discussed and may provide a structure for a future research programme in the field of predictive toxicology. Keywords Critical biological targets · Adverse outcome pathways · Safety assessment · Read-across · Ab initio prediction
Introduction The development of non-animal methodology to evaluate the potential for a chemical to cause systemic toxicity is one of the grand challenges of modern science. Heretofore, systemic (organ) toxicity has been evaluated using animal models in which the outcome of the experiment is the production of disease states that are extrapolated to predict R. S. Thomas National Center for Computational Toxicology, U.S. Environmental Protection Agency, 109 T.W. Alexander Drive, Room B210I, Mail Code: B205‑01, Research Triangle Park, NC 27711, USA C. Mahony Procter and Gamble Technical Centres Limited, Whitehall Lane, Egham, Surrey TW20 9NW, UK M. Whelan Institute for Health and Consumer Protection (IHCP), European Commission Joint Research Center (JRC), Via E. Fermi 2749, TP202, 21027 Ispra, VA, Italy
13
16
adverse effects in humans from exposure to the chemical. This approach has formed the basis for hazard and risk assessments, and public health decisions based on these assessments, so much so that animal test data are central to virtually all chemical, pesticide and pharmaceutical regulation worldwide. Until recently, this has also been the case for cosmetics, but in March 2013, the EU imposed a marketing ban for cosmetic products that contain ingredients tested on animals after that date (European Commission 2013). The heavy reliance of regulatory agencies on information derived from animal testing presents a significant challenge to the development of non-animal replacements, in that the regulatory schemes are unlikely to change until it can be shown that alternative data streams can fulfil the same purpose with comparable confidence. Although tough challenges still need to be addressed, it is important to acknowledge how far the field has come in transitioning to a new way of identifying and characterising toxicological hazard and predicting safety. Within the EU, the European Commission’s Framework Programmes (FP) for research have been the principle driver, contributing more than 200 million Euros over roughly the last 20 years in the development of nonanimal approaches to safety assessment in a variety of sectors. The output of many projects funded under FP6 and FP7 has been reviewed and described in a series of reports by the FP7 coordination action AXLR8 (www.axlr8.eu). In the USA, major programmes such as ToxCast (Dix et al. 2007; Kavlock et al. 2012) and Tox21 (Collins et al. 2008; Tice et al. 2013) being undertaken by government agencies have embraced the vision and strategy described by the seminal NRC (2007) report and have come a long way in making it a reality. With an aim to combine expert thinking from both sides of the Atlantic, AXLR8 organised a series of international workshops to understand how to accelerate ‘the transition to a toxicity pathway-based paradigm for chemical safety assessment through internationally co-ordinated research and technology development’. The output of these important discussions culminated in the formulation of a comprehensive set of recommendations to guide future research programmes (AXLR8 2012). The first phase of the SEURAT (‘Safety Evaluation Ultimately Replacing Animal Testing’) initiative, SEURAT-1 (www.seurat-1.eu), will conclude its 5-year research programme in December 2015. With an overall budget of 50 million Euros, equally financed by the European Commission and Cosmetics Europe, SEURAT-1 has paved the way for the application of modern toxicological science in safety assessment and has addressed a number of AXLR8 priorities. An important commitment of the SEURAT-1 cluster has been to report on the effectiveness and evolution of the SEURAT research strategy,
13
Arch Toxicol (2015) 89:15–23
primarily through publication of an annual book (Gocht and Schwarz 2014). The guiding principle of the SEURAT strategy is to adopt a toxicological mode-of-action framework to describe how any substance may adversely affect human health. Through the formulation of case studies, this is being developed to the proof-of-concept level within SEURAT-1 (Gocht et al. 2014). In essence, the adverse outcome pathway (AOP) framework (OECD 2013) is being employed to integrate mechanistic data for the purposes of grouping/read-across and the prediction of quantitative points of departure needed for safety assessment.
