The kinetic Direct Peptide Reactivity Assay (kDPRA): Intra- and Inter-Laboratory Reproducibility in a Seven-Laboratory Ring Trial

While the skin sensitization hazard of substances can be identified using non-animal methods, the classification of potency into UN GHS sub-categories 1A and 1B remains challenging. The kinetic direct peptide reactivity assay (kDPRA) is a modification of the DPRA wherein the reaction kinetics of a test substance towards a synthetic cysteine-containing peptide are evaluated. For this purpose, several concentrations of the test substance are incubated with the synthetic peptide for several incubation times. The reaction is stopped by addition of monobromobimane, which forms a fluorescent complex with the free cysteine of the model peptide. The relative remaining non-depleted amount of peptide is determined. Kinetic rate constants are derived from the depletion vs concentration and time matrix and used to distinguish between UN GHS sub-category 1A sensitizers and test substances in sub-category 1B/not classified test substances. In this study, we present a ring trial of the kDPRA with 24 blind-coded test substances in seven laboratories. The intraand inter-laboratory reproducibility were 96% and 88%, respectively (both for differentiating GHS Cat 1A sensitizers from GHS Cat 1B/not classified). Following an independent peer review, the kDPRA was considered to be acceptable for the identification of GHS Cat 1A skin sensitizers. Besides GHS Cat 1A identification, the kDPRA can be used as part of a defined approach(es) with a quantitative data integration procedure for skin sensitization potency assessment. For this aim, next to reproducibility of classification, the quantitative reproducibility and variability of the rate constants were quantified in this study. This is an Open Access article distributed under the terms of the Creative Commons Attribution 4.0 International license (http://creativecommons.org/licenses/by/4.0/), which permits unrestricted use, distribution and reproduction in any medium, provided the original work is appropriately cited.


Introduction
The mechanism underlying skin sensitization is complex, but it is well understood and described in the adverse outcome pathway by OECD (OECD, 2014). During the last decade significant progress has been made in the field of non-animal tests and several test methods to address skin sensitization have meanwhile been adopted as OECD test guidelines (TGs). To date however, the regulatory accepted test methods were validated to address the skin sensitization hazard but not potency, while potency information is essential for risk assessment purposes. The molecular initiating event in the adverse outcome pathway for skin sensitization is the covalent binding to skin proteins and in 2015 a first method to address this key event has been adopted as OECD test guideline 442C. This method, the direct peptide reactivity assay (DPRA), uses two synthetic peptides (one containing a cysteine, one containing a lysine residue) that are incubated with a single concentration of the test substance. After 24 hours incubation, the remaining, non-depleted peptide concentrations are determined using HPLC. All currently adopted non-animal OECD TG methods to assess skin sensitization including the DPRA provide information on the skin sensitizing hazard but cannot be used as a stand-alone tests to address skin sensitization (OECD, 2019) or sufficiently address skin sensitizing potency (Wareing et al. 2017). A score-based approach for identification of GHS 1A chemicals had been proposed, however, ALTEX preprint published June 10, 2020 doi:10.14573/altex.2004291 2 reproducibility of the underlying scores derived from quantitative data in the DPRA and the hClat assay was not yet documented (Nukada et al., 2013).
The kinetic direct peptide reactivity assay (kDPRA) is a modification of the OECD adopted in chemico DPRA (described in Appendix I of OECD TG 442C, (OECD, 2019)). The kDPRA uses the cysteine containing test peptide (Ac-RFAACAA-COOH; Cys) also used in the DPRA, while it does not use a lysine containing peptide. The final concentration of the Cys peptide (0.5 mM) and the reaction medium (25% acetonitrile in phosphate buffer) is identical in the kDPRA and in the DPRA. While the DPRA measures only at one concentration of the test substance (5 mM in the Cys peptide reaction mixture) and at one, not exactly defined, time point (≥ 24 h), the kDPRA performs parallel reactions at five test substance concentrations (5, 2.5, 1.25, 0.625 and 0.3125 mM) and at six defined time-points (10, 30, 90, 150, 210 and 1440 min) at 25 ± 2.5ºC. Residual concentration of the Cys peptide after the respective reaction time is measured by stopping the reaction by the addition of monobromobimane (mBrB). The highly reactive and non-fluorescent mBrB rapidly reacts with unbound cysteine moieties of the model peptide to form a fluorescent complex. The remaining non-depleted peptide concentration can thus be determined and the depletion vs. time and concentration matrix is used to calculate rate constants.
It has been shown that kDPRA differentiates GHS Cat 1A sensitizers from GHS 1B/ not classified substances with a balanced accuracy of 85% (based on 180 test substances), with a sensitivity of 84% (38/45), and a specificity of 86% (116/135) relative to LLNA results (Natsch et al., 2020). In addition, the prediction of human skin sensitization for 123 test substances that fall within kDPRA's applicability domain has a balanced accuracy of 76%, a sensitivity of 64% (21/33), and a specificity of 89% (80/90) (Natsch et al., 2020). On the basis of the overall data available (n=180), kDPRA's applicability domain was shown to include a variety of organic functional groups, the full range of skin sensitization potencies (as determined in in vivo studies), and diverse physicochemical properties.
After setting up a standard operating procedure to conduct the kDPRA and to perform rate constant calculations, the primary goal of this study was to assess the transferability of the method (using six test substances) to five naïve labs and then assess intraand inter-laboratory reproducibility of the method using 24 blind-coded substances. The reproducibility of the assay based on the log-transformed rate constants as well as the classification reproducibility to differentiate Cat 1A sensitizers from Cat 1B sensitizers/ not classified substances according to UN GHS were evaluated.

