Current knowledge of xenobiotic metabolism in the skin, associated in vitro and in silico models and their application to Next Generation Risk Assessments

Highlights

• •Update on skin xenobiotic metabolizing enzymes and in vitro skin metabolism models.
• •Use of skin metabolism data in NGRAs from the perspective of the cosmetic industry.
• •Increasing efforts to integrate skin metabolism data into PBK models.
• •Challenges and questions remain – standardization, guidelines, in silico models.

Abstract

In Next Generation Risk Assessments (NGRAs) of cosmetic ingredients, evaluations of systemic exposure after topical application range from conservative screening estimates to higher tier kinetic models. The latter need robust and predictive in silico and in vitro skin absorption and metabolism models to ensure regulatory confidence. This review provides an update on skin xenobiotic metabolizing enzymes (XMEs) and in vitro skin metabolism models of varying complexities. It is intended primarily for safety assessors and study planners who need to select and interpret skin-metabolism NGRA approaches for cosmetic ingredients.

Qualitative information helps to inform subsequent tiers of the NGRA workflow by identifying potential reactive intermediates, while structural and SAR information helps assess bioactivation and detoxification pathways. Quantitative data should be of good quality, correctly interpreted and extrapolated to in vivo using appropriate equations and scaling factors to provide a relevant indication of first-pass skin metabolism. This can be influenced by access to XMEs, extrinsic and intrinsic factors. Challenges and open questions remain regarding standardized metabolism assays and related guidelines, as well as the development of in silico models to predict skin metabolites and skin bioavailability for integration into Physiologically Based Kinetic (PBK) models. Despite advances in technologies, efforts are still needed to improve our understanding of skin XMEs and their location, and to develop suitable models to predict and measure metabolic pathways. There are increasing efforts to integrate skin metabolism data into PBK models for dermal exposure. Further research into skin metabolism is needed for NGRAs of topically applied cosmetic ingredients.

1. Introduction

Cosmetic ingredients registered in the EU can no longer be evaluated for safety using traditional safety assessments employing in vivo data. As a result, ingredients are evaluated by cosmetics companies in so-called Next Generation Risk Assessments (NGRAs), based solely on information gained from in silico and in vitro assays i.e. New Approach Methodologies (NAMs) (Pallocca et al., 2022). Following the 2013 EU ban on animal testing for cosmetics, substantial efforts have been made to develop non-animal approaches for cosmetic ingredient safety assessment. NGRA, as described by Berggren et al., 2017, provides a possible framework for integrating fit-for-purpose NAMs, as also reflected in the Scientific Committee on Consumer Safety (SCCS) Notes of Guidance (NoG) (SCCS, 2023). Although NGRA continues to be strengthened through testing, validation, and practical application, it has not yet been formally adopted by an EU risk-assessment body as an alternative framework for cosmetic ingredient safety assessment. Since topical application represents the primary route of consumer exposure to cosmetics, characterizing dermal bioavailability is essential for human safety assessment. However, dermal bioavailability is not merely a function of physical permeation; it can be actively modulated by first-pass skin metabolism. As part of the International Collaboration on Cosmetics Safety (ICCS), a review of the literature was conducted to lay the essential groundwork for research projects dedicated to quantifying skin metabolism and integrating these experimental data into physiologically based kinetic (PBK) models and in vitro-to-in vivo extrapolations (IVIVE). Crucially, this review highlights how skin metabolism data can be used across different assessment tiers, from qualitative metabolite identification (bioactivation vs. detoxification) to quantitative biophysical parameters needed to refine internal exposure estimates.

Despite early assumptions that skin functions only as a metabolically inert barrier to invading organisms and chemicals, it is now known that the skin is metabolically active and can change the kinetic profile of a chemical due to so-called “first-pass dermal metabolism” (Goebel et al., 2009). For example, an early study showed that cortisone was metabolized to hydrocortisone, mainly in the dermis (Malkinson et al., 1959). Hydrocortisone was also shown to be metabolized to different metabolites, although they were not identified at the time due to analytical limitations. Since then, reconstructed human epidermis (RHE) models developed in the 1980s have increased our knowledge of skin metabolism (Roguet, 2002). It is now known that the skin exhibits a range of metabolic pathways, from Phase 1 functionalization to Phase 2 conjugation (Hewitt et al., 2013; Kazem et al., 2019; Oesch et al., 2014; Van Eijl et al., 2012). As a result of the development of RHEs and new sensitive analytical methods, skin metabolism is now recognized as an important aspect of drug design (Pyo and Maibach, 2019), drug efficacy and the safety of topically applied products (Madden et al., 2017). The impact of skin metabolism was demonstrated by Mavon and Jacques-Jamin (2013), who measured the metabolism of three chemicals with different LogP values in native skin. As the logP increased, the overall dermal delivery (i.e. the total amount of parent and metabolites) was higher in fresh metabolically active skin compared to frozen metabolically inactive skin. The difference was attributed to the contribution of metabolites to overall delivery, which increased with LogP. Metabolism converts lipophilic chemicals into more hydrophilic metabolites, which are more water-soluble, allowing them to diffuse across the epidermis and dermis. Thus, first pass skin metabolism can increase the overall systemic exposure to a chemical and its metabolites, but it can also reduce the exposure to the parent compound (Mavon and Jacques-Jamin, 2013).

Current safety assessments of cosmetic ingredients rely solely on skin absorption and metabolism data from NAMs to help conclude on the toxicity profile of the parent and/or metabolites and, where necessary, to select follow-up higher tier assays (Ebmeyer et al., 2024). There are several important aspects to consider in an NGRA, including whether the metabolic pathway(s) at the local and systemic level leads to more toxic metabolites than the parent chemical (i.e. bioactivation) or less toxic metabolites (i.e. detoxification). Both types of pathways are present in the skin (Hewitt et al., 2013). For example, pro-hapten skin sensitizers require oxidative metabolism to short lived, reactive electrophilic intermediate(s) in the skin to cause sensitization (Gerberick et al., 2009). By contrast, pathways like Phase 2 N-acetylation (which are known to metabolize oxidative aromatic amine hair dyes (Zeller and Pfuhler, 2014)) can detoxify chemical precursors directly in the skin. Therefore, qualitative information from in silico and in vitro models can help to identify potential reactive intermediates, as well as bioactivated and/or detoxified metabolites, which drives the subsequent higher tier activities in the NGRA workflow. For systemic toxicity, first-pass skin metabolism determines whether the intact parent compound or its stable metabolites reach systemic circulation (Ebmeyer et al., 2024).

An assessment of skin metabolism not only depends on whether a chemical is a substrate of the enzymes present in the skin, it is also dependent on the fraction of chemical reaching the viable skin layers containing enzymatic activities and remaining there long enough to be metabolized (i.e., the mean residence time). Thus, to derive a better-informed estimation of the local and systemic bioavailability of a chemical, the rate of metabolism in the skin, the localization of the associated xenobiotic metabolizing enzymes (XMEs) in the different skin layers, as well as the rate of permeation of the chemical should be considered (Boderke et al., 1997, 2000; Quantin et al., 2022). If quantitative data are generated, it needs to be correctly interpreted and, if needed, extrapolated to in vivo using appropriate in vitro-to-in vivo extrapolations (IVIVEs) using scaling factors to provide a relevant and robust indication of first-pass dermal metabolism as a chemical diffuses across the skin layers.

Several case studies have been published showcasing NGRAs according to the workflow defined by the SEURAT-1 (Safety Evaluation Ultimately Replacing Animal Testing) project (Berggren et al., 2017). In all case studies describing tiered workflows for the internal Threshold of Toxicological Concern (iTTC) approach (Najjar et al., 2023), read-across assessments (Grégoire et al., 2024; Lester et al., 2023; Najjar, Kühnl, et al., 2024; Ouedraogo et al., 2022), ab initio assessments (Ebmeyer et al., 2024), as well as occupational safety (Dawick et al., 2025), the measurement of metabolism, particularly skin metabolism, was a key aspect in refining the exposure and thus, the overall safety assessment. Therefore, the NAMs need to be relevant, reliable and fit-for-purpose to increase confidence in the safety assessment.

As more case studies on NGRAs for cosmetic ingredients are published, there is increasing reliance on the development of PBK models as higher-tier refinement tools to be able to predict the plasma concentrations of the parent and/or metabolites under realistic consumer use (Cable et al., 2025; Ebmeyer et al., 2024; Middleton et al., 2022; Najjar, Kühnl, et al., 2024). While there are several open source (e.g., httk, PKSim) and commercial PBK (e.g., GastroPlus, Simcyp PBPK Simulator) platforms available, only a few include a skin module to predict plasma concentrations of the parent chemical via the dermal route. Moreover, PBK models incorporating first-pass metabolism in the skin and concentrations of metabolites are generally lacking. A major priority for non-animal risk assessments of chemicals exposed via the dermal route is thus to be able to develop robust predictive PBK platforms that incorporate skin absorption and metabolism.

