A quantitative in silico model for predicting skin sensitization using a nearest neighbours approach within expert‐derived structure–activity alert spaces

Dermal contact with chemicals may lead to an inflammatory reaction known as allergic contact dermatitis. Consequently, it is important to assess new and existing chemicals for their skin sensitizing potential and to mitigate exposure accordingly. There is an urgent need to develop quantitative non‐animal methods to better predict the potency of potential sensitizers, driven largely by European Union (EU) Regulation 1223/2009, which forbids the use of animal tests for cosmetic ingredients sold in the EU. A Nearest Neighbours in silico model was developed using an in‐house dataset of 1096 murine local lymph node (LLNA) studies. The EC3 value (the effective concentration of the test substance producing a threefold increase in the stimulation index compared to controls) of a given chemical was predicted using the weighted average of EC3 values of up to 10 most similar compounds within the same mechanistic space (as defined by activating the same Derek skin sensitization alert). The model was validated using previously unseen internal (n = 45) and external (n = 103) data and accuracy of predictions assessed using a threefold error, fivefold error, European Centre for Ecotoxicology and Toxicology of Chemicals (ECETOC) and Globally Harmonized System of Classification and Labelling of Chemicals (GHS) classifications. In particular, the model predicts the GHS skin sensitization category of compounds well, predicting 64% of chemicals in an external test set within the correct category. Of the remaining chemicals in the previously unseen dataset, 25% were over‐predicted (GHS 1A predicted: GHS 1B experimentally) and 11% were under‐predicted (GHS 1B predicted: GHS 1A experimentally). Copyright © 2017 John Wiley & Sons, Ltd.     

Assessing the potency of respiratory allergens: uncertainties and challenges

In contrast to skin sensitisation, there are no accepted tests for the identification of chemicals or proteins with the potential to cause sensitisation of the respiratory tract. Although progress has been made, the assessment of respiratory sensitisation potential remains associated with significant challenges and uncertainties. Nevertheless, there is interest in determining whether it is possible to assess the relative potency of respiratory sensitisers. The second Adaptation to Technical Progress (ATP) to the EU Classification, Labelling and Packaging (CLP) Regulation recently introduced changes to criteria for classification and labelling of chemicals and preparations, bringing it in line with the 3rd revision to the UN Globally Harmonised System of Classification and Labelling of Chemicals (GHS). Among other things, the second ATP introduces sub-categories for respiratory and skin sensitisers, discriminating between strong sensitisers and other sensitisers. Here we examine whether such categorisation of protein and/or chemical respiratory allergens is realistic and/or feasible. For this purpose comparisons have been drawn with skin sensitisation, where potency categorisation has now been widely accepted and successfully integrated into the regulatory process. The conclusion drawn is that, on the basis of the currently available information, potency categorisation for respiratory sensitisers is premature and could potentially be misleading.     

Science based guidance for the assessment of endocrine disrupting properties of chemicals

The European legislation on plant protection products (Regulation (EC) No. 1107/2009) and biocides (Directive 98/8/EC), as well as the regulation concerning chemicals (Regulation (EC) No. 1907/2006 ‘REACH’) only support the marketing and use of chemical products on the basis that they do not induce endocrine disruption in humans or non-target species. However, there is currently no agreed guidance on how to identify and evaluate endocrine activity and disruption. Consequently, an ECETOC task force was formed to provide scientific criteria that may be used within the context of these three legislative documents. Specific scientific criteria for the determination of endocrine disrupting properties that integrate information from both regulatory (eco)toxicity studies and mechanistic/screening studies are proposed. These criteria combine the nature of the adverse effects detected in studies which give concern for endocrine toxicity with an understanding of the mode of action of toxicity so that adverse effects can be explained scientifically. The criteria developed are presented in the form of flow charts for assessing relevant effects for both humans and wildlife species. In addition, since not all chemicals with endocrine disrupting properties are of equal hazard, assessment of potency is also proposed to discriminate chemicals of high concern from those of lower concern. The guidance presented in this paper includes refinements made to an initial proposal following discussion of the criteria at a workshop of invited regulatory, academic and industry scientists.     

