The role of in silico assessment in skin sensitization testing

Published: August 29, 2017

Dr Donna Macmillan, Senior Scientist at Lhasa Limited, looks at the role of in silico assessment in skin sensitisation testing in the wake of EU regulation 1223/2009.

The introduction of EU regulation 1223/2009, which ensures that any cosmetic ingredient on the EU market is demonstrably safe for use but at the same time restricts animal experiments placed significant challenges on the cosmetics industry, not least those of compliance and innovation.1,2According to reports under the Cosmetics Directive, the number of animals used for cosmetics in the EU has significantly reduced from 8,988 animals in 2004 to 1,510 in 2008 and 344 in 20091 as a direct consequence of this regulation.

Animal testing on cosmetics and cosmetic ingredients has been prohibited in the EU since 11th March 2009. Since this date, the marketing in the EU of any cosmetics and/or ingredients tested on animals has also been banned, except for very complex endpoints such as repeated-dose toxicity, reproductive toxicity and toxicokinetics. On the 11th March 2013, a complete ban on the marketing of cosmetics and/or ingredients came into force, regardless of the availability of non-animal alternatives.

The development of innovative cosmetic ingredients, post-2013, poses a unique challenge, in that, part of the required safety assessment data is likely to be missing due to the restriction on animal testing. For example, for truly novel ingredients there will be an absence of existing data, and data from other regulatory frameworks which could be used in place of new data. For ingredients new to the cosmetics industry but used in other sectors, existing data may be available but not always for the relevant endpoint or exposure route.

EU regulation 1223/2009 can also have an impact on existing ingredients as new safety concerns may arise after the product is already in use as a cosmetic ingredient. Furthermore, if known ingredients are to be used in new types of products e.g. hair dye, additional safety questions may need to be tackled which were not relevant before.

Skin sensitization is an important toxicological endpoint which leads to allergic contact dermatitis and is therefore imperative as a safety assessment in the cosmetic industry, as many cosmetics are designed to be applied to the skin e.g. face cream, shower gel, sun cream. As a result of the 2009 ban on in vivo animal testing for cosmetic reasons, a concerted effort has been made to develop non-animal alternatives to predict skin sensitisation. Until this point, the ‘gold standard’ methods to predict skin sensitisation were the in vivo LLNA (local lymph node assay) and GPMT (guinea pig maximisation test) although three in chemico/in vitro assays – DPRA; ARE-Nrf2 luciferase and h-CLAT – are currently recommended by ECVAM and have published OECD Guidelines.

It is generally accepted that no single non-animal method (whether in chemico, in vitro or in silico) can be used as a standalone method to replace animal models for the prediction of skin sensitization potential. This has led to the publication of many defined approaches (data generated by non-animal methods – also known as integrated testing strategies (ITS)) to predict skin sensitization e.g. Lhasa’s ITS which uses an in silico prediction with up to two in chemico/in vitro assays to assign a query chemical as a sensitizer/non-sensitizer.3 These defined approaches can be used in integrated approaches to testing and assessment (IATA)4 which characterise chemical hazard and/or risk for regulatory decision-making by integrating results from multiple approaches e.g. QSAR, read-across, in chemico/in vitro assay results. Although the OECD (Organisation for Economic Co-operation and Development) has provided guidelines on how to report IATA5 and published 12 case studies6 which could be used within a skin sensitization IATA, they currently do not recommend any particular ITS or IATA for the prediction of skin sensitization. This can make users apprehensive about choosing the ‘correct’ IATA, and hesitant when difficulties arise e.g. how confident can the user be when assays provide conflicting results (one says sensitizer, one says non-sensitizer)? This can make it difficult to change the view held by many that non-animal methods are inferior or less trustworthy than these in vivo methods – one of the biggest challenges of all.

The use of historical toxicity data may help to change this view. Predictive software and chemical databases hold a wealth of public toxicity data which may render it unnecessary to test a cosmetic ingredient. For example, Derek Nexus7 is an expert prediction tool containing structural alerts written by Lhasa scientists to reflect observed structure-activity relationships of toxicity. Each alert, when activated, represents a class of chemicals shown to induce a toxicological effect e.g. nickel can induce skin sensitization. A user can then access all the toxicity information and mechanistic rationale collated by Lhasa experts, alongside all relevant references, and eliminate the need for additional testing of that class of chemical.

In terms of innovation, predictive software supports new product development by streamlining the development process. If a user can screen cosmetic ingredients at the very beginning of a product’s life cycle it is less likely for an adverse outcome to be discovered when it is brought to market and significant time, effort, and money has been spent. Instead, materials with predictions of toxicity can be replaced much earlier with non-toxic ingredients and expedite the product development process.

The use of predictive software in skin sensitization testing brings its own challenges. For example, when using publicly available data, not all chemical space can be covered and a lot of this data has been exhausted for model building. By introducing proprietary data, the chemical space coverage increases. This highlights the importance of data partnerships – by encouraging collaboration and sharing of new data and ideas, we can improve the available predictive software and databases and potentially find a novel way to predict skin sensitization effectively.

But of course, predictive software is only as good as the data used to build the model. Errors (including assay variability/reproducibility) in the dataset must be understood before building predictive software; if a model is built on poor quality data (a noisy dataset), then this will reduce how predictive a model can be.

In essence, predictive software should be transparent and it should be easy to understand how, or from which data, the toxicity prediction has been derived. All predictive software should be assessed for performance against a dataset not previously used to build the model (a test set) to get a true sense of the predictivity of the model for unseen data e.g. as illustrated by the validation of Lhasa’s LLNA EC3 prediction model.8

It is important to know the strengths and limitations of all toxicological tests in order to make an informed decision about the most appropriate DA or IATA to use for a specific chemical. No test is perfect for predicting skin sensitization, including the animal tests. I do believe the future of testing in the cosmetic sector will focus on the method which can predict the least number of false negatives, as these have the potential to harm human health. In order for in silico assessment to fill that role it remains vital that companies and organisations recognise the benefits and get on board with data collaboration.


1. European Commission. Impact assessment on the animal testing provisions in regulation (EC) 1223/2009 on cosmetics, 2013.
2. European Commission. Report from the commission to the European parliament and the council on the development, validation and legal acceptance of methods alternative to animal testing in the field of cosmetics (2013-2015), 2016.
3. DS Macmillan et al. Regul Toxicol Pharmacol 2016;76:30.
4. OECD. Guidance document for the use of adverse outcome pathways in developing IATA for skin sensitisation, 2016.
5. OECD. Guidance document on the reporting of defined approaches and individual information sources to be used within IATA, 2016.
6. OECD. Annex I: case studies to the guidance document on the reporting of defined approaches and individual information sources to be used within integrated approaches to testing and assessment (IATA) for skin sensitisation, 2016.
7. Derek Nexus, www.lhasalimited.org/products/derek-nexus.htm
8. SJ Canipa et al. J Appl Toxicol 2017;37:985.



Copyright © 2017 Mack Brooks Speciality Publishing.