Towards an Animal-Free Human Health Assessment: Starting from the Current Regulatory Needs

Societal concern for animal welfare and scientific concerns about the predictive power of animal models for the human situation are driving forces for the development of animal-free approaches for the safety testing of chemicals. A paradigm shift towards an assessment of human health risks that is fully based on animal-free approaches is not foreseen within the next decades. To accelerate the use of animal-free innovations (in vitro experiments, in silico methods, etc.) in the EU, it has multiple advantages to simultaneously work towards a new risk assessment paradigm and to aim the development of animal-free methods at better meeting current regulatory needs. To achieve this, a multi-stakeholder collaboration is needed already in the development phase, where regulators can inform on the regulatory needs and the criteria for acceptance. As a first step, the present paper discusses what information is needed within the context of four areas of chemical safety assessment in the EU: 1) classification, labelling and packaging, 2) the derivation of health-based guidance values and product limits, 3) risk assessments of exposure situations of concern and 4) addressing specific topics of societal concern. Further agreements on the level of detail and uncertainty, robustness, predictive value, reproducibility and validation are a prerequisite to develop tools that can be trusted and that will be legally binding. This is an Open Access article distributed under the terms of the Creative Commons Attribution 4.0 International license (http://creativecommons.org/licenses/by/4.0/), which permits unrestricted use, distribution and reproduction in any medium, provided the original work is appropriately cited.


Introduction
In everyday life, humans come into contact with a variety of chemicals in a wide range of uses and applications, including food additives, food contact materials, industrial chemicals, plant protection products, biocides and cosmetics. For many decades, the safety of human exposures has been systematically evaluated by studying potential intrinsic adverse health effects in animal experiments and by extrapolating this information to human exposure scenarios and/or by using the intrinsic hazardous properties in risk management measures. However, already in 1959 Russell and Burch drew attention to the ethical aspects of animal experiments and introduced the concept of the three R's, replacement, reduction and refinement of animal experiments (Russell and Burch, 1959). Although initially slowly acknowledged, the societal concern for animal welfare has significantly grown in the past few decades. At the same time, there is an increasing concern about the use of animal experiments for scientific reasons, i.e., whether the animal species used most often are an appropriate model to predict adverse health effects in humans (Schiffelers et al., 2007). Multiple reasons underlie this concern. The biology and physiology of rats and mice, the animal species used mostly for testing, may differ significantly from that of humans. In addition, human endpoints, e.g., a general ill-feeling such as headache, psycho-somatic and learning effects are not covered by the endpoints generally investigated in animal experiments. A third scientific reason also meets present societal concerns and more or less follows from the previous reason. It relates to the question "Can chemicals in humans' daily environment be responsible for human diseases such as Alzheimer's, Parkinson, obesity and Attention Deficit Hyperactivity Disorder (ADHD), some of which are accompanied by high costs for human society?". The animal experiments as required and traditionally performed in accordance with the OECD Guidelines for Testing of Chemicals or the EU Test Methods are, for various reasons, only designed to cover certain predetermined endpoints, and are neither able nor developed to cover all potential relevant human diseases. These experiments are very resource intensive in terms of costs and man-hours which limits for instance the number of animals and thereby the statistical power to detect adverse effects. Further, the suitability of the animal species tested as a predictive model for potential human health effects is seldom verified beforehand. 1 In addition, development of innovative nanomaterials and the transition towards a circular economy (e.g. green chemistry or non-toxic environments) require adequate cheap and rapid methods for toxicity screening. To enable industries to implement safe-and sustainable-by-design approaches, new methodologies are needed to assess safety and sustainability throughout the entire life-cycle but that do not require extensive data gathering and toxicity testing.
For these reasons, initiatives from both a scientific as well as a societal drive to reduce the number of animal experiments are emerging with the final aim to fully assess the safe use of chemicals by animal-free tests and approaches (NCad, 2016). This is reflected in the increasing research for animal-free approaches, the development of new strategies consisting of (innovative) animal-free methods and tools, based mainly on in silico models and in vitro tests and the development of strategic roadmaps for the transition towards an animal-free approach (Adeleye et al., 2015;Berggren et al., 2017;Blaauboer et al., 2016;Daneshian et al., 2016;Embry et al., 2014;FDA, 2017;Gocht et al., 2015;ICCVAM, 2018;Krewski et al., 2010).
It is often stated that the current legislative frameworks do not accept animal-free or in vitro data. It is true that for the assessment of risks of chemical substances for man and environment, animal studies are commonly performed. Alternatives for these studies may only be applied if the legal frameworks for the assessments explicitly offer a possibility for them. In a report for the Dutch Ministry of Economic Affairs, RIVM has analysed ten European frameworks for assessment of chemical substances to see whether the possibility to use alternatives is a hindrance. The investigation showed that, in a strict sense, there are no legal barriers (Heringa et al., 2014). However, the absence of legal barriers in most frameworks does not mean the road is free to apply the alternatives yet. This is mainly because of the lack of validated methods to predict endpoints such as repeated-dose toxicity, reproductive toxicity, etc., as validation and regulatory (e.g. OECD) acceptance of a method is required by all these regulations.