Recommendations for next steps Drawing on the experience gained in SEURAT-1 and appreciating international advancement in both basic and regulatory science, we reflect here on how SEURAT should evolve scientifically to ultimately realise its vision. Essentially, we propose that further research and development should be directed along two complementary and interconnecting work streams (Fig. 1) that are elaborated below. The first work stream would focus on developing new ‘paradigm’ approaches for regulatory science. The goal here is the identification of ‘critical biological targets’ relevant for toxicity and to test their suitability to be used as anchors for predicting toxicity. These critical targets will not be restricted to effects on distinct organs or particular types of toxic effects. The idea is that independent of the nature of the adverse outcome of interest, be it cancer or developmental toxicity or acute or repeated dose toxicity, the disturbance of pathways related with these critical biological targets would indicate a likelihood for adversity and the dose at which this would occur. The aim would be to establish, at a proof-of-concept level, the suitability of this ‘critical target concept’ to improve the predictive power of mechanism-based toxicity testing methods, which might be applied in a future regulatory safety assessment. The central aspect of this work stream is to give up the definition of adversity at the organ level and to identify new pointsof-departure for a future safety assessment paradigm at the molecular scale. The second work stream would focus on integration and application of new approach methods for hazard (and risk) assessment within the current regulatory ‘paradigm’, aiming for acceptance of animal-free testing strategies by regulatory authorities (i.e. translating scientific achievements into regulation). In essence, this would be to complete the SEURAT-1 ‘conceptual framework’ (see below) for combining evidence, in particular from new approach methods, in a biologically rational manner to qualitatively and quantitatively predict traditional organ-based toxicity.
Arch Toxicol (2015) 89:15–23
17
Characterising toxicity
SEURAT Safety Evaluation Ultimately Replacing Animal Testing
SEURAT-1 Towards the replacement of repeated dose systemic toxicity testing Specific Aim: Proof-of-concept demonstrating mode-of-action informed generation and integration of mechanistic data for safety assessment
SEURAT-1 follow-up Two interconnecting work streams to develop a new regulatory paradigm for safety assessment while applying animal-free approaches to satisfy information requirements of current regulatory frameworks
Work Stream I:
Work Stream II:
Characterising Toxicity
Translation into Regulation
Specific aims:
Specific aims:
Development of a new toxicological hazard assessment framework based on critical biological targets rather than apical health-effect endpoints.
Development of guidance and case studies for implementation of in silico and in vitro methods for grouping/read-across and deriving no-effect levels.
Prove concepts, explore uncertainties and understand implications for regulation.
Application and acceptance in ‘real world’ regulatory decision making processes.
Fig. 1 Realising the SEURAT vision: a two-pronged approach to advance predictive toxicology and safety assessment to assure the highest levels of protection for human health while avoiding animal testing
Specific implementation of the framework includes contributing additional evidence of the biological basis for ‘read-across’ and ab initio development of a safety assessment that relies only on the new in silico and in vitro methods. The aim is to develop guidance (e.g. at OECD level) for the implementation of non-standard in silico and in vitro methods into regulatory risk assessment. The application in ‘real-world’ decision-making processes is an essential deliverable. In this way, this work stream would be a continuation of the SEURAT-1 programme, which would evolve beyond the proof-of-concept stage and finalise the work started in SEURAT-1, demonstrating the widespread implementation of mode-of-action-based reasoning in chemical safety assessment and regulatory decision-making. The conceptual underpinning of these two work streams builds on the current state-of-the-art in different fields, as discussed in the following sections, as both work streams encompass characterising toxicity and translating new tools and thinking into regulatory decision-making.
From animal testing to adverse outcome pathways to ‘critical biological targets’ In future, the potential toxicological hazard of a chemical will be determined through the rational integration of complementary information on its chemical structure, biological activity and pharmacokinetic behaviour (toxicokinetics) derived primarily from in vitro high-throughput/ content screening and computational modelling. Practice is already shifting away from conventional animal testing based on pathological observation after high-dose exposure to a knowledge-informed paradigm where a chemical’s toxicity is profiled in terms of its potential to impair basic biological function at relevant exposure levels. Chemicals can interfere with normal biological processes or pathways at the molecular level through a multitude of different mechanisms. Selectivity of action can vary, for example, from non-selective binding to intracellular proteins to selective agonism/antagonism of a particular nuclear or other receptor. Of course, selectivity can also be described at higher levels of biological organisation such as target tissue, organism life-stage or sub-population, with these aspects depending not only on mechanistic selectivity at the molecular level but also on exposure conditions and a chemical’s pharmacokinetics. When profiling chemical toxicity at the molecular or pathway level, one can envision three possible cases. The first would be the instance where biological activity data indicate specific activity against a single biological pathway whose significance for health and adverse outcome is well characterised. For example, a compound shown to be a specific anti-androgen would be expected to adversely affect male reproductive development and to have effects on fertility. Computational models of pharmacokinetics can predict bounding estimates of maximum blood concentrations and elimination times, and comparison of relative potency versus other anti-androgens would permit an estimate of the concentration (or dose) that would be expected to have an effect in vivo. The second case is the instance where the chemical is entirely non-selective (i.e. affects a large number of biological pathways), with the safety assessment then being driven by determination of the local concentration at which these effects take place, adjusted to in vivo dose using pharmacokinetic models. The third case will be the most challenging, which involves chemicals that affect more than one biological pathway, but are not totally non-selective. Such chemicals with pleiotropic behaviour will probably require more sophisticated systems biology approaches to predict which types of toxicity might be anticipated, and at what concentrations.
