Multivariate models for prediction of human skin sensitization hazard

One of the Interagency Coordinating Committee on the Validation of Alternative Method's (ICCVAM) top priorities is the development and evaluation of non‐animal approaches to identify potential skin sensitizers. The complexity of biological events necessary to produce skin sensitization suggests that no single alternative method will replace the currently accepted animal tests. ICCVAM is evaluating an integrated approach to testing and assessment based on the adverse outcome pathway for skin sensitization that uses machine learning approaches to predict human skin sensitization hazard. We combined data from three in chemico or in vitro assays – the direct peptide reactivity assay (DPRA), human cell line activation test (h‐CLAT) and KeratinoSens™ assay – six physicochemical properties and an in silico read‐across prediction of skin sensitization hazard into 12 variable groups. The variable groups were evaluated using two machine learning approaches, logistic regression and support vector machine, to predict human skin sensitization hazard. Models were trained on 72 substances and tested on an external set of 24 substances. The six models (three logistic regression and three support vector machine) with the highest accuracy (92%) used: (1) DPRA, h‐CLAT and read‐across; (2) DPRA, h‐CLAT, read‐across and KeratinoSens; or (3) DPRA, h‐CLAT, read‐across, KeratinoSens and log P. The models performed better at predicting human skin sensitization hazard than the murine local lymph node assay (accuracy 88%), any of the alternative methods alone (accuracy 63–79%) or test batteries combining data from the individual methods (accuracy 75%). These results suggest that computational methods are promising tools to identify effectively the potential human skin sensitizers without animal testing. Published 2016. This article has been contributed to by US Government employees and their work is in the public domain in the USA.     

Prediction of skin sensitization potency using machine learning approaches

The replacement of animal use in testing for regulatory classification of skin sensitizers is a priority for US federal agencies that use data from such testing. Machine learning models that classify substances as sensitizers or non‐sensitizers without using animal data have been developed and evaluated. Because some regulatory agencies require that sensitizers be further classified into potency categories, we developed statistical models to predict skin sensitization potency for murine local lymph node assay (LLNA) and human outcomes. Input variables for our models included six physicochemical properties and data from three non‐animal test methods: direct peptide reactivity assay; human cell line activation test; and KeratinoSens™ assay. Models were built to predict three potency categories using four machine learning approaches and were validated using external test sets and leave‐one‐out cross‐validation. A one‐tiered strategy modeled all three categories of response together while a two‐tiered strategy modeled sensitizer/non‐sensitizer responses and then classified the sensitizers as strong or weak sensitizers. The two‐tiered model using the support vector machine with all assay and physicochemical data inputs provided the best performance, yielding accuracy of 88% for prediction of LLNA outcomes (120 substances) and 81% for prediction of human test outcomes (87 substances). The best one‐tiered model predicted LLNA outcomes with 78% accuracy and human outcomes with 75% accuracy. By comparison, the LLNA predicts human potency categories with 69% accuracy (60 of 87 substances correctly categorized). These results suggest that computational models using non‐animal methods may provide valuable information for assessing skin sensitization potency. Copyright © 2017 John Wiley & Sons, Ltd.     

Probabilistic hazard assessment for skin sensitization potency by dose–response modeling using feature elimination instead of quantitative structure–activity relationships

Supervised learning methods promise to improve integrated testing strategies (ITS), but must be adjusted to handle high dimensionality and dose–response data. ITS approaches are currently fueled by the increasing mechanistic understanding of adverse outcome pathways (AOP) and the development of tests reflecting these mechanisms. Simple approaches to combine skin sensitization data sets, such as weight of evidence, fail due to problems in information redundancy and high dimensionality. The problem is further amplified when potency information (dose/response) of hazards would be estimated. Skin sensitization currently serves as the foster child for AOP and ITS development, as legislative pressures combined with a very good mechanistic understanding of contact dermatitis have led to test development and relatively large high‐quality data sets. We curated such a data set and combined a recursive variable selection algorithm to evaluate the information available through in silico, in chemico and in vitro assays. Chemical similarity alone could not cluster chemicals' potency, and in vitro models consistently ranked high in recursive feature elimination. This allows reducing the number of tests included in an ITS. Next, we analyzed with a hidden Markov model that takes advantage of an intrinsic inter‐relationship among the local lymph node assay classes, i.e. the monotonous connection between local lymph node assay and dose. The dose‐informed random forest/hidden Markov model was superior to the dose‐naive random forest model on all data sets. Although balanced accuracy improvement may seem small, this obscures the actual improvement in misclassifications as the dose‐informed hidden Markov model strongly reduced " false‐negatives" (i.e. extreme sensitizers as non‐sensitizer) on all data sets. Copyright © 2015 John Wiley & Sons, Ltd.     

