Integrated Testing Strategy for the Safety of Botanical Ingredients: A Case Study with German Chamomile Constituents

Abstract

Introduction:

German chamomile is a botanical ingredient commonly used in cosmetics, thus determination of skin sensitization effects of German chamomile constituents is critical for the safety of consumers. Nonetheless, a systematic investigation of skin sensitization potential of chamomile constituents is lacking. Non-animal methods for skin sensitization hazard evaluation have been progressively accepted as attractive alternatives to conventional animal models, especially when used in an integrated fashion.

Materials and Methods:

In the present work, 30 constituents of German chamomile were investigated for skin sensitization using in silico, in chemico, and in vitro methods, including classification-quantitative structure–activity relationship (c-QSAR) models (ADMET predictor and CASE Ultra), an expert knowledge-based system (Derek Nexus), the Direct Peptide Reactivity Assay (DPRA), the high-throughput assay with dansylated cysteamine (HTS-DCYA), the KeratinoSens^™ assay, and the human Cell Line Activation Test (h-CLAT).

Results and Discussion:

Identical classification was found for 14 compounds upon comparison between computational and experimental methods. Seven compounds (umbelliferone, apigenin, kaempferol, isorhamnetin, nerol, α-terpinene, and carvone) were positive in both models, that is, in silico and experimental settings. Seven other compounds (caffeic acid, t-ferulic acid, cosemetin, hyperoside, α-terpineol, α-bisabolol, and chamazulene) were determined to be non-sensitizers instead.

Conclusions:

Among the compounds positive in experimental settings, umbelliferone and farnesene should be regarded as a potential concern because of their positive classification and significant concentration in German chamomile.

Introduction

German chamomile (Matricaria chamomilla L., syn. Matricaria recutita or Chamomilla recutita) is one of the most popular botanical ingredients used in cosmetics and personal care products. Chamomile extracts are added to topical formulations as a skin conditioning agent, cosmetic biocide, fragrance, or flavoring agent.¹ German chamomile belongs to the Compositae family, along with several species with documented skin sensitization adverse effects, viz., feverfew (Tanacetum parthenium), tansy (Tanacetum vulgare), and arnica (Arnica montana).²

Although commonly reputed to be a gentle ingredient, allergic reactions to the whole plant and products containing chamomile have been observed without information on the probable causative agents.^3–7 Herniarin has been proposed as a possible source of skin sensitization.⁶ Rare adverse reactions to α-bisabolol-based formulations have been reported in children,^8,9 although α-bisabolol is generally considered a non-sensitizer.¹⁰ Evidence gathered from testing with non-animal methods endorsed the dermal sensitization potential of a degradation product of tonghaosu, a volatile polyacetylene that can constitute up to 30% of German chamomile essential oil.¹¹

As the use of animal-based methods, such as the murine local lymph node assay (LLNA), is being replaced with non-animal approaches, there is little to no information available about complex mixtures such as botanical ingredients and their finished products using non-animal alternative methods. Even the literature on safety assessment in animal models is quite scarce considering the predominance of botanical ingredients in cosmetics.^12–14

The exact chemical composition of German chamomile depends on several variables, including chemotypes and edaphic conditions. Both nonvolatile and volatile organic components (VOCs) can be found in cosmetic preparations, in the form of extracts or essential oils, respectively. Based on the analysis of the Mintel Global New Products Database,¹⁵ the most prevalent use of German chamomile (M. chamomilla) as a cosmetic ingredient is in the form of either flower or whole plant extracts.

Water, alcohol, mineral oil, apricot (Prunus armeniaca) kernel oil, and glycols of propylene (or butylene) extracts of chamomile are frequently used as sources of cosmetic raw ingredients.¹ The extracts obtained from German chamomile are typically enriched in flavonoid glucosides (cosemetin, rutin, quercitrin), nonglycosylated flavonoids (apigenin, isorhamnetin, kaempferol, luteolin acids, quercetin), organic acids (chlorogenic and caffeic acids), and coumarins (herniarin, umbelliferone).¹⁶

The main VOCs in German chamomile essential oils include non-oxygenated terpenes (e.g., farnesene), oxygenated sesquiterpenoids (bisabolane-type), dioxa-spiro polyacetylenes (tonghaosu and derivatives), fatty acid methyl esters, and proazulenes (matricin). Depending on the chemotype, essential oils of chamomile can contain as high as 41.45% of α-bisabolol.¹ The characteristic blue color of some German chamomile essential oils is due to chamazulene, which is an artifact formed upon degradation of matricine during the hydrodistillation process.¹⁷ Bisabolol oxides A and B, α-bisabolone oxide A, α- and β-farnesene, and fatty acid methyl esters (methyl palmitate) also contribute to the chemical fingerprinting, depending on the chemotype.^18,19

Along with major marker constituents, chamomile ingredients may contain minor amounts of ubiquitous compounds. Carvone, α-terpinene, α-terpineol, and both pinenes are fragrance ingredients known to elicit skin sensitization, either as a hapten or as potential pre-/pro-haptens.²⁰ None of these fragrance ingredients is a major constituent of German chamomile, and they typically occur in concentrations below 0.1%.¹ Nonetheless, these compounds are widespread in natural essential oils, as well as synthetic raw materials used in cosmetics. For example, (4R)-(−)-carvone is the main constituent of spearmint (Mentha spicata),²¹ whereas the 4S-(+)-stereoisomer is the main constituent of caraway (Carum carvi) and dill (Anethum graveolens) seed oils.²² Monoterpenoids α-terpinene, α-terpineol, and α-pinene are also commonly occurring in many essential oils, and are particularly abundant in tea tree oil (Melaleuca alternifolia).²³ When multiple botanical ingredients containing known sensitizers are used in combination with German chamomile, the cumulative amount of that particular compound in the final product should be taken into consideration.

The Adverse Outcome Pathway (AOP) for skin sensitization describes four main biological responses involved in the early induction stages of allergic contact dermatitis (ACD). At the present, validated non-animal experimental methods have been developed to address individual key events (KEs) that lead to skin sensitization and cannot be used as stand-alone alternatives to LLNA.²⁴ For this reason, in the present work, an in silico approach and four non-animal experimental methods were applied to evaluate the skin sensitization ability of various phytochemical classes of constituents found in M. chamomilla. The panel of chosen methods included preliminary in silico screening of chamomile constituents to build a priority list of structures for further experimental analysis.

In silico screening of German chamomile constituents was performed using two classification-quantitative structure–activity relationship (c-QSAR) models constructed under two different environments (i.e., ADMET Predictor^™ v. 9.5 and CASE Ultra 1.8.0.2) and an expert knowledge-based system (Derek Nexus 6.0.1). The data generated from these three in silico predictions were combined using a simple majority rule (SMR; positive classification if two or three predictions were positive, and vice versa for negative predictions; otherwise, inconclusive). The SMR approach is comparable to the Weight of Evidence (WoE) approach. The term SMR will be used herein to distinguish the final prediction obtained by combining multiple in silico data from the “2 out of 3” integrated approach adopted for the classification of non-animal experiments.

Further experimental research on the selected list of compounds was carried out using two in chemico methods to characterize the ability of the tested articles to trigger the first KE, meaning the haptenation process.²⁵ The chosen chemical methods were the Direct Peptide Reactivity Assay (DPRA)²⁶ and the High-Throughput Screening using dansyl cysteamine (HTS-DCYA).^27,28 The DPRA method is currently the gold standard for characterization of KE1, whereas the HTS-DCYA methods was developed in house to provide a rapid and higher throughput screening alternative. The HTS-DCYA was herein used as an alternative classification aid when DPRA quantification is not feasible (e.g., for mixtures) or inconclusive.

In vitro experiments to characterize the effect of the test articles on KE2 (elicitation of inflammatory pathways in keratinocytes) and KE3 (activation of inflammatory pathways in dendritic cells) were then performed using validated methods, namely the KeratinoSens^™ (KS) and the human Cell Line Activation Test (h-CLAT), respectively.^29,30 Experimental data gathered for the three KEs were integrated using a binary classification to establish a final skin sensitization opinion based on a “2 out of 3” integrated strategy approach.

Material and Methods

In silico predictors

Two c-QSAR models for skin sensitization based on LLNA data, constructed under different environments, were obtained from two different commercial software programs, that is, ADMET Predictor (version 9.5; Simulations-Plus, Inc., https://www.simulations-plus.com/software/admetpredictor) and CASE Ultra (version 1.8.0.2; MultiCASE, Inc., www.multicase.com/case-ultra). An expert knowledge-based system was obtained from Lhasa Limited (Derek Nexus 6.0.1; www.lhasalimited.org). All three in silico predictors were used through a CFSAN Research Collaboration Agreement (RCA).

c-QSAR model (ADMET Predictor, v. 9.5)

ADMET Predictor is a commercially available computational program that rapidly estimates several ADMET properties of chemicals using their molecular structures. Its predictive c-QSAR/r-QSAR (regression-quantitative structure–activity relationship) models are grouped into four modules: physicochemical, biopharmaceutical, metabolism, and toxicity. The toxicity module contains c-QSAR/r-QSAR models of various toxicological endpoints, including a c-QSAR model of skin sensitization based on LLNA data. This c-QSAR model was constructed using a large and diverse data set (n = 292; 66.44% positive), molecular and structural descriptors, and ANNE algorithm, while keeping aside 20% of the data as an external test set. The developer has reported the following statistics (Sensitivity/Specificity/Concordance) for this model: training set = 92.8/90.3/89.4% and test set = 82.9/93.3/85.7%.

The ANNE c-QSAR model comprised a group of 33 submodels on which predictions are based. The method for the ANNE c-QSAR model development has already been described in detail in the literature.³¹ This program uses a voting method rather than an averaging method for combining the submodel results to determine the confidence estimates in individual output classification.