2
Materials and Methods # 2.1 Test substance selection Test substances for the ring trial for evaluation of transferability and reproducibility were selected based upon their published characterization for potency in mice and humans (Basketter et al., 2014;ICCVAM, 2011).

2.1.2
Inter-and Intra-laboratory trial (Reproducibility assessment) The 24 test substances used during the blind-coded testing included 2 correct negatives and 21 correct positives in the Cys-only DPRA according to literature data (ICCVAM, 2011). The test set intentionally also included 1 sensitizer known to be negative / minimally reactive in the DPRA (i.e., dihydrocoumarin). The 24 test substances are listed in Table 1 including their protein reaction mechanistic domain assignment. This set of 24 test substances is strongly biased for positive test substances (sensitizers)this is with clear intention, as the kDPRA is intended for potency discrimination within test substances rated reactive. Two negative test substances were nevertheless included to also assess reproducibility for non-reactive test substances. # Abbreviations: A, alanine; ACN, acetonitrile; C, cysteine; CAS RN, Chemical Abstracts Service Registry Number; Cat, category; Cys, cysteine; CN, correct negative; CP, correct positive; DNCB, 2,4-Dinotrochlorobenzene; dp, depletion; DPRA, direct peptide reactivity assay; DSA05, dose per skin area that produces a positive response in 5% of the tested humans; EC3, estimated concentration leading to a stimulation index of 3 in the local lymph node assay; EGDMA, ethylene glycol dimethacrylate; F, phenylalanine; FN, false negative; FP, false positive; GHS, Globally Harmonized System of Classification, Labeling and Packing of Chemicals, HPLC, high performance liquid chromatography; ICCVAM, Interagency Coordinating Committee on the Validation of Alternative Methods; k, kinetic rate constant; kDPRA, kinetic DPRA; l, liquid; LLNA, local lymph node assay; Lys, lysine; MA, Michael acceptor; mBrB, monobromobimane; n, number of chemicals; NC, negative control; OECD, Organization for Economic Co-operation and Development; OECD TG, OECD test guideline; PC, positive control; QP, quinone precursor; R, arginine; RC, reactive carbonyl; s, solid; SB, Schiff 'base former; SC, substance control; SD, standard deviation; SNAr, Aromatics reacting by nucleophilic substitutions

2.2
Participating laboratories This ring trial validation study was conducted by a total of seven laboratories (in alphabetical order): BASF SE Experimental Toxicology and Ecology (Germany), Charles River Laboratories Den Bosch BV (The Netherlands), Givaudan Schweiz AG (Switzerland), Institute for In Vitro Sciences, Inc. (USA), L'Oréal Research & Innovation (France), National Institute of Public Health (Czech Republic), Procter & Gamble (USA). Two of the labs were the lead labs (BASF SE Experimental Toxicology and Ecology (Germany), Givaudan Schweiz AG (Switzerland)), responsible for protocol authorship, organization of the test-substance selection and procurement and statistical evaluations. The additional five laboratories were naïve to the kDPRA (thereof two were also naïve to the standard DPRA according to OECD TG 442C, (OECD, 2019)).

