Discussion of overall results and learnings

In this project, we explored approaches and developed recommendations for formation of (sub-) groups of substances with the ultimate aim of filling data gaps by read-across. Some of the case examples did not aim at data gap filling, but their category formation approaches were still found to provide valuable learnings generally applicable for forming (sub-)groups with this purpose.

The ECHA Read-across Assessment Framework is tailored for the regulatory use of read-across under the REACH Regulation (ECHA 2017). It, therefore, ultimately is designed for regulatory evaluation of applied read-across cases. Examples of use could be for assessing the regulatory needs of a group of registered substances, assessing registration dossier compliance where read-across is used to adapt the standard information requirements or for assessing read-across used in restriction or SVHC proposals.

The OECD GD 194 on Grouping of Chemicals aims to encompass different possibilities and interpretations of predicting properties of chemicals if adequate information is not available. The guidance describes approaches known to have been accepted or developed by regulators and other approaches, used e.g. by registrants under voluntary programs.

In general, all case examples followed the overall grouping principles as described in both the OECD GD 194 and the ECHA RAAF. However, both the scope of case examples, the number of steps they fulfil in accordance with the recommendations and the level of information provided at each step differed widely between the case examples, as the objectives of the case examples were different. For development of (sub-)groups with the aim of data gap filling, it is recommended to use the ECHA RAAF or the OECD GD 194 to guide the process.

Optimally, grouping exercises aiming at data gap filling are initiated with broad groups of substances and well-argued inclusion and exclusion criteria to have as wide a structural domain and as many potential source substances as possible. Exclusion of substances based on arguments like use profiles or lack of CLP classification or REACH registration is not recommended since substances may be valuable source substances for data gap filling even if they are e.g. not registered under REACH. Thus, inclusion of substances in groups should not depend on their use or current regulatory status, as their use may change over time. Highly regulated substances are, furthermore, often data rich and this data may prove valuable in forming the category and in read-across to fill data gaps for other category members.

The case examples show that the decision about which substances to include in the beginning of a grouping exercise is often not well argued/described and in most cases therefore judged to be non-exhaustive. Along the same lines, the reasons for exclusion of substances from a group are in some cases not explained or justified. It is therefore not transparent and does not follow the recommendations. For example, in some cases, substances are excluded because they are not registered under REACH or classified according to CLP, or they have different use profiles than the rest of the substances. Even though some substances are not relevant to the specific regulatory purpose, e.g. if they are currently not registered under REACH, are strictly regulated already, or have another use profile than the rest of the substances, they can still be valuable source substances or they may in the future become important target substances e.g. in “regrettable substitutions”. The broader the grouping, the more solid basis for subsequent (sub)category formation.

Substance characterization is of high importance and found in the case examples to have room for improvement. Often, reporting of impurities and in some cases also chemical structures and identifiers as CAS RN, are lacking. Mono-constituent substances can have significant impurity profiles as one main constituent has to be present to at least 80%, i.e. up to 20% may be impurities (ECHA 2017). The lack of information on impurities and in some cases chemical structures and identifiers as CAS RN may be a challenge to subsequent data gap filling by read-across.

Multi-constituent substances and UVCBs have more complex compositions and represent a specific challenge which is not covered by this report. Thorough substance characterization is an important element for assuring a solid basis for category formation. The OECD GD 194 has a specific chapter targeting guidance on UVCBs and ECHA has a RAAF specifically targeting UVCBs and ECHA has developed a specific read-across assessment framework for multi-constituent substances, but more detailed consideration of this is out of scope of this report.

Preferably, sub-grouping aimed at data gap filling by read-across should be based on structural similarity considerations relevant to the endpoint under scrutiny.

Ultimately, knowledge is available that substantiates the importance of specific molecular moieties for induction of specific mechanisms that can lead to toxicological effects. The case example on brominated flame retardants demonstrates how (Q)SAR predictions can be used to search for similar effect profiles across (sub-)groups of structurally similar substances and to expand (sub-)groups to include additional relevant substances. Different sub-groups may be relevant to form depending on the endpoint(s) under evaluation and the molecular sub-structures of relevance for the specific underlying mechanisms of action.

Care should be taken when basing (sub-)groups on in silico and/or in vitro data. If in silico and/or in vitro information is used as the basis for sub-grouping, ADME/metabolism of the individual substances is important to consider when extrapolating from in vitro/in silico to in vivo. The resulting (sub-)grouping may be different if in vivo data or reliable ADME data are also included/available. For example, some substances may activate the estrogen receptor in vitro but not induce a uterotrophic effect in vivo, or vice versa, with one possible explanation being metabolic activation or de-activation of the parent compound. The extrapolation between in vitro and in vivo is thus important to keep in mind in the formation of (sub-)groups.