The default voting threshold for output classification is the SMR: if more than half of the 33 network votes are positive (k > 16.5; where k is the number of positive votes), the compound in question is classified as positive; otherwise, it is classified as negative. However, the voting threshold may shift in either direction for the unbalanced dataset to enhance the prediction ability of the model. Therefore, the confidence in positive predictions increases by increasing the number of positive votes from 17 to 33, that is, the positive prediction with minimum confidence (if k = 17) and with maximum confidence (if k = 33). Similarly, confidence in negative predictions increases by decreasing the number of positive votes from 16 to 0, that is, the negative prediction with minimum confidence (if k = 16) and with maximum confidence (if k = 0).

To gain further in-depth knowledge, readers are referred to the specific publication on the estimation of classification uncertainty for ensemble models.³² To enhance prediction reliability, predictions with <60% confidence were not considered, that is, reported as inconclusive.

c-QSAR model (CASE Ultra, version v. 1.8.0.2)

CASE Ultra is a commercial program that incorporates several predictive c-QSAR models relevant to skin sensitization. These include models for the prediction of protein binding, antioxidant response element (ARE) activation in keratinocytes, activation of dendritic cells, LLNA outcomes, and ACD induction in humans and guinea pigs. These models are statistical in nature and were derived with an algorithm based on the MultiCASE methodology.^33–35

The MultiCASE algorithm generates numerous structural fragments from a set of molecules for an endpoint of interest and identifies fragments with statistical relevance. By doing so, the CASE Ultra model encodes both the structural fragments either related to the toxicity (positive alert) or hinder the toxicity (deactivating alerts). On fragment representations, a number next to the heavy atom denotes its hybridization (e.g., C2 is for sp2 carbon) and a lowercase heavy atom symbol (e.g., “c”) is for aromatic atoms.³⁴

In the present study, we have applied their LLNA model (v. 1.7.0.5.1690.450) that was developed using 1690 chemicals, and have 83.1% sensitivity and 78% specificity. CASE Ultra calculates a probability score, known as the probability of being positive (PP) score. For skin sensitization in mice (LLNA) predictions, if this score is <40%, then the chemical is predicted negative, and if >55%, then the chemical is predicted positive. Compounds with scores between positive and negative thresholds are considered inconclusive (“unknown”). Therefore, confidence in positive predictions increases by increasing the value of PP from 55% to 100%. Similarly, confidence in negative predictions increases by decreasing the value of PP from 40% to 0%. Some substances may be flagged as being outside the applicability domain (OAD) and are also considered unknown.

Expert system (Derek Nexus, version 6.0.1)

Derek Nexus is a commercial, expert knowledge-based system for toxicity predictions, containing rules derived from both the public and proprietary data. It contains 100 structural alerts for the assessment of skin sensitization, including in humans, and effectively predicts the standard in vivo assays (e.g., LLNA and guinea pig maximization test). This system was used to evaluate German chamomile constituents against the skin sensitization endpoint, selecting mouse as the in vivo model.

If the predicted skin sensitization outcome was defined as certain/probable/plausible/equivocal, the test chemical was considered as a sensitizer. To enhance prediction reliability, equivocal predictions were not considered, that is, reported as inconclusive. Derek Nexus also provides an EC3 potency prediction for those compounds which triggered a skin sensitization alert. If more than one skin sensitization alert was activated, the most potent EC3 value and corresponding alert were used. If no skin sensitization alert was fired, the test chemical was considered as a non-sensitizer.³⁶

Chemical selection

A comprehensive compilation of chamomile constituents (n = 283) was reduced to 246 by removing duplicates/isomeric mixtures. The modified list was subjected to an in silico screening for skin sensitization (LLNA) using two commercial c-QSAR models (ADMET predictor and CASE Ultra) and an expert knowledge-based system (Derek Nexus).

The in silico predictions were then combined using an SMR method and chemicals were divided into various pools of chemical classes. Finally, structurally diverse chemicals (both positive and negative classifications) were selected with the following considerations: in silico prediction confidence and commercial availability, with emphasis on the major constituents of German chamomile. From this selection, a reasonable number of German chamomile's constituents (n = 30) was obtained that cover various representative chemical classes, including coumarins, flavonoid aglycones and glucoside derivatives, phenolics, bisabolene- and azulene-type sesquiterpenoids, monoterpenes, as well as minor compounds with known skin sensitization effects (Fig. 1).

FIG. 1.

Structures of the 30 German chamomile constituents tested.

Chemicals and reagents

The compounds tested (Table 1) were acquired from various sources.

Table 1.

Selected German Chamomile Constituents

No.	CAS No.	Compound	Class
1	93-35-6	Umbelliferone	Coumarin
2	531-59-9	Herniarin	Coumarin
3	10597-60-1	Hydroxytyrosol	Phenolic
4	490-79-9	2,5-Dihydroxybenzoic acid	Phenolic
5	530-57-4	Syringic Acid	Phenolic
6	331-39-5	Caffeic acid	Phenolic
7	537-98-4	t-Ferulic acid	Phenolic
8	520-36-5	Apigenin	Flavonoid
9	491-70-3	Luteolin	Flavonoid
10	849061-97-8	Quercetin hydrate	Flavonoid
11	520-18-3	Kaempferol	Flavonoid
12	480-19-3	Isorhamnetin	Flavonoid
13	490-46-0	(-)-Epicatechin	Flavonoid
14	578-74-5	Cosemetin (apigenin 7-glucoside)	Flavonoid O-glucoside
15	482-36-0	Hyperoside (quercetin 3- β-D-galactoside)	Flavonoid O-glucoside
16	522-12-3	Quercitrin (quercetin 3- α-L-rhamnoside)	Flavonoid O-glucoside
17	153-18-4	Rutin	Flavonoid O-diglucoside
18	123-35-3	Myrcene	Monoterpene
19	106-25-2	Nerol	Monoterpene
20	7785-70-8	α-Pinene	Monoterpene
21	127-91-3	β-Pinene	Monoterpene
22	99-86-5	α-Terpinene	Monoterpene
23	98-55-5	α-Terpineol	Monoterpene
24	99-49-0	L-Carvone	Monoterpene
25	125037-13-0	Farnesene (isomer mixture)	Sesquiterpene
26	4602-84-0	Farnesol	Sesquiterpene
27	515-69-5	α-Bisabolol	Sesquiterpene
28	529-05-5	Chamazulene	Sesquiterpene
29	489-84-9	Guaiazulene	Sesquiterpene
30	505-32-8	Isophytol	Diterpene

Herniarin, α-bisabolol, umbelliferone, quercetin hydrate, guaiazulene, 2,5-dihydroxybenzoic acid, t-ferulic acid, nerol, isophytol, syringic acid, (-)-epicatechin were purchased from Tokyo Chemical Industry (Tokyo, Japan). Luteolin was purchased from Indofine Chemical Company, Inc. (Hillsborough, NJ). Caffeic acid, myrcene, α-terpinene, α-terpineol, α- and β-pinenes, L-carvone, farnesol, coumarin, rutin, and cinnamaldehyde were from Sigma-Aldrich. Kaempferol was obtained from Chromadex (Los Angeles, CA). Apigenin, chamazulene, hyperoside, quercitrin, isorhamnetin, farnesene (mixture of isomers), and hydroxytyrosol were from Carbosynth (Berkshire, United Kingdom). Cosemetin was obtained from isolation and characterization at the National Center for Natural Products Research, the University of Mississippi. Massoia lactone was purchased from The Good Scent Company (Oak Creek, WI).

The DPRA peptides (Ac-RFAACAA-COOH, Cat. No. RSC998, and Ac-RFAAKAA-COOH, Cat. No. RSC999) were purchased from RS Synthesis (Louisville, KY). A Zorbax SB-C-18 3.5 μm column (2.1 × 100 mm) was used and analyses were performed by high performance liquid chromatography (HPLC) using the Agilent 1290 Infinity HPLC system (Agilent Technologies, Santa Clara, CA). The dansylated cysteamine (DCYA) was synthesized from DCYA disulfide as described previously.³⁷

DPRA method

The DPRA experiments were performed as described by the Organization for Economic Cooperation and Development (OECD) Guidelines TG 442C.²⁶

Briefly, each peptide stock solution was prepared to 0.667 mM in the respective buffer (100 mM phosphate buffer pH 7.5 for Cys-peptide, and ammonium acetate pH 10.2 for Lys-peptide). Samples were solubilized in acetonitrile (ACN) at final concentrations of 20 mM (for Cys-DPRA) or 100 mM (for Lys-DPRA). The test articles were tested in triplicate by mixing 750 μL peptide with 250 μL of compound for 24 hours in amber glass vials at room temperature. The HPLC analyses and peptide quantification were performed as described in the OECD guideline.

Negative percentages of depletion were considered as a “0” depletion. The samples were analyzed and classified according to the cysteine 1:10/lysine 1:50 prediction model (threshold 6.38%) or the cysteine 1:10 prediction model when coelution with the lysine peptide was observed (threshold 13.89%). Cinnamaldehyde was used as a positive control.

HTS-DCYA method

The fluorescent readings were acquired on a SpectraMax M5 Multi-Mode Microplate Reader (Molecular Devices, Sunnyvale, CA). Data were acquired and processed using SoftMax Pro 5 (Molecular Devices) and Microsoft Excel 2013 software. The experiments were performed as previously described by Avonto et al.²⁸ Cinnamaldehyde and massoia lactone were used as moderate and strongly reactive controls, and coumarin was used as negative control. Samples with Reactivity Index (RI) >7.0 were considered positive.

KS assay

The assay was performed following OECD testing guideline 442D, using the KS cell line obtained from Givaudan SA (Vernier, Switzerland). The cells were cultured in Dulbecco's modified Eagle's medium supplemented with GLUTAMAX (Thermo Fisher Scientific), 10% fetal bovine serum (FBS; Atlanta Biologicals, Flowery Branch, GA), and G418 (500 μg/mL) at 37°C, 5% CO₂, and 95% humidity. Upon reaching a confluency of 80%–90%, the cells were seeded in 96-well white plates (10,000 cells/well) for the luciferase assay. Cells were also seeded in a clear 96-well plate for the MTT (3-[4,5-dimethylthiazol-2-yl]-2,5-diphenyltetrazolium bromide; Sigma-Aldrich) assay.