Another important issue is that the currently developed methods are not sufficiently suitable to assess the safety of chemicals, i.e., they do not always sufficiently address the regulatory questions to be answered in a safety assessment. In relation to the safety assessment of chemicals, a large number of different questions can be posed. For instance: is this chemical an endocrine disruptor? Is this chemical harmful for children or newborns? What is the effect of a combination of pesticides present on fruit? At what dose will this chemical have an effect? The question 'does this chemical cause cancer' might imply a number of sub-questions, e.g., does this chemical cause cancer in a specific exposure situation (via skin contact, or via inhalation; route specific toxicity and dose-response relationship), does this chemical cause cancer in combination with other chemicals (cocktail effect), or does this chemical cause cancer in an individual (susceptibility)? Therefore, it is important to specify the information that is needed, and be aware of the specific safety assessment question that needs to be answered.
In this present publication we focus on the current regulatory needs, that is, the current frameworks for which information on the effects of chemicals on human health needs to be generated to assure their safe production, application and use.
Dependent on the regulatory framework and situation with the substance (e.g., already on the market or not), the questions above might or might not need to be answered. Anyhow, it is important to realize that answering the different regulatory questions requires different pieces of information, and thus may require specific tests or testing strategies. We have distinguished 4 different types of regulatory questions: 1) What are the hazardous properties of the chemical? Is it a sensitizer, does it cause cancer etc, (see section 2.1)?) 2) What is a safe level of exposure, so safe use of the chemical can be assured (see section 2.2)? 3) What effects can be expected when people have been exposed to a higher level than the safe level (see section 2.3)? 4) Does a certain chemical cause a certain disease, such as e.g., Alzheimer's or Parkinson disease, that is not covered directly by the hazardous properties of question 1? It will therefore, be described what regulatory safety (or risk) assessment actually entails, to discuss what the current information requirements are from the viewpoint of the regulatory context in the EU, and from there, what this will mean for the focus and development of animal-free innovations, rather than to discuss the innovations themselves. Sometimes a qualitative outcome may suffice, while in other situations specific strategies combining multiple tools are needed to get a quantitative outcome. These aspects will be of importance if the current regulatory needs are to be met. This paper, therefore, intends to stimulate the discussions on how to bridge the gap between the regulatory needs for safety or risk assessment and to provide input for the development of robust and fit-for-purpose animal-free innovations and thereby accelerate the use of animal-free tools in the regulatory human safety assessments on the short term.

Current assessment of human health risks in a regulatory context in the EU: information needs
In the EU, regulatory bodies are involved in human health risk assessments at multiple levels. The safety evaluations may regard the whole supply chain (manufacture, distribution, storage, use via recycling/circular use up to waste disposal) and may be performed a priori, i.e., before a chemical is introduced on the market, whether or not in a new application, or a posteriori for a current exposure situation for which a concern has risen. Depending on the use of a chemical, market ALTEX preprint published February 25, 2020 doi:10.14573/altex.1912041 3 authorization is evaluated within respective regulatory frameworks. Committees or panels of agencies such as ECHA and EFSA evaluate potential human health risks. At a Member State level, national authorities may request for a health risk assessment of specific exposure situations of concern. The type and detail of information required for such an assessment will depend on the regulatory context of the evaluation. For the purpose of the development of new methods and strategies to replace animal experiments in a regulatory context, the question to be answered is why animal experiments are being performed in the first place, i.e., what information is to be obtained from them and how will the information be used. This is important to realize for the development of animalfree innovations to increase the acceptability in regulatory processes.
Animal data used a priori are meant to provide information on potential human health effects such that a safe production and use of a chemical can be assured. In this setting, animal data serve the purpose of a safety assessment before a chemical is introduced on the market. The amount of data required and thus the type of animal experiments depends on the regulatory framework in question. An extensive toxicity profile is required for e.g., plant protection products and biocidal products but in other frameworks the data requirements are less extensive and may be dependent on the level of production or use, or on the expected level of exposure.
Animal data may also be used a posteriori to evaluate whether a current exposure scenario for which a concern has risen might result in human health risks. A posteriori risk assessments of exposure scenarios of societal concern generally are within the compass of governmental responsibility. These are scenarios for situations either arising from a concern for high exposure (or a different route of exposure) to a chemical or for a specific hazard endpoint, or from both. A risk assessment then is performed to answer the question whether the exposure scenario introduces (or has introduced) a human health risk, and if so, what health risks can be expected and to what extent in terms of type and severity of effects and the size and nature of the population affected. Finally, it will be evaluated what measures need to be taken to reduce exposure. Recent examples include acrylamide and the rubber infill in artificial grass pitches (ECHA, 2017a;RIVM, 2017;Svensson et al., 2003;Tareke et al., 2002; US EPA/ATSDR/CPSC, 2016). In practice, such a risk assessment needs to be performed only for a relatively small number of chemicals as compared to the tens of thousands of chemicals present in the environment. However, these are the situations that often receive the highest media attention and create high societal concern. In these situations, there generally is no time to generate new data and to perform a risk assessment, one has to rely on information that is available for the chemical in question.
Often a tiered approach is followed with data requirements increasing with anticipated exposure. One reason for such an approach is the aspect of proportionality of data requirements in relation to the level of production and/or use. Data requirements may also be restricted because of the limitations of an animal experiment. For instance, not all relevant human endpoints can be studied in traditional toxicological animal experiments. It is therefore noted that the official data requirements in a framework may be less than the actual information need for a complete and adequate safety assessment from a scientific point of view.