13
18
Mining and application of mechanistic knowledge for the purposes of chemical testing and assessment is far from routine or straightforward. The adverse outcome pathway (AOP) framework (Ankley et al. 2010) has been developed specifically to tackle this problem. The intent of an AOP is to identify the key steps, from molecular initiating event to the adverse outcome, usually at the organ level. Extensive guidance on how to develop and evaluate AOPs (OECD 2013) and a dedicated public knowledge base (www.aopkb.org) to facilitate scientific crowd-sourcing and peer review are already available. The AOP approach will likely be the state of the art for some time to come. However, at some point, we will have enough data through systematic mapping of toxicological pathways and processes across different levels of biological organisation that it will be possible to use those data to support predictions about a variety of adverse outcomes that have a common molecular initiating event. For example, an inhibitor of the signalling protein sonic hedgehog can be expected to have adverse effects on nervous system development in the embryo and effects on skin or hair growth in the adult. Consequently, significant inhibition of sonic hedgehog activity can be taken as an indication of adversity per se, as adverse effects on development of the nervous system can be assumed based on knowledge of the central role of this gene in development. A large number of such critical targets and their components are already known from basic biology studies. An additional rich source of information may come from the analysis of human disease states, where the molecular defects causing the disease are identified. In this case, a chemical producing the same molecular defect by interference with the critical target can be predicted to cause the disease when present at sufficiently high concentration. Altogether, deeper understanding of these critical targets may open the door for shifting the point of departure for safety assessment away from the organ level and towards the molecular initiating events. Note that the approach of ‘critical biological targets’ may not be required for chemicals interacting non-selectively, as their effects can be predicted without detailed knowledge on all their cellular target interactions. For pharmaceuticals, however, toxicity needs still to be defined at the organ level as the physician needs to know where toxicity might be expected in the patient, but for industrial and environmental chemicals, future safety assessment approaches may focus more on interferences of chemicals with ‘critical biological targets’ (molecules or pathways). Methods Further characterising the landscape of toxicological pathway responses of biological systems needs a suite of advanced research methods and tools. Apart from understanding and describing toxicodynamic processes in both
13
Arch Toxicol (2015) 89:15–23
a qualitative and a quantitative manner, the intention is to identify new biomarkers that are sensitive to interactions between chemicals and ‘critical cellular targets’ that can be used as a basis for toxicity prediction and setting safe exposure levels. With this goal in mind, there are a number of scientific and technological research areas that will need to be developed and/or perfected, such as those described below. Chemoinformatics Chemoinformatics is the use of informatics to identify similarities between chemicals that can be exploited for read-across (translational part). Fertile areas for chemoinformatic research will include further development of expert rules to identify suitable analogues for read-across, prediction of metabolism based on chemical structure, use of other chemical features besides two-dimensional structure to identify similarities and organisation of chemicals into a mode of action ontology. This last need will require X-ray crystallography data on a number of toxicologically relevant proteins (small molecule receptors, enzymes, ion channels, etc.) with a variety of ligands for these proteins, preferably with low to medium potency to approximate what might be expected from chemicals used in cosmetics, where high-potency molecules are avoided. Chemoinformatics can also be used to elucidate chemical structural features that will lead to prediction of common function/ common targets. High‑throughput screening High-throughput screening (HTS) methods are already well developed. The three highest priorities for HTS batteries are to determine the number of assays needed to cover the universe of possible toxicological mechanisms, to continue to evaluate a large number of chemicals that cover a wide range of chemical ‘space’ and to develop statistical approaches to integrate the large amount of data into interpretable conclusions. This latter goal is likely to involve approaches such as Bayesian statistics that support the interplay between assay results and expert judgment. Various international research programmes are using these methods (ToxCast, Dix et al. 2007; Kavlock et al. 2012), the Human Toxome project (Bouhifd et al. 2014), and collaborations with these initiatives should be developed, in particular in the context of the part focusing on the identification of ‘critical cellular targets’. High‑content screening Methods such as toxicogenomics are attractive because virtually all toxic responses are preceded by changes in gene
Arch Toxicol (2015) 89:15–23
