Access the full text.
Sign up today, get DeepDyve free for 14 days.
Loading next page...
/lp/ou_press/deriving-a-no-expected-sensitization-induction-level-for-fragrance-GSyeP2jZRm
- Publisher
- Oxford University Press
- Copyright
- © The Author(s) 2018. Published by Oxford University Press on behalf of the Society of Toxicology. All rights reserved. For permissions, please e-mail: journals.permissions@oup.com
- ISSN
- 1096-6080
- eISSN
- 1096-0929
- DOI
- 10.1093/toxsci/kfy135
- Publisher site
- See Article on Publisher Site
Abstract
Abstract Cosmetic regulations prohibit animal testing for the purpose of safety assessment and recent registration, evaluation and authorization of chemicals guidance states that the local lymph node assay (LLNA) in mice shall only be conducted if in vitro data cannot give sufficient information for classification and labeling. However, Quantitative Risk Assessment for fragrance ingredients requires an NESIL (no expected sensitization induction level), a dose not expected to cause induction of skin sensitization in humans. In absence of human data, this is derived from the LLNA and it remains a key challenge for risk assessors to derive this value from nonanimal data. Here we present a workflow using structural information, reactivity data and KeratinoSens results to predict an LLNA result as a point of departure. Specific additional tests (metabolic activation, complementary reactivity tests) are applied in selected cases depending on the chemical domain of a molecule. Finally, in vitro and in vivo data on close analogues are used to estimate uncertainty of the prediction in the specific chemical domain. This approach was applied to three molecules which were subsequently tested in the LLNA and 22 molecules with available and sometimes discordant human and LLNA data. Four additional case studies illustrate how this approach is being applied to recently developed molecules in the absence of animal data. Estimation of uncertainty and how this can be applied to determine a final NESIL for risk assessment is discussed. We conclude that, in the data-rich domain of fragrance ingredients, sensitization risk assessment without animal testing is possible in most cases by this integrated approach. skin sensitization, risk assessment, alternative methods, QRA, point of departure The field of nonanimal testing for skin sensitization has flourished over the past decade, leading to the publication by the Organisation for Economic Cooperation and Development (OECD) of the adverse outcome pathway for skin sensitization, which has divided the sensitization process into mechanistic key events (OECD, 2012). Three test guidelines covering such mechanistic events were published (OECD, 2015a, b, 2016b), which were mainly developed and evaluated for hazard identification. For hazard assessment, these methods overall give a similar balanced accuracy of around 70% when compared with the local lymph node assay (LLNA) in mice and slightly better performance when compared against human data (Hoffmann et al., 2018; Natsch et al., 2013; Urbisch et al., 2015). Hence there is broad agreement that these methods cannot be used as stand-alone tools and that information from several methods needs to be integrated to perform sensitization hazard identification and potency prediction in a weight of evidence approach. Most published work on sensitizer potency assessment has focused on the attribution of chemicals to potency classes (Cottrez et al., 2016; Jaworska et al., 2015; Zang et al., 2017; Zeller et al., 2017), be it the two classes 1A and 1B implemented in the global harmonized system for Classification and Labeling or the 4 ECETOC (ECETOC, 2003) classes based on the LLNA. However, the most important application of potency assessment is for the risk assessment of chemicals when used in consumer products. Currently, the Quantitative Risk Assessment (QRA) for fragrance materials (Api et al., 2008) is used by industry and is a well-defined approach (DA) starting with an NESIL (no expected sensitization induction level). The NESIL is expressed in µg/cm2 and it is used along with safety assessment factors (SAFs) to derive an acceptable exposure level (AEL) of the chemical in a particular product type. The different SAFs are based on: inter-individual variability, the target site of product application, and the use frequency and type of product. The original QRA was recently updated as QRA 2 with refined SAF values (Basketter and Safford, 2016; IDEA, 2016) and taking aggregated exposure from multiple product types applied to the same skin site into account. For materials used for a long time, existing data were used to derive the NESIL (Lalko and Api, 2008). This follows a weight of evidence approach taking into account all available data, including a no observed effect level (NOEL) from a human maximization test (HMT) or a human repeat insult patch test study (HRIPT) (McNamee et al., 2008) and other available animal (eg LLNA) and read-across data. Especially for materials tested or marketed over the last two decades, the LLNA data have become the key information source to derive the NESIL. Historically, a confirmatory human study on selected molecules at a dose close to or below the EC3 of the LLNA (concentration for 3-fold lymphocyte proliferation) has been performed to confirm the NESIL. It has to be kept in mind that from HRIPT studies with a maximal test concentration, which was selected based on an LLNA EC3 value, the human NOEL is largely influenced by the LLNA result. However, many historical human studies were performed to support actual use levels and may have been tested at a maximal concentration well below the expected NOEL. Therefore, the majority of historically available human data provide only an NOEL and not a lowest observed effect value (LOEL). In case both an NOEL and an LOEL are available, sometimes large spacing factors (up to tenfold) were used between the applied test doses. The ability to use the LLNA to generate new data has greatly diminished due to new regulations. For example, in vitro data are prescribed as the key information requirement by the European chemical agency (ECHA), which requests that animal data are only generated if absolutely necessary (ECHA, 2016b). In addition, Cosmetic Regulations in the EU and elsewhere prohibit animal testing for the purpose of safety assessment. Therefore, an alternative process to derive an NESIL for risk assessment purposes is urgently needed. Defined nonanimal testing approaches (DA) using a data integration procedure (DIP) on several test results are one potential way forward (OECD, 2016a) to transform nonanimal data into a quantitative estimate of sensitizer potency. Such proposals were made by us (Natsch et al., 2015) and others (Hirota et al., 2018; Jaworska et al., 2015; Tsujita-Inoue et al., 2014). Such a DIP can in principle directly yield a point of departure (PoD), but it can also be applied as part of an Integrated Approach to Testing and Assessment (IATA) and combined with other lines of evidence, eg structural alerts identified by expert judgment or in silico tools, read-across to related substances or targeted additional testing. Such an integrated assessment ideally gives additional information on prediction uncertainty: Are similar chemicals predicted correctly by the chosen approach? Are suspected modes of action (eg by metabolic activation) relevant to the chemical under investigation? Here we report application of a DA with a DIP previously published (Natsch et al., 2015) along with additional lines of evidence in an IATA to predict a “most likely LLNA EC3” for fragrance chemicals which serves as PoD. An uncertainty assessment on the models used and benchmarking with close analogues enables us to transform this PoD into an NESIL for risk assessment. This approach is exemplified with specific case studies on (1) three chemicals with no in vivo data which were later tested in the LLNA; (2) existing chemicals with both human and LLNA data; and (iii) new chemicals for which this assessment is deemed sufficient for risk assessment. These case studies reflect our approach as we move from animal testing into a risk assessment based on alternative information only. MATERIALS AND METHODS Data Sources of Existing Data Most in vivo and in vitro data on existing chemicals are from our previous publication (Natsch et al., 2015) or from the data compilation by Urbisch et al. (2015) and Jaworska et al. (2015). Further in vivo data were obtained from the database of the Research institute on fragrance materials. KeratinoSens, DPRA, and LC-MS–Based Peptide Reactivity Additional KeratinoSens data were generated according OECD guideline 442d and additional direct peptide reactivity assay (DPRA) data according to OECD 442c. In selected cases, the reaction product from the DPRA was measured in parallel with HPLC-UV as requested by the guideline and with liquid chromatography coupled to mass spectrometry (LC-MS). LC-MS–based peptide reactivity data with the peptide Cor1-C420 were measured as described before (Natsch and Gfeller, 2008). This assay informs both on peptide depletion and on peptide adduct formation with the test chemical. Kinetic peptide depletion data with the Cor1-C420 peptide were obtained and expressed as rate constants as outlined earlier (Natsch et al., 2015). To obtain a kinetic rate constant for adduct formation, additional LC-MS data were generated for selected chemicals: The samples were prepared as for the 24 h endpoint determination (Natsch and Gfeller, 2008), but they were measured repeatedly with LC-MS at 1, 3, 7, 10, 13, 16, and 19 h after the start of the reaction. The adduct peak at the different time point was integrated. Because synthetic standards of the specific peptide adducts of different molecules are not available, the exact response factor for the peptide adduct is not known. Here we used the peak of the adduct with γ-damascone as reference: γ-damascone triggers 100% peptide depletion of the Cor1-C420 peptide after 10 h with the quantitative formation of a single adduct peak. This peak was used to calculate % adduct formation for closely related chemicals. This approach assumes the same response factor and is therefore an approximation. In Silico Prediction by TIMES SS All chemicals were run through the TIMES in silico prediction tool, version 2.27.17 (Patlewicz et al., 2014). For chemicals predicted as sensitizers, the mechanistic class was used as an information source to attribute chemicals into domains for application of local models. KeratinoSens and LC-MS–Based Peptide Reactivity With Metabolic Activation For chemicals with a phenolic substructure tested negative in KeratinoSens and forming no adducts, we recommended to perform an additional test with an external metabolic activation system (see Figure 4 in Natsch and Haupt (2013)). The KeratinoSens assay in presence of S9 system was run if a chemical fulfilled these requirements as described before (Natsch and Haupt, 2013). In addition, the LC-MS experiment was repeated in the presence of metabolic activation (human