In the next sections, it will be discussed what information, that presently is generated using animal experiments, basically is needed within the context of an a priori regulatory safety assessment, i.e., for Classification, Labelling and Packaging (CLP) and for the derivation of a Health Based Guidance Value (HBGV). The focus in this publication will be on the information that is needed within the legal frameworks in the EU, which is somewhat different from, for example, the regulatory situation in the United States (Busquet and Hartung 2017;Krewski et al., 2010). Subsequently, possibilities of animal-free tools for the purpose of an a posteriori risk assessment will be pointed out, as well as the need to investigate the possibilities to address topics of current societal concern.
2.1 A priori safety assessment: Classification of chemicals (CLP) Within the EU, the mandatory hazard classification of chemicals and mixtures being manufactured, imported, used or distributed within the EU is laid down in the Classification, Labelling and Packaging (CLP) Regulation (EC) 1272/2008 (EC 2008). Currently, the harmonised classification and labelling of approximately 4000 chemicals is included in this regulation. The criteria of the international UN Globally Harmonised System of Classification and Labelling of Chemicals (GHS) are incorporated in the EU CLP Regulation. Classified chemicals are labelled with appropriate Hazard (H-) and Precautionary (P-) statements. H-statements relate to specific hazards and P-statements deal with actions to prevent exposure, actions to respond to exposure, with storage conditions and with disposal. These statements aim to warn for specific hazards and to limit or reduce exposure often without knowledge and/or consideration of what a safe level of exposure actually might be.
This regulation guarantees that employees and consumers in the EU have clear information about the hazards of chemicals by mandating the classification and labelling of chemical substances and mixtures (EC 2008). Chemicals proposed to be classified for carcinogenicity, germ cell mutagenicity or reproductive toxicity (CMR) categories 1A, 1B or 2, or for respiratory sensitization, or chemicals that are active substances in the meaning of Regulation EC 1107/2009 (i.e. PPP) or Regulation (EU) No 528/2012 (biocidal products) shall normally be subject to harmonised classification and labelling.
Currently, the hazard evaluation under CLP is based on available data that is generated within the various legal frameworks (such as PPPs, biocidal products and REACH) and performed per endpoint. CLP therefore does not require additional testing (animal or animal-free) for the purpose of classification for human health hazards (ECHA 2017b), and no conclusion is drawn for an endpoint for which no (sufficient) information has been generated in the various legal frameworks. Dependent on the hazardous properties of the chemical, a restriction or ban might apply for the whole supply chain: i.e. from manufacture, distribution, storage, and use up to waste disposal. The classification of chemicals therefore also triggers risk reduction measures in other legislations, as many European and national legislations are directed at the safe use of hazardous chemicals. Examples are legislations concerning consumer products (cosmetics, toys, cleaning products), the protection of workers, transport safety and external safety.
The regulatory need for this type of safety assessment is thus to obtain information on the intrinsic hazards of chemicals. While the majority of the tests as currently presented for evaluation of human health hazard assessment concerns ALTEX preprint published February 25, 2020 doi:10. 14573/altex.1912041 in vivo tests, in vitro tests are accepted and used for a limited number of endpoints as stand-alone for hazard classification. Table 1 presents an overview of in vitro tests that are currently recognized and regulatorily accepted for classification purposes. In this context, it is noteworthy that in vitro tests for skin corrosion and irritation have been introduced at the level of UN GHS in 2019 (United Nations 2019a) and that presently, an in vitro test for eye irritation is under discussion (United Nations 2019b). Furthermore, non-testing approaches such as grouping and QSARs are also accepted. As can be seen from Table 1, classification based on non-animal tests are common practice for the more simple endpoints, but not for the more complex, systemic endpoints. These tests are not regarded as stand-alone tests, but may be used in a weight of evidence approach. These tests should not be used currently to sub-categorize skin sensitizers into subcategories 1A and 1B (ECHA 2017b). c Positive in vitro tests must be accompanied with a chemical structure similarity to known germ cell mutagens to be accepted for a classification, and then only a category 2 classification can be obtained.
Classification on intrinsic hazardous properties is performed in a qualitative way for most endpoints (e.g., yes/no toxic to reproduction, yes/no carcinogenic), with the exception of the endpoints Specific Target Organ Toxicity Single Exposure (STOT-SE)/Repeated Exposure (STOT-RE) for which quantitative exposure boundaries apply. What thus basically is needed when moving towards classification based on animal-free tools, is a data-generating strategy that aims to assess whether a chemical has the potential to induce a specific health effect. Such a strategy will consist of a number of integrated tools that, in combination, provides information for classification; the basic steps are depicted in Figure 1. For local effects, kinetics generally play no or just a limited role. In case of a systemic effect, the role of kinetics becomes of importance and appropriate kinetic tools need to become part of the strategy. For the demonstration of a potential intrinsic effect of the chemical it is important to know what the potential toxic agent will be (parent compound or metabolite) and whether the toxic agent reaches the target tissue(s) (absorption, distribution). For this purpose, kinetic and hazardous properties need to be generated in an integrated and coherent way. The actual exposure conditions in humans are not of importance. Quantitative extrapolation to a realistic human exposure scenario is not necessary for classification, which may call for less strict demands on the tools to be developed than in case Quantitative In vitro -In vivo Extrapolation (QIVIVE) is required. The present classification categories based on in vivo animal experiments are generally linked to a specific severity of an effect; boundaries have been defined for assignment into these categories. Obviously, when animal experiments are replaced by (a strategy of) animal-free innovations these classification boundaries and criteria may need to be redefined, based on the endpoint observed in a new strategy or method. For instance, where skin corrosion criteria for the in vivo acute dermal irritation/corrosion test (OECD TG 404) are based on visible destruction of skin tissue in test animals, the criteria for in vitro methods are based on, e.g., the integrity of the skin barrier function (OECD TG 430 and 435) or on tissue viability (OECD TG 431) (United Nations 2019a).

