Comparison between endocrine activity assessed using ToxCast/Tox21 database and human plasma concentration of sunscreen active ingredients/UV filters

Abstract

Sunscreen products are composed of ultraviolet (UV) filters and formulated to reduce exposure to sunlight thereby lessening skin damage. Concerns have been raised regarding the toxicity and potential endocrine disrupting (ED) effects of UV filters. The ToxCast/Tox21 program, that is, CompTox, is a high-throughput in vitro screening database of chemicals that identify adverse outcome pathways, key events, and ED potential of chemicals. Using the ToxCast/Tox21 database, octisalate, homosalate, octocrylene, oxybenzone, octinoxate, and avobenzone, 6 commonly used organic UV filters, were found to have been evaluated. These UV filters showed low potency in these bioassays with most activity detected above the range of the cytotoxic burst. The pathways that were most affected were the cell cycle and the nuclear receptor pathways. Most activity was observed in liver and kidney-based bioassays. These organic filters and their metabolites showed relatively weak ED activity when tested in bioassays measuring estrogen receptor (ER), androgen receptor (AR), thyroid receptor, and steroidogenesis activity. Except for oxybenzone, all activity in the endocrine assays occurred at concentrations greater than the cytotoxic burst. Moreover, except for oxybenzone, plasma concentrations (C_max) measured in humans were at least 100× lower than bioactive (AC₅₀/ACC) concentrations that produced a response in ToxCast/Tox21 assays. These data are consistent with in vivo animal/human studies showing weak or negligible endocrine activity. In sum, when considered as part of a weight-of-evidence assessment and compared with measured plasma concentrations, the results show these organic UV filters have low intrinsic biological activity and risk of toxicity including endocrine disruption in humans.

Solar radiation is comprised of ultraviolet radiation (UVR) of wavelengths between 100 and 400 nm (Diffey, 2002). In dermatological photobiology, these wavelengths are subdivided into UVC (100–290 nm), UVB (290–320 nm), and UVA (320–400 nm). The UVR that reaches the earth is composed of UVB and UVA; UVC is filtered by the ozone layer and, as such, not relevant when considering human exposure to solar UVR. The acute skin effects of UVR exposure are erythema, tanning, and immunosuppression, whereas chronic exposure can lead to photoaging and skin cancer (Lautenschlager et al., 2007; Matsumura and Ananthaswamy, 2004; Taylor and Sober, 1996).

Sunscreen products are composed of UV filters and formulated for different purposes, for example, recreational use, daily facial application, and sunscreen with insect repellant. The UV filters in such products absorb photon energy thereby reducing the dose of solar UVR and lessening skin damage (Guan et al., 2021; Lautenschlager et al., 2007; Nash, 2006). The current focus in the development of broad-spectrum sunscreens is spectral uniformity protecting equally against both high-energy UVB and UVA (Diffey and Brown, 2012; Dudley et al., 2021; Surber and Osterwalder, 2021).

Sunscreens and the UV filters used in such products have a long history of use spanning a period of several decades. In this time, there have been numerous in vitro, preclinical (animal), and clinical studies demonstrating the effectiveness of sunscreen active ingredients alone and combined in preventing UV-induced skin damage (Gasparro et al., 1998; Olsen et al., 2017; Young et al., 2017). Among the most notable studies are the population-based prospective epidemiological studies conducted by Green and coworkers showing the benefits of using a sunscreen product containing 2 UV filters, that is, avobenzone and octinoxate, in preventing the development of skin cancer and photoaging (Green et al., 1999; Iannacone et al., 2014; van der Pols et al., 2006). However, there have also been long running controversies including the effects on vitamin D synthesis as well as questions regarding endocrine effects of individual UV filters, particularly oxybenzone (Mustieles et al., 2023; Passeron et al., 2019). Recently, the human safety of sunscreen active ingredients has come under increased scrutiny in several regulatory jurisdictions including the United States and Europe. In 2019, the Food and Drug Administration (FDA) reclassified 12 organic UV filters, including the most commonly used ones namely octisalate, homosalate, octocrylene, oxybenzone, octinoxate, and avobenzone, citing a need for additional safety data (FDA, 2019). The European Commission has included several UV filters in their “list of endocrine disruptors” requesting additional evaluations (Endocrine disruptors | European Commission [europa.eu]). These concerns were amplified following publication of human pharmacokinetic studies conducted by researchers at FDA showing organic UV filters used as sunscreen active ingredients in different formulations can be detected in the plasma following topical application (Matta et al., 2019, 2020). The clinical relevance of the UV filter absorbed systematically has not been investigated, particularly pertaining to toxicological effects including endocrine disruption (ED).

The field of toxicology is making increased use of in vitro assays to determine the toxicological potential of chemicals in vitro by identifying adverse outcomes pathways (AOP), key events (KE), and ED (Knapen et al., 2015; Simon et al., 2014; Villeneuve et al., 2014). The CompTox dashboard is a freely accessible database created and maintained by the U.S. Environmental Protection Agency (EPA). The dashboard contains information for over 1.2 million chemicals from Tox21 and EPA’s Toxicity Forecaster (ToxCast) high-throughput screening (HTS) program. The ToxCast database includes in vitro bioactivity data from over 1800 chemicals used in a variety of settings including pharmaceutical, pesticide, industrial, cosmetic, and personal care tested in hundreds of assays. The Endocrine Disruption Screening Program (EDSP21) is a subset of the ToxCast data that focuses on ED (U.S. EPA). The data from the in vitro bioassays can be extrapolated to estimate the human dose at which a chemical of interest influences biological pathway(s) (Judson et al., 2011).

The current work investigates the endocrine activity of 6 of the most commonly used organic UV filters worldwide, that is, octisalate, homosalate, octocrylene, oxybenzone, octinoxate, and avobenzone (Chaiyabutr et al., 2021; Wang et al., 2013) using in vitro bioassay data from ToxCast/Tox21 database. The concentration needed to elicit an in vitro response was compared with plasma concentrations, that is, C_max, measured following topical application under maximal usage conditions (Matta et al., 2019, 2020). Additionally, we evaluated the importance of careful curation of high-throughput in vitro data from screening programs when using such results for toxicological and/or clinical assessments.

