Assessing Indoor Dust Interference with Human Nuclear Hormone Receptors in Cell-Based Luciferase Reporter Assays

Abstract

Background:

Per- and polyfluoroalkyl substances (PFAS), organophosphate esters (OPEs), and polybrominated diphenyl ethers (PBDEs) are hormone-disrupting chemicals that migrate from building materials into air and dust.

Objectives:

We aimed to quantify the hormonal activities of 46 dust samples and identify chemicals driving the observed activities.

Methods:

We evaluated associations between hormonal activities of extracted dust in five cell-based luciferase reporter assays and dust concentrations of 42 measured PFAS, OPEs, and PBDEs, transformed as either raw or potency-weighted concentrations based on Tox21 high-throughput screening data.

Results:

All dust samples were hormonally active, showing antagonistic activity toward peroxisome proliferator-activated receptor ($\text{PPAR}\mathrm{\gamma }2$) (100%; 46 of 46 samples), thyroid hormone receptor ($\text{TR}\mathrm{\beta }$) (89%; 41 samples), and androgen receptor (AR) (87%; 40 samples); agonist activity on estrogen receptor ($\text{ER}\mathrm{\alpha }$) (96%; 44 samples); and binding competition with thyroxine (${\mathrm{T}}_4$) on serum transporter transthyretin (TTR) (98%; 45 samples). Effects were observed with as little as $4\mathrm{\mu }\mathrm{g}$ of extracted dust. In regression models for each chemical class, interquartile range increases in potency-weighted or unknown-potency chemical concentrations were associated with higher hormonal activities of dust extracts (potency-weighted: $\mathrm{\Sigma }\text{PFAS}–\text{TR}\mathrm{\beta }$, $\uparrow 28\%$, $p<0.05$; $\mathrm{\Sigma }\text{OPEs}–\text{TR}\mathrm{\beta }$, $\uparrow 27\%$, $p=0.08$; $\mathrm{\Sigma }\text{PBDEs}–\text{TR}\mathrm{\beta }$, $\uparrow 20\%$, $p<0.05$; $\mathrm{\Sigma }\text{PBDEs}–\text{ER}\mathrm{\alpha }$, $\uparrow 7.7\%$, $p=0.08$; unknown-potency: $\mathrm{\Sigma }\text{OPEs}–\text{TTR}$, $\uparrow 34\%$, $p<0.05$; $\mathrm{\Sigma }\text{OPEs}–\text{AR}$, $\uparrow 13\%$, $p=0.06$), adjusted for chemicals with active, inactive, and unknown Tox21 designations.

Discussion:

All indoor dust samples exhibited hormonal activities, which were associated with PFAS, PBDE, and OPE levels. Reporter gene cell-based assays are relatively inexpensive, health-relevant evaluations of toxic loads of chemical mixtures that building occupants are exposed to. https://doi.org/10.1289/EHP8054

Introduction

Materials inside buildings contain many hormone-disrupting chemicals, including flame retardants (FRs) and per- and polyfluoroalkyl substances (PFAS) (Lucattini et al. 2018). As unbound additives, these chemicals can leach out of products and accumulate in the dust (Allen et al. 2008b; Mitro et al. 2016; Rauert et al. 2014; Tokranov et al. 2019) that we unintentionally ingest and breathe (Johnson-Restrepo and Kannan 2009; Poothong et al. 2020; Xu et al. 2016). In fact, FRs and PFAS have been detected in the urine or blood of over 90% of Americans (Calafat et al. 2007; Ospina et al. 2018; Sjödin et al. 2008).

PFAS are a class of over 4,700 extremely persistent chemicals (OECD 2018) applied as stain-, grease-, or water-resistant coatings to carpet, furniture, clothing, cookware, and disposable food packaging (Sunderland et al. 2019). PFAS are associated with adverse human health effects on thyroid function (Rappazzo et al. 2017; Xiao et al. 2020), metabolism (including overweight/obesity, diabetes, insulin resistance, and high cholesterol) (Lin et al. 2019; Sunderland et al. 2019), fetal development (Liew et al. 2018; Xiao et al. 2020), and the immune system (Rappazzo et al. 2017), and possibly kidney and testicular cancer (Barry et al. 2013; Stanifer et al. 2018). Even though two of the most widely known toxic PFAS were voluntarily phased out of production by manufacturers in the United States starting in the early 2000s, the numerous replacement PFAS are also of concern to human health (Wang et al. 2013, 2015, 2017).

Chemical FRs have been added to foam furniture, carpet, electronics, and building insulation (Cooper et al. 2016; Jinhui et al. 2017; Kemmlein et al. 2003). One type, polybrominated diphenyl ethers (PBDEs), were largely phased out by 2013 in the United States (Dodson et al. 2012). However, old PBDE-containing products are still in use after decades and can be recycled into new products (Abbasi et al. 2015). In addition, organophosphate esters (OPEs) are often used as PBDE replacements and as plasticizers. Research has found the adverse health effects to include thyroid dysfunction, poor pregnancy outcomes, infertility, and impairment of cognitive or reproductive development for both PBDEs (Allen et al. 2016; Boas et al. 2012; Choi et al. 2019; Czerska et al. 2013; Johnson et al. 2013; Linares et al. 2015; Mumford et al. 2015; Vuong et al. 2018) and, more recently, OPEs (Carignan et al. 2017, 2018; Doherty et al. 2019a, 2019b; Meeker and Stapleton 2010; Messerlian et al. 2018; Preston et al. 2017; Wang et al. 2020).