Evaluation of in vitro assays for the assessment of the skin sensitization hazard of functional polysiloxanes and silanes

The skin sensitization potential of chemicals has traditionally been evaluated in vivo according to OECD testing guidelines in guinea pigs or the mouse local lymph node assay. There has lately been a great emphasis on establishing in vitro test methods reflecting the key biological events in the adverse outcome pathway (AOP) for skin sensitization as published by the OECD. Against this background, a group of 8 polysiloxanes and silanes, seven of them aminofunctionalised, for which in vivo data were already available, has been tested in vitro in the direct peptide reactivity assay (DPRA), the KeratinoSens™ and the human cell line activation test (h-CLAT) and in the modified myeloid U937 skin sensitization test (mMUSST) as far as technically feasible. The main objective of the programme was to determine the utility of these systems for this heterogeneous group of silicone-based substances, recognizing that some substances are outside the assays applicability domains. The presented data provided some interesting mechanistical insights into the performance of these assays for functionalised siloxanes and silanes. The data also allow for a preliminary evaluation of proposed integrated testing strategies (ITS) to determine the skin sensitization potential of chemicals which were not considered in the training sets of the respective ITS.     

Test battery with the human cell line activation test,direct peptide reactivity assay and DEREK based on a 139 chemical data set for predicting skin sensitizing potential and potency of chemicals

To develop a testing strategy incorporating the human cell line activation test (h‐CLAT), direct peptide reactivity assay (DPRA) and DEREK, we created an expanded data set of 139 chemicals (102 sensitizers and 37 non‐sensitizers) by combining the existing data set of 101 chemicals through the collaborative projects of Japan Cosmetic Industry Association. Of the additional 38 chemicals, 15 chemicals with relatively low water solubility (log Kow > 3.5) were selected to clarify the limitation of testing strategies regarding the lipophilic chemicals. Predictivities of the h‐CLAT, DPRA and DEREK, and the combinations thereof were evaluated by comparison to results of the local lymph node assay. When evaluating 139 chemicals using combinations of three methods based on integrated testing strategy (ITS) concept (ITS‐based test battery) and a sequential testing strategy (STS) weighing the predictive performance of the h‐CLAT and DPRA, overall similar predictivities were found as before on the 101 chemical data set. An analysis of false negative chemicals suggested a major limitation of our strategies was the testing of low water‐soluble chemicals. When excluded the negative results for chemicals with log Kow > 3.5, the sensitivity and accuracy of ITS improved to 97% (91 of 94 chemicals) and 89% (114 of 128). Likewise, the sensitivity and accuracy of STS to 98% (92 of 94) and 85% (111 of 129). Moreover, the ITS and STS also showed good correlation with local lymph node assay on three potency classifications, yielding accuracies of 74% (ITS) and 73% (STS). Thus, the inclusion of log Kow in analysis could give both strategies a higher predictive performance. Copyright © 2015 John Wiley & Sons, Ltd.     

Probabilistic derivation of the interspecies assessment factor for skin sensitization

An interspecies sensitization assessment factor (SAF) is used in the quantitative risk assessment (QRA) for skin sensitization when a murine-based NESIL (No Expected Sensitization Induction Level) is taken as point of departure. Several studies showed that, on average, the murine sensitization threshold is in good correspondence with that determined in humans. However, on an individual level, the murine and human sensitization thresholds may differ considerably. In this study, the interspecies SAF was quantified by using a probabilistic approach, to be able to take these cases into account. As expected, the geometric means of the probability distributions of murine and human sensitization threshold ratios were close to one, but taking the 95 ^th percentile of these distributions resulted in an interspecies SAF of 15. By using this value, one is sure that with 95% probability, the sensitization threshold determined in mice does not underestimate the human threshold. It can be concluded that a murine-based NESIL requires the use of an interspecies SAF (of 15) in the QRA for skin sensitization, to correct for the differences between mice and humans. This empirically derived interspecies SAF contributes to refinement of the risk assessment methodology.     