A priori safety assessment: Derivation of guidance values and limit values
EU regulatory frameworks where safety evaluations are performed for chemicals (or products) include REACH, biocidal products, PPPs, novel foods, feed additives, food improvement agents such as enzymes, additives and flavourings, food contact materials, veterinary medicinal products and cosmetics. The main aim of the safety evaluation in (most of) these frameworks is to derive a kind of exposure limit, such as an HBGV or a product limit, that is used to control exposure in order to ensure the safe production, manufacturing and use of a chemical. Although the regulatory information requirements differ between the frameworks, they share the same purpose, i.e., establishing conditions for safe production and/or use. Currently, the establishment of an HBGV is (mostly) based on a reference point for the critical effect, being a no observed adverse effect level (NOAEL) or benchmark dose (BMD) or benchmark concentration (BMC) from an animal experiment, by applying assessment factors. In an in vivo experiment with intact animals, all physiological processes, including kinetics, and selected endpoints are implicitly addressed.
If guidance values or limit values are to be derived via an animal-free approach, then what information is basically needed for their derivation and how could this information be obtained? What elements should hereto be included in a strategy for the derivation of HBGVs or product limits? Generally speaking, as a minimum prerequisite, information about the kinetics of a chemical and the critical effect(s) is needed. Kinetic information is needed to assess the absorption of the chemical, to identify the potential toxic agent (parent compound or metabolite) and to determine to what extent organs and tissues are exposed to the toxic agent. In contrast to the information needed for CLP, the kinetic information and the occurrence of the critical effect (exposure-response) need to be determined quantitatively.
The assessment of an HBGV or limit value via an animal-free approach may, for instance, follow a stepwise strategy as depicted in Figure 2. The first step can be considered as a screening step in which, based on kinetic information and a basic screening of hazardous properties, potential critical effect(s) can be identified for the most toxic agent (parent compound or metabolite). To achieve this, a screening approach is needed in which information on kinetics (i.e., potential for a possible toxic agent to reach specific tissues) and hazardous endpoints can be generated in an integrated way. This is in fact already to some extent in line with the current approach used for genotoxicity testing, where three complementary in vitro assays are required, at least in a first tier, and each is performed in absence and presence of metabolizing liver enzymes (S9). The screening step for derivation of an HBGV may, for instance, consist of a battery of tools through which information can be generated for multiple potential endpoints. This step needs to be designed such that potential hazards will not be missed. By including knowledge about physiological and biological processes that may lead to adverse effects, such as for example the adverse outcome pathway (AOP) network, one or more effects may be identified that can be considered as a potential critical endpoint for HBGV derivation.
In a second step, these identified critical effects may be subject of a more detailed and quantitative evaluation by adequate in vitro assays and/or in silico models to establish exposure-response curves from which internal BMCs or BMDs can be derived. Also in this step, the generation of hazard information needs to be conducted in combination with that of kinetic information. Kinetic information can be used to optimize the exposure conditions in the test designs to bring them in line with realistic human exposure scenarios and to determine an adequate dose metric. For the derivation of an HBGV, the internal dose or concentration that represents a threshold for a potential critical effect, needs to be extrapolated to an external exposure level by QIVIVE. It may also be extrapolated to an in vivo internal blood concentration, or other relevant biomarker, if guidance values, such as for instance a Biological Limit Value (BLV), are designed for human biomonitoring. The determination of the most critical effect to serve as point of departure for an HBGV should be based on comparison of the in vivo BMC or BMD (external or internal). In a third and final step, an HBGV or BLV is derived from the most relevant and critical BMC or BMD.
It is acknowledged that in principle, such an approach does require that adequate animal-free tools are available for multiple relevant endpoints but the advantage is that a chemical only needs to be tested in detail for the identified critical endpoints. Obviously, CMR properties need to be included in the screening step for critical endpoints. However, if CMR properties are a reason for non-authorization, a qualitative assessment may suffice. The current practice is to address uncertainties in the derivation of guidance values, among which the extrapolation from animals to humans, by application of assessment factors. Similarly, uncertainties resulting from the use of an animalfree approach (e.g., the use of cell or organ systems instead of an intact organism or the extrapolation by QIVIVE) can be taken into account by application of assessment factors. Correction for a high source of uncertainty due to the interspecies differences does not apply here if information obtained with human cells or tissues can be used.
Over the past few years, several stepwise or tiered approaches have been suggested to evaluate potential health risks or to perform a safety assessment for chemicals within specific regulatory frameworks (e.g., (Adeleye et al., 2015;Berggren et al., 2017;Blaauboer et al., 2016;Dent et al., 2018;Desprez et al., 2018;Embry et al., 2014;Pastoor et al., 2014)). These strategies serve different purposes and do not (always) focus on the derivation of HBGVs but are mentioned to illustrate how animal-free strategies and methods can be combined in a tiered approach to generate answers to specific regulatory questions.
