The role of chemistry in developing understanding of adverse outcome pathways and their application in risk assessment

Abstract

The Adverse Outcome Pathway (AOP) conceptual framework has been presented as a logical sequence of events or processes within biological systems which can be used to understand adverse effects and refine the current risk assessment practice. This approach shifts the risk assessment focus from traditional apical endpoints to the development of a mechanistic understanding of a chemicals effect at a molecular and cellular level. In order to obtain this level of detail, chemistry in all its disciplines has a key role to play. Measurement techniques will be important in understanding chemical characterisation, free concentration and exposure at the site of interest. Such measurements will be vital in developing structure-based toxicological alerts and informing predictive models. This paper explores the areas where chemistry will be influential in the development of AOPs.

---

Dr Steve Gutsell

Dr Steve Gutsell, Ph.D., has been with Unilever's Safety and Environmental Assurance Centre for over 9 years. With a background in Organic Chemistry and expertise in the area of Computational Chemistry, specifically using predictive methods such as (Quantitative) Structure–Activity Relationships ((Q)SAR), Read Across and other techniques to predict both toxicological and ecotoxicological endpoints from chemical structure. He has published several papers in this area and presented at numerous international scientific and regulatory meetings. Recent areas of interest include how pathways-based approaches can be used to create novel risk assessments for consumer products.

Dr Paul Russell

Dr Paul Russell, Ph.D. CChem MRSC, has over 15 years industrial experience in analytical chemistry, working in the pharmaceutical industry and contract research before joining Unilever's Safety and Environmental Assurance Centre in 2004. He is a technical specialist in liquid chromatography and mass spectroscopy and has published a number of peer reviewed articles in this area and regularly presents at international conferences. Dr Russell currently has a specific focus on the development of mechanistic chemistry based approaches to support pathways based toxicological risk assessments for new materials. He is also Secretary of the Separation Science Group of the Royal Society of Chemistry.

---

Introduction

Toxicological and ecotoxicological risk assessment of chemicals is undertaken for a number of reasons. The risk assessments that inform safety and regulatory decision making around marketing products require a high degree of accuracy. They should consider a detailed assessment of the route and extent of exposure and consider all relevant potential effects on human health and the environment. This type of assessment is often subject to a high level of scrutiny through internal management processes and external peer review. In comparison, safety risk assessments performed early in the product development process are often used as a means to narrow the number of lead chemicals being considered. As such they are usually performed at a less detailed (or ‘screening’) level compared to assessments completed in support of marketing a product containing a new chemical (or an existing chemical in a different product). Nonetheless they are an important step in product development. This screening type of assessment may be carried out on a relatively simple hazard basis or, if the potential exposure is known, it may use threshold-based tools such as the Threshold of Toxicological Concern (TTC).¹ At this stage throughput is often more important than very high levels of accuracy. Hence, in silico and in vitro methods are heavily used in this early screening context. For classification and labelling purposes a more precautionary approach is often favoured as the focus tends to be on protecting workers and members of the public in an occupational/transportation scenario.

The science that underpins the safety risk assessment of consumer goods is currently undergoing one of the largest paradigm shifts in recent history.^2,3 To a large extent this shift is being catalysed by the need to assess increasing numbers of chemicals with fewer resources. Increasing public and political concerns regarding the use of animal studies for assuring the safety of new chemicals have also driven extensive research into non-animal methodologies.⁴ Whilst progress has been made for several toxicological endpoints there still remain many gaps in the knowledge. There is also a desire to reduce the uncertainty inherent in many current risk assessment practices e.g. from extrapolation of effects interspecies, from interindividual variation, and from one route of exposure to another. Many of these uncertainties are addressed through the application of factors (uncertainty/safety/assessment) to the risk assessment equations. However, the extent to which these factors accurately account for such uncertainties is, at best, variable.⁵

Numerous approaches have been put forward to address individual elements of this overall challenge.⁶ However, until recently it was extremely difficult to see how these individual pieces of the puzzle might fit together to achieve the overall aims. The pathways concept attempts to do just that. Whilst different terminology and levels of detail have developed in different areas depending on the scope of the approach⁷ (see Fig. 1), the basic premise is the same: by understanding more about the chemical and biological mechanisms involved at different levels of biological organisation and how they are related it should be possible to predict outcomes at higher levels of organisation from information obtained at lower levels.

Fig. 1 Scope of pathways approaches (adapted from Crofton 2010).

To illustrate the pivotal role that chemistry as a multi-faceted discipline has to play in these pathways-based approaches, this paper will focus on the main areas of application in an Adverse Outcome Pathways (AOP) based approach to refining risk assessment.