This review extends the knowledge of skin metabolism described in multiple reviews (Eilstein et al., 2014; Gibbs et al., 2007; Hewitt et al., 2013; Oesch et al., 2014; Pyo and Maibach, 2019). To fill the knowledge gaps, a literature search was conducted between April and October 2025 using PubMed and the Web of Science. Key words and combinations of terms included “skin metabolism”, “human”, “xenobiotic”, “topical product”, “cosmetics”, “drug”, “pesticide”, “in silico”, “in vitro”, “in vivo”. Papers related to “animal”, “fish”, “mouse” and other animal species were excluded. Since this review addresses the integration of skin metabolism in NGRAs of topically applied ingredients, the literature search was filtered to include references relating to healthy skin; therefore, keywords such as “disease”, dermatitis”, “psoriasis” were excluded. Lastly, metabolic activities reviewed focused on those occurring in the skin, thus papers describing metabolic capabilities on the surface of the skin, such as those related to the microbiome were not considered.

This review also provides updated information and perspectives from the cosmetic industry, in which the majority of ingredient exposures occur via the dermal route. Examples of in silico and in vitro skin models used in NGRA case studies from the Cosmetics Europe Long Range Science Strategy (LRSS) program and/or in the current ICCS program are described. Therefore, this review serves as essential groundwork for ongoing ICCS initiatives on skin metabolism, PBK modeling and IVIVE, as a platform to disseminate these collective insights and, crucially, to identify knowledge gaps and challenges in achieving them. To achieve this, the review systematically evaluates skin XME localization, experimental in vitro models and viability criteria, and in silico structure generators required for quantitative PBK integration in NGRAs. The review is intended primarily for safety assessors and study planners who need to select and interpret skin-metabolism approaches, while also providing a current overview of available methods.

2. Xenobiotic metabolizing enzymes in skin and their localization

Metabolism is only effective if the chemical interacts with an enzyme; therefore, it must penetrate the skin and reach the XME-containing skin layers in order to be metabolized. Researchers have attempted to determine the location of XMEs in the skin for many years. Early studies were conducted in the 50 s by Montagna, who published a series of articles on the “Histology and Cytochemistry of human skin”. These articles covered many aspects of the skin, including the distribution of some metabolic activities, such as monoamine oxidase (Yasuda and Montagna, 1960), β-glucuronidase (Montagna, 1957), as well as non-specific esterases (Montagna and Formisano, 1955), phosphorylase and amylo-1,6-glucosidase (Ellis and Montagna, 1958), which are mainly localized in viable epidermis and skin appendages (i.e., hair follicle, sweat and sebaceous gland). These studies were based on incubations of skin slices with a prototypical substrate, and the metabolite formed was localized using a simple reaction reagent. For example, glucose-1-phosphate was used as a substrate for phosphorylase activity, and the glycogen formed was localized using Lugos’s iodine solution (Ellis and Montagna, 1958). Others have investigated the localization of XMEs using techniques such as immunostaining (Cheung et al., 2003, 1999) or immunofluorescence (Baron et al., 2001), or by comparing them in different models (Bachhav and Kalia, 2009). Today, XMEs characterization can be conducted at three levels: at the transcriptomic level, proteomic content level and at the functional level using a prototypical substrate (Hewitt et al., 2013). It should be kept in mind that the presence of a gene or protein does not always mean the XME is functional (Pushparajah et al., 2008; Rodríguez-Antona et al., 2002); therefore, the most relevant endpoint for metabolism studies is the ability of the test system to metabolize a prototypical substrate.

Supplementary Table S1 summarizes Phase 1 and 2 XMEs reported to be present in the skin, along with the substrate used and, where stated, their localization. Results from multiple studies indicate metabolic activity in the skin is mainly associated with the epidermis. For example, investigations measuring the esterase-mediated hydrolysis of prednisolone 21-acetate in skin models representing different skin layers were reported by Bätz et al. (2013). They showed that the rate of metabolite formation was in the order: reconstructed human skin (epidermis and dermis) = RHE > excised human skin and keratinocytes > fibroblasts. Hikima et al. (2001) showed that esterase-mediated metabolism of prednisolone 21-acetate was 12 times higher in the epidermis, mainly in the basal layer, compared to the dermis. This ratio is similar to the reported relative oxygen consumption in the epidermis and dermis (4.53 ± 1.39 and 0.49 ± 0.12 µl O₂/mg/h, respectively (Pyo and Maibach, 2019)). At the cellular level, sebaceous cells in mice exhibit the greatest O-dealkylation, hydroxylase and UDP-glucuronyltransferase (UGT) activities, followed by differentiated keratinocytes and then basal cells (Steinsträsser and Merkle, 1995). Metabolism in other skin locations, such as the dermis, should not be overlooked, especially since some enzymes are exclusively expressed in the dermis e.g., CYP2A6 (Oesch et al., 2014). Conversely, if an XME activity is observed in keratinocytes, it does not mean that the same activity does not occur in fibroblasts present in the dermis.

Studies using ex vivo human skin explants have revealed the presence of several Phase 1 XMEs, including alcohol dehydrogenase (ADH), aldehyde dehydrogenase (ALDH), 7-ethoxycoumarin O-dealkylation (ECOD), as well as the epoxidation of cinnamyl alcohol, reduction of the aldehyde group of vanillin (Géniès et al., 2019) and azo reduction of dyes (Collier et al., 1993). Others have reported hydroxylated metabolites of arachidonic acid and linoleic acid in vivo in rabbits and in incubations with ex vivo human skin explants (van de Sandt et al., 1995). N-Dealkylation and aromatic hydroxylation of propranolol were also observed using microsomes prepared from human skin (Ademola et al., 1993). In all these examples, neither the enzyme nor their localization were known, including microsomal incubations, which were prepared using full-thickness skin. Metabolism profiles of 7-ethoxycoumarin, cinnamyl alcohol and vanillin were compared between ex vivo human skin explants and skin S9 prepared from a RHE model (Géniès et al., 2020). The main metabolites were observed in both models but some additional metabolites were only observed in human skin explants. Two possible reasons for this difference are, firstly, this could indicate that the additional metabolites were formed only in the dermis since S9 was obtained from an epidermal model (RHE). Secondly, different analytical methods were used: radiolabeled chemicals were used to study metabolites produced in human skin explant incubations, whereas a triple quadrupole LC/MS/MS system was used for parent quantification and metabolite identification for RHE S9 incubations. The use of radioactive chemicals is more sensitive and specific for the detection of known and unknown metabolites in a biological matrix, leading to a higher number of metabolites identified.

A recent study reported the metabolism of 47 chemicals in liver and skin S9 (Eilstein et al., 2020). Of these, it was only possible to measure intrinsic clearance for 38% of the chemicals in skin S9 compared to 66% in liver S9. Thus, hepatic metabolism is clearly more efficient than that of the skin, especially given the greater abundance of XMEs present in the liver compared with the skin (van Eijl et al., 2012). However, this does not mean that hepatic clearance is always greater than skin clearance. For example, N-acetyl transferase (NAT) (Eilstein et al., 2020) or esterase (Obringer et al., 2023) activities have been shown to be higher in skin than in liver. The difference in esterase activities could be due to different carboxylesterase (CES) isoforms expressed at different levels in the skin and liver (Lester et al., 2021). CES2 is mainly expressed in the skin and has greater affinity for lipophilic substrates with large alcohol groups, whereas CES1 has greater abundance than CES2 in liver and preferentially metabolizes substrates with a large acyl group and small alcohol groups (Obringer et al., 2021).

Several reviews have compared XME activities in skin from different species (Oesch et al., 2014), different skin models (Eilstein et al., 2014; Hewitt et al., 2013), as well as specific XMEs in native skin (Eilstein et al., 2014; Géniès et al., 2019). Notably, results from assays measuring genes, proteins or activities are not always consistent. For example, NAT expression in the skin is readily measured at the mRNA level; however, the protein itself is often below the limit of quantification (van Eijl et al., 2012)), even though NAT activities can be measured using prototypical reference substrates (Götz, Pfeiffer, et al., 2012). Despite advances in various technologies, the precise localization of many XMEs in different skin layers remains largely unknown. Moreover, several contradictory conclusions have been made regarding skin enzyme activities (Oesch et al., 2014), which may be related to the different experimental designs and endpoints used. This makes it difficult to directly compare results across studies; however, they highlight where more work is needed to clarify opposing information.