After incubating for 24 hours, the medium was replaced with 1% FBS antibiotic-free medium. Before the assay, the samples were dissolved in dimethyl sulfoxide (DMSO; 200 mM) and diluted to desired concentrations in 1% FBS antibiotic-free medium before adding to the cells. Then, the cells were treated with test samples and, at the end of the incubation period (48 hours), washed with calcium and magnesium-free Dulbecco's phosphate-buffered saline (DPBS; Thermo Fisher Scientific), and incubated with lysis buffer (Promega, Madison, WI) for 20 minutes at room temperature. Firefly luciferase reagent (Promega) was added, and the luminescence was immediately measured on a SpectraMax M5 plate reader.

Increase in luciferase activity was calculated in sample-treated cells as compared with DMSO-treated cells (negative control). To determine the cell viability under similar experimental conditions, the medium was removed from the clear 96-well plate and the cells were incubated with MTT solution (0.5 mg/mL in serum-free media) for 4 hours at 37°C. The dye was removed and DMSO was added to each well to dissolve the blue formazan produced by the cells, the color of which is read at 570 nm on a SpectraMax M5 plate reader. Samples resulting in TC50 (the concentration of the compound that reduces the cell viability to 50%) values above 2000 μM were considered nontoxic. Cinnamaldehyde was used as positive control and DMSO (1%) was the negative control. The response was calculated in terms of EC1.5, which was calculated using the following equation ^29,38:

where C_a is the concentration above 1.5-fold induction; C_b is the concentration below 1.5-fold induction; I_a is the fold induction above 1.5-fold induction; and I_b is the fold induction below 1.5-fold induction. Compounds with EC1.5 above 2000 μM were classified as non-sensitizers.

Human Cell Line Activation Test

The h-CLAT was performed in THP-1 cells (obtained from ATCC, Manassas, VA) following OECD guideline 442E. The cells were cultured in RPMI 1640 media supplemented with 10% FBS, 0.05 mM 2-mercaptoethanol, and 1% antibiotic–antimycotic (Fisher Scientific). The test chemicals were dissolved in DMSO (250 mg/mL). The final DMSO concentration in the assay medium did not exceed 0.2% v/v.

For cytotoxicity testing before activation assay, THP-1 cells were added to 24-well plates (1 × 10⁶ cells per mL per well) and exposed to various concentrations of a test chemical up to 500 μg/mL. The cells were then washed twice with FACS buffer (DPBS with 0.1% bovine serum albumin [BSA]), stained with propidium iodide (Sigma-Aldrich), and analyzed by flow cytometry, using a FACSCalibur equipped with CELLQUEST software. The concentration of the test chemical showing 75% cell viability, termed CV75, was calculated based on the analysis of viable cells.

For the cell activation assay, THP-1 cells (1 × 10⁶ cells per mL per well in 24-well plates) were incubated for 24 hours with various concentrations of the test sample, with the highest concentration being 1.2 × CV75. Following exposure, the cells were first washed with FACS buffer then resuspended and washed with blocking buffer containing 0.01% globulins Cohn fraction II/III (Sigma-Aldrich). The cells were then incubated for 30 minutes at 4°C with the following monoclonal antibodies: APC mouse IgG1 (clone: MOPC-21) from BD Pharmingen, mouse anti-human CD54, ICAM-1/FITC (clone: 6.5B5) from Dako, and FITC mouse anti-human CD86 (clone: 2331 FUN-1) from BD Pharmingen. The cells were washed and stained with Propidium Iodide (Sigma-Aldrich) and the fluorescence intensity of the viable cells was analyzed using the FACSCalibur.

The relative fluorescence intensities (RFIs) of CD86 and CD54 were calculated. If the RFI of CD86 or CD54 was >150% or 200%, respectively, at any dose in at least two experiments, the test chemical was judged as a sensitizer. DNCB (1-chloro-2,4-dinitrobenzene; Sigma-Aldrich) was included as a positive control in each experiment.

Integrated approach

The results obtained from the four experimental methods were combined using a “2 out of 3” approach, where two positive results were considered sufficient for classification as “sensitizer” (1) versus “non-sensitizers ” (0). For KE1, only one data was considered for classification purposes, being DPRA and HTS-DCYA applied to the same KE. In case of classification disagreement, DPRA results overruled HTS-DCYA results. HTS-DCYA results were used in place of DPRA when nonconclusive results were obtained using DPRA (e.g., if quantification was not achievable due to peaks overlapping).

Results

In silico screening of German chamomile constituents and selection of test chemicals

As a first step, a comprehensive compilation of chamomile constituents was gathered from different sources, including SciFinder, Dictionary of Natural Products database, CIR reviews as well as in-house data.^18,39–41 A list of 283 compounds was compiled and SMILES strings for each component were generated, with stereochemical information when available.

The comprehensive list was reduced to 246 by removing duplicates/isomeric mixtures. This restricted list was then subjected to an in silico screening for skin sensitization (LLNA) using two commercial c-QSAR models (ADMET predictor and CASE Ultra) and an expert knowledge-based system (Derek Nexus). These in silico extrapolations are associated with prediction confidence (quantitative or qualitative) and OAD flag (if any). Prediction results are as follows: ADMET Predictor (Positive = 120, Negative = 94, Inconclusive = 14, OAD = 18), CASE Ultra (Positive = 100, Negative = 95, Inconclusive = 47, OAD = 4), and Derek Nexus (Positive = 61, Negative = 121, Inconclusive = 63, OAD = 1). These in silico predictions were then combined using an SMR that resulted in the following classification results: Positive = 82, Negative = 91, Inconclusive = 73.

Based on SMR and structural consideration, the above selection of German chamomile constituents (n = 246) was further narrowed. Emphasis was given to structurally diverse chemicals, in silico predictions (both positive and negative classifications), and sample availability. Priority was given to nonvolatile constituents since polar extracts of German chamomile are most commonly used in cosmetics and personal care products.¹ Finally, 30 constituents of German chamomile were selected as test articles (Table 1) to cover all representative chemical classes, including coumarins, flavonoid aglycones, and glucoside derivatives, phenolic compounds, bisabolane-type sesquiterpenoids, terpenes, azulenes, as well as minor compounds with known skin sensitization effects.⁴²

In silico classification of selected test chemicals

The in silico LLNA predictions for the selected 30 German chamomile constituents (Fig. 1) are summarized in Table 2, along with the SMR results, and experimental LLNA data (if available).

Table 2.

In Silico Skin Sensitization Local Lymph Node Assay Prediction of 30 German Chamomile Constituents

S. No.	Compound	Experimental skin sensitization		ADMET Predictor^a		Case Ultra^b		Derek Nexus^c		SMR^d
S. No.	Compound	LLNA	References	Prediction	Confidence (%)	Prediction	PP (%)	Prediction	Reasoning	SMR^d
1	Umbelliferone			0	97	1	57.8	1	Plausible	Positive
2	Herniarin			Inconclusive	Inconclusive	0	30.5	1	Plausible	Inconclusive
3	Hydroxytyrosol	1	Kern et al.⁷⁰	1	64	1	86.4	1	Plausible	Positive
4	2,5-Dihydroxybenzoic acid			0	97	Inconclusive	Inconclusive	1	Plausible	Inconclusive
5	Syringic acid			0	97	0	11.9	0	NMUF	Negative
6	Caffeic acid			0	75	1	80.3	0	NMUF	Negative
7	t-Ferulic acid			0	89	1	73.2	0	NMUF	Negative
8	Apigenin			0	97	1	72.5	1	Plausible	Positive
9	Luteolin			0	97	1	87.8	1	Plausible	Positive
10	Quercetin			0	82	1	93.1	1	Plausible	Positive
11	Kaempferol			0	97	1	83.2	1	Plausible	Positive
12	Isorhamnetin			0	93	1	90	1	Plausible	Positive
13	(-)-Epicatechin			Inconclusive	Inconclusive	Inconclusive	Inconclusive	1	Plausible	Inconclusive
14	Cosemetin			0	97	0	2.1	Inconclusive	Equivocal	Negative
15	Hyperoside			0	97	0	32.2	1	Plausible	Negative
16	Quercitrin			0	97	1	64.4	1	Plausible	Positive
17	Rutin			OAD	OAD	1	64.4	1	Plausible	Positive
18	Myrcene			1	97	Inconclusive	Inconclusive	Inconclusive	Equivocal	Inconclusive
19	Nerol	1	Ryan et al.⁷²	1	79	1	56.6	1	Certain	Positive
20	α-Pinene			1	93	1	66.6	Inconclusive	Equivocal	Positive
21	β-Pinene			1	88	1	77.6	0	NMUF	Positive
22	α-Terpinene	1	Kern et al.⁷⁰	1	95	0	23.9	1	Certain	Positive
23	α-Terpineol			0	78	0	24.3	Inconclusive	Equivocal	Negative
24	L-Carvone	1	Kern et al.⁷⁰	1	99	1	87.1	1	Plausible	Positive
25a	α-Farnesene			OAD	OAD	Inconclusive	Inconclusive	Inconclusive	Equivocal	Inconclusive
25b	β-Farnesene			1	99	Inconclusive	Inconclusive	Inconclusive	Equivocal	Inconclusive
25	Farnesene^e			Inconclusive	Inconclusive	Inconclusive	Inconclusive	Inconclusive	Equivocal	Inconclusive
26	Farnesol	1	Api et al.⁷¹	1	83	1	56.6	1	Certain	Positive
27	α-Bisabolol			0	64	0	38.1	Inconclusive	Equivocal	Negative
28	Chamazulene			1	93	0	35.6	0	CUF	Negative
29	Guaiazulene			1	90	0	35.6	0	CUF	Negative
30	Isophytol			0	82	1	61.9	0	NMUF	Negative
Positive/negative/inconclusive/OAD				11/16/3/2 (10/16/3/1)^f		17/9/6/0 (17/9/4/0)^f		17/7/8/0 (17/7/6/0)^f		15/10/7/0 (15/10/5/0)^f

Certain = there is proof that the proposition is true. CUF = features in the molecule are found in non-alerting sensitizers in the Lhasa skin sensitization negative prediction dataset. The prediction remains negative and the misclassified features are highlighted to enable the negative prediction to be verified by expert assessment. Equivocal = there is an equal weight of evidence for and against the proposition. NMUF = the query structure does not match any structural alerts or examples for skin sensitization in Derek. Additionally, the query structure does not contain any unclassified or misclassified features. Plausible = the weight of evidence supports the proposition.