Materials and methods

 

Chemical library screened ToxCast/Tox21 library

The ToxCast/Tox21 database is a program organized and maintained by several agencies including the U.S. National Toxicological Program (NTP) and the U.S. EPA. EDSP21 is an in vitro HTS program publicly available within Tox21 that is focused on measures of the endocrine system (Attene-Ramos et al., 2013). The EDSP21 screening program is designed to identify and prioritize estrogen, androgen, and thyroid disruptors as well as steroidogenesis, the so-called EATS. At the time of this study, the total number of chemicals in the Tox21 database was greater than 10 000 (Toxicity Forecasting | Safer Chemicals Research | U.S. EPA). Compound classes screened in EDSP21 include pesticides, pharmaceutical/over-the-counter (OTC) actives such as UV filters, industrial, food additives, consumer goods and environmental chemicals, selected on the basis of environmental hazard or exposure concerns and amenable for in vitro screening, for example, volatility, log P, and solubility (Lynch et al., 2018). Each chemical in the screen undergoes a rigorous quality check (QC) to ensure the reliability of the data generated in the test screening (Richard et al., 2021). This QC check is as part of the Tox21 project, and the QC results are made available to the public (https://tripod.nih.gov/tox21/samples).

Organic UV filters screened using the ToxCast/Tox21 database

The UV filters, octisalate (2-ethylhexyl salicylate), homosalate (3,3,5-trimethylcyclohexyl 2-hydroxybenzoate), octocrylene (2-ethylhexyl 2-cyano-3,3-diphenylprop-2-enoate), oxybenzone (2-hydroxy-4-methoxybenzophenone), octinoxate (2-ethylhexyl methoxycinnamate), and avobenzone (4-tert-butyl-4′-methoxydibenzoylmethane), were evaluated for their toxicity/bioactivity and AOPs using results in the Tox21/EDSP21 database. These 6 chemicals are common organic UV filters used in sunscreen products and as photostabilizers for a variety of products. The organic UV filters used in this study as well as their respective molecular weights and structures are listed in Table 1.

Table 1.

Chemical identifiers and structures

UV Filter	CAS No.	Formula	Molecular Weight (g/mol)	Chemical Structure
Octisalate	118-60-5	C15H22O3	250.3
Homosalate	118-56-9	C16H22O3	262.3
Octocrylene	6197-30-4	C24H27NO2	361.5
Oxybenzone	131-57-7	C14H12O3	228.2
Octinoxate	5466-77-3	C18H26O3	290.4
Avobenzone	70356-09-1	C20H22O3	310.4

Structures from ChemSpider. https://www.chemspider.com/.

The identity of plasma and urinary metabolites of the 6 UV filters were taken from literature sources based on in vitro, animal, and human studies and evaluated for activity in ToxCast/Tox21 database (Bury et al., 2019a,b,c; Guesmi et al., 2020; Hiller et al., 2019a,b; Huang et al., 2019; Klotz et al., 2019; Okereke et al., 1993; Tarazona et al., 2013). Specifically, the metabolites identified in the references and assessed were: (1) salicylic acid and 2-ethylhexanol for octisalate, (2) salicylic acid and 3,3,5-trimethylcyclohexanol for homosalate, (3) ethyl 2-cyano-3,3-diphenylacrylate the analog of 2-cyano-3,3-diphenylacrylic acid for octocrylene, (4) 2,4-dihydroxybenzophenone (DHB), 2,2′-dihydroxy-4-methoxybenzophenone (DHMB), and 2,3,4-trihydroxbenzophenone (THB) for oxybenzone, and (5) 4-methoxycinnamic acid and 2-ethylhexanol for octinoxate. Avobenzone metabolites were not found in ToxCast/Tox21 database and therefore, not assessed.

Screening assays

The bioassays used in this study were obtained from the version 3.3 of the ToxCast data. The assays that were most sensitive, that is, lowest ACC/AC50, to the UV filters are listed in Table 2 and a summary of each is provided. A complete list of active assays in which the UV filters were tested is presented in Supplementary Tables 1 (EDSP21) and 2 (all bioassays). The assays that are part of the EDSP battery were selected for comparison with human plasma (C_max) concentrations of the 6 UV filters obtained in Matta et al. (2019, 2020) and are briefly described.

Table 2.

List of EDSP assays used to assess activity of UV filters for comparison to human plasma concentrations

Assay Name	EDSP Assay Type	Assay Endpoint ID (invitrodb: aeid)	Cell Line	Tissue	Organism	Assay Mode
ATG_ERE_CIS_up	Estrogen	75	HepG2	Liver	Human	Agonist
TOX21_ERa_LUC_VM7_Agonist		788	VM7	Breast	Human	Agonist
TOX21_AR_LUC_MDAKB2_Antagonist_0.5nM_R1881	Androgen	1816	MDA-kb2	Breast	Human	Antagonist
ATG_AR_TRANS_up		115	HepG2	Liver	Human	Agonist
TOX21_TR_LUC_GH3_Antagonist	Thyroid	804	GH3	Pituitary	Rat	Antagonist
NCCT_TPO_AUR_dn		1508	Tissue-based cell free	Thyroid	Rat	Antagonist

ATG_ERE_CIS_up

ATG_ERE_CIS_up is a human liver cell line-based system that utilizes HepG2 cells (variant HG19 with enhanced xenobiotic metabolism) to evaluate transcription factor (TF) activity in response to chemical perturbations at the estrogen response element (ERE). The HepG2 is a cell culture derived from a liver hepatoma lobectomy which has been cloned and transfected with multiple reporter transcription unit (MRTU) constructs that are regulated by a cis promoter. The cells are treated with the test chemical for 24 h in 24-well plates. Increased transcription activity at the ERE is measured by fluorescent intensity produced by the transcribed mRNA. 17b-Estradiol is used as a positive control in the assay (Martin et al., 2010; Romanov et al., 2008).

