Comparative Analysis of Test Methods for Predicting Dermal Sensitization Potency in Essential Oils: A Component-Based Prediction Model, GARDskin Dose-Response Assay, and Local Lymph Node Assay

1. Introduction

The Global Harmonized System of Classification and Labeling of Chemicals (GHS) provides standards for categorizing and communicating the hazards of dermally sensitizing substances [1]. These guidelines classify dermal sensitizers as substances that induce an allergic response following skin contact. Reactions may occur after acute, repeated, and/or delayed exposure to dermal sensitizers and include allergy induction and allergy elicitation reactions. Data to classify dermal sensitizers comes from human, animal, and in vitro research. Mixtures, in the absence of data on the complete or a similar material, are classified as dermal sensitizers if they contain ≥0.1% of a Skin Sensitizer Category 1A constituent or ≥1.0% of a Skin Sensitizer Category 1B constituent. All dermally sensitizing substances are labeled as H317 Skin Sensitizers with the warning “may cause allergic skin reaction.”

When a GHS hazard is identified, a risk assessment is performed to model the likelihood of that hazard occurring in a practical setting. Dermal sensitization risk is based on the dermal sensitizing potency of a material and the end-user’s exposure [2,3]. If the material is a mixture, then compositional information about the material subcomponents and their dermal sensitization potency are often used to estimate risk [4]. In doing so, risk assessors assume that a single constituent or constituent group drives the dermal sensitization effect of a mixture based on the observation that cross-reactivity between sensitizing compounds is rare [5].

Historically, dermal sensitization risk assessments were based on results from the murine LLNA, which characterizes a material’s dermal sensitization dose-response effect and determines a point of departure (POD) known as the no expected sensitization induction level (NESIL) [6]. NESIL values are used to inform safe starting exposures for human repeat insult patch tests (HRIPT), and a low-risk exposure threshold is confirmed clinically using the confirmation of no induction in humans (CNIH) protocol [7]. While the LLNA is regarded as a refined method within the 3Rs principles to replace, refine, and reduce the use of animals in research, new approach methodologies/methods (NAMs) [8], such as the Skin Allergy Risk Assessment-Integrated Chemical Environment (SARA-ICE) and the dose-response adaptation of the Genomic Allergen Rapid Detection for skin sensitizers (GARDskin DR), use in vitro models that address technical limitations associated with the LLNA [9,10,11,12,13,14,15,16,17,18,19].

GARDskin DR is based on the GARDskin assay outlined in the Organisation for Economic Co-operation and Development Test Guideline (OECD TG) 442E [20], which is included in the OECD TG 497 [21] “2 out of 3” defined approach as a method to test for key event 3 in the identification of dermal sensitization hazards. Studies have demonstrated that GARDskin DR-derived NESIL predictions for discrete compounds are generally consistent with values derived from human and LLNA data, supporting its potential utility for evaluating the dermal sensitizing potency of natural complex substances (NCS) [12,13,22,23,24].

Many essential oils and other NCSs are classified as skin sensitizers under the GHS criteria due to the presence of dermally sensitizing constituents. While the dermally sensitizing constituents present in NCS materials may be clinically significant at high concentrations, the risk of dermal sensitization can be mitigated by dilution. Many essential oils are used in topical consumer products and are in increasing demand and availability for direct consumer use. Consequently, robust approaches for evaluating dermal sensitization risk are needed to support recommendations for safe use of essential oils and are especially important for developing new varieties and uses.

There is a small yet growing body of dermal sensitization NAM research for NCSs [12,23,25,26,27,28,29]. This study adds to that body of research by comparing dermal sensitizing potency values for eight essential oil materials generated using three separate methods: GARDskin DR, LLNA, and a CP model that estimates a material’s NESIL based on the concentration and sensitizing potency of its discrete constituents. This approach provides insight into the applicability of both GARDskin DR for evaluating the dermal sensitization properties of essential oils and explores the reliability of a simple model for predicting the dermal sensitization potency of mixtures using constituent data.

2. Materials and Methods

2.1. Test Materials

Essential oils from cedarwood (CAS 8000-27-9), cinnamon bark (CAS 8007-80-5), clove bud (CAS 8000-34-8), geranium (CAS 8000-46-2), lavender (CAS 90063-37-9), lemongrass (CAS 8007-02-1), spearmint (CAS 8008-79-5), and tea tree (CAS 68647-73-4) were provided by dōTERRA, UT, USA. The test materials were authenticated by P.S. and A.P. Descriptions of the test materials are shown in Table 1.