There is considerable evidence that PFAS, PBDEs, and OPEs are hormone-disrupting chemicals. Because nuclear hormone receptors regulate critical genes, their signaling disruption can lead to reproductive (e.g., infertility), developmental (e.g., abnormal fetal growth), and metabolic (e.g., obesity or diabetes) diseases. Certain PFAS and PBDEs or OPEs have been shown to activate estrogen receptor $\mathrm{\alpha }$ ($\text{ER}\mathrm{\alpha }$) (Du et al. 2013; Hamers et al. 2006; Ren and Guo 2013; Suzuki et al. 2013); suppress peroxisome proliferator-activated receptor $\mathrm{\gamma }$ ($\text{PPAR}\mathrm{\gamma }$) (U.S. EPA 2019; Wen et al. 2016), androgen receptor (AR) (Hamers et al. 2006; Klopčič et al. 2016; Orton et al. 2014; Suzuki et al. 2013), and thyroid hormone receptor $\mathrm{\beta }$ ($\text{TR}\mathrm{\beta }$) (Du et al. 2013; Klopčič et al. 2016; Kollitz et al. 2018; Ren and Guo 2013) and to interfere with thyroid hormone serum transport (Hamers et al. 2006; Rosenmai et al. 2021; Weiss et al. 2009). $\text{ER}\mathrm{\alpha }$ regulates the development and maintenance of breast and uterine tissue, as well as the cardiovascular system, female reproductive cycle, and bone density (Delfosse et al. 2015; Grimaldi et al. 2015). AR plays an important role in male sexual development differentiation and spermatogenesis (Delfosse et al. 2015; Grimaldi et al. 2015). $\text{PPAR}\mathrm{\gamma }$ regulates fat storage, lipid metabolism, and insulin sensitivity and can produce anti-inflammatory effects (Delfosse et al. 2015; Grimaldi et al. 2015). $\text{TR}\mathrm{\beta }$ is crucial for normal development, growth, metabolism, and brain function (Grimaldi et al. 2015). Serum transporter transthyretin (TTR) is important for delivering the thyroid hormone thyroxine (${\mathrm{T}}_4$) across the blood–brain barrier and placenta (Grimm et al. 2013). When a chemical competitively binds to TTR instead, free ${\mathrm{T}}_4$ is more readily eliminated from the body, and that chemical could move into important target tissues (Grimm et al. 2013; Ishihara et al. 2003).

Cell-based assays are an emerging method to quantify the total activation or suppression of hormone receptors by complex environmental mixtures of hormone-disrupting chemicals. Compared with traditional targeted laboratory approaches that measure each chemical in a mixture individually, cell-based assays of dust are inexpensive, rapid, and statistically simple to model. Hormonal activities in assays of dust also reflect the combined effects from co-exposures of all hormone-disrupting chemicals in the sample, including unmeasurable chemicals and unknown regrettable substitutes. The assays account for any mixture effects, such as when a chemical’s effect is triggered, enhanced, or reduced in the presence of another chemical (Kollitz et al. 2018; Vandermarken et al. 2016).

Few studies have measured the activities of dust toward nuclear hormone receptors using cell-based assays (Chou et al. 2015; Fang et al. 2015; Hamers et al. 2020; Kassotis et al. 2019; Suzuki et al. 2007; Vandermarken et al. 2016). For example, Suzuki et al. (2013) reported that certain measured PBDEs or OPEs in household dust were probable contributors to $\text{ER}\mathrm{\alpha }$ activation and AR suppression. Kollitz et al. (2018) found significant correlations between PBDE or OPE levels and $\text{TR}\mathrm{\beta }$ antagonism in household dust even though the 12 measured FRs were not active when tested in isolation, demonstrating possible mixture effects and influence from unmeasured chemicals. To our knowledge, there are currently no published studies relating PFAS concentrations to bioactivities in dust.

Hormone receptor activity of a mixture is a function of not only each chemical’s concentration, but also its potency. The increase in available high-throughput screening assay data, such as the Tox21 database for individual potencies of almost 10,000 chemicals (U.S. EPA 2019; Huang et al. 2016), has recently enabled studies to integrate information on chemical concentrations and their respective potencies in order to identify key contaminants driving the total bioactivities of water samples (Blackwell et al. 2017). To our knowledge, this type of potency-weighted exposure evaluation using high-throughput screening data has not been done with chemicals in dust.

The objectives of this study were to a) quantify hormonal activities of 46 indoor dust samples; b) identify associations between measured PFAS, PBDE, and OPE chemicals and hormonal activities of dust; and c) evaluate potency-weighted chemical concentration calculations as a method to determine which measured chemicals are driving the observed effects of dust mixtures.