Adverse outcome pathways

The AOP conceptual framework⁸ (see Fig. 2) originally emerged from the environmental risk assessment community and can be defined as a sequence of events from the exposure of an individual to a chemical through to an understanding of the adverse effect at the individual level (for human health) or population level (for ecotoxicological effects).⁹ AOPs span multiple levels of biological organisation and despite often being depicted as a linear series of events (see the narcosis AOP for aquatic toxicity depicted in Fig. 2B)^10,11 between the Molecular Initiating Event (MIE) and the apical endpoint, may be non-linear, and complex (see the AOP for skin sensitisation in mammals in Fig. 2C).¹²

Fig. 2 (A) General AOP framework, (B) narcosis AOP for aquatic toxicity, (C) skin sensitisation AOP in mammals. (A) and (B) adapted from Ankley et al. 2010 (ref. 6). (C) Adapted from Fig. 3 ‘Flow diagram of the pathways associated with skin sensitization’ (p. 27) from OECD (2012) Series on Testing and Assessment No. 168 ENV/JM/MONO(2012)10/PART1 The Adverse Outcome Pathway for Skin Sensitisation Initiated by Covalent Binding to Proteins http://search.oecd.org/officialdocuments/displaydocumentpdf/?cote=env/jm/mono(2012)10/part1&doclanguage=en.

The initial interaction, or MIE, can provide a link to subsequent outcomes/effects at different levels of biological organisation and other key dimensions such as gender or life stage. It is readily accepted that a single MIE can lead to multiple adverse outcomes and vice versa. Prior to the occurrence of an MIE it is critical to understand the chemical source and conditions of the exposure scenario in order to fully understand the interaction occurring at the MIE itself. This includes an understanding of the possible speciation, metabolism and/or degradation of a chemical prior to the MIE which may render it more or less active.

The AOP approach allows commonalities to be leveraged across the human health and environmental domains, breaking complex toxicology down into focussed biological and chemical processes which allow the influence of a chemical and its derivatives to be more readily understood. This includes building an appreciation of the influence of kinetics at each stage of a pathway to enable qualitative modelling to become quantitative.¹³ Consideration of this dose–response, or response–response, relationship at each organisational level will be required to determine whether the biological self-protection processes are overcome, resulting in an adverse rather than adaptive response mechanism.¹⁴ The degree to which this complexity at the intervening levels of organisation is required to be understood will vary according to the needs of the assessment in question. For example, it could be that an understanding of the effects of a chemical at an organelle level are sufficient to inform an early screening risk assessment.

Chemistry in all its disciplines (i.e. experimental, predictive, theoretical and computational) has long been a key enabler in safety risk assessment. The ability to link an adverse effect to a measured dose of a defined molecular structure requires the use of chemical techniques, typically analytical experimental methods, to measure the input and output of such a study.

In order to meet the demands of a pathways-based approach to risk assessment an in-depth understanding of many currently poorly understood biological processes will be required. This challenge is made easier through the inherent structured approach that the AOP framework presents. The application of existing chemistry based technologies applied in novel ways and maximising the use of established knowledge will allow a bespoke, fit for purpose, integrated, approach to testing/predictions to be developed.

(Quantitative) structure–activity relationships

One traditional interface between chemistry and toxicology has been the development and application of (Quantitative) Structure Activity Relationships ((Q)SAR). Since the mid 19^th century it has been known that the structure of a chemical holds clues as to its potential effects on biological systems.¹⁵ The use of in silico models to predict toxicity has been applied to varying degrees for many years. QSARs are widely accepted risk assessment tools for the prediction of environmental fate and effects, partly as a result of the pioneering work done by the US EPA.¹⁶ In the human health area the use of (Q)SARs has been somewhat more cautious. The historical application of (Q)SAR in toxicology has focussed on attempting to predict apical endpoints at the individual organism level (e.g. LC₅₀, EC₃, NOEL, etc.) from chemical structure. Whilst this has had some success for several well defined mechanisms of toxicity (e.g. skin sensitisation^17,18), it has generated mixed results when applied to more complicated, less well understood, endpoints related to systemic effects.¹⁹

Advances in computer processing power are allowing the calculation of quantum chemical parameters, even from ab initio techniques, on standard computing equipment. This has greatly increased the accessibility of 3-dimensional descriptors necessary to model protein–ligand interactions and Quantum Mechanical (QM) descriptors related to electrophilic/nucleophilic reactivity. On the negative side the number of descriptors that can be calculated in seconds is now vast and can result in QSAR models that are very difficult to interpret. In addition, the use of complex descriptors and statistical methods with little thought as to the mechanistic interpretation of a model has greatly damaged the reputation of QSAR techniques. The need to improve transparency and interpretation of QSAR models have been widely accepted,²⁰ but there has still been a proliferation of models published which do not add to the understanding of the mechanisms involved in toxicological processes.²¹ The result is that for human health endpoints QSAR methods have largely been restricted to use as early screens or as a means of prioritising chemicals for subsequent testing.