2.2. GC-MS Analysis

Essential oils were analyzed by GC-MS using a Shimadzu GCMS-QP2010 Ultra operated in the electron ionization mode (electron energy = 70 eV), scan range = 40–400 atomic mass units, scan rate = 3.0 scans/s, and GC-MS solution software (Shimadzu Scientific Instruments, Columbia, MD, USA). The GC column was a ZB-5 fused silica capillary column with a (5% phenyl)-polymethylsiloxane stationary phase and a film thickness of 0.25 μm, a length of 30 m, and an internal diameter of 0.25 mm (Phenomenex, Torrance, CA, USA). The carrier gas was helium with a column head pressure of 552 kPa and flow rate of 1.37 mL/min. The injector temperature was 250 °C and the ion source temperature was 200 °C. The GC oven temperature was programmed for 50 °C initial temperature, then temperature was increased at a rate of 2 °C/min to 260 °C. A 7% w/v solution of the sample was prepared in dichloromethane and 0.1 μL was injected with a splitting mode (30:1). Identification of the oil components was based on their retention indices determined by reference to a homologous series of n-alkanes, and by comparison of their mass spectral fragmentation patterns with those reported in the literature [30] and an in-house library [31].

2.3. Component-Based Prediction Model

The component-based prediction utilized in this study is a deterministic model that estimates the dermal sensitization potency of a mixture based on the concentration and potency of a single-most driving constituent. The model calculates the dermal sensitization point of departure value of a mixture (NESIL_CP) as the minimum value of the set of ratios between each constituent’s percent abundance in the mixture (b_i) and its weight of evidence (WOE) NESIL (c_i).

$$For\left\{b_1,b_2,\dots ,b_n\right\}\mathrm{a}\mathrm{n}\mathrm{d}\left\{c_1,c_2,\dots ,c_n\right\}$$

$${NESIL}_{CP}=\mathrm{m}\mathrm{i}\mathrm{n}\left\{{\displaystyle \frac{c_1}{b_1}},{\displaystyle \frac{c_2}{b_2}},\dots ,{\displaystyle \frac{c_n}{b_n}}\right\}$$

(1)

The model also yields a value for the normalized percentage of assessed constituents for dermal sensitization (nPAC_ds). This value is calculated as the sum of the relative percent abundance (b_i) of constituents with sufficient data to classify dermal sensitization hazard and conclude a WOE NESIL (i.e., c≠Ø), divided by the percent abundance of constituents in the mixture.

$${nPAC}_{ds}={\displaystyle \frac{{\sum }_{i|c\ne \varnothing }{b}_i}{\sum b_i}}$$

(2)

2.4. GARDskin DR

Testing was performed in accordance with the experimental setup of the GARDskin Dose-Response described in Gradin et al. (2021) [24], which is based on the validated GARDskin protocols as outlined OECD TG 442E [20]. The difference between the two protocols is that while GARDskin utilizes a single concentration to determine hazard, the GARDskin Dose-Response includes additional concentrations to construct a dose-response curve for prediction of continuous potency. In brief, the test system consists of a human myeloid dendritic-like cell line (SenzaCells™) where each test chemical is exposed to the cells in an appropriate vehicle. Test chemicals are prepared for cellular stimulations according to established GARDskin protocols including solubility and cytotoxicity assessment, selection of vehicle and identification of top input concentration (GARD input concentration) for the dose curve (decrease of approximately 10% in cell viability compared to unstimulated controls). Following determination of the top input concentration, five additional concentrations were generated using a serial dilution factor of 0.5. Each test chemical was evaluated in two independent experiments, yielding biological replicates for each concentration.

After 24 h of exposure, cells were harvested and total RNA was extracted and subjected to quality control. Gene expression levels of the GARD Prediction Signature, comprising a predefined set of biomarker genes previously described in the literature [32], were quantified using NanoString nCounter technology (NanoString Technologies, Seattle, WA). Raw gene expression data were processed using the GARD Data Analysis Application as described in the supporting documents to the TG of GARDskin [20]. Decision Values (DVs) were generated using a Support Vector Machine model and assigned to each test chemical and concentration.