Methods

Study Design

We collected dust samples from 46 rooms across 21 buildings at a university in the United States during January to March 2019. The rooms included 22 common spaces, 6 office suites, and 18 classrooms distributed across different renovation statuses. Approximately half of the samples ($n=22$) were collected from rooms renovated between 2017 and 2019, with furniture and carpet specified to be free of all chemical FRs and PFAS. The remaining samples ($n=24$) were collected from carpeted rooms that constituted a similar distribution of room types as the other 22 samples and that had been renovated with conventional furniture as recently as possible. Building construction years ranged from 1863 to 2018 ($\text{median}=1966$) and years of last renovation ranged from 2001 to 2019 ($\text{median}=2017$). Further details about the rooms and sample collection are provided in our previous manuscript (Young et al. 2021).

Dust Sample Collection

We collected a dust sample within each third of each room in order to have sufficient dust mass for three different laboratories (one for chemicals, one for all the cell-based assays, and one for other research). For each sample, we vacuumed all floor surfaces within that one-third sampling area, including underneath furniture, for 10 min. We vacuumed dust into a cellulose extraction thimble (Whatman International) secured with a nitrile rubber O-ring (McMaster-Carr) in a crevice tool attached to a vacuum cleaner (Dyson CY18), following a previously published protocol (Allen et al. 2008a). Thus, the dust came into contact only with the crevice tools, which were cleaned with hot water and isopropyl alcohol between samples. The thimbles were stored in polypropylene centrifuge tubes in polyethylene bags at $-13°\mathrm{C}$ until analysis. For field blanks, we carried four unopened centrifuge tubes with thimbles to field sites on various sampling days.

Cell-Based Luciferase Reporter Gene Assays

The dust samples and field blanks were analyzed in chemically activated luciferase gene expression (CALUX) assays by BioDetection Systems. Based on known or suspected mechanisms of human toxicity for PFAS and FRs, we chose the following assays: antagonism of $\text{TR}\mathrm{\beta }$, AR, and $\text{PPAR}\mathrm{\gamma }2$; agonism of $\text{ER}\mathrm{\alpha }$; and interference of ${\mathrm{T}}_4$ binding to transthyretin. We initially tested 10 samples for $\text{PPAR}\mathrm{\gamma }$ agonism too, but no agonism was detected, so we did not proceed further (Figure S2bi,bj).

These luciferase reporter gene assays employed human female osteosarcoma cell lines (U2OS) stably transfected with the firefly luciferase reporter gene, whose expression was controlled by activation of specific nuclear hormone receptors under study (Sonneveld et al. 2005). When any agonistic chemicals in the dust extract activated a specific receptor, it would trigger expression of luciferase, and that enzyme produced light (luminescence) in the presence of added luciferin substrate. By contrast, antagonistic chemicals would have inhibited the added agonist from activating the receptor, thereby reducing light induction. The intensity of light was measured with a luminometer and was directly proportional to the degree of receptor activation. In the TTR-${\mathrm{T}}_4$ interference assay, the chemicals competed with a fixed concentration of ${\mathrm{T}}_4$ to bind the transport protein TTR, and some ${\mathrm{T}}_4$ would be replaced. The amount of ${\mathrm{T}}_4$ still bound to TTR was then separated out, and its activation of $\text{TR}\mathrm{\beta }$ was quantified in the $\text{TR}\mathrm{\beta }$ agonism assay.

As demonstrated in Figure S1, the measured luminescence in a particular assay was benchmarked against a reference compound to calculate a final result for each sample in units of microgram equivalents per gram. This unit can be interpreted as for a given mass (in grams) of dust, the mass (in micrograms) of the reference compound that produces the same level of activity. The reference compounds were potent and selective agonists or antagonists and measured alongside the samples in the agonism or antagonism assays, respectively (Table 1). A full dose–response curve for each reference compound was fitted from the activities of eight separate serial dilutions using the Hill equation. Then, the benchmarked activities (in microgram equivalents per gram) of the dust extracts were calculated by interpolating a certain concentration of dust extract onto the calculated calibration curve of the reference compound. For agonistic activity for each dust extract, we used the data point for the lowest sample concentration that produced a response above the limit of quantification (LOQ). The result was then the ratio of the reference compound concentration in medium to the sample concentration at that same measured response level. Whereas an actual measured point was used for the sample concentration, the reference concentration was interpolated from the calculated dose–response curve. For antagonism, we used the lowest sample concentration that produced the highest response $<80\%$ of the maximal response (i.e., $>20\%$ inhibition) of the reference compound. The result was then the ratio of the reference to sample concentration at that measured response level for that sample. Although different response levels were used in the calculations for different samples, the results were comparable because of the interpolation onto the reference curve and because the chosen response levels were targeted to be within the linear range of the reference curve. Dose–response curves shifted further to the left indicated higher potency.

Table 1.

Summary statistics for the hormonal activities of 46 indoor dust samples in luciferase reporter gene assays.