A chemical's electrophilic reactivity is thought to be responsible for a number of different potential adverse effects. Whilst the use of computational approaches to understand reactivity and develop predictive models has become more prevalent,²² the need for high quality experimental measurements of reactivity has been largely overlooked. To allow the shift to an AOP-based approach to risk assessment and the development of predictive models, the interdependence between experimental measurements and predictive modelling (at each level of organisation) requires increased emphasis. Clearly this area offers many opportunities for chemists and (eco)toxicologists to work together to develop high quality data sets of chemical properties to act as training sets for predictive models.

Often simpler models that take the form of structure–activity relationships are more accepted as inputs to risk assessment as they are transparent and easy to implement. There are many examples of collections of these SARs that have been combined into either commercial or free software tools.^23,24 A further application of SAR-type rules is in read across. This approach is widely applied as a means to fill gaps in hazard data packages for chemicals with closely related analogues which do possess data. There is much debate around the best approaches to read across and its limitations. Again the structured nature of the AOP approach can be used to provide increased evidence of the validity of a read across proposal. It will also allow informed decisions to be made about the need for further testing at higher levels of biological organisation.

Chemical characterisation

Advances in analytical technology are allowing lower detection limits to be reached. With lower limits, the number of chemicals detected in products and the environment increases.²⁵ In addition more complete profiles of individual chemicals are being obtained including previously undetected low level contaminants present. High quality chemical characterisation data is therefore essential in any risk assessment (pathways based or otherwise) to understand what constitutes the actual ‘exposure’ at any given level of biological organisation.

An effective characterisation of a chemical should give detailed information about its physical and chemical composition along with identification and quantification of any impurities present. A thorough knowledge of a chemical under investigation is the foundation for good experimental design. It also helps with interpretation of the results which is imperative for the success of any in vivo or in vitro study. However, this is an issue that is often not addressed sufficiently, or even overlooked completely. Inputs into predictive models such as (Q)SARs tend to be single chemical structures, yet the experiments that generate the parameters which feed into these models for a given chemical, either knowingly or unknowingly, rarely have such high purity.

A comprehensive characterisation study may include chromatographic techniques (HPLC, GC), spectroscopic techniques (NMR, FTIR, MS) along with traditional wet chemistry measurements (pH, pK_a, log P) but will more likely be a combination of a number of these. Low sensitivity techniques such as Nuclear Magnetic Resonance (NMR) and Fourier Transform Infra-Red (FTIR) spectroscopy may be employed early on in the research process to support screening activities. More sensitive techniques (i.e. Mass Spectrometry (MS)), although equally capable at the screening stage, can be used to unravel more complex problems such as identification of low level impurities or metabolites. Biological based in vitro assays, like in vivo assays, involve multiple stresses on a dosed chemical (e.g. metabolism, degradation, binding, pH effects) and even for a well characterised chemical the outcome is often complex to interpret. It is certainly not possible to fully understand the results of an experiment without sound knowledge of the chemical that you are presenting and analytical chemistry has a key role to play here.

Chemical purities often need to be determined to low levels and the role of any salts or counter-ions understood to predict or interpret results from in vitro assays. Impurities that are detected and unavoidable should likewise be characterised and NMR spectroscopy is particularly effective here as it has both qualitative and quantitative capabilities, although this may then lead to subsequent MS studies if sensitivity is ultimately found to be the limiting factor. Generally with modern equipment the lower sensitivity techniques are adequate for detecting impurities at levels suitable to support in vitro assays. According to the International Conference on Harmonisation (ICH) guidelines for pharmaceuticals,²⁶ identification of impurities below 0.1% (for a 2 g day⁻¹ dose) is not necessary unless there is evidence of their toxicity. The level at which impurities can be detected will depend largely on the impurity in question and the detection system, but when attempting to gain mechanistic understanding from in vitro assays a simple awareness of the potential effect of low level impurities can be an advantage when interpreting results.²⁷ Hence a pragmatic evidence-based approach should be taken to the characterisation and risk assessment of low level impurities guided by experimental methods and tools such as the TTC.²⁸

Biological interactions are often driven by a chemicals three dimensional conformation and hence knowledge of structural isomers is important, albeit these can be challenging to detect. Chemical stability should be well understood, not only in its natural physical state, but also in solution, formulation and within the relevant biomimetic system.

Time spent understanding the chemical of interest will facilitate better understanding of any experimental output through the reduction of uncertainty. In any multi-parameter experimental assay it is critical to identify potential sources of variability and ensure any controlled variables such as the input chemical are well understood. This in turn leads to the development of robust predictive models by improving the quality of the data used to build them.