To determine dose–response potency, the lowest concentration eliciting a positive classification response, termed the cDV₀-value, was identified, as described by Gradin et al. (2021) [24]. Briefly, DVs were plotted against stimulation concentrations to evaluate concentration–response relationships. For substances showing a concentration-dependent increase in DVs that crossed the binary classification threshold within the tested range, the cDV₀ was estimated by linear interpolation between the two concentrations bordering the threshold.

Following determination of chemical-specific cDV₀ values, these were converted to NESIL values (µg/cm²) using a linear regression model as described in Gradin et al. (2024) [13]. The model was trained on a composite potency metric developed to address uncertainties in both human and animal reference data, using an errors-in-variables approach integrating LLNA EC3 values with available human NESIL data.

2.5. LLNA Potency Measures

LLNA potency measures were collected from published literature [33,34,35,36,37] (see Table B1 in Appendix B for details). LLNA EC3 values were converted to NESIL estimates in µg/cm² by multiplying the percentage values by 250 [38].

2.6. NESIL Comparison and Statistics

Association between NESIL derivation methods was assessed using Spearman’s rank correlation coefficient, and agreement was evaluated using Bland-Altman analysis to estimate mean bias and 95% limits of agreement. Overall differences among methods were examined using Friedman’s test, with concordance quantified by Kendall’s W. Statistical analyses were conducted in Python (version 3.14.0) and verified in R (version 4.5.2) using the blandr (v0.6.0) and DescTools (v0.99.60) packages. Figures showing NESIL comparisons and Bland-Altman plots were created using Python (version 3.14.0).

3. Results

3.1. GC-MS Analysis

Constituent profiles for compounds appearing at ≥1% in the 8 assessed materials are shown in Table A1, Table A2, Table A3, Table A4, Table A5, Table A6, Table A7 and Table A8 in Appendix A. The main constituents identified in cedarwood oil are alpha-cedrene (32%), cis-thujopsene (20%), cedrol (11%), and widdrol (10%). The main constituent(s) in cinnamon oil is trans-cinnamaldehyde (64%), in clove bud oil are eugenol (73%) and eugenol acetate (15%), in geranium oil are citronellol (32%), geraniol (11%), and citronellyl formate (10%), in lavender oil are linalool (32%) and linalyl acetate (28%), in lemongrass oil are geranial (41%) and neral (31%), in spearmint oil are carvone (59%) and limonene (22%), and in tea tree oil are terpinene-4-ol (41%), gamma-terpinene (21%), and alpha-terpinene (11%).

3.2. Component-Based Prediction Model

The predicted primary exposure limiting constituent in each material determined by the CP is shown in Table 2 along with its weight of evidence NESIL, and relative percent in the raw material. The model predicts that dermal sensitization properties are driven by cedrene (alpha-cedrene and beta-cedrene isomers combined) in cedarwood oil, trans-cinnamaldehyde in cinnamon bark oil, eugenol in clove bud oil, citronellyl formate in geranium oil, linalyl acetate in lavender oil, citral (neral and geranial isomers combined) in lemongrass oil, and alpha-terpinene in tea tree oil.

The concentrations of the exposure limiting constituents in the 8 test materials range from 10-73% (median: 48.5%) and constituent Research Institute for Fragrance Materials (RIFM) WOE NESIL values range from 590-10,000 µg/cm² (median: 3050 µg/cm²). nPAC_ds values range from 26-92% (median: 75%). NESIL_CP values range from 900-64,000 µg/cm² (median: 8700 µg/cm²).

3.3. GARDskin DR

Of the 8 materials that were evaluated in the GARDskin DR assay, 7 generated a dose-curve, i.e., were identified as skin sensitizers, while 1 was identified as a non-sensitizer, i.e., the DV did not exceed the threshold at any tested concentration.

For substances with an identified cDV₀, this value was entered into a linear regression model describing the relationship between experimentally derived cDV₀ values and skin sensitizing potency, yielding continuous potency predictions in µg/cm². As shown in Table 3, each cDV₀ with corresponding predicted NESIL value can be seen. Additionally, assayed concentrations, vehicle, and determined solubility are shown. Furthermore, Figure 1 illustrates all the experimentally derived dose-curves with corresponding DV/concentration relationship for each material.