3. Determinants of first-pass dermal metabolism

3.1. Chemical penetration and access to XMEs

Skin bioavailability is determined by the (a) ability of the chemical to penetrate into the skin, (b) distribution into different skin layers, (c) transformation into metabolites and (d) elimination of parent and/or metabolites from the skin into the systemic circulation. The outermost layer of the skin is the stratum corneum (SC), which acts as the main barrier to physical, biological and chemical attack. After the chemical passes through the SC, it must diffuse across the epidermis and dermis. As it passes into the dermis, it can reach the systemic circulation through the vascular network localized in dermis. The mathematical description of these processes involved in skin bioavailability is complex. The distribution of a chemical across the skin depends on the different pathways involved, such as via the SC and/or hair follicles, as well as the heterogeneity of the skin and metabolic activities within the different layers. Some parameters depend on the formulation e.g., coefficients related to the SC, whereas parameters describing diffusion across the epidermis and dermis are only related to chemical properties (Golbamaki et al., 2025). Permeation of a chemical through the different skin layers occurs at different rates and extents. The partition and diffusion coefficients of the chemical for each skin layer, which can be measured using dedicated experimental designs (Ellison et al., 2020), determines how long the chemical is co-located with the enzymes. Thus, knowledge of the location of XMEs involved in the metabolism of a chemical, along with its clearance, partition and diffusion coefficients within the skin layers will help to determine the extent of first-pass metabolism in the skin.

3.2. Extrinsic and intrinsic factors

It is reported that there are 20 different factors that can affect skin absorption (Law et al., 2020), which in turn can also impact skin metabolism. Sources of variability which can all affect in vivo and ex vivo assays include chemical properties, the formulation type, the exposure scenario (e.g., dose, duration and frequency of application) and inter-individual variability. These factors can be categorized as: extrinsic factors i.e., related to exposure (UV, chemical exposure etc.) and intrinsic factors i.e., related to skin biology and human variability (anatomical area, age, gender, genetic etc.). Table 1 summarizes literature reports in which different factors were observed to affect skin metabolism, mostly based on gene expression analysis. Some findings were not entirely consistent, e.g., the effect of age, some depended on the experimental conditions, (e.g., UV exposure), and some relied on a link between enzymatic activity and gene expression. However, these provide a general indication of reasons for variability,

Table 1.

Examples of studies showing factors affecting skin metabolism and reported conclusions.

Parameter	Skin Model	Protocol	Observed effect	Reference
UV exposure	Human in vivo (biopsies)	24 h UVR exposure 290–400 nm, 770 mW/cm², 3.9–48 J/ cm²	Enzyme induction (COX-2, CYP2E, GSTP1 or MRP1) Variable from subject to subject. No CYP1A1 induction CYP1B1 induced in a few subjects over 26	Smith et al. (2003)
		24 h UVB exposure	Induction of CYP1A1 and CYP1B1 mRNA	Katiyar et al. (2000)
UV & Psoriasis	Human ex vivo (biopsies)	24 h UVR exposure 290 nm, 770 mW/cm², 3.9–48 J/ cm²	Greatest induction than on normal skin	Smith et al. (2003)
XME induction	Human in vivo (biopsies)	Treatment with crude coal tar (12%) or retinoic acid (0.025%) for 96 h	Induction CYP2S1 mRNA	Smith et al., 2003a, Smith et al., 2003b
	Ex vivo neonatal rat skin	Topical treatment with Benzo[a]pyrene	Induction of AHH activity
	Human Keratinocytes	Treatment with dexamethasone over 3 days at 0.1 µM	Induction of CYP3A4 mRNA	Baron et al. (2001)
	Human Keratinocytes	Treatment with benzanthracene over 2 days at 10 µM	Induction of CYP1A1 mRNA	Baron et al. (2001)
	HaCaT	Treatment with retinoids and UV radiation	Induction CYP2S1 mRNA	McNeilly et al. (2012)
	Human Keratinocytes, RHE, HaCaT	Treatment with a mixture of inducers, rifampicin, dexamethasone and 6-(4-chlorophenyl:imidazo[2,1-b]thiazole-5-carbaldehyde O-(3,4-dichlorobenzyl)oxime for 1 to 3 days	Induction of CYP1A/1B, CYP2B activities	Götz, Hewitt, et al. (2012), et al. (2012)
Age	Ex vivo human skin from young (25–55) vs middle aged (>60) donors	17 enzymes from glycolysis, pentose phosphate pathway, citric acid cycle and amino acid metabolism	Decrease of enzymatic activity to a maximum of 15 to 25%. Higher activity loss for enzymes from the citric acid cycle	Salfeld (Physiologie und Pathophysiologie (einschliesslich Histologie) der Altershaut./1966)
	Ex vivo human skin from 63 subjects from 1 month to 90 years	14 enzymes activity	No difference	Yamasawa et al. (1972)
	Human skin S9 from 87 donors from 17 to 90 years old	Phase II activities (NAT, UGT, SULT, COMT, GST)	Negative correlation with age only for GST	Spriggs et al. (2018)
Gender	Human skin S9 from 87 donors: 72 women, 17 men	Phase II activities (NAT, UGT, SULT, COMT, GST)	No difference with gender	Spriggs et al. (2018)
Isophorm	Subcellular fractions obtained from ex vivo human skin (10 donors)	NAT activities measured with p-Toluidine	Inter-individual variability with NAT activity ranking from 0.63 to 3.03 nmol/min/mg protein	(Götz, Pfeiffer, et al., 2012)
Anatomical area	Ex vivo human skin from different anatomical area	Hydrocortisone 5a-reductase activity	Observed only in foreskin	Steinsträsser and Merkle (1995)
Skin condition	Human skin biopsies (in vivo)	Gene expression	Gene upregulated UGT1A7, UGT1A9, UGT1A10, CYP4F12, and CYP2E1 in Atopic Dermatitis but not in Ichthyosis Vulgaris skin	Blunder et al. (2018)

4. Experimental in vitro models for studying skin metabolism

Over the past twenty years, numerous in vitro models have been used to assess skin metabolism. The in vitro models by which skin metabolism has been measured in various NGRA case studies varied, and included native human skin, RHE tissues, isolated cells and subcellular fractions. Each in vitro skin model has its own advantages and limitations: cost, availability of the model, complexity of use, uncertainty, and the IVIVE relationship. Here we provide a brief overview of different skin metabolism models currently available from commercial suppliers, along with their advantages and limitations. The advantages and limitations of each model are summarized in Table 2 and are described below.

Table 2.

Uses, advantages and disadvantages of different skin models.

Skin model	Uses	Main strengths	Key limitations
Ex vivo skin explants	Quantify dermal absorption, metabolism, distribution, and flux under relevant use conditions; measure formulation effects Measuring metabolism (and toxicity) of topically-applied chemicals	Closest model to in vivo native human skin; simultaneously captures barrier function and skin metabolism. Maintains XMEs longer than standard Franz cell set-up; conditions maximize metabolism	Limited availability of tissue; higher cost; donor variability lost over a few days not a long-term incubation model unless modified (e.g. NativSkin); higher cost; incubation conditions (temperature, occluded, humidified atmosphere) not comparable with consumer use; continued metabolism by underside of the skin increases uncertainty; not possible to measure metabolism kinetics i.e. intrinsic clearance separately from penetration kinetics
Non-viable (frozen) human skin	Passive permeation control under relevant use conditions; measure formulation effects	Represents passive transport; stable and practical	Limited availability of tissue; no XME-mediated metabolism (apart from carboxylesterases and peptidases);
RHE model	Measuring metabolism (and toxicity) of topically-applied chemicals	Stratified epithelium; express Phase 1 and 2 XMEs; supports repeated topical dosing; compatible with irritation/sensitization assays; can be used in organ-on-chip microfluidic systems; good reproducibility due to industrial scale production standardization	Only represents the epidermal layer
Full-thickness reconstructed skin models	Human-relevant for measuring metabolism (and toxicity) of topically-applied chemicals under relevant use conditions; measure formulation effects	Contains a stratum corneum, epidermis and dermis supports (nearest RS model to native human skin); express Phase 1 and 2 XMEs; repeated topical dosing; compatible with genotoxicity assays; can be used in organ-on-chip microfluidic systems; good reproducibility due to production standardization	Higher cost compared to RHE and S9 models;
Primary human keratinocytes	Screening for metabolic stability (intrinsic clearance); epidermal cell specific metabolism and mechanisms	Easy to isolate and culture; represent cellular metabolism (i.e. considering the plasma membrane); can measure intracellular and extracellular metabolites	Altered phenotype/enzyme expression in culture; no stratum corneum barrier; no dermal component
HaCat cells	Screening for metabolic stability (intrinsic clearance); Epidermal cell specific metabolism and mechanisms	A single donor with virtually unlimited supply of cells; good reproducibility; can be used in higher throughput; represents cellular metabolism; can measure intracellular and extracellular metabolites	Immortalized cell line with altered phenotype/enzyme expression compared to primary keratinocytes; a single donor (does not reflect donor variability)
Dermal fibroblasts	Measures contribution of dermal metabolism	Essy to isolate and culture; good model representing lower skin layer	Less relevant to first contact with a topical chemical; no barrier
Human RHE/skin S9 fractions	Screening for metabolic stability (intrinsic clearance); suited for comparing chemical series	Easy to use; scalable (i.e. converting values for mg protein/nm to whole skin); enables cofactor-dependent testing; relatively economical; suitable for testing the impact of different Phase 1 and 2 cofactors	Whole skin S9 is technically more challenging to prepare than cell suspensions; lack cellular compartmentalization; requires cofactor supplementation; only short (hours) incubations possible; may overestimate clearance due to a lack of plasma membrane; careful selection of controls required, especially for lipophilic chemicals (to account for solubility and non-specific binding)
Skin microsomes	Screening for metabolic stability (intrinsic clearance)	Easy to use; scalable; suitable for testing the impact of different Phase 1 and 2 cofactors	Low yield and technically challenging to prepare; requires cofactor supplementation; only short (hours) incubations possible; may overestimate clearance due to a lack of plasma membrane; careful selection of controls required, especially for lipophilic chemicals (to account for solubility non-specific binding)

