Case study on developmental toxicity of mehylhexanoic acid. OECD Series on testing and assessment (IATA case studies). No. #325

Organisation for Economic Co-operation and Development ENV/JM/MONO(2020)21 Unclassified English - Or. English 9 October 2020 ENVIRONMENT DIRECTORATE JOINT MEETING OF THE CHEMICALS COMMITTEE AND THE WORKING PARTY ON CHEMICALS, PESTICIDES AND BIOTECHNOLOGY Cancels & replaces the same document of 24 September 2020 Case study on the use of Integrated approaches to testing and assessment for READ-ACROSS BASED FILLING OF DEVELOPMENTAL TOXICITY DATA GAP FOR METHYL HEXANOIC ACID Series on Testing and Assessment No. 325 The corresponding annexes are available under the following cotes: ENV/JM/MONO(2020)21/ANN2 andANN3. JT03466608 OFDE This document, as well as any data and map included herein, are without prejudice to the status of or sovereignty over any territory, to the delimitation of international frontiers and boundaries and to the name of any territory, city or area. 2  ENV/JM/MONO(2020)21 Unclassified ENV/JM/MONO(2020)21  3 Unclassified OECD Environment, Health and Safety Publications Series on Testing and Assessment No. 325 CASE STUDY ON THE USE OF INTEGRATED APPROACHES TO TESTING AND ASSESSMENT FOR READ-ACROSS BASED FILLING OF DEVELOPMENTAL TOXICITY DATA GAP FOR METHYL HEXANOIC ACID Environment Directorate ORGANISATION FOR ECONOMIC CO-OPERATION AND DEVELOPMENT Paris 2020 4  ENV/JM/MONO(2020)21 Unclassified About the OECD The Organisation for Economic Co-operation and Development (OECD) is an intergovernmental organisation in which representatives of 37 industrialised countries in North and South America, Europe and the Asia and Pacific region, as well as the European Commission, meet to co-ordinate and harmonise policies, discuss issues of mutual concern, and work together to respond to international problems. Most of the OECD’s work is carried out by more than 200 specialised committees and working groups composed of member country delegates. Observers from several countries with special status at the OECD, and from interested international organisations, attend many of the OECD’s workshops and other meetings. Committees and working groups are served by the OECD Secretariat, located in Paris, France, which is organised into directorates and divisions. The Environment, Health and Safety Division publishes free-of-charge documents in twelve different series: Testing and Assessment; Good Laboratory Practice and Compliance Monitoring; Pesticides; Biocides; Risk Management; Harmonisation of Regulatory Oversight in Biotechnology; Safety of Novel Foods and Feeds; Chemical Accidents; Pollutant Release and Transfer Registers; Emission Scenario Documents; Safety of Manufactured Nanomaterials; and Adverse Outcome Pathways. More information about the Environment, Health and Safety Programme and EHS publications is available on the OECD’s World Wide Web site (www.oecd.org/chemicalsafety/). This publication was developed in the IOMC context. The contents do not necessarily reflect the views or stated policies of individual IOMC Participating Organisations. The Inter-Organisation Programme for the Sound Management of Chemicals (IOMC) was established in 1995 following recommendations made by the 1992 UN Conference on Environment and Development to strengthen co-operation and increase international coordination in the field of chemical safety. The Participating Organisations are FAO, ILO, UNDP, UNEP, UNIDO, UNITAR, WHO, World Bank and OECD. The purpose of the IOMC is to promote co-ordination of the policies and activities pursued by the Participating Organisations, jointly or separately, to achieve the sound management of chemicals in relation to human health and the environment. ENV/JM/MONO(2020)21  5 Unclassified © OECD 2020 Applications for permission to reproduce or translate all or part of this material should be made to: Head of Publications Service, RI[email protected], OECD, 2 rue André-Pascal, 75775 Paris Cedex 16, France OECD Environment, Health and Safety Publications This publication is available electronically, at no charge. Also published in the Series on testing and Assessment link For this and many other Environment, Health and Safety publications, consult the OECD’s World Wide Web site www.oecd.org/chemicalsafety/ or contact: OECD Environment Directorate, Environment, Health and Safety Division 2, rue André-Pascal 75775 Paris cedex 16 France Fax : (33-1) 44 30 61 80 E-mail : ehs[email protected]rg 6  ENV/JM/MONO(2020)21 Unclassified Forward OECD member countries have been making efforts to expand the use of alternative methods in assessing chemicals. The OECD has been developing guidance documents and tools for the use of alternative methods such as (Q)SAR, chemical categories and Adverse Outcome Pathways (AOPs) as a part of Integrated Approaches for Testing and Assessment (IATA). There is a need for the investigation of the practical applicability of these methods/tools for different aspects of regulatory decision-making, and to build upon case studies and assessment experience across jurisdictions. The objective of the IATA Case Studies Project is to increase experience with the use of IATA by developing case studies, which constitute examples of predictions that are fit for regulatory use. The aim is to create common understanding of using novel methodologies and the generation of considerations/guidance stemming from these case studies. This case study was developed by EU ToxRisk project (BIAC) for illustrating practical use of IATA and submitted to the 2019 review cycle of the IATA Case Studies Project. This case study was reviewed by the project team. The document was endorsed at the 4th meeting of the Working Party on Hazard Assessment in June 2020. The following case study was also reviewed in the project in 2019: 1. CASE STUDY ON USE OF AN INTEGRATED APPROACH TO TESTING AND ASSESSMENT (IATA) AND NEW APPROACH METHODS TO INFORM A THEORETICAL READ-ACROSS FOR DERMAL EXPOSURE TO PROPYLPARABEN FROM COSMETICS, ENV/JM/MONO(2020)16. 2. CASE STUDY ON THE USE OF INTEGRATED APPROACHES FOR TESTING AND ASSESSMENT FOR SYSTEMIC TOXICITY ARISING FROM COSMETIC EXPOSURE TO CAFFEINE, ENV/JM/MONO(2020)17. 3. CASE STUDY ON THE USE OF INTEGRATED APPROACHES FOR TESTING AND ASSESSMENT FOR 90-DAY RAT ORAL REPEATED-DOSE TOXICITY OF CHLOROBENZENE-RELATED CHEMICALS, ENV/JM/MONO(2020)18. 4. CASE STUDY ON THE USE OF INTEGRATED APPROACHES FOR TESTING AND ASSESSMENT TO INFORM READ-ACROSS OF PALKYLPHENOLS: REPEATED-DOSE TOXICITY, ENV/JM/MONO(2020)19. 5. CASE STUDY ON THE USE OF INTEGRATED APPROACHES TO TESTING AND ASSESSMENT FOR PREDICTION OF A 90 DAY REPEATED DOSE TOXICITY STUDY (OECD 408) FOR 2-ETHYLBUTYRIC ACID USING A READ-ACROSS APPROACH FROM OTHER BRANCHED CARBOXYLIC ACIDS, ENV/JM/MONO(2020)20. 6. CASE STUDY ON THE USE OF INTEGRATED APPROACHES TO TESTING AND ASSESSMENT FOR IDENTIFICATION AND CHARACTERISATION OF PARKINSONIAN HAZARD LIABILITY OF DEGUELIN BY AN AOPBASED TESTING AND READ ACROSS APPROACH, ENV/JM/MONO(2020)22. ENV/JM/MONO(2020)21  7 Unclassified 7. CASE STUDY ON THE USE OF INTEGRATED APPROACHES TO TESTING AND ASSESSMENT FOR MITOCHONDRIAL COMPLEX-III-MEDIATED NEUROTOXICITY OF AZOXYSTROBIN - READ-ACROSS TO OTHER STROBILURINS, ENV/JM/MONO(2020)23. These case studies are illustrative examples, and their publication as OECD monographs does not translate into direct acceptance of the methodologies for regulatory purposes across OECD countries. In addition, these cases studies should not be interpreted as official regulatory decisions made by the authoring member countries. In addition, a considerations document summarising the learnings and lessons of the review experience of the case studies is published with the case studies: REPORT ON CONSIDERATIONS FROM CASE STUDIES ON INTEGRATED APPROACHES FOR TESTING AND ASSESSMENT (IATA) -Fifth Review Cycle (2019) -, ENV/JM/MONO(2020)24. This document is published under the responsibility of the Joint Meeting of the Chemicals Committee and Working Party on Chemicals, Pesticides and Biotechnology. 8  ENV/JM/MONO(2020)21 Unclassified Abstract / Synopsis / Executive summary 2-Methylhexanoic acid (MHA) is a compound for which developmental and reproductive toxicity test (DART) data is taken to be lacking. We have searched for structural analogues that have this data in order to explore the possibility to read across information of these source chemicals to MHA. The following structural related aliphatic carboxylic acids were selected that have in vivo developmental and/or reproductive toxicity data: 2-ethylhexanoic acid (EHA), 2-propylpentanoic acid (VPA), 2-propylheptanoic acid (PHA), 2ethylbutanoic acid (EBA), 4-pentenoic acid (PA), 2-propyl-4-pentenoic acid (4-ene-VPA), and 2-dimethylpentanoic acid (DMPA). Some of these analogues proved to be clear developmental toxicants, i.e. VPA, PHA, EHA, and 4-ene-VPA, while others were identified as not being toxic to development, i.e. EBA, PA, and DMPA; i.e. they did or did not induce neural tube defects upon in vivo exposure. Thus, structural similarity alone doesn’t allow a conclusion on the developmental toxicity of MHA. Therefore, we have also tested MHA and all the selected source chemicals in a battery of in vitro tests with clear relevance to developmental toxicity, i.e. the Zebrafish Embryo Test (ZET), mouse Embryonic Stem cell Test (mEST), iPSC-based neurodevelopmental model (UKN1), and a series of CALUX Reporter assays, that we combined with toxicokinetic models to calculate effective cellular concentrations and associated in vivo exposure doses. With these new approach methodologies (NAM) we wanted to explore whether they could correctly predict the in vivo developmental toxic properties of these aliphatic carboxylic acids, and thus could be used to predict the in vivo developmental toxicity of MHA itself. This data would also allow to further explore the relationship between structure and developmental toxicity within this series of aliphatic carboxylic acids. For that reason, we have also tested 2-methylpentanoic acid (MPA) in these NAM, despite the absence of in vivo data. We have also investigated the potential to inhibit histone deacetylase in ZET, mEST, and UKN1 models, as this enzyme is postulated to be the molecular initiating target leading to neural tube defects observed with these analogues. The NAM results show that VPA, PHA, EHA, and 4-ene-VPA were correctly predicted as in vivo developmental toxicants, and EBA, and DMPA as non-developmental toxicants. The NAM results suggest that MHA may not be fully negative for developmental toxicity. ENV/JM/MONO(2020)21  9 Unclassified Table of Contents Forward .................................................................................................................................................. 6 Abstract / Synopsis / Executive summary ........................................................................................... 8 1. Introduction ..................................................................................................................................... 12 2. Purpose ............................................................................................................................................. 14 2.1. Purpose of use (targeted regulatory framework) ......................................................................... 14 2.2. Target chemical / category definition / source chemicals ........................................................... 14 Target chemical .............................................................................................................................. 14 Category definition ......................................................................................................................... 15 Source chemicals ............................................................................................................................ 15 2.3. Endpoint(s) for which the read-across is performed ................................................................... 15 2.4. Exposure information.................................................................................................................. 15 3. Hypothesis for the category approach and selected source chemicals ........................................ 16 3.1. Chemical identity and composition ............................................................................................ 19 3.2. Physical-chemical properties and other molecular descriptors ................................................... 19 3.3. In vivo data source chemicals ..................................................................................................... 20 3.4. Kinetics: Absorption, distribution, metabolism and excretion ................................................... 23 Development of PBPK models ...................................................................................................... 24 Sensitivity analyses, parameter uncertainty and model assumptions ............................................. 27 Reverse dosimetry to establish oral equivalent dose in human and mouse .................................... 28 Metabolism ..................................................................................................................................... 28 3.5. Mode/Mechanism of action or adverse outcome pathways (MOA/AOP); including experimental (NAM) data and in silico models – e.g. prediction of MIEs, key events ..................... 31 3.6. Chemical/biological interaction .................................................................................................. 32 3.7. Responses found in alternative assays (e.g., experimental (NAM) data, in silico) ..................... 32 3.8. Information obtained from other endpoints ................................................................................ 32 Profiling for DNA and protein binding, and skin sensitisation ...................................................... 32 Profiling for skin and eye irritation ................................................................................................ 33 3.9. Information on fate in the environment (hydrolysis, biodegradation) ........................................ 33 3.10. The route and duration of expected exposure. .......................................................................... 33 4. Data gap filling and Justification ................................................................................................... 34 4.1. Methodology ............................................................................................................................... 34 4.1.1. In vitro DART battery models ............................................................................................. 34 4.1.2. In silico models .................................................................................................................... 42 4.1.3. Classification and data analysis models ............................................................................... 43 4.2. Results ......................................................................................................................................... 44 4.2.1. In vitro DART battery models ............................................................................................. 45 4.2.2. In silico models .................................................................................................................... 53 4.2.3. Classification and data analysis models ............................................................................... 53 4.3. Justification ................................................................................................................................. 60 4.3.1. In vitro DART battery models ............................................................................................. 61 4.3.2. In silico models .................................................................................................................... 67 4.3.3. Classification and data analysis models ............................................................................... 68 16  ENV/JM/MONO(2020)21 Unclassified 3. Hypothesis for the category approach and selected source chemicals MHA is a 2-branched aliphatic carboxylic acid that lacks developmental toxicity data. In the above chapter, it is indicated that we consider other 2-branched aliphatic carboxylic acids with comparable chain length suitable source chemicals for reading across these health effects to MHA, provided that these chemicals themselves have adequate developmental toxicity data. It is assumed that by keeping as strict inclusion criteria for source chemicals only aliphatic carboxylic acids branched at position 2, and by selecting aliphatic carboxylic acids with comparable chain length, that both phys-chem properties, that drive toxicokinetic processes, and structural features, that may trigger toxicodynamic processes, the associated toxicokinetic and -dynamic properties for these source chemicals will be only minimally deviating. It is recognised that the category only contains one member with negative in vivo data; which introduces an uncertainty. Broadening the category criteria, however, to potentially increase the number of negative (and positive) members was considered, but not decided for, as this would certainly increase the uncertainty of the structural similarity basis of the read across. The source chemicals selected in this study allow the read across to MHA to be classified as a mere interpolation rather than extrapolation, as MHA appears to be a carboxylic acid with intermediate aliphatic chain lengths: see below table depicting target and source chemicals. The target chemical MHA has two aliphatic chains at position 2 with carbon chain lengths of resp. n=1, and n=4: see Table 1. ENV/JM/MONO(2020)21  17 Unclassified Table 1. Target and source chemicals of this read across study depicted with their chain lengths; also some structurally closely related control chemicals are depicted. CAS-RN Abbreviated name 591-80-0 PA 1185-39-3 DMPA 88-09-5 EBA 97-61-0 MPA 4536-23-6 MHA 149-57-5 EHA 99-66-1 VPA 31080-39-4 PHA 1575-72-0 4-ene-VPA Structure Chain lengths at position 2 3 / 0 / 0 3 / 1 / 1 2 / 2 / 0 3 / 1 / 0 4 / 1 / 0 4 / 2 / 0 3 / 3 / 0 5 / 3 / 0 3 / 3 / 0 Identity / function control control source in vitro target source source source control In vivo response / potency neg neg neg no data ? pos: + pos: +++ pos: +++ pos: ++ Structural similarity RDKit fingerprint (%) - - 59 83 100 82 69 77 - From the above table one can see that source chemicals directly on the left (MPA) and right hand side (EHA) of MHA are having only one carbon-atom chain length difference, and those more distant to MHA some may differ two carbons in length in one of the chains. The structural similarity of source chemicals to the target is provided as well, using RDKit fingerprints (Daylight-like topological fingerprint, and Tanimoto similarity metric) showing the closest similarity of MPA and EHA relative to MHA. In de data-matrix also results of some other similarity tools (FCFP4 Fingerprint [Circular fingerprint based on the Morgan algorithm and feature invariants], Layered Fingerprint [An experimental substructure-matching fingerprint], and RDKit Descriptors are shown: all showing the closest similarity of MPA and EHA. The control chemicals have less structural similarity to the target; their role in this read across assessment is to provide additional positive and negative in vitro effect profiles in the applied models. When this composition of the category is compared to the scenario options in the Read Across Assessment Framework (RAAF; ECHA, 2017), scenario number 4 seems most applicable here: a category approach, where the read across is based on different compounds that have the same type of effect(s), and where quantitative variations may exist in the strength of effect(s) observed among source substances. The RAAF indicates that prediction for the target chemical here could be based on an observed regular pattern within the category or on a worst-case approach. 