Assay end point	Abbreviation	Reference compound	Percentage detected	GM (GSD)	Median	Range among detected	Units
Peroxisome proliferator-activated receptor $\mathrm{\gamma }2$ antagonism	$\text{PPAR}\mathrm{\gamma }$	GW9662 (chemical)	100% (46 of 46)	0.554 (1.92)	0.580	0.150–2.90	$\mathrm{\mu }\mathrm{g}\text{-eq}/\mathrm{g}$
Estrogen receptor $\mathrm{\alpha }$ agonism	ER	$17\mathrm{\beta }\text{-estradiol}$ (natural hormone)	96% (44 of 46)	2.21 (2.38)	1.76	0.287–22.0	$\text{ng-eq}/\mathrm{g}$
Thyroid hormone receptor $\mathrm{\beta }$ antagonism	TR	Deoxynivalenol (mycotoxin)	89% (41 of 46)	68.7 (2.30)	80.8	12.8–370	$\mathrm{\mu }\mathrm{g}\text{-eq}/\mathrm{g}$
Androgen receptor antagonism	AR	Flutamide (medication)	87% (40 of 46)	105 (2.26)	104	27.5–434	$\mathrm{\mu }\mathrm{g}\text{-eq}/\mathrm{g}$
Thyroid hormone transport interference	TTR-${\mathrm{T}}_4$	Perfluorooctanoate (PFAS chemical)	98% (45 of 46)	104 (2.90)	141	15.2–626	$\mathrm{\mu }\mathrm{g}\text{-eq}/\mathrm{g}$

Note: Activities by sample can be found in Table S2 and Figure S2. AR, androgen receptor; ER, estrogen receptor; GM, geometric mean; GSD, geometric standard deviation; PFAS, per- and polyfluoroalkyl substances; $\text{PPAR}\mathrm{\gamma }2$, peroxisome proliferator-activated receptor; TR, transport protein; TTR-${\mathrm{T}}_4$, transthyretin–thyroxine.

Exposure of Cell-Based Assays to Dust Extracts

To prepare the dust samples for the cell assays, the samples were first sieved with a $1\text{-mm}$ mesh and extracted by accelerated solvent extraction (ASE) using hexane and acetone (1:1, vol/vol). An average $0.43\mathrm{g}$ of dust was extracted for each sample. After gentle evaporation under nitrogen, the hexane/acetone extracts were dissolved in $100\mathrm{\mu }\mathrm{L}$ of dimethyl sulfoxide (DMSO). Five-point serial dilutions ($1\times ,3\times ,10\times ,30\times ,\text{\hspace{0.17em}and\hspace{0.17em}}100\times$) of each final extract were then prepared in DMSO. The $3\times$ dilution was made with $8\mathrm{\mu }\mathrm{L}$ of $1\times$ dilution plus $16\mathrm{\mu }\mathrm{L}$ DMSO, and the subsequent dilutions were made with $36\mathrm{\mu }\mathrm{L}$ DMSO plus $4\mathrm{\mu }\mathrm{L}$ of a prior level of dilution (using the $1\times$ dilution to make the $10\times$ dilution, using the $3\times$ dilution to make the $30\times$, using the $10\times$ to make the $100\times$, using the $30\times$ to make the $300\times$, etc.). The final DMSO concentration during exposure of the cells to the prepared serial dilutions was 0.1% in the hormone receptor assays, based on $1\mathrm{\mu }\mathrm{L}$ of the dilution in $1\mathrm{m}\mathrm{L}$ of medium.

For the nuclear hormone receptor assays, CALUX cells were cultured in Dulbecco’s Modified Eagle Medium/Nutrient Mixture F12 (DMEM/F12) without phenol-red (#31331-028; Gibco) supplemented with 7.5% charcoal-stripped fetal calf serum (FCS), nonessential amino acids (#11140-03; Gibco) and $10\mathrm{U}/\mathrm{mL}$ penicillin and $10\mathrm{\mu }\mathrm{g}/\mathrm{mL}$ streptomycin (culture medium). Cells were maintained at standardized conditions (37°C, 5% ${\mathrm{CO}}_2$, high humidity) and subcultured every 3–4 d.

For agonistic analysis, cultured CALUX cells were trypsinized and resuspended in assay medium (DMEM/F12 medium supplemented with 5% charcoal-stripped FCS, nonessential amino acids, and $10\mathrm{U}/\mathrm{mL}$ penicillin and $10\mathrm{\mu }\mathrm{g}/\mathrm{mL}$ streptomycin) at a final concentration of $1\times {10}^5\text{\hspace{0.17em}cells}/\mathrm{mL}$. Assay medium for $\text{TR}\mathrm{\beta }$ antagonism analyses did not contain dextran-coated charcoal-stripped FCS. Resuspended cells were seeded in 96-well plates and incubated for 16–24 h under standardized conditions. Following incubation, the medium was removed, and the cells were incubated with $200\mathrm{\mu }\mathrm{L}$ of exposure medium containing assay medium supplemented with the dilution series of dust extract or reference compound in 0.1% DMSO. All dilutions were tested in triplicate. Following $24\pm 2\mathrm{h}$ of incubation of exposed cells under standardized conditions, the incubation plates were removed from the incubator, the exposure medium was removed, and the cells were lysed using $30\mathrm{\mu }\mathrm{L}/\text{well}$ of a Triton-lysis buffer. Luciferase activity in cellular lysates was measured for 4 s using a luminometer (TriStar LB941; Berthold).