Accounting for data variability,a key factor in in vivo/in vitro relationships: application to the skin sensitization potency (in vivo LLNA versus in vitro DPRA) example

When searching for alternative methods to animal testing, confidently rescaling an in vitro result to the corresponding in vivo classification is still a challenging problem. Although one of the most important factors affecting good correlation is sample characteristics, they are very rarely integrated into correlation studies. Usually, in these studies, it is implicitly assumed that both compared values are error‐free numbers, which they are not. In this work, we propose a general methodology to analyze and integrate data variability and thus confidence estimation when rescaling from one test to another. The methodology is demonstrated through the case study of rescaling the in vitro Direct Peptide Reactivity Assay (DPRA) reactivity to the in vivo Local Lymph Node Assay (LLNA) skin sensitization potency classifications. In a first step, a comprehensive statistical analysis evaluating the reliability and variability of LLNA and DPRA as such was done. These results allowed us to link the concept of gray zones and confidence probability, which in turn represents a new perspective for a more precise knowledge of the classification of chemicals within their in vivo OR in vitro test. Next, the novelty and practical value of our methodology introducing variability into the threshold optimization between the in vitro AND in vivo test resides in the fact that it attributes a confidence probability to the predicted classification. The methodology, classification and screening approach presented in this study are not restricted to skin sensitization only. They could be helpful also for fate, toxicity and health hazard assessment where plenty of in vitro and in chemico assays and/or QSARs models are available. Copyright © 2016 John Wiley & Sons, Ltd.     

Computational approaches for skin sensitization prediction

Abstract

Drugs, cosmetics, preservatives, fragrances, pesticides, metals, and other chemicals can cause skin sensitization. The ability to predict the skin sensitization potential and potency of substances is therefore of enormous importance to a host of different industries, to customers’ and workers’ safety. Animal experiments have been the preferred testing method for most risk assessment and regulatory purposes but considerable efforts to replace them with non-animal models and in silico models are ongoing. This review provides a comprehensive overview of the computational approaches and models that have been developed for skin sensitization prediction over the last 10 years. The scope and limitations of rule-based approaches, read-across, linear and nonlinear (quantitative) structure–activity relationship ((Q)SAR) modeling, hybrid or combined approaches, and models integrating computational methods with experimental results are discussed followed by examples of relevant models. Emphasis is placed on models that are accessible to the scientific community, and on model validation. A dedicated section reports on comparative performance assessments of various approaches and models. The review also provides a concise overview of relevant data sources on skin sensitization.     

Anchoring molecular mechanisms to the adverse outcome pathway for skin sensitization: Analysis of existing data

Allergic contact dermatitis (ACD) is a hypersensitivity immune response induced by small protein-reactive chemicals. Currently, the murine local lymph node assay (LLNA) provides hazard identification and quantitative estimation of sensitizing potency. Given the complexity of ACD, a single alternative method cannot replace the LLNA, but it is necessary to combine methods through an integrated testing strategy (ITS). In the development of an ITS, information regarding mechanisms and molecular processes involved in skin sensitization is crucial. The recently published adverse outcome pathway (AOP) for skin sensitization captures mechanistic knowledge into key events that lead to ACD. To understand the molecular processes in ACD, a systematic review of murine in vivo studies was performed and an ACD molecular map was constructed. In addition, comparing the molecular map to the limited human in vivo toxicogenomic data available suggests that certain processes are similarly triggered in mice and humans, but additional human data will be needed to confirm these findings and identify differences. To gain insight in the molecular mechanisms represented by various human in vitro systems, the map was compared to in vitro toxicogenomic data. This analysis allows for comparison of emerging in vitro methods on a molecular basis, in addition to mathematical predictive value. Finally, a survey of the current in silico, in chemico, and in vitro methods was used to indicate which AOP key event is modeled by each method. By anchoring emerging classification methods to the AOP and the ACD molecular map, complementing methods can be identified, which provides a cornerstone for the development of a testing strategy that accurately reflects the key events in skin sensitization.     

