Workshop Report 29

Syndicate Session 2: Classification of chemicals according to MOA and chemical activity or other dose metrics for chemicals with specific mode of action

Participants
Moderator:     J. Hermens
Moderator     R. Altenburger
Rapporteur    D. Salvito
M. Cronin
S. Dyer
F. Fischer
M. Galay-Burgos
N. Kramer
V. Otton
E. Roex
P. Thomas
L. Vergauwen
D. Villeneuve

Introduction
While the chemical activity concept has been applied in the analysis of toxicity data for baseline toxicity, this concept could also be valuable for compounds with other modes of action (MOA). A recent ECETOC report on “Activity based relationships for aquatic ecotoxicology data”  listed examples of estimated activities for baseline toxicants as well compounds with other MOA (ECETOC, 2013). Classification into MOA is an essential element in analysing toxicity data. In addition to classification based on chemical structure, other novel techniques such as “omics” and high throughput screening (HTS) can become powerful tools in the analyses of MOA and within the adverse outcome pathways approach (AOPs). In addition to chemical activities, other dose metrics can be appropriate to analyse and understand differences in toxicity of compounds with a MOA beyond baseline toxicity.
Objectives
The objectives of this WG 2 were:
To determine the extent of chemical and toxicological domain for the use of the chemical activity concept as it is applied to neutral non-polar organic chemicals and to compounds with modes of action beyond baseline toxicity for both acute and chronic ecotoxicological effects.
To explore alternative methods for classifying the toxicological mode of action for chemicals, including the role of adverse outcome pathways in classification and to explore alternative dose metrics, and to assess the role of chemical activity as a potential complementary approach.

The participants discussed the following three themes during three breakout sessions:
Theme 1: Data for chemical activity (beyond baseline toxicity)
Theme 2: Modes of action (MOA) and classification
Theme 3: (Quantitative) adverse outcome pathways (AOP) – chemical activity and other dose metrics.
A brief introductory text has been prepared in advance for each of the three themes and finalised during the workshop (see Background information below). Each participant has agreed with this text.
Background information
Theme 1: Data for chemical activity (beyond baseline toxicity)
The theme was introduced with a presentation of the data from the ECETOC report “Activity based relationships for aquatic ecotoxicology data”. The presentation included acute and chronic algae, daphnids and fish effect data for MOA 1 and 2.
Acute fish tox data for fathead minnow and guppy
The chemical activity concept is already applied in the analysis and prediction of effect data of chemicals that act via non-polar baseline toxicity (see working group 1). Compounds with other MOAs are often “more toxic” (potent) than these base-line toxicants, at least if the toxicity data are interpreted on a Kow scale (Russom et al., 1997; Verhaar et al., 1992) or as a plot of effect concentrations versus the sub-cooled liquid solubility (ECETOC, 2013). Comparing effect data of compounds with chemical activities is more direct because comparisons can be made simply based on one parameter instead of a Kow regression. The EPA fathead minnow LC50 data represent a high quality dataset from which various toxicological modes of action can be assessed (Russom et al., 1997). Mackay (Mackay et al., 2014), for instance, have plotted the data from the EPA fathead minnow database demonstrating the applicability domain of the chemical activity approach for baseline toxicants (see Figure 4.2.1). From this study, as well a number of other publications, it is generally accepted that acute baseline toxicity occurs within an activity range of between 0.01 and 0.1. In a recently published ECETOC report, the chemical activity approach was applied to acute toxicity data for a whole range of chemicals covering different MOAs (ECETOC, 2013).

Figure 4.2.1. Fathead minnow acute toxicity data – chemical activity against water solubility. LC50 data from EPA (Russom et al., 1997). Figure reproduced from (Mackay et al., 2014)