Predictions from commercial model of ADMET Predictor^™ v9.5.

Predictions from commercial model of Case Ultra 1.8.0.2.

Predictions from Derek Nexus: 6.0.1. (an expert knowledge-based system).

Combined in silico predictions using an SMR (SMR is alike to the “2 out of 3” approach); 1 and 0 represent positive and negative, respectively.

The in silico prediction of “Farnesene” is based on its components α-Farnesene and β-Farnesene. Note: Table 2 contains in silico results of 32 chemicals that include two additional components (α-Farnesene and β-Farnesene) of Farnesene (an isomer mixture).

Data of α-Farnesene and β-Farnesene were excluded.

CUF, contains unclassified features; LLNA, local lymph node assay; NMUF, no misclassified or unclassified feature; OAD, outside the applicability domain; PP, probability of being positive; SMR, simple majority rule.

Thirty-three percent (10 out of 30) compounds were predicted positive from the ADMET predictor, while 56% (17 out of 30) compounds were predicted positive from both CASE Ultra and Derek Nexus. Farnesene is an isomeric mixture of α- and β-farnesene, therefore in silico predictions were performed on both forms. α-Farnesene was classified as OAD, although β-farnesene was predicted positive by the ADMET predictor. In contrast, both α- and β-farnesene were assigned as inconclusive by CASE Ultra and Derek Nexus. Thus, farnesene results were considered inconclusive in in silico predictions. When combining these three in silico results using an SMR approach, 50% (15 out of 30) of the compounds were classified as positive.

Syringic acid was predicted negative by all three in silico models, and therefore classified as negative, while four compounds (carvone, farnesol, hydroxytyrosol, and nerol) were predicted positive by all three, and therefore considered positive. Four other compounds (cosemetin, α-bisabolol, α-terpineol, and hyperoside) were predicted negative by two in silico predictors, and therefore classified as negative based on SMR. Although these four compounds were predicted negative by both c-QSAR models, the Derek Nexus prediction was positive for hyperoside and inconclusive for cosemetin, α-bisabolol, and α-terpineol based on their skin sensitization reasoning “equivocal.” The positive prediction by Derek Nexus for hyperoside, with “Plausible” reasoning, is based on the two alerts: (1) 1,2-dihydroxybenzene or derivative and (2) resorcinol or precursor; and predicted LLNA EC3 = 0.33% (strong sensitizer).

Three compounds (caffeic acid, t-ferulic acid, and isophytol) were predicted negative by ADMET predictor and Derek Nexus, but positive by CASE Ultra. Therefore, these three compounds were classified as negative based on SMR. Positive predictions of these three compounds by CASE Ultra are supported by their positive/deactivating alerts: four positive alerts [fragments: c-OH, c:c(:cH)-OH, cH:cH:c(-OH):c(:cH)-OH, and C2H-C2(-OH) = O] and one deactivating alert [fragment: c:cH:cH:c(:cH)-C2H] were identified with calculated probability (PP = 80.3%) for caffeic acid. Four positive alerts [fragments: c-OH, c:c(:cH)-OH, C3H3-O-c(:cH:c):c(:cH)-OH and C2H-C2(-OH) = O] and two deactivating alerts [fragment: c:cH:cH:c(:cH)-C2H and C3H3-O-c(:cH):c] were identified with calculated probability (PP = 73.2%) for t-ferulic acid, while three positive alerts [fragments: C3H3-C3H(-C3H3)-C3H2, C3H2-C3H2-C3H(-C3H3)-C3H2-C3H2-C3H2-C3H(-C3H3)-C3H2-C3H2, and C2H = C2H2] and one deactivating alert (fragment: C3-OH) were identified with calculated probability (PP = 61.9%) for isophytol.

α-Terpinene was predicted positive by ADMET predictor and Derek Nexus, but negative by CASE Ultra, and therefore classified as positive by the SMR. The negative prediction for α-terpinene by CASE Ultra was associated with the absence of an identifiable positive alert in the test chemical. One deactivating alert [fragment: C3H3-C3H(-C3H3)-C2(-C3H2) = C2H] was found instead. β- and α-pinenes were positive based on SMR because these compounds were positive in both c-QSAR models. Instead, Derek Nexus predicted a negative outcome for β-pinene (“NMUF” [no misclassified or unclassified feature] reasoning) and inconclusive for α-pinene (“equivocal” reasoning).

Although seven compounds (umbelliferone, apigenin, luteolin, quercetin, kaempferol, isorhamnetin, and quercitrin) were classified as non-sensitizers by ADMET predictor with high confidence (97%–82%), they were predicted positive by the remaining two predictors: CASE Ultra and Derek Nexus (“plausible” reasoning). Therefore, negative predictions by ADMET predictor were overridden in the SMR, and these seven compounds were classified as positive. Similarly, chamazulene and guaiazulene were predicted positive by ADMET predictor but negative by the remaining two predictors and were classified as negative. Rutin was predicted positive by CASE Ultra and Derek Nexus (“plausible” reasoning) but OAD in ADMET predictor, therefore it was classified as positive in the SMR prediction.

Although herniarin was considered inconclusive by the ADMET predictor based on the low prediction confidence (<60%), contradicting predictions were obtained by the remaining two predictors. Derek Nexus predicted herniarin as positive with “plausible” reasoning because it matched three alerts (resorcinol or precursor, α,β-unsaturated ester or precursor, and vinylic or allylic anisole), and predicted LLNA EC3 = 0.87% (strong sensitizer). Therefore, contrasting classification led to an inconclusive result for herniarin. Similarly, 2,5-dihydroxybenzoic acid was predicted negative by ADMET predictor, inconclusive by CASE Ultra, and positive by the Derek Nexus (“plausible” reasoning because it matched one alert for “hydroquinone or derivative”), and thus it was assigned as inconclusive.

Epicatechin was inconclusive in both c-QSAR models, while predicted positive in Derek Nexus, therefore, overall considered as inconclusive. The Derek Nexus-positive prediction for epicatechin (with “plausible” reasoning) was related to the presence of one alert (i.e., 1,2-dihydroxybenzene or derivative), therefore, the predicted LLNA EC3 was 0.19% (strong sensitizer). Similarly, myrcene was predicted positive by ADMET predictor but inconclusive CASE Ultra and Derek Nexus, and therefore classified as inconclusive.

Characterization of KE1

Direct Peptide Reactivity Assay

Results of Cys- and Lys-peptide depletion obtained in the DPRA method are summarized in Table 3. Fifteen compounds were positive in the DPRA. Out of those resulting positive for peptide depletion, the majority (13/15) was more reactive toward the Cys-heptapeptide compared with the Lys-heptapeptide. 2,5-Dihydroxybenzoic acid was reactive with either peptide alike. Umbelliferone was the only compound with stronger reactivity observed toward Lys-peptide (24.78%) than Cys-peptide (6.94%). Among the most reactive compounds, syringic acid, α-terpinene, and farnesene were reactive against both cysteine and lysine peptides, with quantitative depletion of Cys-peptide.

Table 3.

Activation of KE1 Results: Direct Peptide Reactivity Assay and High-Throughput Screening Using Dansyl Cysteamine Classification of German Chamomile Constituents

No.	Compound	DPRA				DCYA
No.	Compound	% Del (Cys)	% Del (Lys)	Average	Classification	RI	Classification
1	Umbelliferone	6.94 ± 1.7	24.78 ± 2.8	15.87	Positive	20.4 ± 10.7	Positive
2	Herniarin	4.45 ± 0.9	1.12 ± 0.8	2.78	Negative	2.8 ± 3.7	Negative
3	Hydroxytyrosol	n.q.	n.q.	n.q.	n.q.	3.3 ± 1.3	Negative
4	2,5-Dihydroxybenzoic acid	19.75 ± 6.4	22.40 ± 0.7	21.07	Positive	12.0 ± 3.8	Positive
5	Syringic acid	91.38 ± 5.8	14.52 ± 0.2	52.95	Positive	−1.6 ± 5.8	Negative
6	Caffeic acid	36.68 ± 5.1	n.q.	36.68^a	Positive	6.4 ± 2.8	Negative
7	t-Ferulic acid	26.53 ± 6.4	n.q.	26.53^a	Positive	0.1 ± 2.7	Negative
8	Apigenin	14.82 ± 1.8	1.67 ± 0.1	8.25	Positive	−6.3 ± 2.9	Negative
9	Luteolin	99.24 ± 0.3	n.q.	99.24^a	Positive	14.9 ± 1.1	Positive
10	Quercetin	98.41 ± 1.1	n.q.	98.41^a	Positive	61.3 ± 1.9	Positive
11	Kaempferol	52.42 ± 7.7	n.q.	52.42^a	Positive	1.4 ± 0.8	Negative
12	Isorhamnetin	n.q.	n.q.	n.q.	n.q.	41.1 ± 2.4	Positive
13	(-)-Epicatechin	91.38 ± 7.0	n.q.	91.38^a	Positive	3.9 ± 4.6	Negative
14	Cosemetin	n.q.	n.q.	n.q.	n.q.	−11.7 ± 2.1	Negative
15	Hyperoside	n.q.	n.q.	n.q.	n.q.	10.8 ± 2.7	Positive
16	Quercitrin	37.15 ± 5.6	n.q.	37.15^a	Positive	10.0 ± 2.6	Positive
17	Rutin	n.q.	n.q.	n.q.	n.q.	3.5 ± 4.9	Negative
18	Myrcene	1.71 ± 2.5	0.31 ± 0.1	1.01	Negative	1.6 ± 3.1	Negative
19	Nerol	2.32 ± 1.4	−0.23 ± 1.4	1.05	Negative	−1.0 ± 3.6	Negative
20	α-Pinene	8.06 ± 1.6	0.74 ± 0.6	4.4	Negative	1.0 ± 1.5	Negative
21	β-Pinene	8.05 ± 1.3	0.04 ± 0.5	4.05	Negative	2.0 ± 1.1	Negative
22	α-Terpinene	99.89 ± 0.1	9.48 ± 1.3	54.69	Positive	0.6 ± 2.6	Negative
23	α-Terpineol	10.61 ± 0.5	−0.21 ± 0.03	5.20	Negative	0.5 ± 6.3	Negative
24	L- Carvone	23.04 ± 0.3	0.85 ± 0.5	11.95	Positive	12.8 ± 1.3	Positive
25	Farnesene	99.82 ± 0.03	25.61 ± 2.3	62.72	Positive	8.9 ± 2.5	Positive
26	Farnesol	4.30 ± 2.3	−0.16 ± 0.3	2.07	Negative	1.5 ± 4.0	Negative
27	α-Bisabolol	10.06 ± 2.0	−0.17 ± 0.6	4.94	Negative	4.1 ± 1.6	Negative
28	Chamazulene	58.14 ± 2.0	1.27 ± 1.5	29.71	Positive	1.1 ± 2.7	Negative
29	Guaiazulene	5.68 ± 1.6	0.73 ± 1.1	3.2	Negative	2.5 ± 4.0	Negative
30	Isophytol	6.48 ± 0.6	−1.02 ± 0.8	2.73	Negative	1.4 ± 4.7	Negative
Pos. Ctrl.	Cinnamaldehyde	78.21 ± 0.7	67.15 ± 2.0			15.3 ± 3.5
Neg. Ctrl.	Coumarin					−0.8 ± 4.7
Total positives					15		9