Insights into possibilities for grouping and read-across for nanomaterials in EU chemicals legislation

This paper presents a comprehensive review of European Union (EU) legislation addressing the safety of chemical substances, and possibilities within each piece of legislation for applying grouping and read-across approaches for the assessment of nanomaterials (NMs). Hence, this review considers both the overarching regulation of chemical substances under REACH (Regulation (EC) No 1907/2006 on registration, evaluation, authorization, and restriction of chemicals) and CLP (Regulation (EC) No 1272/2008 on classification, labeling and packaging of substances and mixtures) and the sector-specific pieces of legislation for cosmetic, plant protection and biocidal products, and legislation addressing food, novel food, and food contact materials. The relevant supporting documents (e.g. guidance documents) regarding each piece of legislation were identified and reviewed, considering the relevant technical and scientific literature. Prospective regulatory needs for implementing grouping in the assessment of NMs were identified, and the question whether each particular piece of legislation permits the use of grouping and read-across to address information gaps was answered.     

Exposure-driven risk assessment: Applying exposure-based waiving of toxicity tests under REACH

The REACH Regulation 1907/2006/EC aims to improve knowledge of the potential risks to humans and the environment of the large number of chemicals produced and used in the EU. The testing requirements are likely to trigger numerous toxicological studies, potentially involving millions of experimental animals, despite the professed goal of REACH to reduce vertebrate testing. It may be necessary therefore to shift emphasis away from animal studies towards more pragmatic strategies, reserving animal tests for the substances of greatest concern. One approach is to waive certain tests based on levels of exposure to the substance. This review explores application of ‘Exposure-Based Waiving’ (EBW) of toxicity studies, with a particular focus on inhalation where possible, considering the potential qualitative and quantitative supporting arguments that might be made, including the use of thresholds of toxicological concern. Incorporating EBW into intelligent testing strategies for substance registration could advance the goals of REACH and the 3Rs (reduction, replacement and refinement of animals in research) by reducing the usage of animals in toxicity tests, whilst maintaining appropriate protection of human health and the environment. However greater regulatory evaluation, acceptance and guidance are required for EBW to achieve its full impact.     

Threshold for classification as a skin sensitizer in the local lymph node assay: a statistical evaluation

For more than 15 years, the murine local lymph node assay (LLNA) has undergone development, evaluation and validation as an alternative approach to the predictive identification of skin sensitizing chemicals. The criteria by which sensitizing chemicals are distinguished from those without significant skin sensitising hazard were developed empirically and were based on experience rather than a mathematical formula or statistical method. The current practice is to classify, as skin sensitizers, those chemicals which at one or more test concentrations stimulate a threefold or greater increase in the proliferative activity in draining lymph node cells. Despite the apparent confirmation of the utility of this approach from the extensive data available, there has not previously been any attempt to substantiate the accuracy of this criterion. In this present investigations, data from 134 chemicals tested in the LLNA and in the guinea pig and/or for which there exists clear evidence relating to human skin sensitization potential, have been subjected to a rigorous statistical evaluation using Receiver Operating Characteristic (ROC) curves. Whether the analysis is based on a comparison with guinea pig or human data, the results indicate that the empirically derived threefold threshold is an acceptable practical value for hazard identification.     

Practical issues on the application of the GHS classification criteria for germ cell mutagens

The Globally Harmonized System of Classification and Labeling of Chemicals (GHS) requires classification of chemicals on germ cell mutagenicity. The Japanese government has conducted GHS classification on about 1400 chemicals in a 2-year project (J-GHS) for implementing GHS domestically. Prior to the classification work, the technical guidance for classification of germ cell mutagens was prepared. This guidance introduces the concept of heritable mutagenicity, and presents detailed criteria for germ cell mutagens, test data to be used, and a practical decision tree for classification. These practical guidance and supporting explanations are useful for non-expert Classifiers (scientists applying the classification criteria). Several issues, however, were identified during the course of J-GHS and in re-evaluating the classification results. These include: (1) the information sources when available data are limited; (2) lack of understanding GHS classification criteria or insufficient review of the information by Classifiers; (3) varying opinions of experts on data quality and weight of evidence, and; (4) decision tree approaches, e.g., inadequacy for use in overall evaluation in some cases. Ideally, classification should be performed by Classifiers with high expertise using high quality information sources. Genetic toxicologists as experts should consider data quality and reliability, and give a critical review of all available information for support of classification. A weight of evidence approach is also required to assess mutagenic potential of chemicals. Critical points for suitable classification for GHS are discussed.     