capture-decran-2016-11-24-a-08-13-56

Figure 4.2.1 reveals some interesting trends. In particular, the chemical activity of the more hydrophobic chemicals (low Sw) with “other modes of action” is closer to the range representing baseline toxicity of 0.01-0.1 (grey band) than the less hydrophobic compounds. Similar trends have been observed for the 96-hour LC50 to guppy of reactive chemicals, for example for reactive acrylates (Freidig and Hermens, 2000; Freidig et al., 1999). The most plausible explanation for these trends is related to internal distribution of these compounds inside the organisms (Figure 3.3.1). The target site for reactive compounds is often in an aqueous environment inside the organism. An example is the interaction of reactive compounds with intracellular glutathione. More hydrophobic compounds will accumulate mainly in the cell membrane and the concentration in a more aqueous phase (intracellular or blood) will be relatively low. For those chemicals, narcosis overrules the more specific MOA and this explains the shift towards the base line activity range.
Figure 4.2.1 also includes data for polar narcosis compounds (MOA 2). Based on internal membrane concentration, these two classes (MOA 1 and 2) merge (Escher and Schwarzenbach, 2002; Vaes et al., 1998). Additional analyses are needed to get more understanding of chemical activity of MOA 1 and MOA 2 compounds.
Theme 2: Chemically-based methods for determining Modes of action (MOA) and classification
The theme was introduced with a presentation on the topic of MOA and classification, which was followed by a presentation on the application of omics in classifying chemicals according to their MOA.
Chemistry-based Modes of action (MOA) and classification
Classifying compounds according to their mode of action is important in the interpretation of ecotox data and in developing predictive models. A good example of a clear classification system is the one that was developed by the EPA Duluth lab (Russom et al., 1997). In this system a number of requirements for the assignment of a MOA to a specific compound are defined, including (a) results from fish acute toxicity syndromes and behaviour studies, (b) joint toxicity data, (c) excess toxicity (Te) and (d) similarity in chemical structure or chemical properties (e.g. reactivity towards nucleophiles) with compounds with a known MOA. This classification system has been applied to the fathead minnow database (Russom et al., 1997).
Most other classification systems are simpler and in fact are based on chemical structure. Structural alerts or rules are then applied to assign a MOA to a chemical. The Verhaar classification scheme (Verhaar et al., 1992) is another system based on four classes of chemicals representing very broad MOAs. Chemicals within these MOA’s include: (i) inert chemicals, (ii) less inert chemicals, (iii) reactive chemicals and (iv) specifically acting chemicals). An updated and improved version for the Verhaar classification scheme was recently published by Enoch et al. (Enoch et al., 2008). Automated versions to classify compounds are developed as part of an OECD toolbox. The OECD toolbox includes several other classification systems for MOA assignment (see http://www.oecd.org/chemicalsafety/risk-assessment/theoecdqsartoolbox.htm). More recently, omic approaches also have been applied in this field. According to Dom, “transcriptomics tools hold considerable promise to be used in biological response based mechanistic profiling of potential (eco)toxicants” (Dom et al., 2012).
Biologically-based mode of action classification
As it pertains to the application of chemical activity as a predictor of toxicity, at present our current understanding and data strongly supports the utility of chemical activity for predicting non-polar narcosis toxicity (Verhaar category 1). Data evaluating the applicability of chemical activity to other modes of action, such as reactive toxicity or specifically-acting toxicities, including those that cause chronic and/or sub lethal adverse effects are currently lacking. Therefore, discrimination of chemicals as predominantly baseline/non-polar narcosis toxicants versus other modes of action can have significant value for determining whether a chemical activity-based toxicity prediction is a sound basis for a risk assessment/risk management decision. The classification strategies described above are largely chemical-structure based. However, emerging biological pathway-based tools have potential to provide complementary or orthogonal approaches for binning chemicals into broad mode of action categories.
As an example, evaluation of US EPA’s ToxCast data set has identified a phenomenon of a “burst” of pathway-based activity at or near the concentrations that elicit overt cytotoxicity (Judson et al., 2015). This “burst” of activity across a wide range of assays largely associated with generalised toxic stress may serve as the high throughput in vitro analog to “baseline toxicity”. Judson et al. have proposed the use of Z-scores to identify assay responses that occur in the region of the non-specific cytotoxic burst and to distinguish these responses from those that may reflect specific biological activities against particular pathways or biological targets (Judson et al., 2015). Examining the overall chemical space encompassed by the ToxCast chemicals, it was broadly identified that pharmaceuticals and pesticides (compounds designed to interact with specific targets) were the most likely to have biological activities at concentrations well below the cytotoxic burst region. In contrast, while industrial chemicals often showed a diversity of biological activities, those were more likely to be reflective of generalised toxic stress and were activated at or very near the cytotoxic burst region. These results all allude to the potential of using such biologically-based high throughput data to differentiate narcosis/baseline type toxicants for which chemical activity approaches can be applied with good predictive confidence, versus those for which more pathway/target specific approaches may be needed.