Different substructures of a molecule may be connected to different mechanisms of action as suggested by Kitamura (2005). When developing sub-groups aimed at data gap filling by read-across with focus on ED properties, it is recommended to consider that different molecular moieties may be relevant to different mechanisms and thereby modes of action and adverse effects. Some overlap may occur, but this should ideally be analysed and described. The complexity increases when considering that many endocrine disruptors act through more than one mechanism/mode of action and that several different mechanisms/ modes of actions can lead to the same adverse effects. In a grouping context, this means that different sub-groups may be relevant to form, depending on the mechanism/mode of action in focus. Special care is therefore recommended in (sub-)grouping of endocrine disruptors, taking this complexity into consideration.

EDs may act through many different mechanisms and modes of action. Depending on the substances in question, a broad range of mechanisms of action, which may possibly be connected to different molecular moieties should therefore ideally be considered as important for (sub-)grouping.

Comparative studies are valuable and could aid to provide the mechanistic understanding and foundation for development of (sub)groups of EDs, including bisphenols. (Q)SAR predictions and molecular docking studies (modelling of preferred orientation of one molecule to another when they form a complex) could be used to substantiate the hypothesis that specific molecular moieties are linked to specific mechanisms or modes of action.

The case examples show that valuable source substances for data gap filling by read-across may be identified among substances which are under regulatory scrutiny (e.g. biocides, plant protection products and substances with proposals for SVHC identification or harmonized classification) or registered under REACH in higher tonnage bands (>100 tpa) since they are generally data rich.

For grouping of endocrine disruptors aimed at data gap filling by read-across, valuable source substances may be identified among substances which are already identified as endocrine disruptors under REACH, BPR or PPPR and in future, CLP. Including substances already identified as EDs (alternatively under regulatory scrutiny due to other effects or registered in high tonnage bands under REACH) in grouping exercises as source substances may thus be of high regulatory relevance.

The isobutylparaben case example demonstrated that similar structures combined with ADME, in silico, in vitro (and if available in vivo) data on endocrine activity can be used as a basis for data gap filling by read-across of adversity info for successful identification of endocrine disruptors.

Development of a comprehensive data matrix with in silico, in vitro and in vivo data is an important tool, which can provide a robust basis for sub-grouping and data gap filling by read-across. Such a data matrix is developed in some, but not in all cases, depending on their specific objectives. Data availability may determine the subsequent formation of sub-groups.

If reliable mechanistic (in silico, in vitro or in vivo) information is available, read- across from group members with additional information on relevant adverse effects could be applied, as demonstrated in the isobutylparaben case. Considerations about ADME should always be included to the extent possible, e.g. by including ADME and metabolism profilers in the in silico modelling. Constructing a data matrix containing both mechanistic, adversity and ADME information is highly recommended for both analogue and category approaches.

Non-monotonic dose response curves, low dose effects, different effect profiles depending on exposure windows and times of investigation and several modes of action with different ones being activated at different dose levels are some of the elements to consider when assessing the hazards and risks of endocrine disruptors. As the substances have complex hazard profiles themselves, creating a data matrix in a grouping of endocrine disruptors must be given special attention.

The use of (Q)SAR predictions are recommended to guide, challenge, support and revise groupings and sub-groupings aimed at data gap filling by read-across. (Q)SARs can provide predictions for all potential group members and thereby help to guide already when developing the read-across hypotheses. (Q)SAR predictions may also aid in identifying molecular moieties of relevance for the source and target substances. I.e., (Q)SARs can help point out molecular sub-structures and other properties of relevance for specific mechanisms/modes of action (sub-)groups of substances. (Q)SARs can also be used to identify additional relevant substances to add to the group based on structural similarity of specific relevance for important molecular moieties.

A perspective on the use of (Q)SARs is that today, many ED-related endpoints are modelled in a binary fashion, i.e. not predicting potency, however it may be a priority in the future to develop (Q)SARs for quantitative predictions. Such modelling may for example be based on results of large screening projects such as ToxCast and Tox21, where many substances have been tested by the same test procedure in the same laboratory.