Inter‐relationships between different classes of chemical allergens

Although allergic sensitization of the respiratory tract induced by chemicals is not as common as skin sensitization, it is nevertheless an important occupational health issue. Respiratory allergy to chemicals, characterized typically by rhinitis and asthma, is associated with considerable morbidity and with related socioeconomic costs. Several experimental approaches have been proposed for the prospective identification of chemical respiratory allergens, but none of these has yet been validated formally. In the absence of a widely accepted method for respiratory allergen identification, it is appropriate and relevant to explore their relationship with skin‐sensitizing chemicals. A series of chemicals known to cause immune‐mediated respiratory allergy in humans has been examined. The majority of the respiratory allergens tested were found to elicit positive responses in one or more standard tests used for the identification of skin‐sensitizing potential (guinea pig maximization test, the Buehler test and/or the local lymph node assay). We suggest that this observation might form the basis of a potentially useful paradigm for initial characterization of the respiratory‐sensitizing potential of chemicals. Specifically, chemicals that fail to elicit positive responses in accepted skin‐sensitization test methods might also be regarded as lacking the inherent potential to cause allergic sensitization of the respiratory tract. Copyright © 2012 John Wiley & Sons, Ltd.     

Novel approach for classifying chemicals according to skin sensitizing potency by non-radioisotopic modification of the local lymph node assay

The murine local lymph node assay (LLNA) is currently recognized as a stand-alone sensitization test for determining the sensitizing potential of chemicals, and it has the advantage of yielding a quantitative endpoint that can be used to predict the sensitization potency of chemicals. The EC3 has been proposed as a parameter for classifying chemicals according to the sensitization potency. We previously developed a non-radioisotopic endpoint for the LLNA based on 5-bromo-2'-deoxyuridine (BrdU) incorporation (non-RI LLNA), and we are proposing a new procedure to predict the sensitization potency of chemicals based on comparisons with known human contact allergens. Nine chemicals (i.e. diphencyclopropenone, p-phenylenediamine, glutaraldehyde, cinnamicaldehyde, citral, eugenol, isopropyl myristate, propyleneglycol and hexane) categorized as human contact allergen classes 1-5 were tested by the non-RI LLNA with the following reference allergens: 2,4-dinitrochlorobenzene (DNCB) as a class 1 human contact allergen, isoeugenol as a class 2 human contact allergen and alpha-hexylcinnamic aldehyde (HCA) as a class 3 human contact allergen. Consequently, nine test chemicals were almost assigned to their correct allergen class. The results suggested that the new procedure for non-RI LLNA can provide correct sensitization potency data. Sensitization potency data are useful for evaluating the sensitization risk to humans of exposure to new chemical products. Accordingly, this approach would be an effective modification of LLNA with regard to its experimental design. Moreover, this procedure can be applied also to the standard LLNA with radioisotopes and to other modifications of the LLNA.     

Correlation between experimental human and murine skin sensitization induction thresholds

Quantitative risk assessment for skin sensitization is directed towards the determination of levels of exposure to known sensitizing substances that will avoid the induction of contact allergy in humans. A key component of this work is the predictive identification of relative skin sensitizing potency, achieved normally by the measurement of the threshold (the “EC3” value) in the local lymph node assay (LLNA). In an extended series of studies, the accuracy of this murine induction threshold as the predictor of the absence of a sensitizing effect has been verified by conduct of a human repeated insult patch test (HRIPT). Murine and human thresholds for a diverse set of 57 fragrance chemicals spanning approximately four orders of magnitude variation in potency have been compared. The results confirm that there is a useful correlation, with the LLNA EC3 value helping particularly to identify stronger sensitizers. Good correlation (with half an order of magnitude) was seen with three-quarters of the dataset. The analysis also helps to identify potential outlier types of (fragrance) chemistry, exemplified by hexyl and benzyl salicylates (an over-prediction) and trans-2-hexenal (an under-prediction).     

Assessing the potency of respiratory allergens: Uncertainties and challenges

In contrast to skin sensitisation, there are no accepted tests for the identification of chemicals or proteins with the potential to cause sensitisation of the respiratory tract. Although progress has been made, the assessment of respiratory sensitisation potential remains associated with significant challenges and uncertainties. Nevertheless, there is interest in determining whether it is possible to assess the relative potency of respiratory sensitisers. The second Adaptation to Technical Progress (ATP) to the EU Classification, Labelling and Packaging (CLP) Regulation recently introduced changes to criteria for classification and labelling of chemicals and preparations, bringing it in line with the 3rd revision to the UN Globally Harmonised System of Classification and Labelling of Chemicals (GHS). Among other things, the second ATP introduces sub-categories for respiratory and skin sensitisers, discriminating between strong sensitisers and other sensitisers. Here we examine whether such categorisation of protein and/or chemical respiratory allergens is realistic and/or feasible. For this purpose comparisons have been drawn with skin sensitisation, where potency categorisation has now been widely accepted and successfully integrated into the regulatory process. The conclusion drawn is that, on the basis of the currently available information, potency categorisation for respiratory sensitisers is premature and could potentially be misleading.     