For antagonistic analysis, the procedure described above was followed with the exception that the exposure medium was supplemented with the nonsaturating, half maximal effective concentration (${\text{EC}}_{50}$) of an agonist [triiodothyronine (${\mathrm{T}}_3$) (#T2877; Sigma-Aldrich), rosiglitazone (#71740; Cayman Chemical Company), and dihydrotestosterone (#S4757; Selleckchem) for $\text{TR}\mathrm{\beta }$, $\text{PPAR}\mathrm{\gamma }$, and AR assays, respectively]. Antagonistic compounds were expected to compete with binding of that agonist to its respective receptor, resulting in a reduction in light emission.

For the TTR-${\mathrm{T}}_4$ binding interference analysis, we used a combination of the TTR-${\mathrm{T}}_4$ competition assay and the $\text{TR}\mathrm{\beta }$ assay to determine the amount of ${\mathrm{T}}_4$ still bound to TTR after exposure. Serial sample dilutions in DMSO were incubated in Tris buffer (pH 8.0; $200\mathrm{\mu }\mathrm{L}$ transferred from a mixture of $140\mathrm{\mu }\mathrm{L}$ of Tris buffer with $500\mathrm{\mu }\mathrm{L}$ of serum-free medium) overnight at 4°C in the presence of TTR ($0.058\mathrm{\mu }\mathrm{M}$) and a fixed concentration of ${\mathrm{T}}_4$ ($0.052\mathrm{\mu }\mathrm{M}$). The final concentration of DMSO in the incubations was 3.2%. Following incubation, TTR-bound ${\mathrm{T}}_4$ was separated from free ${\mathrm{T}}_4$ by placing the incubation mixture on cooled Bio-Gel P-6DG columns (#150-0739; Bio-Rad) after which the columns were centrifuged for 1 min at $120\times g$. The eluate (TTR-bound ${\mathrm{T}}_4$) was added to the assay medium, after which $\text{TR}\mathrm{\beta }$ CALUX cells were exposed for 24 h, as described above. The procedure for the TTR-${\mathrm{T}}_4$ assay is presented in Figure 2 of the manuscript by Collet et al. (2019).

Before evaluating the samples in the CALUX nuclear hormone receptor assays, prepared serial dilutions (with 1% DMSO) were first evaluated for cytotoxicity (cell line #83; BioDetection Systems). Unlike other assays, this cell line continuously expressed luciferase, and any cell death reduced the amount of light emitted, which we measured with the luminometer. Sample extract dilutions that caused a 20% reduction in light were considered cytotoxic and were excluded from assessment because the lack of cell viability in the test system could be misinterpreted as antagonism (van der Linden et al. 2014). The cytotoxicity assay was conducted with a cell density of $\mathrm{10,000}\text{\hspace{0.17em}cells}/\mathrm{mL}$ and treatment length of 24 h.

To quantify CALUX analysis results, full dose–response dilution series of the assay-specific reference compounds were included on each plate. The reference compounds were $17\mathrm{\beta }\text{-estradiol}$ (#E2758; Sigma-Aldrich) for $\text{ER}\mathrm{\alpha }$ agonism, flutamide for AR antagonism (#F9397; Sigma-Aldrich), GW9662 for $\text{PPAR}\mathrm{\gamma }$ antagonism (#70785-50; Cayman), deoxynivalenol for $\text{TR}\mathrm{\beta }$ antagonism (#D0156; Sigma-Aldrich), perfluorooctanoate (#171468; Sigma-Aldrich) for TTR-${\mathrm{T}}_4$ binding interference, and tributyltin acetate (#8216500050; Merck) for cytotoxicity.

The method LOQs for antagonism were defined as the concentration of reference compound resulting in 80% of its maximal response. For agonism, the LOQs were calculated as the average of the DMSO solvent blank plus 10 times the standard deviation of the triplicate measurements of the solvent blank. Each plate had separate solvent blanks, so different samples could have a slightly different LOQ depending on which plate they were analyzed on. For samples with no dilutions producing a response above the LOQ, the LOQ was reported, corrected to represent the first dilution not showing cytotoxicity if needed. The average LOQs for the assays were $0.099\text{ng-eq}/\mathrm{g}\text{-dust}$ for $\text{ER}\mathrm{\alpha }$ agonism, $8.3\mathrm{\mu }\mathrm{g}\text{-eq}/\mathrm{g}\text{-dust}$ for AR antagonism, $13\mathrm{\mu }\mathrm{g}\text{-eq}/\mathrm{g}\text{-dust}$ for $\text{TR}\mathrm{\beta }$ antagonism, $0.022\mathrm{\mu }\mathrm{g}\text{-eq}/\mathrm{g}\text{-dust}$ for $\text{PPAR}\mathrm{\gamma }$ antagonism, and $0.94\mathrm{\mu }\mathrm{g}\text{-eq}/\mathrm{g}\text{-dust}$ for TTR-${\mathrm{T}}_4$ binding interference, without taking into account the cytotoxicity of certain sample dilutions. The catalog numbers for the assay cell lines are BioDetection Systems #44 for AR antagonism, #60 for $\text{ER}\mathrm{\alpha }$ agonism, #82 for $\text{PPAR}\mathrm{\gamma }$ antagonism, and #88 for $\text{TR}\mathrm{\beta }$ antagonism.