2.3
Kinetic direct peptide reactivity assay (kDPRA) 2.3.1 Procedure Test substances were dissolved in acetonitrile (ACN) or in pH 7.5 phosphate buffer, if not soluble in ACN, to yield stock solutions of 20 mM. Thereafter, dilution series of 20, 10, 5, 2.5 and 1.25 mM were prepared.
The kDPRA consisted of the following steps: In case of ACN-soluble substances, 120 µL of 0.667 mM Cys-peptide solution in pH 7.5 phosphate buffer were added to each well of a black 96-well plate. Next, 40 µL of the respective substance solution were added to each well. This yielded 0.5 mM peptide concentration and substance concentrations of 5, 2.5, 1.25, 0.625 and 0.3125 mM (final ratios of peptide: test substance = 1:10, 1:5, 2:5, 4:5, 8:5). All substances were tested in triplicate within the same run.
In case of pH 7.5 phosphate buffer-soluble substances, 80 µL of 1.0 mM Cys-peptide solution in phosphate buffer (pH 7.5) were added to each well of a black 96-well plate. Next, 40 µL of ACN were added and finally 40 µL of the respective substance solution. This yielded the same composition of samples as for the ACN-soluble substances described above.
Each 96-well plate comprised control samples as follows: 12 wells of a negative control (NC), containing the peptide and vehicle; 12 wells of blank control (BC), containing pH 7.5 phosphate buffer (without peptide) and the vehicle; 1 sample per concentration of the positive control (PC) cinnamic aldehyde and 1 sample per substance and concentration of a substance control (SC), containing the respective test substance and the buffer but no peptide. The SC served for identification of interference of the test substance with the fluorescence measurement and as a background measurement.
The plates were sealed with impermeable foil directly after application of the substance and were shaken for 5 min on a plate shaker and thereafter incubated in the dark at 25°C±2.5°C. The substances were incubated for 10, 30, 90, 150, 210 and 1440 min. After the respective reaction time, each test run was stopped by the addition of 40 µL of 3 mM of the fluorescence dye, i.e. mBrB solution (diluted in acetonitrile). Highly reactive non-fluorescent mBrB rapidly reacts with unbound cysteine moieties of the model peptide to form a fluorescent complex. The higher the intensity in fluorescence, the more cysteine moieties remain unbound after the respective reaction time and the less peptide-reactive is the test substance.
The substance solutions containing mBrB were then further incubated for 5 min in the dark on a plate shaker, and fluorescence then detected using an excitation filter of 390 nm and an emission filter of 480 nm.
Fluorescence intensities were normalized relative to the substance without the peptide (SC), as well as the phosphate buffer and acetonitrile (BC; background fluorescence). Relative peptide depletion was expressed as percent decrease in relation to the mean of the NC wells. Further, a pair-wise comparison of each substance concentration group with the NC was performed using the Welch t-test (two-sided) for the hypothesis of equal means.

Data evaluation
Depletion values were further evaluated, if at the highest concentration of one reaction time point the criteria for positivity was reached (13.89% Cys-peptide depletion, based on the Cys-only prediction model described in OECD TG 442C (OECD, 2019)) and if the depletion was statistically significantly different (p < 0.05) from NC.
For each incubation time, the remaining (non-reacted) amount of Cys-peptide was determined, the natural logarithm taken and plotted against the respective substance concentration. For each time point for which the regression line gives a correlation >0.9, the obtained slope was divided by the incubation time to determine the second order reaction rate constant k in [min -1 mM -1 ] (Roberts and Natsch, 2009). This value is transformed to the rate constant in [s -1 M -1 ] and the logarithm is taken. The maximum value observed at any time point is taken as the log kmax and is used for the further evaluation.

Acceptance criteria
The results of a 96-well test plate were considered valid if the following conditions were met: • Positive control (PC) the log kmax of the PC at 90 min should be within the following range: -1.75 M -1 s -1 to -1.40 M -1 s -1 If no log kmax was obtained at 90 min, the value at 150 min could be taken into account and should lie in the following range: -1.90 M -1 s -1 to -1.45 M -1 s -1 • Vehicle control (VC): The coefficient of variation of the 12 VC values of a plate should be <12.5% for 5 of the 6 time points. If one or more of these criteria were not met a repetition of the run was considered. Further, the runs for substances were repeated if non-linear behavior of results was obtained in order to exclude data bias due to artifacts, e.g. pipetting errors.