3.4. NESIL Comparisons

Side-by-side comparisons of NESIL values derived from CP, GARDskin DR, and LLNA are shown in Figure 2 and Table 4. NESIL_CP values correlate well with NESIL_GARD and NESIL_LLNA (Spearman’s correlation coefficient: 0.881 (p = 0.007) and 0.929 (p = 0.007), respectively). However, the correlation between NESIL_GARD and NESIL_LLNA is moderate and not statistically significant (Spearman’s correlation coefficient: 0.607 (p = 0.17)). Bland-Altman plots are shown in Figure 3. Log transformed bias and 95% limits of agreement values are as follows: CP vs. GARDskin DR: 0.35 (-0.22–0.92), CP vs. LLNA: 0.49 (-0.06–1.04), and GARDskin DR vs. LLNA: 0.18 (-0.75–1.12). Friedman’s test shows a statistically significant difference among the methods (χ²(2) = 7.71, p = 0.021), with post hoc tests showing a significant difference between NESIL_CP compared to NESIL_GARD and NESIL_LLNA. However, there is an overall strong statistical concordance between the methods as indicated by Kendall’s W (0.857, p = 0.017). Given the limited number of materials tested in the current study, these findings should be interpreted descriptively.

4. Discussion

Overall, the findings from this study show concordance between dermal sensitization potency values derived using the CP, GARDskin DR, and LLNA approaches. Results show strong correlation between the NESIL values from the CP compared to both GARDskin DR and LLNA. However, the NESIL_CP values found in this study are greater than their corresponding NESIL_GARD and NESIL_LLNA values for the materials tested by a factor of about 2- to 3-fold. Additionally, results of this study show a non-significant correlation between GARDskin DR and LLNA. It is important to note that these results should be interpreted descriptively, as the small number of samples tested is insufficient to support generalizable findings. Nevertheless, the results of this explorative study are informative and provide important insights into research surrounding the utility of GARDskin DR and CP for determining the dermal sensitization potency of complex mixtures.

Previous research has determined that the standard deviation of the log-transformed LLNA EC3 is between 0.15–0.31 [47,48]. In an evaluation of reproducibility, the residual standard deviation of the GARDskin DR assay was reported to be 1.8-fold [12]. While the NESIL_CP values in this study are often greater than their corresponding NESIL_GARD and NESIL_LLNA values, the absolute log difference between these values is < 0.2 for 4/8 of the materials tested. This is the case for NESIL_CP vs. NESIL_GARD in clove bud oil, lavender oil, and tea tree oil and for NESIL_CP vs. NESIL_LLNA in lemongrass oil. Additionally, the absolute log difference is < 0.2 for the NESIL_GARD vs. NESIL_LLNA in spearmint oil and geranium oil. For these cases, close agreement between dermal sensitization potency values in only 2 of the test methods provides an interesting opportunity explore the potential reasons for variance in the 3rd method and to gauge the practical significance of its difference.

The NESIL_GARD and NESIL_LLNA values for spearmint oil are nearly equivalent (2020 and 2050 µg/cm², respectively), whereas the NESIL_CP is over 2-fold higher (4400 µg/cm²). The CP determined that carvone is the exposure limiting constituent and is found at a concentration of 60% in the sample of spearmint oil tested. Spearmint oil specifically contains L-carvone (CAS 6485-40-1) for which RIFM has reported a WOE NESIL of 2600 µg/cm² based on human data [45]. Results from 3 separate LLNA studies show NESIL_LLNA values for carvone ranging from 2675–3250 µg/cm² [45,49,50]. With regard to L-carvone in GARDskin DR, Grandin et al. report a NESIL_GARD of 1560 µg/cm² [13] and Lee et al. report 2680 µg/cm² [23]. From these data we see that the dermal sensitizing potency of spearmint oil in GARDskin DR and LLNA is as sensitizing, if not slightly more, than its constituent parts alone. This is intriguing because the dermal sensitization effect of a mixture is conventionally believed to be driven by a single constituent or constituent group [5]. While the tested sample of spearmint oil does contain other constituents with sensitizing potential (limonene, 1,8-cineole, myrcene, and beta-pinene), concentrations of these constituents are below levels that would be expected to be sensitizing. Further investigation will be necessary to determine if the dermal sensitization potentiation observed in LLNA and GARDskin DR testing of spearmint oil compared to L-carvone is significant, and if so, by what mechanism it occurs.