For quality assurance and quality control, all dust sample extracts, reference compound series, and solvent blanks were analyzed in triplicate with an acceptable maximum coefficient of variation defined as $<15\%$. Each plate contained its own reference compound series and solvent blanks. The four field blanks were almost all below the LOQ for all five assays (plus cytotoxicity) or otherwise well below the minimum detected response of the samples, except that one field blank had a detected response against $\text{TR}\mathrm{\beta }$ that was about half the median of the samples (all three other blanks had responses below the LOQ) (Table S1). We subtracted average field blank responses from the sample responses, as described in the “Statistical Analyses” section. More details are described in previous studies for the CALUX assay procedures for AR antagonism and $\text{ER}\mathrm{\alpha }$ agonism (Sonneveld et al. 2005), $\text{PPAR}\mathrm{\gamma }$ antagonism (Gijsbers et al. 2011), $\text{TR}\mathrm{\beta }$ antagonism, and TTR-${\mathrm{T}}_4$ binding interference (Collet et al. 2019).

Chemical Analyses of Dust

The dust samples and field blanks were analyzed for 15 PFAS (Kim and Kannan 2007), 19 OPEs (Kim et al. 2019), and 8 PBDEs (Johnson-Restrepo and Kannan 2009) by following published protocols. Specifically, the measured PFAS were perfluorooctane sulfonate (PFOS), perfluorooctanoate (PFOA), perfluorohexanoate (PFHxA), perfluorohexane sulfonate (PFHxS), perfluorooctane sulfonamide (FOSA), perfluoroheptanoate (PFHpA), perfluoropentanoate (PFPeA), perfluorononanoate (PFNA), perfluorobutane sulfonate (PFBS), perfluorodecane sulfonate (PFDS), perfluorobutanoate (PFBA), perfluorodecanoate (PFDA), perfluoroundecanoate (PFUnDA), perfluorododecanoate (PFDoDA), and $n$-methyl perfluorooctane sulfonamidoacetic acid (N-MeFOSAA). PBDE analytes included congeners 28, 47, 99, 100, 153, 154, 183, and 209. The OPE analytes were tris(2-butoxyethyl) phosphate (TBOEP), tris(1-chloro-2-propyl) phosphate (TCIPP), tris(1,3-dichloro-2-propyl) phosphate (TDCIPP), triphenyl phosphate (TPHP), tris(2-chloroethyl) phosphate (TCEP), 2-ethylhexyl diphenyl phosphate (EHDPP), isodecyl diphenyl phosphate (IDDP), tri-iso-butyl phosphate (TIBP), tripropyl phosphate (TPP), cresyl diphenyl phosphate (CDPP), tert-butylphenyl diphenyl phosphate (BPDP), tri-n-butyl phosphate (TNBP), tetrakis(2-chloroethyl) dichloroisopentyl diphosphate (V6), bisphenol A bis(diphenyl phosphate) (BDP), resorcinol bis(diphenyl phosphate) (RDP), tris(2-ethylhexyl) phosphate (TEHP), tris(methylphenyl) phosphate (TMPP), triethyl phosphate (TEP), and tris(p-tert-butylphenyl) phosphate (TBPHP).

First, the dust samples were sieved through a $150\text{-}\mathrm{\mu }\mathrm{m}$ stainless steel mesh. Then, the samples ($0.2–0.5\mathrm{g}$) were spiked with $30\text{\hspace{0.17em}ng}$ each of labeled surrogate standard mixture and extracted using methanol ($3\mathrm{mL}$) with mechanical oscillation (1 h) followed by ultrasonication (30 min). The resultant extracts were centrifuged ($\mathrm{3,500}\times g$, 10 min) and transferred into new polypropylene tubes. The extraction procedure was repeated twice with acetonitrile ($3\mathrm{mL}$) and ethyl acetate ($3\mathrm{mL}$), and then the extracts were combined and evaporated to $3\mathrm{mL}$ under a gentle stream of nitrogen and divided into three aliquots for analysis of PFAS, OPEs, and PBDEs. The aliquots were evaporated to near dryness and were reconstituted with $200\mathrm{\mu }\mathrm{L}$ of different solvents: methanol for PFAS, water:methanol (4:6; vol/vol) for OPEs, and hexane for PBDEs. The extracts were filtered through $0.2\text{-mm}$ nylon filters into glass vials prior to instrumental analysis.

OPEs were analyzed with high-performance liquid chromatography (HPLC) coupled with electrospray ionization (ESI) triple quadrupole mass spectrometry (ESI-MS/MS), using electrospray positive ionization multiple reaction monitoring. PBDEs were analyzed using a gas chromatographer coupled with a mass spectrometer (GC-MS) under electron impact ionization mode. PFAS were analyzed using HPLC coupled with ESI-MS/MS. Target PFAS were monitored by multiple reaction monitoring mode under negative ionization. Limits of detection ranged from $0.1–0.8\text{\hspace{0.17em}ng}/\mathrm{g}$ for OPEs, $0.09–4.5\text{\hspace{0.17em}ng}/\mathrm{g}$ for PBDEs, and $0.06–1.5\text{\hspace{0.17em}ng}/\mathrm{g}$ for PFAS.