While the current assumption is that the “cytotoxic burst” phenomenon is an in vitro analog to baseline toxicity, this assumption has not been explored experimentally. This represents an interesting topic for research moving forward. In the near term, one could envision two efforts which might provide insights into the validity of this assumption. First, based on the data of Judson (Judson et al., 2015), concentrations associated with the “cytotoxic burst” could be expressed as chemical activity to test the hypothesis that these concentrations would be equivalent to activity in the 0.1-0.01 range. Second, it would be useful to apply structure-based MOA classification schemes to the ToxCast chemical library and examine the agreement (or lack thereof) between chemical structure-based identification of putative baseline (MOA 1,2) toxicant and biologically-based identification of baseline toxicants as based on the cytotoxic burst analysis. Kramer (Kramer et al., 2009) compared in vitro and fathead minnow in vivo toxicity and found that not all specifically acting chemicals were poorly predicted with in vitro cytotoxicity, indicating baseline toxicity and excess toxicity are both leading to the observed toxicity. Active metabolites have been little considered in vitro and lack of consideration may lead to misclassifying chemicals in MOA classes.
Beyond the utility for discriminating predominantly baseline or non-specific toxicants (which may include certain reactive MOAs) from those with potential to interact with specific biological targets/pathways at much lower concentrations, high throughput toxicology datasets can also provide a finer resolution classification of chemicals that fall into broad Verhaar category IV or excess toxicity categories. It would be expected that solubility-based chemical activity would not necessarily be a robust predictor of potency for many of these specific modes of toxicity. An illustrative example is the case of the stereoselectivity of many enzymes and receptors. While stereoisomers would have similar solubility they can have dramatically different biological potency. In these cases structural features and/or physical/chemical properties other than those closely linked to solubility would be needed to describe the chemical space likely to interact potently with these targets. Nonetheless, there may prove to be certain targets for which chemical activity may be good predictors. These would likely be targets that are found in membranes or other lipid-rich regions of the cell and are fairly promiscuous in terms of the chemical structures with which they bind or react. A potential research exercise would be to examine correlations between chemical activity and potency of ToxCast chemicals in specific assays and identify those for which a strong relationship exists. One could then examine the localisation and function of those targets in more detail and begin to investigate whether there is a scientifically-plausible theoretical basis on which to expect that activity-based predictions would have value for predicting chemical potency against those targets. This finer resolution of MOA categorisation based on activity in various pathway-based high-throughput toxicology assays can be mapped to the concept of molecular initiating events (MIE). Whereas baseline toxicants can be expected to act similarly on a broad range of organisms, life stages, sexes, to cause overt mortality through non-specific membrane interactions, more specifically-acting toxicants may show considerable selectivity in terms of the taxa, life stages, sexes, etc. that are sensitive/susceptible to their effects. The AOP framework is intended to establish and describe the scientifically-credible links between perturbation of a particular MIE, as may be captured/assessed via a high throughput assay, and the downstream biological consequences that may be expected within specific biological domains. Divergent effects and sensitivities among these different biological domains can be represented in an AOP network, allowing this finer resolution definition of chemical mode(s) of action to be linked to relevant hazards. While chemical activity can be viewed as a useful predictive framework for linking non-selective baseline toxicity to a fairly universal outcome of acute lethality, AOPs provide the framework to link these more specific toxicities to their more specific and selective outcomes. Thus, the approaches are complementary rather than redundant in the context of an overall predictive framework for chemical safety evaluation.
The use of toxicogenomics for understanding the mechanisms underlying chemical toxicity in (eco)toxicology has become common practice. Generally, the application of transcriptomics is more routine and advanced than the application of proteomics and metabolomics in toxicological studies. In particular, use of QPCR for targeted measurement of transcript levels of genes known to be implied in the toxic mechanism of interest is now routine in many toxicological studies. In addition, measurements of the whole transcriptome, or at least larger subsets, are often performed using microarray or next generation sequencing techniques. There are two main ways in which such toxicogenomics datasets can help to identify the MOA of previously uncharacterised chemicals. 1) The MOA is inferred from direct biological interpretation of the data based on toxicological knowledge, for example through pathway analysis identifying enrichment of a receptor activated pathway. 2) A more literal application of the classification concept is to use clustering algorithms to group chemicals according to their expression profiles or signatures without necessarily knowing the functions of the genes contributing to the classification.  Although the use of toxicogenomics data for regulatory applications can be envisaged in the long term, for instance aiding in the selection of appropriate QSAR models, early attempts to apply omics-based classification strategies have been limited due to several factors. Firstly, the resolution for distinguishing among MOAs depends on intrinsic limitations of the techniques applied. More importantly, the vast number of MOAs that exist, the different levels of definition of MOAs that are being used (e.g. endocrine disruption in general versus oestrogen agonism, oestrogen antagonism, androgen agonism etc.), and the – often limited – numbers of MOAs and chemicals per MOA that are included in profiling analyses contribute to a generally low resolving power. Furthermore, the use of outgroups containing chemicals with MOAs strongly differing from those of interest (a standard practice in more traditional classification approaches such as phylogenetic analyses) has often been neglected, although using outgroups as a reference improves the interpretation of differences and similarities among chemicals. Additionally, both technical (e.g. inter-lab variability, poor standardisation of protocols) and biological (e.g. age, time, handling stress, exposure concentration) variability have complicated these efforts. Due to these complexities, there has been an ongoing debate about the potential of toxicogenomics for classification of chemicals according to their MOA, with only rare success stories in ecotoxicology.