The NESIL_CP and NESIL_LLNA values for lemongrass oil are similar (1900 and 1625 µg/cm², respectively), whereas the NESIL_GARD is somewhat lower (703 µg/cm²). The CP determined that citral is the exposure limiting constituent and is found at a concentration of 73% in the tested sample of lemongrass oil. RIFM has reported a WOE NESIL of 1400 µg/cm² for citral based on LLNA and human data [44], whereas GARDskin DR shows a NESIL of about 330 µg/cm² for citral [13,23]. The relative dermal sensitization potency of citral in GARDskin DR and LLNA closely mirrors the pattern we see for lemongrass oil in the current study. Again, this is intriguing because despite the fact that the tested lemongrass oil contains dermal sensitizing compounds other than citral (e.g., geraniol, geranyl & neryl acetate, isocitral, linalool, and 6-methyl-5-hepten-2-one), these compounds do not appear to enhance the dermal sensitization potency of the overall mixture.

The NESIL_CP and NESIL_GARD values for clove bud oil are relatively similar (8100 and 5690 µg/cm², respectively), whereas the NESIL_LLNA is over 3-fold smaller (1775 µg/cm²). The CP determined that eugenol is the exposure limiting constituent and is found at a concentration of 73% in the tested sample of clove bud oil. Indeed, eugenol is the only confirmed dermal sensitizer in clove bud oil. The RIFM WOE NESIL reported for eugenol is 5600 µg/cm² based on human data [41]. The same report cites a weighted mean NESIL_LLNA of 2827 µg/cm² [41]. NESIL_LLNA values for eugenol reported in the literature range from 1350–6275 µg/cm² [35,51]. Considering the range of potency values reported in the literature for eugenol and the fact that only one LLNA study could be identified for clove bud oil, it is reasonable to assume that repeated LLNA testing of clove bud oil may yield a range of NESIL values similar to what we see with eugenol. If this is the case, then it would be expected that the NESIL_LLNA range for clove bud oil would extend into the NESIL_CP and NESIL_GARD values determined in the current study. Together with the roughly proportional correlation we see between the published NESIL_GARD value for eugenol (2824 µg/cm²) [13,24] and the clove bud oil findings from this study when adjusting for the concentration of eugenol in clove bud oil, we are confident that the NESIL_GARD of 5690 µg/cm² reported here is a sufficiently conservative point of departure for the dermal sensitization of clove bud oil.

For lavender oil the NESIL_CP and NESIL_GARD values are relatively similar (35,300 and 23,800 µg/cm², respectively) whereas the NESIL_LLNA is 2- to 3-fold smaller (9000 µg/cm²). Linalyl acetate is determined by the CP to be the exposure limiting constituent and is found at a concentration of 28% in the tested sample of lavender oil. The RIFM WOE NESIL for linalyl acetate is 10,000 µg/cm² based on human data, with NESIL_LLNA values ranging from 400–6250 µg/cm² [43,52]. A NESIL_GARD value for linalyl acetate could not be found in the published literature. Again, based on the range of NESIL_LLNA values for linalyl acetate reported in the literature it may reasonably be expected for repeated LLNA testing of lavender oil to yield a range of values that extends into the to the NESIL_CP and NESIL_GARD values determined in this study.

The NESIL_CP and NESIL_GARD values for tea tree oil are relatively similar (20,600 and >30,000 µg/cm², respectively) and far exceed the NESIL_LLNA (3066 µg/cm²). Tea tree oil is classified as a CLP 1B skin sensitizer based on results in from a Draize HRIPT showing a 1% (3/309 subjects) sensitization induction response rate with occlusive applications of 5%, 25%, and 100% tea tree oil and four LLNA studies showing EC3 values ranging from 4.4–25.5% (NESIL: 1100–6375 µg/cm²) [37,53,54]. In a 2025 opinion from the Scientific Committee on Consumer Safety (SCCS) on tea tree oil, the lowest EC3 value was selected from these LLNA studies as a point of departure (4.4%, NESIL = 1100 µg/cm²) despite the test material in this particular study being oxidized tea tree oil and it being known that oxidized tea tree oil is more sensitizing than fresh tea tree oil [53,55]. The current study used a NESIL_LLNA value converted from the geometric mean of the EC3 values from the four available LLNA studies on tea tree oil (3066 µg/cm²), but still, this value is nearly an order-of-magnitude lower than the NESIL_GARD (>30,000 µg/cm²) and NESIL_CP (20,600 µg/cm²). Alpha-terpinene is the exposure limiting constituent determined by the CP and is present at 11% in the sample of tea tree oil tested in the current study. The single LLNA study on alpha-terpinene yielded a NESIL_LLNA of 2225 µg/cm² [50,56], which is in alignment with the RIFM WOE NESIL of 2200 µg/cm² for alpha-terpinene [46]. At this point in time alpha-terpinene has not been evaluated in GARDskin DR.