One major topic one should be aware of, is the fact that frameworks that do not require data depend on the data obtained within the frameworks that do require data. An example is exposure values used in emergency response planning or land-use planning. Another example is the derivation of occupational exposure levels (OELs), although it is acknowledged that within REACH, derived no-effect levels (DNELs) for the worker are also being derived. Therefore, it must be kept in mind that the information generated for the derivation of HBGVs for chemicals within a specific regulatory framework should be sufficient to derive OELs and perform emergency response and land-use planning, as well.

A posteriori risk assessment of exposure situations of concern
As mentioned before, a posteriori risk assessments for the evaluation of human health risks in a historic, current or emerging exposure scenario for which a concern has risen are performed for a relatively small number of chemicals. The exposure can be the main trigger to perform such a risk assessment, e.g., in case of a fire or leakage at an industrial site or in case a contaminant in food is detected, such as the recent acrylamide case (Tareke et al., 2002;van Bruggen et al., 2010). Another example may be unrecognized high emissions of a chemical during a long time period. A trigger for a full risk assessment may also come from public concern about the hazardous properties of a chemical compound, e.g., the use of rubber infill in artificial football pitches (RIVM 2017). In the case that the need for a full risk assessment is caused by concern about exposure, knowledge about the critical effect of a chemical and/or a safe level no longer suffices. People who are exposed want to know what health risks they can expect from that exposure. Therefore, quantitative information on the exposureresponse relationship is needed, not only for the critical effect but also for other effects that might occur at higher exposure levels. Such an evaluation requires detailed knowledge about the toxicity profile of a chemical, preferably in relation to the specific characteristics of the exposure scenario in question, e.g., route of exposure, frequency, duration, and concentration/dose. When a public concern about specific hazardous properties of a chemical arises, in principal, a complete insight in the toxicity profile is not necessarily needed. Quantitative information about the exposure-response relationship for the hazardous endpoint(s) of interest suffices. Since exposure has already occurred (in contrast to most a priori risk assessments), available epidemiological data can provide an important additional source of information.
Whether the high level of exposure or the public concern triggers the risk assessment, there usually is only limited time to come up with an answer and one has to rely on the data available. The steps that need to be taken are reflected in step  Figure 3. The characteristics of the exposure scenario of interest is described in detail to determine and search for the appropriate kinetic and hazard information that best fits to address potential health risk resulting from that scenario. If the available information is insufficient for risk assessment, information gaps are identified. Generation of additional information to fill these gaps (step 2) can only be performed when this is possible within a short time frame (and if sufficient budget is available), depending on the urgency to address the concern. Animal experiments are then not the appropriate way forward; especially animal studies with repeated exposures are time-consuming. Thus, animal-free innovations are needed to generate data about potential effects that might occur in the exposure scenario of concern such that potential effects in relation to the exposure scenario of interest can be evaluated and appropriate exposure-effect relationships can be assessed. In these situations, in silico tools (including PBPK-models) may also become of importance, even more than in vitro assays, as these will also require time to perform, to analyse and to interpret. The more so, when in silico tools can be developed that are sufficiently flexible to mimic the exposure conditions under evaluation, thereby making it possible to apply available information directly to the exposure scenario for which the risk assessment is performed. Development of such flexible tools will be of great advantage for risk assessments since the exposure conditions of an emerging situation (e.g., route of exposure, frequency, duration and concentration) often cannot be foreseen. Finally, a quantitative risk assessment can be performed for the exposed populations in step 3. Thus, to serve the purpose of risk assessments of past, occurring or emerging exposures, it would be favourable if tools are being developed that can aid a risk assessment by quickly generating new information for a chemical or by supporting the extrapolation of available data to the specific exposure scenario under evaluation. Up till now, incorporation of PBPK-modelling has proven its value in several risk assessments (Andersen 2003;Bos et al., 2006;Clewell et al., 2005;Mielke and Gundert-Remy, 2012). However, the development of these types of models requires a large amount of good quality data from in vitro and/or in vivo tests.

Addressing specific topics of societal concern
A large potential benefit of animal-free approaches is that specific topics can be addressed that cannot or only limitedly be tackled by information from animal experiments. Especially topics that are presently subject of societal discussions may benefit, such as whether exposure to chemicals may be related to development of contemporary diseases such as diabetes, cardiovascular diseases, behavioural effects as ADHD, or autism, Alzheimer's, etc (Bellanger et al., 2015;Gwinn et al., 2017). Animal models currently used for regulatory purposes do not provide information on these endpoints. In addition, also questions with regard to effects of combined exposures to (mixtures of) multiple chemicals are not sufficiently addressed when using animal studies (Gwinn et al., 2017). The current safety or risk assessment paradigm based predominantly on animal experiments is oriented towards the health effects of single chemicals, whereas in practice, the human body is exposed to a cocktail of chemicals. These topics favour a system in which human relevant cell systems for new endpoints and human relevant technologies are used. Lessons can be learned from advances in human medicine research and disease models. In addition, combinations of in vitro and in silico tools are needed in which the mixture effect of many different combinations of chemicals can be studied. This information need will not be easily met. However, from a societal and also from a healtheconomic point of view, it is worthwhile to explore the possibilities of new methods and methodologies to include this information in human safety and risk assessment (Pistollato et al., 2016).