Sub-categorisation of skin corrosive chemicals by the EpiSkin™ reconstructed human epidermis skin corrosion test method according to UN GHS: Revision of OECD Test Guideline 431

The EpiSkin™ skin corrosion test method was formally validated and adopted within the context of OECD TG 431 for identifying corrosive and non-corrosive chemicals. The EU Classification, Labelling and Packaging Regulation (EU CLP) system requires the sub-categorisation of corrosive chemicals into the three UN GHS optional subcategories 1A, 1B and 1C. The present study was undertaken to investigate the usefulness of the validated EpiSkin™ test method to identify skin corrosive UN GHS Categories 1A, 1B and 1C using the original and validated prediction model and adapted controls for direct MTT reduction. In total, 85 chemicals selected by the OECD expert group on skin corrosion were tested in three independent runs. The results obtained were highly reproducible both within (>80%) and between (>78%) laboratories when compared with historical data. Moreover the results obtained showed that the EpiSkin™ test method is highly sensitive (99%) and specific (80%) in discriminating corrosive from non-corrosive chemicals and allows reliable and relevant identification of the different skin corrosive UN GHS subcategories, with high accuracies being obtained for both UN GHS Categories 1A (83%) and 1B/1C (76%) chemicals. The overall accuracy of the test method to subcategorise corrosive chemicals into three or two UN GHS subcategories ranged from 75% to 79%. Considering those results, the revised OECD Test Guideline 431 permit the use of EpiSkin™ for subcategorising corrosive chemicals into at least two classes (Category 1A and Category 1B/1C).     

Risk assessment of endocrine active chemicals: Identifying chemicals of regulatory concern

The European regulation on plant protection products (1107/2009) (EC, 2009a), the revisions to the biocides Directive (COM[2009]267) (EC, 2009b), and the regulation concerning chemicals (Regulation (EC) No. 1907/2006 ‘REACH’) (EC.2006) only support the marketing and use of chemical products on the basis that they do not induce endocrine disruption in humans or wildlife species. In the absence of agreed guidance on how to identify and evaluate endocrine activity and disruption within these pieces of legislation a European Centre for Ecotoxicology and Toxicology of Chemicals (ECETOC) task force was formed to provide scientific criteria that may be used within the context of these three legislative documents. The resulting ECETOC technical report (ECETOC, 2009a) and the associated workshop (ECETOC, 2009b) presented a science-based concept on how to identify endocrine activity and disrupting properties of chemicals for both human health and the environment. The synthesis of the technical report and the workshop report was published by the ECETOC task force (Bars et al., 2011a and Bars et al., 2011b). Specific scientific criteria for the determination of endocrine activity and disrupting properties that integrate information from both regulatory (eco)toxicity studies and mechanistic/screening studies were proposed. These criteria combined the nature of the adverse effects detected in studies which give concern for endocrine toxicity with an understanding of the mode of action of toxicity so that adverse effects can be explained scientifically. A key element in the data evaluation is the consideration of all available information in a weight-of-evidence approach. However, to be able to discriminate chemicals with endocrine properties of low concern from those of higher concern (for regulatory purposes), the task force recognised that the concept needed further refinement. Following a discussion of the key factors at a second workshop of invited regulatory, academic and industry scientists (ECETOC, 2011), the task force developed further guidance, which is presented in this paper. For human health assessments these factors include the relevance to humans of the endocrine mechanism of toxicity, the specificity of the endocrine effects with respect to other potential toxic effects, the potency of the chemical to induce endocrine toxicity and consideration of exposure levels. For ecotoxicological assessments the key considerations include specificity and potency, but also extend to the consideration of population relevance and negligible exposure. It is intended that these complement and reinforce the approach originally described and previously published in this journal (Bars et al., 2011a and Bars et al., 2011b).     