Cys-model applied; cinnamaldehyde was used as positive control.

DCYA, dansylated cysteamine; DPRA, Direct Peptide Reactivity Assay; n.q., not quantifiable; RI, Reactivity Index.

The flavonoids luteolin, quercetin, and epicatechin were strongly reactive against the Cys-peptide, but Lys-depletion was not quantifiable. At the end of the 24-hour incubation, the chromatograms contained numerous peaks and a very noisy baseline, presumably related to the degradation of compounds/peptides in the assay conditions. Similarly, the Lys-depletion of caffeic acid, ferulic acid, kaempferol, and quercitrin were not quantifiable, while reaction against Cys-peptide was moderate, and hence only the Cys-peptide model was applied for reactivity classification.

The DPRA method also failed to quantify the reactivity of hydroxytyrosol, isorhamnetin, cosemetin, hyperoside, and rutin due to poor baseline resolution and generation of several peaks in either aqueous buffer solutions, hence no meaningful quantification was achieved.

Of the compounds resulting in higher Cys-depletion, 10 out of 13 were either aromatic acids or flavonoids. Nonoxygenated compounds, farnesene, chamazulene, and α-terpinene resulted in Cys-depletion even in the absence of a structural alert. α-Terpinene is known to act as a prehapten through chemical oxidation,^43,44 while reactivity of chamazulene is unclear considering the low solubility in the aqueous buffers and the contrasting classification of the structural analog, guaiazulene. Farnesene is known to rapidly oxidize with the generation of hydroperoxide derivatives.⁴⁵

High-throughput screening using dansyl cysteamine

In contrast to the DPRA method, where the remaining unreacted peptide is quantified, the HTS-DCYA method enables direct quantification of the thiol adducts as fluorescent derivatives. The shorter reaction times reduce the risk of degradation/oxidation of the thiol functional group, and hence non-specific reactions. In this study, fluorescent quantification of chamomile constituents not quantifiable with DPRA was achieved by fluorescence emission instead (Table 3). Nine compounds were identified as reactive with DCYA. Caffeic, ferulic, and syringic acids did not contribute to the formation of any DCYA adducts. Similarly, apigenin, chamazulene, epicatechin, kaempferol, and α-terpinene were also not reactive.

Quercetin was the most reactive compound, followed by isorhamnetin and umbelliferone.

Further HPLC investigation using a diode-array-detector combined with a mass spectrometer (HPLC-DAD-MS) was performed on reactive flavonoids (quercetin, isorhamnetin, hyperoside) and the non-reactive kaempferol to confirm whether the fluorescence response was related to the presence of DCYA adducts. None of the reactive samples resulted in the presence of chromatographic peaks with the expected m/z for DCYA-adducts. An increased concentration from the peak corresponding to DCYA dimer was observed in all samples instead, except for kaempferol. The abundance of DCYA dimer estimated by HPLC-DAD-MS correlated with the degree of fluorescence response observed in the HTS screening. Challenges in the reliable assessment of flavonoids were also encountered in the course of DPRA experiments, and it is a well-known challenge for in vitro screening.^46,47

Characterization of KE2 and 3

The KE2 of the skin sensitization AOP is the activation of proinflammatory pathways in keratinocytes. The effect of German chamomile constituents on KE2 was examined through the KS assay,²⁹ which measures the activation of the Nrf2-ARE pathway. Fifty-three percent of the compounds (16 out of 30) resulted positive for KE2 elicitation in this assay. Most of the flavonoids and aromatic nonvolatile compounds tested in KS assay did not elicit activation of the Nrf2-ARE pathway in vitro, including the strong, chemically reactive quercetin. Among the main VOCs, α-bisabolol and chamazulene were found negative, while farnesene was positive with EC1.5 of 1 μM. Coumarins (herniarin and umbelliferone), apigenin, and its 7-O-glucoside derivative were also found to activate proinflammatory pathways, with EC1.5 values of 14.17, 28.65, 1.79, and 44.23 μM, respectively (Table 4).

Table 4.

Activation of KEs 2 and 3: KeratinoSens and Human Cell Line Activation Test Results for German Chamomile Constituents

		KeratinoSens			h-CLAT
No.	Compound	EC 1.5 (μM)	TC50 (μM)	Classification	CV75 (μg/mL)	CD54/CD86	Classification
1	Umbelliferone	28.65 ± 2.14	>2000	Positive	500	Positive/negative	Positive
2	Herniarin	14.17 ± 2.18	800 ± 200	Positive	250	Negative/negative	Negative
3	Hydroxytyrosol	51.74 ± 5.2	405 ± 5	Positive	15.625	Negative/negative	Negative
4	2,5-Dihydroxybenzoic acid	>2000	>2000	Negative	1000	Negative/negative	Negative
5	Syringic acid	987.76 ± 12.12	>2000	Positive	1000	Negative/negative	Negative
6	Caffeic acid	>2000	168 ± 37.5	Negative	1000	Negative/negative	Negative
7	t-Ferulic acid	>2000	>2000	Negative	1000	Negative/negative	Negative
8	Apigenin	1.79 ± 1.16	30 ± 10	Positive	7.8	Negative/negative	Negative
9	Luteolin	>2000	47 ± 8	Negative	7.8	Negative/negative	Negative
10	Quercetin	>2000	85 ± 15	Negative	7.8	Negative/negative	Negative
11	Kaempferol	4.72 ± 1.57	60 ± 0	Positive	7.8	Negative/negative	Negative
12	Isorhamnetin	1.62 ± 0.31	83 ± 7.5	Positive	250	Negative/positive	Positive
13	(-)-Epicatechin	8.76 ± 0.87	1050 ± 0	Positive	1000	Positive/negative	Positive
14	Cosemetin	44.23 ± 16.21	>2000	Positive	500	Negative/negative	Negative
15	Hyperoside	>2000	>2000	Negative	500	Negative/negative	Negative
16	Quercitrin	>2000	1050 ± 0	Negative	7.8	Negative/negative	Negative
17	Rutin	>2000	17 ± 1	Negative	250	Positive/negative	Positive
18	Myrcene	>2000	180 ± 10	Negative	1000	Negative/negative	Negative
19	Nerol	19.55 ± 0.58	800 ± 0	Positive	125	Positive/negative	Positive
20	α-Pinene	>2000	385 ± 5	Negative	500	Positive/negative	Positive
21	β-Pinene	>2000	365 ± 15	Negative	500	Negative/negative	Negative
22	α-Terpinene	11.66 ± 0	155 ± 35	Positive	1000	Positive/positive	Positive
23	α-Terpineol	>2000	>2000	Negative	250	Positive/negative	Positive
24	L-Carvone	72.86 ± 13.23	980 ± 80	Positive	125	Positive/negative	Positive
25	Farnesene	1 ± 0	34 ± 8.5	Positive	7.8	Negative/negative	Negative
26	Farnesol	4.25 ± 0.97	41 ± 1	Positive	31.25	Negative/negative	Negative
27	α-Bisabolol	>2000	105 ± 15.5	Negative	31.25	Positive/negative	Positive
28	Chamazulene	>2000	46 ± 3	Negative	3.9	Negative/negative	Negative
29	Guaiazulene	7.45 ± 1.24	10 ± 1	Positive	15.6	Positive/negative	Positive
30	Isophytol	4.42 ± 0.8	10 ± 0	Positive	1000	Positive/negative	Positive
Total positives				16			12

h-CLAT, human Cell Line Activation Test.

The KE3, which is considered essential in the skin sensitization inflammatory pathway, is the expression of cell surface markers (CD54 and CD86) in dendritic cells.³⁰ The h-CLAT method is based on in vitro evaluation of the expression of CD54/CD86 in THP-1 cells, a human monocytic leukemia cell line, using a flow cytometric method.