Implementation of animal-free innovations in the EU: next steps
As pointed out, societal concerns for reasons of animal welfare as well as scientific data-gaps and concerns are driving forces for the development of animal-free innovations. Use of animal-free innovations potentially can improve the safety and risk assessment for certain aspects where currently animal studies are not requested or conducted for various reasons. In addition, an animal-free approach including appropriate in silico tools may provide the flexibility to better address specific human exposure scenarios in practice. Building tools based on human physiology and biology and the use of human tissues and cells may improve the relevance of the outcome for the assessment of human health risks (Burden et al., 2015). The ideal picture would be that developments of animal-free innovations will finally result in an approach that makes it possible to determine a complete profile of toxic effects and of human diseases for every chemical coming to or already on the market, with acceptable costs in terms of resources and time. While the final timelines and costs are still unclear, developments based on, e.g., AOPs, ontology mapping, virtual organs and body-on-a-chip may ultimately lead to approaches that provide such toxicity profiles. An assessment of human health risks based on such an approach will require a shift towards a new safety assessment paradigm, one that is based on human biology and physiology rather than the present evaluation of individual endpoints in animal experiments (Marx et al., 2016;Staal et al., 2017). Interesting and challenging initiatives and proposals are already being presented with 'the virtual human' as the ultimate remote prospect . New information on chemical toxicity could be derived by artificial intelligence mining existing data sets, combined with the information from the new technologies mentioned above (Hartung 2019). The shift to a new safety assessment based on human biology and physiology will need a regulatory acceptance process that is rather complex. Schiffelers et al. (2014) have applied the multilevel perspective on technology transitions (Geels, 2002) to describe the influences at the micro-, meso-and macro level needed to induce regulatory acceptance ( Figure  4). Innovations take place at the micro level, where new methods are developed, tested and validated. The meso level entails the current situation or regime, where institutions with existing rules and regulations find their place and influence the acceptance of innovations. Societal features, such as the social attitude towards animal welfare and also the political culture, are present at the macro level. Developments at the macro level (e.g., concern for animal welfare and about the predictive power of animal models) and at the micro level (e.g., increasing availability and applicability of animal-free innovations) are important drivers to stimulate the necessary changes at the meso level for regulatory acceptance. However, discussions are needed at and between these three levels to assure a broad acceptance of safety and risk assessment based on animal-free innovations. Discussion topics include the uncertainty about the value of the new animal-free innovations (micro level), the lack of harmonization of regulatory requirements and acceptance criteria (meso level) and the high levels of risk aversion and the concern for animal welfare (macro level) (Schiffelers et al., 2014). In optimizing the process, critical drivers are commitment, communication, cooperation and co-ordination. However, despite the huge efforts undertaken, it is not realistic to expect that such an approach will be possible for all chemicals entering the market, at least not in the foreseeable future (Sewell et al., 2017). It will take decades before a paradigm shift towards a safety assessment fully based on animal-free approaches will have been achieved. It may turn out to be disadvantageous for the development of animal-free innovations if significant steps in the replacement of animal experiments by animal-free innovations cannot be taken in the meantime, i.e., if the implementation of these innovations are postponed until it is possible to generate a complete, quantitative profile of human health risks. The momentum towards a transition may be lost if the practical advantages of animal-free approaches for an adequate assessment of human health risks will not become clear in short term. At the moment, a lot of effort is put into the development of new methods (micro level). At the same time, application of animal-free innovations in the current safety and risk assessments may provide the proof of principle for new developments. Therefore, next to the developments towards a new safety or risk assessment paradigm, implementation of animal-free developments in the current human safety assessments should be stimulated.
There is a request both from the developers of new tools and technologies (micro-level) and the urge to reduce animal testing (macro-level) to speed up the process of human risk assessment based on human biology and physiology. This is mainly the role of the regulatory implementation and acceptance (meso-level), for which co-ordination and collaboration at all three levels (micro, meso and macro) is key.

Need for co-ordination and collaboration
For an innovation to be successfully taken up in regulatory safety testing, it is important to start from a foreseen need and to determine the kind and level of detail of the information required for that need rather than to first develop and finalize a sophisticated tool before seeking where it can best be applied. The basic needs for safety assessments (e.g., CLP, derivation of HGBVs and product-limits) and risk assessments of exposure situations of concern have been discussed. Depending on the regulatory context, the regulatory needs clearly vary and different criteria are applicable to the level of detail and of uncertainty of the data to be generated. For one purpose, qualitative information may suffice while for other purposes a quantitative evaluation such as QIVIVE may be required. Especially when based on animal-free innovations, fulfilment of a regulatory need will often require information that is obtained by different scientific fields and thus requires a multidisciplinary approach. At least for systemic endpoints, new tools for hazard evaluation and kinetics need to be developed in coherence. For instance, if the hazard evaluation appears to be optimally generated by a combination of different techniques, such as omics, computational tools and/or in vitro assays, the question is how these are to be combined and how and where kinetics needs to be included. The tools to be developed should be complementary and compatible.