Examples of identifying and regulating endocrine disruptors in groups are starting to emerge. The work conducted by ECHA to assess the regulatory needs of groups of chemical substances and publish the findings, is an important first step in future groupings of chemicals, including for regulatory purposes. In this report, it is exemplified by the bisphenol ARN. The data base of substances registered under REACH is a valuable tool for creating ‘pre-assessments’ of groupings as it is possible to identify substances for which relevant experimental data is available and use these substances as source chemicals to form groups. (Q)SARs can help provide the basis for including relevant structurally similar substances from a much larger data pool than the REACH registration database.

Adapting the data requirements under the data-generating regulations such as the REACH Regulation, the Plant Protection Products Regulations and the Biocidal Products Regulation could support identification of endocrine disruptors through grouping and read-across. Including an in vitro test battery for endocrine activity at all tonnage levels under the REACH regulation would improve the basis for developing groups of endocrine disruptors based on structural similarity and mechanistic information, and subsequently fill data gaps on adversity by use of read-across from high tonnage level substances with relevant in vivo adversity information available.

When the common data platform on chemicals under the Commissions ‘one substance, one assessment’ legal initiative becomes operational, it will further enable use of chemicals data across legislative sectors, which is expected to increase information about effects of endocrine disruptors applicable for use in groupings and read-across exercises.

Similar adversity can be induced through different mechanisms in the same adverse outcome pathway. Several publications have shown dose additivity in the observed adverse effects of endocrine disruptors after combined exposure to different endocrine disruptors with both similar and dissimilar modes of action (Hass et al., 2012; Kortenkamp et al., 2009; Kortenkamp et al., 2012),

If it, within a grouping assessment, is established that members of the category act by different mechanisms but lead to the same adverse effect, instead of separating them, these sub-groups of endocrine disruptors with a Mode of Action (MoA) via e.g. the estrogenic (E), androgenic (A) and steroidogenic (S) modalities should be considered together in a combined exposure assessment to assess the cumulative risks from these substances and restrict them, as appropriate. The regulatory relevance of cumulative risk assessment persists independently of advances to also adopt harmonized classifications of groups of endocrine disruptors.

In 2014, EFSA established cumulative assessment groups (CAGs) of pesticides on the basis of their toxicological profile, including for effects on the thyroid. It was concluded that the developed grouping methodology can be applied even when the underlying biochemical events mediating the effects are not understood (EFSA 2014). Grouping in CAGs had four levels, reflecting an increasing amount of adversity and mechanistic data. Criteria for CAG levels 1–4 were: 1: toxicological target organ, followed by 2: common specific phenomenological effect on the target organ, 3: common mode of action and 4: common mechanism of action. In the assessment of more than 220 pesticide active substances it was evident that CAGs generally could not be refined beyond level 2 as sufficient information on mode- or mechanism of action was available only for a limited number of chemicals. The report acknowledged that an agreed inventory of mode of actions, as well as a defined set of criteria for ways of characterizing or predicting modes of action in data-poor situations was missing (EFSA 2014). In 2019 EFSA further developed and updated CAGs for two specific effects on the thyroid (EFSA 2019). Work to establish adverse outcome pathways has since accelerated and is prioritized in the scientific community, the OECD and in the EU PARC and research and innovation funding programs such as Horizon 2020 and Horizon Europe.

Forming cumulative assessment groups (CAGs) for endocrine disruptors could be valuable to explore. While doing this, the CAGs could in theory be broadened to include more substances than only those with available adversity information. This could be done by first using each substance included in a preliminary CAG to search for structurally similar substances, secondly develop a data matrix to identify substances for which mechanistic data is available to substantiate data gap filling of adversity info by read-across and finally include these additional substances in the CAG.

In the PFAS restriction proposal, currently under evaluation in ECHA, a common concern for the group of up to 10,000 PFAS is very high persistence due to the strong carbon-fluorine bond (ECHA 2023b). All PFAS are either persistent themselves or degrade to other persistent PFAS, making the carbon-fluorine bond the structural similarity for the group. In the future, specific molecular moieties (like the carbon-fluorine bond for PFAS) may be identified as decisive for ED activity/effects, either within specific groups of substances or more generally applicable. Comparative studies like Kitamura (2005) as well as docking analyses and (Q)SAR modelling are valuable tools for exploring this.

Current practice on cumulative assessment groups and scientific knowledge about endocrine disruptors point to the possibility to develop adverse outcome pathways to be used, among others, for identification of endocrine disruptors in the future. Until then, the review of the case examples shows some general recommendations for future groupings on endocrine disruptors with the aim of data gap filling by read-across.