liver microsomes; HLM) as described in Ball et al. (2011): 0.5 mM of test peptide, 2.5 mM NADPH, and 0.2 mM of test compound were dissolved in a total volume of 200 µl of Na-Phosphate buffer (20 mM, pH 7.4) and amended with 10 µl of HLM (Gentest, Woburn, MA). Samples were mixed by inversion, then incubated at 37°C without shaking for 2 h. The reaction was stopped by adding 60 μl of acetonitrile (ACN). Samples were centrifuged, filtered and analyzed by LC-MS for new adduct peaks. As controls, parallel samples were prepared without the addition of HLM. Cinnamic aldehyde (a direct-acting hapten) was used as positive control in the direct reactivity assay and the pro-hapten eugenol was the positive control for the assay with metabolic activation. Kinetic Reactivity With Butylamine for Aldehydes For chemicals with an aldehyde functionality, additional kinetic data for Schiff base formation with butylamine (BA) were generated with a method slightly modified from a test described before (Natsch et al., 2012). The test chemicals were dissolved at 10 mM in ACN and serial dilutions down to 0.625 mM were prepared in polypropylene microtiter plates. To 100 µl of these dilutions an equal volume of BA dissolved at 0.5 mM in ACN containing 2 M of H2O was added. Samples were incubated for 20 min, 1 h, 3 h, and 6 h at 22°C and the reaction was stopped by adding 50 µl of a 4 mM solution of Fluram (Fluka, Buchs, Switzerland). The fluorescence due to the reaction of the free amine groups with Fluram was measured (excitation at 381 nm, emission at 470 nm). The depletion of the free amine groups was used to calculate rate constants (Supplementary information document 5). Sulfotransferase Activation Assay Activation of benzyl alcohols and conjugated alcohols by sulfotransferase (SULT) under participation of the cofactor 3′-phosphoadenosine-5′-phosphosulfate (PAPS) to form a metabolite with a sulfoxide leaving group is a new mode of action (MoA) probably relevant for skin sensitization (Charpentier et al., 2018; Moss et al., 2016). To assess the potential to covalently modify sulfanyl groups in peptides, benzyl alcohols and related chemicals (0.2 mM) were incubated with rat liver S9 fractions (1 mg protein/ml) in 0.1 mM Tris buffer, pH 7.4 as metabolic system, along with the SULT cofactor PAPS (2 mM) and glutathione (GSH) (1 mM) as trapping agent to indicate sulfanyl group binding to form GSH adducts. Samples were analyzed by LC-MS as described for peptide reactivity experiments. As no synthetic reference standards are available, only peak areas are given assuming a similar response factor when comparing the different test compounds (Supplementary information document 6). Prediction of LLNA EC3 Values With Linear Regression All the in vitro data (EC 1.5, EC3 and IC50 from KeratinoSens; rate constant from peptide reactivity) were normalized and used in the regression equations as described before (Natsch et al., 2015). Several models are available to allow prediction of EC3 values as described in the publication. Specific models have been generated for several reactivity domains to improve predictivity based on common reactivity pathways. The Michael acceptor model is described below as well as a refinement of the model for aldehydes. The global model is intended to be used where a specific model is not available. Global model For the global model, equation 7 described before (Natsch et al., 2015) was used: pEC3=0.04+0.38×log Knorm+0.25×log EC1.5norm+0.25×log IC50norm−0.19×log VPnorm, (1) where Knorm is the normalized rate constant for depletion of the Cor1-C420 peptide, EC1.5norm is the normalized concentration for 1.5-fold luciferase induction in KeratinoSens, IC50norm is the normalized concentration reducing viability of the KeratinoSens cells by 50%, and VPnorm is the normalized vapor pressure. This model yielded a misprediction of the LLNA of 3.45-fold on either side (geometric mean for 244 chemicals). Michael acceptor model For Michael acceptors, we used the regression equation on 43 chemicals (Natsch et al., 2015) which excludes pentaerythritol triacrylate, a highly reactive chemical which is false-negative in the LLNA based on human data, and which is clearly positive in in vitro tests. pEC3=0.51+0.25×log Knorm+0.28×log EC3norm−0.29×log VPnorm. (2) This equation gives the best fit (R2 = 68.3%, n = 43, geometric mean of the misprediction for 43 chemicals is 1.81-fold) and it uses the EC3norm value (concentration for 3-fold induction of luciferase from KeratinoSens). For a few chemicals investigated here, a positive KS result is obtained, but a 3-fold induction is not reached due to cytotoxicity. These chemicals were assessed with a slightly modified equation taking advantage of the EC1.5 value. This model gives only slightly lower predictivity on all Michael acceptors assessed: pEC3=0.22+0.28×log Knorm+0.29×log EC1.5norm−0.28×log VPnorm. (3) This equation gives the best fit (R2 = 65.3%, n = 43, 1.84-fold geometric mean misprediction). Aldehydes model The previously published predictive equation for aldehydes is: pEC3=0.545+0.298×log EC3norm+0.535×log IC50norm−0.221×log VPnorm. (4) This equation is based on KeratinoSens data only, and does not contain a reactivity parameter, as quantitative peptide reactivity did not statistically correlate to potency for aldehydes. This is due to the fact that peptide depletion by aldehydes is mainly due to oxidization of Cys-residues and adduct formation is not routinely observed. We therefore tested reaction rates for 44 aldehydes with available LLNA data with BA in ACN in presence of 1 M of H2O to obtain a quantitative amine reactivity parameter. Adding this parameter to regression analysis, a new prediction model for aldehydes was derived: pEC3=0.357+0.240×log Knorm Butylamine+0.788×log IC50 norm−0.129×log VPnorm–0.375×c log P. (5) This equation gives the best fit (R2 = 55.3%, n = 44). Where Knorm Butylamine is the normalized reaction constant of aldehydes with BA, IC50 is cytotoxicity in KeratinoSens, and c log P is the calculated octanol-water partition coefficient. This model is mainly influenced by reactivity with BA, cytotoxicity and physicochemical parameters. It predicts the LLNA EC3 for aldehydes with a 1.9-fold misprediction (geometric mean)/2.4-fold arithmetic mean, whereas the previous model in equation (4) gives an misprediction of 2.0-fold (geometric mean)/3.7-fold (arithmetic mean). The details on the study to derive this refined aldehyde model are given in Supplementary information document 5. RESULTS Description of the IATA to Perform Risk Assessment The general scheme of the workflow of the IATA is shown in Figure 1. For each chemical, KeratinoSens data, DPRA and kinetic peptide reactivity data are collected. Peptide adduct formation is monitored in the LC-MS assay to determine direct reactivity with the test peptide and to discriminate direct reactivity from peptide oxidation. A prediction by TIMES SS completes the initial data set. Based on the structural alert obtained by expert judgment, reactivity results (adduct formation) and TIMES, chemicals are attributed to reactivity domains. We then apply the regression models for sensitizer potency assessment (Natsch et al., 2015) to derive a predicted LLNA EC3 value based on a global model and, if applicable, a local model. This approach can be viewed as a “DA to testing and assessment” with the regression models as a “DIP” according recent OECD terminology (OECD, 2016a). Figure 1. View largeDownload slide The overall IATA incorporating our previously published defined approach combined with additional evidence and uncertainty assessment. Figure 1. View largeDownload slide The overall IATA incorporating our previously published defined approach combined with additional evidence and uncertainty assessment. Based on structural alerts a decision is made whether further tests are available which may improve the reliability of the prediction. The following tests were available for this purpose: (1) peptide reactivity in presence of HLM as metabolic system for phenolic prohaptens (Ball et al., 2011); (2) KeratinoSens in presence of Arochlor induced rat S9 fractions for phenolic prohaptens(Natsch and Haupt, 2013); (3) amine reactivity test for aldehydes; (4) a kinetic assessment of peptide adduct formation with LC-MS; and (5) GSH reactivity in presence of SULT co-factor PAPS. All results are summarized in a summary table as shown below for the reference chemical Citral (Figure 2). This leads to an estimated LLNA EC3, which is used as a PoD. Figure 2. View largeDownload slide Case study on Citral—assessment follows the Scheme shown in Figure 1. Figure 2. View largeDownload slide Case study on Citral—assessment follows the Scheme shown in Figure 1. To gain information on the uncertainty of the prediction, the database for molecules with available in vivo and in vitro data (n > 400) is searched for closely related structures. This structure-based search is based on the chemical functional group suspected to be involved in reactivity and the most closely related structures are selected. The rationale for selecting these analogues is always given. We then perform the prediction for these molecules by the DA and the DIP, and compare the outcome with the actual in vivo data. These comparisons help to estimate uncertainty of predictions in this very specific domain and thus also for the target chemical. In addition, it indicates whether the local or global model best predicts the close relatives and which model is more conservative. This read-across based uncertainty assessment is summarized in a second table, as outlined below again for Citral in Figure 2. Both tables for each molecule discussed in this paper are presented in Supplementary information documents 1–4. The final PoD derived from the analysis in the first table can then be adjusted based on the uncertainty assessment to arrive at a final NESIL used in QRA. The rationale for possible adjustment factors is discussed in the Discussion section and Figure 6. Case Studies on Existing Molecules With Congruent LLNA and Human Data For a limited number of older fragrance molecules with a clear skin sensitization potential, human data from HRIPT or HMT with an actual LOEL and an NOEL value do exist. For most of these also a positive LLNA study is available. These latter molecules were selected here as case studies on data rich molecules: they allow testing how close the PoD from the IATA is to an NESIL derived from either human or LLNA data, and this PoD can then also be compared with the human LOEL. Here we first show this approach in detail for Citral (Figure 2). This substance is very data-rich with both the human and the LLNA NESIL derived from multiple studies and it was therefore also used as an exemplary molecule when setting up QRA (Api and Vey, 2008; Lalko and Api, 2008). Citral can in principle react with skin proteins by both Michael addition (MA) or by a Schiff base MoA. Experimental data with the LC-MS peptide assay prove the MA MoA, whereas the Schiff base formation test indicates weak amine reactivity (weaker as compared with the