Relatively few of the constituents in the tested sample of tea tree oil have sufficient data to determine dermal sensitization potency (nPAC_ds = 26%), and perhaps a constituent other than alpha-terpinene may be driving the dermal sensitizing effect of tea tree oil. However, it is worth noting that tea tree oil has been found to be non-sensitizing in the in vitro KeratinoSens assay and several in vivo assays (guinea pig maximization test, human maximization test, and semi-occluded HRIPT) [53]. As such, based on the OECD TG 497 “2 out of 3” defined approach on skin sensitization, tea tree oil would not be classified as a dermal sensitizer [21]. Further investigations exploring the dermal sensitization of tea tree oil using other in vitro methods and evaluating the potency of alpha-terpinene, gamma-terpinene, and terpinene-4-ol in GARDskin DR may help to reconcile the differences seen in the dermal sensitization potency of tea tree oil in the LLNA vs. the GARDskin DR and CP methods.

For cedarwood oil NESIL_GARD < NESIL_LLNA < NESIL_CP (3120, 5500, and 9300 µg/cm², respectively). Cedrene (alpha & beta isomers) are determined by the CP to be the exposure limiting constituents in the tested sample of cedarwood oil and are found at a combined concentration of 38%. However, the tested sample of cedarwood oil also contains the dermal sensitizer cedrol at 11%, a level that is expected to have a significant sensitizing effect. Cedrene has a RIFM WOE NESIL of 3500 µg/cm² based on human data but has not been tested in the LLNA [39]. Cedrol has a RIFM WOE NESIL of 2000 µg/cm² based on human data and has been tested in a single LLNA showing a NESIL_LLNA of 4750 µg/cm² [57]. Additionally, neither cedrol nor cedrene presently has GARDskin DR reference data, and we are unable to determine if these constituents exhibit a combinatorial effect on the dermal sensitization potency of cedarwood oil at this time. Furthermore, cedrene and cedrol are the only two constituents in the tested sample of cedarwood oil with sufficient data to determine a dermal sensitization potency value (nPAC_ds = 50%), and it is potentially possible that another constituent may be the primary driver in dermal sensitization potency. Conducting further testing to evaluate cedrene, cedrol, and cedarwood oil constituents in the GARDskin DR assay may provide additional insights.

For geranium oil the NESIL_CP (64,000 µg/cm²) is greater than the NESIL_GARD and NESIL_LLNA (12,240 and >12,500 µg/cm², respectively). It should be noted that this NESIL_LLNA value is based on a single LLNA study where the EC3 for geranium oil was found to be >50%, the highest concentration tested [35]. Citronellyl formate is determined by the CP to be the exposure limiting constituent and is present at 10% in the tested sample of geranium oil. RIFM concluded a WOE NESIL of 6400 µg/cm² for citronellyl formate based on human and LLNA studies conducted on citronellyl formate and its read-across material citronellyl butyrate [42]. Citronellyl formate and citronellyl butyrate have each been tested in a single LLNA showing NESIL_LLNA values of 8087 µg/cm² and 6600 µg/cm², respectively [42]. At this time citronellyl formate has not been evaluated using GARDskin DR.

While we cannot completely exclude the possibility that another constituent in the tested sample of geranium oil may be driving the dermal sensitization potency, this seems unlikely considering that the dermal sensitization properties of most of the material constituents are well-described (nPAC_ds = 77%). Neither can we exclude the possibility that the presence of other dermal sensitizers in the tested sample of geranium oil (citronellol, geraniol, menthone, isomenthone, linalool, geranyl formate, geranyl butyrate, rose oxide, and citral) may potentiate its dermal sensitization effect in the GARDskin DR assay. It will be interesting to see in future testing if the potency of citronellyl formate in GARDskin DR compared to geranium oil is proportionate to the 10% concentration that this constituent is present in the test material.