expression, and the pattern of gene expression can be diagnostic of mode of action and ‘critical biological targets’. Furthermore, because gene expression analysis usually covers the entire genome, the methodology is able to detect off-target effects provided that the offending target or pathway is expressed in the model system(s) being tested. There is already a reasonable amount of research to show that toxicogenomic responses from in vitro systems are rich enough to recapitulate in vivo responses, but more needs to be done to fully define the potential and limitations of in vitro systems as the platforms for toxicogenomics. Another important area for research is to develop higher-throughput methods, such as the connectivity map approach now used for therapeutic agents (Lamb et al. 2006). In addition, the development or adaptation of biochemical high-content/ high-throughput methods can be envisaged: methods that are able to detect and quantitatively describe chemical–target interactions can be further developed. Systems biology approaches One of the most significant challenges in extrapolating from in vitro systems to organismal responses is bridging the gap between responses at the molecular and cellular level and responses at levels of organisation at which diseases are manifested. The relationship between events at the molecular level and those at the organ level is decidedly nonlinear; e.g. interaction of a chemical with its endogenous molecular target is almost certainly non-threshold, but responses at higher levels require millions of chemical–target interactions. If this threshold is not reached, then an adverse response is not produced. The difficulty is in identifying at what point, and how, these nonlinearities occur. Higher-order tissue models, such as three-dimensional tissue culture, tissue chips and bioengineered microfluidic systems that connect different tissue types, are all in various stages of development and probably hold the key to bridging the distance between molecular effects and adverse responses. These methods will need considerably more development in order to be useful with any sort of throughput. Regardless of how well these methods are developed, throughput is likely to be low. Therefore, it will be important to couple development of these methods with computational systems biology models that will support simulated experiments in order to identify the key events at the molecular level that can then be modelled in vitro. Cellagent-based models, ordered differential equation and other models exist now but must be further developed in order to be fully applicable to toxicology. Pharmacokinetics There are already several excellent in silico pharmacokinetic models available for QVIVE and reverse dosimetry
19
(Clewell 1995; Rotroff et al. 2010). Additional research challenges in this area are the development of in silico methods for predicting dermal absorption from finite doses of chemical, in silico approaches for estimating variables like protein binding, hepatic clearance and renal clearance that now must be determined empirically, and molecular dynamics approaches for better determining the actual chemistry of dermal absorption.
Translating new tools and thinking into regulatory decision‑making There are two primary needs for characterising the hazardous properties of chemical substances, namely to screen a large numbers of substances to identify groups with particular characteristics for further action and to assess a specific substance for a defined purpose, e.g. to fill a ‘data gap’ and establish safe use from a risk assessment. The degree of uncertainty tolerated in the prediction depends on the regulatory purpose; therefore, assessing and communicating uncertainty is a key element (WHO 2014). In general, higher certainty is needed to assess specific individual substances than for screening sets of chemicals for priority setting. Regulators generally set standards for the information on the properties of chemicals, whether from standard tests or non-standard approaches; hence, if the prediction does not meet the standard, it is not fit for purpose. Chemical categories and read‑across Initial application of alternative approaches for toxicity testing has been focusing more towards using the data to inform the biological similarity of chemicals in a ‘readacross’, i.e. when the known toxicological properties of an already-tested ‘source’ substance are ‘read-across’ to an untested ‘target’ chemical and ‘chemical categories’ of targets and source(s). Read-across is primarily based on considerations of similarity of structural and physical chemistry between molecules that indicate that they are likely to be handled by the body in a similar way (similar pharmacokinetics) and to have similar biological activity (Enoch et al. 2008; Wu et al. 2010). Sometimes the support for the read-across is sufficiently strong that no additional data need be generated, such as when there are good data sets on several closely related analogues, all of which indicate similar toxicity and potency, or when the differences in chemical structure between the target chemical and its analogue(s) are trivial. However, when data are more limited or structural differences are greater, additional evidence may be needed to support a definitive conclusion of the target chemical’s toxicity that is adequately robust for the particular purpose of the prediction (Patlewicz et al.