structurally related, weak sensitizer citronellal; Supplementary information document 5), hence this latter MoA is considered of lower importance (ie conferring lower sensitizer potency) and the molecule is predicted with the global and the MA model. The related molecules farnesal and safranal are the closest analogues in the database (β-alkyl-substituted, αβ-unsaturated aldehydes). They are predicted with the local model within a 2-fold error on the conservative side. Therefore the predicted EC3 of 6.8% from the local model is used as PoD (1700 µg/cm2) and is considered certain based on accuracy for the related molecules. Three (overlapping) sets of LLNA studies on Citral are in the literature, indicating an LLNA EC3 of 5.7% (weighted average of 11 studies (Lalko and Api, 2008), 9.3% (median of 6 studies [Hoffmann, 2015]) and 8.8% (dose-response modeling of 7 studies [Bil et al., 2017]). The first is used in the current IFRA (international fragrance association) standard (1425 µg/cm2). The same NESIL is also derived from HRIPT studies in humans, indicating that the PoD derived from the IATA for citral and both human and animal data come to very similar conclusions. A summary of the IATA-derived PoD and the result of the uncertainty assessment for an additional 14 fragrance molecules with reported human LOEL are presented in Table 1, along with both the human and the animal derived NESIL currently used by IFRA. All details for these case studies are given in Supplementary information document 1. Table 1. Case studies on Sensitizers With Congruent Human and LLNA Data Leading to Similar NESIL Chemical NESIL Human (Human NOEL) (µg/cm2) Human LOEL (µg/cm2) NESIL/EC3 LLNA (µg/cm2) PoD IATA (µg/cm2) Uncertainty Assessment IATA PoD Model Used and Discussion of IATA PoD Adjustement Factor to Derive NESIL IATA-derived NESIL (µg/cm2) Citral 1400 3876 1414 1700 High certainty MA model 2 850 Phenylacetaldehyde 590 1180 962 1250 High certainty New aldehyde model 2 625 Cinnamic aldehyde 591 775 262 575 High certainty MA model 2 288 Cinnamic alcohol 3000 4724 5250 5425 High certainty, predictions of analogues on conservative side Global model, different possible metabolic routes not taken into account 2 2712 Isoeugenol 250 775 498 400 Limited; analogues well predicted Due to prehapten nature of molecule, little experience for read-across, thus conservative assessment with global model chosen 2 if taking conservative model 200 2-Phenyl-propionaldehyde 388 1938 1575 2400 High certainty New aldehyde model, underestimation human NESIL, partly due to high spacing factor human test 2 1200 2-Hexyliden cyclopentanone 300 500 600 1100 High certainty MA model 2 550 Safranal 39 394 1875 850 High certainty MA model; Note: Underestimation human NESIL, partly due to high spacing factor human test 2 425 Perillaaldehyde 709 2760 2175 800 High certainty MA model 2 400 Hydroxycitronellal 5000 5906 5612 12750 High certainty New aldehyde model; relatively high amine reactivity, but low cytotoxicity 2 6375 Methyl-2-octynoate 118 194 112.5a 222 Global model indicates higher certainty, certain with global model Global model 2 if taking conservative model 111 Damascenone 100 (group level)b 500 (group level) 308 475 High certainty MA model 2 238 3-Propylidenphtalide 945 2760 925 6625 High uncertainty based on analogues Global model, known underprediction for amine binding chemicals 10 662 Carvone 2657 18 898 2650 3925 High certainty MA model. Note: High spacing factor human test 2 1963 Tropional 11 811 15 000 4100 5850 High certainty New aldehyde model 2 2925 Chemical NESIL Human (Human NOEL) (µg/cm2) Human LOEL (µg/cm2) NESIL/EC3 LLNA (µg/cm2) PoD IATA (µg/cm2) Uncertainty Assessment IATA PoD Model Used and Discussion of IATA PoD Adjustement Factor to Derive NESIL IATA-derived NESIL (µg/cm2) Citral 1400 3876 1414 1700 High certainty MA model 2 850 Phenylacetaldehyde 590 1180 962 1250 High certainty New aldehyde model 2 625 Cinnamic aldehyde 591 775 262 575 High certainty MA model 2 288 Cinnamic alcohol 3000 4724 5250 5425 High certainty, predictions of analogues on conservative side Global model, different possible metabolic routes not taken into account 2 2712 Isoeugenol 250 775 498 400 Limited; analogues well predicted Due to prehapten nature of molecule, little experience for read-across, thus conservative assessment with global model chosen 2 if taking conservative model 200 2-Phenyl-propionaldehyde 388 1938 1575 2400 High certainty New aldehyde model, underestimation human NESIL, partly due to high spacing factor human test 2 1200 2-Hexyliden cyclopentanone 300 500 600 1100 High certainty MA model 2 550 Safranal 39 394 1875 850 High certainty MA model; Note: Underestimation human NESIL, partly due to high spacing factor human test 2 425 Perillaaldehyde 709 2760 2175 800 High certainty MA model 2 400 Hydroxycitronellal 5000 5906 5612 12750 High certainty New aldehyde model; relatively high amine reactivity, but low cytotoxicity 2 6375 Methyl-2-octynoate 118 194 112.5a 222 Global model indicates higher certainty, certain with global model Global model 2 if taking conservative model 111 Damascenone 100 (group level)b 500 (group level) 308 475 High certainty MA model 2 238 3-Propylidenphtalide 945 2760 925 6625 High uncertainty based on analogues Global model, known underprediction for amine binding chemicals 10 662 Carvone 2657 18 898 2650 3925 High certainty MA model. Note: High spacing factor human test 2 1963 Tropional 11 811 15 000 4100 5850 High certainty New aldehyde model 2 2925 Summary of case studies shown in detail in Supplementary information document 1. Chemicals were selected for which sensitization in humans was observed and an LOEL is available as well as a positive LLNA. a Three studies, <125; 112.5; >250. b Several studies on different damascone isomers, RIFM applied a WoE LOEL and NOEL to damascones. Table 1. Case studies on Sensitizers With Congruent Human and LLNA Data Leading to Similar NESIL Chemical NESIL Human (Human NOEL) (µg/cm2) Human LOEL (µg/cm2) NESIL/EC3 LLNA (µg/cm2) PoD IATA (µg/cm2) Uncertainty Assessment IATA PoD Model Used and Discussion of IATA PoD Adjustement Factor to Derive NESIL IATA-derived NESIL (µg/cm2) Citral 1400 3876 1414 1700 High certainty MA model 2 850 Phenylacetaldehyde 590 1180 962 1250 High certainty New aldehyde model 2 625 Cinnamic aldehyde 591 775 262 575 High certainty MA model 2 288 Cinnamic alcohol 3000 4724 5250 5425 High certainty, predictions of analogues on conservative side Global model, different possible metabolic routes not taken into account 2 2712 Isoeugenol 250 775 498 400 Limited; analogues well predicted Due to prehapten nature of molecule, little experience for read-across, thus conservative assessment with global model chosen 2 if taking conservative model 200 2-Phenyl-propionaldehyde 388 1938 1575 2400 High certainty New aldehyde model, underestimation human NESIL, partly due to high spacing factor human test 2 1200 2-Hexyliden cyclopentanone 300 500 600 1100 High certainty MA model 2 550 Safranal 39 394 1875 850 High certainty MA model; Note: Underestimation human NESIL, partly due to high spacing factor human test 2 425 Perillaaldehyde 709 2760 2175 800 High certainty MA model 2 400 Hydroxycitronellal 5000 5906 5612 12750 High certainty New aldehyde model; relatively high amine reactivity, but low cytotoxicity 2 6375 Methyl-2-octynoate 118 194 112.5a 222 Global model indicates higher certainty, certain with global model Global model 2 if taking conservative model 111 Damascenone 100 (group level)b 500 (group level) 308 475 High certainty MA model 2 238 3-Propylidenphtalide 945 2760 925 6625 High uncertainty based on analogues Global model, known underprediction for amine binding chemicals 10 662 Carvone 2657 18 898 2650 3925 High certainty MA model. Note: High spacing factor human test 2 1963 Tropional 11 811 15 000 4100 5850 High certainty New aldehyde model 2 2925 Chemical NESIL Human (Human NOEL) (µg/cm2) Human LOEL (µg/cm2) NESIL/EC3 LLNA (µg/cm2) PoD IATA (µg/cm2) Uncertainty Assessment IATA PoD Model Used and Discussion of IATA PoD Adjustement Factor to Derive NESIL IATA-derived NESIL (µg/cm2) Citral 1400 3876 1414 1700 High certainty MA model 2 850 Phenylacetaldehyde 590 1180 962 1250 High certainty New aldehyde model 2 625 Cinnamic aldehyde 591 775 262 575 High certainty MA model 2 288 Cinnamic alcohol 3000 4724 5250 5425 High certainty, predictions of analogues on conservative side Global model, different possible metabolic routes not taken into account 2 2712 Isoeugenol 250 775 498 400 Limited; analogues well predicted Due to prehapten nature of molecule, little experience for read-across, thus conservative assessment with global model chosen 2 if taking conservative model 200 2-Phenyl-propionaldehyde 388 1938 1575 2400 High certainty New aldehyde model, underestimation human NESIL, partly due to high spacing factor human test 2 1200 2-Hexyliden cyclopentanone 300 500 600 1100 High certainty MA model 2 550 Safranal 39 394 1875 850 High certainty MA model; Note: Underestimation human NESIL, partly due to high spacing factor human test 2 425 Perillaaldehyde 709 2760 2175 800 High certainty MA model 2 400 Hydroxycitronellal 5000 5906 5612 12750 High certainty New aldehyde model; relatively high amine reactivity, but low cytotoxicity 2 6375 Methyl-2-octynoate 118 194 112.5a 222 Global model indicates higher certainty, certain with global model Global model 2 if taking conservative model 111 Damascenone 100 (group level)b 500 (group level) 308 475 High certainty MA model 2 238 3-Propylidenphtalide 945 2760 925 6625 High uncertainty based on analogues Global model, known underprediction for amine binding chemicals 10 662 Carvone 2657 18 898 2650 3925 High certainty MA model. Note: High spacing factor human test 2 1963 Tropional 11 811 15 000 4100 5850 High certainty New aldehyde model 2 2925 Summary of case studies shown in detail in Supplementary information document 1. Chemicals were selected for which sensitization in humans was observed and an LOEL is available as well as a positive LLNA. a Three studies, <125; 112.5; >250. b Several studies on different damascone isomers, RIFM applied a WoE LOEL and NOEL to damascones. Table 1 indicates that in most cases the IATA PoD is close to the LLNA and human-derived NESIL. We calculated fold-differences between the different predictions. Overall the IATA PoD is 1.4-fold (geometric mean) higher than the LLNA NESIL and 2.4-fold higher than the human NESIL, whereas the LLNA NESIL is 1.7-fold higher than the human NESIL. When comparing to the human LOEL, the IATA derived PoD correlates well to the human LOEL (Figure 3). The regression line of the logarithmic plot has a y-intercept of 0.29, indicating the IATA PoD is 2-fold below human LOEL. Figure 3. View largeDownload slide Correlation of IATA PoD and LLNA EC3 values with human LOEL for chemicals in Table 1. Figure 3. View largeDownload slide Correlation of IATA PoD and LLNA EC3 values with human LOEL for chemicals in Table 1. There is one case of a strong misprediction for both LLNA and IATA: Safranal has a 22-fold higher PoD by IATA and 