To stimulate the introduction of animal-free innovations for the assessment of human health risks, it is therefore of utmost importance to intensify the dialogue between the different stakeholders involved. What then is needed, is the development of a safety assessment strategy describing the information needs to answer a specific regulatory question, the steps to be taken and the criteria that should be met. To achieve such an integrated assessment strategy that is fit for purpose, a multi-disciplinary collaboration, including regulators, scientists, and representatives from industries, is preferred already in the developmental stage of animal-free innovations (Benfenati et al., 2016;Burden et al., 2015;Meek and Lipscomb, 2015;Sewell et al., 2017). In this, there is an important role in store for regulatory officers, risk assessors as well as risk managers to clearly formulate the regulatory questions to be answered and to define the criteria involved (Bell et al., 2018;Punt et al., 2018). Scientists can put forward the most recent developments and their possibilities. It can be decided in consultation how the regulatory need can best be addressed such that compatible tools can be developed and criteria be defined that should be met for the respective regulatory goals. Working in collaboration and combining regulatory needs and technical possibilities will provide the best assurance for implementation and application of animal-free innovations within the field of regulatory safety testing within the foreseeable future. Involvement of regulatory officers already in the development phase of animalfree innovations will further contribute to the creation of a broad support for acceptance of these innovations and to strategies that will be fit for the purpose of safety and risk assessment and that are trusted, accepted and legally binding.

Acceptance and implementation of animal-free innovations
Whether implementation of animal-free innovations for a priori safety assessments in regulatory frameworks or for a posteriori risk assessment of chemicals will be successful will further depend on several factors. First of all, there should be no legal boundaries in the European frameworks for the use of animal-free innovations. It has been determined that this is in principle indeed the case (Heringa et al., 2014). An important hurdle for regulatory acceptance rather appears to be a lack of reproducible and robust methods (Sewell et al., 2017). For instance, to sufficiently address health effects from repeated exposures, a testing strategy consisting of a combination of advanced methods is required to identify critical effects and to determine the toxicity profile of a chemical. Unless such a strategy will encompass a sufficient number of relevant endpoints, an animal experiment will still be performed instead. Secondly, a fundamental change in mind-set is necessary for a successful shift from an assessment where animal testing is still seen as the gold standard, towards a safety assessment that is completely based on animal-free innovations (Schiffelers et al., 2007;Sewell et al., 2017). Gathering data from animal experiments for several endpoints has been the way forward for many decades, and the experience over the past decades with methods based on animal models has created a certain degree of trust in animal data and has provided at least a certain feeling of safety for the current approaches. It has also created a basis for making legal decisions, and new methods that are to be implemented should cover at least the same legal basis and acceptance as the one they are replacing. This has created some reserve towards implementation of animal-free approaches in safety and risk assessment. We know what we have and to blaze a new and uncertain trail may lead to a road of trial and error, which is considered not to be the best way forward when human health is concerned.
It is acknowledged that there are uncertainties involved in using animal experiments for human safety or risk assessment; these uncertainties are generally accounted for by assessment factors. Similarly, the use of animal-free ALTEX preprint published February 25, 2020 doi:10.14573/altex.1912041 innovations will also be accompanied with uncertainties, but these will be other uncertainties with which we have little or no experience. Starting the introduction of animal-free innovations in a stepwise approach provides the opportunity to gain practical experience, and thus trust in the predictive power of robust animal-free innovations. At the same time, it can be ensured that they meet the requirements of the regulatory frameworks and the needs of regulators and risk assessors. Gradual implementation of practical, straightforward, and fit for purpose animal-free approaches also provides possibilities for interim adjustments and improvements where considered necessary. Additional issues of importance that need to be addressed during the development include determination of an acceptable level of uncertainty, robustness, reproducibility and transparency. To create trust, the issues regarding uncertainty, robustness, reproducibility, predictive power and transparency need to be agreed upon and accepted by the respective stakeholders. Combining building blocks existing of complex animal-free innovations, such as omics, organs-on-a-chip, computational tools, ontology mapping etc., into an approach that meets a specific regulatory need, will require a multidisciplinary approach (Benfenati et al., 2016). Such building blocks may be difficult to understand for someone who is not an expert in the multiple scientific fields involved, but nevertheless does have to evaluate these building blocks individually and in conjunction and has to judge them on their own merits. It is therefore of importance that end-users of these tools, such as regulatory officers, risk assessors and risk managers are acquainted with the new technologies, and in addition already are involved during their development. This further supports that new developments, in combination with the mentioned requirements they should meet, need to be established via a dialogue of multiple stakeholders (ECHA, 2018;Punt et al., 2018).
Another important issue for the implementation of animal-free innovations is that the validation process needs to be reconsidered (Burgdorf et al., 2019;NAS, 2017;Piersma et al., 2018;Tralau et al., 2015). Animal experiments are considered to be the gold standard for validation but this is in contradiction with one of the scientific reasons to replace animal experiments: they may not always be the appropriate model for human safety and risk assessment. In addition, when new topics such as new endpoints, or combined or aggregated exposures are introduced, the amount of presently available animal data may be too limited for validation. Therefore, development of animal-free innovations should go hand in hand with research on how to validate these innovations and to develop appropriate methods for validation.
A possibility to meet the reserve and concerns towards introduction of animal-free innovations for safety evaluation is the introduction of a trusted assessment environment, or a safe harbour, where an open dialogue between industry and regulatory authorities, a safety assessment of a chemical or a product, based on data obtained with animal-free methods can be encouraged.