18  ENV/JM/MONO(2020)21 Unclassified Table 1 also shows the in vivo response data of these chemicals: some appear negative in vivo (‘neg’), some are positive (‘pos’), though with different potency. This in vivo data is further described below in section 3.3. ENV/JM/MONO(2020)21  19 Unclassified An interesting observation is that the analogues on the right side of MHA are all positive in vivo, while of the two analogues on the left, the one with in vivo data, i.e. EBA, is negative. As indicated above we have also included ‘outlier’ structures because, though structurally less similar, they have relevant in vivo data. Because of the in vivo data of the category analogues, we have positioned the in vivo negative control chemicals PA, and DMPA on the left, and the in vivo positive control 4-ene-VPA on the right side of this category list. 3.1. Chemical identity and composition The chemical identity of the target chemical is described in section ’Target chemical(s) / category definition’ in chapter 2.2. The chemical identity of the source and control chemicals are described in the data-matrix (folders ‘Source and Target cmpds’, and ‘Physchem’). All chemicals investigated in this report are of analytical grade and purchased by one partner as central facility (IFADO, Dortmund, Germany), and from there distributed to the different collaborating laboratories. 3.2. Physical-chemical properties and other molecular descriptors The physico-chemical properties and other molecular descriptors of MHA and the source and control chemicals are provided in Table 1b (and listed in the data-matrix in Annex I in folder ‘Physchem’). From this it appears that from fully left (EBA) to fully right (PHA) within the category there is a trend for all these parameters (rounded values): molecular weights of the sources range from 116 to 172 g/mol, melting and boiling points increase from 15 to 59, and from 196 to 269 oC, respectively. LogPow ranges from 1,7 to 3,2 (experimental) or from 2,0 to 3,9 (predicted). Water solubility shows, for only some available, a decrease going from left (EBA) to right (PHA), i.e. from 18 to 0,3 g/L. The pKa constant values appear rather similar: experimentally derived values go from 4,7 (EBA) to 4.6 (VPA), while predicted values are similar for all 6 members in the category, i.e. 4.8, indicating these are all weak acids. Vapour pressures reduce when going from left to right (0.19 for EBA to 0,005 mmHg for PHA at room temperature. Henry’s Law constant was experimentally only determined for EHA, being 0,29 Pa m3/mol, while predicted values increase from 0,16 for EBA to 0,54 Pa m3/mol for PHA. The conclusion is that these compounds are all non-volatile, well to reasonable water soluble, will not bio-accumulate, and are weak acids of quite comparable strength. 20  ENV/JM/MONO(2020)21 Unclassified Table 1b. Physchem properties of target and source chemicals of this read across study CAS-RN Abbreviated name 591-80-0 PA 1185-39-3 DMPA 88-09-5 EBA 97-61-0 MPA 4536-23-6 MHA 149-57-5 EHA 99-66-1 VPA 31080-39-4 PHA 1575-72-0 4-ene-VPA Molecular weight 102,133 130,187 116,08 116,08 130,1 144,12 144,12 172,15 142,2 Melting point (oC; pred. EPISUITE) 14,76 24,61 15,24 15,24 26,62 37,72 37,72 59,15 36,44 Boiling point (oC; pred. EPISUITE) 187,75 207,77 195,8 195,8 215,45 234,2 234,2 268,98 232,83 logPow (exp. EPISUITE) 1,39 - 1,68 1,8 1,8 2,64 2,75 3,2 - logPow (pred. EPISUITE) 1,56 2,43 1,98 1,98 2,47 2,96 2,96 3,94 2,82 Water solubility (mg/L; exp.) 24000 (25°C) - 18000 (20°C) - - 2000 (20°C) 2000 (20°C) 275.6 (25°C) - pKa (exp.) - - 4.71 - - 4.7 4.6 - - pKa (pred. ACD/Percepta) 4,84 4,9 4,8 4,8 4,8 4,8 4,8 4,8 4,7 Vapour pressure (mm Hg; exp.) - - 0.188 (25°C) - - 0.03 0.0458 (25°C) 0.0048 - Henry's Law Constant (Pa m3/mol; pred. EPISUITE (bond method)) 0,130 0,229 0,162 0,172 0,229 0,304 0,304 0,535 0,226 Henry's Law Constant (Pa m3/mol; exp. EPISUITE) - - - - - 0,289 - - - experimental (exp.) or predicted (pred.) values and source models listed; ‘-‘: no data 3.3. In vivo data source chemicals As shown in Table 1 for 4 of the 5 source chemicals, and for the controls (named ‘outliers’ in report template) we do have in vivo data on developmental toxicity. For MPA we do not have this data but we included it in this study to explore the relationship between structure and developmental toxicity within this series of aliphatic carboxylic acids, as it may show support for a trend that may come out of this investigation. The in vivo data of this 4 source and 3 control chemicals all come from a standard teratogenicity protocol developed by Nau and colleagues to further study the induction of neural tube defects by VPA: this is the most apparent lesion in mouse, and also observed in humans exposed to VPA (Nau et al., 1981, 1986). In short, in this standard teratogenicity protocol female NMRI mice (28-32 g), housed under specific-pathogen-free conditions (diet and tapwater ad libitum), were mated with males for periods of 2 hr (8 to 10 AM), and a subsequent 24 hr examined for vaginal plugs; if found, this was designated as Day 0 of gestation. Groups of 15 pregnant animals were exposed to test compounds at day 8 of gestation (by single sc injection) at 400 or 600 mg/kg bw in water (10 ml/kg bw), and were examined on day 18 of gestation for implantation sites, and each live foetus was individually weighed and inspected for the presence of the neural tube defect exencephaly. Effects on visceral and skeletal systems were not examined. The outlier substances DMPA, 4-ene-VPA, and PA were tested in this same teratogenicity protocol (Nau et al., 1981, 1986, 1991; Eikel et al., 2006; Courage-Maguire et al., 1997; Dearden et al., 2013). These studies showed that aliphatic carboxylic acids have different potencies of inducing neural tube ENV/JM/MONO(2020)21  21 Unclassified defects, and that some of them even were inactive in this respect. Table 2 shows the potency of induction of exencephaly by the aliphatic carboxylic acids: of the source chemicals, EBA is negative, while EHA, VPA and PHA are positive, though at different potencies, as indicated below. Table 2. Teratogenic potency grading in NMRI mouse exencephaly model (Eikel et al., 2006) Teratogenic potency Dose range (mmol/kg bw) Exencephaly rate (%) Description 0 > 3.0 0 No teratogenic potency detectable + 2.0 - 3.0 1-5 Low teratogenic potency ++ 2.0 - 3.0 5 - 25 Lower teratogenic potency than VPA +++ 2.0 - 3.0 25 - 60 Equal teratogenic potency to VPA ++++ 1.0 - 2.0 40 - 60 Higher teratogenic potency than VPA The above NMRI mouse study results are included in the data-matrix in Annex I under the header ‘target endpoint 2’. By comparing structure and potency in Table 1, the data suggest that side-chain length is related to potency: PHA, having the longest side chains within the category has the highest potency, while EBA with the smallest side chains is inactive. EHA, with in between side chain length has the lowest potency. For MHA, and MPA, that are in between EHA, and EBA with regard to chain length, we have no in vivo data, and it is, therefore, unclear whether they possess any potential for developmental toxicity in vivo: this read across case study is to fill this gap. There is no clear explanation for this structure-related potency, other than that chain-length probably is a factor in affinity for binding to the critical target that leads to the in vivo observed effects. Only for VPA and EHA other in vivo developmental toxicity study data were available. For VPA, in fact there is a wealth of data available. A series of developmental toxicity studies in mice in which neurodevelopmental parameters were included are described (Paulson et al., 1985; Padmanabhan, 1996; Sonoda et al., 1990; Turner et al., 1990). In these studies, at dose levels ranging from 200-600 mg/kg bw/day, neurodevelopmental effects (exencephaly, reduced head size, neural tube closure defects, microscopically disorganisation of neuroepithelium) were observed. The lowest LOAEL that was described in mice was 200 mg/kg bw/day (Padmanabhan et al., 1996). The lowest NOAEL that was described in mice was <200 mg/kg bw/day. At this concentration, exencephaly and neural tube closure effects were observed. In addition, in the study of Paulson et al. (1985), and Sonoda et al. (1990), exencephaly was observed at 340 and 560 mg/kg bw/day and 600 mg/kg bw/day, respectively. In general, at concentrations that induced neurodevelopmental effects, a wide range of other developmental effects (among others foetal death, resorptions, growth retardations, craniofacial and skeletal malformations, cardiovascular effects, urogenital anomalies) as well as maternal toxicity (reduced food/water intake, reduced weights) were observed. In addition to the neurodevelopmental toxicity studies, a series of developmental toxicity studies was described in which no neurodevelopmental parameters were studied (ECHA website). Overall, the lowest LOAEL for ‘general’ developmental toxic effects in mice was 160 mg/kg bw/day. There were two rat studies in which neurodevelopment parameters of VPA were studied (Ong et al., 1983; Frisch et al., 2009). In the study of Ong et al., no neurodevelopmental effects were observed at concentrations as high as 600 mg/kg bw/day whereas in the study of Frisch et al., functional and structural effects were observed at 720 mg/kg bw/day, which is considered as the lowest LOAEL for neurodevelopmental effects of VPA in rats. In 22  ENV/JM/MONO(2020)21 Unclassified addition, a series of developmental toxicity studies was performed in which no neurodevelopmental parameters were included. Overall, general developmental effects of VPA in rats were observed at 100 mg/kg bw/day and higher. For VPA one rabbit study was reported that investigated neurodevelopment parameters. In this study, the NOAEL for neurodevelopmental toxicity was >350 mg/kg bw/day, whereas the NOAEL for developmental toxicity was 50 mg/kg bw/day (Petrere et al., 1986). No maternal toxicity was reported. In a prenatal developmental toxicity study with New Zealand White rabbits (without neurodevelopmental endpoints), the NOAEL for prenatal developmental toxicity was 150 mg/kg bw/day (ECHA website). At higher concentrations of 250 and 350 mg/kg bw/day, the incidence of resorptions was increased and foetal weight was decreased. Hendrickx et al. (1988) reported a study with Rhesus monkeys treated with sodium valproate during gestation days 21-50. The NOAEL for neurodevelopmental toxicity was 75 mg/kg bw/day, at >100 mg/kg bw/day, the incidence of foetuses with a reduced head circumference was increased. Other developmental toxic effects were observed at concentrations higher than 20 mg/kg bw/day (craniofacial skeletal effects, growth retardation, embryoand foetal mortality). At all concentration tested, maternal toxicity was observed. For EHA, a series of rat pre-natal developmental toxicity studies and a series of rat reproductive toxicity studies, including an extended-one generation reproductive toxicity study (EOGRTS) according to OECD guideline 443, was available (ECHA website). In one of the pre-natal developmental toxicity studies, dilation of the lateral ventricles of the brain was observed in foetuses of dams treated with 500 mg/kg bw/day, which might be considered as a neurodevelopmental toxic effect. At this concentration, other developmental effects (reduced foetal weight, skeletal effects) as well as maternal toxic effects were observed. In the EOGRTS study, no neurodevelopmental effects were observed at concentrations up to and including 800 mg/kg bw/day (ECHA website). At this concentration, no other reproductiveand developmental toxic effects were observed whereas slight maternal toxicity (body weight, food consumption, kidney and liver pathology) was observed. In two other reproductiveand developmental toxicity studies neurodevelopmental parameters were not included but developmental toxicity (effects on pup weights, litter size, physical development) was observed at dose levels as low as 100 mg/kg bw/day (ECHA website; Pennanen et al., 1992; Bui et al., 1998). In a prenatal developmental toxicity study with New Zealand White rabbits, the animals were treated with 0-250 mg/kg bw/day of EHA (ECHA website). No developmental toxic effects were observed, whereas maternal toxicity was observed at dose levels of 25 mg/kg bw/day and higher. Relevant NOAELs of the above studies are included in the data-matrix in Annex I under the header ‘target endpoint 1’. Only for VPA also human data is available indicating that the use of valproate during pregnancy is associated with foetal death and major congenital malformations. High dosages (>1000 mg/d) were associated with significant greater risks of congenital malformations than lower dosages (Diav-Citrin et al., 2008; Jentink et al., 2010). Malformations associated with maternal valproate use include among others neural tube defects (particularly spina bifida), cerebral, craniofacial, skeletal, cardiovascular and urogenital effects (hypospadias). The mouse model appears a suitable animal model reflecting similar findings such as among others neurodevelopmental effects (exencephaly, ENV/JM/MONO(2020)21  23 Unclassified reduced head size, neural tube closure defects) at dose levels ranging from 200-600 mg/kg bw/day (Nau et al., 1991; Turner et al., 1990; Sonoda et al., 1990). As a final remark to this study data, it is noted that some of the aliphatic carboxylic acids have asymmetric carbon atoms, and thus may exist as racemates: i.e. MPA, MHA, EHA, PHA, and 4-ene-VPA. The in vivo data in the NMRI mouse model were generated with the racemates, as far as data on this was retrievable; therefore, all in vitro tests described in this report have been performed with racemates of these chemicals as well. 3.4. Kinetics: Absorption, distribution, metabolism and excretion In the folder ‘ADME-Toxicokinetics’ in the data-matrix in Annex I the ADME characteristics ‘unbound fraction’ (fu), hepatic intrinsic and in vivo clearance (CLint and CL, respectively), as well as total and unbound steady state volume of distribution (Vss and Vu,ss, respectively) for the category members (target and source chemicals) and controls (named ‘outliers’ in report template) are listed. The models from which these values were derived will be described below. Physiologically based pharmacokinetic modelling Physiologically-based pharmacokinetic (PBPK) modelling and simulation was applied as part of the read-across with two primary aims: 1) To predict the pharmacokinetics of source and target compounds in the read-across based on in vitro to in vivo extrapolation (IVIVE) PBPK and allometric scaling. 2) Determine the concentrations of compounds in target tissues, namely here the foetal-placental compartment, and through reverse dosimetry translate in vitro toxicity assay data to an oral equivalent dose. PBPK models for read-across compounds were developed in the human and animal Simcyp Simulators (V17r1, Certara UK Ltd. Simcyp Division, Sheffield, UK; www.simcyp.com) production of each version of the Simcyp simulator has been described in detail (Jamei et al., 2013). In developing these PBPK models for the read-across study, the following aspects were considered as suggested in the WHO PBPK guidance (WHO publication Harmonisation Project Document No. 9. Characterisation and Application of Physiologically based PharmacoKinetic Models in Risk Assessment): 1) The source or target compound was assumed to be the toxic moiety (i.e. plasma and tissue levels of formed metabolites were not routinely considered in the PBPK models). 2) Metabolism is thought to be the major clearance pathway of the compounds in this read-across study and the metabolic clearance in humans was predicted using an IVIVE approach, scaling in vitro intrinsic clearance determined in human primary hepatocytes. 3) The physiology (i.e. tissue weights and blood flow rates) of the species of interest were the default values in the Simcyp human and animal simulators (V17r1, Certara UK Ltd. Simcyp Division, Sheffield, UK). 