4.1. Ex vivo native skin models

Of all the in vitro models used to evaluate skin metabolism, the closest to in vivo are biopsies of native human skin or, alternatively, pig skin, referred to as ex vivo explants (Géniès et al., 2020, 2019, 2025a, 2025b; Jacques et al., 2014). When the appropriate set up is used, the metabolism and absorption of a chemical can be evaluated simultaneously. Classically, skin absorption is evaluated using the static Franz Cell apparatus, flow through diffusion cells (Hewitt et al., 2020, 2000; Yourick and Bronaugh, 2000) or side-by-side cells (Liu et al., 1994). These models have been used to demonstrate the presence of Phase 1 activities, such as ADH (Ademola et al., 1993; Liu et al., 1994), hydroxylation (Ademola et al., 1993), esterase (Hewitt et al., 2000), as well as Phase 2 activities such as NAT and sulfotransferase (SULT) activities (Yourick and Bronaugh, 2000). A split-thickness model was proposed by Altenbach et al. (2006), who used heat separated epidermis sheets mounted on side-by-side diffusion cells according to a specific design called “reflection boundary”. The SC was in contact with an impermeable barrier consisting of a plastic foil. The epidermis was directly exposed to the chemical and metabolites were collected in the same compartment. A dipeptide Tyr-Phe was evaluated with this set up, which clearly demonstrated that the dipeptide was hydrolyzed into corresponding amino-acids.

Whatever the type of diffusion cell, the conditions under which the models are kept can stress the skin, decrease its viability and thus reduce the skin metabolic activity (De Wever et al., 2015). To overcome this constraint, an alternative to the Franz cell was proposed by Rutten et al. (1990), whereby a fresh piece of skin was cultured on an insert with a microporous membrane. van de Sandt et al. (1993) updated the system with a glass ring glued on the skin surface to define the exposure area. Despite not needing a diffusion cell, they observed alterations in the histology of the epidermis and upper dermis in areas directly under the rings. This method was further improved in similar setup developed by Jacques et al. (2014), which has been used to evaluate formulation effects on skin metabolism and absorption of cosmetics ingredients (Géniès et al., 2025b). Another ex vivo skin model, called NativeSkin®, was developed to simplify the setup (De Wever et al., 2015). In the NativeSkin® model, fresh skin is embedded in a nutritive gel to maintain its physiology, viability, barrier function and metabolic capability over 7 days. This model has been used to investigate skin barrier dysfunction, skin diseases (e.g. atopic dermatitis, skin cancer) and associated therapies (Clancy et al., 2026; D'Avino et al., 2026; Lefebvre et al., 2026)

In all the test systems using skin explants, the skin explant is covered with a lid to prevent evaporation of the receptor medium, thus resulting in occluded conditions, which can increase skin penetration. Moreover, the skin is typically placed in CO₂ incubator maintained at 37 °C with saturated humidity. The higher temperature can also increase penetration and metabolism compared to standard skin penetration assays, in which the skin is maintained at 32 °C under normal hydration conditions. The design of the skin explant studies (Géniès et al., 2019, 2025a; Jacques et al., 2014) aimed to maximize metabolism, whereby the static conditions allowed parent reaching the culture media below the skin to always be in contact with the underside of the skin, resulting in further metabolism. This increased bioavailability adds uncertainty but should be considered a conservative parameter in the exposure assessment of an NGRA. In addition, the presence of a lid may cause the SC to be highly hydrated, as observed in occlusive conditions. Since lipophilic chemicals are more affected by occlusion than hydrophilic chemicals (Hafeez and Maibach, 2013), this factor should be considered when extrapolating to the in vivo situation.

4.2. Reconstructed human epidermal models

The main limitations associated with ex vivo human skin explants are the limited availability and variability of the human skin. To overcome these limitations, 3D RHE models offer a suitable alternative. Produced on an industrial scale in a reproducible manner, RHE models have been used as an alternative to human skin in metabolism studies (Gibbs et al., 2007; Wiegand et al., 2014). Many Phase 1 and 2 XMEs are reported to be expressed in these models (Brandmair et al., 2024; Chen et al., 2021; Hu et al., 2010; Kazem et al., 2019; Kühnl et al., 2021; Luu-The et al., 2009), albeit based on the expression of XME mRNA. Ideally, activities need to be confirmed by evaluating the metabolism of prototypical reference substrates (Eilstein et al., 2015; Gibbs et al., 2007; Götz, Pfeiffer, et al., 2012; Wiegand et al., 2014). For example, Eilstein et al. (2014) measured esterases, UGTs, glutathione-S-transferase (GSTs) and NAT activities in RHEs using systemic exposure i.e. the substrate was added to the medium below the RHE model. The measured V_max represented several processes, including the diffusion of the chemical from the receptor compartment to the epidermal layer, metabolism by the XME, and back diffusion of metabolite into the receptor compartment. Thus, this V_max did not represent the metabolism activity according to the definition of Michaelis-Menten kinetics. Nevertheless, the in vitro clearance measured in RHE models, defined as the ratio, V_max/K_m, normalized to the amount of protein, was within 2-fold of those measured on native human skin (Eilstein et al., 2014). These results support the use of RHE models for investigating skin metabolism.

One caveat of the RHE model is that it does not replicate all the metabolic activities present in the skin since fibroblasts, sebaceous glands, hair follicles and even melanocytes present in the epidermis are absent from RHE models. It is possible to use a full-thickness model including fibroblasts (e.g. the Phenion Full Thickness model (Wiegand et al., 2014)); however, to date, there are no in vitro models which account for the diversity of cells present in the skin as potential sources of metabolism. Despite this limitation, RHE models provide good representative test systems for measuring skin metabolism, as keratinocytes are primarily responsible for the skin’s metabolic activity. In recent years, other skin models have been developed using technologies such as bioprinting (Gao et al., 2021; Quílez et al., 2020), 3D electrospun scaffolds (Wu et al., 2024) with the aim of overcoming some of the limitations of classical 3D models e.g., lower barrier function and the absence of appendage. While these show promise, these models are not well characterized regarding their metabolic capabilities.

4.3. Isolated skin cells

Skin cells, including sebaceous cells, basal keratinocytes, differentiated keratinocytes, can be relatively easily isolated from skin biopsies, making it possible to measure XME activities in specific cell types (Steinsträsser and Merkle, 1995). Data from incubations with these cells can help to investigate cell-specific pathways, although to obtain comparable results, the data should be normalized to the cell number.

Studies reported in the 90 s (e.g., Steinsträsser and Merkle (1995) focused on keratinocytes, as these are the most abundant cells in the skin and they are present in the main site of metabolism i.e. the epidermis (Pyo and Maibach, 2019; Quantin et al., 2022). Unlike subcellular fractions, which require the addition of a mixture of cofactors for most XMEs and lack cellular compartmentalization, human epidermal keratinocytes are intact cells and thus retain XMEs and related cofactors. This makes them a potentially more predictive and representative model for a full range of skin metabolic pathways occurring in vivo. Two types of keratinocytes have been used for metabolism studies: normal human epidermal keratinocytes (NHEK, primary keratinocytes) or immortalized keratinocytes (HaCat).

Monolayer primary keratinocyte cultures are widely used for basic skin biology research. They can be cultured under appropriate conditions to maintain a differentiated state. Neonatal human epidermal keratinocytes (HEKn) and adult NHEK have been used to evaluate the metabolism of chemicals (Bergström et al., 2007; Rukthong et al., 2020; Steinsträsser and Merkle, 1995), as well as CYP induction (Götz, Hewitt, et al., 2012; Wiegand et al., 2014). Keratinocytes can be incubated as simple monolayer cultures in multi-well plates or adapted to mimic the application scenario. For example, Rukthong et al. (2020) cultured HEKn on a permeable support and then treated them topically with a solution containing the test chemical (a-mangostin and gartanin). The permeation through the monolayer was monitored over time and the metabolites quantified.