TOX21_AR_LUC_MDAKB2_Antagonist

Tox21_AR_LUC_MDAKB2_Antagonist is a human breast cancer cell line-based assay used to identify compounds that will inhibit androgen receptor (AR) signaling. The MDA-kb2 AR-luc cell line was derived from a metastatic breast carcinoma cell line MDA-MB-453 that has high levels of ARs. The cell line utilizes a luciferase reporter gene construct downstream for the endogenous AR. Cytotoxicity of the test chemical was tested in parallel using CellTiter-Fluor assay. The MDA-kb2 AR-luc cells are exposed to the test compounds for 24 h in 1536-well plates in the presence of 0.5nM_R1881 (agonist) and measured for luminescence from AR gene expression. Nilutamide was used as a positive control in the assay (Wilson et al., 2002).

TOX21_ERa_LUC_VM7_Agonist

ToX21_ERa_LUC_VM7_Agonist is a cell-based assay utilizing the human breast tissue cell line, VM7 to measure estrogen agonist activity. VM7-Luc-4E2 are cells derived from a breast carcinoma cell line transfected with 3 EREs upstream of a luciferase reporter. The immortalized cell line endogenously expresses receptors for estrogen (α and β) and progesterone. The cells are plated in 1536-well plates and treated with the test compound for 24 h. Increased estrogen agonist activity is measured by increased luminescence. 17b-Estradiol is used as a positive control in the assay (Huang et al., 2014, 2016).

TOX21_TR_LUC_GH3_Antagonist

ToX21_TR_LUC_GH3_Antagonist assay is based on the rat pituitary tumor cell line, GH3, and is used to investigate thyroid receptor antagonist activity. GH3 cells express both isoforms of thyroid hormone (α and β) at high amounts. The cell line has routinely been used to investigate the thyroid disrupting compounds (Gutleb et al., 2005). To measure the TR antagonist activity, the loss of signal was measured against thyroid agonist (T3). The cell viability was tested in parallel with the thyroid receptor activity of the Tox21 chemical library compounds. GH3.TRE-Luc cells were exposed to the test compounds in 1536-well plates in the presence of T3 agonist. Cell viability was measured using the CellTiter-Fluor system. The thyroid receptor antagonistic activity was measured through decreased luminescence (Freitas et al., 2011).

ATG_AR_TRANS_up

ATG_AR_TRANS_up assay utilizes human the HepG2 liver cell line to measure the gain-of-signal function as related to the AR. The HepG2 is transfected with multiple reporter transcription unit (MRTU) constructs that are regulated by a trans promoter. Each RTU expresses a GAL4-AR that regulates the transcription of a reporter sequence. The cells are treated with the test compound for 24 h in a 24-well plate. Increased transcription activity is measured by increased fluorescent intensity. 6a-Fluorotestosterone used as a positive control in the assay (Martin et al., 2010).

NCCT_TPO_AUR_dn

NCCT_TPO_AUR_dn is a single readout assay that utilizes microsomes from rat thyroid gland in a tissue-based cell-free assay. The assay measures signal loss by measuring thyroperoxidase (TPO) activity as detected with fluorescence intensity signals by fluorescence using the Amplex UltraRed technology as a form of enzyme activity reporter. The measurements are taken every 30 min in a 384-well plate. Methimazole is used as a key positive control (Paul et al., 2014).

Comparing blood plasma concentrations with AC50/ACC

Plasma concentrations of the 6 UV filters were obtained from randomized clinical studies following a Maximal Usage Trial (MUsT) study design (Matta et al., 2019, 2020). Briefly, the study participants were healthy individuals (both men and women) randomized into 4 different groups treated with 4 different sunscreen product formulations containing a mixture of UV filters. The sunscreen formulations were applied at 2 mg/cm² to 75% of the total body surface area, 4 times per day, that is, once every 2 h (Matta et al. 2019), and once during the first day and 4 times for 3 days (Matta et al. 2020). Blood samples were collected at various timepoints to determine plasma concentration of the sunscreen active ingredients, reported in ng/ml for both single application and maximal use scenarios. Plasma concentration from the studies were converted from ng/ml to µg/ml and then to µM and are presented in Table 3.

Table 3.

The mean plasma concentration (C_max) of the sunscreen active ingredients following single and repeat application as reported in Matta et al. (2019, 2020)

	Plasma concentration, µg/ml		Plasma concentration, µM
UV Filter (Conc. %) [Formulation]	Single Use	Maximal Use	Single Use	Maximal Use
Octisalate (5%) [nonaerosol spray]	0.0014	0.0058	0.0056	0.0232
Homosalate (15%) [aerosol spray]	0.0076	0.0231	0.0290	0.0881
Octocrylene (6%) [lotion]	0.0015	0.0078	0.0041	0.0216
Oxybenzone (4%) [lotion]	0.0942	0.2581	0.4128	1.1310
Octinoxate (7.5%) [nonaerosol spray]	0.0020	0.0079	0.0069	0.0272
Avobenzone (3%) [LOTION]	0.0016	0.0071	0.0052	0.0229

The range of plasma concentrations and coefficient of variation (%) for the UV filters reported by Matta et al. were as follows: octisalate (2.3–16.7 ng/ml, 77.4%), homosalate (8.7–67.7 ng/ml, 68%), octocrylene (2.6–38.7 ng/ml, 87.1%), oxybenzone (131.3–498.1 ng/ml, 53%), octinoxate (2.6–30.6 ng/ml, 86.5%), and avobenzone (2.9–28 ng/ml, 73.9%). The highest concentration of the sunscreen active ingredient is indicated in brackets () and type of formulation is in square brackets []. For purposes of comparison, ng/ml values reported by Matta et al. were converted to µg/ml and µM.