24  ENV/JM/MONO(2020)21 Unclassified Development of PBPK models Where sufficient input data was available, PBPK models were developed for each of the source and target compounds; modelled compounds are EHA, MPA, MHA, EHA, and VPA. The PBPK model development work-flow is summarised in Figure 1 below and exemplar model input parameters are shown in Table 3; the parameters for each compound specific PBPK model in each of the species of interest are tabulated in Annex III. ENV/JM/MONO(2020)21  25 Unclassified Figure 1. Human and model species physiologically-based pharmacokinetic model development work-flow. 32  ENV/JM/MONO(2020)21 Unclassified for these analogues: analogue accumulation within the foetus, oxidative stress, and folate antagonism are other possible suggested mechanisms (Lloyd, 2013). With regard to HDAC inhibition as molecular initiating event, two postulated AOPs are submitted to AOPWiki, one for craniofacial development (#274), and one for neural tube defects (#275), both having HDAC inhibition as MIE. Linking in vitro assays to MIE and /or KE of these AOPs is not quite justified yet. Nonetheless, we have monitored histone deacetylase (HDAC) inhibition by the aliphatic carboxylic acids investigated in this study in our experimental models to verify the correlation between HDAC inhibition and exencephaly induction. Because of this state of knowledge of neural tube defects underlying mode(s) of action (and developmental toxicity in general), and the state of knowledge of the applicability domains of the applied models, an attempt to link or select assays on this basis for this case study was considered not opportune. The results are presented and discussed in chapters 4. and 5. , respectively. 3.6. Chemical/biological interaction Section 3.8 below shows the OECD QSAR Toolbox profiler analysis results for the aliphatic carboxylic acids investigated in this study, that characterise the expected chemical-biological interactions for these chemical structures. Quite explicit characteristics are DNA and protein binding potential, but also potential for sensitisation, and irritation are reflecting expected chemical-biological interactions. From this analysis, it can be concluded that there are no great differences expected in the biological interactions of these category chemicals, which is another argument in support of their grouping. 3.7. Responses found in alternative assays (e.g., experimental (NAM) data, in silico ) The in silico responses are discussed in sections 3.6 and 3.8. For some of the chemicals investigated in this study quite some data in alternative assays is available in public literature. However, this data simply is too much to describe and discuss here. Moreover, a more realistic situation for most target chemicals is that this data is actually non-existent. Therefore, we simulate this case here, and only present and discuss the data we generate in this study: this is described in chapters 4. and 5. , respectively. 3.8. Information obtained from other endpoints To further characterise, the biological similarity of the category members profilers of the OECD QSAR toolbox were used to gather information on DNA and protein binding, skin sensitisation, and skin and eye irritation. Profiling for DNA and protein binding, and skin sensitisation None of the grouped compounds has an alert for DNA and protein binding, and skin sensitisation. This is taken as similar behaviour of all compounds grouped in this study (Table 5). It is recognised though that the applied profilers flag the “presence of alerts” and that the absence of known alerts should not be interpreted as a lack of toxicity. As such, “negative predictions” should be taken cautiously. ENV/JM/MONO(2020)21  33 Unclassified Table 5. OECD Toolbox profiler predictions for DNA and protein binding, and on skin and eye irritation CAS-RN Abbreviated name 88-09-5 EBA 97-61-0 MPA 4536-23-6 MHA 149-57-5 EHA 99-66-1 VPA 31080-39-4 PHA Chromosomal aberrations OASIS - - - - - - DNA binding OASIS OECD - - - - - - - - - - - - Sensitisation GHS OASIS h-CLAT - - - - - - - - - - - - - - - - - - Protein binding OASIS OECD CYS potency (DPRA 13%) GHS potency LYS potency (DPRA 13%) - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - Skin irritation excl. BfR rules incl. BfR rules u + u + u + u + u + Group C oa Eye irritation excl. BfR rules incl. BfR rules u oa u oa u oa u oa u oa Group C oa - : no alerts, not reactive or unreactive; + : alert or reactive; oa: out of applicability domain Profiling for skin and eye irritation All compounds are flagged as irritating to skin, because they contain an aliphatic acid. This corresponds well to the pKa values of about 4.5 for all compounds. This alert is not indicating any dissimilarities between the target and source compounds. Only PHA flags an alert for the exclusion rules for eye/skin irritation (i.e. this molecule should not be an irritant). It must be noted that such an alert is based on an estimated melting point and it is known that QSPR methods are characterised by prediction errors well in excess of the error of experimental measurements. Therefore, the uncertainty attached to the pertinence of such an alert is rather high (Dearden et al., 2013). From this data, it can be concluded that MHA and its analogues on both sides have no predicted dissimilarities in reactivity profile. All this data is also included in the data-matrix of Annex I (folder ‘In chemico’). 3.9. Information on fate in the environment (hydrolysis, biodegradation) Not applicable for the read-across case described here. 3.10. The route and duration of expected exposure. Not applicable for the read-across case described here. 34  ENV/JM/MONO(2020)21 Unclassified 4. Data gap filling and Justification In this chapter the in vitro and in silico methods used and results generated with these methods within the context of this read across approach for MHA will be shortly described. Section ‘Methodology’ will name and describe the methods, as well as their suitability regarding the defined purpose of this study. Further details on these methods is described in referenced annexes at the back of this report. The section ‘Results’ will describe the results that have been generated with these described methods. Further details on these results is described in referenced annexes at the back of this report. The section ‘Justification’ will describe how the generated data are interpreted within the context of this study of assessing the developmental properties of MHA via reading across from structural analogues. This section will also evaluate the uncertainties of the data generated. 4.1. Methodology In this section we will first of all describe the battery of experimental models applied in this study, followed by in silico models for prediction of biological activity, in silico models of toxicokinetic behaviour, and finally models for interpretation and classification of the results. 4.1.1. In vitro DART battery models The experimental models applied are the ZET, the ZET reporter, the mEST and UKN1 models, and the CALUX reporter models. The suitability of the model and the applied scoring approaches will be outlined as well. ZET Suitability The zebrafish embryo (Danio rerio) holds great promise as a model for developmental toxicity testing. The model is emerging as a candidate lower vertebrate animal model capable of filling the gap between high(er) throughput in vitro cellular assays and conventional preclinical animal testing. The embryonic/larval zebrafish model offers an intact whole animal model, providing the possibility to obtain information on dysmorphic effects comparable to the ones that may occur in rodent in vivo studies, with many of the advantages of in vitro systems, making it a suitable model organism for medium/high throughput screening (Ali et al., 2011). The model covers a broad developmental period (from fertilisation to hatching and early larval stage), including processes as gastrulation and neurulation and the development of major organ systems. It is relatively easy to monitor the morphology and evaluate effects on developmental since the embryos are transparent. In addition, for many years the zebrafish is a model to study (basic) embryology, so the development of the embryo is well defined (Kari et al., 2007). During these early developmental stages, the zebrafish embryos are not defined as protected under European legislation for animal welfare, as they are still yolk-dependent and thus non-free-feeding. In addition, fertilisation and development of the zebrafish occur ENV/JM/MONO(2020)21  35 Unclassified outside the body (oviparous), there is no need to sacrifice parental animals to obtain the embryos. Method Basically the method followed is that described in OECD guideline 236 and DB-ALM Fish Embryo Test 2013. Details are described in Annex II. As deviation from this OECD 236 guideline the following is noted: as described (Braunbeck et al., 2015; Braunbeck and Lammer, 2006; Embry et al., 2010; Lammer et al., 2009) embryos are raised (and exposed) until an age of 96 hpf. However, based on the wording of current EU animal welfare legislation (Directive 2010/63/EU of the European Parliament and of the counsel on the protection of animals used for scientific purposes), in cases of inconclusive observations, exposure may be extended to 120 hpf (Strähle et al., 2012). In order to classify specific endpoints of the test substances, this extension has been performed in this study. Calculation of EC/LC values and generating of graphs: EC and LC values were calculated with ToxRat Prof. Vers. 2.10. It also generated graphs for every observed effect, however, for a clear presentation the graphs of Annex II were created and edited with SigmaPlot 13.0. For the calculation of the EC/LC values and for the data of the graphs ToxRat Prof. Vers. 2.10, the total amount of affected embryos of every test concentration at every time point was used. HDAC inhibition assay: the standard protocol for HDAC inhibition measurements applied in this project did not work out for zebrafish embryo’s; one explanation may be that different HDAC isozymes are involved in this species. Scoring After the end of the test at 120 hpf, the number of embryos exhibiting specific effects were scored. Subsequently, from each dose effect graph EC10 values were calculated using ToxRat Professional version 2.10 and SigmaPlot 13.0 software. The EC10 was defined as the concentration at which there was a 10% increase of the incidence of a monitored effect, relative to control. Subsequently, the calculated EC10 values of all compounds were compared for ranking the compounds for their potency to induce the monitored effects. ZET Reporter assay Suitability The suitability of the ZET has already been described above. This reporter assay focusses on one specific aspect of the developing zebrafish larvae: the deterioration of facial cartilage structures, called cranio-facial deformation. Method Basically, the method followed is that described in OECD guideline 236 and DB-ALM Fish Embryo Test 2013. Test solutions are refreshed every 24 hours. The measurement takes place at 120 hpf. The method relies on a transgenic zebrafish collagen reporter line, which expresses the fluorescent protein mCherry under the control of the collagen 2 a (col2a) promoter (Hammond and Schulte-Merker, 2009), and on the uniform positioning of larvae for imaging afforded by the Vertebrate Automated Screening Technology (VAST) bioimaging platform (Union Biometrica, USA). The basic raw data is fluorescent 36  ENV/JM/MONO(2020)21 Unclassified ventral-view images of zebrafish larvae at 5 days post fertilisation. The images are analysed by measuring the angle formed by the ceratohyal bone (CHA), a characteristic v-shaped structure clearly visible in ventral views of 5 DPF larvae, as the increase in this angle provides a quantifiable proxy for the deterioration of the morphological development of facial cartilage structures. For more details, see Annex II. HDAC inhibition assay: the standard protocol for HDAC inhibition measurements (applied to mEST, and UKN1 cells) did not work out for zebrafish embryo’s; one explanation may be that different HDAC isozymes are involved in this species. Scoring After the end of the test at 120 hpf, the embryos are monitored for the CHA value. Thus, the basic analysis unit is degrees of angle. Data normalisation is based on the normalisation to mean measurements in sibling controls, defined as 0% response, while 100%, the complete deterioration of the ceratohyal angle, is defined as 180° where the two halves of the ceratohyal forms a straight line (see Annex II). The EC10 was defined as the concentration at which there was a 10% deviation of this CHA value relative to control. Subsequently, the calculated EC10 values of all compounds were compared for ranking the compounds for their potency to induce the monitored effects. mEST Suitability The mouse embryonic stem cell test (mEST) is used as an in vitro model for the screening of embryotoxicity, based on a blastocyst-derived permanent embryonic mouse D3 cell line (ESC), derived from mouse 129 strains and was validated by the European Centre for the Validation of Alternative Methods (ECVAM). During this formal validation, the results obtained with the mEST concordances between the embryotoxic potential derived from the in vitro data and from the in vivo data were good, and were reproducible, both within and among different laboratories that performed the test. The ability of the test to discriminate a diverse group of chemicals, being strongly, weakly embryotoxic or non-embryotoxic, has been clearly demonstrated, and a predictivity of 100% was obtained with strongly embryotoxic chemicals (Genschow et al. 2002, 2004). Method The assay execution and data evaluation was performed according to the process described in the DB-ALM protocol provided in Annex II. The assay principle is based on the assessment of chemical-induced cytotoxicity and the inhibition of differentiation from mouse embryonic stem cells into cardiomyocytes during a ten days substance treatment. The assay is defined through the three following endpoints: Inhibition of growth (cytotoxicity) of 3T3 fibroblast cells, which represent differentiated cells, and growth inhibition of undifferentiated ESC cells after 10 days of treatment. This is determined by the use of dehydrogenase enzymes present in the intact mitochondria of living cells to convert yellow soluble substrate 3-(4,5-dimethythiazol-2-yl)-2,5-diphenyl Tetrazolium bromide (MTT), into a dark blue insoluble formazan product, which gets sequestered within the cells and is detected quantitatively at 570 nm using a absorbance reader, after solubilizing the cell membrane. For this IC10 values are derived, i.e. inhibitory concentration of cell viability at 10% of the maximum dose response. ENV/JM/MONO(2020)21  37 Unclassified A third endpoint is the inhibition of differentiation of ESCs (embryonic stem cells) into cardiomyocytes after 10 days of treatment. The beating of these cells is evaluated by microscopy. Here the ID10 is derived, i.e. inhibitory concentration of differentiation at 10% maximum dose response. HDAC inhibition assay: short description (see Annex II for details): ca 2.5 million D3 cells were lysed and the lysate containing the HDAC enzymes was then centrifuged for 10 mins at 10,000g. Supernatant was collected and diluted 1 in 6 with assay buffer. Reaction plate setup with the inhibitors (VPA analogues) and positive control (TSA) and incubated for 1hr at 37 degrees Reaction stopped with a 1 μM TSA solution and then incubated for 15 minutes. Thereafter, the plate is read in the Tecan infinite M200 PRO reader at 360nm excitation and 460nm emission wavelength. Scoring From the differentiation assay and the cytotoxicity assays, half maximum dose response values were calculated (ID50, and IC50 values resp.) and entered into a statistical evaluation developed from the modified prediction model used by Scholz et al. (1999), from which a prediction score was obtained to characterise chemicals as negative (D12_3 < 0.5), or positive (D12_3 > 0.6) with regard to embryotoxicity: see Annex II for details. Besides, relative ID10 values of chemicals were evaluated to assess their relative differentiation inhibition potential. UKN1 Suitability UKN1 model is a human neural embryonic stem cell test, corresponding to weeks 3-4 of human foetogenesis, to assess neurodevelopmental toxicity induction by chemical or drug exposure, and established with positive controls, classifiers for compound classes, biomarkers, extensive transcriptomics background data and epigenetic endpoints (Krug et al. 2013; Waldman et al 2014). Method UKN1 models the process of neural induction and patterning. In order to model this process human induced pluripotent stem cells (iPSC) are differentiated by dual SMAD inhibition for 6 days (Balmer, 2012; Krug, 2013). The differentiation protocol is based on previously published protocol (Chambers, 2009) and a detailed test method description and SOP is provided in Annex II. The differentiated cells on day 6 correspond to neuroectodermal progenitors (NEP) as present during neural plate and neural tube formation. Therefore, disturbances of differentiation and patterning can be modelled by the test method. To this end, the differentiating cells are exposed to the compounds during the whole process of differentiation from day 0 to day 6. Neural tube defects are thought to be induced by wrong differentiation tracks during neurulation and in chicken; it has been shown that TSAinduced neural tube defects are due to induction of neural crest cell fade instead of NEP (Murko, 2010; Murko, 2013). Therefore, the idea of the present test method is that we may detect neural tube defect inducing chemical by following gene expression changes. This includes e.g. spina bifida, 38  ENV/JM/MONO(2020)21 Unclassified anencephaly or encephalocele. The scientific rational of the test method is based on the hypothesis that gene expression alterations induce a wrong differentiation of NEP cells or a changed differentiation track. For example if the iPSC cells do not differentiate into epithelial cells but in mesenchymal cells, the neural tube cannot be closed and severe congenital malformations may occur. The UKN1 test method is mainly related to AOP #275 (implemented in the frame of the EUToxRisk project): “HDAC inhibition leads to neural tube defects” and measures two key events of this AOP. The first KE is “alterations of gene expression” and this is measured on day 6 by real time quantitative PCR (RT-qPCR) in order to follow the expression of 6 key player genes having a major role in the maintenance of pluripotency (OCT4, NANOG), neural induction (PAX6, OTX2) and a neural crest marker (TFAP2) which indicates wrong differentiation. A compound is counted as hit if PAX6 and OTX2 are downregulated and TFAP2 is upregulated. The second KE measured by UKN1 is “altered differentiation”. To this end, we investigate if the NEP can be differentiated into the next developmental stage, which is represented in vitro by the so-called rosettes. The compound is removed on day 6 and the cells get further differentiated for another 8 days. On day 11, the cells are seeded in 96-well plates and are stained with two markers (ZO1, GM130) that mirror the specific morphology of the rosettes (Waldmann, 2014). The cells are imaged by an automated microscope and the number of rosettes per well is determined by an algorithm that segments the rosettes and counts the number of rosettes per well. If the number of rosettes is reduced the compound is counted as a hit. If both KE are affected by a compound we conclude on a DNT potential of the compound. To determine the testing concentration for the measuring of KE1 and KE2 a viability based concentration response curve is measured and the EC10 is determined by a 4 parameter model fit. For the measuring of KE1 and KE2 (as described above) the ½ x EC10, EC10 and 2x EC10 was used. HDAC inhibition assay: short description (see Annex II for details): ca 2.5 million D3 cells were lysed and the lysate containing the HDAC enzymes was then centrifuged for 10 mins at 10,000g. Supernatant was collected and diluted 1 in 6 with assay buffer. Reaction plate setup with the inhibitors (VPA analogues) and positive control (TSA) and incubated for 1hr at 37 degrees. Reaction stopped with a solution containing 1 μM TSA trypsin, and then incubated for 15 minutes. Thereafter, the plate is read in the Tecan infinite M200 PRO reader at 360nm excitation and 460nm emission wavelength. Scoring The prediction model to classify the compound as positive or negative is as follows: 1) PAX6 and OTX2 gene expression must be down-regulated (compared to control) and AP2 up-regulated. 