The testing results of German chamomile compounds for activation of KE3 (Table 4) are in overall accordance with the activation of KE2. Twelve compounds were found positive for at least one CD marker, with flavonoids less reactive than VOCs. For other compounds, the lack of proinflammatory effects in vitro was not unexpected. Chamomile has been shown to possess anti-inflammatory and antiallergic properties and both volatile and nonvolatile constituents are known to contribute to such effects.⁴⁸ It is used as a traditional remedy for mild skin irritation and inflammation.⁴⁹ Chamomile exposure to macrophages has been reported to protect against oxidative stress and exhibit cytoprotective properties through the Nrf2-ARE pathway.⁵⁰

Binary classification of German chamomile constituents using in chemico/in vitro results

The results of each individual test are summarized in Table 5 as binary classification, along with the integrated experimental results. The integrated results were obtained using a “2 out of 3” approach based on the positive outcomes of KEs 1, 2, and 3. For KE1, DPRA results overrode DCYA results in case of opposite classification. When DPRA results were inconclusive, DCYA results were used for the classification of KE1.

Table 5.

Classification of German Chamomile Constituents Based on Experimental “2 out of 3” Integrated Approach and In Silico Results (Simple Majority Rule)

S. No.	Ingredient name	Experimental in chemico data^a		Experimental in vitro data			In silico (SMR)
S. No.	Ingredient name	DCYA	DPRA	KS	h-CLAT	2 Out of 3
1	Umbelliferone	1	1	1	1	Positive	Positive
2	Herniarin	0	0	1	0	Negative	Inconclusive
3	Hydroxytyrosol	0	n.q.	1	0	Negative	Positive
4	2,5-(OH)₂-benzoic acid	1	1	0	0	Negative	Inconclusive
5	Syringic acid	0	1	1	0	Positive	Negative
6	Caffeic acid	0	1	0	0	Negative	Negative
7	t-Ferulic acid	0	1	0	0	Negative	Negative
8	Apigenin	0	1	1	0	Positive	Positive
9	Luteolin	1	1	0	0	Negative	Positive
10	Quercetin	1	1	0	0	Negative	Positive
11	Kaempferol	0	1	1	0	Positive	Positive
12	Isorhamnetin	1	n.q.	1	1	Positive	Positive
13	(-)-Epicatechin	0	1	1	1	Positive	Inconclusive
14	Cosemetin	0	n.q.	1	0	Negative	Negative
15	Hyperoside	1	n.q.	0	0	Negative	Negative
16	Quercitrin	1	1	0	0	Negative	Positive
17	Rutin	0	n.q.	0	1	Negative	Positive
18	Myrcene	0	0	0	0	Negative	Inconclusive
19	Nerol	0	0	1	1	Positive	Positive
20	α-Pinene	0	0	0	1	Negative	Positive
21	β-Pinene	0	0	0	0	Negative	Positive
22	α-Terpinene	0	1	1	1	Positive	Positive
23	α-Terpineol	0	0	0	1	Negative	Negative
24	L-Carvone	1	1	1	1	Positive	Positive
25	Farnesene	1	1	1	0	Positive	Inconclusive
26	Farnesol	0	0	1	0	Negative	Positive
27	α-Bisabolol	0	0	0	1	Negative	Negative
28	Chamazulene	0	1	0	0	Negative	Negative
29	Guaiazulene	0	0	1	1	Positive	Negative
30	Isophytol	0	0	1	1	Positive	Negative
Positive/negative/inconclusive		9/21/0	15/10/5	16/14/0	12/18/0	12/18/0	15/10/5

Integrated results based on one result for each KE. If DPRA quantification was not possible, DCYA results were used for KE1.

KE, key event; KS, KeratinoSens^™.

All OECD methods resulted in about 40%–53% of compounds classified as positive when taken alone, although, individual compounds resulted in opposite classification across all methods, with few exceptions. When using a “2 out of 3” approach, 12 compounds were classified as positive. Carvone and umbelliferone were the only two compounds scoring four out of four positive tests across all the KEs. Epicatechin, isorhamnetin, and α-terpinene were positive in at least one in chemico assay and both in vitro methods. Four compounds (2,5-dihydroxybenzoic acid, quercetin, luteolin, and quercitrin) were reactive in both in chemico methods, but did not activate further KEs in vitro, and were therefore considered negative.

Among the VOCs, marker compounds α-bisabolol, and chamazulene were considered non-sensitizers based on integrated data, whereas farnesene was positive in all in chemico and in vitro assays, except for h-CLAT.

Discussion

In silico classification (Table 2) suggests that the majority of aromatic compounds have opposite classification from two c-QSAR models (ADMET predictor and CASE Ultra). It is difficult to explain the exact reason behind these results because both the c-QSAR models were constructed using different training sets, algorithms, and environments. However, it may be explained in terms of chemical class and common positive alerts.

Eight out of 10 compounds (predicted positive by CASE Ultra and negative by ADMET) can be separated into two chemical classes: aromatic acids (e.g., caffeic acid, and t-ferulic acid) and flavonoids (e.g., apigenin, luteolin, quercetin, kaempferol, isorhamnetin, and quercitrin). Three positive alerts [c-OH, c:c(:cH)-OH, and C2H-C2(-OH) = O] were responsible for the positive classification of aromatic acids (caffeic acid and t-ferulic acid) by CASE Ultra. Similarly, flavonoids (apigenin, luteolin, quercetin, kaempferol, isorhamnetin, and quercitrin) were classified as positive by CASE Ultra based on four positive alerts [c-OH, c:c(:cH)-OH, cH:c(:cH)-OH, and C2-O-c].

It is worth noting that this class of compounds was also difficult to characterize chemically, with opposite outcomes in the two chemical methods used for KE1. Such incongruences are typically seen for unstable compounds and pro-oxidants, which tend to result in positive DPRA, but less reactive in the DCYA method, due to differences in the reaction conditions and incubation timing. Evidence of skin sensitization from flavonoids does exist, although they are quite scarce considering their occurrence.⁵¹ Also, they are typically regarded as beneficial compounds, antiallergic, anti-inflammatory, and antipruritus.^52–55

Flavonoids such as epicatechin, quercitrin, apigenin, isorhamnetin, and kaempferol (log p-values are 0.51, −0.64, 2.92, 2.38, and 3.11, respectively) varies in polarity and dermal absorption properties.^56–58 Flavonoid glycosides are typically too polar to be able to cross the stratum corneum, although the formulation and concentration may have a significant impact on in situ availability, with related consequences for skin sensitization.^59,60 The limited evidence on the side effects of flavonoids could thus be related to the relatively scarce concentration in commercial formulations and low dermal availability. Nonetheless, with the booming of green and natural ingredients, new beneficial activities associated with this class of biomolecules,^61–67 and advances in topical formulations,^68,69 we cannot exclude a rise in reported side effects as a consequence of increased consumer exposure.

Based on the literature survey, carvone, farnesol, hydroxytyrosol, nerol, and α-terpinene were found to be skin sensitizers in LLNA.^70–72 Of these five compounds, four compounds (carvone, farnesol, hydroxytyrosol, and nerol) were predicted positive by all the three in silico predictors and therefore classified as positive.

On the contrary, α-terpinene was predicted positive in ADMET predictor (95% confidence) and Derek Nexus but negative in CASE Ultra. Derek Nexus predicted α-terpinene as positive with reasoning “Certain” because it was present in their database, matched alert (“conjugate diene”), and LLNA EC3 = 8.9% (moderate sensitizer). In contrast, CASE Ultra predicted α-terpinene as negative because no positive alert was identified in the structure, rather one deactivating alert [fragment: C3H3-C3H(-C3H3)-C2(-C3H2) = C2H] was found. This fragment decreases the PP by 11.7% if no other deactivating features are present. The calculated probability (PP = 23.9%) is lower than the model's current classification threshold of 40% for negative prediction. Since α-terpinene was predicted positive by two in silico predictors, it was classified as positive in the SMR prediction, confirming the experimental LLNA reports.

It is worth noting that the current in silico models selected are apt to predict LLNA outcome. The possibility of false-positive/negative LLNA predictions was minimized using two LLNA c-QSAR models (constructed under two different environments) and an expert knowledge-based system in an SMR. However, reliable LLNA predictions may still provide false-positive/negative results relative to human data for some specific group of chemicals where low concordance between LLNA and human data exists.

Based on in silico results, 30 chemicals were further tested experimentally, using non-animal methods to assess the activation of 3 main KEs. In silico SMR results of 14 compounds agree with the experimental results, providing high confidence in the skin sensitization predictions of both computational and experimental approaches. Seven compounds (umbelliferone, apigenin, kaempferol, isorhamnetin, nerol, α-terpinene, and carvone) were flagged as skin sensitizers, while the remaining seven compounds (caffeic acid, t-ferulic acid, cosemetin, hyperoside, α-terpineol, α-bisabolol, and chamazulene) were considered non-sensitizers.

Four flavonoids (luteolin, quercetin, quercitrin, and rutin) were overall positive in in silico predictions but negative using an integrated approach. In practice, in the absence of in vivo data or human evidence, these four flavonoids are likely non-sensitizers. The calculated log p-values were 2.427, 1.958, 0.035, and −0.870, respectively (S+log P; ADMET Predictor, v. 9.5), suggesting a diverse range of polarity and dermal absorption properties. Thus, formulation- and concentration-related skin sensitization consequences may not be ruled out.

Other three compounds (syringic acid, guaiazulene, and isophytol) were categorized as negative by in silico predictions but positive using the “2 out of 3” strategy. Three flavonoids (apigenin, isorhamnetin, and kaempferol) were positive in in silico predictions and also found capable of triggering multiple KEs in experimental settings; however, their dermal penetration data are largely missing, and epidemiological evidence gives little support to the hypothesis that flavonoids may be responsible for skin sensitization adverse events when considering their widespread occurrence across plant species. The classification of nerol, α-terpinene, and carvone as skin sensitizers are further supported by their experimental LLNA data.^70,72 In contrast, α-terpineol is an established contact allergen in humans, thus, it may be regarded as a false negative in LLNA.²⁰

In silico results of hydroxytyrosol and farnesol in SMR differ from the overall experimental classification. However, in silico results may override the integrated data based on the experimental LLNA reports, and thus they are considered as skin sensitizers.^70,71 Farnesol is also listed among 26 fragrance allergens of primary concern.²⁰ Similarly, α- and β-pinene were predicted positives in in silico predictions but negatives in chemical and cell-based experiments. Both pinenes are known skin sensitizers, although the lack of reactive domains suggests that abiotic or biotic activation may be required, which may explain the observed negative experimental outcome.