Data analysis

HTS in vitro assay results for each UV filter were evaluated from the ToxCast v3.3 MySQL database which contains the results of the ToxCast processing library (TCPL) analysis. Two activity measures were used for this analysis: (1) the ACC—Activity Concentration Cutoff, which is the concentration where the winning model meets a TCPL defined cutoff threshold, and (2) the AC50—Activity Concentration at 50% of maximal activity, which is the concentration where the winning model is at 50% of the maximum response. A “hit”, that is, response in an assay, was determined by analysis of the TCPL and used to assign an active/inactive status for each chemical/assay pair. For chemicals that were tested multiple times in the same assay, an overall “hit” was assigned if 50% or more of the assays were active. The assay with the lowest AC50 was chosen as the representative for the chemical/assay pair. If less than 50% of the assays were active, the chemical/assay pair was considered inactive. A summary plot was built for each active chemical/assay pair that shows the winning TCPL model fit layered over the concentration/response data.

The distribution of ACCs and AC50s for active “hits” across all chemicals in ToxCast was assessed for each assay in which one of the UV filters was active. Each chemical/assay pair was compared with the appropriate distribution to calculate a percentile rank of the ACC/AC50 relative to all other actives in ToxCast database. This allowed for an easy way to gauge the relative potency of a given chemical/assay pair compared with other chemicals tested in the ToxCast in the same assay.

The TCPL analysis calculates a cytotoxicity threshold for each chemical tested in the ToxCast program which was included in our analysis. This threshold is calculated by using the results of 86 cytotoxicity-specific assays. If more than 5% of these assays are active for a chemical, a cytotoxicity threshold is defined for the chemical by taking the median AC50 of the active assays and subtracting 3 standard deviations to get a lower limit (Judson et al., 2016). For chemicals where less than 5% of the cytotoxicity assays are hits, a cytotoxicity threshold of >1000 µM is assigned. The cytotoxicity threshold was included as means to flag cases where the ACC or AC50 for a chemical/assay pair is above the threshold, indicating the possibility that non-specific cytotoxicity effects are interfering with/driving the results and caution in interpreting the assay is warranted.

The concentration-response curves from each assay for the 6 UV filters were individually evaluated by visual inspection resulting in exclusion of assays with multiple “flags” associated with borderline activity, low efficacy, curve overfitting, noise, etc. Curves that did not have a clear concentration-response in the bioassays were also excluded. For this analysis, analytical quality control cautions were evaluated and considered acceptable if there was a clear concentration-response curve in the bioassays in response to the UV filters. The list of endocrine bioassays for each UV filter after visual analysis is shown in Supplementary Table 1.

In vitro bioactivity (ACC and AC50) for the UV filters that were obtained from ToxCast/Tox21 (reported as µM) were compared with the plasma concentration taken from Matta et al. (2019, 2020) (Table 3).

Results

 

ToxCast/Tox21 (EDSP21) data analysis

The UV filters, octisalate, homosalate, octocrylene, oxybenzone, octinoxate, and avobenzone, were identified in the V3.3 of the ToxCast Database following a search of CompTox. Each UV filter was tested in a minimum of 235 assays and less than 19.6% of the assays were reported as active (Table 4). All the assay “hits” with the respective AC50 and ACC for each of the UV filters are represented in box and whisker plots and compared with the maximal plasma levels from Matta et al. (2019, 2020) as presented in Supplementary Figures 1 and 2. The full list of assay “hits” are presented in Supplementary Table 2.

Table 4.

UV filters tested in ToxCast/Tox21 and accessible through the online CompTox database

	CompTox
UV Filter	Tested	Active	Active %
Octisalate	577	43	7.5
Homosalate	453	42	9.3
Octocrylene	235	40	17.0
Oxybenzone	633	62	9.8
Octinoxate	241	5	2.1
Avobenzone	235	46	19.6

The number of assays in which the sunscreen actives were tested, and the active assays are reported. ToxCast/Tox21 is a living database with the number of assays and chemicals updated over time. All active bioassay data are presented in Supplementary Table 2.

All the UV filters were tested in the multiple endocrine assays as determined from evaluation of the EDSP21 dataset, which is a subset of the ToxCast/Tox21 data. The endocrine endpoints evaluated in EDSP21 included estrogen receptor (ER), AR, thyroid hormone, and steroidogenesis activity. The UV filters were active in only a subset of endocrine assays (Table 5). Following careful inspection of each concentration-response curve, the number of active assays for each UV filter was reduced, and, for octinoxate, no endocrine activity was seen (Supplementary Figs. 3 and 4).

Table 5.

Organic UV filters EDSP21 assays

	Estrogen			Androgen			Thyroid			Steroidogenesis
UV Filter	Tested	Active	Curated	Tested	Active	Curated	Tested	Active	Curated	Tested	Active	Curated
Octisalate	8	3	3	9	0	0	8	0	0	2	0	0
Homosalate	18	9	7	14	5	4	9	2	1	26	2	0
Octocrylene	6	1	1	8	2	2	6	1	1	2	1	1
Oxybenzone	18	10	7	14	2	2	12	2	0	26	3	3
Octinoxate	6	0	0	8	0	0	9	1	0	2	0	0
Avobenzone	6	3	2	8	1	1	6	2	1	2	0	0

All 6 sunscreen active ingredients were tested in the EDSP21 dataset. The endpoints tested for each sunscreen include estrogen receptor (ER), androgen receptor (AR), thyroid hormone activity, and steroidogenesis. The table presents the number of assays in which the sunscreen active ingredients were tested, those that were “active,” and the number of assays left following curation.

Tissues and signaling pathways

The endocrine endpoints (modalities), target family, and target family sub-type that were potentially affected by the UV filters were evaluated and presented in Figure 1. Endpoints, target family, and target family subtype are represented as a percentage of the total endocrine assays with confirmed activity. There was varied endocrine activity across the UV filters with the main endocrine endpoint affected being the ER, except for octocrylene (Figure 1A). As seen in Figure 1B, nuclear receptor target family was affected by octisalate, homosalate, and avobenzone as would be expected because they only had activity in ER, AR, and thyroid assays which are all nuclear receptor mediated (Dahlman-Wright et al., 2006; Flamant et al., 2006; Lu et al., 2006). Finally, as illustrated in Figure 1C, oxybenzone and octocrylene showed activity in nuclear receptor in addition to steroidal and CYP target family pathways. The steroidal and CYP pathway activity is in response to effects on steroidogenesis (Chakraborty et al., 2021). The Target family subtype predominantly affected by the UV filters was steroidal activity.

