Donna Macmillan, Henrik Johansson, Olivia Larne, Malin Lindstedt
1. Granary Wharf House, 2 Canal Wharf, Leeds, LS11 5PS
2. SenzaGen, Lund, Sweden
3. Lund University, Lund, Sweden
There has been a significant drive to reduce, refine and replace animal models for the prediction of skin sensitization. This is in part due to the implementation of EU regulation 1223/20091 which prohibits the sale and marketing of any cosmetics and cosmetic ingredients which have been tested on animals, alongside REACH2 and CLP3 regulations which state that non-animal methods must be exhausted prior to considering the use of animal tests. The use and availability of non-animal methods is ever-increasing and 3 assays have been validated by the OECD thus far; the in chemico DPRA, the in vitro KeratinoSens™ and the in vitro h-CLAT. A number of other assays are undergoing OECD validation, including the GARDskin assay (Genomic Allergen Rapid Detection), a dendritic cell-based assay which identifies skin sensitizers from 200 genomic biomarkers4. However, it is generally accepted that no single non-animal method can be used as a standalone approach to replace animal models such as the murine local lymph node assay (LLNA). The focus has instead turned to combining multiple in chemico/in vitro/in silico assays and/or molecular descriptors to derive a more accurate assessment of hazard or risk, known as integrated testing strategies (ITS)5. The GARDskin assay has demonstrated high predictivity and has been reported as ready to use in an ITS6, therefore, it was decided to investigate the effect on performance when GARD was used in combination with Derek Nexus – and to compare these results against Derek with the DPRA, KeratinoSens™ and h-CLAT.
Using Derek skin sensitization predictions in combination with in chemico/in vitro assay results has a beneficial effect when predicting the LLNA outcome. GARDskin in particular performs extremely well when used with Derek in a conservative call approach. Human sensitization is more challenging to predict and GARDskin performs less well for this compared to predicting the LLNA – attributed to the small number of chemicals with both GARDskin and human data (n = 57), in addition to the positive bias in the GARD dataset (70%). However, the addition of Derek predictions clearly improve assay performance. Future work will focus on repeating this analysis on a larger, more balanced dataset.
Poster at Lhasa website