Another possibility is the use of an appropriate measure of post-marketing control of exposure as a safety net. This can be achieved by the introduction of BLVs and the set-up of health surveillance systems. New animal-free approaches will require more and more detailed biological and kinetic information. This will provide the opportunity to derive adequate BLVs which will be better and more adequate exposure estimates than external HBGVs, especially in case of exposure via multiple routes and/or from multiple sources. BLVs describe the actual internal exposure. Setting up surveillance systems to regularly measure these BLVs and perform basic health assessment will help to control the exposure more adequately. Furthermore, the information thus obtained can in its turn be used to evaluate the adequateness of the safety evaluation. It may help to further fill in data gaps and to fine-tune the safety assessment.

Summary and conclusions
Development of animal-free innovations that are directly aimed to meet the present regulatory needs has multiple advantages (Burden et al., 2015). It may contribute to a reduction of animal use in the short term. Instead of an uncontrolled proliferation of new animal-free innovations, the development of animal-free innovations will be focused on what is really needed from a regulatory point of view, leading to fit-for-purpose tools. This will advance the possibilities for their use within the regulatory safety assessments. Agreements among the respective stakeholders on the level of detail and uncertainty, robustness, predictive value, reproducibility and method for validation are a prerequisite to come to tools that can be trusted and that will be legally binding (Bell et al., 2018;Punt et al., 2018). To achieve this, clear agreements are necessary on criteria that the tools should meet as well as on the interpretation of the outcome of the tools and the resulting conclusions, decisions and consequences.
To identify the regulatory information needs, different areas of safety or risk assessment were discerned that presently rely on data from animal experiments, including CLP and derivation of HBGVs and product limits. In the EU, the majority of chemicals are currently regulated by minimizing exposure via CLP by restricting the manufacture, distribution, use and storage of a chemical and/or the derivation of HBGVs. As a third and fourth area, risk assessment of human exposure scenarios of concern and addressing societal issues were identified, respectively.
In principle, CLP is based on the intrinsic hazard of a chemical combined with agreements on a categorization based on the test results. Simply put, a mutual agreement on, for example, a combination of individual (qualitative) in vitro and/or in silico hazard and kinetic assays with information about the human physiology and biology, and the establishment of new or adapted classification criteria may be sufficient for the classification of chemicals under an animal-free evaluation system.
For the derivation of HBGVs and product limits, basic, quantitative information is required about the kinetics of a compound and its critical effect (including the dose-response relationship to derive an appropriate point of departure). An approach then may be the development of a screening strategy to first identify the most toxic agent(s) and the most probable critical endpoints. The screening step may subsequently be followed by a more detailed quantitative evaluation of the selected endpoints to estimate a quantitative relationship between an appropriate target dose metric and the respective health effects. This relationship can be translated to an external or internal estimate of safe human exposure by QIVIVE. In a third ALTEX preprint published February 25, 2020 doi:10.14573/altex.1912041 11 step, an HBGV or BLV may then be derived from this safe exposure estimate. Each of the three steps will require different tools (in vitro kinetics, in vivo kinetics, in silico) with different levels of quantification and certainty. The consequence of animal-free approaches in safety assessment may be that limited data become available for an a posteriori quantitative risk assessment in case that, for instance, an HBGV is exceeded. However, it should be realized that present HBGVs are not always based on a complete toxicity database either. A complete toxicity profile is seldom available and often data gaps are identified. Risk assessments of emerging exposure situations may be subject to constraints of time. It would then be favourable if animal-free innovations are being developed that are able to quickly generate data, especially if these innovations are sufficiently flexible to address a variety of exposure scenarios.
The fourth area regards the inclusion of specific topics in the assessment of human health risks that until now cannot or not fully be addressed by animal experiments. Animal-free innovations may be developed to evaluate whether chemicals may cause a human disease thereby improving the safety assessment process. The assessment of human health risks from combined exposures to chemicals often is limited to chemicals with a similar mode of action, such as the group of organophosphates and the inhibition of cholinesterase. Also exposure to well-known mixtures such as mineral oil or white spirit has been subject of toxicity studies. But exposure to combinations of chemicals also includes other areas of interaction, such as induction or inhibition of metabolising enzymes which may affect the activation or deactivation of other chemicals. Animal-free innovations enabling faster data generation (and against reasonable costs) could allow more assessments of combined exposures.
A multi-stakeholder dialogue is needed to develop animal-free tools that meet the present regulatory needs. Therefore, close involvement of regulators in the development of animal-free innovations is a prerequisite to assure that the regulatory needs are clearly defined and that criteria for a robust assessment of human health following chemical exposure will be met (Benfenati et al., 2016;ECHA, 2018;Meek and Lipscomb, 2015;Punt et al., 2018). This dialogue should also address additional issues, including how to create trust and confidence in assessments based on animal-free innovations, agreement upon how to deal with uncertainties involved and the level of robustness, reproducibility and transparency needed. Experience and thus trust can be gained with animal-free tools and models aimed at the present regulatory needs, which can be brought forward in the development of more complex innovations. Therefore, a gradual implementation may aid the phasing out of animal testing and pave the road for more complex and sophisticated innovations. This may accelerate the developments of animal-free innovations as well as acceptance of these innovations in human safety or risk assessment in a regulatory context.