From these limitations and experiences, it has become clear that large-scale efforts with an important bio-informatics component are needed. Useful advances have been made in other scientific fields and recently also in ecotoxicology that could aid in using toxicogenomics for classification of chemicals, with regulatory applications following in the longer term. One example is the MNI (MOA by Network Identification) approach used by Ergün et al. (2007) to study prostate cancer in a clinical setting. The authors used a large multi-cancer transcriptional expression dataset containing different sources of variation (drug treatments, cell lines, patient samples) in order to construct a co-expression network reflecting the average behaviour of genes in cancer. Subsequently, they used this network as a background to filter out genes that are specific (i.e., respond differently from what is expected based on the network) for different stages of prostate cancer. They were thus able to build a molecular classifier to distinguish between non-recurrent and recurrent primary prostate cancer, which is of high diagnostic value. Using a large database and relatively simple mathematical models, they achieved a resolving power that would not have been possible based on standard transcript expression analysis workflows. A recent example in ecotoxicology is the study of Antczak (Antczak et al., 2015) in which Daphnia magna were exposed to 26 organic chemicals to study the mechanism for basal toxicity. The mechanisms involved in narcosis, especially related to sub lethal effects, are still poorly understood, although the use of QSARs for prediction of baseline acute toxicity (i.e., mortality) has become common practice. The authors built a network integrating physicochemical features of the chemicals with affected pathways, and pathways with organismal toxicity. The results indicated a link between transcriptional changes involved in intracellular calcium mobilisation and narcosis. They validated these findings by showing that exposure to a calcium ATPase pump inhibitor was able to reproduce a large part of the differential expression signature of narcotics.
For an appropriate use of QSARs for predicting narcosis toxicity (acute and chronic, lethal and sub lethal), it is essential to determine whether chemicals act through narcosis or have (additional) specific MOA(s). In this respect, regulatory agencies are increasingly demanding biological mechanistic information to support MOA designation to justify the use of a QSAR model. Since aquatic toxicity tests using algae, invertebrates and fish embryos are considered alternative testing methods, they could be used to collect acute toxicity data on new unknown compounds without the need to use animals. Subsequently, analysis of the concordance between predicted toxicity based on acute QSAR models for narcosis and the acute experimental data may suggest that the chemical acts through narcosis. However, this MOA designation would be solely based on mortality data and would not take sub lethal effects into account. Subsequent application of acute-to-chronic ratios for prediction of chronic toxicity may not always be appropriate, since other mechanisms as well as inter-species differences in sensitivity may become more important after long-term exposure to low chemical concentrations. For these reasons, toxicogenomics data could play a role in biologically supporting the proper MOA designation needed to justify the use of QSARs. Combined with in vitro assays and in vivo measurements of sub lethal endpoints at different levels of biological organisation (biochemistry, physiology, behaviour, etc.), toxicogenomics are considered as an important source of biological support for increasing confidence in risk assessment. Given the maturing methodologies, the increasing scale of studies, and the growing bioinformatics component, a revival of the use of toxicogenomics for classifying chemicals according to MOA may be underway.
Theme 3: Quantitative adverse outcome pathways (AOP) – chemical activity and other dose metrics
The theme was introduced with a presentation on the topic of Quantitative Adverse Outcome Pathways and a presentation on Dose metrics.
Mode of action is a very broad term and does not directly refer to the underlying processes. If we take reactive chemicals or alkylating agents, for example: there are numerous more specific processes or targets that such compounds may interfere with (DNA, a specific enzyme etc.). The terminology applied within the context of AOP (Ankley et al., 2010; Russom et al., 2014; Villeneuve et al., 2014; Villeneuve and Garcia-Reyero, 2011) is more precise. According to Ankley (Ankley et al., 2010) and the OECD (OECD, 2013): “an Adverse Outcome Pathway (AOP) is an analytical construct that describes a sequential chain of causally linked events at different levels of biological organisation that lead to an adverse health or ecotoxicological effect (see Figure 4.2.3). AOPs are the central element of a toxicological knowledge framework being built to support chemical risk assessment based on mechanistic reasoning”.
The AOP concept is useful to interpret and study the effects of chemicals in a hazard assessment. In hazard characterisation and risk assessment we need also quantitative information about dose-response relationships as well as toxicokinetic data, including uptake, internal distribution and biotransformation. For a more complete analysis of the whole process from the dose to the overall effect, information about rate or equilibrium constants of the underlying intermediate steps is needed as well. We could refer here to quantitative AOP analysis. An example of a quantitative AOP was recently published by Villeneuve (Villeneuve et al., 2015). Such complete analyses are also presented in so-called Toxicokinetic/Toxicodynamic models (TKTD). A nice example of such a more complete quantitative AOP or TK-TD model is a study from Kretschmann (Kretschmann et al., 2012) for the effects of organophosphates. The TK-TD model used includes parameters for all underlying processes such as  uptake, elimination, biotransformation and interaction with the target.