HaCaT cells were developed as a model for skin metabolism (Steinsträsser and Merkle, 1995). An advantage of this cell type is that it is a reproducible source of one cell type in large quantities. HaCaT cells have been successfully used for evaluating different metabolic pathways e.g., GST (Jacquoilleot et al., 2015) and NAT (Goebel et al., 2009; Manwaring et al., 2015). The main disadvantage is that these are transformed cells, with some differences in enzymatic activity. For example, Jacquoilleot et al. (2015) compared GST activities in HaCat and NHEK cells and ex vivo human skin by normalizing activities to either the amount of cellular protein (cells) or tissue weight (human skin). As the normalization factors were different, it was difficult to compare GST basal levels. However, a comparison of GST based on apparent V_max values indicated that GST activities in human skin were many thousand-fold lower than in HaCaT cells.

4.4. Subcellular fractions

A simple and practically less challenging skin metabolism model is skin homogenate. A technical challenge arises when preparing dermis homogenates as it is an elastic biological tissue that cannot be easily homogenized using standard Potter’s homogenizers. This can be overcome by making a skin powder using cryogenic grinding. This technique has been used to prepare homogenates from split thickness human cadaver skin stored at 4 °C (Roy and de Groot, 1994) or at −70 °C (Rohatagi et al., 1997), fresh human adult foreskin (M. Foldvari and C. J. Oguejiofor, 1997) or frozen abdominal human skin (Sugino et al., 2017). The homogenates can be directly used in incubations (Rohatagi et al., 1997)) or they can be centrifuged to remove large debris before incubating (M. Foldvari and C. J. Oguejiofor, 1997; Roy and de Groot, 1994; Sugino et al., 2017). Skin homogenates can be supplemented with cofactors (M. Foldvari and C. J. Oguejiofor, 1997) or incubated without cofactors (Sugino et al., 2017), with the design of the study depending on the enzymatic activity investigated. Sugino et al. (2017) evaluated the hydrolysis of phthalate esters, thus the centrifuged homogenates did not require cofactors. By contrast, M. Foldvari and C. J. N. Oguejiofor (1997) studied the metabolism of Prostagladin E1, the metabolism of which is mediated by Phase 1 XMEs; therefore, NAD+ was included in the centrifuged homogenates.

One disadvantage of skin homogenates is that the XMEs are diluted due to the presence of extracellular matrix. Therefore, an alternative is to prepare more concentrated subcellular fractions. S9 and microsomes can be obtained by homogenizing the skin and subjecting it to different centrifugation speeds. The main difference between these two different subcellular fractions is the absence of cytosolic enzymes in microsomal preparations, leading to some activities being absent or inactivated e.g., flavin-monooxygenases (Richardson et al., 2016). Thus, skin S9 is suitable for evaluating a wider range of enzymatic activities, whereas microsomes are more useful when evaluating Phase 1 CYP- or UGT-mediated metabolism. Both subcellular fractions are easy to use for metabolic stability assays, whereby the chemical is simply incubated in the presence of the subcellular fraction and relevant cofactors (Table 3). For example, Lester et al. (2021) evaluated the contribution of Phase 1 and 2 XMEs (other than esterases) to the metabolism of 16 parabens by conducting the incubations in the absence and presence of a cocktail of cofactors (NADPH, acetyl-CoA, reduced glutathione (GSH), UDP-glucuronic acid and adenosine 3′-phosphate 5′ phosphosulfate). Subcellular S9 fractions can also be supplemented with a cocktail of protease inhibitors to protect the intracellular proteins. For example, phenylmethylsulfonylfluorid (PMSF) is a serine protease inhibitor (serine hydrolase inactivator) commonly used in the preparation of cell lysates. When added to skin S9 during the incubation, it can increase the working lifetime, allowing the measurement of lower intrinsic clearance chemicals (Nichols et al., 2021). However, while protease inhibitors can increase CYP-specific activities, they can also inhibit esterase and peptidase enzymes (Nichols et al., 2021).

Table 3.

Cofactors used with skin S9 and microsome with corresponding enzymatic activity.

Cofactor	Chemical name	Enzymatic Activity	S9	Microsomes
NADPH	Nicotinamide adenine dinucleotide phosphate	CYP	√	√
NAD	Nicotinamide adenine dinucleotide	CYP, ADH, ALDH	√	√
AcetylCoA	Acetyl-Coenzyme A	NAT	√	-
UDPGA	Uridine 5′-diphosphoα-d-glucuronic acid	UGT	√	√
PAPS	3′-Phosphoadenosine-5′- phosphosulphate	SULT	√	-
GSH	Glutathione	GST	√	-
SAM	S-adenosyl-l-methionine	COMT	√	-

A disadvantage of skin microsomes is that the production yield is very low, which means that these preparations require a large amount of skin to obtain sufficient amounts of microsomes. By contrast, the preparation of skin S9 needs less skin; however, the way in which it is prepared can affect enzymatic activities. This was investigated by Spriggs et al. (2018), who measured the impact of the conditions of storage of the skin after surgery. This showed that maintaining the skin at 25 °C or 37 °C for 24 h before preparing the S9 decreased XME activities. To retain optimal metabolism activity, the fresh ex vivo skin should be used within 2 h post-surgery or immediately frozen in liquid nitrogen and stored at −80 °C until use. The use of an Ultra Turrax blender was preferred over a cell crusher in liquid nitrogen, with the observation that skin S9 from fresh skin was generally difficult to achieve. To overcome this constraint, another source of skin S9 can be used e.g., S9 derived from RHE models (Eilstein et al., 2020; Lester et al., 2021). Spriggs et al. (2018) also recommended preparing fresh skin S9 and not freezing it for storage before use. However, this recommendation is very restrictive and is not supported by data from others. For example, Phase 1 and 2 activities remain constant in liver S9 over 10 years storage at −70 °C and up to 10 freeze-thaw cycles (Richardson et al., 2016). This observation should also be valid for skin S9. Nevertheless, the paper by Spriggs et al. (2018) highlights the importance of the methods used to prepare skin S9 to guarantee robust results.

Experiments using S9 fractions and microsomes are simpler to carry out than those involving tissues (skin or RHE) or cells. Nevertheless, the clearance measured using subcellular fractions is often higher than that observed with other intact whole cell models. This is because the cellular structure is lost; therefore, the chemical does not need to cross the cell membrane and is in direct contact with the enzyme. Conversely, if the chemical is a substrate of an efflux transporter, clearance may be underestimated. Thus, when extrapolating data derived from subcellular fractions to in vivo conditions an additional level of uncertainty is introduced.

4.5. Optimization of skin metabolism assay designs

Importantly, in vitro incubation methods can be impacted by the loss of parent chemical not related to metabolism, including volatility, precipitation, non-specific binding or non-enzymatic reactions. To address these, Li et al. (2023) defined a list of criteria for conducting clearance measurements using hepatocytes, which can be applied for clearance measurements for other test systems. The aim of these criteria is to confirm that loss of chemical over time is only related to metabolism.

• •The change in negative control over the duration of experiment should be lower than 20%.
• •The relative standard deviation (CV%) of the negative control at t0 should be below 20%.
• •The concentration at t0 in control should be within ±20% of the nominal concentration or within ±20% of the assay concentration at t0.

The negative control corresponds to an incubation without enzymatic activity. Typically, this is the same incubation without cells, S9 or microsomes. However, a disadvantage of these buffer-only controls is that non-specific binding to the plastic can take place due to an absence of protein (which helps to solubilize the chemical). An alternative is to inactivate the enzymes by using heated deactivated cells, S9 etc. With S9 and microsomes, a third option is to incubate the chemical with the sub-cellular fraction but without cofactors. This cannot be used, of course, if the chemical is hydrolyzed by esterases since the cofactor for this reaction is water. Despite the inclusion of controls some clearance measurements remain a challenge due to non-specific binding of highly lipophilic properties or residual hydrolysis (Grégoire et al., 2024). When technical difficulties arise, the additional level of uncertainty in the in vitro clearance measurement should be accounted for in the NGRA.

Clearance can also be influenced by non-covalent binding, which occurs in the skin as well as in the liver (De Buck et al., 2007). The extent of non-covalent binding depends on the type of proteins present in the skin layer: keratin in the SC (Norlén and Al-Amoudi, 2004), albumin in the epidermis and dermis (Hummer et al., 2021; Rabilloud et al., 1988) and collagen in the dermis (Gelse et al., 2003). Non-covalent binding may reduce the diffusion coefficient and result in a longer residence time in the skin layers, which could allow XMEs more time to metabolize the parent compound. By contrast, if only the free fraction of a chemical is available for metabolism, high non-specific binding may limit the metabolic rate (as observed in hepatic clearance models (De Buck et al., 2007). Developing validated, standardized methods to accurately determine or model the free faction of a chemical in the skin represents a critical ongoing research priority for dermal IVIVE. These two opposing effects can make it difficult to anticipate the overall impact of non-covalent protein binding on skin metabolism. Another more recent finding is that uptake and efflux transporters are expressed in the skin (Osman-Ponchet et al., 2014). While the impact of transporters on the clearance of chemicals in the liver has been extensively investigated, this area of research is relatively new to skin metabolism.