2) The means of the fold change values are calculated (with down-regulated values are multiplied with -1) = gene expression score 3) The rosettes formation are transformed from % of control (which is set to 100%) to a scale from 1 (no inhibition of rosettes formation) to 20 (no rosettes are detectable) = RoFo score. The rational behind this scaling is that in this data set the maximal gene regulation was 17.5 fold. Therefore, as the RoFo score should ENV/JM/MONO(2020)21  39 Unclassified be weighted in the same range as gene expression, we have chosen 20 for maximal inhibition of rosettes formation. 4) The mean of the gene expression score and the RoFo score is then calculated = UKN1 score The average UKN1 score of positive controls (known in vivo DNT effect) was 15.7 (SD 2.6) and 4.7 (SD 3.4) for the negatives. Based on SD, the following 3 classes were defined: UKN1 score classification < 7 no hit (='0' in the data matrix) 7 – 13 no or weak hit (= 0.5 in the data matrix) > 13 hit (= 1 in the data matrix) CALUX reporters Suitability The CALUX panel is based on the U2-OS osteosarcoma cell line. This cell line does not endogenously express most nuclear receptors and/or metabolic enzymes. Therefore, this cell line is ideally suited to study the interaction of a compound with a specific receptor or pathway, without interference of metabolism or receptor cross-talk. This makes interpretation of the results relatively straightforward, and allows comparison of compounds on the level of MIEs. The CALUX panel consist of 26 cell-based reporter gene assays. Each cell line measures the activation or inhibition of one specific nuclear hormone receptor or cell signalling pathway (see below Figure 4). In general CALUX assays focus on molecular initiating and early key events that are the primary target of toxicants (van der Burg et al.,2013; Becker et al.,2015), and involved in cellular events key in cell growth and organismal development, while deregulation can cause developmental disorders and diseases like cancer (NRC, 2000; Hanahan and Weinberg, 2011). The panel is unique, in that it contains highly selective human cell based assays having a particularly strong coverage of assays for nuclear hormone receptors and pathways relevant for developmentaland reproductive toxicity (van der Burg et al.,2013; Sonneveld et al.,2005, 2006, 2011; Piersma et al.,2013; van der Burg et al.,2015a; Lewin et al.,2015), including a broader coverage within developmental and reproductive toxicity, but also covering additional relevant nuclear receptors (glucocorticoid-, retinoid-, peroxisome proliferator activated-, dioxin-, pregnane X-, liver X-, and progesterone receptors). In addition key pathways assays have been added including reporter gene assays for AP1 (fos/jun complexes, activated via MAPK pathways), NF-kappaB, TCF, nrf-2, p21 and p53 (van der Burg et al.,2013; Piersma et al.,2013; van der Linden et al.,2014; Gijsbers et al.,2013). Clearly most of these pathways are involved in quite generic responses including developmental and reproductive toxicity. The assay panel has been used to predict developmental toxicity successfully, either alone or in combination with other assays (Piersma et al., 2013; Van der Burg et al., 2015a,b). Also, a control cell line, called cytotox CALUX has been generated which constitutively expresses the same luciferase gene under control of the same expression plasmid which is used to generate reporter gene assays for steroid receptors (i.e. pSG5). This assay has been proved to be a very suitable control for high throughput screening (HTS). All CALUX assays have now been automated in high throughput format. Data storage and analysis has been set up and more than 600 compounds have been screened in the CALUX 40  ENV/JM/MONO(2020)21 Unclassified panel. Finally, the CALUX HTS panel endogenously expresses little metabolic activity but can be run with and without S9 metabolic fractions, allow assessment of involvement of metabolism in chemical effects (van Vugt-Lussenburg et al., 2018). The assay results can be used to identify compounds activating similar MIE’s, and is therefore well suited for read across studies. Furthermore, the results provide clues for the MoA of the different compounds. Therefore, a direct link with an AOP (if available) can be made. Role of HDAC inhibition on activation profile CALUX reporters HDAC is a positive regulator of gene expression in multiple settings. HDACs are recruited by a variety of transcription factor corepressor complexes and are believed to repress transcription by reducing the level of acetylation of core histones, thereby altering chromatin structure. HDAC inhibition, therefore, will result in an increase in activity of several reporters and cell signalling pathways. The CALUX activity profile of chemicals could also be explained by their activity as HDAC inhibitors, rather than direct interaction with the pathways (Piersma et al. 2013). Method The CALUX panel of 26 cell based reporter gene assays are depicted in the table below. Table 6. The CALUX panel used in this Case Study. Cell line Endpoint Cell line Endpoint Cytotox CALUX Cytotoxicity PPARg CALUX Peroxisome proliferator receptor agonists ERa CALUX (ago/anta) Estrogen receptor (ant)agonists AhR CALUX Aryl hydrocarbon receptor agonists AR CALUX (ago/anta) Androgen receptor (ant)agonists Hif1a CALUX Chemical hypoxia response PR CALUX (ago/anta) Progesterone receptor (ant)agonists TCF CALUX Wnt/TCF pathway activation GR CALUX (ago/anta) Glucocorticoid receptor (ant)agonists AP1 CALUX AP1 pathway activation/cell cycle control TRb CALUX (ago/anta) Thyroid receptor (ant)agonists ESRE CALUX Endoplasmic reticulum stress RAR CALUX Retinoid acid receptor agonists NFkB CALUX Activation of NF-kB pathway (immune response) LXR CALUX Liver X receptor agonists Nrf2 CALUX Oxidative stress PXR CALUX Pregnane X receptor agonists p21 CALUX Transcription of p21 inhibitor of cell cycle progression PPARa CALUX Peroxisome proliferator receptor agonists p53 GENTOX CALUX P53-dependent pathway activation/genotoxicity PPARd CALUX Peroxisome proliferator receptor agonists CALUX assay principle used for this read across study is shown below in Figure 5. For the current study, DBALM Protocol n° 197: Automated CALUX reporter gene assay procedure was used; this method is described in the combined OECD 211/ DB-ALM template (Annex I). This involves exposure to the study compounds in a concentration range of 1E-3M (1 mM, as maximum tested concentration) to 1E-9M (1 nM, as lowest tested concentration) of 26 different CALUX assays, carried out by a liquid handling robot. ENV/JM/MONO(2020)21  41 Unclassified Figure 5. CALUX assay principle. In 2005 we already developed a panel of mechanism-based CALUX assays to assess hormonal activity of compounds (Sonneveld et al., 2005), a panel which has shown to be highly predictive for such activities in experimental animals (Sonneveld et al., 2006, 2011). Starting in 2005, several of these assays have been successfully engaged in extensive validation exercises as alternatives to animal experiments via EURL-ECVAM, OECD, and others (OECD, 2009, 2013a, 2013b, van der Burg et al., 2010a, 2010b). All assays are validated in house and at client laboratories. Further formal validations have been initiated in the area of thyroid disruption, metabolic activation of endocrine assays, while SOPs, including one on the assay panel automation, have been submitted to the EURL-ECVAM database of alternative methods DB-ALM. Scoring Reported values are lowest effect concentrations (LEC) in Log(M). Depending on the assay, LECs are defined as (Figure 6, next page): - the concentration where the test compound causes an activation/agonist effect equal to 10% of the maximum effect elicited by the test’s reference compound (PC10, Figure 6A); - the concentration where the test compound causes an antagonist effect equal to 20% of the maximum effect elicited by the test’s reference compound (PC20, Figure 6B); - the concentration where the test compound elicits pathway activation 1.5-fold above background (FI 1.5, Figure 6C) 48  ENV/JM/MONO(2020)21 Unclassified Table 9. Nominal and total embryo EC10 values (in μM) in ZET CHA Reporter assay. Abbreviated name PA DMPA EBA MPA MHA EHA VPA PHA 4-e-VPA CHA effect nominal >2000 1,2 174 8562 194 >80 12 2 1 7 total embryo >500 44 143 49 >20 2 0.5 0.1 1 1) '>'means: not detectable up to this highest tested dose; 2) recently 4 , it was noticed that for this dose the medium pH value will have dropped to below pH 6.5 (the OECD Guideline 236 indicated border), and the depicted total embryo EC10 value will be an underestimation of the true embryo value. Corresponding mouse and human OED values Mouse and human OED values for this CHA effect are depicted below in Table 10 for the five compounds for which PBPK models were available. The OED values correspond to the indicated in vitro specified effect concentrations, i.e. it is the external dose needed to achieve such an internal equipotent target concentration. Please, note that no CHA effect was observed for MHA up to the highest tested dose (indicated by the ‘>’sign), which is a relatively low dose as at higher dose levels survival was substantially reduced. The associated OED values were generated just for comparison purposes. Table 10. ZET reporter total embryo EC10 values (in μM), and corresponding OED values (μmol/kg) in mouse and human. Abbreviated name EBA MPA MHA EHA VPA CHA effect ZET total embryo 143 2 49 >20 1 2 0.5 Mouse OED range 1154-2472 388-775 >44-71 7-15 0.14-0.28 Human OED range 1034-1077 370-388 >88-92 5-7 0.73-0.94 1) '>'means: not detectable up to this highest tested dose; 2) recently4, it was noticed that for this dose the medium pH value will have dropped to below pH 6.5 (the OECD Guideline 236 indicated border), and the depicted total embryo EC10 value will be an underestimation of the true embryo value. From this table one can see that the OED values for MHA and EHA are somewhat in between EBA, and MPA on the one hand (highest OED values), and VPA on the other hand (lowest OED value): EHA clearly is closer to VPA, while MHA is closer to EBA, and MPA; how close this is cannot be determined, as it was not observed at the indicated (highest tested) dose in this model. mEST Full dose response data description and standard analysis results are provided in Annex II, while relevant ID10 values for D3 cells, and IC10 values for cytotoxicity of D3 and 3T3 cells, as well as HDAC inhibition and OED data are depicted in the data-matrix (Annex I). In vitro medium unbound ID10, IC10 and EC10 values For EBA MPA and the ‘negative control’ DMPA, no effects were observed on differentiation up to nominal concentrations of 3000 µM, and thus, medium unbound ID10 concentrations were not derived for them (see below Table 11). Medium unbound ID10 values for VPA, and PHA were 189, and 116 µM, respectively. EHA and MHA followed 4 See footnote 1, page 45. ENV/JM/MONO(2020)21  49 Unclassified with medium unbound ID10 values of 406, and 875 µM, respectively. The ‘positive control’ 4-ene-VPA had a medium unbound ID10 value of 212 μM, while the other ‘negative control’ PA showed a medium unbound ID10 value of <78 μM. Table 11. Nominal and medium unbound ID10, IC10 and EC10 values (in μM) in mEST. Abbreviated name PA DMPA EBA MPA MHA EHA VPA PHA 4-e-VPA ID10 D3 nominal 72 >3000 1 >3000 >3000 913 547 275 278 328 medium unbound 71 >2490 >2910 >2880 875 406 189 116 212 IC10 D3 nominal <78 >3000 2064 2713 >10000 <313 415 294 343 medium unbound 70 >2490 2000 2601 >9580 300 285 123 222 IC10 3T3 nominal 79 >3000 >3000 1838 6834 9568 >3000 1084 354 medium unbound 78 >2490 >2910 1762 6552 7101 >2060 452 229 HDAC EC10, day 4 nominal 950 >10000 4640 7380 >9580 340 50 40 390 medium unbound 935 >8290 4496 7076 >10000 252 34 17 253 HDAC EC10, day 10 nominal 30 3680 2130 1500 2400 160 40 30 50 medium unbound 30 3051 2064 1438 2301 119 27 13 32 1) '>'means: not detectable up to this highest tested dose. The corresponding medium unbound IC10 values in D3 were as follows: 2000, and 2601 µM for EBA, and MPA, respectively, while not detectable for the ‘negative control’ DMPA, i.e. comparable or somewhat lower as compared to their ID10 values. Those analogues having the lowest medium unbound ID10 values, i.e. VPA, and PHA also had the lowest medium unbound IC10 values: 285, and 123 µM, respectively. Of the remaining two analogues, EHA and MHA, the medium unbound IC10 values were 232 μM, and undetectably high, respectively. The ‘positive control’ 4-ene-VPA had a medium unbound IC10 value of 222 μM, while the ‘negative control’ PA showed a medium unbound ID10 value of <78 μM. For 3T3 cells, medium unbound IC10 values were as follows: for EBA, MPA and MHA, and the controls PA, DMPA, and 4-ene-VPA they were more or less comparable to the medium unbound IC10 values for D3 cells. For the other analogues, EHA, VPA, and PHA, these were considerably higher as compared to their corresponding values in D3 cells. HDAC inhibition Concerning HDAC inhibition in D3 cells the following results were observed. All compounds tested appeared to be more active when monitored at the 10th day of incubation. Still, EBA, MPA and MHA were virtually inactive in this respect: their medium unbound IC10 values were 2064, 1438, and 2301 μM, respectively. Medium unbound IC10 values of the other analogues EHA, VPA, and PHA were 119, 27, and 13 μM, respectively. The ‘positive control’ 4-ene-VPA had a medium unbound IC10 value of 32 μM, while the ‘negative controls’ DMPA and PA showed medium unbound IC10 values of 3051 and 30 μM, respectively. Please, note that HDAC monitoring used a different experimental protocol, and thus absolute dose descriptor values achieved here should not be compared with absolute descriptor values achieved for functional effects. Corresponding mouse and human OED values Mouse and human OED values for D3 differentiation, and HDAC inhibition are depicted below in Table 12 for the five compounds for which PBPK models were available. The 50  ENV/JM/MONO(2020)21 Unclassified OED values correspond to the indicated in vitro specified effect concentrations, i.e. it is the external dose needed to achieve such an internal equipotent target concentration. Table 12. mEST medium unbound ID10 and HDAC EC10 values (in μM), and corresponding OED values (mmol/kg) in mouse and human. Abbreviated name EBA MPA MHA EHA VPA D3 ID10 medium unbound >3000 1 >3000 875 406 189 Mouse OED range >26.0-31.8 >25.2-30.1 2.7-2.8 3.8-4.1 0.21-0.27 Human OED range >4.3-4.6 >4.5-4.7 0.84-0.95 0.23-0.36 0.11-0.17 HDAC EC10, day 10 medium unbound 2064 1438 2301 119 27 Mouse OED range 18.52-22.60 12.58-15.01 7.12-7.32 1.12-1.19 0.028-0.042 Human OED range 3.08-3.27 2.25-2.37 2.2-2.51 0.07-0.1 0.014-0.024 1) '>'means: not detectable up to this highest tested dose. Both MHA and EHA are somewhat in between EBA, and MPA on the one hand (highest OED values), and VPA on the other hand (lowest OED value). The OED value for differentiation inhibition for MHA and EHA are quite close for mouse, while it seems to be somewhat lower for EHA for humans, close to that of VPA. For HDAC inhibition, OEDs for mouse and human are much lower for EHA as they are for MHA. UKN1 Concentration response curves for viability are shown in Annex II, as well as an overview of graphs of expression of PAX6, OTX2 and AP2 genes, and the endpoint rosette formation at the specified viability concentrations. OED values are included in Annex I as well. In vitro medium unbound EC10 values The below Table 13 shows that cell viability reduction for EBA, MPA, and MHA is not observed up to the highest tested nominal dose of 5000 µM, which corresponds to medium unbound EC10 values