Figure 4.2.3. Different steps in a quantitative adverse outcome pathway (AOP). Modified from (Ankley et al., 2010; OECD, 2013).

capture-decran-2016-11-24-a-08-15-30

Because they were intended to link molecular-level biological perturbations elicited by chemicals to hazards considered relevant to risk assessment and regulatory decision-making (e.g., impacts on human health or survival, growth, or reproduction of wildlife populations) AOPs are bioactivity-based, not chemical specific. They are employed to generalise and predict the pattern(s) of effects any chemical that perturbs a particular target/MIE (with sufficient potency and duration) could be expected to produce. The challenge to applying AOPs in predictive toxicology then is understanding what constitutes sufficient potency and duration of interaction for different chemicals. Thus, while the AOP itself does not explicitly consider chemical-specific properties, application of AOPs must consider those properties in order to assess how much of a delivered chemical (in a laboratory/bioassay context) or ambient concentration can actually reach the target site of action and for how long.
In the prediction of the overall organism responses from the applied dose, a number of modelling approaches are relevant and these include:
Toxicokinetic and Toxicodynamic (TK/TD)
Physiologically based Toxicokinetic (PBPK)
Reverse Dosimetry
Dose-Response and Biologically Based Dose-Response (DR)
These same models are also important in the prediction of in vivo effects from in vitro data and there is much similarity in quantitative AOP and quantitative in vitro-in vivo extrapolation (QIVIVE). There are several examples, also in high throughput screening in the ToxCast program, where these models are applied in the extrapolation of in vitro to in vivo effects. In these examples, concentrations in blood are set equal to a concentration in an in vitro assay. The in vivo toxicity is then predicted from in vitro data using reverse dosimetry. Similar studies have been published by Louisse et al. (Louisse et al., 2010). Bioavailability issues and kinetic analyses of partitioning processes are essential in these approaches and here chemical activity may play an important role as well (Wetmore et al., 2012).
The dose of a compound is an essential element in quantitative AOP, TKTD modelling and in quantitative in vitro-in vivo extrapolations (see figure 4.2.3). A number of dose metrics can be applied, including nominal and total concentrations, as well as freely dissolved and internal concentrations (Escher and Hermens, 2004; Groothuis et al., 2015). Also chemical activity is a powerful metric to express the dose. In particular in multi-compartment systems, chemical activity has its strengths. As indicated by Reichenberg (Reichenberg and Mayer, 2006), “the chemical activity of a substance – as well as its fugacity – is by definition the same throughout a system that has reached thermodynamic equilibrium. In that case, the measured chemical activity in one phase applies to the other phases as well. This is true regardless of the degree of heterogeneity, the number and diversity of sorptive sites, and the organic matter quality”. Because of this, chemical activity is very useful in the interpretation of toxicity data from in vitro assays, where total concentrations leading to an effect may be affected by protein binding, while chemical activity and freely dissolved concentration should be independent of that (Armitage et al., 2014; Kramer et al., 2010).
For some specific compounds or MOA’s (or AOP’s), effects are related to a time integrated dose and this will particularly be the case when the MIE is based on an irreversible mechanism. In those cases, a time integrated exposure, such as an area under the curve (AUC) or target occupation is a more suitable dose metric (Gülden et al., 2010; Legierse et al., 1999). Also in TKTD modelling, the factor time is inherently included in the analyses and modelling of effect data (Ashauer and Brown, 2008; Ashauer et al., 2015).
To the extent that chemical activity or other exposure metrics could improve the accuracy/precision with which the dose and duration of chemical exposure at the target site of the MIE are described, their application could be expected to improve and enhance the utility of quantitative AOPs for predictive toxicology. These approaches could provide more precise definition of point of departure, as a generalisable description of an equivalent dose and duration of chemical exposure, needed to produce the effects observed for a reference compound. In the context of weight of evidence evaluation for AOPs, the use of dose-metrics that can more readily account for significant differences in study design, would aid the evaluation of whether apparent deviations from dose-response concordance among different key events is simply an explainable result of disparate study designs or whether it actually represents grounds for reducing confidence in the causal relationships depicted in the AOP and/or outright rejection of that AOP. The bottom-line relative to quantitative application of AOPs in predictive risk assessment is that higher precision dosimetry estimates, whether they be derived from chemical activity, free concentration estimates, IVIVE, AUC or other approaches, should aid and strengthen the quantitative application of AOP knowledge for predictive risk assessment applications.
At present, these concepts remain largely unexplored. However, the first case studies in the quantitative application of AOPs are underway (Villeneuve et al., 2014). Initial sets of toxicodynamic model predictions which capture key feedback and compensatory mechanisms known to operate along the reproductive endocrine axis have been generated using the simple steady state assumption that chemical concentration in water is equivalent to the free concentration in plasma and that relative potency at the target site can be defined simply based on nominal concentration in an in vitro assay. Should those initial case studies fail to produce reasonable estimates of in vivo biological response, a logical next step would be to redo the model simulations using more sophisticated approaches to predict the internal dose in the organism from the external concentration in the exposure media and/or apply more refined in vitro dose-metrics in an attempt to provide a more accurate characterisation of potency that more accurately considers how much chemical actually reached the target site within the bioassay. These are near-term case studies that could provide insights into the extent to which toxicokinetic considerations and alternative dose-metrics could improve the accuracy of quantitative AOP-based predictions.
Summary of resulting discussions
The participants discussed the following three themes during three breakout sessions:
Theme 1: Data for chemical activity (beyond baseline toxicity)
Theme 2: Modes of action (MOA) and classification
Theme 3: (Quantitative) adverse outcome pathways (AOP) – chemical activity and other dose metrics.
Theme 1: Data for chemical activity (beyond baseline toxicity)
The following questions were discussed during the workshop:
Can we identify suitable data sets with effect concentrations from which chemical activities can be derived, or are there publications that report chemical activities (for chemicals beyond non-polar baseline toxicity)?
Can the chemical activity approach be helpful in the identification of compounds with modes of action other than non-polar baseline toxicity?
Can the chemical activity approach be useful to interpret differences in effect concentrations of MOA1 and MOA2 chemicals?
Are there enough data to estimate the range in chemical activity of compounds with a certain mode of action? What are the advantages of applying a chemical activity concept to compounds with specific modes of action?
Can we link the chemical activity approach to the TTC concept (threshold of toxicological concern)?
Can we apply chemical activities in the interpretation and understanding of effects of complex mixtures with baseline toxicity only and multiple MOAs; as well as of individual compounds with “multiple” modes of action?