Several coumarins, especially psoralenes, are mostly regarded as phototoxic or photosensitizers.⁷³ Conflicting reports regarding the skin sensitization potential have been described in the literature. Both umbelliferone and herniarin were considered non-sensitizing in guinea pig and open epicutaneous tests.⁷⁴ Further investigations identified umbelliferone as a weak sensitizer,⁷⁵ while herniarin was identified as a candidate allergen in patch test studies.⁶ In the present study, herniarin was found to only activate KE2, while umbelliferone consistently activated all KEs studied. It is worth remembering that the photosensitization mechanism differs from skin sensitization, and photosensitizers are typically out of the domain of applicability of skin sensitization methods. Herniarin is also considered a strong photosensitizing and phototoxic compound, although to a lesser degree compared with psoralen-type coumarins.⁷³

Myrcene was inconclusive in in silico and did not activate any experimental KE tested. Therefore, it may be considered a nonsensitizer. Myrcene is also a fragrance ingredient that has not been established as a contact allergen in humans.²⁰ Herniarin was inconclusive in in silico predictions, and only activated one KE in vitro, but it is known to elicit skin sensitization in humans.⁶ Also, 2,5-dihydroxybenzoic acid was inconclusive in in silico but classified as negative based on the “2 out of 3” strategy, thus likely a non-sensitizer to skin. Epicatechin and farnesene were inconclusive in in silico predictions but the positive experimental data suggest they are likely skin sensitizers.

Conclusions

Due to the widespread occurrence of German chamomile as an ingredient in cosmetic formulations, in silico predictions were performed on its constituents to evaluate their potential skin sensitization ability. Based on in silico results and occurrence in German chamomile, 30 constituents were selected as test chemicals for further confirmatory studies using a combination of 4 in chemico and in vitro methods. Among the 30 chemicals tested, 50% (15 out of 30) were classified as positive in computational studies using a combination of two c-QSAR models and an expert knowledge-based system.

Based on integrated results for the activation of the 3 main KEs, 12 out of 30 chamomile compounds were classified as positive. Out of the 12 positive compounds, 4 (carvone, α-terpinene, guaiazulene, and nerol) are minor constituents of German chamomile essential oils or extract, and their final concentration in the product should be considered when other ingredients containing the same chemicals are formulated with German chamomile.

When comparing in silico and experimental results, 47% agreement (14 out of 30) was found between in silico (SMR) and WoE results. Seven of the 14 compounds (umbelliferone, apigenin, kaempferol, isorhamnetin, nerol, α-terpinene, and carvone) were positive for skin sensitization by either computational and experimental evidence, while the remaining 7 compounds (caffeic acid, t-ferulic acid, cosemetin, hyperoside, α-terpineol, α-bisabolol, and chamazulene) were classified as non-sensitizers. α-Terpineol is an established contact allergen in humans thus it may classify as a false negative in LLNA.²⁰ Agreement between in silico (SMR) and integrated experimental results may provide higher confidence in evaluating the skin sensitization potential of chemicals. However, it may not always be true as relative to human data, for example, α-terpineol was classified as non-sensitizers in both in silico (SMR) and experimental (WoE), but it is an established contact allergen in humans.

Among the main German chamomile constituents, farnesene and umbelliferone are the main constituents of concern.^73–75 Herniarin and umbelliferone can be found in German chamomile extracts in concentration up to 915 and 290 ppm, respectively.¹ α- and β-farnesenes are major constituents of German chamomile essential oils, with concentrations reaching up to 30% and 50%, respectively, depending on the chemotype. Due to their abundance in chamomile extracts, more studies are needed to characterize their potential bioavailability upon topical application and dermal absorption, to perform quantitative risk assessments.

Footnotes

Acknowledgment

The authors would like to dedicate this article to the memory of Dr. Jon Parcher for his life-long contributions.

Authors' Contributions

C.A. and J.A. performed the chemical assays; O.D. performed in vitro testing; R.P.V. performed in silico studies; N.S., S.I.K., A.G.C., and I.A.K. contributed to data reviewing; C.A., R.P.V., and Z.W. wrote the article.

Disclaimer

The content of this publication does not necessarily reflect the views or policies of the U.S. Food and Drug Administration (US FDA), Department of Health and Human Services, nor does mention of commercial products, or organizations imply endorsement by the U.S. Government. This article reflects the current thinking and experience of the authors.

Author Disclosure Statement

No competing financial interests exist.

Funding Information

Financial support was provided by the “Science-Based Authentication of Botanical Ingredients” funded by the Center for Food Safety and Applied Nutrition, US FDA grant number 5U01FD004246.

References

Johnson

, Boyer

, Bergfeld

, et al. Amended safety assessment of Chamomilla recutita-derived ingredients as used in cosmetics. Int J Toxicol, 2018:37(3 Suppl):51S–79S.

Paulsen

, Andersen

, Hausen

. Sensitization and cross-reaction patterns in Danish Composite-allergic patients. Contact Dermatitis, 2001:45; 197–204.

Ketel

. Allergy to Matricaria chamomilla. Contact Dermatitis, 1987:16; 50–51.

Lundh

, Hindsen

, Gruvberger

, et al. Contact allergy to herbal teas derived from Asteraceae plants. Contact Dermatitis, 2006:54; 196–201.

Brandão

MGL

, Freire

, Vianna-Soares

. Surveillance of phytotherapeutic drugs in the State of Minas Gerais. Quality assessment of commercial samples of chamomile. Cad Saúde Pública, 1998:14; 613–616.

Paulsen

, Otkjaer

, Andersen

. The coumarin herniarin as a sensitizer in German chamomile [Chamomilla recutita (L.) Rauschert, Compositae]. Contact Dermatitis, 2010:62; 338–342.

Paulsen

. Contact sensitization from Compositae-containing herbal remedies and cosmetics. Contact Dermatitis, 2002:47; 189–198.

Jacob

, Hsu

. Reactions to Aquaphor^®: is bisabolol the culprit?. Pediatr Dermatol, 2010:27; 103–104.

Jacob

, Matiz

, Herro

. Compositae-associated allergic contact dermatitis from bisabolol. Dermatitis, 2011:22; 102–105.

10.

Andersen

. Final report on the safety assessment of bisabolol. Int J Toxicol, 1999:18(3 Suppl);33–40.

11.

Avonto

, Rua

, Lasonkar

, et al. Identification of a compound isolated from German chamomile (Matricaria chamomilla) with dermal sensitization potential. Toxicol Appl Pharmacol, 2017:318; 16–22.

12.

, Yan

G-Y

, Zhao

Y-L

, et al. Investigation of the dermal sensitizing potential of traditional medical extracts in local lymph node assays. Exp Biol Med, 2009:234; 306–313.

13.

Menné Bonefeld

, Nielsen

, Gimenéz-Arnau

, et al. An immune response study of oakmoss absolute and its constituents atranol and chloroatranol. Contact Dermatitis, 2014:70; 282–290.

14.

Basista-Sołtys

, Filipek

, Zadrożna

, et al. Sensitization properties of propolis and balsam of peru in Guinea Pig Maximization Test (GPMT). Food Nutr Sci, 2017:8; 714–724.

15.

Mintel Global New Products Database. https://www.mintel.com/global-new-products-database. Accessed May 10, 2019.

16.

Raal

, Orav

, Püssa

, et al. Content of essential oil, terpenoids and polyphenols in commercial chamomile (Chamomilla recutita L. Rauschert) teas from different countries. Food Chem, 2012:131; 632–638.

17.

Schilcher

, Imming

, Goeters

. Active chemical constituents of Matricaria chamomilla. L. syn. Chamomilla recutita (L.) Rauschert. In: Chamomile: Industrial Profiles., Franke R and Schilcher H (eds), pp. 56–76. New York: CRC Press; 2005.

18.

Wang

, Avula

, Wang

Y-H

, et al. An integrated approach utilising chemometrics and GC/MS for classification of chamomile flowers, essential oils and commercial products. Food Chem, 2014:152; 391–398.

19.

Rubiolo

, Belliardo

, Cordero

, et al. Headspace-solid-phase microextraction fast GC in combination with principal component analysis as a tool to classify different chemotypes of chamomile flower-heads (Matricaria recutita l.). Phytochem Anal, 2006:17; 217–225.

20.

SCCS (Scientific Committee on Consumer Safety). Opinion on fragrance allergens in cosmetic products, 26–27 June 2012.

21.

Akhila

, Banthorpe

, Rowan

. Biosynthesis of carvone in Mentha spicata. Phytochemistry, 1980:19; 1433–1437.

22.

Ravid

, Bassat

, Putievsky

, et al. Isolation and determination of optically pure carvone enantiomers from caraway (Carum carvi L.), dill (Anethum graveolens L.), spearmint (Mentha spicata L.) and Mentha longifolia (L.) Huds. Flavour Fragr J, 1987:2; 95–97.

23.

Brophy

, Davies

, Southwell

, et al. Gas chromatographic quality control for oil of Melaleuca terpinen-4-ol type (Australian tea tree). J Agric Food Chem, 1989:37; 1330–1335.

24.

European Chemicals Agency (ECHA). Guidance on information requirements and chemical safety assessment. In: Chapter

.7a: Endpoint Specific

Guidance

. Version 6.0

July

. 2017. ECHA (ed). Helsinki; 2017. Available at: https://echa.europa.eu/documents/10162/13632/information_requirements_r7a_en.pdf

25.

Vinken

. The adverse outcome pathway concept: a pragmatic tool in toxicology. Toxicology, 2013:312; 158–165.

26.

OECD (2021), Test

. 442C: In Chemico skin sensitisation: Assays addressing the Adverse Outcome Pathway key event on covalent binding to proteins, OECD Guidelines for the Testing of Chemicals, Section 4, OECD Publishing, Paris, https://doi.org/10.1787/9789264229709-en.