A dataset on 145 chemicals tested in alternative assays for skin sensitization undergoing prevalidation

Skin sensitization is a key endpoint for cosmetic ingredients, with a forthcoming ban for animal testing in Europe. Four alternative tests have so far been submitted to ECVAM prevalidation: (i) MUSST and (ii) h‐Clat assess surface markers on dendritic cell lines, (iii) the direct peptide reactivity assay (DPRA) measures reactivity with model peptides and (iv) the KeratinoSens^TM assay which is based on detection of Nrf2‐induced luciferase. It is anticipated that only an integrated testing strategy (ITS) based on a battery of tests might give a full replacement providing also a sensitization potency assessment, but this concept should be tested with a data‐driven analysis. Here we report a database on 145 chemicals reporting the quantitative endpoints measured in a U937‐ test, the DPRA and KeratinoSens^TM . It can serve to develop data‐driven ITS approaches as we show in a parallel paper and provides a view as to the current ability to predict with in vitro tests as we are entering 2013. It may also serve as reference database when benchmarking new molecules with in vitro based read‐across and find use as a reference database when evaluating new tests. The tests and combinations thereof were evaluated for predictivity, and overall a similar predictivity was found as before on three‐fold smaller datasets. Analysis of the dose–response parameters of the individual tests indicates a correlation to sensitization potency. Detailed analysis of chemicals false‐negative and false‐positive in two tests helped to define limitations in the tests but also in the database derived from animal studies. Copyright © 2013 John Wiley & Sons, Ltd.     

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.     

Proposal for a risk assessment methodology for skin sensitization based on sensitization potency data

In this paper, we propose a quantitative risk assessment methodology for skin sensitization aiming at the derivation of 'safe' exposure levels for sensitizing chemicals, used e.g., as ingredients in consumer products. Given the limited number of sensitizers tested in human sensitization tests, such as the human repeat-insult patch test (HRIPT) or the human maximization test (HMT), we used EC3 values from the local lymph node assay (LLNA) in mice because they provide the best quantitative measure of the skin sensitizing potency of a chemical. A comparison of LLNA EC3 values with HRIPT and HMT LOEL, and NOEL values was carried out and revealed that the EC3, expressed as area dose, can be used as a surrogate value for the human NOEL in risk assessment. The uncertainty/extrapolation factor approach was used to derive (a) an 'acceptable non-sensitizing area dose' (ANSAD) to protect non-allergic individuals against skin sensitization and (b) an 'acceptable non-eliciting area dose' (ANEAD) to protect allergic individuals against elicitation of allergic contact dermatitis. For ANSAD derivation, interspecies, intraspecies and time extrapolation factors are applied to the LLNA EC3. For ANEAD derivation, additional application of a variable sensitization-elicitation extrapolation factor is proposed. Values for extrapolation factors are derived and discussed, the proposed methodology is applied to the sensitizers methylchloroisothiazolinone/methylisothiazolinone, cinnamic aldehyde and nickel and results are compared to published risk assessments.     

Non-animal methods to predict skin sensitization (II): an assessment of defined approaches**

Skin sensitization is a toxicity endpoint of widespread concern, for which the mechanistic understanding and concurrent necessity for non-animal testing approaches have evolved to a critical juncture, with many available options for predicting sensitization without using animals. Cosmetics Europe and the National Toxicology Program Interagency Center for the Evaluation of Alternative Toxicological Methods collaborated to analyze the performance of multiple non-animal data integration approaches for the skin sensitization safety assessment of cosmetics ingredients. The Cosmetics Europe Skin Tolerance Task Force (STTF) collected and generated data on 128 substances in multiple in vitro and in chemico skin sensitization assays selected based on a systematic assessment by the STTF. These assays, together with certain in silico predictions, are key components of various non-animal testing strategies that have been submitted to the Organization for Economic Cooperation and Development as case studies for skin sensitization. Curated murine local lymph node assay (LLNA) and human skin sensitization data were used to evaluate the performance of six defined approaches, comprising eight non-animal testing strategies, for both hazard and potency characterization. Defined approaches examined included consensus methods, artificial neural networks, support vector machine models, Bayesian networks, and decision trees, most of which were reproduced using open source software tools. Multiple non-animal testing strategies incorporating in vitro, in chemico, and in silico inputs demonstrated equivalent or superior performance to the LLNA when compared to both animal and human data for skin sensitization.     