5. Challenges and open questions

5.1. Standardized skin metabolism assays and guidelines

While guidelines are available for the conduct of ex vivo skin absorption studies (OECD, 2004; SCCS, 2010), there are currently no dedicated guidelines for measuring skin metabolism or when to conduct the assay using viable skin. The current SCCS NoG (SCCS, 2023) address the wider aspects of the potential metabolism of cosmetic ingredients (including dermal metabolism); however, there is no guidance on how to conduct the experiments e.g. the types of skin models, protocols, or controls to use. The OECD 428 Test Guideline (OECD, 2004) includes a brief sentence regarding skin metabolism: “When skin metabolism is being investigated, freshly excised skin should be used”, mentioning that the viability of the skin should be maintained during the entire experiment. The SCCS guideline (SCCS, 2010) also does not provide any additional information on this topic, other than to exclude frozen skin for skin metabolism studies. Notably, while most XMEs are inactivated by freezing, some XMEs are still active in frozen skin, especially esterases (Sugino et al., 2017). This means that some metabolism can still take place via esterases or peptidases present in frozen skin (Kraeling et al., 2015). To reconcile these practical realities with regulatory expectations, the OECD 428 Test Guideline could be updated to align with the current knowledge of skin XMEs and improved technologies. Additional advice is also needed regarding the selection of the most appropriate and practical in vitro model and associated analytical technique and assay controls for a given question (measurement of clearance, metabolite identification). Moreover, information is needed on how to prepare the in vitro models and design the assay conditions using a standardized protocol, with clear outcome and quality criteria.

The selection of the metabolism model listed in Table 3 should follow a pathway-specific strategy: frozen skin, skin S9 fractions or homogenates without cofactor supplements for Phase 1 hydrolysis pathways (carboxylesterases, peptidases), skin microsomes for Phase 1 CYP- or Phase 2 UGT-mediated metabolism, whereas a range of Phase 1 and 2 mediated pathways can be evaluated using intact tissues e.g., fresh ex vivo skin and RHE tissues (without an exogenous cofactor supplementation), or S9 fractions or tissue homogenates supplemented with a mix of cofactors e.g., NADPH, UDPGA, PAPS, Acetyl-CoA. Relevant considerations include the skin compartment and metabolic pathway of interest, the need for qualitative or quantitative information, preservation of cellular organization and cofactors, exposure conditions, analytical sensitivity, and compatibility with IVIVE or PBK modelling. Even in the absence of a dedicated guideline, the design and reporting of skin metabolism studies should be driven by a clearly defined scientific question and the intended use of the resulting data. The selected test system and experimental conditions should therefore be justified as fit for purpose. This justification should explain why a particular model was selected and why alternative approaches were considered less appropriate. The integrity of kinetic data is paramount; therefore, rigorous experimental controls are needed to ensure the robustness of the intrinsic clearance measurements. Appropriate controls, model-specific quality criteria, and transparent reporting of limitations and uncertainties are particularly important for non-guideline studies.

5.2. Development of sensitive analytical methods

A particularly challenging aspect of skin metabolism assays is the extraction of the analytes from the skin sample, especially ex vivo explants or full thickness reconstructed skin models. This is partly due to the particularly robust structure of the skin (making it difficult to homogenize) and, for some chemicals, partly due to non-reversible covalent binding to proteins, with collagen being a major component of the skin.

As for many assays, the sensitivity of the analytical method can impact the accuracy of the measurement of skin metabolism. The "absence of evidence is not evidence of absence" remains a relevant caveat in metabolite profiling. The non-detection of a predicted species may stem from suboptimal chromatographic resolution or the sensitivity limits of the mass spectrometer. Analytical protocols should reach the stringency defined by the Food and Drug Administration bioanalytical method validation guidelines (FDA, 2018), achieving established benchmarks for accuracy, precision, selectivity, and robustness. Typically, a mass balance of 100 ± 10% for radiochemical assays and 100 ± 15% for LC-MS/MS are required. Consequently, a metabolite representing 5–10% of the initial dose could remain undetected even within a "validated" mass balance.

For metabolite identification, High-Resolution Mass Spectrometry (HRMS) remains the gold standard for elucidating metabolic pathways. Structural elucidation, specifically the identification of regioselective hydroxylation or conjugation, is significantly enhanced by the strategic application of stable isotope labeling (VandenHeuvel, 1986) and integrated in silico platforms optimized for metabolic hotspot prediction (Madden et al., 2020).

5.3. Development of in silico skin metabolite identification tools

In silico models have been developed for predicting liver metabolites (Madden et al., 2020) but there are fewer models that specifically predict metabolic pathways in skin. One way to predict skin metabolites is to use an in silico liver metabolite model to identify potential metabolic pathways and then check whether the XME involved is also present in skin. This is associated with some uncertainty, since activity may not be linked to the relative abundance of the XME e.g. NAT1 is highly active in skin despite the protein not being detected (van Eijl et al., 2012). More recent metabolite predictive algorithms have been adapted to inform skin metabolism predictions. These are divided into rule-based expert systems, 3D structure-based biophysical models, and machine learning hazard classifiers. The primary regulatory and industrial standards for predicting physical metabolite structures are TIMES-SS (Patlewicz et al., 2007; Roberts et al., 2007) and the free version, OECD QSAR Toolbox (both utilizing the rule-based Skin Metabolism Simulator) and MetaSite (from Molecular Discovery) (Słoczyńska et al., 2015). Unlike empirical rule-based systems, MetaSite represents a sophisticated 3D docking-based approach. It uses Molecular Interaction Fields coupled with a dedicated SKIN Consensus Model that weights predictions based on CYP isoforms physiologically predominant in human skin (such as CYP1A1, CYP2E1, and CYP3A5). The accuracy of this platform has been validated in independent cosmetic case studies, successfully predicting the specific biotransformation pathways of complex topical ingredients like retinoids (Słoczyńska et al., 2015). Despite the advances in skin metabolites prediction models, a persistent challenge remaining is the lack of quantitative kinetic outputs i.e., the Michaelis-Menten parameters, V_max and K_m, despite the provision of probability scores for metabolic "hotspots".

5.4. Models used to predict reactive intermediates

The assessment of reactive metabolites, specifically those forming transient, electrophilic species, is notoriously complex since they cannot be directly detected by traditional analytical methods. An example of this analytical uncertainty is resorcinol. Although classified as a weak sensitizer requiring metabolic activation (pro-hapten), comprehensive studies using radiolabeled resorcinol in viable human skin failed to isolate reactive intermediates (Géniès et al., 2019). The metabolic profile was dominated exclusively by Phase 2 detoxification (sulfonation and glucuronidation). Therefore, although the detection of GSH conjugates is an indication for reactivity, the absence of GSH-conjugates is not definitive proof of safety. Despite the development of promising in silico metabolite prediction models such as TIMES-SS, their predictivity varies depending on the reaction being simulated (OECD, 2012). Therefore, toxic reactive intermediates that elude experimental confirmation may be best evaluated using “trapping”, whereby the test chemical is incubated with liver microsomes and stable isotope labeled GSH, potassium cyanide (CN) and semicarbazide (SCZ) as trapping agents (Rousu et al., 2009). The reactive metabolites are trapped as GSH-, CN- or SCZ-adducts and detected using liquid chromatography/HRMS or ultra-performance liquid chromatography/time-of-flight mass spectrometry (UPLC/TOFMS) based on the observation of twin ions separated by 3 Da in the mass spectra, which allows the highly specific identification of GSH-, CN- or SCZ-related signals. While these methods exist for liver-derived reactive metabolites, there are no equivalent assays using skin microsomes, possibly because this would not add to the binary outcome of the assay – i.e., whether the chemical has the potential to form reactive metabolites.

5.5. Development of IVIVE skin modules for systemic exposure assessment

Until recently, most in silico skin models predicted only the penetration of non-metabolized chemicals through the skin into the receptor fluid (Grégoire et al., 2021). However, recent work has endeavored to incorporate skin metabolism kinetics into in silico penetration models, which are then integrated into a dermal module in a PBK model (Coleman et al., 2020; Dancik et al., 2013; Ellison et al., 2025; Golbamaki et al., 2025; Najjar, Lange, et al., 2024).