13
20
2014). In these instances, the read-across process can be considered to be a hypothesis-generating exercise, in which the hypothesis falls into one of two broad categories: (a) either the target compound is metabolised to an entity that has already been tested for toxicity or (b) the target compound and/or its metabolites (if applicable) are of comparable biological activity to the source compound and/ or metabolites, leading to a comparable toxicological profile. Methods already exist for the first category (Wu et al. 2010). It is the second, much larger category that requires the new tools that are being developed to understand biological action at the molecular and cellular level. The hypothesis that the source and target substance(s) (of the parent substance and/or the parent substance’s metabolites) all act via a common mode of action (i.e. have comparable toxicodynamics) can be tested using methods that evaluate biological responses at a molecular or cellular level, such as toxicogenomics or high-throughput batteries such as those employed by ToxCast. The results of these assays, with pharmacokinetic adjustments, may be sufficient to support the read-across in many instances. There is a complication, however, in that some aspects of toxicity are likely to require higher-order model systems that can predict nonlinear responses that may occur at different levels of biological organisation (e.g. one may need a sufficient number of cells to be damaged before an effect at the tissue level is manifested) or that require complex interactions between different cell types. This issue is pertinent both to the read-across justifications and also to the ab initio predictions discussed in the next section. There are a number of potential approaches to address these questions, but all are likely to require higher-order in silico systems biology models and sophisticated cell culture systems. Non-selective chemicals, however, tend to affect a large number of biological processes in HTS systems such as ToxCast, at or just below a level that is frankly cytotoxic (Thomas et al. 2013). Additional consideration will therefore need to be given as to how to characterise the hazard for such chemicals from alternative data streams, although it may be sufficient to simply indicate that there is an apparent lack of specific adverse effect signals, and regulation will be based on potency. Ab initio prediction of toxicity A more challenging application of alternative approaches for toxicity testing is their use in developing a full quantitative chemical risk assessment. A full chemical risk assessment requires a comprehensive evaluation of the potential for a chemical to affect physiological function in virtually every organ/tissue in the body. Ultimately, of course, the goal will be to have an integrated set of models that are sophisticated enough to conduct a full chemical
13
Arch Toxicol (2015) 89:15–23
risk assessment, but these models are not yet available. In the near term, a more pragmatic application of alternative approaches may be required. For non-pharmaceutical chemicals, the majority appear to act via non-selective interactions with biological targets and cellular pathways (Thomas et al. 2013). In other words, most chemicals interact with a relatively large number of targets and pathways within a narrow, high-concentration range. For these chemicals, fully characterising the multiple mechanisms that are likely contributing to the adverse response may not be efficient. Application of alternative approaches could be used to predict which chemicals are likely non-selective in their interactions with biological systems and the concentration at which they begin to perturb biological activity. The link to tissue- and organ-level adverse effects would then be inferred. Pharmacokinetic models could then be used to convert this concentration into an administered dose to derive a point-of-departure for a risk assessment (Judson et al. 2011). For chemicals showing selectivity for a specific biological target or pathway, a hypothesised modeof-action would be constructed. Alternative approaches such as specific in vitro assays or structure–activity models could be used to examine key events in the mode of action and identify qualitatively whether the key event is likely to occur and, if so, quantitatively at what concentration it is triggered. Computational modelling would allow integration of the key events and dynamic simulation of the potential health outcomes. Pharmacokinetic models could then be used to convert this concentration into an administered dose to derive a point-of-departure for a risk assessment. Note that both the assessment of biological activity and the pharmacokinetic estimates are aided significantly by the availability of chemical analogues that have already been tested for toxicity in traditional models. Assessment frameworks We are still not at the point where chemical regulation can be based on this approach; however, it is highly likely that some or all of the tools can be used to support chemical safety-related decisions as part of an integrated testing approach. In order to move forward from this point on, it is important not just to develop tools, but to develop and apply them in the context of a framework, whether it is read-across or de novo prediction, that allows the prediction of hazard and risk in humans. Future research will need to be coordinated such that individual model systems are integrated with others and serve the purpose of estimating not only the possible outcomes from an exposure, but also the probability of the outcome at a specific dose level, taking into account population differences. There are a variety of frameworks that have been proposed for the purpose of predicting human toxicity without
Arch Toxicol (2015) 89:15–23 Fig. 2 ‘Conceptual framework’ as a structure to guide assessors in devising a fit-for-purpose ‘bespoke’ Integrated Assessment Strategy that combines information from predictive tools with a stated protection goal (Gocht et al. 2014; with modifications)
21
Pieces of evidence and initial considerations • Purpose of the assessment • Exposure context • Expert knowledge and judgement based on existing evidence / data
Hypothesis generation regarding mode of action
General adversities
Type of adversity
Toxicokinetics
Definition of relevant dose range
Determination of point of departure
Organ specific adversities
Assessment of ADME Properties
Toxicodynamics
Toxicodynamics
• Many biological targets (based on chemical structure, e.g. alkylating agents) • Specific targets present in many cells / tissues / organs (e.g. AhR-pathway)
• Target organ: full assessment based on Adverse Outcome Pathway (AOP)1 • Non-target organ: limited assessment
Quantitative in vitro to in vivo extrapolation
Overall assessment
Evaluation
Including uncertainties and knowledge gaps
Improve assessment if necessary
Result
Use of prediction for pre-defined purpose With consideration of acceptable uncertainty
1) The
steps p in the AOP ((molecular initiating g event,, keyy events)) will be assessed using a selection of tools including in silico predictions and in vitro tests.
