Registration Dossier

Administrative data

Description of key information

In-chemico/in-vitro testing:

Direct Peptide Reactivity Assay:

The purpose of this study (based on the OECD guideline for the testing of chemicals, In chemico Skin Sensitisation: Direct Peptide Reactivity Assay (DPRA), OECD/OCDE document TG 442C) was to assess the reactivity and sensitizing potential of the test substance.

Solutions of the test substance were successfully analysed by the validated DPRA analytical method in both Cysteine and Lysine containing synthetic peptides.

The overall result places the test substance in the “no or minimal reactivity” class and therefore it is predicted to be a potential non-skin sensitizer.

In-silico/QSAR assessment:


Prediction result

TOPKAT non extended


TOPKAT extended





EC3 = not concluded

OECD Toolbox





Alert for Schiff base formation

Alert for Michael Acceptor

From the predictions with all the above models, there is no clear consensus between the models and a general lack of reliability in the predictions. There is a certain case for a weight of evidence that GalHueShield HCS is likely to be skin sensitising. However as there were multiple deficiencies with a majority of the predictions, leading most to be considered low reliability, it was determined that this evidence was insufficient to conclude on classification without further study. It was determined however that there is a likelihood of pro-haptens being of significance for the substance, and thus assessing for metabolic potential may be necessary.


Dermal absorption:

Based on the substances properties an assessment of toxicokinetic behaviour suggests dermal absorption is expected to be limited. GalHueshield HCS is not a skin irritant so enhanced penetration through the skin due to local damage is not expected. The available data on the substance suggests it is of low toxicity and possibly low systemic uptake.



Key value for chemical safety assessment

Skin sensitisation

Endpoint conclusion
Additional information:

Appraisal of QSAR Modelling:

As the substance is a UVCB, the prediction of the skin sensitising potential of GalHueShield HSCwas performed on the primary constituent:

IUPAC name: 1-Docosanaminium, N-[3-[[3-(4-methoxyphenyl)-1-oxo-2-propen-1-yl]amino]propyl]-N,N-dimethyl-,chloride (1:1)

CAS: 880645-41-0

Common name: Methoxycinnamidopropyl Behendimonium Chloride

Predictions were made with BIOVIA Discovery Studio (TOPKAT) 4.5, VEGA NIC 1.1.4 (CAESAR), OECD QSAR Toolbox 4.2, and DEREK Nexus 5.0.2. In addition, results from Toxtree 2.6.13 were also checked but not used in consideration in this assessment. The TOPKAT model for skin sensitisation was extended by including data from the Envigo database.

DEREK, CAESAR and OECD Toolbox predicted GalHueShield HCS to be sensitising. Using the Envigo model extension for TOPKAT and the non-extended TOPKAT model for skin sensitisation however resulted in a negative result.

While there is some confidence in the positive prediction of DEREK, the predicted values for EC3 are based on very small datasets. There is moderate confidence in the Toolbox prediction, however there is a considerable lack of similarity between the target compound and the nearest neighbours. The CAESAR prediction from VEGA highlights several issues with its own applicability domain and is thus considered unreliable. The TOPKAT models showed poor statistics also, with the extended model also falling outside of the applicability domain.

The negative prediction by the non-extended TOPKAT is characterised by some uncertainties with regard to the poor similarity displayed by the most similar structurers in the training set, thus indicating poor confidence in the prediction. Nevertheless, the prediction statistics (probability, enrichment, baysian score/best split difference, concordance of measured data and accuracy of predictions) were all high indicating some amount of confidence in the prediction.

The extended TOPKAT model also derived a negative prediction. Similarly to the non-extended model, there were issues with similarity, but also multiple flaws identified as being outside of the allowed limits of the domain. While the other statistics were similarly reassuring for this extended model as for the non-extended model, the issues relating to the applicability domain mean that this model should not be considered reliable.

Of the four most similar structures in CAESAR’s training set, all four are sensitizers. They have an average structural similarity to the query structure of 69.3 % according to the software.Upon investigation using the OECD toolbox to profile the similar compounds, protein binding by OASIS with autoxidation simulation and skin metabolism simulation shows that the similar structures do show similar mechanistic domain as the query structure but only one follows both identified mechanisms.Uncertainty is indicated by the absence of fragments not found in the compounds of the training set, and also a descriptor for the target was identified as being outside of the scope of the training set. Therefore the model concludes the prediction was outside of its domain. As the query structure was not within the applicability domain of the model, and there is some uncertainty in the model, the prediction was considered to be low reliability.

The DEREK skin sensitisation module indicated two separate potential mechanisms for the compound. The first being considerations around the quaternary nitrogen moeity, which DEREK reports as a direct acting hapten acting through ion pair formation. The model states that this alert alone should only be considered equivocal, but there is reference to higher lipophilicty contributing to the likelihood of the material being skin sensitising. The model also shows the potential for sensitisation highlighted by an alert for vinylic or allylic anisole. The model goes on to list several potential mechanisms by which this may induce sensitisation from the literature via formation of pre/pro-haptens. However it points out that the most likely mechanism will follow enzymatic cleavage following absorption in the skin to for the phenol, after which abstraction of the hydroxyl hydrogen to generate a reactive phenolic radical species, stabilised by the vinyl group.

The EC3 prediction from DEREK based on the n-quat alert shows only a single substance in the dataset which was significantly dissimilar to the target compound and only identified as a week sensitizer by LLNA (30%). Via the vinyl anisole mechanism however, the prediction is marginally improved owing to having two substances with three data points in total, with LLNA values of 2.3, 18 and 20%. Though these substances were also dissimilar from the target, the degree of dissimilarity was in both instances this was too few data points to derive a predicted value. These conclusions were considered reliable.

While no prediction is made by the software, no EC3 result from the similar compounsds was less than 2%.

Prediction with OECD Toolbox was performed manually as the automated workflow would not continue for these charged compounds. GalHueSHield HCS triggered an alert for protein binding for skin sensitisation by oasis, for Michael addition, and also alerts for Schiff base formation when considering skin metabolism products.

Primary grouping was made with the Protein binding alerts for skin sensitization by oasis, followed by sub-categorisations using (i) protein binding alerts for skin sensitisation by OASIS with skin metabolism simulator, (ii) protein binding alerts for skin sensitisation by OASIS with autoxidation simulator, and (iii) structural similarity. This returned the 4 nearest structures, though their similarity was only >30%, with an average of 40% similarity to the target, which indicates some uncertainty. Furthermore, concordance between the measured results can only be considered moderate as there are two positive and two negative results (though the collection of data prior to structural similarity grouping was predominately positive). The query structure is within the applicability domain of all subcategories, which affords some confidence in the prediction. As a result the model is assumed to be only moderately reliable.

Justification for classification or non-classification