The first step in scaling skin (or liver) metabolism data to in vivo clearance values is to calculate the in vitro intrinsic clearance value, which can be conducted using two methods. If multiple concentrations are tested, the Michaelis-Menten equation can be fitted to determine the K_m and V_max, and thus, the in vitro clearance (V_max/K_m). The second method is based on the disappearance of the parent compound, which is tested at concentrations below the expected Km of the XME involved in the metabolism of the substrate (typically 1–5 µM). The curve is then fitted with a first order equation to calculate the half-life, taking into account the volume of the incubation and the protein or cell concentration.

In the next step in an IVIVE for liver, the in vitro intrinsic clearance is converted to the in vivo intrinsic clearance using a scaling factor of 99 × 10⁶ hepatocytes per g liver (Barter et al., 2007), 101 mg S9 protein per g liver (Ahire et al., 2022) or 32–40 mg microsomal protein per g liver (Barter et al., 2007; Hakooz et al., 2006). A similar scaling factor has been proposed for scaling in vitro skin clearance values. While data from recombinant enzymes can be integrated into a liver PBK model, this is not possible for the skin because of the paucity of proteomics data regarding specific XME abundance (pmol/mg protein) within different skin layers, as current tissue-wide databases frequently omit skin values (Basit et al., 2020). Until new ideas and concepts of how to incorporate in vitro data using appropriate scaling factors develop, PBK developers must rely on simple assumptions for XME abundance. However, for keratinocyte incubations, assuming that metabolism is effective only in the epidermis, the cell abundance in the epidermis has been proposed to be 6.02 × 10⁶ cells/cm² (Manwaring et al., 2015). For skin S9 incubations, a protein abundance of 5.68 mg/cm³ was derived, with the mean volume of the epidermis calculated using the mean thickness of epidermis multiplied by the exposure area (Ellison et al., 2025). The work of Ellison et al. is being extended as part of a current ICCS project, in which samples of human skin from varying donor demographics (ethnicity, age, sex) and areas of the body have been collected. After recording the surface area and weight of the skin, it was processed to generate skin S9 according to a defined protocol. The yield of S9 (mg/cm²) is being measured, along with enzyme activity for key Phase 1 and 2 XMEs. The XME activities are being repeated over time to examine the effects of freeze/thaw or storage time. The data will also be examined to make a preliminary evaluation of differences between sexes, ethnicities and/or ages.

When calculating hepatic clearance, equations using restricted in vivo intrinsic hepatic clearance (De Buck et al., 2007) include the unbound fraction in the plasma, the blood-to-plasma ratio and the blood flow to the liver. By contrast, when scaling skin metabolism, this type of equation is not applicable since skin metabolism is effective only during the diffusion through the skin and isn’t affected by blood flow.

The next step in building an in silico skin bioavailability model is to determine the metabolic flux, which requires precise knowledge of the chemical concentration at the enzymatic site. To this end, the time and concentration profile in skin layers exhibiting metabolic activities can be predicted using partition and diffusion coefficients of the different skin layers. In the skin model developed by Dancik et al. (2013), the skin is divided into three distinct compartments i.e., the SC, the viable epidermis and the dermis. The movement of a chemical across the skin is described by Fick's laws of diffusion and partition coefficients in the different layers. The method described by Dancik et al. (2013) was also used in a skin module for genistein and daidzein (Najjar, Lange, et al., 2024). The skin compartment was a multilayered slab, each corresponding to a skin layer (SC, epidermis, and dermis), through which the chemical diffuses into the bloodstream. The skin metabolism rate was initially predicted using in silico QSARs; however, the metabolism rate, which was extensive, was adjusted using ex vivo skin data. The final human PBK model for daidzein was used to convert the in vitro Point of Departure (PoD) to an external dose in a body lotion formulation. This PBK model was a central component of the NGRA read-across case study.

There are reports of PBK models with a skin module for topically applied chemicals without directly incorporating in vitro skin metabolism data. For example, van Osdol et al. (2024) built a PBK model for clobetasol 17-propionate (CP) based on available measured in vivo pharmacokinetics data obtained by Open Flow Microperfusion (OFM) (Bodenlenz et al., 2016). This method monitors the concentration of drug in the dermis over time after topical application in vivo. At the time of the publication, skin metabolism data for CP were unavailable, although based on the chemical structure, hydrolysis of the propionate was expected. Therefore, the pharmacokinetics information was only with respect to the parent compound, CP, and any potential metabolism was already included in the measurement.

In addition to the concentration of a chemical in a skin layer, the extent of its metabolism depends on the rate of biotransformation and length of time it spends in contact with epidermis i.e. the mean residence time (MRT). The integration of MRT information in the skin allows the calculation of the amount of metabolite(s) formed. In the PBK model developed by Ellison et al. (2025), it was assumed that metabolism occurs only in the epidermis; therefore, the MRT in the epidermis was estimated based on the partition and diffusion coefficients in the epidermis, calculated according to the method described by Dancik et al. (2013). This model was able to estimate systemic concentrations of both the parent chemical, l-menthyl-d-lactate, and its ester hydrolysis metabolites following topical application. For other chemicals, measured skin metabolism data may not be relevant to the overall first-pass effect. For example, the UV filter, oxybenzone, was metabolized to a glucuronide conjugate by skin S9; however, the half-life of the parent was very long (856 min) compared to the permeability across the skin (Golbamaki et al., 2025); therefore, it is unlikely to be metabolized as the MRT in epidermis is shorter than the rate of metabolism. Consequently, the skin module in the PBK model did not incorporate the skin S9 metabolism data (Golbamaki et al., 2025). Another example of this is a PBK model for octyl methoxycinnamate, which is extensively metabolized in the skin (Li et al., 2023). As pointed out by the authors, the lack of a robust IVIVE relationship did not allow the inclusion of skin metabolism in the PBK platform used (e.g. TCAT model from GastroPlus 9.8 (Simulations Plus). Despite this limitation, the model was validated according to available human pharmacokinetics data (Matta et al., 2020). If metabolism is efficient in other skin layers such as the dermis, sebaceous glands or hair follicles, the corresponding MRT should also be determined using data generated from relevant in vitro model (e.g., fibroblasts for the dermis).

To perform a safety assessment for a large population, ranges of parameters affecting skin metabolism such as age, gender, skin quality, and genetic and environmental factors can also be considered. To validate a PBK model, simulations should ideally be compared with in vivo clinical data. This has been performed for several UV filters (Golbamaki et al., 2025; Li et al., 2023) using data generated from FDA clinical trials of sunscreens (Matta et al., 2020); however, the availability of such data is rare for cosmetics ingredients, making this a particular challenge when using these models in NGRAs. As long as in vivo pharmacokinetics data are available to build and validate a PBK model, knowledge of skin metabolism rate is not mandatory, unless the question is related to the formation of a given metabolite. When no in vivo pharmacokinetics data are available, within the NGRA context, the robustness of the PBK model is strongly related to the in vitro model used.

The appropriate complexity and required confidence of a skin or PBK model depend on its intended purpose. Simple or exploratory models may be sufficient for screening, hypothesis generation, comparison of scenarios, or identification of influential parameters. By contrast, models used for quantitative prediction of parent or metabolite exposure, derivation of safety margins, or decision-making should be supported by physiologically relevant structure, reliable parameterization, sensitivity and uncertainty analyses, and evaluation against independent data where available. Model complexity should therefore increase only when needed to address the assessment question and reduce decision-relevant uncertainty; unnecessary complexity may introduce poorly informed parameters without improving confidence. The model purpose, assumptions, domain of applicability, data quality, and remaining uncertainties should be documented transparently, and the level of confidence should be considered when interpreting the predicted exposure.

6. Use of skin metabolism data for safety assessments

The identification and/or quantification of metabolism is relevant for safety assessment, with all NGRA workflows defining metabolism as a key step when evaluating systemic toxicity as well as local toxicity (Berggren et al., 2017). Fig. 1 provides examples of how the most appropriate skin model could be selected based on the needs of the NGRA. This can follow very different paths depending on the chemical and whether the assessment is an ab initio (perhaps only needing information on one chemical) or a read-across (RAX, which may require testing of multiple chemicals). The same type of assay may also be applied to different tiers of an NGRA e.g. a comparison of the metabolism of multiple chemicals may be used in an early Tier to help select suitable analogues (Grégoire et al., 2024) or later, when grouping chemicals with a common metabolite to the contribution of its overall systemic exposure (Najjar et al., 2025). For this reason, a comprehensive NGRA workflow describing when and how skin metabolism data could be measured and interpreted is too complex for the scope of this paper. Therefore, the figure starts with an example of the kind of data that may be needed, which leads to 1–2 simple questions to help decide the initial model to use. Follow-up steps are suggested in a second step using a different model if the first assay does not enable a clear conclusion e.g. no metabolism or clearance. It is assumed here that a sensitive analytical method is available for the test chemicals. These are suggestions and not hard and fast rules, since additional considerations may influence the choice of model, such as low solubility, high non-specific binding or cytotoxicity.

Fig. 1.

Examples of how the most appropriate skin model could be selected based on the needs of the NGRA. The examples shown assume an appropriate analytical method is available.