44-fold higher by the LLNA versus the human NOEL, however this is significantly influenced by the spacing factor of 10 in the human study (NOEL 39, LOEL 394 µg/cm2), and therefore the discrepancy between the IATA PoD and the human LOEL is much smaller. The second chemical rated as far too weak is 3-propylidenphtalide, with a 7-fold higher PoD by IATA than indicated by both LLNA and human data. This chemical is predominantly binding to amine groups as shown by LC-MS test, DPRA and the BA binding test. The low reliability of this assessment is also indicated by the uncertainty assessment on two related amine-binding molecules (see Supplementary information). Case Studies on Existing Molecules With LLNA and/or Human Data Available With Partial or Conflicting in Vivo Evidence The selection of fragrance chemicals with complete human (NOEL and LOEL) and animal in vivo evidence as for the chemicals in Table 1 is limited. Next we specifically selected cases of molecules with existing human and animal data, but partly conflicting evidence or lacking human LOEL, to test what risk assessment measures would be taken when only following the IATA presented here as compared with human or animal data. The specific reason for selecting these more difficult cases is indicated in Table 2. The summary of the NESIL derived by the IATA and the NESIL from human or LLNA evidence is given in Table 2. The detailed case studies are reported in document 2 of the Supplementary information. Table 2. Case Studies for Existing Chemicals With Partially Contrasting Human and LLNA Data Chemical Selection Criteria NESIL Human (Human NOEL) (µg/cm2) NESIL LLNA (µg/cm2) PoD IATA (µg/cm2) Uncertainty Assessment IATA PoD Model Used and Discussion of IATA PoD Ajuste-ment Factor to Derive NESIL IATA-derived NESIL (µg/cm2) Cinnamyl nitrile Structurally different (nitrile), LLNA versus human differences 1060 (LOEL 1938) No NESIL, non-sensitizing (>2500; systemic toxicity > 10%) 9750 High certainty MA model; more protective than LLNA, but less protective than human data 2 4875 HMPCC (Lyral) Test approach for substance with high clinical human sensitization rate 4000 (No LOEL)a 4275 5800 High certainty, predictions of analogues on conservative side New aldehyde model, all three assessment give similar PoD 2 2900 Methyl-2-nonynoate Very low human NESIL 24 (LOEL 118) 625 222 Global model indicates higher certainty, certain with global model Global model, lower PoD than LLNA, higher compared with human which is heavily influenced by the large spacing between tested doses 2 111 Benzyl salicylate Very potent in LLNA, but not in humans 17 700 (No LOEL) 2) 725 4700 High certainty Global model, PoD between LLNA and human data, consistent with low human sensitization frequency 2 2350 Iso-methylionone Big difference LLNA versus human 70 000 (No LOEL) 2) 5450 Non-sensitizing High certainty MA model, no NESIL, not reactive, consistent with very low human sensitization potential from HRIPT and clinical data n.a. No NESIL, non-sensitizer Anisylalcohol Good LLNA and human correlation for NESIL, but only due to lacking LOEL and low max. test conc. in human test 3450 (No LOEL) 2) 1475 Non-sensitizing High certainty for LLNA; but underestimation for human sensitization of analogues Lack of reactivity consistent with lack of very low human sensitization potential, Prohapten MoA only recognized by additional SULT assay and assessment of metaboliteb n.a. No NESIL, non-sensitizer Benzyl alcohol LLNA NS but LOEL in humans 5900 (LOEL 8858) >12 500, No NESIL, non-sensitizing Non-sensitizing High certainty Human sensitization potential not recognized by in vitro test battery, recognized by additional SULT assayb n.a. No NESIL, non-sensitizer Chemical Selection Criteria NESIL Human (Human NOEL) (µg/cm2) NESIL LLNA (µg/cm2) PoD IATA (µg/cm2) Uncertainty Assessment IATA PoD Model Used and Discussion of IATA PoD Ajuste-ment Factor to Derive NESIL IATA-derived NESIL (µg/cm2) Cinnamyl nitrile Structurally different (nitrile), LLNA versus human differences 1060 (LOEL 1938) No NESIL, non-sensitizing (>2500; systemic toxicity > 10%) 9750 High certainty MA model; more protective than LLNA, but less protective than human data 2 4875 HMPCC (Lyral) Test approach for substance with high clinical human sensitization rate 4000 (No LOEL)a 4275 5800 High certainty, predictions of analogues on conservative side New aldehyde model, all three assessment give similar PoD 2 2900 Methyl-2-nonynoate Very low human NESIL 24 (LOEL 118) 625 222 Global model indicates higher certainty, certain with global model Global model, lower PoD than LLNA, higher compared with human which is heavily influenced by the large spacing between tested doses 2 111 Benzyl salicylate Very potent in LLNA, but not in humans 17 700 (No LOEL) 2) 725 4700 High certainty Global model, PoD between LLNA and human data, consistent with low human sensitization frequency 2 2350 Iso-methylionone Big difference LLNA versus human 70 000 (No LOEL) 2) 5450 Non-sensitizing High certainty MA model, no NESIL, not reactive, consistent with very low human sensitization potential from HRIPT and clinical data n.a. No NESIL, non-sensitizer Anisylalcohol Good LLNA and human correlation for NESIL, but only due to lacking LOEL and low max. test conc. in human test 3450 (No LOEL) 2) 1475 Non-sensitizing High certainty for LLNA; but underestimation for human sensitization of analogues Lack of reactivity consistent with lack of very low human sensitization potential, Prohapten MoA only recognized by additional SULT assay and assessment of metaboliteb n.a. No NESIL, non-sensitizer Benzyl alcohol LLNA NS but LOEL in humans 5900 (LOEL 8858) >12 500, No NESIL, non-sensitizing Non-sensitizing High certainty Human sensitization potential not recognized by in vitro test battery, recognized by additional SULT assayb n.a. No NESIL, non-sensitizer Summary of case studies shown in detail in Supplementary information document 2. a No sensitization was observed, only NOEL available which is dependent on maximal tested concentration. b Benzyl alcohols are known as substructure in drugs causing immunological cutaneous reactions due to activation by sulfotransferases (SULTs). SULT activation results for benzyl and anisic alcohols shown in Supplementary information document 6. Table 2. Case Studies for Existing Chemicals With Partially Contrasting Human and LLNA Data Chemical Selection Criteria NESIL Human (Human NOEL) (µg/cm2) NESIL LLNA (µg/cm2) PoD IATA (µg/cm2) Uncertainty Assessment IATA PoD Model Used and Discussion of IATA PoD Ajuste-ment Factor to Derive NESIL IATA-derived NESIL (µg/cm2) Cinnamyl nitrile Structurally different (nitrile), LLNA versus human differences 1060 (LOEL 1938) No NESIL, non-sensitizing (>2500; systemic toxicity > 10%) 9750 High certainty MA model; more protective than LLNA, but less protective than human data 2 4875 HMPCC (Lyral) Test approach for substance with high clinical human sensitization rate 4000 (No LOEL)a 4275 5800 High certainty, predictions of analogues on conservative side New aldehyde model, all three assessment give similar PoD 2 2900 Methyl-2-nonynoate Very low human NESIL 24 (LOEL 118) 625 222 Global model indicates higher certainty, certain with global model Global model, lower PoD than LLNA, higher compared with human which is heavily influenced by the large spacing between tested doses 2 111 Benzyl salicylate Very potent in LLNA, but not in humans 17 700 (No LOEL) 2) 725 4700 High certainty Global model, PoD between LLNA and human data, consistent with low human sensitization frequency 2 2350 Iso-methylionone Big difference LLNA versus human 70 000 (No LOEL) 2) 5450 Non-sensitizing High certainty MA model, no NESIL, not reactive, consistent with very low human sensitization potential from HRIPT and clinical data n.a. No NESIL, non-sensitizer Anisylalcohol Good LLNA and human correlation for NESIL, but only due to lacking LOEL and low max. test conc. in human test 3450 (No LOEL) 2) 1475 Non-sensitizing High certainty for LLNA; but underestimation for human sensitization of analogues Lack of reactivity consistent with lack of very low human sensitization potential, Prohapten MoA only recognized by additional SULT assay and assessment of metaboliteb n.a. No NESIL, non-sensitizer Benzyl alcohol LLNA NS but LOEL in humans 5900 (LOEL 8858) >12 500, No NESIL, non-sensitizing Non-sensitizing High certainty Human sensitization potential not recognized by in vitro test battery, recognized by additional SULT assayb n.a. No NESIL, non-sensitizer Chemical Selection Criteria NESIL Human (Human NOEL) (µg/cm2) NESIL LLNA (µg/cm2) PoD IATA (µg/cm2) Uncertainty Assessment IATA PoD Model Used and Discussion of IATA PoD Ajuste-ment Factor to Derive NESIL IATA-derived NESIL (µg/cm2) Cinnamyl nitrile Structurally different (nitrile), LLNA versus human differences 1060 (LOEL 1938) No NESIL, non-sensitizing (>2500; systemic toxicity > 10%) 9750 High certainty MA model; more protective than LLNA, but less protective than human data 2 4875 HMPCC (Lyral) Test approach for substance with high clinical human sensitization rate 4000 (No LOEL)a 4275 5800 High certainty, predictions of analogues on conservative side New aldehyde model, all three assessment give similar PoD 2 2900 Methyl-2-nonynoate Very low human NESIL 24 (LOEL 118) 625 222 Global model indicates higher certainty, certain with global model Global model, lower PoD than LLNA, higher compared with human which is heavily influenced by the large spacing between tested doses 2 111 Benzyl salicylate Very potent in LLNA, but not in humans 17 700 (No LOEL) 2) 725 4700 High certainty Global model, PoD between LLNA and human data, consistent with low human sensitization frequency 2 2350 Iso-methylionone Big difference LLNA versus human 70 000 (No LOEL) 2) 5450 Non-sensitizing High certainty MA model, no NESIL, not reactive, consistent with very low human sensitization potential from HRIPT and clinical data n.a. No NESIL, non-sensitizer Anisylalcohol Good LLNA and human correlation for NESIL, but only due to lacking LOEL and low max. test conc. in human test 3450 (No LOEL) 2) 1475 Non-sensitizing High certainty for LLNA; but underestimation for human sensitization of analogues Lack of reactivity consistent with lack of very low human sensitization potential, Prohapten MoA only recognized by additional SULT assay and assessment of metaboliteb n.a. No NESIL, non-sensitizer Benzyl alcohol LLNA NS but LOEL in humans 5900 (LOEL 8858) >12 500, No NESIL, non-sensitizing Non-sensitizing High certainty Human sensitization potential not recognized by in vitro test battery, recognized by additional SULT assayb n.a. No NESIL, non-sensitizer Summary of case studies shown in detail in Supplementary information document 2. a No sensitization was observed, only NOEL available which is dependent on maximal tested concentration. b Benzyl alcohols are known as substructure in drugs causing immunological cutaneous reactions due to activation by sulfotransferases (SULTs). SULT activation results for benzyl and anisic alcohols shown in Supplementary information document 6. For these