Can we identify suitable data sets with effect concentrations from which chemical activities can be derived, or are there publications that report chemical activities (for chemicals beyond non-polar baseline toxicity)?
Only a few studies interpret acute toxicological effect data using the chemical activity concept (Mackay et al., 2014; McCarty et al., 2013; Mayer and Reichenberg, 2006). However, several data sets in the open literature report effect concentrations that can be applied to derive chemical activity using experimental or estimated data for subcooled liquid solubility. Many of these data sets include non-polar organic compounds (MOA 1) (McGrath and Di Toro, 2009), while other sets report data for MOA 1, 2, 3 and 4 chemicals (Barron et al., 2015; Russom et al., 1997; Verhaar et al., 1992). Also the European Chemicals Agency (ECHA) has developed toxicity databases. Other data sets are known to exist but need to be made available.
The group recommended that for calculating reliable chemical activity data, high quality toxicity data as well as reliable experimental or estimated values for solubility (or subcooled liquid solubility) are needed. More detailed information about estimation of solubility is presented in WG3.
Can the chemical activity approach be helpful in the identification of compounds with modes of action other than non-polar baseline toxicity?
As discussed in WG1, chemical activities related to acute effects on survival for compounds that act via MOA 1 are in a rather narrow range (between 0.01 and 0.1). There were indications from a recently published ECETOC report (ECETOC, 2013) that chronic activities could also be interpreted using the chemical activity concept, based on data obtained from regulatory studies, and that, as expected, the chemical activities derived were lower than those for chemical activities calculated based on acute toxicity data. However, it is notable that the accuracy of the relationship was hampered by data quality and transparency related to methods used to allocate substances as MoA 1 or MoA 2. It was concluded that if for a given chemical the calculated chemical activity is lower than 0.001, especially at lower log Kow (<4), that there is a high likelihood that the chemical has a more specific MOA, for instance MoA 4 in the Verhaar system or is a more reactive compound, i.e. MoA 3. Consequently, based on the analysis reported in the ECETOC report, it can be argued that the chemical activity concept is useful in differentiating chemicals as being either baseline toxicants or as having the potential to illicit excess toxicity. As log Kow increases above 4, the difference in chemical activities between baseline and excess toxicity is less obvious.

Can the chemical activity approach be useful to interpret differences in effect concentrations of MOA1 and MOA2 chemicals?
Initial analyses that are reported in the ECETOC report (ECETOC, 2013) have shown that chemical activities of compounds that act via so-called polar narcosis (MOA 2) are lower than the range of 0.01-01. The group agreed that a more detailed analysis of the data is needed to improve the overall understanding of the difference between MOA 1 and MOA 2. It is not clear why chemical activity for MOA 2 compounds is lower than the range 0.01-0.1 and how this relates to the observation that internal membrane concentrations for MOA 1 and MOA 2 chemicals are very similar.
The group agreed that this topic will be explored in more detail in a follow-up manuscript planned to be prepared following publication of this ECETOC workshop report.
Are there enough data to estimate the range in chemical activity of compounds with a certain mode of action? What are the advantages of applying a chemical activity concept to compounds with specific modes of action?
Data directly assessing the chemical activity of chemicals with modes of action beyond MoA 1 and 2 are currently not readily available. Nonetheless, WG2 participants acknowledged that there could be advantages in applying the chemical activity concept for chemicals with specific modes of action associated with them. Expressing toxicity data using chemical activities will very easily show an “enhanced” toxicity as compared to the 0.01-0.1 range that is valid for MOA 1. Applying the chemical activity beyond MoA 1 and 2, however remains a challenging area, and is revisited in the responses to other questions posed to the group. A more detailed analysis of MOA 3 and 4 chemicals is needed to arrive at clear conclusions about the applicability of the activity concept to compounds from MOA 3 and 4 classes. Such an analysis will be performed and results will be presented in a manuscript planned to be prepared following publication of this ECETOC workshop report. It was recognised that chemicals may have multiple MOA’s, especially with respect to chronic exposure, and that for acute exposures baseline toxicity may present itself prior to a specific MOA, particularly in instances where the exposure is relatively short, and the concentration in the cell membrane is much higher than the concentration at the target site. This will occur in particular for the more hydrophobic chemicals.
Can we link the chemical activity approach to the TTC concept (threshold of toxicological concern)?
During the meeting, a short presentation was given regarding an application of the chemical activity concept in relation to the concept of threshold of toxicological concern (TTC), in particular for chemicals whose only mode of action is baseline toxicity. Using aqueous concentrations, QSAR models and sensitivity distribution analyses have been developed in the literature to predict no-effect concentrations at the ecosystem level. In this approach, hazardous concentration for 5 % of the species (HC5) has been proposed as a no-effect level for MOA 1 compounds (van Leeuwen et al., 1992). An initial analyses of HC5 values shows that recalculating HC5 for baseline toxicants (i.e. MOA 1) into chemical activities leads to a constant chemical activity for the NOEC(ecosystem). This approach is useful within a TTC concept. At present, the TTC for the aquatic environment for MOA I is based on aqueous concentrations and is determined by the most toxic (most hydrophobic) compound (De Wolf et al., 2005). ). A TTC based on chemical activity adds value in that the approach acknowledges the effect of hydrophobicity.  The group regarded this as an interesting approach and suggested to work this out in more detail.
The approach will be presented in more detail in an upcoming manuscript planned to be prepared following publication of this ECETOC workshop report.
It must be emphasised that such an approach is useful for MOA 1, and the applicability to other MOA’s is not feasible.
Can we apply chemical activities in the interpretation and understanding of effects of complex mixtures with baseline toxicity only and multiple MOAs; as well as of individual compounds with “multiple” modes of action?
There was a clear consensus that the chemical activity concept is very useful in interpreting and estimating the effects of complex mixtures that are solely comprised of MOA 1 and 2 chemicals, since chemical activities can simply be added (see also discussion from WG1). WG2 noted that the applicability of the chemical activity concept to mixtures consisting of chemicals that act as MOA 3 and 4 is currently outside the applicability domain and not feasible at the moment. An exception is that in mixtures that contain compounds from MOA 1 and other modes of action, the chemical activity approach is useful to predict the contribution of all compounds (including the contributions from MOA 3 and 4) to the baseline toxicity.
Theme 2: Modes of action (MoA) and classification
The following questions were discussed during the workshop:
Which major modes of action (MOA) are relevant in analysing ecotox data?
How can we classify compounds according to their mode of action? What kinds of classification systems are available and how reliable are they? Do we need more precise and specific classifications?
New approaches in classification are using “omics data”. Are there many examples and how do they relate to more traditional classification schemes?
How can chemistry based approaches impact on definition and identification of MOA?
Are the current classifications schemes for acute MOA appropriate for chronic endpoints?
Which major modes of action (MOA) are relevant in analysing ecotox data?
Examples of major modes of action that are well documented in ecotoxicological research are: baseline toxicity (non-polar, polar, ester, amine); uncouplers; electrophilic (can be sub-categorised); specific: Central Nervous System (CNS), Acetylcholine-esterase inhibitors, other neurotoxicity, oestrogenicity, photosynthesis inhibition.