Molecular initiating events

The (MIE) can be defined as the initial point of chemical–biological interaction within the organism that starts the AOP. MIEs can be thought of as the “gateway” to potential AOPs. Through an understanding of the chemical attributes (SAR) required to invoke an MIE it could be possible to filter the number of potential pathways that need to be considered in the risk assessment for a given chemical.⁷ Whilst this binary interpretation of MIEs may prove useful in screening level risk assessments, a more quantitative understanding of MIEs is desirable to understand the relationship between the potency (dose response) of a chemical, with respect to a given MIE, and/or the downstream effects at a higher level of biological organisation. To develop this quantitative understanding of MIEs and subsequently predictive models, it is necessary to develop suitable training sets of chemicals with measured dose response characteristics. As the MIE can be distilled to a chemical interaction that is necessary for downstream events to occur, it is possible to investigate MIEs in their purest sense. One can imagine that a cycle of wet-lab experiments to quantify MIEs and in silico modelling of the resulting data is needed to reach a point where a chemical can not only be grouped according to the MIEs that it could be associated with, but also a series of associated MIE potencies can be predicted (see Fig. 3).

Fig. 3 Wet/dry cycle for development of in silico models.

The predictivity and applicability domain of such models will clearly improve as the size of the training sets increases. Pragmatic decisions will need to be made as to when a model is fit for purpose. This will obviously depend on the risk assessment decision in question.

As the AOP conceptual framework is predicated by the need to understand the links between effects at different levels of organisation it follows that it should be possible to relate events at the MIE level to those further downstream on the pathway. Indeed this has been the fundamental concept behind the development of many (Q)SARs. By understanding the quantitative (or qualitative) interconnections between the levels of organisation it will become apparent how far downstream MIE information alone will allow predictions to be made. This should prevent the production of models that attempt to make predictions without a mechanistically plausible connection to an appropriate endpoint.

Once again the importance of a thorough understanding of the actual amount of chemical eliciting a response at the MIE should not be underestimated as this will underpin any predictions of events further down the pathway.

In vitro to in vivo extrapolation

Sensitive in vitro (cell-based and/or in chemico) assays are important for determining biological responses and extrapolating thresholds for determining safe consumer concentrations and understanding the potential burden being placed on ecosystems. There are clear limitations in the use of in vitro assays to mimic whole organism behaviour, not least because in vitro assays generally consider single or small numbers of cells with limited target sites rather than a holistic organism.^29,30 One approach to address this shortcoming is to use large batteries of diverse in vitro assays, such as the ToxCast programme.³¹ Their output is then integrated into large datasets for interpretation using bioinformatic approaches. An alternative approach is the development of more complex assays which include multiple biological processes such as tissue engineering,³² or person on a chip.³³

It is evident that in vitro assays are seen as a critical tool in advancing our knowledge of AOP's so long as there outputs are well understood in terms of their toxicological significance.³⁴ Traditionally the nominal dose introduced to the system has been used to describe chemical exposure within in vitro assays without consideration of how much chemical is available to act on a cell, or a target within a cell. This introduces significant error when extrapolating from in vitro toxicity data to in vivo scenarios where the exposure at the target site is expected to be inherently different. Risk assessment approaches can be refined through quantitatively understanding the free chemical available to act within the in vitro experiment.³⁵ This has been identified as a key priority by the U.S. National Research Council in their strategic vision for toxicity testing in the 21^st century.³Fig. 4³⁶ illustrates the factors affecting free concentration include binding to components in the cell media such as serum proteins^37–40 and binding to the glass or plastic in solid supports and labware.^41–43 Chemicals are often dosed into in vitro assays in mixed solvent systems to aid their solubility, but subsequent introduction into a wholly aqueous cellular system can cause precipitation, again reducing the amount of chemical freely available. In addition, non-target binding can occur within the cell itself (i.e. to membranes and organelles) and cell metabolism or degradation can lead to a reduction in the applied dose of a chemical.

Fig. 4 In vitro to in vivo extrapolation. Understanding the interactions and an awareness of the dynamic equilibria occurring within in vitro assays is essential when using data to extrapolate to in vivo situations (adapted from Kramer et al. 2012). Emphasis should be placed on understanding the free concentration available to reach the target and have an effect.

These potential in vitro losses will apply to both human and environmental toxicology assays. The chemical specific equilibria that exist between non-targeted binding and losses make the determination of actual target concentrations particularly challenging with any analytical sampling procedures potentially adding further complexity. The development of small volume extraction techniques such as solid phase micro extraction^44,45 can be employed to provide highly specific, matrix compatible extractions at focussed target sites within an assay with minimal disruption to the in vitro assay equilibria.⁴⁶

In order to reproducibly and accurately predict in vivo effects from in vitro data, a key enabler for AOP-based toxicological risk assessment, these non-target interactions are important parameters to consider. A sequence of in vitro experiments should be designed to determine the free concentration, whether by direct analysis or by developing a quantitative understanding of some or all of the non-targeted interactions described in Fig. 4. Physical chemical parameters (measured and predicted) of the chemical can potentially be used to predict many of these interactions. Through an improved understanding of free concentration the validity of the many predictive models that will be developed from in vitro assay data will be improved.