Prediction model
The maximum rate constant observed, log kmax, is used in the kDPRA to distinguish between two levels of skin sensitization potency, i.e. to discriminate between GHS subcategory 1A from GHS subcategory 1B/ not classified. The prediction model was developed from a dataset on 180 substances with LLNA refence data (Natsch et al., 2020): Reaction rate kDPRA prediction log kmax ≥ -2.0 GHS subcategory 1A log kmax < -2.0 GHS subcategory 1B or not classified

Transfer phase
For the transfer phase, the naïve laboratories received the protocol, an evaluation sheet and run validity criteria, and a preliminary proficiency range based on previous results of the lead labs with the same test substances. Two telephone conferences were held to clarify questions related to the conduct of the assay but no hands-on training was conducted.

Blind coded testing
In total 24 different test substances were assessed under GLP-like conditions during the blind coded testing in seven different laboratories. Test-substance procurement, blind coding and distribution were conducted by an external service (BioTeSys GmbH, Esslingen, Germany). All seven participating labs tested all 24 test substances in one repetition (inter-laboratory reproducibility). Further, out of the 24 test substances a random subset of 12 test substances was tested in two additional repetitions (with a different code for each run) in three or four labs (additional intra-laboratory reproducibility). Thus, for 12 test substances this intra-laboratory comparison was conducted in 3 labs and for the remaining 12 test substances in 4 labs (in total 12 x 3 + 12 x 4 = 84 intra-laboratory comparisons). This was fully randomized, so no laboratory received the same test substances for intralaboratory reproducibility.

Results and discussion
To illustrate a typical result in the kDPRA assay, the depletion matrices for DNCB determined in the seven participating laboratories during the transfer phase are shown in Fig. 1. For each exposure time point, the natural logarithm of the remaining Cys-peptide (in per cent relative to the vehicle control) is plotted against test-substance concentration.

Transferability
The five naïve labs tested six (non-coded) known sensitizers and the positive control (PC) cinnamic aldehyde to establish the assay within their labs and to familiarize with the study protocol. Overall, the results reported were very similar to those obtained by the two lead labs (Fig. 2) with the exception of formaldehyde that was much less reactive in all three repetitions performed at Lab E.
The kDPRA was easily transferable to five naïve labs without hands-on training and the SOP was found sufficiently detailed to perform the test in all participating laboratories. There were no significant technical obstacles specific to the method. Fig. 2 indicates that the 6 test substances first tested in the two lead laboratories resulted in very similar log kmax values when tested in the 5 naïve labs. Moreover, the standard deviation from the four experiments in the two lead labs is similar to the standard deviation in the 11 experiments in the 5 naïve labs, and thus the variability was not significantly increased by moving from the lead labs to the naïve labs.

3.2
Blind-coded testing Once the laboratories had successfully tested the six substances of the transfer phase, they progressed to testing the 24 blindcoded substances. After the test of the blind-coded substances were completed at the seven participating laboratories, all spreadsheets containing the blind-coded data were collected by the lead labs and provided to BioTeSys as external and independent data repository site before the code for un-blinding was provided by the latter. Analysis and biostatistics on the decoded data were then conducted at Givaudan.

3.2.1
Reproducibility of positive control (PC) For the PC cinnamic aldehyde, the log kmax values and the rate constants at a fixed time (90 min and 150 min) were reported from each experiment (summarized in Table 2.) The average log kmax of all valid runs during the blind-coded testing phase was at -1.35, and this value is identical to the value obtained as average value of all labs in the transfer phase, thus very comparable results were obtained in both phases of this study. The average log kmax value varied between -1.15 and -1.51 for the seven labs. The standard deviation for intra-laboratory reproducibility of the log kmax was between 0.14 and 0.23, similar to the average standard deviation obtained for all test substances in the intra-laboratory reproducibility (0.158; see subchapter 3.2.2 intra-lab reproducibility).
The inter-laboratory variability in the PC was even lower for the rate constants derived at 90 and 150 min which are used as acceptance criteria, and log k90 min varied between -1.53 and -1.64 for the seven labs (Tab. 2), while the intra-laboratory standard deviation was between 0.04 and 0.08. The overall standard deviation for all runs was at 0.04.
Log k150 min varied between -1.62 and -1.75 for the seven labs (Tab. 2), while the intra-laboratory standard deviation was between 0.03 and 0.09. The overall standard deviation for all runs is at 0.04.  The k90min value is the rate used to decide on acceptability of an experiment. In case no rate is calculated at 90 min (reaction not linear or not statistically significant) then k150min can be considered. Only in 4 of 148 runs during the blind-coded testing labs had to report 150 min value instead of 90 min value.