the use of animal data (Judson et al. 2011; Thomas et al. 2013; Pastoor et al. 2014). An example that emerged within SEURAT-1 is illustrated in Fig. 2. The aim was to develop a flexible approach to deliver fit-for-purpose assessment of the toxicological properties of a substance, taking into account its properties and the (regulatory) purpose for the prediction. This approach can be used as a basis for rational combination of information derived from predictive tools to support a safety assessment process or decision to achieve a stated protection goal. This framework is intended to set out a structure to guide assessors in devising a fit-forpurpose (or ‘bespoke’) Integrated Assessment and Testing Approach (IATA) for the particular circumstances and case. The overall outcome is anticipated to be robust as it is not based on single pieces of evidence, but rather a weight of evidence combined in a biologically rational manner.
In brief, the following steps should be considered when using this framework (Gocht et al. 2014): Decide the degree of confidence needed for prediction (low degree of confidence may be acceptable in case of well-controlled and low human exposure). Examine existing knowledge (toxicological studies, ‘read-across’ from chemical or biological analogues, QSARs and structural alerts, expert judgement). Distinguish between ‘general chemicals’ (expected to be non-selective in interacting with biological targets) and drugs/pesticides (designed to be selectively biologically active). Two parallel lines of consideration: (I) ‘general’ adverse effects not associated with a particular organ and (II) organ-based adverse effects.
13
22
Arch Toxicol (2015) 89:15–23
Consideration of toxicokinetics/toxicodynamics (for both lines of consideration). Extrapolation from in vitro to in vivo concentrations. Effects on organs can be assessed by (several) AOP’s incorporating existing knowledge and with new data as a combination of in vitro assays (‘-omics’ data, etc.) and in silico predictions in a battery of tools. The ‘conceptual framework’ can be used as a basis for designing the implementation of non-animal-based technology in read-across studies to be used in a regulatory context. If the read-across case is already robust, then it can be strengthened by examining the possibility of confounding effects due to pharmacokinetic differences that are attributable to small differences in physical chemistry. Pharmacokinetic in silico models exist that can serve this purpose, but there is a research and development need to improve these methodologies. The conceptual framework can also be used as a basis for combining evidence in a biologically rational manner for ab initio prediction of the toxicological properties of substances, taking into account the expected or experimentally determined selectivity of the chemical for its interactions with biological targets and processes. Depending on the intended purpose of the assessment, the process may be iterative to address specific uncertainties or fill known data gaps. Case studies Case studies are a powerful tool in heightening collaboration and harnessing knowledge and data from multiple sources. For maximum impact, they require industrial relevance to extract the required knowledge and show translation for decision-making. Implicit in their formulation, for either read-across or ab initio scenarios, is a common agreement of test compounds and where applicable dose ranges for testing thereby providing a common goal. The formulation of a regulatory framework that formalises read-across is envisaged in the next 5 years, which would require collaborative efforts of industrial toxicologists and regulators. It will be necessary to test against this framework with case studies to evaluate its applicability and demonstrate its utility. Further developments of AOP’s may be a component in this context and would justify the involvement of researchers from academia. However, the primary goal is not to develop new toxicity testing methods, but to formulate a guidance document about how to perform read-across assisted by already existing in vitro and in silico methods and publicly available databases. This building block is fully based on the concept developed within SEURAT-1 and would expand the applicability domain of tools and methods that were already developed within SEURAT-1 as well offer opportunity to harness methods developed elsewhere.