molecules, the picture appears more complex. When comparing against LLNA, two chemicals (benzyl alcohol and HMPCC) are rated similar, although for two chemicals a more conservative assessment is made by the IATA (cinnamyl nitrile and methyl-nonynoate) and for three chemicals the LLNA would predict a clearly lower NESIL (anisic alcohol, benzyl salicylate, and isomethylionone). When looking at the human data, the outcome is quite different: For the latter three molecules no human sensitization was observed in predictive tests, and clinical data indicate low to very low frequency of positive reactions, hence the human data align better with the IATA results as compared with the LLNA. On the other hand, human data are still more conservative than the IATA PoD for cinnamyl nitrile (6-fold), methyl-nonynoate (10-fold, but there is a large spacing factor in the human test) and benzyl alcohol (not predicted as sensitizers by IATA; see discussion on potentially unusual mode of action by SULT activation in document 6 in Supplementary information). Case Studies on Three Novel Molecules, Later Challenged by LLNA Data In the discovery process on fragrance materials a few years back (2013–early 2016, ie at a time when REACH (registration, evaluation and authorization of chemicals) regulation was still considering LLNA as mandatory information source), most new fragrance chemicals in our internal discovery program (n > 500) were internally assessed with the DIP presented here. Three market candidates were then studied in more detail as IATA case studies. Only after the assessment presented here was made and fully documented, these candidates were submitted to the LLNA in order to complete the dataset required by REACH regulations at the time. This offered the unique opportunity to challenge the prediction with an actual LLNA result and compare the RA based on LLNA or IATA, without the potential bias of post hoc analysis. We show here the summary of these case studies in Table 3, along with some experimental data from the additional testing with the targeted tests to further illustrate the IATA approach (the detailed cases studies are again in Supplementary information document 3). Table 3. Risk Assessment for Three New Molecules Without Animal Data—Later Challenged by LLNA Chemical Structure TIMES Prediction KS Result Peptide Reactivity Additional Tests PoD IATA (µg/cm2) Uncertainty Assessment IATA PoD Adjuste-ment Factor to Derive NESIL IATA-derived NESIL (µg/cm2) LLNA Resulta α-Methyl-δ-damascone Strong sensitizer, α,β-carbonyl compound with polarized double bonds Negative Cor1C420: 6.8% depletion; <0.5% direct MA adduct; DPRA negative Kinetic adduct formation, Figure 4, 4000-fold reduced reaction rate versus damascones. EC3 60%; 15 000 µg/cm2 Predictions with local model indicate low uncertainty, especially. for human data 2 7500 Negative, EC3 >25%; >6250 µg/cm2 2,6-Dimethylcyclohexyl-crotonate Weak sensitizer, α,β-carbonyl compound with polarized double bonds Negative Cor1C420: 3.7% depletion; ca. 5% direct MA adduct; DPRA low category Kinetic adduct formation, Figure 4, 100-fold reduced reaction rate versus damascones. EC3 30%–40%; 11 000 µg/cm2 Predictions with local model indicate low uncertainty 2 5500 Positive, EC3 21%; 5450 µg/cm2 (E)-3-ethoxy-4-hydroxybenzaldehyde O-methyl oxime Parent: non-sensitizerMetabolite: strong sensitizer, quinoide oxime structure Negative Cor1C420: 5.7% depletion; no adduct; DPRA negative Test in presence of metabolic system:Small trace of peptide adduct in presence of microsomes, positive in S9 KS (see text) Very weak in presence of metabolic activation only; EC3 estimated 30%–50%, 7500 µg/cm2. High certainty by global model for four tested analogues;Remaining uncertainty due to metabolic activation 2 3750 Negative, EC3 >25%; >6250 µg/cm2 Chemical Structure TIMES Prediction KS Result Peptide Reactivity Additional Tests PoD IATA (µg/cm2) Uncertainty Assessment IATA PoD Adjuste-ment Factor to Derive NESIL IATA-derived NESIL (µg/cm2) LLNA Resulta α-Methyl-δ-damascone Strong sensitizer, α,β-carbonyl compound with polarized double bonds Negative Cor1C420: 6.8% depletion; <0.5% direct MA adduct; DPRA negative Kinetic adduct formation, Figure 4, 4000-fold reduced reaction rate versus damascones. EC3 60%; 15 000 µg/cm2 Predictions with local model indicate low uncertainty, especially. for human data 2 7500 Negative, EC3 >25%; >6250 µg/cm2 2,6-Dimethylcyclohexyl-crotonate Weak sensitizer, α,β-carbonyl compound with polarized double bonds Negative Cor1C420: 3.7% depletion; ca. 5% direct MA adduct; DPRA low category Kinetic adduct formation, Figure 4, 100-fold reduced reaction rate versus damascones. EC3 30%–40%; 11 000 µg/cm2 Predictions with local model indicate low uncertainty 2 5500 Positive, EC3 21%; 5450 µg/cm2 (E)-3-ethoxy-4-hydroxybenzaldehyde O-methyl oxime Parent: non-sensitizerMetabolite: strong sensitizer, quinoide oxime structure Negative Cor1C420: 5.7% depletion; no adduct; DPRA negative Test in presence of metabolic system:Small trace of peptide adduct in presence of microsomes, positive in S9 KS (see text) Very weak in presence of metabolic activation only; EC3 estimated 30%–50%, 7500 µg/cm2. High certainty by global model for four tested analogues;Remaining uncertainty due to metabolic activation 2 3750 Negative, EC3 >25%; >6250 µg/cm2 Summary of case studies shown in detail in Supplementary information document 3. a LLNA conducted after this assessment was made and documented. Table 3. Risk Assessment for Three New Molecules Without Animal Data—Later Challenged by LLNA Chemical Structure TIMES Prediction KS Result Peptide Reactivity Additional Tests PoD IATA (µg/cm2) Uncertainty Assessment IATA PoD Adjuste-ment Factor to Derive NESIL IATA-derived NESIL (µg/cm2) LLNA Resulta α-Methyl-δ-damascone Strong sensitizer, α,β-carbonyl compound with polarized double bonds Negative Cor1C420: 6.8% depletion; <0.5% direct MA adduct; DPRA negative Kinetic adduct formation, Figure 4, 4000-fold reduced reaction rate versus damascones. EC3 60%; 15 000 µg/cm2 Predictions with local model indicate low uncertainty, especially. for human data 2 7500 Negative, EC3 >25%; >6250 µg/cm2 2,6-Dimethylcyclohexyl-crotonate Weak sensitizer, α,β-carbonyl compound with polarized double bonds Negative Cor1C420: 3.7% depletion; ca. 5% direct MA adduct; DPRA low category Kinetic adduct formation, Figure 4, 100-fold reduced reaction rate versus damascones. EC3 30%–40%; 11 000 µg/cm2 Predictions with local model indicate low uncertainty 2 5500 Positive, EC3 21%; 5450 µg/cm2 (E)-3-ethoxy-4-hydroxybenzaldehyde O-methyl oxime Parent: non-sensitizerMetabolite: strong sensitizer, quinoide oxime structure Negative Cor1C420: 5.7% depletion; no adduct; DPRA negative Test in presence of metabolic system:Small trace of peptide adduct in presence of microsomes, positive in S9 KS (see text) Very weak in presence of metabolic activation only; EC3 estimated 30%–50%, 7500 µg/cm2. High certainty by global model for four tested analogues;Remaining uncertainty due to metabolic activation 2 3750 Negative, EC3 >25%; >6250 µg/cm2 Chemical Structure TIMES Prediction KS Result Peptide Reactivity Additional Tests PoD IATA (µg/cm2) Uncertainty Assessment IATA PoD Adjuste-ment Factor to Derive NESIL IATA-derived NESIL (µg/cm2) LLNA Resulta α-Methyl-δ-damascone Strong sensitizer, α,β-carbonyl compound with polarized double bonds Negative Cor1C420: 6.8% depletion; <0.5% direct MA adduct; DPRA negative Kinetic adduct formation, Figure 4, 4000-fold reduced reaction rate versus damascones. EC3 60%; 15 000 µg/cm2 Predictions with local model indicate low uncertainty, especially. for human data 2 7500 Negative, EC3 >25%; >6250 µg/cm2 2,6-Dimethylcyclohexyl-crotonate Weak sensitizer, α,β-carbonyl compound with polarized double bonds Negative Cor1C420: 3.7% depletion; ca. 5% direct MA adduct; DPRA low category Kinetic adduct formation, Figure 4, 100-fold reduced reaction rate versus damascones. EC3 30%–40%; 11 000 µg/cm2 Predictions with local model indicate low uncertainty 2 5500 Positive, EC3 21%; 5450 µg/cm2 (E)-3-ethoxy-4-hydroxybenzaldehyde O-methyl oxime Parent: non-sensitizerMetabolite: strong sensitizer, quinoide oxime structure Negative Cor1C420: 5.7% depletion; no adduct; DPRA negative Test in presence of metabolic system:Small trace of peptide adduct in presence of microsomes, positive in S9 KS (see text) Very weak in presence of metabolic activation only; EC3 estimated 30%–50%, 7500 µg/cm2. High certainty by global model for four tested analogues;Remaining uncertainty due to metabolic activation 2 3750 Negative, EC3 >25%; >6250 µg/cm2 Summary of case studies shown in detail in Supplementary information document 3. a LLNA conducted after this assessment was made and documented. α-Methyl-δ-damascone is closely related to the moderate sensitizer δ-damascone. However, a detailed kinetic peptide reactivity study with a kinetic assessment of adduct formation (see Figure 4), indicates that this molecule has a 4000-fold reduced reaction rate as compared with damascones. Adduct formation for such a slowly reacting molecule is more sensitive than peptide depletion assessment, however quantification here is based on the response factor in LC-MS analysis of a very closely related adduct. However, low reactivity is also confirmed by low peptide depletion for the Cys-peptide and the Cor1 peptide. The low reactivity also translates to a negative KeratinoSens result. IATA prediction would indicate a very weak sensitization potential with an EC3 of 60%, and uncertainty assessment indicates high certainty from results on the related molecules δ-damascone and α-methyl-ionone. Actual LLNA result was EC3 > 25%, confirming this assessment (the maximal test concentration of 25% was set based on systemic effects; piloerection, hunched posture at higher doses in a pre-study). Figure 4. View largeDownload slide Kinetic assessment of peptide adduct formation for two new fragrance candidate molecules in Table 3 in comparison to current benchmarks. The candidates have 100- to 4000-fold reduced reactivity versus benchmarks. Figure 4. View largeDownload slide Kinetic assessment of peptide adduct formation for two new fragrance candidate molecules in Table 3 in comparison to current benchmarks. The candidates have 100- to 4000-fold reduced reactivity versus benchmarks. 