How can we classify compounds according to their mode of action? What kinds of classification systems are available and how reliable are they?    4.  How can chemistry based approaches impact on definition and identification of MOA?
Verhaar and Russom are coded classification schemes and readily available but should be used with appropriate caveats, in particular attention should be paid to check the chemical applicability domain for untested compounds. It was recognised that Verhaar classes offer a simple approach to group MoAs and due to its simplicity, that this method could be used as a basis for a more complete scheme with some modifications:
Subcategorisation of the major classes could be valuable (for example different classes for reactive compounds etc.).
Updating of chemical information is needed.
For specifically acting and reactive chemicals, a secondary consideration would be what is the target of action and can chemical activity be applied in a useful and relevant manner.
It was noted that chemistry based approaches can be a combination of alerts and physico-chemical properties. It was also recognised by the group that chemical reactivity data (glutathione (GSH) depletion, Direct Peptide Reactivity Assay (DPRA), although optimised for skin sensitisation) as well as information from other endpoints (skin sensitisation) could be helpful but need to be supported experimentally. The group concluded that biologically based HTS assays (high throughput screening) can be useful in identifying MOAs. It was noted that while the current assumption is that the “cytotoxic burst” phenomenon is an in vitro analog to baseline toxicity, this assumption has not been explored experimentally. This represents an interesting topic for research moving forward.
The group recommended to create a separate activity to amend existing classification schemes for non-baseline toxicity chemicals and to consider the inclusion of esters (perhaps, as in Russom scheme, to include as a sub-class of non-polar narcosis). Such an activity should also include a discussion on improvement of classification systems of MOA, including chemistry based approaches (see above), high throughput screening and omics approaches (see question 3).
New approaches in classification are using “omics data”. Are there many examples and how do they relate to more traditional classification schemes?
The group discussed new approaches in classification based on “omics data”. There are two main ways in which omics such as toxicogenomics datasets can help to identify the MoA of previously uncharacterised chemicals. 1) The MoA is inferred from direct biological interpretation of the data, based on toxicological knowledge. 2) A more literal application of the classification concept is to use clustering algorithms to group chemicals according to their expression profiles. Although regulatory agencies are increasingly demanding biological mechanistic information to support MoA designation to justify the use of QSAR models, regulatory application of toxicogenomics data is at this point limited, largely due to high technical and biological variability. Large scale efforts based on bio-informatics are needed to overcome the current limitations, and it is further recommended that efforts should be focused on increasing the applicability of omics for identifying MoAs of untested chemicals (including assigning chemicals as either MoA 1 or 2) and specifically on sub-lethal effects and species differences.
Are the current classifications schemes for acute MOA appropriate for chronic endpoints?
The group also discussed the applicability of the chemical activity concept to chronic toxicity data. The following questions were raised:
For acute baseline toxicants does an ACR of 10:1 hold, and if so can chemical activity be applied to determine chronic endpoints?
Do specifically acting chemicals have a higher, or more variable ACR?
How can we accommodate species differences in ACR and effects?
Time was too short to discuss these questions in detail, although some positive evidence was provided supporting the last point. It was noted that more chemical activity research should be performed on chronic effects. This aspect will be elaborated upon in the manuscript which is planned as a product of the workshop.
Theme 3: (Quantitative) adverse outcome pathways (AOP) – chemical activity and other dose metrics
The following questions were discussed during the workshop:
General
Are the concepts of adverse outcome pathway (AOP) and molecular initiating event (MIE) useful in classification of chemicals (which could lead to improved ability to study relationships with chemical activity)?
Can the chemical activity concept provide a complementary approach towards an improved understanding of an adverse outcome pathway (AOP)?
Has chemical activity the potential to link exposure with the molecular initiating event (MIE) in a single metric?
Specific
Is the chemical activity concept useful for all kinds of target sites and target site environments?
Can chemical activity be applied as valuable input parameter in a quantitative AOP, a TKTD model and in in vitro-in vivo extrapolations?
Are there alternative promising dose metrics or parameters for the evaluation of the effects of reactive compounds or of compounds with specific modes of action?