27.

US patent WO2015200870A22015. Methods for detecting and categorizing skin sensitizers.

28.

Avonto

, Chittiboyina

, Rua

, et al. A fluorescence high throughput screening method for the detection of reactive electrophiles as potential skin sensitizers. Toxicol Appl Pharmacol, 2015:289; 177–184.

29.

OECD (2018), Test

. 442D: In vitro skin sensitisation: ARE-Nrf2 luciferase test method, OECD Guidelines for the Testing of Chemicals, Section 4, OECD Publishing, Paris, https://doi.org/10.1787/9789264229822-en.

30.

OECD (2018), Test

. 442E: In vitro skin sensitisation: In vitro skin sensitisation assays addressing the Key Event on activation of dendritic cells on the Adverse Outcome Pathway for skin sensitisation, OECD Guidelines for the Testing of Chemicals, Section 4, OECD Publishing, Paris, https://doi.org/10.1787/9789264264359-en.

31.

Verma

, Matthews

. Estimation of the chemical-induced eye injury using a weight-of-evidence (WoE) battery of 21 artificial neural network (ANN) c-QSAR models (QSAR-21): part I: irritation potential. Regul Toxicol Pharmacol, 2015:71; 318–330.

32.

Clark

, Liang

, Lee

, et al. Using beta binomials to estimate classification uncertainty for ensemble models. J Cheminform, 2014:6; 34.

33.

Klopman

. MULTICASE 1. A hierarchical computer automated structure evaluation program. QSAR, 1992:11; 176–184.

34.

Chakravarti

, Saiakhov

, Klopman

. Optimizing predictive performance of CASE Ultra expert system models using the applicability domains of individual toxicity alerts. J Chem Inf Model, 2012:52; 2609–2618.

35.

Graham

, Gealy

, Macina

, et al. QSAR for allergic contact dermatitis. QSAR, 1996:15; 224–229.

36.

Canipa

, Chilton

, Hemingway

, et al. A quantitative in silico model for predicting skin sensitization using a nearest neighbours approach within expert-derived structure–activity alert spaces. J Appl Toxicol, 2017:37; 985–995.

37.

Chittiboyina

, Avonto

, Rua

, et al. Alternative testing methods for skin Sensitization: NMR spectroscopy for probing the reactivity and classification of potential skin sensitizers. Chem Res Toxicol, 2015:28; 1704–1714.

38.

EURL-ECVAM (2014). Recommendation on the KeratinoSens™ assay for skin sensitisation testing. Available at: http://ihcp.jrc.ec.europa.eu/our_labs/eurlecvam/eurl-ecvam-recommendations/recommendation-keratinosens-skin-sensitisatio

39.

Jones

, Avula

, Wang

Y-H

, et al. Investigating sub-2 μm particle stationary phase supercritical fluid chromatography coupled to mass spectrometry for chemical profiling of chamomile extracts. Anal Chim Acta, 2014:847; 61–72.

40.

Avula

, Wang

Y-H

, Wang

, et al. Quantitative determination of phenolic compounds by UHPLC-UV-MS and use of partial least-square discriminant analysis to differentiate chemo-types of Chamomile/Chrysanthemum flower heads. J Pharm Biomed Anal, 2014:88; 278–288.

41.

Avonto

, Wang

, Chittiboyina

, et al. Hydroxylated bisabolol oxides: evidence for secondary oxidative metabolism in Matricaria chamomilla. J Nat Prod, 2013:76; 1848–1853.

42.

Heisterberg

, Menné

, Johansen

. Contact allergy to the 26 specific fragrance ingredients to be declared on cosmetic products in accordance with the EU cosmetics directive. Contact Dermatitis, 2011:65; 266–275.

43.

Rudbaeck

, Bergstroem

, Boerje

, et al. α-Terpinene, an antioxidant in tea tree oil, autoxidizes rapidly to skin allergens on air exposure. Chem Res Toxicol, 2012:25; 713–721.

44.

Avonto

, Chittiboyina

, Wang

, et al. In chemico evaluation of tea tree essential oils as skin sensitizers: impact of the chemical composition on aging and generation of reactive species. Chem Res Toxicol, 2016:29; 1108–1117.

45.

Anet

. Autoxidation of α-farnesene. Austr J Chem, 1969:22; 2403–2410.

46.

Pohjala

, Tammela

. Aggregating behavior of phenolic compounds—a source of false bioassay results?. Molecules, 2012:17; 10774–10790.

47.

Papadopoulou

, Green

, Frazier

. Interaction of flavonoids with bovine serum albumin: a fluorescence quenching study. J Agric Food Chem, 2005:53; 158–163.

48.

, Xu

, Yao

. Anti-inflammatory and anti-allergic effects of German chamomile (Matricaria chamomilla L.). J Essent Oil-Bear Plants, 2011:14; 549–558.

49.

Gardiner

Chamomile (Matricaria recutita, Anthemis nobilis): Clinician Information Summary. Boston, MA: The Longwood Herbal Task Force, The Center for Holistic Pediatric Education and Research; 2000. Available at: http://www.longwoodherbal.org/chamomile/chamomile.cis.pdf

50.

Bhaskaran

, Shukla

, Kanwal

, et al. Induction of heme oxygenase-1 by chamomile protects murine macrophages against oxidative stress. Life Sci, 2012:90; 1027–1033.

51.

Hausen

, Schmalle

, Marshall

, et al. 5,8-Dihydroxyflavone (primetin) the contact sensitizer of Primula mistassinica Michaux. Arch Dermatol Res, 1983:275; 365.

52.

Choi

, Kim

S-H

. Rutin suppresses atopic dermatitis and allergic contact dermatitis. Exp Biol Med, 2013:238; 410.

53.

Jeon

, Kim

, Kang

, et al. Anti-inflammatory and antipruritic effects of luteolin from Perilla (P. frutescens L.) leaves. Molecules, 2014:19; 6941.

54.

Kim

, Lim

, Kang

, et al. Aceriphyllum rossii extract and its active compounds, quercetin and kaempferol inhibit ige-mediated mast cell activation and passive cutaneous anaphylaxis. J Agric Food Chem, 2014:62; 3750–3758.

55.

Weng

, Zhang

, Asadi

, et al. Quercetin is more effective than cromolyn in blocking human mast cell cytokine release and inhibits contact dermatitis and photosensitivity in humans. PLoS One, 2012:7; e33805.

56.

Rothwell

, Day

, Morgan

. Experimental determination of octanol-water partition coefficients of quercetin and related flavonoids. J Agric Food Chem, 2005:53; 4355–4360.

57.

Chen

J-L

. Comparision of succinate- and phthalate-functionalized etched silica hydride phases for open-tubular capillary electrochromatography. J Chromatogr A, 2009:1216; 6236–6244.

58.

Xie

, Luo

, Duan

, et al. Phytic acid enhances the oral absorption of isorhamnetin, quercetin, and kaempferol in total flavones of Hippophae rhamnoides L. Fitoterapia, 2014:93; 216–225.

59.

, Birt

. In vivo and in vitro percutaneous absorption of cancer preventive flavonoid apigenin in different vehicles in mouse skin. Pharm Res, 1996:13; 1710–1715.

60.

Alonso

, Ramon

, Lozano

, et al. Percutaneous absorption of flavan-3-ol conjugates from plant procyanidins. Drugs Exp Clin Res, 2004:30; 1–10.

61.

Shivasaraun

, Sureshkumar

, Karthika

, et al. Flavonoids as adjuvant in psoralen based photochemotherapy in the management of vitiligo/leucoderma. Med Hypotheses, 2018:121; 26–30.

62.

Arct

, Pytkowska

. Flavonoids as components of biologically active cosmeceuticals. Clin Dermatol, 2008:26; 347–357.

63.

Gendrisch

, Esser

, Schempp

, et al. Luteolin as a modulator of skin aging and inflammation. BioFactors, 2021:47; 170–180.

64.

Jichkar

, Sahasrabuddhe

, Kankhar

. Quercetin: a natural flavonoid used in cosmetics products. World J Pharm Pharm Sci, 2020:9; 1123–1134.

65.

Saewan

, Jimtaisong

. Photoprotection of natural flavonoids. J Appl Pharm Sci, 2013:3; 129.

66.

Svobodova

, Psotova

, Walterova

. Natural phenolics in the prevention of UV-induced skin damage. a review. Biomed Pap, 2003:147; 137–145.

67.

Zhu

, Gao

. The use of botanical extracts as topical skin-lightening agents for the improvement of skin pigmentation disorders. J Invest Dermatol Symp Proc, 2008:13; 20–24.

68.

Nagula

, Wairkar

. Recent advances in topical delivery of flavonoids: a review. J Controlled Release, 2019:296; 190–201.

69.

Sharma

, Dogra

, Thakur

, et al. Phytosomes, a novel strategy to improve the bioavailability of flavonoids: a review. Eur J Biomed Pharm Sci, 2019:6; 209–218.

70.

Kern

, Gerberick

, Ryan

, et al. Local lymph node data for the evaluation of skin sensitization alternatives: a second compilation. Dermatitis, 2010:21; 8–32.

71.

Api

, Basketter

, Cadby

, et al. Dermal sensitization quantitative risk assessment (QRA) for fragrance ingredients. Regul Toxicol Pharmacol, 2008:52; 3–23.

72.

Ryan

, Chaney

, Kern

, et al. The reduced local lymph node assay: the impact of group size. J Appl Toxicol, 2008:28; 518–523.

73.

Kaidbey

, Kligman

. Photosensitization by coumarin derivatives: structure-activity relationships. Arch Dermatol, 1981:117; 258–263.

74.

Hausen

, Schmieder

. The sensitizing capacity of coumarins (I). Contact Dermatitis, 1986:15; 157–163.

75.

Masamoto

. Sensitization and crossreaction of simple coumarins. Yakugaku Zasshi, 2001:121; 97–103.