Physiologically based pharmacokinetics

Physiologically Based Pharmacokinetic (PBPK) models seek to simplify the complex processes involved in the Adsorption, Distribution, Metabolism and Elimination (ADME) of chemicals in a given species. They have been largely developed for application within the pharmaceutical sector, but are also finding increasing application in risk assessment of industrial chemicals.⁴⁷ In their simplest form PBPK models can be thought of as a series of compartments (corresponding to organs) with interconnections to represent blood flow (or other circulatory processes). Through a series of differential equations they allow the calculation of the quantity of chemical in a given compartment at a given time. In so doing these models allow for a more sophisticated understanding of chemical exposure which is relevant to understanding the risk of potential adverse events.

The inputs to PBPK models can be categorised as chemical independent (e.g. organ weights, blood flow etc.) and chemical dependent (e.g. plasma protein binding, various partition coefficients and metabolic clearance rates). A clear barrier to the use of this kind of model in environmental risk assessment is the range of different species that need to be considered. However, many elements of PBPK modelling are similar to those used in environmental fate models that also attempt to predict the kinetics of a chemical as it partitions to various environmental compartments and possibly biodegrades.

Whilst many QSAR models exist to provide some of the chemical dependent inputs to PBPK models for those chemicals without experimental values, the applicability domain of such models is somewhat restricted. In addition, methods to predict metabolism rates using either in silico or in vitro models are currently limited.⁴⁸ It is also fair to say that many of the QSARs themselves rely on predicted descriptors as input (e.g. log P).⁴⁹ In this situation it is clear that errors may be introduced into the final PBPK predictions at several stages. Acquisition of accurate measured data wherever possible as input to the PBPK modelling process will clearly remove some of the early sources of potential error, or at least assist in highlighting the source of error, and result in more reliable outputs.

One of the major barriers to the use of PBPK modelling to understand relevant exposure from home and personal care products (amongst others) is the lack of a reliable method to predict skin penetration.⁵⁰ As with other topics discussed previously there is a need for further quality data to be generated to facilitate the development of predictive models.⁵¹ This issue is further complicated by the knowledge that the nature of the formulation/vehicle in which a chemical is applied can have a large impact on the penetration. The current state of in silico modelling for topical exposure is very basic. This is largely due to the lack of data obtained under consistent comparable conditions and data designed to elucidate the effect of vehicle on penetration. Clearly this is an area that warrants further experimental work.⁵² In particular the influence of the physical–chemical properties of formulations on penetration or absorption should be investigated.

Conclusions

AOP-based approaches are gaining acceptance as a conceptual framework under which novel risk assessment techniques will be developed across both environmental and human health communities. More than ever the fundamental chemistry of interactions with biological systems will be pivotal to the success of any new methods that are developed under the AOP framework. Identification and development of SARs based on knowledge of MIEs (gateways to AOPs) will provide an invaluable tool to facilitate screening large numbers of new and existing chemicals by narrowing the number of AOPs that need to be assessed in more detail for a given chemical. These SARs will also provide a basis for mechanistically transparent read-across by allowing the formation of groups or categories around a biological pathway. This approach is already proving useful for filling hazard data gaps in, for example, REACH dossiers.⁵³

Experimental quantification of MIEs will lead to the development of QSARs for both the MIE itself and downstream effects. This quantitative understanding of an MIE together with an understanding of the events at higher levels of biological organisation will enhance the mechanistic interpretation of QSARs used to predict biological “endpoints” by limiting predictions to only those effects that are plausibly linked to the MIE. This should increase the credibility of QSAR approaches if they are presented as part of an integrated pathways-based risk assessment.

Generating data from multiple assays at different levels of biological organisation will result in a plethora of data that will require interpretation. Developments in the field of systems biology are being facilitated by advances in measurement science that enable the monitoring of biological processes at a molecular level in real time.²⁶ In addition such analytical technologies, when applied correctly within chemical characterisation studies, can inform exposure to high sensitivities. The development of an understanding of metabolism, degradation, speciation and fundamental physical–chemical properties are all challenges that can be addressed using the latest analytical chemistry approaches.

It is recognised that in vitro experimental results will be key to building mechanistic understanding of AOPs. However, when extrapolating in vitro results to in vivo risk assessment scenarios care should be taken to fully understand assay variables and the free concentration in vitro. In turn this improved chemical specific input to PBPK models will facilitate conversion of consumer exposure to free concentrations at target sites. The next step in the ambition would be to obtain a quantitative dose response at these target sites which will be key to the success of a truly refined AOP-based risk assessment.

The AOP conceptual framework is already proving useful in providing biological plausibility to observed correlations between chemical structure and biological effects. As more detail of more pathways becomes available this utility will also increase. Quantitative modelling of an entire pathway is a long term ambition. This will only be achievable after first understanding the pathway in a qualitative fashion. It will then be necessary to quantify the relevant interactions through the development of suitable assays, the results of which can be fed into holistic predictive models.