of >4487, >4319, and >4321 µM, respectively. Unbound medium EC10 values for EHA, VPA, and PHA, are 184, 215, and 43 µM, respectively, i.e. much lower. The ‘positive control’ 4-ene-VPA had a medium unbound IC10 value of 121 μM, while the ‘negative controls’ DMPA and PA showed medium unbound IC10 values of 1164 and 1710 μM, respectively. Table 13. Nominal and medium unbound EC10 values (in μM) in UKN1 model. Abbreviated name PA DMPA EBA MPA MHA EHA VPA PHA 4-e-VPA EC10 viability, nominal 1807 2042 >5000 >5000 >5000 417 575 263 363 medium unbound 1710 1164 >4476 >4319 >4321 184 215 43 121 HDAC EC10, nominal 90 1240 1040 1190 2370 160 90 30 110 medium unbound 85 707 931 1028 2048 70 34 5 37 At these EC10 values gene expression and rosette formation were as follows. Those analogues having relatively low EC10 values, i.e. EHA, VPA, PHA, and 4-ene-VPA, all had clear reductions in rosette formation, and the anticipated gene expression changes, with the exception of PHA: this analogue did not show the gene expression changes accompanying inhibition of rosette formation. Prediction scores were positive for EHA, VPA, and 4-ene-VPA (UKN1 scores all >16.3), while PHA was scored as weakly positive ENV/JM/MONO(2020)21  51 Unclassified (UKN1 score of 11.9). The compounds with the relatively high EC10 values of > 5000 µM, EBA, MHA, and MPA, all have somewhat different profiles for gene expression, and inhibition of rosette formation: while EBA has no effects in these two parameters, MHA shows some gene expression changes and inhibition of rosette formation. Finally, MPA shows some of the anticipated gene expression changes, but without inhibition of rosette formation. The associated prediction scores classified EBA and MPA as being negative, while MHA was classified as unclear/weakly positive (UKN1 score of 7.9). Of the medium potent analogues with regard to viability, DMPA, and PA (EC10 values resp. 2042, and 1807 µM, and UKN1 scores of 9.6, and 17.6 resp.), PA showed most clearly showed gene expression changes, and rosette formation inhibition. Based on these profiles PA was classified as positive, DMPA as unclear/weakly positive. HDAC inhibition Concerning HDAC inhibition (monitored at day 6) the following results were observed. EBA, MPA, and MHA had relatively high EC10 medium unbound IC10 values of 1040, 1190, and 2370, respectively. Medium unbound IC10 values of the other analogues EHA, VPA, and PHA clearly were lower. i.e. 160, 90, and 30 μM, respectively. The ‘positive control’ 4-ene-VPA had a medium unbound IC10 value of 110 μM, while the ‘negative controls’ DMPA and PA showed medium unbound IC10 values of 1240, and 90 μM, respectively. Please, note that HDAC monitoring used a different experimental protocol, and thus absolute dose descriptor values achieved here should not be compared with absolute descriptor values achieved for functional effects. Corresponding mouse and human OED values Mouse and human OED values for UKN1 cell viability, and HDAC inhibition are depicted below in Table 14 for the five compounds for which PBPK models were available. The OED values correspond to the indicated in vitro specified effect concentrations, i.e. it is the external dose needed to achieve such an internal equipotent target concentration. Please, note that the ‘>’ sign on front of the OED values are to indicate that these were highest tested doses at which these effects have not been observed; these values were generated for comparison purposes only. Table 14. UKN1 medium unbound cell viability EC10 and HDAC EC10 values (in μM), and corresponding OED values (mmol/kg) in mouse and human. Abbreviated name EBA MPA MHA EHA VPA Cell viability EC10 medium unbound >4476 >4319 >4321 184 215 Mouse OED range >40-48 >38-45 >13.4-13.7 1.7-1.85 0.24-0.31 Human OED range >6.7-7.1 >6.8-7.1 >4.13-4.71 0.10-0.16 0.12-0.19 HDAC EC10 medium unbound 931 1028 2048 70 34 Mouse OED range 8.4-10.2 9-10.7 6.34-6.51 0.66-0.70 0.035-0.049 Human OED range 1.39-1.47 1.61-1.70 1.96-2.23 0.04-0.06 0.021-0.028 EBA and MPA show highest OED values both for mouse, as well as for human, both for viability reduction, as well as HDAC inhibition. MHA has comparable OED values as EBA, and MPA, apart from mouse viability reduction, where its OED is an order of magnitude lower. VPA clearly has the lowest OED values of all, equalled by EHA, apart from mouse HDAC inhibition, where its OED is an order of magnitude higher. All OED values are clearly below those of MHA, mostly an order of magnitude. 52  ENV/JM/MONO(2020)21 Unclassified CALUX Reporters In vitro cell-total LEC values Assays important for the endpoint under investigation, i.e. induction of neural tube defects, are expected to show a positive response to the in vivo positive category compounds EHA, VPA and PHA, and a negative response to the in vivo negative category compound EBA. Six assays responded in this respect: p21, PXR, TCF, ESRE, p53 GENTOX, and anti-PR, identified all these four compounds. When the control compounds (classified as outliers), i.e. DMPA, 4-ene-VPA and PA, being in vivo negative, positive, and negative, respectively, are also considered, only one assay, p21, identified all seven compounds correctly in this respect. The other five assays, PXR, TCF, ESRE, p53 GENTOX and anti-PR, all misclassified one of these outliers. PPARa responded to all 9 compounds. Twelve assays didn’t respond to any of the 9 compounds. The six statistically best classifying CALUX assays, p21, PXR, TCF, ESRE, p53 GENTOX and anti-PR, show a trend where the potency of the compound increases with chain length (see data matrix): the lowest effect concentration (LEC) for EHA > VPA > PHA (only for ESRE LEC VPA= LEC PHA). Of the thirteen responding assays, none was responding to MHA, similar as the observation for EBA. To the negative control DMPA, only the anti-AR assay responded. Also, to the structurally most closely related MPA, for which no in vivo data exist, none of the assays responded. Table 15. CALUX result table. Results are displayed as LECs in LogM. The abbreviations used for the assays are explained in Table 6. Corresponding mouse and human OED values Mouse and human OED values for CALUX reporter specified in vitro effect values were calculated for the five compounds for which PBPK models were available. The OED values correspond to the indicated in vitro specified effect concentrations, i.e. it is the external dose needed to achieve such an internal equipotent target concentration. These are not depicted here, but can be viewed in Annex I. They do not lead to different conclusions as described above. ENV/JM/MONO(2020)21  53 Unclassified 4.2.2. In silico models QSAR for prediction in vivo induction of exencephaly in rodents A more detailed description of the results is provided in Annex II. This model uses both structural characteristics as well as HDAC inhibiting capacity in UKN1 cells of the tested compounds. Based on this QSAR–linear discriminant analysis the known in vivo active analogues VPA, EHA, PHA, and 4-ene-VPA have high posterior probabilities for being active and the model, therefore, predicts them as inducers of exencephaly. The other analogues MHA, MPA, PA, and DMPA are predicted with high posterior probabilities as being non-inducers of exencephaly. EBA is predicted to be a weak inducer of exencephaly. The predictions for these negative compounds are in line with their reported in vivo observation with respect to exencephaly, save for MPA, which has not been tested for this endpoint. Toxicokinetic models Bio-kinetic modelling The results of the bio-kinetic modelling, i.e. transforming nominal concentrations into embryo-total (ZET/reporter), cell-total (CALUX reporters), and medium unbound (mEST, UKN1) is incorporated into the results description section 4.2.1 for the various experimental models (see above). PBPK modelling The results of the reverse PBPK modelling, i.e. transforming the bio-kinetic modelling derived dose descriptors into OEDs is incorporated into the justification section 4.3.2 for the various experimental models (see below). From this data, i.e. predicted OEDs corresponding to these in vitro target concentrations generated for the various specified models, it can be concluded that OED values for humans are in general lower than those for mice. This suggests humans build up higher target concentrations at similar external dose as compared to mice; for the 5 compounds explored here, this differs per compound and per in vitro model system. Concerning the differences among chemicals with regard to effective delivery 5 at the target: for mouse this is predicted to be most efficiently done by VPA, followed by MHA and MPA and EBA, with EHA being least efficient, across all models. The difference between MHA and EHA being a factor of 2 to 3, MHA being more effective. For human predictions differences are smaller and different, EHA and VPA being most efficient, shortly followed by MHA, and subsequently by MPA, and EBA. Here differences between MHA and EHA being around 20 to 40 percent only, with EHA being more effective. 4.2.3. Classification and data analysis models Dempster-Shafer Theory approach The Dempster-Shafer theory (DST) (Shafer G., 1976; Dempster AP, 1967) that was applied to the data is an extension of generalised Bayesian statistical inference in which evidence 5 relative delivery efficacy: ratio of {in vitro EC10} and associated {in vivo OED} for a chemical. 54  ENV/JM/MONO(2020)21 Unclassified can be associated with multiple sources. DST represents a rigorous decision-theory approach that provides a framework to generate predictions, estimate the uncertainty associated with each prediction, and combine multiple sources of evidence resulting in a weight-of-evidence (WoE) prediction by quantitatively accounting for the reliability of each of the individual sources (details in Annex III). In general, this decision theory will support the decision making process done by the toxicologist. In this submission, DST was used to combine the evidence from different in vitro assays for the source compounds in order to provide a WoE estimate for the target compound with respect to the in vivo neurodevelopmental toxicity outcome. Binary data were used, indicating active/inactive per assay result. The different assay sets are depicted below in the column ‘assay set‘ of Table 16. First the subset of assays were identified, which gives the best validation results from a leave-one-out (LOO) cross validation. This LOO validation enables to detect the reliability, positive prediction accuracy (PPV) and negative prediction accuracy (NPV). The LOO procedure is a standard approach for selecting a set of sources (in this case assays) based only on the training set (in this case the source compounds). DST analysis indicated that developmental and neurodevelopmental effects is predicted with 100% certainty from the current in vitro assay results. Using the correlations between in vivo and in vitro outcomes from the assays for the 7 source compounds the in vivo predicted outcome for the target compound in all 3 analysis is that MHA is not an in vivo developmental and/or neurodevelopmental toxicant (Table 16; Annex III). The same conclusion was obtained for MPA, though lacking in vivo data, but included for its structural similarity to MHA, and for exploring its in vitro response profile. Table 16. Results from DST analysis on target compound prediction. Bayesian Automatic Classification approach The results of the training of the classifier are given in Annex III. ENV/JM/MONO(2020)21  55 Unclassified In vitro 10% effect measurements If only the best tests and relevant to neurodevelopmental toxicity are used (ZET_pericardial, ZET_eyes, ZET_jitter, UKN1_viability), the classifier assigns MHA to category negative with probability 93% of the case and strongly positive with probability 7%. See Figure 7 for the assignment of MHA by those classifiers. If only the best tests and relevant to neurodevelopmental and developmental toxicity are used (All five ZET read outs, ZET reporter for cranio-facial deformation, mEST cell viability at day 3, UKN1 cell viability and all 14 CALUX assays), the classifier assigns MHA to category negative with probability 90% of the case and strongly positive with probability 10%. See Figure 8 for the assignment of MHA by those classifiers. The strongly positive assignments come from the two indicated tests. Figure 7. Results of automatic classification of MHA (red crossed circle) given its response in the various assays relevant to the underlying AOP. The black circles mark the positions of the source chemicals used for training the classifier. Corresponding in vivo OEDs, mouse If only the best tests and relevant to neurodevelopmental toxicity are used (ZET_pericardial, ZET_eyes, ZET_jitter, UKN1_viability), the classifier assigns MHA to category negative with probability 39%, mildly positive with probability 37%, positive 56  ENV/JM/MONO(2020)21 Unclassified with probability 22%, and strongly positive with probability 1%. See Figure 8 for the assignment of MHA by those classifiers. If only the best tests and relevant to neurodevelopmental and developmental toxicity are used (All five ZET read outs, ZET reporter for cranio-facial deformation, mEST cell viability at day 3, UKN1 cell viability and all 14 CALUX assays), the classifier assigns MHA to category negative with probability 20%, mildly positive with probability 44%, positive with probability 30% and strongly positive with probability 4%. See Figure 8 for the assignment of MHA by those classifiers. The low number of training chemicals (only three) is probably responsible for the uncertainty in the predictions. The pharmacokinetic correction also seems to increase the predicted toxicity. Figure 8. Results of automatic classification of MHA (red crossed circle) given its response in the various assays relevant to the underlying AOP, after pharmacokinetic correction for the mouse. The black circles mark the positions of the source chemicals used for training the classifier. Corresponding in vivo OEDs, human If only the best tests and relevant to neurodevelopmental toxicity are used (ZET_pericardial, ZET_eyes, ZET_jitter, UKN1_viability), the classifier assigns MHA to category negative with probability 69%, positive with probability 18%, and strongly positive with probability 3%. See Figure 9 (next page) for the assignment of MHA by those classifiers. ENV/JM/MONO(2020)21  57 Unclassified If only the best tests and relevant to neurodevelopmental and developmental toxicity are used (All five ZET read outs, ZET reporter for cranio-facial deformation, mEST cell viability at day 3, UKN1 cell viability and all 14 CALUX assays), the classifier assigns MHA to category negative with probability 50%, mildly positive with probability 41%, positive with probability 8%, and strongly positive with probability 1%. See Figure 9 (next page) for the assignment of MHA by those classifiers. The low number of training chemicals (only three) is probably responsible for the uncertainty in the predictions. Figure 9. Results of automatic classification of MHA (red crossed circle) given its response in the various assays relevant to the underlying AOP, after pharmacokinetic correction for the mouse. The black circles mark the positions of the source chemicals used for training the classifier. For MPA, the analogue without in vivo data included because of its close structural similarity to the target chemical MHA, also predictions were made to see whether it would fit the trend in the category, i.e. going from in vivo positives at the right side (PHA, VPA, and EHA in decreasing potency) to the in vivo negatives at the left side (EBA), and to verify whether it would indirectly provide support to any read-across to MHA. Corresponding Figures to the below described analysis results are to be found in Annex III. Based on EC10 values for neurodevelopmental toxicity tests (see above with MHA), the classifier assigns MPA to category negative with probability 99%, and to strongly positive with probability 1%. If all tests relevant to neurodevelopmental and developmental toxicity are used (see above with MHA), the classifier assigns MPA to category negative with probability 96% and to strongly positive with probability 4%. Based on the correspondingly derived in vivo OEDs for mouse for neurodevelopmental toxicity tests (see above with MHA), the classifier assigns MPA to category negative with 64  ENV/JM/MONO(2020)21 Unclassified patterning, differentiation), but not processes and cell types that occur later in embryonic development, such as neurite outgrowth or neural crest cells or mature neurons. Uncertainties of this method are also caused due to limited metabolic capacity of the cells (leading to false negatives when metabolite is toxicant, or false positives if parent is toxicant), as well as the absence of an in vivo physiological distribution system. The laboratory is well experienced with this model (high reproducibility, low variability). The method is not fully evaluated concerning sensitivity and specificity, as there are only few DNT positive compounds identified in humans, mainly based on epidemiological studies. Assessment Comparing in vitro medium unbound concentrations for EC10 viability, for induction of marker genes expression, and for inhibition of rosette formation with in vivo exencephaly induction potential for the investigated analogues shows the following picture: those analogues inducing exencephaly in vivo, i.e. EHA, VPA,and PHA, and the ‘positive control’ 4-ene-VPA, have relatively low EC10 values of 184, 215, 43, and 121 µM, respectively, show clear inhibition of rosette formation at these concentration levels, and marker gene expression changes at somewhat higher concentrations. PHA, though, behaves somewhat differently in this latter respect: this analogue doesn’t induce expression changes of marker genes at levels clearly inhibiting rosette formation. The prediction (UNK1) score does classify EHA, VPA, and 4-ene-VPA as positive, and PHA as weakly positive neurodevelopmental toxicants. The in vivo negative compounds, i.e. the analogue EBA, and the ‘negative controls’ DMPA, and