Subcategorization of skin sensitizers into UN GHS categories using GARDskin Dose-Response

Presented at 2025 SOT

Download a copy in PDF

 

Conclusion

GARD®skin Dose-Response provides accurate and robust UN GHS subclassifications, with performance and reproducibility levels comparable to, or surpassing, those of current in vitro counterparts.

​

Abstract

Background and Purpose:
Proactive identification and characterization of sensitization hazards and risks are central aspects of risk assessment of chemicals. Current legislations and trends in predictive toxicology advocate a transition from in vivo methods to new approach methodologies (NAM:s). For the purpose of hazard identification, numerous successful innovations have led to both the development and validation of several NAM:s for assessment of chemical skin sensitizers. Similarly, subcategorization of skin sensitizers according to requirements defined by United Nations Globally Harmonized System (UN GHS) has recently been demonstrated to be achievable by so-called Defined Approaches (DA:s), purely constituting NAM data sources. However, the ability of NAM:s to perform UN GHS-associated subcategorization may be considered subjects of optimization, both in terms of performance, applicability and the number of data sources required to obtain robust and accurate results.

The GARDskin Dose-Response (DR) method, adapted from the conventional OECD TG 442E method GARDskin, provides quantitative potency assessment of skin sensitizers. The method has been demonstrated to be useful for various potency-associated problem situations, including for definition of a point of departure (PoD) for downstream risk assessment, prediction of LLNA potency categories and weight of evidence (WoE)-based categorization of test chemicals. However, the ability of the method to contribute to UN GHS subcategorization, i.e., categorizing skin sensitizers into category 1A (strong) and category 1B (weak) sensitizers, has as of to date not been described. Here, a meta-analysis of all available and peer-reviewed GARDskin DR data is presented, aiming to describe the performance and reproducibility with which the GARDskin DR method can be used for UN GHS subcategorization.

Methods:
The GARDskin method (OECD TG 442E) is an in vitro assay for assessment of chemical skin sensitizers. The method provides binary hazard identification of skin sensitizers by evaluation of transcriptional patterns of an endpoint-specific genomic biomarker signature, comprising 196 genes, referred to as the GARDskin Genomic Prediction Signature (GPS), in the SenzaCell cell line. Final classifications are provided by a machine-learning prediction algorithm in the form of decision values (DV), the sign of which is evaluated by the prediction model; Any test chemical with a positive mean DV is classified as a skin sensitizer. Conversely, any test chemical with a negative mean DV is classified as non-skin sensitizer.

The GARDskin DR is an expanded adaptation of the conventional GARDskin method, in which test chemicals are evaluated by the GARDskin prediction algorithm in an extended range of concentrations, in order to investigate the dose-response relationship between GARDskin DVs and test chemical concentration. As such, it provides a quantitative estimation of sensitizing potency, referred to as cDV0, which corresponds to the lowest required dose able to generate a positive mean DV.

As cDV0 is typically reported as experimentally derived concentrations in the units of molar concentration or in the unit of μg/ml, improved interpretation may be facilitated by conversion to a predicted potency value (PPV) in the unit of µg/cm2, as recently described. As such, a PPV can readily be used to predict a human NESIL-value, derived from e.g. LLNA EC3 or human NOEL-values. Lastly, a PPV can be evaluated with respect to the UN GHS classification threshold of 500 µg/cm2 (corresponding to an LLNA EC3 of 2%). As such, a test chemical with a PPV smaller than 500 µg/cm2 is classified as a cat. 1A sensitizer. Conversely, a test chemical with a PPV greater than 500 µg/cm2 is classified as a non-cat. 1A sensitizer, which in the presence of positive GARDskin results is indicative of a cat. 1B sensitizer.

In the present study, a meta-analysis was conducted considering all available data from published GARDskin DR resources, which comprises > 150 unique test chemicals all of which have been assayed blindly in partner-controlled studies. The dataset was cross-referenced for UN GHS categories according to both human and LLNA data, as extracted from the Annex 2 of the Supporting document to the Guideline (GL) on Defined Approaches (DAs) for Skin Sensitization (GL 497). Taken together, the union of available data and GL 497 references included 87 test chemicals, with 69 and 38 test chemicals having available LLNA and human subcategory references, respectively. In addition, 42 chemicals had multiple results from repeated testing in 2-6 separate experiments, allowing also for an estimation of reproducibility.

Results:

The ability of the GARDskin DR method to perform UN GHS-associated subcategorization of skin sensitizers was estimated by evaluating published data with respect to the classification threshold of 500 µg/cm2 (corresponding to an LLNA EC3 of 2%). Considering all available published data with corresponding reference values (human and/or LLNA) available in the Annex 2 of the Supporting document to GL 497, results indicated that GARDskin DR was able to subcategorize test chemicals as cat. 1A or non-cat. 1A with a predictive accuracy of 88.6% and 91.3%, when considering human (N = 38) and LLNA (N = 69) references, respectively. Furthermore, from repeated testing of N = 42 chemicals, results indicate that the method was 92.9% reproducible, with concordant subcategorizations for 39 of the 42 chemicals. For this subset of data, test chemicals had been repeatedly tested in 2 (N = 22), 3 (N = 15), 4 (N = 4) or 6 (N = 1) independent experiments.

When put into context, these estimates indicate that GARDskin DR performs at levels comparable to, or surpassing, those of the DA:s of GL 497, when studying the same datasets, in terms of predictive accuracy and reproducibility.

 

Conclusions:
Taken together, this herein presented study indicates the usefulness of the GARDskin DR method to contribute to the prediction of skin sensitizing potency and associated UN GHS subcategorizations. Of particular note, the method is estimated to be comparable to existing DA:s, as described in GL 497, while being based on a single method. Should the regulatory context approve of classification and labelling based on individual data sources to the same extent as for DA:s, it is foreseen that such a testing strategy could potentially allow for resource-effective testing, with maintained high robustness and accuracy.