Figure 1.

The endocrine endpoints (modalities), target family, and target family subtype affected by the UV filters. A, The percentage of modalities that have biological activity when UV filters are tested in endocrine assays. B, Intended target families that are affected by the UV filters. C, Intended target subfamilies that are affected by the UV filters. There was no endocrine activity for octinoxate and therefore it is not shown.

Comparison of AC50/ACC from ToxCast with plasma C_max from Matta et al

The in vitro endocrine activity data were reported as both the ACC and AC50 and for illustrative purposes, the most sensitive assays for each UV filter are presented in Table 6. The in vitro bioactivity data were used for comparing the measured plasma C_max (unbound or free) for both single and repeat application following maximal exposure conditions as reported by Matta et al. (2019, 2020). The UV filter concentrations needed to produce a response in the endocrine assays were greater than the highest plasma concentrations of UV filters measured under MUsT conditions. Octisalate, homosalate, octocrylene, and avobenzone have AC50 values above their respective “cytotoxic burst,” the concentration reflective of general cellular toxicity. Oxybenzone has an AC50 below cytotoxicity, however, a cytotoxic burst value of 1000 µM is the highest concentration reported in the assays (Table 6). Octinoxate does not have a suitable AC50/ACC for comparison and, therefore, is not presented.

Table 6.

UV filter AC50/ACC values compared to human plasma concentrations following single or maximal usage conditions (13–16 applications)

					Matta et al. (2019, 2020) (µM)
UV Filter	Assay (Most Sensitive)	Cytotoxic burst (µM)	AC50 (µM)	ACC (µM)	Single Application	MUsT (13 to 16)
Octisalate	TOX21_ERa_LUC_VM7_Agonist	5.86	16.48	17.64	0.0056	0.023
Homosalate	ATG_ERE_CIS_up	12.44	14.20	8.86	0.029	0.088
Octocrylene	TOX21_AR_LUC_MDAKB2_Antagonist_0.5nM_R1881	9.01	12.71	6.28	0.0041	0.022
Oxybenzone	ATG_ERE_CIS_up	1000.00	8.28	4.92	0.41	1.13
Avobenzone	TOX21_TR_LUC_GH3_Antagonist	8.28	35.04	26.02	0.0051	0.023

The plasma concentration of the UV filters are below levels that elicit endocrine responses in the most sensitive assays. Octinoxate did not have an A50/ACC value after further evaluation.

The difference between the in vitro endocrine activity for the UV filters and the human plasma concentrations increased after examination of each assay and removal of the active “hits,” that is, curation, due to borderline activity, low efficacy, curve overfitting, noise, etc. Assays that did not have a clear concentration-response curve were also excluded. As a result, octinoxate, which had one overall active hit, was deemed not to have any endocrine activity in the assays after the data was curated. This highlights the importance of careful examination of data from the ToxCast/Tox21 database to ensure confidence in the biological relevance and accuracy of the in vitro to in vivo extrapolation. All active assays for each UV filter compared with the plasma C_max before and after curation are summarized in Figures 2A and 2B and Supplementary Figures 3 and 4.

Figure 2.

Box and whisker plots of in vitro activity for each UV filter compared with the plasma concentration, C_max, of each sunscreen active taken from Matta et al. (2019, 2020) (represented by black dots). A, Distribution of active assays for each UV filter before curation of the data. B, Distribution of the active assays for each UV filter after curation of the data. Octinoxate did not have any active assay results after curation.

To further illustrate this point, for avobenzone, 6 endocrine active assays were identified (Figure 3A). After careful examination, as described in the Materials and methods, the number of assays decreased to 4 active endocrine assays (Figure 3B). Among these 4 assays, the lowest concentration of avobenzone that produced a response was 26.016 µM (ACC) in 804 TOX21_TR_LUC_GH3. All active assays were above the cytotoxic burst indicating general cell toxicity rather than specific activity on endocrine pathways. The concentrations needed to trigger an endocrine activity were also significantly higher than the plasma C_max of 0.0229 µM from Matta et al. (2019, 2020). Additionally, avobenzone activity in assay 804 TOX21_TR_LUC_GH3 was at the 59th percentile indicating low potency, since the higher the percentile, the lower the potency in the assay.

Figure 3.

Box and whisker plots of in vitro activity for avobenzone before and after data curation compared with the plasma concentration, C_max, taken from Matta et al. (2019, 2020) (represented by black dash line). A, Endocrine activity of avobenzone in all assays (total of 6). B, Endocrine activity of avobenzone in active assays after curation (total of 4).

Evaluation of sunscreen active ingredient metabolites

The plasma and/or urinary metabolites of 5 of the 6 UV filters were evaluated for endocrine activity as presented in Table 7. Metabolites for avobenzone or structurally related analogs have not been evaluated in ToxCast/Tox 21 and therefore were not included in this analysis. One of the technical challenges of ToxCast/Tox21 data is the lack of physiologically relevant metabolic competence in many of the cell-based and cell-free assays (Thomas et al., 2018). This prevents holistic analysis of parent compounds in such bioassays. As such, based on existing in vitro, animal, and human data, plasma and/or urinary metabolites of the sunscreen active ingredients were found to have been tested in the ToxCast/Tox21 bioassays. We focused on the main metabolites, that is, those that could be over 10% of the parent compound in human plasma (FDA, 2020). Because the metabolites were not measured in the work by Matta et al., we did not compare AC50/ACC with human plasma concentrations.

Table 7.