Exposure based waiving: The application of the toxicological threshold of concern (TTC) to inhalation exposure for aerosol ingredients in consumer products

The inhalation toxicology studies available in the public domain have been reviewed to establish a database for inhalation toxicology and derive thresholds of toxicological concern (TTC) for effects in the respiratory tract and systemically for Cramer class 1 and 3 chemicals. These TTCs can be used as the basis for developing an exposure based waiving (EBW) approach to evaluating the potential for adverse effects from exposure to ingredients in aerosol products, used by consumers. The measurement of consumer exposure in simulated product use is key to the application of an exposure based waiving approach to evaluating potential consumer risk. The detailed exposure evaluation for aerosol ingredients with defined use scenarios, in conjunction with an evaluation of the potential structure activity relationship for toxicity and the TTCs for inhalation exposure could be used to waive undertaking inhalation toxicology studies under REACH. Not all classes of chemicals are suitable for such an approach, but for chemicals with a predictable low potential toxicity, and very low levels of exposure, this approach, could reduce the amount of inhalation toxicology studies required for the implementation of the European REACH legislation. Such an approach is consistent with the concept of developing ‘intelligent testing strategies’ for REACH.     

Assessment of reproductive toxicity under REACH

The European REACH regulation requires the evaluation of reproductive toxicity in screening tests according to OECD TG 421 and 422 for substances above the tonnage level of 10 tons/year. The overall aim of this paper is to increase flexibility in combination with a reduced number of experimental animals. Therefore, in contrast to the existing approach the registrant should have the possibility to file a dossier for a substance at the level of 10 tons/year and above also on the basis of data from a developmental toxicity study (OECD TG 414) plus a full-scale subacute toxicity study (OECD TG 407 according to the 1995 protocol). The proposed new test strategy takes into account overall considerations of duty of care and animal welfare. It enables an assessment of developmental toxicity on a definitive instead of a screening level. Registrants should be allowed to select between these two options, either the existing approach (OECD TG 421/407 and alternatively TG 422) or the approach proposed in this paper (OECD TG 407 plus TG 414).     

Skin sensitization potency of isoeugenol and its dimers evaluated by a non-radioisotopic modification of the local lymph node assay and guinea pig maximization test

Allergic contact dermatitis is the serious unwanted effect arising from the use of consumer products such as cosmetics. Isoeugenol is a fragrance chemical with spicy, carnation-like scent, is used in many kinds of cosmetics and is a well-known moderate human sensitizer. It was previously reported that the dimerization of eugenol yielded two types of dimer possessing different sensitization potencies. This study reports the differences in skin sensitization potencies for isoeugenol and two types of dimer, beta-O-4-dilignol and dehydrodiisoeugenol (DIEG), as evaluated by the non-radioisotopic local lymph node assay (non-RI LLNA) and guinea pig maximization test. In the guinea pig maximization test, isoeugenol, beta-O-4-dilignol and DIEG were classified as extreme, weak and moderate sensitizers, respectively. As for the results of non-RI LLNA, the EC3 for isoeugenol, beta-O-4-dilignol and DIEG were calculated as 12.7%, >30% and 9.4%, respectively. The two types of isoeugenol dimer showed different sensitizing activities similar to the case for eugenol dimers. A reduction of sensitization potency achieved by dimerization may lead to developing safer cosmetic ingredients. Isoeugenol dimers are not currently used for fragrance chemicals. However, the dimerization of isoeugenol may yield a promising candidate as a cosmetic ingredient with low sensitization risk. The data may also provide useful information for the structure-activity relationship (SAR) in skin sensitization.     

Classification of sensitizing and irritative potential in a combined in-vitro assay

We have developed a coculture system which in parallel indicates the sensitizing and irritative potential of xenobiotics. The assay is named loose-fit coculture-based sensitization assay (LCSA) and may be performed within 5 days. The system is composed of human monocytes that differentiate to a kind of dendritic cells by 2-day culturing in the presence of allogenic keratinocytes. The culture medium is enriched by a cocktail of recombinant cytokines. On day 3, concentration series of probes are added. On day 5, cells are harvested and analyzed for expression range of CD86 as a marker of sensitizing potential and for uptake of the viability stain 7-AAD as a marker of irritative potential. Estimation of the concentration required to cause a half-maximal increase in CD86 expression allowed quantification of sensitizing potential, and estimation of the concentration required to reduce viability to 50% allowed quantification of irritative potential. Examination of substances with known potential resulted in categorization of test scores. To evaluate our data, we have compared results with those of the validated animal-based sensitization test, the murine local lymph node assay (LLNA, OECD TG 429). To a large extent, results from LCSA and from LLNA achieved analogous grouping of allergens into categories like weak-moderate-strong. However, the new assay showed an improved capacity to distinguish sensitizers from non-sensitizers and irritants. In conclusion, the LCSA contains potential to fulfil the requirements of the EU's programme for the safety of chemicals “Registration, Evaluation, Authorisation and Restriction of chemical substances” (REACH, 2006) to replace animal models.     