The 64,000 µg/cm² NESIL_CP for the tested sample of geranium oil is very high, especially considering that the maximum NESIL thresholds for LLNA and GARDskin DR are 25,000 and 30,000 µg/cm², respectively. Indeed, the 64,000 µg/cm² NESIL_CP for geranium oil reported in this study is well beyond the testable scale, and an upper NESIL_CP threshold needs to be defined for the CP model. It should be noted that NESIL values around this magnitude are not completely unheard of, for example the WOE NESIL reported for both benzyl benzoate and cis-3-hexenyl benzoate is 59,000 µg/cm² [58,59]. Additionally, the maximum exposure to a neat material in a CNIH study is approximately 61,000 µg/cm² (0.3 g test material applied in a 25 mm Hill Top Chamber patch) [7]. With this said, the upper NESIL thresholds for preclinical and clinical testing are clearly skewed, especially when interpreted on a linear scale. The current study attempts to overcome this limitation by quantifying comparisons between NESIL values on a logarithmic scale. However, until a well-defined methodology to normalize sensitization potency outcomes between the LLNA, GARDskin DR, and CNIH is established, it must be understood that in the upper range, approximately >10,000 µg/cm², NESIL values have wider variation between methods.

It should also be noted that geranium oil is highly variable in composition. While the sample of geranium oil tested in this study aligns with an Egyptian chemotype [60], it is unknown if differences in the African chemotype tested in the reference LLNA [35] significantly impact the dermal sensitization potency. Indeed, this is an important caveat for all of the essential oil LLNA reference values cited in the current analysis as NCSs are known to vary significantly based on their source [61], and their characterization is often poorly described in pharmacological and toxicological literature [62,63].

For cinnamon bark oil, the NESIL_CP (900 µg/cm²) is greater than the NESIL_GARD (189 µg/cm²) by a factor of about 4.8-fold (Δlog = +0.68). No LLNA was available for cinnamon bark oil. Trans-cinnamaldehyde is the exposure limiting constituent determined by the CP and is present in the tested sample of cinnamon bark oil at 64%. The RIFM WOE NESIL for trans-cinnamaldehyde is 590 µg/cm² based on human data, and the weighted mean NESIL_LLNA for trans-cinnamaldehyde is 262 µg/cm² [40]. In GARDskin DR trans-cinnamaldehyde showed a NESIL of 159 µg/cm² [13,24]. Based on the relative congruence between GARDskin DR and LLNA for trans-cinnamaldehyde, it is likely that NESIL values for cinnamon bark oil from these test methods may be similar. As such, the example of cinnamon bark oil from the current study demonstrates that the conservative nature of GARDskin DR is preserved in the testing of complex mixtures. Despite this, a more permissive WOE NESIL for trans-cinnamaldehyde has been concluded based on high quality testing for dermal sensitization in humans as well as the extensive and well-documented history of safe use for this compound when diluted. It is not entirely impossible that a constituent of unknown dermal sensitization potency in cinnamon bark oil may be driving the dermal sensitization effect in GARDskin DR. However, this seems unlikely since the dermal sensitization properties of most of the constituents in the tested sample of cinnamon bark oil (nPAC_ds = 82%) have been well-characterized. Altogether, this indicates that the NESIL_CP for cinnamon bark oil may be a more suitable value for human health risk assessment than the NESIL_GARD value determined in this study.

5. Conclusions

In conclusion, we see an overall concordance in this comparative analysis between the outputs of three separate methods to measure dermal sensitization potency for a set of complex mixtures. We also detected a significant increase in the dermal sensitization potency of the complex mixtures tested compared to the potency that would be assumed from the individual constituents expected to be the primary drivers in dermal sensitization. While descriptive in nature, these findings suggest that 1) essential oils are within the applicability domain of the GARDskin DR method to predict the dermal sensitization potency of mixtures and 2) the dermal sensitizing potency of an essential oil may be greater in nonclinical testing than what is predicted from the WOE NESIL values and concentrations of its discrete components. Further GARDskin DR testing of other mixtures with well-characterized dermal sensitization effects is needed to confirm these findings. Thorough characterization of the chemical composition of each mixture, as well as the sensitizing effects of each mixture’s constituents, will be needed to evaluate the conditions under which the dermal sensitization potency of a mixture may be different from its separate components. Additionally, it may be valuable to examine the applicability of essential oils in SARA-ICE to explore the potential of obtaining an additional comparison of dermal sensitization potency. As the use of NAMs becomes increasingly accepted for defining the dermal sensitization of materials, the findings from this and future investigations may have important implications for how risk assessors weigh dermal sensitization data for mixtures and their discrete parts.