2,6-Dimethylcyclohexyl-crotonate is a potential MA. It has a weak reactivity in the LC-MS test (100-fold reduced vs. δ-damascone; Figure 4). Based on reaction rate only, EC3 of 26% was predicted, whereas MA model predicts EC3 of 43%. Actually measured LLNA shows weak sensitization potential with EC3 = 21%. A similar crotonate registered under REACH was predicted with an EC3 of 47%, whereas LLNA result is EC3 > 50% (ECHA, 2016a). The third case study is (E)-3-ethoxy-4-hydroxybenzaldehyde O-methyl oxime. This chemical is structurally related to ethyl-vanillin, eugenol, isoeugenol and benzaldoxime. TIMES predicts it as a prohapten, able to form a quinoide oxime structure. The chemical is negative in KeratinoSens and does not form peptide adducts, and thus is rated a non-sensitizer by the global model. However, due to predicted prohapten MoA, further tests were made. In an LC-MS test, traces (ca. 0.3%) of peptide adducts in presence of HLM are observed (Adduct ([M + H]+ =1071.4 indicates direct adduct – 2 mass units, consistent with quinone methide formation and subsequent reaction with test peptide). Quantitatively, reactivity is much lower compared with eugenol (40% adduct in presence of HLM). The KeratinoSens in presence of S9 is positive, too. Expert judgment came to the following conclusion: Based on KS and LC-MS very weak sensitizer in presence of metabolic activation only, EC3 estimated 30%–50%, PoD 7500 µg/cm2. The LLNA then indeed showed no sensitization up to maximal test concentration of 25% (EC3 > 25%; NESIL > 6250 µg/cm2). Based on four analogues including vanillin and benzaldoxime, a low uncertainty for the global model is found. Case Studies on Four Novel Molecules to Derive an NESIL in Absence of LLNA Data Chemicals from our discovery pipeline in 2016–2017 are presented as further case studies, for which no LLNA is currently planned due to changed REACH guidance. These cases thus illustrate how four new chemicals are assessed in absence of animal data. One is predicted as a moderate sensitizer, one as weak and two as non-sensitizers. For summary see Table 4, details are given here for ethyl (Z)-2-acetyl-4-methyltridec-2-enoate predicted as moderate sensitizer (Figure 5), whereas the full case studies on the other three molecules are in Supplementary information document 4. Table 4. Risk Assessment for Four New Molecules Without Animal Data Chemical Structure TIMES Prediction KS Result Peptide Reactivity Additional Tests PoD IATA (µg/cm2) Uncertainty Assessment IATA PoD Adjustement Factor to Derive NESIL IATA-derived NESIL (µg/cm2) Ethyl (Z)-2-acetyl-4-methyltridec-2-enoate Strong sensitizer, α,β-carbonyl compound with polarized double bonds Positive Cor1C420: 14% depletion: 14%; direct MA adduct; DPRA low category Not needed EC3 5.1% (moderate), 1250 µg/cm2 Global model used (conservative), low uncertainty for global model for two close analogues 2 625 2,4,7-Trimethyloct-6-en-1-ol Parent non-sensitizer, metabolite weak sensitizer, hydroperoxide formation Negative Cor1C420: 61%–100% depletion due to peptide oxidation; no adduct; DPRA negative Weak amine reactivity for corresponding aldehyde ⇒ no pro-hapten risk for aldehyde MoA Non-sensitizer, no NESIL High certainty based on two assessed analogues, hydroperoxide formation not relevant if properly stored (Kern et al., 2014) n.a. No NESIL, non-sensitizer (E)-2,4,7-trimethylocta-2,6-dienal Strong sensitizer; α,β-aldehydes Negative Cor1C420: 11% depletion; 0.3% MA adduct; DPRA moderate category due to peptide oxidation only (0.2% adduct) Weak amine reactivity, lower than citronellal, low risk for aldehyde MoA EC3 31%, 7750 µg/cm2 High certainty based on three assessed analogues; reduced sensitization versus citral shown by all tests 2 3875 Ethyl cyclohexanecarboxylate Non-sensitizer Negative Not reactive Not needed Non-sensitizer, no NESIL High certainty based on three assessed analogues n.a. No NESIL, non-sensitizer Chemical Structure TIMES Prediction KS Result Peptide Reactivity Additional Tests PoD IATA (µg/cm2) Uncertainty Assessment IATA PoD Adjustement Factor to Derive NESIL IATA-derived NESIL (µg/cm2) Ethyl (Z)-2-acetyl-4-methyltridec-2-enoate Strong sensitizer, α,β-carbonyl compound with polarized double bonds Positive Cor1C420: 14% depletion: 14%; direct MA adduct; DPRA low category Not needed EC3 5.1% (moderate), 1250 µg/cm2 Global model used (conservative), low uncertainty for global model for two close analogues 2 625 2,4,7-Trimethyloct-6-en-1-ol Parent non-sensitizer, metabolite weak sensitizer, hydroperoxide formation Negative Cor1C420: 61%–100% depletion due to peptide oxidation; no adduct; DPRA negative Weak amine reactivity for corresponding aldehyde ⇒ no pro-hapten risk for aldehyde MoA Non-sensitizer, no NESIL High certainty based on two assessed analogues, hydroperoxide formation not relevant if properly stored (Kern et al., 2014) n.a. No NESIL, non-sensitizer (E)-2,4,7-trimethylocta-2,6-dienal Strong sensitizer; α,β-aldehydes Negative Cor1C420: 11% depletion; 0.3% MA adduct; DPRA moderate category due to peptide oxidation only (0.2% adduct) Weak amine reactivity, lower than citronellal, low risk for aldehyde MoA EC3 31%, 7750 µg/cm2 High certainty based on three assessed analogues; reduced sensitization versus citral shown by all tests 2 3875 Ethyl cyclohexanecarboxylate Non-sensitizer Negative Not reactive Not needed Non-sensitizer, no NESIL High certainty based on three assessed analogues n.a. No NESIL, non-sensitizer Summary of case studies shown in detail in Supplementary information document 4. Table 4. Risk Assessment for Four New Molecules Without Animal Data Chemical Structure TIMES Prediction KS Result Peptide Reactivity Additional Tests PoD IATA (µg/cm2) Uncertainty Assessment IATA PoD Adjustement Factor to Derive NESIL IATA-derived NESIL (µg/cm2) Ethyl (Z)-2-acetyl-4-methyltridec-2-enoate Strong sensitizer, α,β-carbonyl compound with polarized double bonds Positive Cor1C420: 14% depletion: 14%; direct MA adduct; DPRA low category Not needed EC3 5.1% (moderate), 1250 µg/cm2 Global model used (conservative), low uncertainty for global model for two close analogues 2 625 2,4,7-Trimethyloct-6-en-1-ol Parent non-sensitizer, metabolite weak sensitizer, hydroperoxide formation Negative Cor1C420: 61%–100% depletion due to peptide oxidation; no adduct; DPRA negative Weak amine reactivity for corresponding aldehyde ⇒ no pro-hapten risk for aldehyde MoA Non-sensitizer, no NESIL High certainty based on two assessed analogues, hydroperoxide formation not relevant if properly stored (Kern et al., 2014) n.a. No NESIL, non-sensitizer (E)-2,4,7-trimethylocta-2,6-dienal Strong sensitizer; α,β-aldehydes Negative Cor1C420: 11% depletion; 0.3% MA adduct; DPRA moderate category due to peptide oxidation only (0.2% adduct) Weak amine reactivity, lower than citronellal, low risk for aldehyde MoA EC3 31%, 7750 µg/cm2 High certainty based on three assessed analogues; reduced sensitization versus citral shown by all tests 2 3875 Ethyl cyclohexanecarboxylate Non-sensitizer Negative Not reactive Not needed Non-sensitizer, no NESIL High certainty based on three assessed analogues n.a. No NESIL, non-sensitizer Chemical Structure TIMES Prediction KS Result Peptide Reactivity Additional Tests PoD IATA (µg/cm2) Uncertainty Assessment IATA PoD Adjustement Factor to Derive NESIL IATA-derived NESIL (µg/cm2) Ethyl (Z)-2-acetyl-4-methyltridec-2-enoate Strong sensitizer, α,β-carbonyl compound with polarized double bonds Positive Cor1C420: 14% depletion: 14%; direct MA adduct; DPRA low category Not needed EC3 5.1% (moderate), 1250 µg/cm2 Global model used (conservative), low uncertainty for global model for two close analogues 2 625 2,4,7-Trimethyloct-6-en-1-ol Parent non-sensitizer, metabolite weak sensitizer, hydroperoxide formation Negative Cor1C420: 61%–100% depletion due to peptide oxidation; no adduct; DPRA negative Weak amine reactivity for corresponding aldehyde ⇒ no pro-hapten risk for aldehyde MoA Non-sensitizer, no NESIL High certainty based on two assessed analogues, hydroperoxide formation not relevant if properly stored (Kern et al., 2014) n.a. No NESIL, non-sensitizer (E)-2,4,7-trimethylocta-2,6-dienal Strong sensitizer; α,β-aldehydes Negative Cor1C420: 11% depletion; 0.3% MA adduct; DPRA moderate category due to peptide oxidation only (0.2% adduct) Weak amine reactivity, lower than citronellal, low risk for aldehyde MoA EC3 31%, 7750 µg/cm2 High certainty based on three assessed analogues; reduced sensitization versus citral shown by all tests 2 3875 Ethyl cyclohexanecarboxylate Non-sensitizer Negative Not reactive Not needed Non-sensitizer, no NESIL High certainty based on three assessed analogues n.a. No NESIL, non-sensitizer Summary of case studies shown in detail in Supplementary information document 4. Figure 5. View largeDownload slide Case study for risk assessment of a new molecule ethyl (Z)-2-acetyl-4-methyltridec-2-enoate in absence of animal data. Figure 5. View largeDownload slide Case study for risk assessment of a new molecule ethyl (Z)-2-acetyl-4-methyltridec-2-enoate in absence of animal data. (Z)-2-acetyl-4-methyltridec-2-enoate is a MA reactive in LC-MS assay, DPRA and KeratinoSens. Based on two close analogues, a conservative assessment with the global model is made, leading to a predicted EC3 of 5.1% and high certainty can be attributed based on analogues. 2,4,7-Trimethyloct-6-en-1-ol is a new alcohol related to citronellol. It may form hydroperoxides according to TIMES, yet this MoA is only relevant after forced oxidation for many weeks and does not apply to the parent molecule of related terpenes, if properly stored (Kern et al., 2014). Alternatively, it might also be oxidized to the corresponding aldehyde. Yet, the aldehyde 2,4,7-trimethyloct-6-en-1-al has a clearly reduced ability to form Schiff-Bases as compared with citronellal, which itself is a very weak sensitizer (EC3 = 60%). Thus, limited conversion in the skin to this aldehyde would not indicate significant sensitization risks. Based on these considerations, the negative results in LC-MS, DPRA and KeratinoSens and the corresponding lack of a significant sensitization potential can be accepted with high certainty. (E)-2,4,7-trimethylocta-2,6-dienal is a MA aldehyde, structurally related to citral. However, KeratinoSens, LC-MS based adduct formation and the amine binding assay all indicate reduced reactivity as compared with citral. Both the MA model and the global model indicate EC3 of 31%–32% which can be used to derive a PoD of 7750 µg/cm2. Accurate prediction for the close analogues citral, farnesal and safranal (αβ-unsaturated aldehydes with alkyl substituent in α- and/or β-position) indicate high certainty. Ethyl cyclohexanecarboxylate is a saturated ester. It is negative in KeratinoSens, DPRA, LC-MS assay and by TIMES assessment, clearly indicating no sensitization risk. The analogue clofibrate is negative in the LLNA, although cydrane and tropicalia have EC3 values ≥ 50%. These results may be false-positive LLNA results often observed when testing chemicals at > 50% and would not indicate that this class of molecules represents a true sensitization risk. Thus, the negative results by all predictive tests would be accepted and high certainty is attributed to this prediction. DISCUSSION Previous research showed how different in vitro and in silico models contribute to sensitizer potency assessment and how it became possible to classify chemicals into potency classes (Jaworska et al., 2015; Natsch et al., 2015; Tsujita-Inoue et al., 2014; Zang et al., 2017; Zeller et al., 2017). Here we now went the additional step in performing a case by case analysis on a total of 22 existing and 7 new fragrance molecules to arrive at a PoD for risk assessment, and to compare this assessment versus animal and human data where available. We focused specifically on fragrance molecules, because the new