In addition to the model, the type of data needed, i.e. qualitative or quantitative, depends on the question being addressed. For example, if the chemical is shown to be metabolized to a reactive metabolite, then it is not necessary to quantify the metabolite (which may not be technically possible anyway) - a qualitative result is sufficient to conclude on whether has skin sensitization potential. This is reflected in the Adverse Outcome Pathway (AOP) for sensitization (OECD, 2012), which requires a covalent reaction with a protein. These reactions with proteins can occur directly with reactive chemicals (haptens) or the chemical requires activation via XMEs (pro-hapten) or non-enzymatic conversion (pre-hapten) (Bergström et al., 2007) e.g., eugenol is oxidized to aldehyde (Jäckh et al., 2012). By contrast, some sensitizers are less potent due to detoxification (e.g., p-phenylenediamine is detoxified by NAT (Nohynek et al., 2005)). At the systemic level, identifying stable biotransformation products helps safety assessors determine whether systemic toxicity is driven by the parent chemical or its metabolites, directly informing structural analog selection for read-across assessments.

For higher-tier systemic safety assessments where internal exposure predictions are required, skin metabolism data provide the necessary input for IVIVE. While integrating dermal clearance into PBK models remains an evolving field, establishing standardized experimental clearance measurements, understanding tissue storage viability, and resolving local protein binding corrections represent indispensable building blocks. When a safety assessment is based on the parent compound, the robustness of the evaluation is inextricably linked to the precision of clearance measurements. This encompasses the analytical stringency of the data, the validity of experimental controls, and the physiological relevance of the chosen in vitro system (e.g., RHE or S9 fractions). For some chemicals, uncertainties can be related to biokinetic issues (Grégoire et al., 2024; Proença et al., 2021), whereas other uncertainties are related to the in vitro model system itself (Eilstein et al., 2020), a lack of reference metabolites for quantification (Liigand et al., 2020) or the method of IVIVE (Laue et al., 2020). Ultimately, these hepatic or extra-hepatic clearance values can be integrated into a PBK structural framework to simulate plasma profiles. If the risk is driven by a metabolite, absolute quantification is the gold standard, whereby a PBK model can help to track concentrations of the metabolite(s) for comparison with bioactivity PoDs. In the absence of quantified amounts of metabolites, assessments rely on qualitative observations, which introduce significant uncertainty into the safety margin.

For specific endpoint toxicity assays, information on XMEs present in skin models can help to interpret the outcome of the in vitro assay in which they are used. For example, the outcome of the skin genotoxicity assay, Reconstructed Skin Micronucleus (RSMN) assay, has been linked to the presence and induction of various CYPs known to be present in the model (Hewitt et al., 2013). By contrast chemicals which are positive in standard in vitro mutagenicity and genotoxicity assays supplemented with liver S9 and Phase 1 cofactors are negative in the RSMN assay due to the presence of detoxifying Phase 2 enzymes in the skin models, which are considered to be more relevant for the in vivo situation (Petrenci et al., 2025). In line with the in vitro skin assays, the chemicals tested by Petrenci et al. (2025) were also negative for in vivo genotoxicity assays (which of course, can no longer be used in the safety assessment of cosmetic ingredients).

Table 4 lists examples of NGRAs in which skin metabolism data were used for the assessment. Skin metabolism can be integrated into safety assessments in different ways. Read-across assessments remain fundamentally centered on the qualitative comparative analysis of structural analogues to the target chemical; therefore, establishment of an IVIVE relationship is generally not a prerequisite. Conversely, transitions toward Quantitative Risk Assessment (QRA) necessitate robust quantitative data for the parent and/or specific metabolites. For these deterministic frameworks, a validated IVIVE correlation is indispensable for scaling intrinsic clearance to systemic or local parameters.

Table 4.

Examples of NGRA case studies on cosmetic ingredients using skin metabolism as input.

Chemical	Skin model used	Context of data usage	Outcome from use of data in the assessment	Reference
L-menthyl d-lactate	Skin S9	Margin of safety based on known metabolite	Risk assessment of l-menthyl d-lactate used skin S9 intrinsic clearance (among other aspects) to estimate internal exposure of its hydrolysis metabolites	Ellison et al. (2025)
Daidzein	ex vivo human skin explants	Read across	The extent of skin metabolism in ex vivo human skin explants was used to select the analogue (genistein) and to adjust systemic exposure due to first-pass metabolism in the skin	Najjar et al., 2024b, Najjar et al., 2024c
Propyl paraben	Skin S9	Read across	The extent of skin metabolism in skin S9 was used to select the most appropriate analog	Ouedraogo et al. (2022)
Benzyl salicylate	Ex vivo explants	Ab initio	Intrinsic clearance in the skin was included in PBK model to track parent chemical and to identify the most relevant metabolite (salicylic acid) used for safety assessment	Ebmeyer et al. (2024)
Aromatic amines	HaCat cells	Skin sensitization	Detoxification of aromatic amine via NAT-mediated metabolism	Manwaring et al. (2015)

7. Conclusions

Safety assessments of topically applied cosmetic products should consider the level of skin absorption of the ingredient according to the exposure scenario and the potential contribution of skin metabolism. Skin metabolism data can provide information on the local and systemic exposure to the parent compound or to a metabolite. It can also support conclusions on biological effects via bioactivation or detoxification.

There is a range of available in vitro skin metabolism models with varying complexities. Whilst metabolism occurs primarily in the epidermis, the contribution of other skin compartments cannot be ruled out; therefore, knowledge of the location of the XMEs likely to be involved in the metabolism of the test chemical will help to select the most appropriate skin model. Subcellular fractions of RHEs or keratinocyte cell models (NHEK, HaCaT) primarily reflect epidermal activity. If biotransformation occurs within the dermis, sebaceous glands, or the pilosebaceous unit, these localized activities will be missed by conventional epidermal assays. In this case, human ex vivo skin explants or full-thickness RHE models are more relevant. The selection skin model will also depend on the aim of the assay e.g. qualitative metabolite identification for analogue selection or intrinsic clearance for IVIVE. Each model has advantages and disadvantages, which should also be considered when selecting a model and when conducting appropriate assay controls to account for technical challenges such as non-specific binding. Even when the most relevant model is used, the safety assessment also depends on the quality of the data. This includes the use and validation of appropriate controls and a suitable and sufficiently sensitive analytical method to ensure the best possible accuracy of the information provided.

Once generated, data need to be correctly interpreted and extrapolated to in vivo using appropriate equations and scaling factors to provide a relevant and robust indication of first-pass dermal metabolism as a chemical diffuses across the skin layers. This can be influenced by the rate of permeation, access to XMEs within different skin layers, extrinsic and intrinsic factors, in addition to the quality of the data. Challenges and open questions remain regarding standardized metabolism assays and related guidelines, as well as the development of in silico models to predict skin metabolites and skin bioavailability for integration into PBK models.

In conclusion, skin metabolism is an essential component of dermal safety evaluations for cosmetic ingredients. A universally applicable framework is not in place yet, and therefore the choice of a test system and the interpretation of dermal metabolism data will need to be justified on a case-by-case basis for consideration in a safety assessment. Qualitative metabolic profiling provides immediate value by identifying local bioactivated chemicals (pro-haptens) and general toxic metabolites without requiring complex kinetic modeling. The generation of robust, reproducible in vitro skin clearance data and the establishment of relevant scaling factors will be indispensable for future PBK models and systemic IVIVEs for dermally exposed chemicals. Continued research into the localization of XMEs within the skin, test model protocols and their selection, and biophysical binding parameters will increase confidence in animal-free human safety assessments.

Funding statement

This work was supported by the International Collaboration on Cosmetics Safety (ICCS) (www.iccs-cosmetics.org).

CRediT authorship contribution statement

Nazanin Golbamaki: Writing – review & editing, Supervision, Funding acquisition, Conceptualization. Sebastien Gregoire: Writing – review & editing, Writing – original draft, Visualization, Data curation, Conceptualization. Nicola J. Hewitt: Writing – review & editing. Daniela Lange: Writing – review & editing. Ans Punt: Writing – review & editing. Sandrine Spriggs: Writing – review & editing. Sara Sepehri: Writing – review & editing. Nadia Heramvand: Writing – review & editing. Andreea Cuciureanu: Writing – review & editing, Project administration. Dave Allen: Writing – review & editing, Supervision, Project administration. Abdulkarim Najjar: Writing – review & editing, Supervision, Funding acquisition, Conceptualization.

Declaration of competing interest

The authors declare the following financial interests/personal relationships which may be considered as potential competing interests:

Nicola J. Hewitt reports financial support was provided by International Collaboration on Cosmetics Safety. Sebastien Gregoire reports financial support was provided by International Collaboration on Cosmetics Safety. If there are other authors, they declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.

Data availability

No data was used for the research described in the article.