Chemical concentrations in the field blanks were all either below the LOD or far below measured concentrations in dust samples. Duplicate analysis of seven dust samples showed that median relative percentage differences were 0% (range: $-62$ to 190%) for PFAS, 0% (range: $-96$ to 52%) for OPEs, and $-3.2\%$ (range: $-50$ to 80%) for PBDEs. This variability likely reflects the natural heterogeneity of dust. Another publication describes the laboratory analyses and results of the chemical analyses in depth (Young et al. 2021).

Potency-Weighted Concentrations of Chemicals

To account for different degrees of hormonal activity of individual chemicals present in dust mixtures, we used previously published information to calculate relative potency factors (RPFs), which are weights for each chemical based on its bioactivity in a given assay compared with other chemicals. For the four antagonism or agonism end points, we downloaded Tox21 data on the in vitro toxicity screening of thousands of chemicals from the U.S. Environmental Protection Agency’s ToxCast Chemistry Dashboard (Huang et al. 2016). We chose one reporter gene assay per end point based on relevance, availability, and cell line sensitivity: “TOX21_PPARg_BLA_antagonist_ratio” (beta-lactamase reporter; human embryonic kidney cells), “TOX21_TR_LUC_GH3_Antagonist” (luciferase reporter; rat pituitary tumor cells), “TOX21_AR_LUC_MDAKB2_Antagonist_0.5nM_R1881” (luciferase; human breast cells), and “TOX21_ERa_LUC_VM7_Agonist” (luciferase; human ovarian cancer cells), respectively.

As measures of potency in the Tox21 tests, we used activity concentrations at cutoff (ACCs) because they are point-of-departure estimates based on the potency of a chemical at a threshold that is predefined for all chemicals for the given assay. By contrast, more traditional chemical concentrations at 50% of their own maximal response (${\text{AC}}_{50}\mathrm{s}$) are estimated at different thresholds for different chemicals and thus cannot be accurately compared (Blackwell et al. 2017). We first inverted each analyte’s ACC and then applied Equation 1 to calculate a unitless RPF for each assay. The “HIT_CALL” and “FLAGS” columns in the database were used to classify chemicals as active (including if borderline) or inactive in an assay.

$$RP{F}_{chemical}=\frac{potenc{y}_{chemical}}{\text{max}{(potency)}_{allanalytes}}.$$ (1)

For the TTR-${\mathrm{T}}_4$ binding interference assay, there was no available Tox21 information, so we used laboratory data to calculate RPFs for PFAS in the exact same luciferase assay (Besselink 2020). These RPFs were calculated as the ${\text{IC}}_{50}$ of the chemical (the concentration at 50% of its maximal response) divided by the maximal ${\text{IC}}_{50}$ (max ${\text{IC}}_{50}$) among our analytes, following Equation 1. An active chemical had an ${\text{IC}}_{50}$ greater than zero.

The RPFs allowed us to develop sums of potency-weighted concentrations for each chemical class in dust for each assay end point (Equation 2), instead of just an unweighted sum that did not account for chemicals in each class having different degrees of both concentrations and potencies. We could also then calculate the percentage that each chemical contributes to the potency-weighted class sum in order to identify important drivers of differences in dust bioactivities. We adapted the methods from exposure–activity ratios used in recent research of chemicals in water (Blackwell et al. 2017) and from toxicity equivalents for dioxin-like activities of chemicals in dust (Tue et al. 2013).

$$Potency–WeightedSumofChemicalClas{s}_{\mathrm{\mu }g/g}={\sum }_{analytes}Concentratio{n}_{\mathrm{\mu }g/g}\times RPF.$$ (2)

Statistical Analyses

We blank-corrected chemical concentrations and dust potencies by subtracting averages of field blanks. Before modeling, we substituted nondetect values with half the LOD (Hornung and Reed 1990), and potencies were log-transformed due to nonnormality (based on Shapiro-Wilk tests and histograms). We conducted several stages of linear regression to first evaluate the unweighted impact of the three chemical classes on the bioactivities and then to model separate contributions of chemicals designated as active, unknown, or inactive, with the goal of a) improving model explanatory power; b) assessing the usefulness of potency-weighted chemical concentration calculations; and c) evaluating the extent of missing data. Because the model-dependent variables were log-transformed, the estimates from the models were transformed back to the linear scale and presented as percentage differences.

We did not have sufficient sample size or statistical power to determine the impact of the building materials renovation on dust bioactivities, given the many measured and unmeasured covariates about other chemicals and products in the rooms that we would have liked to control for. However, we conducted a simple model with a three-level variable to determine if renovation and product selection influenced bioactivities: a) spaces in older buildings (built before the first 2004 PBDE phase-out) and meeting historic, stringent flammability standards; b) partially renovated spaces that likely had less FR contamination from legacy building materials or furniture; and c) spaces in post-2004 or fully renovated buildings with furniture and carpet specified as free of all PFAS and chemical FRs.

Statistical significance was evaluated at the $\mathrm{\alpha }=0.05$ level. Suggestive evidence was evaluated at $\mathrm{\alpha }=0.10$. All analyses were conducted in R (version 3.6.1; R Core Development Team).