General
Are the concepts of adverse outcome pathway (AOP) and molecular initiating event (MIE) useful in classification of chemicals (which could lead to improved ability to study relationships with chemical activity)?
Can the chemical activity concept provide a complementary approach towards an improved understanding of an adverse outcome pathway (AOP)?
Has chemical activity the potential to link exposure with the molecular initiating event (MIE) in a single metric?
A brief discussion was held on the linkage between MOA classification and the AOP (adverse outcome pathway) construct. It was agreed that AOP can support MOAs. AOP may help identify MIEs which could form the basis of classification schemes, AOP may help understand interspecies differences and chronic toxicity. It was also noted that few reliable AOPs exist at this time and that more research on AOPs is required.
The group agreed that chemical activity may be applied as a valuable input parameter in a quantitative AOP as it generally improves quantifying exposure and may also give information about the bioavailability of a compound. Chemical activities thereby (i) offer the potential to better define “points of departure” from where the organism can handle the perturbation to where the perturbation is fatal to the organism, (ii) offer an improved ability to use diverse studies to evaluate overall concordance of the relationships depicted in the AOP and (iii) represents a critical link of environmental concentrations to potency.
The group does not support the suggestion that chemical activity can be linked to the molecular initiating event (MIE) in one single metric. Only for baseline toxicity, chemical activity represent a single metric because there is no specific target molecule and the potency for all baseline toxicity compounds is similar. For other modes of action, effects are related to a number of processes (uptake, interaction with a target or receptor) and this cannot be related to one single metric.

Specific
Is the chemical activity concept useful for all kinds of target sites and target site environments?
Can chemical activity be applied as valuable input parameter in a quantitative AOP, a TKTD model and in in vitro-in vivo extrapolations?
Are there alternative promising dose metrics or parameters for the evaluation of the effects of reactive compounds or of compounds with specific modes of action?
Chemical activities, free concentration (measured or modelled approaches) and internal concentration in an organ or cell are promising dose metrics to describe the bioavailable fraction. For in vitro-in vivo extrapolations, chemical activity forces consideration of the bioavailable fraction.
Dose metrics such as concentrations or chemical activities are useful in the analysis of effects of chemicals with a key event that represents a reversible interaction. For irreversible interactions with the target site, as is often the case for MOA 3 compounds, time is an important variable as well because “damage” is cumulative for irreversible interactions. In those cases, time integrated exposure measures such as area under the curve (AUC) could be an interesting dose metric (Gülden et al., 2010).
It was suggested that TKTD modelling is an interesting approach to analyse or predict effects of MOA 3 and MOA 4 chemicals as it includes parameters for all underlying processes including, uptake, elimination, biotransformation, interaction with the target, etc. It may also prove useful to classify compounds into distinguishable MOA subgroups.
Conclusions
AOPs and biologically based assays will be useful to differentiate further MOA classes 3 and 4 which in turn open these roads for chemical activity to improve bioavailability metrics;
Verhaar classes offer a simple approach to differentiate between expected specific, reactive and MOA 1 and 2 for acute exposure with some recommendations (see below) if the chemical domain is covered;
In specifically acting and reactive classes a secondary consideration would be what is the target of action and can chemical activity be applied in a useful relevant manner;
Chemical activity seems best applied to MOA 1 and 2 with significant research effort necessary to see to what extent the application can be broadened to MOA 3 and 4;
Initial analyses of HC5 values have shown that recalculating HC5 into chemical activities leads to constant “threshold of toxicological concern”. This concept is interesting and could be useful in risk assessment;
New approaches in classification based on “omics data” are promising. The variability in omics data is presently high and the regulatory applicability is still limited. Large scale efforts with an important bio-informatics component are needed to overcome the current limitations and it is further recommended that efforts should be focused on increasing the applicability of omics for identifying MOAs of untested chemicals (including assigning MOA 1 chemicals) and specifically on sub-lethal effects and species differences.