13
While the conceptual framework depicted in Fig. 2 was primarily developed for repeated dose toxicity, the work still to be done in this building block must not be restricted to this particular endpoint. Applying the conceptual framework in other areas (such as developmental and reproductive toxicity) would induce additional uncertainty that can be addressed by adapting the framework to the particular toxicological endpoint and use case of the chemical and extending the toolbox to encompass other tools and methods. Case studies can therefore offer opportunity to translate research output to solutions for decision-making but will also be an important tool in highlighting gaps in the research and areas of uncertainty. As such, they serve a purpose in both the shorter term to lend credence to a regulatory framework for read-across as well the longer term to bear out ab initio predictions. A number of case studies are being progressed in SEURAT-1 (Gocht et al. 2014) in part in collaboration with ToxCast but the grand challenge will be to harness more global cooperation still.
The road ahead The success of any research programme relies heavily on the effectiveness, coherence and clarity of its research strategy. Essentially, it is the glue that binds efforts together and directs the collective momentum of all actors towards the achievement of common goals. Here, we suggest a research programme that combines the translation of implementation of scientific achievements from the last 5 years into regulatory practice with the refinement and further development of the research strategy developed within SEURAT-1. Modern tools are already fit to support many regulatory decisions. The most obvious applications are in more robust read-across approaches and the prioritisation of chemicals for further testing. The development of a generally applicable framework for readacross is achievable within 5 years. In parallel, further development and integration of the tools may eventually support application to a full risk assessment, in particular to data-poor chemicals. Success in at least some of the research areas outlined above will go a long way towards replacement of animal testing for toxicity. While initial forays into prediction will rely to a large extent on readacross from 50 years of animal test data, at some point, particularly with the development of a large foundation of HTS and high-content data combined with computational systems biology and pharmacokinetic modelling, it will be possible to elucidate the toxicity of chemicals with no analogues in the animal testing database. The ultimate achievement will be to predict toxicity based on these initial molecular interactions and cellular responses in a way
Arch Toxicol (2015) 89:15–23
that incorporates the nonlinear, stochastic and emergent properties of biological systems. Acknowledgments This work has received funding from the European Community’s Seventh Framework Programme (FP7/2007–2013) under grant agreement no 267044 (project acronym ‘COACH’) and has received financing from Cosmetics Europe.
References Ankley GT, Bennett RS, Erickson RJ, Hoff DJ, Hornung MW, Johnson RD, Mount DR, Nichols JW, Russom CL, Schmieder PK, Serrano JA, Tietge JE, Villeneuve DL (2010) Adverse outcome pathways: a conceptual framework to support ecotoxicology research and risk assessment. Environ Toxicol Chem 29(3):730–741 AXLR8 (2012) Alternative testing strategies: progress report 2012 & AXLR8-3 workshop report on a ‘roadmap to next generation safety testing under horizon 2020’. Self-publishing, Berlin. http://axlr8.eu/assets/axlr8-progress-report-2012.pdf. Accessed 14 Nov 2014 Bouhifd M, Hogberg HT, Kleensang A, Maertens A, Zhao L, Hartung T (2014) Mapping the human toxome by systems toxicology. Basic Clin Pharmacol Toxicol 115(1):24–31 Clewell HJ 3rd (1995) The application of physiologically based pharmacokinetic modeling in human health risk assessment of hazardous substances. Toxicol Lett 79(1–3):207–217 Collins FS, Gray GM, Bucher JR (2008) Toxicology. Transforming environmental health protection. Science 319(5865):906–907 Dix DJ, Houck KA, Martin MT, Richard AM, Setzer RW, Kavlock RJ (2007) The ToxCast program for prioritizing toxicity