UV filter metabolite biological activity in endocrine assays compared with the plasma concentration of the parent sunscreen

UV Filter	Metabolite Name	Assay (Most Sensitive)	AC50, µM	ACC, µM	Cytotoxic Burst, µM
Octisalate	Salicylic acid	ATG_AR_TRANS_up	1.003	1.093	1000.0
	2-Ethylhexanol	—	No activity	No activity	1000.0
Homosalate	Salicylic acid	ATG_AR_TRANS_up	1.003	1.093	1000.0
	3,3,5-Trimethylcyclohexanol	—	No activity	No activity	1000.0
Octocrylene	2-cyano-3,3-diphenylacrylic acid (CPAA)	No information	No information	No information	No information
	Ethyl 2-cyano-3,3-diphenylacrylatea	TOX21_AR_LUC_MDAKB2_Antagonist_0.5nM_R1881	6.650	4.900	11.08
Oxybenzone	2,4-dihydroxybenzophenone (DHB)	NCCT_TPO_AUR_dn	4.191	0.275	20.01
	2,2′-dihydroxy-4-methoxybenzophenone (DHMB)	TOX21_AR_LUC_MDAKB2_Antagonist_0.5nM_R1881	12.975	7.984	8.78
	2,3,4-Trihydroxbenzophenone (THB)	NCCT_TPO_AUR_dn	1.838	0.322	15.77
Octinoxate	4‐methoxycinnamic acid (MCA)	—	No activity	No activity	1000.0
	2-Ethylhexanol	—	No activity	No activity	1000.0

The cytotoxic burst concentration of each of the UV filters and metabolites is included.

^aEthyl 2-cyano-3,3-diphenylacrylate is a CPAA analog.

Octisalate is metabolized to salicylic acid and 2-ethylhexanol and detectable in urine. Other minor metabolites (less than 10%) including 2-ethyl-5-hydroxyhexyl 2-hydroxybenzoate (5OH-EHS), 2-ethyl-5-oxohexyl 2-hydroxybenzoate (5oxo-EHS), and 5-(((2-hydroxybenzoyl)oxy)methyl)heptanoic acid (5cx-EHS) are further metabolized to salicylic acid (Bury et al., 2019a,b). Salicylic acid had an AC50 value of 1.003 µM and an ACC value of 1.093 µM in assay ATG_AR_TRANS_up. 2-Ethylhexanol had no activity in the EDSP21 ED assays.

Homosalate is metabolized primarily to 3,3,5-Trimethylcyclohexanol and salicylic acid in human and rat liver microsomes (Guesmi et al., 2020). The 3,3,5-trimethylcyclohexanol was not active in any of the ED assays. As stated above for octisalate, salicylic acid had an AC50 value of 1.003 µM and an ACC value of 1.093 µM in assay ATG_AR_TRANS_up.

Octinoxate is metabolized to 4‐methoxycinnamic acid (MCA) and 2-ethylhexanol (Huang et al., 2019). MCA was identified as a urinary metabolite. Neither 4‐methoxycinnamic acid (MCA) nor 2-ethylhexanol had activity in the endocrine-related assays.

The main urinary metabolite for octocrylene is 2-cyano-3,3-diphenylacrylic acid (CPAA) (Bury et al., 2018, 2019c; Klotz et al., 2019). There is limited information on CPAA ED from ToxCast data. Consequently, ethyl 2-cyano-3,3-diphenylacrylate was used as structurally similar compound in the CompTox database (Tanimoto index 0.88). Ethyl 2-cyano-3,3-diphenylacrylate was the chemical with the highest structural similarity and was tested in at least 1 endocrine assay. Ethyl 2-cyano-3,3-diphenylacrylate had an AC50 of 6.650 µM and ACC of 4.900 µM in assay TOX21_AR_LUC_MDAKB2_Antagonist_0.5nM_R1881.

Oxybenzone has 3 major plasma and urinary metabolites: DHB, DHMB, and THB (Okereke et al., 1993; Tarazona et al., 2013). The AC50 for DHB was 4.191 µM and the ACC was 0.275 µM NCCT_TPO_AUR_dn. The AC50 for DHMB was 12.975 µM and the ACC was 7.984 µM in assay TOX21_AR_LUC_MDAKB2_Antagonist_0.5nM_R1881 in assay. The AC50 for THB was 1.838 µM and the ACC was 0.322 µM in assay.

Discussion

Sunscreen products containing UV filters have been used for decades to reduce the harmful effects of sunlight. Their use, as part of a sun safe strategy, is supported by NGOs, for example, Skin Cancer Foundation, government agencies, for example, HHS Office of the Surgeon General, and healthcare professionals such as the American Academy of Dermatology (Narla and Lim, 2020). Recent pharmacokinetic trials have found that 6 commonly used organic sunscreen active ingredients were detectable in human plasma at concentrations above 0.5 ng/ml, an FDA-defined “safety” threshold (Matta et al., 2019, 2020; Wang and Ganley, 2019). Because organic UV filters have been reported to have endocrine effects, that is, endocrine disruptors or EDs based on in vitro and/or animal models (Lorigo et al., 2018), we used the ToxCast/Tox21 dataset to investigate in vitro bioactivity activation thresholds and compare with human plasma concentrations for the 6 sunscreen actives evaluated in MUsTs.

In our study, endocrine bioactivity by sunscreen active ingredients focused on the “EAT(S) pathways” which include estrogen, androgen, thyroid, and, to a lesser extent, steroidogenesis. The EAT(S) modalities are well characterized with bioassays to evaluate agonist/antagonist activity of estrogen, androgen, and thyroid receptors in addition to steroidogenesis (Martyniuk et al., 2022). Importantly, only oxybenzone and homosalate were evaluated completely in EDSP21. Additional in vitro bioassays for octisalate, octocrylene, octinoxate, and avobenzone would be needed to base any conclusion solely on this dataset. The 6 UV filters in this study, however, have been evaluated for their ED potential using a variety of in silico, in vitro, and in vivo studies (Huang et al., 2021; Krause et al., 2012; Witorsch and Thomas, 2010).