PA, have all clearly higher medium unbound EC10 values, i.e. >4476, 1164, and 1710 µM, respectively, than in vivo positives. EBA and DMPA also show inhibition of rosette formation at these higher concentrations, but do not change expression of marker genes, while PA shows both gene expression changes, and rosette inhibition at the EC10 viability level. MPA, without in vivo data, also had a high medium unbound EC10 value of >4319 µM (and a UKN1 score of 6.3), showing the gene expression changes, but no inhibition of rosette formation. It may be seriously doubted whether these high concentrations are achieved in vivo. MHA showed some gene expression changes, and inhibition of rosette formation (giving a UKN1 score of 7.9), but only at a relatively high medium unbound EC10 value of >4321 µM. As MHA is the least potent of all analogues in HDAC inhibition, with an EC10 of 2048 µM, this activity evidently is not involved in the observed inhibition of rosette formation (i.e. for this chemical). Conclusion The effects of MHA, resembles some of the in vivo negative compounds. Also, its relatively low HDAC inhibition potential supports the conclusion that MHA would be expected to be negative for induction of exencephaly in vivo. The prediction score at a relatively high concentration supports this conclusion. ENV/JM/MONO(2020)21  65 Unclassified CALUX Reporters Uncertainty All assays have the same cellular background of U2-OS cells, that have very low expression levels of endogenous receptors, no cross-talk with other receptors, and, therefore, are highly specific and responsive. Uncertainties of this method are also caused due to limited metabolic capacity of the cells (leading to false negatives when metabolite is toxicant, or false positives if parent is toxicant), as well as the absence of an in vivo physiological distribution system. If information on bio(in)activation is required, a metabolic module can be added to the CALUX assay to study the activity of metabolites. Solubility was well monitored, while for bioavailability it was assumed that extracellular free concentrations mimic those intracellularly. The laboratory is well experienced with this model (high reproducibility, low variability). Assessment When comparing the exencephaly-inducing capability in vivo of the test compounds with response induction in CALUX reporter assays, the following picture emerges: the in vivo positive compounds EHA, VPA, PHA, and the positive control 4-ene-VPA, are active, i.e. having their LEC at ≤ 1 mM concentration, in 8, 7, 13, and 6 of the 25 assays (leaving out PPARa as non-discriminative here), while for the in vivo negative compounds EBA, DMPA, and PA this is 0, 1 and 6, respectively. Additionally, those being in vivo most potent, VPA and PHA, also show the lowest LEC values in the CALUX assays: VPA has LEC values ≤ 0.1 mM for 3 assays, for PHA this is for 6 assays, while none of the LEC values for EHA (and the positive control 4-ene-VPA), is ≤ 0.1 mM. MHA doesn’t trigger any assay (not counting PPARa, that is triggered by all), and would be grouped with EBA, and DMPA, not with EHA, VPA, PHA, and 4-ene-VPA, which implies it will most probably be negative in vivo as well. A similar conclusion can be achieved when considering the chain length of the 6 analogues EBA, MPA, MHA, EHA, VPA, and PHA, and their potential to activate assays: for EBA, MPA, MHA no activations are observed up to 1 mM, while EHA, VPA, and PHA activate 8, 7, and 13 assays up to 1 mM, with VPA, and PHA activating 3, and 6, respectively, even at 0.1 mM. Thus, with increasing chain length, compounds show more potency in activating CALUX assays. Partial least squares discriminant analysis (PLS-DA) was performed on the CALUX results to build a predictive model for the CALUX panel. A statistical model was built using the analogues that have in vivo data, i.e. PHA, VPA, 4-ene-VPA, EHA, PA, DMPA and EBA. Subsequently, MPA and MHA were predicted based on this model. The PLS-DA model was able to differentiate between the in vivo positive compounds (+1) and the in vivo negative compounds (-1). The CALUX assays identified as most relevant for building this model were ESRE, p21, TCF, PR-anti, PXR, p53 GENTOX; these are the same six assays identified as key factors when considering the ‘false positive/negative’ rate. The PLS-DA model predicted MHA as negative in vivo (Figure 12). 66  ENV/JM/MONO(2020)21 Unclassified Figure 12. PLS-DA analysis; statistical model was built using PHA, VPA, 4-ene-VPA, EHA, PA, DMPA and EBA. MHA, and MPA were predicted based on this model. Role of HDAC inhibition on activation profile VPA is known to inhibit histone deacetylase (HDAC), a negative regulator of gene expression in multiple settings. HDACs are recruited by a variety of transcription factor corepressor complexes and are believed to repress transcription by reducing the level of acetylation of core histones, thereby altering chromatin structure. HDAC inhibition therefore results in an increase in activity of several reporters and cell signalling pathways. The CALUX activity profile of VPA and its analogues could be explained by their activity as HDAC inhibitors, rather than direct interaction with the pathways (Piersma et al. 2013). The CALUX assays that are predominantly activated by the VPA analogues are p21, TCF, ESRE, p53 GENTOX, PXR and anti-PR. All nuclear receptor-mediated transcriptional responses are regulated by HDACs. The reason why not all receptors are being activated by the HDAC inhibitors may be because different HDACs exist with different selectivities. Possibly the valproates inhibit a specific class only. Alternatively, it could be that some background receptor activity may be needed to see an effect of HDAC-inhibition. In our assays, background activity is very low, and possibly with low concentrations of specific ligand the activation may become visible. The four cell signalling pathways p21, p53, TCF (Wnt) and ESRE, on the other hand, have been previously reported to be upregulated by VPA through HDAC inhibition (Phiel et al. 2001, Hoti et al. 2006, Paradis and Hales Barbara 2015, Segar et al. 2017). HDAC inhibition studies with mEST and UKN1 models (Table 11 and Table 13) have shown that five of the compounds involved in this read-across act as HDAC inhibitors. See Figure 13; all four in vivo active compounds VPA, EHA, PHA and 4-ene VPA are identified as HDAC inhibitors, and are active on all four selected CALUX assays. The in vivo negative compound PA was already defined as an outlier based on its structure, its ENV/JM/MONO(2020)21  67 Unclassified CALUX profile and the PLS-DA analysis; also with respect to HDAC inhibition, it behaves as an outlier, since it is positive here as well. Figure 13. Relationship between CALUX reporter activation for TCF, ESRE, p21 and p53 and HDAC inhibition properties of case study chemicals. CALUX: grey cells = not active; values = LEC in Log(M). HDACi: green cells = not active; red cells = active with EC50< 2mM. Conclusion Based on the above statistical reasoning, structure-based analysis, as well as PLS-DA, MHA is predicted as in vivo negative for exencephaly induction. 4.3.2. In silico models QSAR for prediction in vivo induction of exencephaly in rodents For the target compound MHA of this read-across case study, the model predicts it as a non-inducer of exencephaly. The three in vivo positive source compounds, EHA, VPA, and PHA, are predicted as inducers of exencephaly, while a fourth source compound, EBA, known to be negative in vivo, is predicted as weak inducer. This span of activity for the source compounds affords all the necessary information to correctly predict the target compound’s ability to induce the exencephaly in vivo. Toxicokinetic models Bio-kinetic modelling The results of the bio-kinetic modelling, i.e. transforming nominal concentrations into embryo-total (ZET/reporter), medium unbound (mEST, UKN1), and cell-total (CALUX reporters) is incorporated into the results description section 4.2.1 for the various experimental models (see above), and included in the discussion and conclusion section 4.3.1. 68  ENV/JM/MONO(2020)21 Unclassified PBPK modelling Differences between chemicals here mainly reside in differences in absorption, plasmabinding and clearance rates. From Annex I it can be deduced that plasma-binding of EHA is higher than that of MHA by a factor of about two (and MPA and EBA equal MHA), while in vivo clearance is almost an order lower for EHA (here MPA and EBA double MHA). Although EHA is one of the more lipophilic compounds in the series it is more ionised and so is less rapidly and less completely absorbed when ingested orally, therefore, to account for this higher doses are required to achieve the effective peak concentrations, this is especially pronounced in the mouse model. The above kinetic aspects are integrated via the reverse dosimetry calculations into OED values. The predicted OED values associated with in vitro target concentrations show that EHA is less effective in building these target concentrations in mouse as compared to MHA, by a factor of 2 to 3, while in human on the other hand EHA is slightly more effective in this, by 20-40%. It should be noted that this predicted species difference in toxicokinetics is driven in part, by the use of a more conservative model for the prediction of oral absorption parameters in the human PBPK models (namely the fraction absorbed, fa, and the first-order absorption rate constant, Ka (1/h); a comparison of the models used to predict oral absorption parameters in human and mouse is detailed in Annex III). For arriving at relative in vivo toxic potency factors for these chemicals these toxicokinetic differences between MHA and the other category members need to be combined with their toxicodynamic differences, especially the relative potency of MHA towards EBA, and EHA, respectively, is of importance to make a reliable prediction on the in vivo (neuro)developmental potency of MHA itself. This comparison is made in section 5.2. 4.3.3. Classification and data analysis models Dempster-Shafer Theory approach Using DST that correlates in vivo observations (i.e. being neurodevelopmental toxicant or not) and in vitro outcomes of the assays for the 7 source compounds, the in vivo predicted outcome for the target compound MHA in all 3 analysis (only neurodevelopmental read outs, all read outs, only CALUX read outs) is that it is not neurodevelopmental toxic. For MPA the same conclusion holds. Bayesian Automatic Classification approach From this classification analysis, the results for the in vitro effect concentrations clearly show MHA to be predicted negative for neurodevelopmental effects induction, as well as induction of developmental and neurodevelopmental effects, with probabilities of 93, and 90%, respectively. When corresponding human OED values are used as input for this, the negative prediction probabilities for neurodevelopmental effects, and for both developmental and neurodevelopmental effects reduce to 69, and 50%, respectively. For mouse these predictions for neurodevelopmental toxicity turn to 39% for negative, 37% for mild positive, 22% for positive, and 1% for strong positive, while predictions for developmental and neurodevelopmental effects turn to 20% for negative, 44% for mild positive, 30% for positive, and 4% for strong positive. For MPA, one of the two structurally closest to MHA (next to EHA), for which we also do not have in vivo data (but that was included for exploring any category trend), the ENV/JM/MONO(2020)21  69 Unclassified corresponding negative prediction probabilities for neurodevelopmental effects, and for both developmental and neurodevelopmental effects, are 99, and 96%, respectively, when based on in vitro data. The OED values derived negative prediction probabilities for this analogue for neurodevelopmental effects, and for both developmental and neurodevelopmental effects, being 90, and 87%, for mouse, respectively, and 58, and 79%, for human, respectively. Biological Fingerprint Classification approach The results show that based on CALUX responses MHA nicely fits in with the in vivo negative compounds. If the CALUX responses are converted into a single response, and combined with responses seen in the other in vitro assays, MHA moves away from the negative compounds to an independent position. 70  ENV/JM/MONO(2020)21 Unclassified 5. Strategy for and integrated conclusion of data gap filling In this chapter, an integrated conclusion is drawn on filling the in vivo data-gap of the target chemical MHA on the basis of all results generated with the methodology described in chapter 4. , and on the basis of the approach and assumptions as described in chapters 2. and 3. As all background information as well as all information generated in the context of this study has intrinsic uncertainty that may have bearing on the final conclusion, all this will be outlined in the first section of this chapter to arrive at a final conclusion with regard to the objective of this study in section 2. Therefore, here we will discuss the following aspects: 1) Hypothesis used for the read across; 2) Structural similarity; 3) Similarity of physio-chemical properties; 4) Similarity of toxicokinetics data; 5) Similarity of other supportive data (e.g. data related to key event); 6) Number of analogues used for the readacross; 7) Quality of the endpoint data used for the read-across; 8) Similarity of the endpoint data (among source chemicals); 9) Concordance and weight of evidence of all data used for justifying the hypothesis, and finally 10) Overall uncertainty of the read across. 5.1. Uncertainty 5.1.1. Hypothesis used for the read-across Source compounds are selected on a firm and transparent structural basis, showing high structural similarity to the target MHA. However, within this well-defined category a remarkable difference in toxicological profile is observed: some show induction of exencephaly in mice (and man), a neurodevelopmental disorder, while others don’t. There is yet no clearcut structural feature to be linked with this specific toxicity. To reliably classify, MHA within this category with regard to this neurodevelopmental toxicity, response data within specific in vitro models is generated. The in vivo data suggest that in this case study structure-activity relationships may be subtle. Compared to the overall uncertainty of this assessment the uncertainty assigned to this aspect is low to medium. 5.1.2. Structural similarity The structural similarity basis of this read across is quite reliable: all category members are 2-branched aliphatic carboxylic acids. The category includes members differing only one carbon-atom with the MHA molecule, i.e. MPA, and EHA, respectively. Nonetheless, the in vivo data suggest that in this case study structure-activity relationships may be subtle. It is noted that some of the aliphatic carboxylic acids have asymmetric carbon atoms, and thus may exist as racemates: i.e. MPA, MHA, EHA, PHA, and 4-ene-VPA. The in vivo data in the NMRI mouse model were generated with the racemates, as far as data on this was retrievable. Also, all in vitro tests described in this report have been performed with racemates of these chemicals as well. Compared to the overall uncertainty of this assessment the uncertainty assigned to this aspect is low. 5.1.3. Similarity of physio-chemical properties (see remark on asymmetric carbon atoms for some aliphatic carboxylic acids under 5.1.2). ENV/JM/MONO(2020)21  71 Unclassified The structural similarity basis of this read across is the 2-branched aliphatic carboxylic acid structure, with only minimal chain-length variations. When glancing at the typical physchem properties going from small chain members to longer ones relatively small ranges are observed (rounded values): 116 to 172 g/mol for molecular weight, 15 to 59oC, and from 196 to 269oC for melting and boiling points, 1,7 to 3,2 or from 2,0 to 3,9 for experimental or predicted LogPow ranges, 18 to 0,3 g/L for water solubility shows, while experimental pKa constant values appear rather similar: 4,7 to 4.6, leading to the conclusion that these compounds are all non-volatile, well to reasonable water soluble, will not bioaccumulate, and are weak acids of quite comparable strength. Compared to the overall uncertainty of this assessment the uncertainty assigned to this aspect is low. 5.1.4. Similarity of toxicokinetics data (see remark on asymmetric carbon atoms for some aliphatic carboxylic acids under 5.1.2). Bio-kinetic modelling The results of the bio-kinetic modelling, i.e. transforming nominal concentrations into embryo-total (ZET/reporter), medium unbound (mEST, UKN1), and cell-total (CALUX reporters) is performed as described in section 4.2.1. For CALUX reporters calibrated models were used. Due to unavailability of bio-analysis data for ZET, mEST, and UKN1 uncalibrated models were used. Compared to the overall uncertainty of this assessment the uncertainty assigned to this aspect is low to medium. Metabolism Using Meteor Nexus biotransformations were predicted for all category analogues (all these compounds were within its domain). Also, a detailed metabolism pattern (first generation) was generated for the two nearest analogues of MHA, being MPA and EBA on one side, and EHA, and VPA on the other side. From this, it is concluded that both biotransformation fingerprints, and concordant metabolite scores are high. Compared to the overall uncertainty of this assessment the uncertainty assigned to this aspect is low to medium. It is remarked here that all in vitro models were metabolically not fully competent (as also indicated at the model descriptions in section 4.3.1). This introduces uncertainties to the dose-response data, and the final conclusions on intrinsic potency. PBPK modelling A trend between fraction unbound and side chain length is observed. Only predicted data for plasma protein binding are available (from four different QSAR models). All compounds are within the models’ applicability domain. The highest and lowest predict ppb value was used for the in vitro to in vivo calculation. By this approach a range of human equivalent doses is predicted, which gives in our view a more realistic picture than taking only one single value. Intrinsic hepatic clearance was measured for all grouped compounds in HepaRG cells. It turned out that the longer chain analogue PHA was outside the applicability domain of 72  ENV/JM/MONO(2020)21 Unclassified these assays, whereas VPA and the majority of the shorter chain analogues could be measured. A functional PBPK model could be build based on in vivo data for one analogue, VPA. This model predicts bioavailable concentrations of VPA in plasma and liver very well. A good prediction was also obtained by parameterising the PBPK model with in vitro values on ppb and intrinsic hepatic clearance. This proof of concept gives confidence in the IVIVE approach used for all analogues in this case study. Compared to the overall uncertainty of this assessment the uncertainty assigned to this aspect is low to medium. 