3.2.2
Intra-laboratory reproducibility The average and the standard deviations of log kmax are shown in Figure 3 for test substances tested in four labs and in Figure 4 for test substances tested in three labs.
For most test substances the intra-laboratory variability was low and the standard deviations between individual runs were very small. For those test substances with very low standard deviation (below 0.3 on the logarithmic scale corresponding to a two-fold difference in the kinetic rate), the values reported from the different laboratories are also very close to each other. The average standard deviation of the 24 test substances in intra-laboratory testing was at 0.158, and the average was < 0.1 for 10 test substances, with further five test substances being non-reactive in all laboratories. These include the two non-sensitizers (4methoxy-acetophenone and chlorobenzene) and 3,4-dihydrocoumarin, which are also non-Cys-reactive in DPRA (Natsch et al. 2013). In addition, phenylbenzoate and bourgeonal were reproducibly non-reactive in kDPRA despite the fact that Cys-depletion had been reported in the DPRA (Bauch et al. 2011) (and unpublished data). This low intra-laboratory variability of the rate constants and quantification of this variability should be considered as an important aspect for uncertainty analysis when these data are later used in defined approaches for potency assessment in a continuous scale. For most validated in vitro assays, only reproducibility of hazard classification was fully documented in validation studies, but not reproducibility of the continuous outputs (peptide depletion, concentration-response for biological marker induction or cytotoxicity).
For four of the 24 test substances (CAR, MHD, PPA, IE), a higher intra-laboratory variability was observed (logarithmic standard deviation between 0.35 and 0.45), but this increased variability is also observed in multiple labs (see below) and it appeared to be intrinsic to certain test substances. While the exact reason for enhanced variability for specific test substances is not known, some explanations can be given. Very slowly reacting test substances like CAR and MHD may have higher variation, as variation over prolonged incubation time may be cumulative. Test substances triggering peptide oxidation like PPA are known to be subject to higher variability. For PPA this had already been observed in the DPRA pre-validation study (reported in (Dimitrov et al. 2016). Finally, pre-haptens spontaneously oxidizing like IE may be more variable, as autoxidation is known to be a self-catalyzed process and hence prone to more stochastic effects.
Intra-laboratory data for prediction of UN GHS sensitizer classes according to the prediction cut-off log kmax -2 are provided in Table 3.
In summary, using the kDPRA to differentiate GHS Cat 1A vs. GHS Cat 1B/not classified, attribution to GHS Cat 1A was consistent in 81 of 84 instances, hence intra-laboratory reproducibility with the classification prediction model for identifying 1A test substances is at 96%. Variable predictions were mostly observed for test substances with a log kmax very close to the classification cut-off.
Based on data in Figure 3 and 4 and in Table 3, the kDPRA has proven to be very reproducible in intra-laboratory testing both when predicting binary classification as well as when considering the numerical log kmax values on a continuous scale.

3.2.3
Inter-laboratory reproducibility Log kmax values obtained during the inter-laboratory reproducibility assessment are summarized in Table 4 and Figure 5. These data demonstrate the inter-laboratory variability of the 7 laboratories for all 24 test substances. For most test substances the interlaboratory reproducibility was high (little spread of the values around the mean), and the average standard deviation for inter-

Fig. 3: Intra-laboratory testing: Variability expressed as average values and standard deviation in repeated intra-laboratory testing (3 times each) in four labs
For test substances not reactive (log kmax < -3.46 corresponding to Cys-depletion of < 13.89% at 5 mM after 24 h) a default value of -3.5 was indicated to allow plotting the results. Abbreviated test substance names (see Table 1) and number attributed to the test laboratory for testing that particular test substance are indicated on the x-axis. The solid red line indicates the cut-off log kmax = -2.0.