testing of environmental chemicals. Toxicol Sci 95(1):5–12 Enoch SJ, Cronin MT, Schultz TW, Madden JC (2008) Quantitative and mechanistic read across for predicting the skin sensitization potential of alkenes acting via Michael addition. Chem Res Toxicol 21(2):513–520 European Commission (2013) Communication from the Commission to the European Parliament and the Council on the animal testing and marketing ban and on the state of play in relation to alternative methods in the field of cosmetics. Brussels. http://ec. europa.eu/consumers/archive/sectors/cosmetics/files/pdf/animal_ testing/com_at_2013_en.pdf. Accessed 14 Nov 2014 Gocht T, Schwarz M (2014) Towards the replacement of in vivo repeated dose systemic toxicity. Self-publishing, Paris. http:/www.seurat-1.eu/pages/library/seurat-1-annual-report.php. Accessed 14 Nov 2014 Gocht T, Berggren E, Ahr HJ, Cotgreave I, Cronin MT, Daston G, Hardy B, Heinzle E, Hescheler J, Knight DJ, Mahony C, Peschanski M, Schwarz M, Thomas RS, Verfaillie C, White A, Whelan M (2014) The SEURAT-1 approach towards animal free human safety assessment. ALTEX. doi:10.14573/altex.1408041 Judson RS, Kavlock RJ, Setzer RW, Hubal EA, Martin MT, Knudsen TB, Houck KA, Thomas RS, Wetmore BA, Dix DJ (2011)
23 Estimating toxicity-related biological pathway altering doses for high-throughput chemical risk assessment. Chem Res Toxicol 24(4):451–462 Kavlock R, Chandler K, Houck K, Hunter S, Judson R, Kleinstreuer N, Knudsen T, Martin M, Padilla S, Reif D, Richard A, Rotroff D, Sipes N, Dix D (2012) Update on EPA’s ToxCast program: providing high throughput decision support tools for chemical risk management. Chem Res Toxicol 25(7):1287–1302 Lamb J, Crawford ED, Peck D, Modell JW, Blat IC, Wrobel MJ, Lerner J, Brunet JP, Subramanian A, Ross KN, Reich M, Hieronymus H, Wei G, Armstrong SA, Haggarty SJ, Clemons PA, Wei R, Carr SA, Lander ES, Golub TR (2006) The connectivity map: using gene-expression signatures to connect small molecules, genes, and disease. Science 313(5795):1929–1935 NRC: National Research Council (2007) Toxicity testing in the 21st century: a vision and a strategy. National Academies Press, Washington DC OECD: Organisation for Economic Co-operation and Development (2013) Guidance document on developing and assessing adverse outcome pathways. OECD environment, health and safety publications—series on testing and assessment, No.184. Paris, France Pastoor TP, Bachmann AN, Bell DR, Cohen SM, Dellarco M, Dewhurst IC, Doe JE, Doerrer NG, Embry MR, Hines RN, Moretto A, Phillips RD, Rowlands JC, Tanir JY, Wolf DC, Boobis AR (2014) A 21st century roadmap for human health risk assessment. Crit Rev Toxicol 44(S3):1–5 Patlewicz G, Ball N, Becker RA, Booth ED, Cronin MT, Kroese D, Steup D, van Ravenzwaay B, Hartung T (2014) Read-across approaches: misconceptions, promises and challenges ahead. ALTEX 31(4):387–396 Rotroff DM, Wetmore BA, Dix DJ, Ferguson SS, Clewell HJ, Houck KA, Lecluyse EL, Andersen ME, Judson RS, Smith CM, Sochaski MA, Kavlock RJ, Boellmann F, Martin MT, Reif DM, Wambaugh JF, Thomas RS (2010) Incorporating human dosimetry and exposure into high-throughput in vitro toxicity screening. Toxicol Sci 117(2):348–358 Thomas RS, Philbert MA, Auerbach SS, Wetmore BA, Devito MJ, Cote I, Rowlands JC, Whelan MP, Hays SM, Andersen ME, Meek ME, Reiter LW, Lambert JC, Clewell HJ 3rd, Stephens ML, Zhao QJ, Wesselkamper SC, Flowers L, Carney EW, Pastoor TP, Petersen DD, Yauk CL, Nong A (2013) Incorporating new technologies into toxicity testing and risk assessment: moving from 21st century vision to a data-driven framework. Toxicol Sci 136(1):4–18 Tice RR, Austin CP, Kavlock RJ, Bucher JR (2013) Improving the human hazard characterization of chemicals: a Tox21 update. Environ Health Perspect 121(7):756–765 WHO: World Health Organization (2014) Guidance document on evaluation and expressing uncertainty in hazard characterization. IPSC harmonisation project document, no. 11, Geneva, Switzerland Wu S, Blackburn K, Amburgey J, Jaworska J, Federle T (2010) A framework for using structural, reactivity, metabolic and physicochemical similarity to evaluate the suitability of analogs for SAR-based toxicological assessments. Regul Toxicol Pharmacol 56(1):67–81
13