Advantage of using CBA/N strain mice in a non-radioisotopic modification of the local lymph node assay

The murine local lymph node assay (LLNA) is currently recognized as a stand-alone test method for determining the skin sensitizing potential of chemicals. It has been incorporated into the official test guidelines published by some authorities, including the OECD. To avoid the use of radioisotopes, efforts have been made recently to develop non-radioisotopic modifications of the LLNA. A non-radioisotopic modification of the LLNA was developed previously using 5-bromo-2'-deoxyuridine (BrdU) incorporation (non-RI LLNA). However, the non-RI LLNA was found to be somewhat less sensitive than the standard assay. This study reports the advantage of using mice of the CBA/N strain in the non-RI LLNA to improve the sensitivity of this method. The non-RI LLNA was performed using CBA/JN and CBA/N mice exposed to one of four confirmed skin sensitizers, 2,4-dinitrochlorobenzene (DNCB), eugenol (EG), isoeugenol (IEG) or alpha-hexylcinnamic aldehyde (HCA), and to one non-sensitizer, propylene glycol (PG). The EC3 values for DNCB, IEG, EG, HCA and PG were calculated to be 0.1%, 9.6%, 40.6%, 45.5% and >50% in CBA/JN mice and 0.08%, 1.9%, 10.7%, 20.3% and >50% in CBA/N mice, respectively. The EC3 values for DNCB, IEG, EG, HCA and PG in the standard LLNA using CBA/Ca mice and radioisotopes were reported elsewhere as being 0.08%, 1.3%, 13.0%, 8.0% and >50%, respectively. The EC3 values derived from the CBA/N mice in the non-RI LLNA were nearly equivalent to the EC3 values obtained using the standard radioisotopic LLNA with CBA/Ca mice. These data suggest that the use of CBA/N mice may provide a realistic opportunity to develop a version of the LLNA that does not have a requirement for the use of radioisotopes, but which nevertheless has sensitivity approaching, or comparable to, the standard method.     

Retrospective analysis of the Draize test for serious eye damage/eye irritation: importance of understanding the in vivo endpoints under UN GHS/EU CLP for the development and evaluation of in vitro test methods

For more than two decades, scientists have been trying to replace the regulatory in vivo Draize eye test by in vitro methods, but so far only partial replacement has been achieved. In order to better understand the reasons for this, historical in vivo rabbit data were analysed in detail and resampled with the purpose of (1) revealing which of the in vivo endpoints are most important in driving United Nations Globally Harmonized System/European Union Regulation on Classification, Labelling and Packaging (UN GHS/EU CLP) classification for serious eye damage/eye irritation and (2) evaluating the method’s within-test variability for proposing acceptable and justifiable target values of sensitivity and specificity for alternative methods and their combinations in testing strategies. Among the Cat 1 chemicals evaluated, 36–65 % (depending on the database) were classified based only on persistence of effects, with the remaining being classified mostly based on severe corneal effects. Iritis was found to rarely drive the classification (<4 % of both Cat 1 and Cat 2 chemicals). The two most important endpoints driving Cat 2 classification are conjunctiva redness (75–81 %) and corneal opacity (54–75 %). The resampling analyses demonstrated an overall probability of at least 11 % that chemicals classified as Cat 1 by the Draize eye test could be equally identified as Cat 2 and of about 12 % for Cat 2 chemicals to be equally identified as No Cat. On the other hand, the over-classification error for No Cat and Cat 2 was negligible (<1 %), which strongly suggests a high over-predictive power of the Draize eye test. Moreover, our analyses of the classification drivers suggest a critical revision of the UN GHS/EU CLP decision criteria for the classification of chemicals based on Draize eye test data, in particular Cat 1 based only on persistence of conjunctiva effects or corneal opacity scores of 4. In order to successfully replace the regulatory in vivo Draize eye test, it will be important to recognise these uncertainties and to have in vitro tools to address the most important in vivo endpoints identified in this paper.