QRA2 was developed for fragrance risk assessment, which in the future should be conducted in absence of animal data. The approach chosen here may not be appropriate for all use categories of chemicals, thus for example hair dyes may need additional tests in an IATA such as eg a quantitative, kinetic test to assess protein binding under oxidative conditions. Therefore, the conclusions from this work may be particularly applicable to fragrance molecules, but a similar approach may also be considered for other areas. Deriving a PoD for Existing Molecules The assessment of the 15 molecules with largely congruent human and animal data indicates that the LLNA and the IATA PoD overall predict the human NESIL with a similar accuracy, yet both assessments are a bit less conservative as compared with the human NOEL. However, it should be kept in mind that the LLNA EC3 and the IATA modeled EC3 are not actually NOEL values but interpolated concentrations in the dose–response curve at which an effect is noted (ie measured or predicted 3-fold stimulation of cell proliferation in the local lymph node), whereas the human NESIL is defined as the NOEL (ie at or below the low end of the dose–response curve). Interestingly, when comparing the EC3 from LLNA and the PoD from the IATA with human LOEL a direct correlation between these values is observed with a logarithmic y-intercept of ca. 0.3 (ie human LOEL is in general 2-fold above IATA PoD) (Figure 3). For the chemicals with discordant human and animal data, human predictive tests and patch test data indicate that moving from the LLNA to the IATA would not lead to an overall loss of consumer protection. Indeed, some reactive chemicals like cinnamyl nitrile and methyl-nonynoate would be assessed even more stringently by the IATA as compared with LLNA. For three chemicals (anisic alcohol, benzyl salicylate, and isomethylionone) higher or no use limits would be set by the IATA as compared with LLNA, however, human data for these three chemicals (predictive tests and clinical data [Schnuch et al., 2007]) indicate only (very) low sensitization potential. The maybe most striking case is HMPCC: the IATA, the LLNA, and the human NOEL all predict the same, relatively high NESIL (4000–5800 µg/cm2), but clinical evidence indicates significant frequency of sensitization (Bruze et al., 2008; Schnuch et al., 2009), which may be due either to too high exposures or the possibility that the NESIL based on animal and human data was estimated too high. Assessment of New Chemicals For the three candidates which were only after the assessment tested in the LLNA, weak or very weak sensitization potency was predicted. 2,6-Dimethylcyclohexyl-crotonate was the most reactive chemical and an EC3 of 30%–40% was predicted, whereas experimental EC3 was then determined at 21%. The other two chemicals were predicted with EC3 of 30%–60% and found to be non-sensitizing at maximal test concentration of 25%. These three cases indicate that the low sensitization risk was correctly predicted. Although from the LLNA no NESIL would be derived for two candidates, we derived for all three molecules a PoD as reactivity or metabolic activation tests indicated still a minimal reactivity, and thus the IATA is slightly more conservative. In two cases, the uncertainty assessment indicated that closely related molecules were predicted with high accuracy and thus uncertainty is low. In the third case (3-ethoxy-4-hydroxybenzaldehyde O-methyl oxime) analogues were very well predicted indicating low uncertainty, too. Still we cannot exactly predict how strongly the very weak metabolic activation observed in presence of liver enzymes would translate to sensitization in the skin. Based on a read-across to the metabolic activation data for the related chemical eugenol, we however predict the molecule to be clearly less sensitizing (ie EC3 ≫ 13%), even when accounting for metabolism. Uncertainty: Additional Adjustment of the NESIL When Using Nonanimal Data A key question is whether QRA based on nonanimal data introduces an additional intrinsic uncertainty which needs to be considered when deriving the AEL. We propose to include uncertainty on a case-by-case basis when setting the final NESIL, and not in the form of a generalized SAF. Thus, uncertainty is assessed for each molecule when deriving the PoD. In cases with low uncertainty, we would use the PoD (ie the predicted EC3 value) and divide it by a factor of two to arrive at a final NESIL. With this adjustment factor applied to the IATA PoD, for most cases with low uncertainty the resulting NESIL is close to the human NESIL and below the human LOEL (Table 1). This adjustment on the conservative side takes into consideration that even for assessments with low uncertainty, the prediction is not completely accurate and it does account for the fact that LLNA EC3 (and hence also the modeled EC3 expressed as the PoD) is not actually an NOEL but it is the interpolated dose between NOEL and LOEL in the LLNA (see above). An NESIL adjusted by a factor of two would thus better reflect the low end of the dose response in the LLNA and the NOEL. In cases with higher uncertainty (underprediction of close analogues), the adjustment factor would be based on the error for the analogues, whereas in the case of lack of close analogues in the database, the adjustment factor would be based on the global model uncertainty. For a schematic view of this approach, see Figure 6. Thus, in the case of 3-propylidenphtalide, the uncertainty assessment indicates significantly higher potency for analogues than the predicted PoD, hence here an adjustment factor of 10 would appear appropriate. In addition, the formation of stable adducts with amine groups in peptides on its own could lead to a more stringent assessment of this molecule (see discussion on importance of amine reactivity for potency in Natsch and Emter (2017)). Figure 6. View largeDownload slide Scheme how the PoD from the IATA can be transformed to an NESIL for risk assessment based on the uncertainty assessment. Figure 6. View largeDownload slide Scheme how the PoD from the IATA can be transformed to an NESIL for risk assessment based on the uncertainty assessment. Benefit of Assessment Early in Chemical Discovery Process The key question when using the non-animal testing strategy for potency assessment is: Can a new approach replace the in vivo assessment (in the LLNA)? However, as exemplified based on the presented case studies, there is an additional benefit. As the alternative testing can be performed on many more molecules and earlier in the discovery process, safer molecules can be selected. Thus α-methyl-δ-damascone and (E)-3-ethoxy-4-hydroxybenzaldehyde O-methyl oxime are two new molecules with very low sensitization risk identified by this screening at early stage. These chemicals are mainly designed for topical leave-on application. On the other hand, moderate sensitization risk was identified for ethyl (Z)-2-acetyl-4-methyltridec-2-enoate, and further development of this molecule was thus limited to fabric care products or rinse-off hair products with low skin exposure. These potential areas of safe use could be identified early in the discovery process, which was not the case at times when the LLNA was only conducted briefly before chemical registration. Use of Complementary Tests Recent developments on non-animal testing were mainly focused on finding the “ideal” stand-alone test with higher predictivity and providing a direct potency estimate (Cottrez et al., 2016; Zeller et al., 2017). Here we present an approach where additional testing may be guided by chemical considerations and domain attributions. Thus the simple kinetic amine binding test can improve predictivity for aldehydes, but would not be proposed as a general test for all chemicals. Similarly, we proposed the S9-KeratinoSens assay only for those chemicals with a prohapten alert (Natsch and Haupt, 2013). This approach always will involve more expert judgment in the assessment, but if conducted in a structured and rational workflow this can be properly documented as shown in the present case studies following a defined IATA. Alternative Approaches The ITS presented here is only one of several approaches for sensitizer potency assessment, and a key question is of course: Did we select the best option and how would the results compare to alternative models, especially those also including dendritic cell activation (Hirota et al., 2013; Jaworska et al., 2015) or in vitro assays specifically developed for potency assessment (Cottrez et al., 2016)? Currently, a complete dataset on all the molecules assessed here is only available for our selected endpoints. However, as more data accumulate this work can be continued with a comparative analysis using different models and it will be interesting to compare the PoD derived from different approaches on the same molecules. However, we currently do not expect large differences as different data integration models use similar data sources, and, more importantly, since different in vitro tests often yield partly redundant data. We had previously extended the regression models used here with the addition of data from the human cell-line activation test (h-Clat). The global model with KeratinoSens and reactivity on 128 molecules gave R2 = 0.61 (average misprediction 3.22-fold), on the same 128 chemicals the model with h-Clat and reactivity gave R2 = 0.64 (average misprediction 3.12-fold), whereas the model including all data (KeratinoSens, h-Clat and reactivity) gave R2 = 0.65 (average misprediction 3.05), hence a very marginal improvement of prediction accuracy. This is obviously due to data redundancy: The two models with either KeratinoSens/reactivity or h-Clat/reactivity correlate very well to each other R2 = 0.88 (Natsch et al., 2015). This analysis indicated that adding a dendritic cell activation assay will not significantly improve prediction, but it also showed that KeratinoSens data could be replaced by h-Clat data. CONCLUSIONS Here we show a structured approach to derive a PoD in the form of a most likely EC3 value. This approach comprises an uncertainty assessment and a combination of the PoD and the uncertainty assessments then allows deriving an NESIL for risk assessment. Based on the case studies presented, the derived NESIL would be close to an NESIL derived from human data in many instances and thus, for the data rich domain of fragrance molecules, risk assessment should be adequately possible without animal testing. Animal tests may still be needed in specific cases, for example if very new type of chemistry is explored, where the uncertainty assessment and the formation of local models as presented here may not be applicable. SUPPLEMENTARY DATA Supplementary data are available at Toxicological Sciences online. FUNDING This work was entirely performed at and supported by Givaudan Schweiz AG, no external funding was received for this study. ACKNOWLEDGMENTS We would like to thank RIFM (the Research Institute for Fragrance Materials) for providing additional in vitro and in vivo data on some chemicals. We thank Philip Kraft, Francis Voirol, and Martin Lovchik, who prepared test chemicals and Hans Gfeller, Anita Hintermeister and Daniel Bieri for analytical assistance. The authors are all employees of Givaudan but receive no other compensation but their salaries for this work and do declare no conflict of interest. REFERENCES Api A. M. , Basketter D. A. , Cadby P. A. , Cano M. F. , Ellis G. , Gerberick G. F. , Griem P. , McNamee P. M. , Ryan C. A. , Safford R. ( 2008 ). Dermal sensitization quantitative risk assessment (QRA) for fragrance ingredients . Regul. Toxicol. Pharmacol . 