To address the myriad of challenges associated with the proposed changes to risk assessment, it will be necessary to bring the combined expertise and techniques of several chemistry-based disciplines together with those from other fields. Each of these disciplines has a role to play, but the exact nature of this role is only just becoming clear and may well be very different to that of the present situation. Rather than attempting to address all the complex questions of an AOP approach at once, an integrated chemistry approach can guide research and prioritise activities in a pragmatic, fit for purpose manner. This would allow the ultimate aims of replacing the use of animal studies, improving efficiency and reducing the uncertainty in current risk assessment practices to be achieved.

Acknowledgements

The authors would like to thank Beate Nicol and Nora Aptula for their contribution and critical reviews of this paper.

Notes and references

1. SCCS, SCHER, SCENIHR, Joint Opinion on the Use of the Threshold of Toxicological Concern (TTC) Approach for Human Safety Assessment of Chemical Substances with focus on Cosmetics and Consumer Products, 8 June 2012.
2. SCENIHR (Scientific Committee on Emerging and Newly Identified Health Risks), SCHER (Scientific Committee on Health and Environmental Risks), SCCS (Scientific Committee on Consumer Safety), Preliminary report on Addressing the New Challenges for Risk Assessment, 8 October 2012.
3. Committee on Toxicity Testing and Assessment of Environmental Agents, and National, R. C. (2007). Toxicity Testing in the 21st Century: A Vision and a Strategy, The National Academies Press.
4. S. Adler, D. Basketter, S. Creton, O. Pelkonen, J. Benthem, V. R. Zuang, K. Andersen, A. Angers-Loustau, A. Aptula, A. Bal-Price, E. Benfenati, U. Bernauer, J. Bessems, F. Bois, A. Boobis, E. Brandon, S. Bremer, T. Broschard, S. Casati, S. Coecke, R. Corvi, M. Cronin, G. Daston, W. Dekant, S. Felter, E. Grignard, U. Gundert-Remy, T. Heinonen, I. Kimber, J. Kleinjans, H. Komulainen, R. Kreiling, J. Kreysa, S. Leite, G. Loizou, G. Maxwell, P. Mazzatorta, S. Munn, S. Pfuhler, P. Phrakonkham, A. Piersma, A. Poth, P. Prieto, G. Repetto, V. Rogiers, G. Schoeters, M. Schwarz, R. Serafimova, H. Tahhti, E. Testai, J. Delft, H. Loveren, M. Vinken, A. Worth and J. M. Zaldivar, Alternative (non-animal) methods for cosmetics testing: current status and future prospects, Arch. Toxicol., 2011, 85(5), 367–485.
5. Interdepartment Group on Health Risks from Chemicals, Uncertainty Factors: Their use in human health risk assessment by UK Government, (2003), Institute for Environment and Health.
6. B. J. Blaauboer, M. D. Barratt and J. B. Houston, The integrated use of alternative methods in toxicological risk evaluation, Altern. Lab. Anim., 1999, 27, 229–237.
7. OECD (2011). Report of the Workshop on Using Mechanistic Information in Forming Chemical Categories. OECD Environment, Health and Safety Publications Series on Testing and Assessment No. 138. ENV/JM/MONO(2011)8.
8. G. T. Ankley, R. S. Bennett, R. J. Erickson, D. J. Hoff, M. W. Hornung, R. D. Johnson, D. R. Mount, J. W. Nichols, C. L. Russom, P. K. Schmieder, J. A. Serrrano, J. E. Tietge and D. L. Villeneuve, Adverse outcome pathways: a conceptual framework to support ecotoxicology research and risk assessment, Environ. Toxicol. Chem., 2010, 29(3), 730–741.
9. OECD, Proposal for A Template, and Guidance on Developing and Assessing the Completeness of Adverse Outcome Pathways, Appendix I, Collection of Working Definitions, http://www.oecd.org/chemicalsafety/testingofchemicals/49963576.pdf.
10. H. J. M. Verhaar, C. J. van Leeuwen and J. L. M. Hermens, Classifying environmental pollutants, Chemosphere, 1992, 25, 471–491.
11. C. L. Russom, S. P. Bradbury, S. J. Broderius, D. E. Hammermeister and R. A. Drummond, Predicting mode of toxic action from chemical structure: acute toxicity in the fathead minnow (Pimephales promelas), Environ. Toxicol. Chem., 1997, 16, 948–967.