Fig. 4: Intra-laboratory testing: Variability expressed as average values and standard deviation in repeated intra-laboratory testing (3 times each) in three labs
For test substances not reactive (log kmax < -3.46, corresponding to Cys-depletion of <13.89% at 5 mM after 24 h) a default value of -3.5 was indicated to allow plotting the results. Abbreviated test substance names (see Table 1) and number attributed to the test laboratory for testing that particular test substance are indicated on the x-axis. The solid red line indicates the cut-off log kmax = -2.0. laboratory comparison is at 0.244, and thus slightly higher as compared to intra-laboratory comparisons. Again, quantification of uncertainty of this rate constant determinations on a continuous scale are an important attribute for uncertainty analysis of subsequent models and risk assessments relying on log kmax values. In general, similar test substances which had higher variability in intra-laboratory testing also exhibited higher inter-laboratory variability (GLY, CAR, MHD, PPA), indicating that it is an intrinsic property of the test substance and not experimental variability due to the experimental procedures.
There is one significant outlier in the whole data-set: Lab E reported very low / no reactivity (in repeated intralaboratory testing) for IU, thus for some unknown reason this lab obtains different results for test substances related to formaldehyde. This outlier is very consistent for IU and formaldehyde in that particular lab and it appears to be linked to the chemistry. Formaldehyde does form a reversible peptide-adduct (data not shown), and for unknown reasons the reaction must have been reversed prior to reaction or during the reaction with mBrB in that particular laboratory.
Benzylidene acetone was tested both in the transfer phase and in the blind-coded phase. The log kmax value from the seven labs in the blind coded phase was at -1.89 ± 0.13, while it was at -1.80 ± 0.12 in the transfer phase, hence a very similar result and similar variability is observed in both open and blinded testing.
3 of 3 Bourgeonal n-r n-r n-r n-r n-r n-r n-r n-r n-r n-r n-r n-r 3 of 3 n-r n-r n-r n-r n-r n-r n-r n-r n-r n-r n-r n-r 3 of 3 Hydroxycitronellal n-r n-r n-r 1B 1B 1B n-r n-r n-r 1B 1B n-r n-r 1B n-r 1B 4 of 4 Imidazolidinyl urea 1A 3 of 3 Phenyl benzoate n-r n-r n-r n-r n-r n-r n-r n-r n-r n-r n-r n-r n-r n-r n-r n-r 4 of 4 1A 4 of 4 4-methoxy-acetophenone n-r n-r n-r n-r n-r n-r n-r n-r n-r n-r n-r n-r n-r n-r n-r n-r 4 of 4 Chlorobenzene 1B n-r n-r n-r n-r n-r n-r n-r n-r n-r n-r n-r 3 of 3 Total 81 of 84 a Abbreviations: n-r: non-reactive b AVG indicates the rating of the test substance by the average log kmax determined from all repetitions in a particular lab, this value is below used for inter-laboratory reproducibility of class prediction n-r n-r n-r n-r n-r n-r n-r Chlorobenzene n-r n-r n-r n-r n-r n-r n-r n-r n-r 1 Abbreviations: AVG, indicates the rating of the test substance by the average log kmax determined from all repetitions in all labs; n-r: non-reactive; SD, standard deviation

Fig. 5: Log kmax values from inter-laboratory testing
Shown are the 7 individual lab results (circles; the average from repeated testing for labs that tested a particular test substance several times), the interquartile range box and the average (horizontal line). The average log kmax from three intra-laboratory experiments was taken to make the prediction When applying the prediction cut-off (log kmax = -2.0) to differentiate GHS Cat 1A vs. GHS Cat 1B/GHS not classified, a consistent result was obtained for 22 test substances (when testing each test substance once, Tab. 5) and for 20 test substances (when testing each test substance three times, Table 6). This results in an inter-laboratory reproducibility for identifying 1A test substances of 92% (for laboratories testing the test substance once) and 83% (laboratories testing the test substances three times), respectively, with an average for the two independent evaluations of 88%.
ALTEX preprint published June 10, 2020 doi:10.14573/altex.2004291 12 When evaluating these values vs. other published validation studies where each test substance was typically tested three times, this analysis is a bit more stringent as 50% of the comparisons are made in 4 labs, and the random chance of congruent results falls from 25% to 13% with testing in 4 labs instead of 3 labs (i.e. the more labs, the higher the chance of producing one deviating result).

Conclusion
The kDPRA proved to be transferable to laboratories without hands-on training and highly reproducible results for the positive control were obtained. The within laboratory reproducibility of the kDPRA for assigning GHS Cat 1A was 96% and the between laboratory reproducibility for 24 test substances was 88%.
The average standard deviation of the logarithmic rate in intra-laboratory testing was at 0.158, which corresponds to a variation around the geometric mean of kmax of 1.44-fold, while the average standard deviation for inter-laboratory comparison is at 0.244, which corresponds to a variation of 1.75-fold. This quantification of variability on a continuous scale will be beneficial for uncertainty analysis in risk assessment.
Following an independent peer review, the kDPRA validation study was considered to demonstrate that this method should be acceptable for the predictive identification of skin sensitization potency categories. The kDPRA has been included in the OECD Work Plan for the Test Guidelines Program for inclusion as an additional method in OECD TG 442C as project 4.317. Details and the draft updated OECD TG 442C are expected to be discussed with the OECD expert group on skin sensitization during summer 2020.