Predictivity of the Kinetic Direct Peptide Reactivity Assay (kDPRA) for Sensitizer Potency Assessment and Subclassification

Several in vitro OECD test guidelines address key events 1–3 of the adverse outcome pathway for skin sensitization, but none are validated for sensitizer potency assessment. The reaction of sensitizing molecules with skin proteins is the molecular initiating event and appears to be rate-limiting as chemical reactivity strongly correlates with sensitizer potency. The kinetic direct peptide reactivity assay (kDPRA), a modification of the DPRA (OECD TG 442C), allows derivation of rate constants of the depletion of the cysteine-containing model peptide upon reaction with the test item. Its reproducibility was demonstrated in an inter-laboratory study. Here we present a database of rate constants, expressed as log kmax, for 180 chemicals to define the prediction threshold to identify strong sensitizers (classified as GHS 1A). A threshold of log kmax 2 offers a balanced accuracy of 85% for predicting GHS 1A sensitizers according to the local lymph node assay. The kDPRA is proposed as a stand-alone assay for identification of GHS 1A sensitizers among chemicals identified as sensitizers by other tests or defined approaches. It may also be used for the prediction of sensitizer potency on a continuous scale, ideally in combination with continuous parameters from other in vitro assays. We show how the rate constant could be combined with readouts of other in vitro assays in a defined approach. A decision model based on log kmax alone has, however, a high predictivity and can be used as stand-alone model for identification of GHS 1A sensitizers among chemicals predicted as sensitizers.


Predictivity of the Kinetic Direct Peptide Reactivity Assay (kDPRA) for Sensitizer Potency Assessment and Subclassification
Supplementary Data 1 Tab. S1: Chemicals that could not be evaluated based on technical limitations Name CAS Observed problem 2-Nitro-1,4-phenylendiamine 5307-14-2 fluorescence quenching   Table S4 according to the Basketter classification For creosol (2-Methoxy-4-methylphenol ) Cat 1A classification is based on a low NOEL value only (i.e. from a study showing no sensitization reactions). Thus this class attribution is due to the arbitrarily chosen low test concentration, and it is highly likely that this chemical would only sensitize at much higher doses as similar molecules like eugenol or dihydroeugenol are weak to moderate sensitizers falling into Cat 1B. Thus, it is probably an incorrect assignment.
Lyral was classified as Cat 1A based on clinical observations, while predictive human tests had not found sensitization reactions. Thus, neither LLNA nor human predictive testing would not have led to 1A. The false-positives include two clear prohaptens (diethylenetriamine and 3-dimethylaminopropylamine) and two pre-haptens (4-phenylenediamine, which reacts more slowly in the kinetic assay as it requires abiotic oxidation and 2-aminophenol).
They include also three Michael acceptor chemicals with lowest observed effect level (LOEL) values close to the human cut-off for Cat 1A chemicals of 500 g/cm 2 , and an extrapolated DSA05 (extrapolated value leading to induction of sensitization in 5% of the panelist) therefore closely below the cut-off. (-Damascone (human LOEL = 500 g/cm 2 ), 2-hexylidene cyclopentanone (human LOEL= 500 g/cm 2 ), Methylanisylidene acetone (human LOEL= 550 g/cm 2 ) which are also 1B in LLNA, so these are clearly borderline chemicals. Phenylacetaldehyde was rated 1B by the Basketter et al. compilation and by the OECD data review, but it is 1A based on the ICCVAM evaluation of the RIFM data and here included in 1A.
Finally, 9 of 12 of these under-predicted chemicals are rated as Cat 1B by the LLNA, too. Thus, overall, only a limited number of important and clear-cut human 1A sensitizers are missed by this refined cut-off (4-phenylenediamine, Diethylenetriamine, Glutaric aldehyde, 2-aminophenol, 3-Dimethylaminopropylamine, 6-Methyl-3,5-heptadien-2-one). Based on all these evaluations, a refined cut-off of log kmax = -2.0 appears as an optimal prediction model to balance accuracy for LLNA and human data.

Alternative calculations for identification of chemicals in the GHS 1A potency class
Since LLNA values are in weight % and kmax values are based on molar concentration, we performed two additional ways of calculating which chemicals have a predicted EC3 value < 2% (i.e. fall into the GHS 1A category) to test the impact of this simplification. a) We transformed the measured k-values to a percentage value (by multiplying kmax with 10 and dividing it by the molecular weight), and then performed the ROC-analysis vs. the LLNA based on kmax values calculated in % b) We used the predictive formula derived by regression analysis (Eq. S1: pEC3 = 2.652 + 0.3491 × log kmax) to derive a predicted pEC3 which was then transformed to the EC3 and used for classification according to the 2% threshold. Approach a): In this analysis a threshold of -3.5 in % -1 s -1 has the best predictivity. With the approach b), no threshold needs to be determined, but rather chemicals are classified by the EC3 values calculated from the predicted pEC3 value according Eq. S1. Table S6 shows the predictivity of the original approach using the threshold of log kmax M -1 s -1 compared to the predictivity with the two alternative calculations. In each case, for 174 of the 180 chemicals the same result is obtained, However, the balanced accuracy is not improved; i.e. slightly reduced (from 85.2% to 84.4% for approach a) and to 83% for approach b). We thus propose to remain with the threshold of log kmax = -2 for a regulatory classification, not last for its simplicity and most important for its predictivity, acknowledging that calculating everything in molar terms may scientifically be a preferred approach.
As the predictivity is optimal with the approach using log kmax [based on M -1 s -1 ] and the threshold -2, this approach is proposed to be taken forward for regulatory use of the kDPRA.  Table S7 shows the chemicals for which the alternative calculations lead to a different outcome. As can be expected, this is the case for chemicals with a log kmax close to the threshold of -2 or those with a relatively low or high molecular weight.