12. Adapted from Fig. 3 ‘Flow diagram of the pathways associated with skin sensitization’ (p. 27) from OECD (2012) Series on Testing and Assessment No. 168 ENV/JM/MONO(2012)10/PART1 The Adverse Outcome Pathway for Skin Sensitisation Initiated by Covalent Binding to Proteins http://search.oecd.org/officialdocuments/display documentpdf/?cote=env/jm/mono(2012)10/part1&doclanguage=en.
13. C. MacKay, M. Davies, V. Summerfield and G. Maxwell (submitted/personal communication). From pathways to people: applying the adverse outcome pathway (AOP) for skin sensitisation to risk assessment, ALTEX.
14. V. J. Kramer, M. A. Etterson, M. Hecker, C. A. Murphy, G. Roesijadi, D. J. Spade, J. A. Spromberg, M. Wang and G. T. Ankley, Adverse outcome pathways and ecological risk assessment: bridging to population-level effects, Environ. Toxicol. Chem., 2011, 30(1), 64–76.
15. A. Crum Brown and T. Frazer, On the connection between chemical constitution and physiological action, Trans. – R. Soc. of Edinburgh, 1868–9, 25, 151–203.
16. S. Bradbury, Quantitative structure–activity relationships and ecological risk assessment: an overview of predictive aquatic toxicity research, Toxicol. Lett., 1995, 79, 229–237.
17. G. Patlewicz, A. Aptula, D. Roberts and E. Uriarte, A minireview of available skin sensitisation (Q)SARs/expert systems, QSAR Comb. Sci., 2008, 27, 60–76.
18. D. Roberts, A. Aptula, M. Cronin, E. Hulzebos and G. Patlewicz, Global (Q)SARs for skin sensitisation – assessment against OECD principles, SAR QSAR Environ. Res., 2007, 18, 343–365.
19. S. Lapenna, M. Fuart-Gatnik and A. Worth, Report of QSAR Models and Software Tools for Predicting Acute and Chronic Systemic Toxicity, JRC Scientific and Technical Reports, EUR24639 EN-2010.
20. A. P. Worth and M. T. D. Cronin, Report of the workshop on the validation of QSARs and other computational prediction models, Altern. Lab. Anim., 2004, 32, 703–706.
21. M. T. D. Cronin and T. W. Schultz, Pitfalls in QSAR, J. Mol. Struct. (THEOCHEM), 2003, 622(1–2), 39–51.
22. S. J. Enoch, C. M. Ellison, T. W. Schultz and M. T. Cronin, A review of the electrophilic reaction chemistry involved in covalent protein binding relevant to toxicity, Crit. Rev. Toxicol., 2011, 41, 783–802.
23. Derek Nexus v.3.0.1., Copyright 2012, Lhasa Limited.
24. Toxtree (Estimation of Toxic Hazard – A Decision Tree Approach, v.2.5.1., Ideaconsult Limited.
25. J. W. DeVries, Chasing “Zero” in Chemical Contaminant Analysis, Food Safety Magazine, August/September, 2006.
26. U.S. Department of Health and Human Services Food and Drug Administration. ICH Guidance for Industry Report Q3A – Impurities in New Drug Substances (Revision 2). 2008.
27. A. Natsch, H. Gfeller, R. Emter and G. Ellis, Use of in vitro testing to identify an unexpected skin sensitizing impurity in a commercial product: a case study, Toxicol. in Vitro, 2010, 24(2), 411–416.
28. R. Kroes, J. Kleiner and A. Renwick, The threshold of toxicological concern concept in risk assessment, Toxicol. Sci., 2005, 86(2), 226–230.
29. D. A. Basketter, J. F. McFadden, F. Gerberick, A. Cockshott and I. Kimber, Nothing is perfect, not even the local lymph node assay: a commentary and the implications for REACH, Contact Dermatitis, 2009, 60(2), 65–69.
30. A. J. Murk, E. Rijntjes, B. J. Blaauboer, R. Clewell, K. M. Crofton, M. L. Dingemans, J. D. Furlow, R. Kavlock, J. Köhrle, R. Opitz, T. Traas, T. J. Visser, M. Xia and A. C. Gutleb, Mechanism-based testing strategy using in vitro approaches for identification of thyroid hormone disrupting chemicals, Toxicol. in Vitro, 2013, 27(4), 1320–1346.
31. R. Kavlock, The future of toxicity testing – the NRC vision and the EPA's ToxCast program national center for computational toxicology, Neurotoxicol. Teratol., 2009, 31(4), 237.
32. V. Y. Soldatow, E. L. LeCluyse, L. G. Griffith and I. Rusyn, In vitro models for liver toxicity testing, Toxicol. Res., 2013, 2(1), 23–39.
33. U. Marx, H. Walles, S. Hoffmann, G. Lindner, R. Horland, F. Sonntag, U. Klotzbach, D. Sakharav, A. Tonevitsky and R. Lauster, ‘Human-on-a-ship’ developments: a translational cuttingedge alternative to systemic safety assessment and efficiency evaluation of substances in laboratory animals and man?, Altern. Anim. Test., 2012, 40(5), 235–227.