52 , 3 – 23 . Google Scholar Crossref Search ADS PubMed Api A. M. , Vey M. ( 2008 ). Implementation of the dermal sensitization Quantitative Risk Assessment (QRA) for fragrance ingredients . Regul. Toxicol. Pharmacol . 52 , 53 – 61 . Google Scholar Crossref Search ADS PubMed Ball N. , Cagen S. , Carrillo J. C. , Certa H. , Eigler D. , Emter R. , Faulhammer F. , Garcia C. , Graham C. , Haux C. , et al. . ( 2011 ). Evaluating the sensitization potential of surfactants: Integrating data from the local lymph node assay, guinea pig maximization test, and in vitro methods in a weight-of-evidence approach . Regul. Toxicol. Pharmacol . 60 , 389 – 400 . Google Scholar Crossref Search ADS PubMed Basketter D. , Safford B. ( 2016 ). Skin sensitization quantitative risk assessment: A review of underlying assumptions . Regul. Toxicol. Pharmacol . 74 , 105 – 116 . Google Scholar Crossref Search ADS PubMed Bil W. , Schuur A. G. , Ezendam J. , Bokkers B. G. H. ( 2017 ). Probabilistic derivation of the interspecies assessment factor for skin sensitization . Regul. Toxicol. Pharmacol . 88 , 34 – 44 . Article). Google Scholar Crossref Search ADS PubMed Bruze M. , Andersen K. E. , Goossens A. ( 2008 ). Recommendation to include fragrance mix 2 and hydroxyisohexyl 3-cyclohexene carboxaldehyde (Lyral) in the European baseline patch test series . Contact Dermatitis 58 , 129 – 133 . Google Scholar Crossref Search ADS PubMed Charpentier J. , Emter R. , Koch H. , Lelièvre D. , Pannecoucke X. , Couve-Bonnaire S. , Natsch A. , Bombrun A. ( 2018 ). Effect of fluorination on skin sensitization potential and fragrant properties of cinnamyl compounds . Chem. Biodivers. 15 , e1800013 , 10.1002/cbdv.201800013. Google Scholar Crossref Search ADS PubMed Cottrez F. , Boitel E. , Ourlin J. C. , Peiffer J. L. , Fabre I. , Henaoui I. S. , Mari B. , Vallauri A. , Paquet A. , Barbry P. , et al. . ( 2016 ). SENS-IS, a 3D reconstituted epidermis based model for quantifying chemical sensitization potency: Reproducibility and predictivity results from an inter-laboratory study . Toxicol. In Vitro 32 , 248 – 260 . Google Scholar Crossref Search ADS PubMed ECETOC ( 2003 ). Contact sensitisation: Classification according to potency. Technical Report No. 87. ECHA ( 2016a ). European Chemicals Agency: REACH registration dossier, 2-methylpentan-3-yl (2E)-but-2-enoate. Available at: https://echa.europa.eu/registration-dossier/-/registered-dossier/16609. Accessed February 1, 2018. ECHA ( 2016b ). Guidance on information requirements and Chemical Safety Assessment Chapter R.7a: Endpoint specific guidance; Draft version 5.0, February 2016. Available at: http://echa.europa.eu/documents/10162/13643/ir_csa_r7a_r7-3_msc_rac_draft_clean_en.pdf. Accessed May 30, 2016. Hirota M. , Ashikaga T. , Kouzuki H. ( 2018 ). Development of an artificial neural network model for risk assessment of skin sensitization using human cell line activation test, direct peptide reactivity assay, KeratinoSens and in silico structure alert parameter . J. Appl. Toxicol . 38 , 514 – 526 . Google Scholar Crossref Search ADS PubMed Hirota M. , Kouzuki H. , Ashikaga T. , Sono S. , Tsujita K. , Sasa H. , Aiba S. ( 2013 ). Artificial neural network analysis of data from multiple in vitro assays for prediction of skin sensitization potency of chemicals . Toxicol. In Vitro 27 , 1233 – 1246 . Google Scholar Crossref Search ADS PubMed Hoffmann S. ( 2015 ). LLNA variability: An essential ingredient for a comprehensive assessment of non-animal skin sensitization test methods and strategies . ALTEX 32 , 379 – 383 . Google Scholar PubMed Hoffmann S. , Kleinstreuer N. , Alepee N. , Allen D. , Api A. M. , Ashikaga T. , Clouet E. , Cluzel M. , Desprez B. , Gellatly N. , et al. . ( 2018 ). Non-animal methods to predict skin sensitization (I): The Cosmetics Europe database . Crit. Rev. Toxicol . 48 , 344 – 358 . Google Scholar Crossref Search ADS PubMed IDEA ( 2016 ). International Dialogue for the Evaluation of Allergens. Available at: http://www.ideaproject.info. Accessed June 22, 2016. Jaworska J. S. , Natsch A. , Ryan C. , Strickland J. , Ashikaga T. , Miyazawa M. ( 2015 ). Bayesian integrated testing strategy (ITS) for skin sensitization potency assessment: A decision support system for quantitative weight of evidence and adaptive testing strategy . Arch. Toxicol . 89 , 2355 – 2383 . Google Scholar Crossref Search ADS PubMed Kern S. , Dkhil H. , Hendarsa P. , Ellis G. , Natsch A. ( 2014 ). Detection of potentially skin sensitizing hydroperoxides of linalool in fragranced products . Anal. Bioanal. Chem . 406 , 6165 – 6178 . Google Scholar Crossref Search ADS PubMed Lalko J. , Api A. M. ( 2008 ). Citral: Identifying a threshold for induction of dermal sensitization . Regul. Toxicol. Pharmacol . 52 , 62 – 73 . Google Scholar Crossref Search ADS PubMed McNamee P. M. , Api A. M. , Basketter D. A. , Frank Gerberick G. , Gilpin D. A. , Hall B. M. , Jowsey I. , Robinson M. K. ( 2008 ). A review of critical factors in the conduct and interpretation of the human repeat insult patch test . Regul. Toxicol. Pharmacol . 52 , 24 – 34 . Google Scholar Crossref Search ADS PubMed Moss E. , Debeuckelaere C. , Berl V. , Elbayed K. , Moussallieh F. M. , Namer I. J. , Lepoittevin J. P. ( 2016 ). In situ metabolism of cinnamyl alcohol in reconstructed human epidermis: New insights into the activation of this fragrance skin sensitizer . Chem. Res. Toxicol . 29 , 1172 – 1178 . Google Scholar Crossref Search ADS PubMed Natsch A. , Emter R. ( 2017 ). Reaction chemistry to characterize the molecular initiating event in skin sensitization: A journey to be continued . Chem. Res. Toxicol . 30 , 315 – 331 . Google Scholar Crossref Search ADS PubMed Natsch A. , Emter R. , Gfeller H. , Haupt T. , Ellis G. ( 2015 ). Predicting skin sensitizer potency based on in vitro data from keratinosens and kinetic peptide binding: Global versus domain-based assessment . Toxicol. Sci . 143 , 319 – 332 . Google Scholar Crossref Search ADS PubMed Natsch A. , Gfeller H. ( 2008 ). LC-MS-based characterization of the peptide reactivity of chemicals to improve the in vitro prediction of the skin sensitization potential . Toxicol. Sci . 106 , 464 – 478 . Google Scholar Crossref Search ADS PubMed Natsch A. , Gfeller H. , Haupt T. , Brunner G. ( 2012 ). Chemical reactivity and skin sensitization potential for benzaldehydes: Can schiff base formation explain everything? Chem. Res. Toxicol . 25 , 2203 – 2215 . Google Scholar Crossref Search ADS PubMed Natsch A. , Haupt T. ( 2013 ). Utility of rat liver S9 fractions to study skin-sensitizing prohaptens in a modified KeratinoSens assay . Toxicol. Sci . 135 , 356 – 368 . Google Scholar Crossref Search ADS PubMed Natsch A. , Ryan C. A. , Foertsch L. , Emter R. , Jaworska J. , Gerberick F. , Kern P. ( 2013 ). A dataset on 145 chemicals tested in alternative assays for skin sensitization undergoing prevalidation . J. Appl. Toxicol . 33 , 1337 – 1352 . Google Scholar PubMed OECD ( 2012 ). The Adverse Outcome Pathway for Skin Sensitisation Initiated by Covalent Binding to Proteins, Part 1: Scientific Evidence. OECD Environment, Health and Safety Publications, Series on Testing and Assessment NO. 168 (ENV/JM/MONO(2012)10/PART1). OECD ( 2015a ). OECD Guideline for the Testing of Chemicals: In Chemico Skin Sensitisation: Direct Peptide Reactivity Assay (DPRA). OECD Testing Guidelines 442c. OECD ( 2015b ). OECD Guideline for the Testing of Chemicals: In Vitro Skin Sensitisation: ARE-Nrf2 Luciferase Test Method. OECD Testing Guidelines 442d. OECD ( 2016a ). Guidance Document on the Reporting of Defined Approaches to be Used Within Integrated Approaches to Testing and Assessment. OECD Guidance Document No. 255. Available at: http://www.oecd.org/officialdocuments/publicdisplaydocumentpdf/? cote=env/jm/mono(2016)28&doclanguage=en. Accessed September 8, 2017. OECD ( 2016b ). OECD Guideline for the Testing of Chemicals: In Vitro Skin Sensitisation: Human Cell Line Activation Test (h-CLAT). OECD Testing Guidelines 442e. Patlewicz G. , Kuseva C. , Mehmed A. , Popova Y. , Dimitrova G. , Ellis G. , Hunziker R. , Kern P. , Low L. , Ringeissen S. , et al. . ( 2014 ). TIMES-SS–recent refinements resulting from an industrial skin sensitisation consortium . SAR QSAR Environ. Res . 25 , 367 – 391 . Google Scholar Crossref Search ADS PubMed Schnuch A. , Uter W. , Dickel H. , Szliska C. , Schliemann S. , Eben R. , Rueff F. , Gimenez-Arnau A. , Loffler H. , Aberer W. , et al. . ( 2009 ). Quantitative patch and repeated open application testing in hydroxyisohexyl 3-cyclohexene carboxaldehyde sensitive-patients . Contact Dermatitis 61 , 152 – 162 . Google Scholar Crossref Search ADS PubMed Schnuch A. , Uter W. , Geier J. , Lessmann H. , Frosch P. J. ( 2007 ). Sensitization to 26 fragrances to be labelled according to current European regulation. Results of the IVDK and Review of the Literature . Contact Dermatitis 57 , 1 – 10 . Google Scholar Crossref Search ADS PubMed Tsujita-Inoue K. , Hirota M. , Ashikaga T. , Atobe T. , Kouzuki H. , Aiba S. ( 2014 ). Skin sensitization risk assessment model using artificial neural network analysis of data from multiple in vitro assays . Toxicol. In Vitro 28 , 626 – 639 . Google Scholar Crossref Search ADS PubMed Urbisch D. , Mehling A. , Guth K. , Ramirez T. , Honarvar N. , Kolle S. , Landsiedel R. , Jaworska J. , Kern P. S. , Gerberick F. , et al. . ( 2015 ). Assessing skin sensitization hazard in mice and men using non-animal test methods . Regul. Toxicol. Pharmacol . 71 , 337 – 351 . Google Scholar Crossref Search ADS PubMed Zang Q. , Paris M. , Lehmann D. M. , Bell S. , Kleinstreuer N. , Allen D. , Matheson J. , Jacobs A. , Casey W. , Strickland J. ( 2017 ). Prediction of skin sensitization potency using machine learning approaches . J. Appl. Toxicol . 37 , 792 – 805 . Google Scholar Crossref Search ADS PubMed Zeller K. S. , Forreryd A. , Lindberg T. , Gradin R. , Chawade A. , Lindstedt M. ( 2017 ). The GARD platform for potency assessment of skin sensitizing chemicals . ALTEX 34 , 539 – 559 . Google Scholar Crossref Search ADS PubMed © The Author(s) 2018. Published by Oxford University Press on behalf of the Society of Toxicology. All rights reserved. For permissions, please e-mail: journals.permissions@oup.com This article is published and distributed under the terms of the Oxford University Press, Standard Journals Publication Model (https://academic.oup.com/journals/pages/about_us/legal/notices)
Journal
Toxicological Sciences – Oxford University Press
Published: Sep 1, 2018