5.1.5. Similarity of other supportive data (e.g. data related to key event) Sections 3.5, 3.6 and 3.8 describe and discuss supportive biological information showing relatively similar biological behaviour of the category members where it concerns HDAC inhibition as common initial key event for neurodevelopmental members (section 3.5), and identical chemical-biological interaction results for all category members for DNA and protein binding, skin sensitisation, and irritation for eyes and skin using OECD Toolbox profilers. Compared to the overall uncertainty of this assessment the uncertainty assigned to this aspect is low. 5.1.6. Number of analogues used for the read-across The number of analogues used is limited, but this is due to the fact that constraints set to source chemicals, being very high structural similarity, and having in vivo data from the same test-protocol. Compared to the overall uncertainty of this assessment the uncertainty assigned to this aspect is low. 5.1.7. Quality of the endpoint data used for the read-across The in vivo data of the source chemicals are discussed in section 3.3, all come from a specific mouse species protocol from one laboratory, excluding variations possibly due to species, laboratory, diet etc. The in vitro and in silico data used to built and support the biological read across in this case study are described in sections 4.1 to 4.3, all demonstrated as suitable for identification of (neuro)developmental toxicants. Compared to the overall uncertainty of this assessment the uncertainty assigned to this aspect is low to medium. 5.1.8. Similarity of the endpoint data (among source chemicals) As a general remark: all generated biological data, be it in vivo or in vitro intrinsically will show variation in responses. Repetition of in vivo studies performed according to OECD standards may show differences in potency of about one order of magnitude between studies, and even differences in specific observed toxicological effects (e.g. Janer et al., 2007, 2008). In this case study in vivo data of the source chemicals come from a specific standard protocol developed by Nau and colleagues, all in vitro data generated within this case study ENV/JM/MONO(2020)21  73 Unclassified to build the biological evidence for read across is performed for source and target chemicals under exactly the same conditions. Compared to the overall uncertainty of this assessment the uncertainty assigned to this aspect is relatively low. 5.1.9. Concordance and weight of evidence of all data used for justifying the hypothesis The structural similarity of these 2-branched aliphatic carboxylic acids has been described as one commonality of the category members. Also, phys-chem properties, and chemicobiological reactivity appear quite comparable. Predictions on metabolism show reasonable similar profiles as well. On the other hand, it should be mentioned that all in vitro models are metabolically not fully competent, which introduces an uncertainty with regard to quantitatively using and comparing in vitro responses for concluding on in vivo responses and deriving OEDs. The in vitro data generated in this case study show quite consistent effects for the category members, both for the target chemical as well as for source chemicals. There are differences in potency of inducing the various monitored effects, though: from this, it is concluded that MHA is most often grouping with the in vivo negative analogue EBA, and with the negative controls PA, and DMPA. Although in some cases it shows somewhat higher potency of inducing in vitro effects than these in vivo negative chemicals, in all cases its effective concentration is higher than that of EHA, mostly about an order of magnitude, if effective at all (i.e. in some models highest tested doses of MHA didn’t show any effects). This is also reflected in the HDAC inhibition potential: MHA clusters with in vivo negative category member EBA, and the negative controls. It is noted that HDAC inhibition on the one hand, and differentiation inhibition in mEST cells, as well as inhibition of rosette formation in UKN1 cells on the other hand, do not fully parallel for MHA: suggesting these two activities are not linked for this chemical (in at least these two models). When applying integrating statistical analysis to this in vitro dataset, DST analysis predicts MHA to be no in vivo developmental and/or neurodevelopmental toxicant. The Bayesian automatic classifier analysis showed similar results for MHA, and MPA, when applied to the in vitro dose descriptors, and the Biological Fingerprint analysis gives a similar conclusion for the CALUX reporter data. When combining this CALUX data with the other in vitro responses, MHA seems to cluster on its own, at a distant from the in vivo positive and negative analogues. The prediction model, built on known qualitatively observed in vivo structure-activity features of VPA analogues and their exencephaly induction on the one hand, HDAC inhibitory potency, and relevant physical-chemical parameters on the other, predicted MHA to be non-exencephaly inducer. The Bayesian automatic classifier analysis is also applied to the OED data that were derived from the in vitro dose descriptors by reverse PBPK modelling. For man this result in a similar conclusion as derived from the in vitro data: MHA clusters more with the in vivo negative members, than with EHA. A somewhat different picture emerges for the mouse though: as a result of a predicted higher delivery efficacy 6 for MHA, as compared to EHA, 6 relative delivery efficacy: ratio of {in vitro EC10} and associated {in vivo OED} for a chemical. 80  ENV/JM/MONO(2020)21 Unclassified Janer G, Hakkert BC, Vermeire T, Piersma AH (2007), A retrospective analysis of the added value of the rat two-generation reproductive toxicity study versus the rat subchronic toxicity study, Reproductive toxicology, Vol. 24, pp. 103-13. Janer G, Slob W, Hakkert BC, Vermeire T, Piersma AH. (2008), A retrospective analysis of developmental toxicity studies in rat and rabbit: what is the added value of the rabbit as an additional test species?, Regul.Toxicol.Pharmacol, Vol. 50, pp. 206-17. Jentink J, Loane MA, Dolk H, Barisic I, Garne E, Morris JK et al. (2010), Valproic acid monotherapy in pregnancy and major congenital malformations, New Engl.J.Med, Vol. 362, pp. 2185-93. Kais B., Schneider K.E., Keiter S., Henn K., Achermann C., Braunbeck T. (2013), DMSO modifies the permeability of the zebrafish (Danio rerio) chorion-implications for the fish embryo test (FET), Aquat. Toxicol, Vo. 140-141, pp. 229-38. Kantola-Sorsa E, Gaily E, Isoaho M, and Korkman M (2007), Neuropsychological outcomes in children of mothers with epilepsy. Journal of the Int. Neuropsychological Society, Vol. 13, pp. 642-652. Kari G., Rodeck U., Dicker AP. (2007), Zebrafish : An Emerging Model System for Human Disease and Drug Discovery, Clinical Pharmacol. & Therapeutics, Vol. 82, pp. 70-80. Kilford P.J., Gertz M., Houston J.B., Galetin A. (2008), Hepatocellular binding of drugs: correction for unbound fraction in hepatocyte incubations using microsomal binding or drug lipophilicity data, Drug Metab Dispos, Vol. 36, pp. 1194-1197. Koren G., Nava-Ocampo AA, Moretti ME, Sussman R, and Nulman I. (2006), Major malformations with valproic acid, Can Fam Physician, Vol. 52, pp. 441–447. Krug AK, et al. (2013), Human embryonic stem cell-derived test systems for developmental neurotoxicity: a transcriptomics approach, Archives of Toxicology, Vol. 87, pp. 123–143. Lammer E., Carr G.J., Wendler K., Rawlings J.M., Belanger S.E., Braunbeck T. (2009), Is the fish embryo toxicity test (FET) with the zebrafish (Danio rerio) a potential alternative for the fish acute toxicity test?, Comp. Biochem. Physiol, Vol. 149C, pp. 196-209. Leist M, Hasigawa N, Rovida C, Daneshiam M, Basketter D, Kiber I, et al. (2014), Consensus report on the future of animal-free systemic toxicity testing, ALTEX, Vol. 31, pp. 341–56. Lewin G, Escher SE, Van der Burg B, Simetska N, Mangelsdorf I (2015), Structural features of endocrine active chemicals – A comparison of in vivo and in vitro data, Reprod. Toxicol, Vol. 55, pp. 81-94. Linden van der SC, von Bergh A, Van Vugt-Lussenburg B, Jonker L, Brouwer A, Teunis M, Krul C and Van der Burg B. (2014), Development of a panel of high throughput reporter gene assays to detect genotoxicity and oxidative stress, Mutation Res, Vol. 760, pp. 23-32. Lloyd KA (2013), A scientific review: mechanisms of valproate-mediated teratogenesis, Bioscience Horizons, Vol. 6, pp. 1-10. Martinez C.S., Feas D.A., Siri M., Igartúa D.E., Chiaramoni N.S., del.V. Alonso S., Prieto M.J. (2018), In vivo study of teratogenic and anticonvulsant effects of antiepileptic drugs in zebrafish embryo and larvae, Neurotox. And Teratox, Vol. 66, pp. 17-24. Muñoz-Espín D., M. Cañamero, A. Maraver, G. Gómez-López, J. Contreras, S. Murillo-Cuesta, A. Rodríguez-Baeza, I. Varela-Nieto, J. Ruberte, M. Collado, M. Serrano (2013), "Programmed Cell Senescence during Mammalian Embryonic Development." Cell, Vol. 155, No.5, pp. 1104-1118. Murko C, Lagger S, Steiner M, Seiser C, Schoefer C, and Pusch O (2013), Histone deacetylase inhibitor trichostatin A induces neural tube defects and promotes neural cvrest specification in the chicken neural tube. Differentiation, Vol. 85, pp. 55-66. ENV/JM/MONO(2020)21  81 Unclassified Murko C, Lagger S, Steiner M, Seiser C, Schoefer C, and Pusch O (2010), Expression of class I histone deacetylases during chick and mouse development. Int. J. Dev. Biology, Vol. 54, pp.1527–1537. Nagel R. (2002), DarT: The embryo test with the zebrafish Danio rerio - a general model in ecotoxicology and toxicology, ALTEX, Vol. 19, Suppl. 1, pp. 38-48. Narotsky MG, Kavlock EZ. (1994), Developmental toxicity and structure–activity rela-tionships of aliphatic acids, including dose–response assessment of valproic acid in mice and rats, Fundam Appl Toxicol, Vol. 22, pp. 251–65. Nau H., Zierer R., Spielmann H., Neubert D., Gansau C. (1981), a new model for embryotoxicity testing: teratogenicity and pharmacokinetics of valproic acid following constant rate administration in the mouse using human therapeutic drug and metabolite concentrations, Life Sci., Vol. 29, pp. 28032814. Nau H., Löscher W. (1986), Pharmacologic evaluation of various metabolites and analogs of valproic acid: teratogenic potencies in mice, Fundamental and applied Toxicology, Vol. 6, pp. 669-676. Nau H, Hawk R-S, Ehlers K (1991), Valproic Acid-Induced Neural Tube Defects in Mouse and Human: Aspects of Chirality, Alternative Drug development, Pharmacokinetics and Possible Mechanisms, Pharmacology & Toxicology, Vol. 69, pp. 310-321. NRC (2000), Scientific Frontiers in Developmental Toxicology and Risk Assessment, NAS Press, National Research Council Washington, D.C. USA. Nostrand, van, J. L., M. E. Bowen, H. Vogel, M. Barna, L. D. Attardi (2017), "The p53 family members have distinct roles during mammalian embryonic development.", Cell Death Differ., Vol. 4. Pp. 575579. (1476-5403 (Electronic)). OECD (2009) Test guideline No.455. Stably transfected human estrogen receptor-transcriptional activation assay for detection of estrogenic agonist-activity of chemicals, OECD Publishing, Paris, https://doi.org/10.1787/9789264076372-en. OECD (2013a) SPSF by the Netherlands: U2-OS cells Transcriptional ERalpha CALUX®- assay for the detection of estrogenic and antiestrogenic chemicals for inclusion in TG455/TG457, OECD, Paris. OECD (2013b) SPSF by the European Commission: Performance-Based Test Guideline on Androgen Receptor Transactivation Assays (AR CALUX), OECD, Paris Ogungbenro, K., Aarons, L., Cresim, Epi, C.P.G. (2014), A physiologically based pharmacokinetic model for Valproic acid in adults and children, Eur J Pharm Sci, Vol. 63, pp. 45-52. Ong LL, Schardein JL, Petrere JA, Sakowski R, Jordan H, Humphrey RR et al. (1983), Teratogenesis of calcium valproate in rats, Fundamental & Applied Toxicology, Vol. 3, pp. 121-6. Ornoy A. (2009), Valproic acid in pregnancy: how much are we endangering the embryo and fetus, Reproductive Toxicology, Vol. 28, pp. 1-10. Padmanabhan R, Ahmed I. (1996), Sodium valproate augments spontaneous neural tube defects and axial skeletal malformations in TO mouse fetuses [corrected].[Erratum appears in Reprod Toxicol 1996 Nov-Dec;10(6):VI], Reproductive Toxicology, Vol. 10, pp. 345-63. Paradis F.-H. and Hales Barbara F. (2015), Valproic Acid Induces the Hyperacetylation of P53, Expression of P53 Target Genes, and Markers of the Intrinsic Apoptotic Pathway in Midorganogenesis Murine Limbs Birth Defects Research Part B: Developmental and Reproductive, Vol. 104, pp. 177-226. Park SJ, Ogunseitan OA, Lejano RP (2013), Dempster‐Shafer Theory Applied to Regulatory Decision Process for Selecting Safer Alternatives to Toxic Chemicals in Consumer Products, Integrated Environmental Assessment and Management, Vol. 10, pp. 12–21. 82  ENV/JM/MONO(2020)21 Unclassified Paulson RB, Sucheston ME, Hayes TG, Paulson GW (1985), Teratogenic effects of valproate in the CD-1 mouse fetus, Archives of Neurology, Vol. 42, pp. 980-3. Pelka, K.E., Henn, K., Keck, A., Sapel B., Braunbeck, T. (2017), Size does matter – Determination of the critical molecular size for the uptake of chemicals across the chorion of zebrafish (Danio rerio) embryos, Aquat. Toxicol, Vol. 185, pp. 1-10. Pennanen S, Auriola S, Manninen A, Komulainen H. (1991), Identification of the Main Metabolites of 2Ethylhexanoic Acid in Rat Urine Using Gas Chromatography-Mass Spectrometry, Journal of Chromatography B: Biomedical Sciences and Applications, Vol. 568, pp. 125-134. Pennanen S, Tuovinen K, Huuskonen H, Komulainen H. (1992), The Development Toxicity of 2Ethylhexanoic Acid in Wistar Rats, Toxicological Sciences, Vol. 19, pp. 505–511. Pennanen S, Kojo A, Pasanen M, Liesivuori J, Juvonen RO, Komulainen H. (1996), CYP Enzymes Catalyze the Formation of a Terminal Olefin from 2-Ethylhexanoic Acid in Rat and Human Liver, Human and Experimental Toxicology, Vol. 15, pp. 435-442. Péry, Alexandre R. R., James Devillers, Céline Brochot, Enrico Mombelli, Olivier Palluel, Benjamin Piccini, François Brion, Rémy Beaudouin. (2014), “A Physiologically Based Toxicokinetic Model for the Zebrafish Danio Rerio.” Environmental Science & Technology, Vol. 48, No. 1, pp. 781–90. https://doi.org/10.1021/es404301q. Petrere JA, Anderson JA, Sakowski R, Fitzgerald JE, De La Igelsia FA (1986), Teratogenesis of calcium valproate in rabbits, Teratology, Vol. 34, pp. 263-9. Phiel C.J., Zhang F., Huang E.Y., Guenther M.G., Lazar M.A., Klein P.S. (2001), Histone Deacetylase is a direct target of valproic acid, a potent anticonvulsant, mood stabilizer, and teratogen, J Biol Chem, Vol. 276, pp. 36734-36741. Piersma A. H., S. Bosgra, M. B. van Duursen, S. A. Hermsen, L. R. Jonker, E. D. Kroese, S. C. van der Linden, H. Man, M. J. Roelofs, S. H. Schulpen, M. Schwarz, F. Uibel, B. M. van Vugt-Lussenburg, J. Westerhout, A. P. Wolterbeek, B. van der Burg (2013), "Evaluation of an alternative in vitro test battery for detecting reproductive toxicants.", Reprod Toxicol, Vol. 38, pp. 53-64. Riebeling, Christian et al. (2011), “The Embryonic Stem Cell Test as Tool to Assess Structure-Dependent Teratogenicity: The Case of Valproic Acid, Toxicological sciences: an official journal of the Society of Toxicology, Vol. 120, No.2, pp. 360-370. Segar K.P., Chandrawanshi V., Mehra S. (2017), Activation of unfolded protein response pathway is important for valproic acid mediated increase in immunoglobulin G productivity in recombinant Chinese hamster ovary cells, J. Biosci Bioeng, Vol. 124, pp. 459-468. Silva MFB, Aires CCP, Luis PBM, Ruiter JPN, Ijlst L, Duran M, Wanders RJA, Tavares de Almeida I. (2008), Valproic Acid Metabolism and its Effects on Mitochondrial Fatty Acid Oxidation: A Review, Journal of Inherited Metabolic Disease, Vol. 31, pp. 205-216. Sipes NS, Martin MT, Reif DM, Kleinstreuer NC, Judson RS, Singh AV, et al. (2011) Predictive models of prenatal developmental toxicity from ToxCast high throughput screening data, Toxicol Sci, Vol. 124, pp. 109–27. Schenk B, Weimer M, Bremer S, van der Burg B, Cortvrindt R, Freyberger A, et al. (2010) The ReProTect Feasibility Study, a novel comprehensive in vitro approach to detect reproductive toxicants, Reprod Toxicol, Vol. 30, pp. 200–18. Scholz G, Genschow E, Pohl I, Bremer S, Paparella M, Raabe H, Southee J, and Spielmann H (1999), Prevalidation of the Embryonic Stem Cell Test (EST) - A new in vitro embryotoxicity test, Toxicology in vitro, Vol.13, pp. 675-681. ENV/JM/MONO(2020)21  83 Unclassified Shafer G (1976) A Mathematical Theory of Evidence, Princeton University Press, ISBN 9780691100425, 314 pp. Sonoda T, Ohdo S, Ohba K, Okishima T, Hayakawa K (1990), Teratogenic effects of sodium valproate in the Jcl: ICR mouse fetus, Acta Paediatrica Japonica, Vol. 32, pp. 502-7. Sonneveld E, Jansen HJ, Riteco JAC, Brouwer A, Van der Burg B. (2005), Development of androgenand estrogen-responsive bioassays, members of a panel of human cell line-based highly selective steroid responsive bioassays, Toxicological Sciences, Vol. 83, pp. 136–48. Sonneveld, E., Riteco, J.A.C., Jansen, H.J., Pieterse, B., Brouwer, A., Schoonen, W.G., Van der Burg, B. (2006), Comparison of in vitro and in vivo screening models for androgenic and estrogenic activities, Toxicol. Sci., Vol. 89, pp. 173-87. Sonneveld, E., Pieterse, B., Schoonen, W., Van der Burg, B. (2011), Validation of in vitro screening models for progestagenic activities: inter-assay comparison and correlation with in vivo activity in rabbits, Toxicology in vitro, Vol. 25, pp. 545-554. Strähle, U., Scholz, S., Geisler, R., Greiner, P., Hollert, H., Rastegar, S., Schumacher, A., Selderslaghs, I., Weiss, C., Witters, H., Braunbeck, T. (2012), Zebrafish embryos as an alternative to animal experiments – A commentary on the definition of the onset of protected life stages in animal welfare regulations, Repro. Tox, Vol. 33, pp. 128-132. Thomas SV, Ajaykumar B, Sindhu K, Francis E, Namboodiri N, Sivasankaran S et al. (2008), Cardiac malformations are increased in infants of mothers with epilepsy, Pediatric Cardiology, Vol. 29, pp. 604-8. Tomson T, Battino D. (2009), Teratogenic effects of antiepileptic medications - Review. Neurologic Clinics, Vol. 27, pp. 993-1002. Tomson T. (2005), Gender aspects of pharmacokinetics of new and old AEDs: pregnancy and breastfeeding, Therapeutic Drug Monitoring, Vol. 27, pp. 718-21. Turner S, Sucheston ME, De Philip RM, Paulson RB (1990), Teratogenic effects on the neuroepithelium of the CD-1 mouse embryo exposed in utero to sodium valproate, Teratology, Vol. 41, pp. 421-42. Vugt-Lussenburg, van, BMA, Van der Lee RB, Man HY, Middelhof I, Brouwer A, Besselink H, Van der Burg B. (2018), Incorporation of metabolic enzymes to improve predictivity of reporter gene assay results for estrogenic and anti-androgenic activity, Reproductive Toxicol., Vol. 75, pp. 40-48. Wagner, C. K. (2008), "Progesterone Receptors and Neural Development: A Gap between Bench and Bedside?", Endocrinology Vol. 149, No. 6, pp. 2743-2749. Walker V and Mills GA (2001) Urine 4-Heptanone: a beta-Oxidation Product of 2-Ethylhexanoic Acid from Plasticisers, Clinica Chimica Acta, Vol. 306, pp. 51-61. Waldmann et al. (2014), Design Principles of Concentration-Dependent Transcriptome Deviations in Drug-Exposed Differentiating Stem Cells, Chem Res Toxicol, Vol. 27, pp. 408-420. Wetendorf, M. and F. J. DeMayo (2012), "The progesterone receptor regulates implantation, decidualization, and glandular development via a complex paracrine signaling network.", Mol Cell Endocrinol, Vol. 357, No. 1-2, pp. 108-18. (1872-8057 (Electronic)). Wiltse J (2005), Mode of action: inhibition of histone deacetylase, altering WNT-dependent gene expression, and regulation of beta-catenin - developmental effects of valproic acid, Critical Reviews in Toxicology, Vol. 35, pp. 727-738. 