Non‐animal assessment of skin sensitization hazard: Is an integrated testing strategy needed,and if so what should be integrated?

There is an expectation that to meet regulatory requirements, and avoid or minimize animal testing, integrated approaches to testing and assessment will be needed that rely on assays representing key events (KEs) in the skin sensitization adverse outcome pathway. Three non‐animal assays have been formally validated and regulatory adopted: the direct peptide reactivity assay (DPRA), the KeratinoSens? assay and the human cell line activation test (h‐CLAT). There have been many efforts to develop integrated approaches to testing and assessment with the “two out of three” approach attracting much attention. Here a set of 271 chemicals with mouse, human and non‐animal sensitization test data was evaluated to compare the predictive performances of the three individual non‐animal assays, their binary combinations and the “two out of three” approach in predicting skin sensitization potential. The most predictive approach was to use both the DPRA and h‐CLAT as follows: (1) perform DPRA – if positive, classify as sensitizing, and (2) if negative, perform h‐CLAT – a positive outcome denotes a sensitizer, a negative, a non‐sensitizer. With this approach, 85% (local lymph node assay) and 93% (human) of non‐sensitizer predictions were correct, whereas the “two out of three” approach had 69% (local lymph node assay) and 79% (human) of non‐sensitizer predictions correct. The findings are consistent with the argument, supported by published quantitative mechanistic models that only the first KE needs to be modeled. All three assays model this KE to an extent. The value of using more than one assay depends on how the different assays compensate for each other's technical limitations. Copyright © 2017 John Wiley & Sons, Ltd.     

LuSens: A keratinocyte based ARE reporter gene assay for use in integrated testing strategies for skin sensitization hazard identification

Allergic contact dermatitis can develop following repeated exposure to allergenic substances. To date, hazard identification is still based on animal studies as non-animal alternatives have not yet gained global regulatory acceptance. Several non-animal methods addressing key-steps of the adverse outcome pathway (OECD, 2012) will most likely be needed to fully address this effect. Among the initial cellular events is the activation of keratinocytes and currently only one method, the KeratinoSens™, has been formally validated to address this event. In this study, a further method, the LuSens assay, that uses a human keratinocyte cell line harbouring a reporter gene construct composed of the antioxidant response element (ARE) of the rat NADPH:quinone oxidoreductase 1 gene and the luciferase gene. The assay was validated in house using a selection of 74 substances which included the LLNA performance standards. The predictivity of the LuSens assay for skin sensitization hazard identification was comparable to other non-animal methods, in particular to the KeratinoSens™. When used as part of a testing battery based on the OECD adverse outcome pathway for skin sensitization, a combination of the LuSens assay, the DPRA and a dendritic cell line activation test attained predictivities similar to that of the LLNA.     

LuSens: A keratinocyte based ARE reporter gene assay for use in integrated testing strategies for skin sensitization hazard identification

Allergic contact dermatitis can develop following repeated exposure to allergenic substances. To date, hazard identification is still based on animal studies as non-animal alternatives have not yet gained global regulatory acceptance. Several non-animal methods addressing key-steps of the adverse outcome pathway (OECD, 2012) will most likely be needed to fully address this effect. Among the initial cellular events is the activation of keratinocytes and currently only one method, the KeratinoSens™, has been formally validated to address this event. In this study, a further method, the LuSens assay, that uses a human keratinocyte cell line harbouring a reporter gene construct composed of the antioxidant response element (ARE) of the rat NADPH:quinone oxidoreductase 1 gene and the luciferase gene. The assay was validated in house using a selection of 74 substances which included the LLNA performance standards. The predictivity of the LuSens assay for skin sensitization hazard identification was comparable to other non-animal methods, in particular to the KeratinoSens™. When used as part of a testing battery based on the OECD adverse outcome pathway for skin sensitization, a combination of the LuSens assay, the DPRA and a dendritic cell line activation test attained predictivities similar to that of the LLNA.