In silico evaluation of the estrogen and androgen activity of the 6 UV filters suggest they are inactive or weakly active, apart from homosalate which had moderate binding affinity in Collaborative Estrogen Receptor (ER) Activity Prediction Project or CERAPP (Supplementary Table 3). The Collaborative Modeling Project for Androgen Receptor (AR) Activity (CoMPARA) data predict that the octisalate, homosalate, octocrylene, and octinoxate are inactive as androgen agonists and antagonists (Mansouri et al., 2020). Oxybenzone and avobenzone are predicted to be inactive as androgen agonists but active as antagonists. CERAPP predicts that octisalate, homosalate, octocrylene, oxybenzone, octinoxate, and avobenzone were inactive or had weak agonist/antagonist estrogenic activity (Mansouri et al., 2016), and consistent with CompTox EDSP21 dashboard.

Several in vitro studies have been performed investigating the estrogen, androgen, thyroid, and steroidogeneic effects of organic UV filters. In vitro data shows octisalate was not estrogenic or antiestrogenic in MCF-7 cells to weakly estrogenic (39 µM, 15% maximal response) and antiestrogenic activity in recombinant (hERα) yeast transactivation assays (Jimenez-Diaz et al., 2013; Kunz and Fent, 2006; Kunz et al., 2006). Octisalate did not exhibit androgenic or antiandrogenic in hAR assay in PALM cells (up to 10 µM) but showed weak androgenic effects in recombinant (hAR) yeast transactivation assays (1/45 000 androgenic activity relative to DHT) (Jimenez-Diaz et al., 2013; Kunz and Fent, 2006). Homosalate showed estrogenic effects (5.5 µM) and no antiestrogenic effects in MCF-7 cells (Jimenez-Diaz et al., 2013). Homosalate was noted to have weak antiandrogen activity (10 µM) with mixed results for androgen activity (Jimenez-Diaz et al., 2013; Kunz and Fent, 2006; Schreurs et al., 2005). Octocrylene showed ER binding and estrogen activity in MCF-7 cells (Matsumoto et al., 2005). Kunz et al. found that octocrylene demonstrated androgenic (EC50 of 627 µM), antiandrogenic (IC50 of 24.5 µM), and antiestrogenic activity (IC50 of 2570 µM) but no estrogenic activity in recombinant (hERα) yeast transactivation assays (Kunz and Fent, 2006). Oxybenzone, which has been under more scrutiny for ED compared with other UV filters (Mustieles et al., 2023), was reported to have estrogenic and androgenic activity in MCF-7 and HEK293 cells at higher concentrations (>10 µM) (Schlumpf et al., 2001; Schreurs et al., 2005). The lowest effect for oxybenzone was 0.5 µM measuring 50% increase in basal pS2 gene transcription in MCF-7 cancer cells (Heneweer et al., 2005). Octinoxate has been reported to have estrogenic activity in vitro (EC50 2.37 μM) and no androgenic effects in HEK293 and MCF-7 cells (Schlumpf et al., 2001; Schreurs et al., 2005). Octinoxate also had steroidogenesis activity in a H295R Steroidogenesis Assay at 10 µM (Strajhar et al., 2017). Avobenzone is reported to have estrogenic (10 µM) and antiandrogenic activity at high concentration (IC50 11 µM) in Schreurs et al. (2005). Avobenzone was also noted to have thyroid-like activity and antithyroid activity in GH3.TRE-Luc cells (EC50 0.001 µM and IC50 71.69 µM) (Klopcic and Dolenc, 2017). Collectively, although there are inconsistencies in the in vitro studies on the 6 UV filters with some studies showing no effect, many reported some activity at µM concentrations for estrogen, androgen, thyroid, or steroidogenic effects.

In our current investigation of ToxCast/Tox21 data, most of the activity observed was in bioassays testing estrogen and androgen effects. A smaller number of assays tested demonstrated disruptions of thyroid hormone activity and steroidogenesis (Figure 1). In general, the activity observed in these assays occurred at a concentration of greater than 6 µM, comparable with the concentrations reported in other published in vitro studies (Huang et al., 2021). This value, that is, 6 µM from Table 6, is based on the ACC, a metric independent of the maximum response and a relative potency estimate (Blackwell et al., 2017; Fay et al., 2018). That said, using an AC10 would be considered more protective than AC50 and the ACC. For the purposes of the present study, we believe ACC and AC50 serve as established, suitable metrics for comparison with human plasma concentrations, knowing that AC10 or AC1 are more conservative. We have avoided identifying “points of departure” for these metrics considering it inappropriate and subject to abuse, that is, lowest “number” to calculate margin of safety without understanding in vitro assay limitations. Regardless, as NAMs become a greater part of risk assessment, it will be critical to develop metrics that reflect reliable and reproducible metrics, including more conservative measures. Importantly, AC50 values for octisalate, homosalate, octocrylene, octinoxate, and avobenzone were greater than the “cytotoxic burst” suggesting general cytotoxicity rather than a specific, mechanistic response. Oxybenzone activity was below the cytotoxic burst because the threshold was set to 1000 µM due to low activity in the cytotoxicity assays. Concentrations above the cytotoxic burst are likely to result in stress responses as the cell nears death rather than disruption of a specific endocrine pathway (Escher et al., 2020). The relative low potency of the UV filters can be inferred by their higher percentile ranking based on their activity in the respective bioassays. The more potent chemicals in the assay are ranked at a lower percentile.

Further evaluation of the ToxCast/Tox21 data suggested that the metabolites of some of these 6 organic UV filters had low ED potential. No activity or concentrations higher than 1 µM (ACC/AC50) were needed to elicit a response in the ToxCast/Tox21 assays for metabolites of octisalate, homosalate, octocrylene, and octinoxate. For octisalate and homosalate, the main metabolite is salicylic acid, a chemical that has been studied extensively in humans (Palmer and Clegg, 2020). No avobenzone metabolites were tested in ToxCast/Tox21. Importantly, as presented in Table 7, oxybenzone metabolites (DHB, and THB) were active at concentrations below 1 µM a proposed internal threshold of toxicological concern or TTC (Najjar et al., 2023), suggestive that metabolism may contribute to effects observed with this UV filter and distinguishing it from the other 5 UV filters.