84  ENV/JM/MONO(2020)21 Unclassified Winiwarter, S., Ax, F., Lennernas, H., Hallberg, A., Pettersson, C., Karlen, A. (2003), Hydrogen bonding descriptors in the prediction of human in vivo intestinal permeability, J Mol Graph Model, Vol. 21, pp. 273-287. Winiwarter, S., Bonham, N.M., Ax, F., Hallberg, A., Lennernas, H., Karlen, A. (1998), Correlation of human jejunal permeability (in vivo) of drugs with experimentally and theoretically derived parameters. A multivariate data analysis approach, J Med Chem, Vol. 41, pp. 4939-4949. ENV/JM/MONO(2020)21  85 Unclassified Annex I. Data-matrix Source and Target Compounds Source 1 Source 2 Target Source 3 Source 4 Source 5 Outlier 1 Outlier 2 Outlier 3 CAS 88-09-5 97-61-0 4536-23-6 149-57-5 99-66-1 31080-39-4 1185-39-3 1575-72-0 109-52-4 Name 2-ethyl butyric acid 2-methyl pentanoic acid 2-methyl hexanoic acid 2-ethyl hexanoic acid 2-propyl pentanoic acid 2-propyl heptanoic acid 2-dimethyl pentanoic acid 2-propyl pentenoic acid pentenoic acid Abbreviation EBA MPA MHA EHA VPA PHA DMPA 4-ene-VPA PA Structure Structure (smiles) CCC(CC)C(=O)O CCCC(C)C(=O)O CCCCC(C)C(=O)O CCCCC(CC)C(=O)O CCCC(CCC)C(=O)O CCCCCC(CCC)C(=O)O CCCC(C)(C)C(=O)O CCCC(CC=C)C(O)=O C=CCCC(=O)O Chain length from position 2 2 / 2 / 0 3 / 1 / 0 4 / 1 / 0 4 / 2 / 0 3 / 3 / 0 5 / 3 / 0 3 / 1 / 1 3 / 3 / 0 3 / 0 / 0 Structural similarity rel. to 4536-23-6 RDKit Fingerprint 59 83 100 82 69 77 – – – FCFP4 Fingerprint 55 83 100 68 62 52 – – – Layered Fingerprint 60 84 100 86 78 81 – – – RDKit Descriptors 72 76 100 76 75 28 – – – Summary of Data Gap Filling Source 1 Source 2 Target Source 3 Source 4 Source 5 Outlier 1 Outlier 2 Outlier 3 CAS 88-09-5 97-61-0 4536-23-6 149-57-5 99-66-1 31080-39-4 1185-39-3 1575-72-0 109-52-4 Name 2-ethyl butyric acid 2-methyl pentanoic acid 2-methyl hexanoic acid 2-ethyl hexanoic acid 2-propyl pentanoic acid 2-propyl heptanoic acid 2-dimethyl pentanoic acid 2-propyl pentenoic acid pentenoic acid 86  ENV/JM/MONO(2020)21 Unclassified Target endpoint 1 (neuro-) developmental toxicity in rodent studies Experimental result: NOAEL rats & mice No studies available No studies available Rats: neurodev.effects at 500 mg/kg bw; dev. effects at 250 mg/kg bw. Mice: (neuro)dev.effects at 200 mg/kg bw Rats: neurodev.effects at 720 mg/kg bw; dev. effects at 100 mg/kg bw. No studies available No studies available No studies available No studies available Integrated conclusion derived result Target endpoint 2 exencephaly-induction in NMRI mice (Nau et al., 1986) Experimental result: Potency relative to VPA Negative No studies available Positive: + Positive: +++ Positive: +++ Negative Positive: ++ Negative Integrated conclusion derived result Molecular Profiling Related to the Analogue Approach Hypothesis Source 1 Source 2 Target Source 3 Source 4 Source 5 Outlier 1 Outlier 2 Outlier 3 CAS 88-09-5 97-61-0 4536-23-6 149-57-5 99-66-1 31080-39-4 1185-39-3 1575-72-0 109-52-4 Name 2-ethyl butyric acid 2-methyl pentanoic acid 2-methyl hexanoic acid 2-ethyl hexanoic acid 2-propyl pentanoic acid 2-propyl heptanoic acid 2-dimethyl pentanoic acid 2-propyl pentenoic acid pentenoic acid Parent chemical Profiler 1 (name, version) Expert system 1 (name, version) Metabolites* Profiler 1 (name, version) Expert system 1 (name, version) Physico-Chemical Data Source 1 Source 2 Target Source 3 Source 4 Source 5 Outlier 1 Outlier 2 Outlier 3 CAS 88-09-5 97-61-0 4536-23-6 149-57-5 99-66-1 31080-39-4 1185-39-3 1575-72-0 109-52-4 ENV/JM/MONO(2020)21  87 Unclassified Name 2-Ethyl butyric acid 2-Methylpentanoic acid 2-Methylhexanoic acid 2-Ethylhexanoic acid Valproic acid 2-Propyl heptanoic acid 2-dimethyl pentanoic acid 4-ene-VPA pentenoic acid Molecular weight 116.08 116.08 130.1 144.12 144.12 172.15 130.187 142.2 102.133 Melting point (deg C; predicted EPISUITE) 15.24 15.24 26.62 37.72 37.72 59.15 24.61 36.44 14.76 Boiling point (deg C; predicted EPISUITE) 195.8 195.8 215.45 234.2 234.2 268.98 207.77 232.83 187.75 logPow (experimental EPISUITE) 1.68 1.8 1.8 2.64 2.75 3.2 1.39 logPow (predicted EPISUITE) 1.98 1.98 2.47 2.96 2.96 3.94 2.43 2.82 1.56 Water solubility (mg/L; experimental) 18000 20°C (41) 2000 20°C (16) 2000 at 20°C (37) 275.6 at 25°C (24) 24000 at 25°C Water solubility (mg/L; human metabolome) pKa (experimental) 4.71 (43) 4.7 (19) 4.6 (39) pKa (predicted ACD/Percepta) 4.8 4.8 4.8 4.8 4.8 4.8 4.9 4.7 4.84 Vapour pressure (mm Hg; experimental) 0.188 at 25°C (44) 0.03 (18) 0.0458 at 25°C (40) 0.0048 (26) Henry's Law Constant (Pa m3/mol; experimental EPISUITE) 0.289 Henry's Law Constant (Pa m3/mol; predicted EPISUITE (bond method)) 0.162 0.172 0.229 0.304 0.304 0.535 0.229 0.226 0.130 ADME-Toxicokinetics Source 1 Source 2 Target Source 3 Source 4 Source 5 Outlier 1 Outlier 2 Outlier 3 CAS 88-09-5 97-61-0 4536-23-6 149-57-5 99-66-1 31080-39-4 1185-39-3 1575-72-0 109-52-4 Name 2-Ethyl butyric acid 2-Methylpentanoic acid 2-Methylhexanoic acid 2-Ethylhexanoic acid Valproic acid 2-Propyl heptanoic acid 2-dimethyl pentanoic acid 4-ene-VPA pentenoic acid Fraction unbound in plasma (human; predicted – Model 1 Mirko-all in domain) 0.348 0.348 0.296 0.138 0.141 0.135 0.282 0.210 0.431 88  ENV/JM/MONO(2020)21 Unclassified Fraction unbound in plasma (human; predicted – Model 2 Mirko-all in domain) 0.46 0.460 0.382 0.245 0.310 0.191 0.471 0.338 0.434 Fraction unbound in plasma (human; predicted – Model 3 Mirko-all in domain) 0.413 0.359 0.265 0.155 0.193 0.081 0.268 0.307 0.592 Fraction unbound in plasma (human; predicted) - Model 4 (Lhasa-all in domain) 0.712 0.671 0.454 0.201 0.138 0.263 NA NA NA Intrinsic Hepatic Clearance (CLint,H; µl/min/106 Hepatocytes) (human; experimental Cyprotex) 9.620 10.200 3.950 0.551 0.219 NA NA NA NA In vivo Clearance (CL; L/h) (human; predicted Simcyp, min fu) 20.42 21.09 9.10 0.85 0.33 NA NA NA NA In vivo Clearance (CL; L/h) (human; predicted Simcyp, max fu) 29.09 29.07 13.68 1.80 0.73 NA NA NA NA In vivo Clearance (CL; mL/min) (mouse; predicted allometric scaling, min fu) 0.795 0.821 0.354 0.033 0.013 NA NA NA NA In vivo Clearance (CL; mL/min) (mouse; predicted allometric scaling, max fu) 1.132 1.131 0.532 0.070 0.028 NA NA NA NA Steady-state volume of distribution, human (Vss; L/kg; predicted) 0.16 0.16 0.15 0.14 0.14 0.12 0.20 0.13 0.35 Steady-state volume of distribution, human (Vu,ss; L/kg; predicted) 0.35 0.35 0.40 0.57 0.45 0.63 0.33 0.39 0.15 Steady-state volume of distribution, mouse (Vss; L/kg; predicted) 0.30 0.17 0.27 0.13 0.14 0.11 0.40 0.24 0.16 Steady-state volume of distribution, mouse (Vu,ss; L/kg; predicted) 0.65 0.37 0.70 0.53 0.45 0.59 0.63 0.70 0.37 ENV/JM/MONO(2020)21  89 Unclassified Intrinsic clearance -Insphero Supporting data related to the target endpoint(s) Source 1 Source 2 Target Source 3 Source 4 Source 5 Outlier 1 Outlier 2 Outlier 3 CAS 88-09-5 97-61-0 4536-23-6 149-57-5 99-66-1 31080-39-4 1185-39-3 1575-72-0 591-80-0 Name EBA MPA MHA EHA VPA PHA DMPA 4-ene-VPA PA In vitro ZET assay Effect: (most sensitive) pericardial and/or yolk oedema morphological observation 24-48 hpf EC10 nominal (µM) 363 373 385 65 38 10 426 Not available yet 568 Effect: small eyes morphological observation 24-48 hpf EC10 nominal (µM) (1000)1 >9002 >1000 (400)1 (400)1 153 540 Not available yet >1400 Effect: jitter/tremor behavioural observation 96-120 hpf EC10 nominal (µM) >1000 >900 442 421 82 16 977 Not available yet >1400 Effect: craniofacial deformation morphological observation 96-120 hpf EC10 nominal (µM) >1000 380 >1000 102 52 21 481 Not available yet 626 Effect: scoliosis/lordosis morphological observation 96-120 hpf EC10 nominal (µM) 892 393 383 651 59 41 663 Not available yet 747 ZET Reporter assay CHA angle 96  ENV/JM/MONO(2020)21 Unclassified human oral dose equivalent (max fu model) (mg/kg) 86.3 7.36 3.5 mouse oral dose equivalent (min fu model) (mg/kg) 14.4 1.36 0.69 mouse oral dose equivalent (max fu model) (mg/kg) 24.1 1.8 0.925 scoliosis/lordosis EC10 Nominal (µM) 892 393 383 651 59 41 663 N/A 747 human oral dose equivalent (min fu model) (mg/kg) 164 75.2 54.2 82.6 7.7 human oral dose equivalent (max fu model) (mg/kg) 154.6 71.4 47.7 53 4.8 mouse oral dose equivalent (min fu model) (mg/kg) 145.5 66.4 56.7 109.9 10.4 mouse oral dose equivalent (max fu model) (mg/kg) 144.2 65.9 46.96 72.7 6.7 scoliosis/lordosis embryototal, model (µM) 144 98 96 130 12 5 166 N/A 187 human oral dose equivalent (min fu model) (mg/kg) 121.2 85.7 57.5 61.1 5.4 human oral dose equivalent (max fu model) (mg/kg) 125.8 89 55.1 47.84 4.2 mouse oral dose equivalent (min fu model) (mg/kg) 21.1 14.9 10.1 8.86 0.83 mouse oral dose equivalent (max fu model) (mg/kg) 36.8 24.9 13.2 11.8 1.11 ZET Reporter assay EC10 Nominal Treatment Conc. (µM) 856 194 N/A 12 1.7 1 174 7 N/A human oral dose equivalent (min fu model) (mg/kg) 157.4 37.1 1.5 0.22 human oral dose equivalent (max fu model) (mg/kg) 148.4 35.2 0.98 0.138 mouse oral dose equivalent (min fu model) (mg/kg) 139.6 32.8 2.02 0.3 mouse oral dose equivalent (max fu model) (mg/kg) 138.4 32.5 1.34 0.19 EC10embryo-total, model (µM) 143 49 N/A 2 0.3 0.1 44 1 N/A ENV/JM/MONO(2020)21  97 Unclassified human oral dose equivalent (min fu model) (mg/kg) 120.30 42.70 0.94 0.135 human oral dose equivalent (max fu model) (mg/kg) 124.90 44.50 0.74 0.105 mouse oral dose equivalent (min fu model) (mg/kg) 20.90 7.45 0.136 0.020 mouse oral dose equivalent (max fu model) (mg/kg) 36.50 12.45 0.18 0.028 mEST assay ID10 D3 µM (medium-unbound) 875 406 189 116 212 71 human oral dose equivalent (min fu model) (mg/kg) 124.00 51.6 24.5 human oral dose equivalent (max fu model) (mg/kg) 108.90 33 16 mouse oral dose equivalent (min fu model) (mg/kg) 129.50 69 33 mouse oral dose equivalent (max fu model) (mg/kg) 107.20 45.3 21 IC10 D3 µM (medium-unbound) 2000 2601 285 123 222 human oral dose equivalent (min fu model) (mg/kg) 367.8 497.50 36.9 human oral dose equivalent (max fu model) (mg/kg) 347.00 472.30 23 mouse oral dose equivalent (min fu model) (mg/kg) 326.20 439.40 50 mouse oral dose equivalent (max fu model) (mg/kg) 323.40 436.00 32 IC10 3T3 µM (mediumunbound) 1762 6552 7101 452 229 78 human oral dose equivalent (min fu model) (mg/kg) 337.0 928.40 901.3 human oral dose equivalent (max fu model) (mg/kg) 320.00 815.30 578 mouse oral dose equivalent (min fu model) (mg/kg) 297.70 969.10 1199 mouse oral dose equivalent (max fu model) (mg/kg) 295.40 802.60 793 98  ENV/JM/MONO(2020)21 Unclassified HDAC inhibition-day 4 EC10medium-unbound (µM) 4496 7076 252 34 17 253 935 human oral dose equivalent (min fu model) (mg/kg) 826.6 1353.30 32 4.4 human oral dose equivalent (max fu model) (mg/kg) 779.20 1284.80 21 3 mouse oral dose equivalent (min fu model) (mg/kg) 733.30 1195.00 43 6 mouse oral dose equivalent (max fu model) (mg/kg) 727.00 1186.30 28.1 4 HDAC inhibition-day 10 EC10medium-unbound (µM) 2064 1438 2301 119 27 13 3051 32 30 human oral dose equivalent (min fu model) (mg/kg) 379.6 275.00 326.10 15 3.49 human oral dose equivalent (max fu model) (mg/kg) 357.80 261.10 286.30 10 2 mouse oral dose equivalent (min fu model) (mg/kg) 336.60 242.90 340.30 20 5 mouse oral dose equivalent (max fu model) (mg/kg) 333.70 241.10 281.90 13.3 3 UKN1 assay Viability EC10medium-unbound (µM) 184 215 43 1164 121 1710 human oral dose equivalent (min fu model) (mg/kg) 23 28 human oral dose equivalent (max fu model) (mg/kg) 15 18 mouse oral dose equivalent (min fu model) (mg/kg) 31 38 mouse oral dose equivalent (max fu model) (mg/kg) 21 24 HDAC inhibition EC10 mediumunbound (µM) 931 1028 2048 70 34 5 707 37 85 human oral dose equivalent (min fu model) (mg/kg) 171.20 196.60 290.20 8.9 4 human oral dose equivalent (max fu model) (mg/kg) 161.40 186.60 254.80 6 3 ENV/JM/MONO(2020)21  99 Unclassified mouse oral dose equivalent (min fu model) (mg/kg) 151.80 173.60 302.90 12 6 mouse oral dose equivalent (max fu model) (mg/kg) 150.50 172.40 250.90 7.8 4 CALUX assays EC10cell-total (µM) min. 41 41 4 7 9 3 24 9 38 max. 41 41 4 70 110 100 38 110 76 human oral dose equivalent (min fu model) (mg/kg) min. 34.4 35.8 2.4 3.29 3.95 max. 34.4 35.8 2.4 32.9 49.5 human oral dose equivalent (max fu model) (mg/kg) min. 35.8 37.3 2.3 2.6 3.2 max. 35.8 37.3 2.3 25.7 38.3 mouse oral dose equivalent (min fu model) (mg/kg) min. 6 6.2 0.42 0.48 0.62 max. 6 6.2 0.42 4.8 7.6 mouse oral dose equivalent (max fu model) (mg/kg) min. 10.5 10.4 0.55 0.63 0.83 max. 10.5 10.4 0.55 6.4 10.14 In chemico Source 1 Source 2 Target Source 3 Source 4 Source 5 Outlier 1 Outlier 2 Outlier 3 CAS 88-09-5 97-61-0 4536-23-6 149-57-5 99-66-1 31080-39-4 1185-39-3 1575-72-0 591-80-0 Name EBA MPA MHA EHA VPA PHA DMPA 4-ene-VPA PA In silico models of OECD QSAR Toolbox DNA binding by OASIS No alert found No alert found No alert found No alert found No alert found No alert found DNA binding by OECD No alert found No alert found No alert found No alert found No alert found No alert found Eye irritation/corrosion Exclusion rules by BfR Undefined Undefined Undefined Undefined Undefined Group C Melting Point > 55 C|Undefined Eye irritation/corrosion Inclusion rules by BfR Inclusion rules not met Inclusion rules not met Inclusion rules not met Inclusion rules not met Inclusion rules not met Inclusion rules not met Protein Binding Potency hCLAT No alert found No alert found No alert found No alert found No alert found No alert found 100  ENV/JM/MONO(2020)21 Unclassified Protein binding alerts for Chromosomal aberration by OASIS No alert found No alert found No alert found No alert found No alert found No alert found Protein binding alerts for skin sensitization according to GHS No alert found No alert found No alert found No alert found No alert found No alert found Protein binding alerts for skin sensitization by OASIS No alert found No alert found No alert found No alert found No alert found No alert found Protein binding by OASIS No alert found No alert found No alert found No alert found No alert found No alert found Protein binding by OECD No alert found No alert found No alert found No alert found No alert found No alert found Protein binding potency GSH Not possible to classify Not possible to classify Not possible to classify Not possible to classify Not possible to classify Not possible to classify Protein binding potency Cys (DPRA 13%) NonConjugated carboxylic acids and esters (non reactive) NonConjugated carboxylic acids and esters (non reactive) NonConjugated carboxylic acids and esters (non reactive) Non-Conjugated carboxylic acids and esters (non reactive) Non-Conjugated carboxylic acids and esters (non reactive) Non-Conjugated carboxylic acids and esters (non reactive) Protein binding potency Lys (DPRA 13%) NonConjugated carboxylic acids and esters (non reactive) NonConjugated carboxylic acids and esters (non reactive) NonConjugated carboxylic acids and esters (non reactive) Non-Conjugated carboxylic acids and esters (non reactive) Non-Conjugated carboxylic acids and esters (non reactive) Non-Conjugated carboxylic acids and esters (non reactive) Skin irritation/corrosion Exclusion rules by BfR Undefined Undefined Undefined Undefined Undefined Group C Melting Point > 55 C|Undefined Skin irritation/corrosion Inclusion rules by BfR Aliphatic acids Aliphatic acids Aliphatic acids Aliphatic acids Aliphatic acids Inclusion rules not met ENV/JM/MONO(2020)21  101 Unclassified Annex II. In vitro models - detailed description of methods and generated data Please refer to the separate publication for full Annex II ENV/JM/MONO(2020)21/ANN2 102  ENV/JM/MONO(2020)21 Unclassified Annex III. In silico models - detailed description of methods and generated data Please refer to the separate publication for full Annex III ENV/JM/MONO(2020)21/ANN3