34. B. J. Blaauboer, K. Boekelheide, H. J. Clewell, M. Daneshian, M. M. L. Dingemans, A. M. Goldberg, M. Heneweer, J. Jaworska, N. I. Kramer, M. Leist, H. Seibert, E. Testai, R. J. Vandebriel, J. D. Yager and J. Zurlo, The use of biomarkers of toxicity for integrating in vitro hazard estimates into risk assessment for humans, ALTEX, 2012, 29(4), 411–425.
35. B. J. Blaauboer, Biokinetic modeling and in vitro–in vivo extrapolations, J. Toxicol. Environ. Health, Part B, 2010, 13, 242–252.
36. N. I. Kramer, M. Krismartina, A. Rico-Rico, B. J. Blaauboer and J. L. M. Hermens, Quantifying processes determining the free concentration of phenanthrene in basal cytotoxicity assays, Chem. Res. Toxicol., 2012, 25(2), 436–445.
37. M. Gülden, S. Mörchel, S. Tahan and H. Seibert, Impact of protein binding on the availability and cytotoxic potency of organochlorine pesticides and chlorophenols in vitro, Toxicology, 2002, 175(1–3), 201–213.
38. M. B. Heringa, R. H. M. M. Schreurs, F. Busser, P. T. Van Der Saag, B. Van Der Burg and J. L. M. Hermens, Toward more useful in vitro toxicity data with measured free concentrations, Environ. Sci. Technol., 2004, 38(23), 6263–6270.
39. E. V. Hestermann, J. J. Stegeman and M. E. Hahn, Serum alters the uptake and relative potencies of halogenated aromatic hydrocarbons in cell culture bioassays, Toxicol. Sci., 2000, 53(2), 316–325.
40. H. Seibert, S. Mörchel and M. Gülden, Factors influencing nominal effective concentrations of chemical compounds in vitro: medium protein concentration, Toxicol. in Vitro, 2002, 16(3), 289–297.
41. J. Riedl and R. Altenburger, Physicochemical substance properties as indicators for unreliable exposure in microplate-based bioassays, Chemosphere, 2007, 67(11), 2210–2220.
42. K. Schirmer, A. G. J. Chan, B. M. Greenberg, D. G. Dixon and N. C. Bols, Methodology for demonstrating and measuring the photocytotoxicity of fluoranthene to fish cells in culture, Toxicol. in Vitro, 1997, 11(1–2), 107–119.
43. R. Schreiber, R. Altenburger, A. Paschke and E. Küster, How to deal with lipophilic and volatile organic substances in microtiter plate assays, Environ. Toxicol. Chem., 2008, 27(8), 1676–1682.
44. H. Lord and J. Pawliszyn, Evolution of solid-phase microextraction technology, J. Chromatogr., A, 2000, 885(1–2), 153–193.
45. S. Ulrich, Solid-phase microextraction in biomedical analysis, J. Chromatogr., A, 2000, 902(1), 167–194.
46. B. Ekwall, Overview of the final MEIC results: II. The in vitro–in vivo evaluation, including the selection of a practical battery of cell tests for prediction of acute lethal blood concentrations in humans, Toxicol. in Vitro, 1999, 13(4–5), 665–673.
47. M. Reddy, R. S. Yang, M. E. Andersen and H. J. Clewell III, Physiologically based pharmacokinetic modeling: science and applications, Wiley-Interscience, 2005.
48. M. Yoon, J. L. Campbell, M. E. Andersen and H. J. Clewell, Quantitative in vitro to in vivo extrapolation of cell-based toxicity assay results, Crit. Rev. Toxicol., 2012, 42(8), 633–652.
49. J. Caldwell, M. Evans and K. Krishnan, Cutting edge PBPK models and analyses: providing the basis for future modeling efforts and bridges to emerging toxicology paradigms, J. Toxicol., 2012, 852384,  DOI:10.1155/2012/852384.
50. T. Peyret and K. Krishnan, QSARs for PBPK modelling of environmental contaminants, SAR QSAR Environ. Res., 2011, 22(1–2), 129–169.
51. M. Davies, R. Pendlington, L. Page, C. Roper, D. Sanders, C. Bourner, C. Pease and C. MacKay, Determining epidermal disposition kinetics for use in an integrated nonanimal approach to skin sensitisation risk assessment, Toxicol. Sci., 2011, 119(2), 308–318.
52. Y. Dancik, M. A. Miller, J. Jaworska and G. B. Kasting, Design and performance of a spreadsheet-based model for estimating bioavailability of chemicals from dermal exposure, Adv. Drug Delivery Rev., 2013, 65(2), 221–236.
53. T. Schultz, Adverse outcome pathways: a way of linking chemical structure to in vivo toxicological hazards, chapter 14 in In Silico Toxicology: Principles and Applications, ed. M. Cronin and J. Madden, Royal Society of Chemistry, 2010, pp. 351–376.

---