An additional step involved in using ToxCast/Tox21 data to evaluate the endocrine activity of UV filters was a careful curation of the assay “hits.” The concentration-response curves for each assay tested against a UV filter were carefully examined. Curves with multiple “flags” for low efficacy, curve overfitting, borderline activity, and noise were excluded. This objective exercise helped exclude positive assays that had no biological relevance.

Studies conducted in vivo are generally consistent with in vitro findings showing effects at high doses, for example, > 100 mg/kg/day. All 6 UV filters were negative for estrogen agonism or antagonism in uterotrophic assays with no increases in uterine weights reported (Klammer et al., 2005; NTP, 2011; Schlecht et al., 2004; Schlumpf et al., 2001). Similarly, Hershberger data are negative for androgenic effects for octisalate, octocrylene, octinoxate, and oxybenzone (NTP, 2012). The lack of estrogen or androgen effects at high dosages up to 1000 mg/kg/day in Hershberger and uterotrophic assays suggests the weak responses observed in in vitro studies are not physiologically relevant. Although repeat dose studies in animal models have shown some evidence of estrogenic activity, the effects are very weak and require very high doses (Schneider et al., 2005; Schreurs et al., 2002; Witorsch and Thomas, 2010). For example, a comprehensive in vivo 1 and 2-generation studies for octinoxate showed no endocrine effects up to 1000 mg/kg/day (Schneider et al., 2005). Nonetheless, of the 6 sunscreen active ingredients evaluated in our study, oxybenzone has the largest number of in vivo animal studies some of which demonstrate a level of endocrine activity (for review, see Mustieles et al. [2023]).

There has been no definitive cause-effect evidence that organic UV filters cause ED in humans. Reports on the ED potential of sunscreen actives typically cite in vitro and rodent studies with high oral doses, for example, >100 mg/kg and the findings of, for example, oxybenzone in human plasma from biomonitoring studies. There are no clinical reports of adverse endocrine effects from the exposure to sunscreen active ingredients in adults (Janjua et al., 2007; Suh et al., 2020). Human exposure data from Matta et al. (2019, 2020) indicate that the organic UV filters are detected at low concentrations in the plasma ranging from 0.0216 µM for octocrylene to levels of 1.131 µM for oxybenzone under exaggerated maximal usage conditions (applied every 2 h, 4 times a day over 75% of the body). This represents >100-fold difference from the lowest concentration needed to trigger activity in the endocrine bioassays treated with the UV filters, again, with the notable exception of oxybenzone with > 6-fold difference.

When considering ED, there have been attempts to identify criteria by which a substance shall be considered as having ED properties (European Chemical et al., 2018). Specifically, and abbreviated, a substance must meet all the following criteria: (1) show an adverse effect, that is, change in morphology, physiology, etc., (2) have an endocrine mode of action, and (3) the adverse effect is a consequence of the endocrine mode of action. Studies in humans have not shown an adverse effect apart from oxybenzone—albeit without a cause-and-effect relationship (Huang et al., 2021). In animal studies, effects have been observed but largely at high doses (>100 mg/kg), which appear to be non-specific events. Finally, the mode of action is absent/missing based on screening evaluations presented in this article. Thus, although endocrine activity may be present, for some of the UV filters, it is too weak to produce any adverse effects and, as such, do not fulfill all the criteria needed to be categorized as an ED chemical.

There are several limitations of the current study. To begin with, the study is focused largely on the EATS modalities of ED of the UV filters and, only homosalate and oxybenzone have the full complement of screening studies (Supplementary Table 3). In the absence of a full complement of assays, any strong conclusion related to the absence of endocrine effects based solely on ToxCast/Tox21 data would be an overstatement (EPA, 2022). The endocrine system is complicated with multiple pathways that can be perturbed by chemicals, that is, glucocorticoid, progesterone, retinoic acid, and peroxisome proliferator-activated receptor signaling. This is particularly true for the thyroid system where in vitro testing in ToxCast/Tox21 database underrepresents pathways in the AOP (Noyes et al., 2019). However, data interrogating some of this signaling pathways indicate that the UV filters have only weak endocrine activity in non-EATS pathways at high concentration. For example, avobenzone upregulated mRNA levels of PPARγ in vitro during adipogenesis in human bone marrow mesenchymal stem cells at concentrations >10 µM (Ahn et al., 2019). Octisalate exhibited progesterone-like activity in human sperm cells at 10 µM (Rehfeld et al., 2018). Next, the assays only consider nominal in vitro concentrations, which may not represent the concentration responsible for the observed effect. As well, and a limitation for many cell-based in vitro assays, there was no evaluation of metabolism, which, at least for oxybenzone may be important. Finally, in their most recent White Paper, EPA describes full estrogen and AR pathway models that constitute a “weight of evidence” for such effects (EPA, 2022). Accordingly, the UV filters were not tested in enough of the in vitro EDSP/ToxCast assays to then evaluate the in vitro data through the ToxCast pathway models (Browne et al., 2015; Kleinstreuer et al., 2017), which is the preferred way to look at this type of data rather than looking at select assays. Despite these limitations, the current study is the first to compare human plasma concentrations of these UV filters to ToxCast bioactivity, an independent, unbiased source of bioactivity.

The use of ToxCast/Tox21 data is an example of “new approach methodologies” (NAMs) that are being evaluated by several regulatory agencies including the EPA, The European Chemicals Agency (ECHA), and Scientific Committee on Consumer Safety for hazard and ultimately risk assessment of chemicals (Parish et al., 2020). As part of a weight-of-evidence analysis, our study shows that the UV filters octisalate, homosalate, octocrylene, octinoxate, and avobenzone have low ED potential risk based on an examination of ToxCast/Tox21 data with weak activity occurring at concentrations >100-fold what would be achieved in human plasma. The exception is oxybenzone where the bioactivity and plasma concentrations are separated by less than 10-fold.

Supplementary data

Supplementary data are available at Toxicological Sciences online.

Declaration of conflicting interests

All authors are full-time employees of The Procter & Gamble Company.