Results

Hormonal Activities of Dust Extracts

All 46 extracted dust samples were hormonally active in at least two of the cell-based assays. Approximately 38 of the extracted dust samples (83%) activated or suppressed all four nuclear hormone receptors, as shown in the full dose–response data in Figure S2. Specifically, 46 (100%) of the 46 samples suppressed $\text{PPAR}\mathrm{\gamma }$, 44 (96%) activated $\text{ER}\mathrm{\alpha }$, 41 (89%) suppressed $\text{TR}\mathrm{\beta }$, 40 (87%) suppressed AR, and 45 (98%) interfered with ${\mathrm{T}}_4$ binding to a serum transport protein at detectable levels (Table 1). All active dust extracts also usually exhibited dose–response monotonic relationships for each assay. Four dust extracts exceeded the maximal response (efficacy) observed for the endogenous estrogen hormone, $17\mathrm{\beta }\text{-estradiol}$ (Figures S2k,l,n). Forty-four samples (96%) either produced a response above half the efficacy of $17\mathrm{\beta }\text{-estradiol}$ ($n=21$) or could not be evaluated at the highest concentration in the dilution series because of cytotoxicity ($n=23$).

Table 1 presents the median hormonal activities in extracted dust samples, and Table S2 provides data for each sample. We observed detectable effects with as little as 3.7, 5.2, 16, or $18\mathrm{\mu }\mathrm{g}$ of extracted dust per well for $\text{ER}\mathrm{\alpha }$, $\text{PPAR}\mathrm{\gamma }$, AR, and $\text{TR}\mathrm{\beta }$, respectively, as well as $30\mathrm{\mu }\mathrm{g}$ of extracted dust per incubate for TTR activity (Table S3). Suppression of $\text{PPAR}\mathrm{\gamma }$ and $\text{TR}\mathrm{\beta }$ by the extracted dust samples were significantly correlated (Spearman $r=0.54$, $p<0.001$). Interference with TTR-${\mathrm{T}}_4$ binding was also significantly correlated with $\text{PPAR}\mathrm{\gamma }$ suppression ($r=0.43$, $p<0.01$), and moderately correlated with $\text{TR}\mathrm{\beta }$ suppression ($r=0.26$, $p=0.084$). All other pairs of nuclear hormone receptor activity were not significantly correlated, with correlation coefficients ranging between 0.13 and 0.25 ($\text{ER}\mathrm{\alpha }–\text{PPAR}\mathrm{\gamma }$: $r=0.21$, $p=0.16$; $\text{ER}\mathrm{\alpha }–\text{AR}$: $r=0.13$, $p=0.39$; $\text{ER}\mathrm{\alpha }–\text{TR}\mathrm{\beta }$: $r=0.24$, $p=0.11$; $\text{ER}\mathrm{\alpha }–\mathrm{T}4\text{-TTR}$: $r=0.25$, $p=0.10$; $\text{PPAR}\mathrm{\gamma }–\text{AR}$: $r=0.25$, $p=0.088$; $\text{AR}–\text{TR}\mathrm{\beta }$: $r=0.21$, $p=0.17$; AR–${\mathrm{T}}_4$-TTR: $r=0.21$, $p=0.16$).

Hormonal Potencies of Chemicals

Twelve of the 15 PFAS analytes, all 19 OPE analytes, and all 8 PBDE analytes were detected in our 46 dust samples. Median dust concentrations by chemical ranged from 0.918 to $193\text{\hspace{0.17em}ng}/\mathrm{g}$ for detected PFAS, 3.55 to $830\text{\hspace{0.17em}ng}/\mathrm{g}$ for PBDEs, and 11.1 to $\mathrm{15,300}\text{\hspace{0.17em}ng}/\mathrm{g}$ for OPEs (Table 2).

Table 2.

RPFs and potency-weighted exposure contributions for each chemical measured in the present study’s dust samples ($n=46$), using Tox21 data on ACCs and hit calls for the agonism/antagonism assays (Huang et al. 2016) or using the laboratory’s data on RPFs for the transport interference assay (Besselink 2020).

Chemical	Exposure levels in dust samples			Bioactivity classification (RPF) [Median percentage contribution to potency-weighted concentration sum of chemical class]
	Percentage of samples $>\text{MDL}$	Median [range (ng/g)]	Median percentage of class sum	$\text{PPAR}\mathrm{\gamma }$ antagonism (Huang et al. 2016)	Thyroid hormone receptor $\mathrm{\beta }$ antagonism (Huang et al. 2016)	Androgen receptor antagonism (Huang et al. 2016)	Estrogen receptor $\mathrm{\alpha }$ agonism (Huang et al. 2016)	Thyroid hormone transport interference (Besselink 2020)a
Per- and polyfluoroalkyl substances (PFAS)
PFHxA	97.8	193 ($<\text{MDL}–\mathrm{2,980}$)	66	Inactive	Inactive	Inactive	Inactive	Active (0.044) [19]
PFOS	97.8	15.2 ($<\text{MDL}–296$)	5.4	Active (0.55) [51]	Active (0.52) [50]	Inactive	Inactive	Active (0.85) [30]
PFOA	73.9	7.63 ($<\text{MDL}–\mathrm{1,520}$)	4.5	Active (0.4) [25]	Inactive	Inactive	Inactive	Active (0.37) [11]
PFHxS	63.0	1.82 ($<\text{MDL}–23.7$)	$<1$	Unknown	Unknown	Unknown	Unknown	Active (1) [3.5]
FOSA	60.9	3.26 ($<\text{MDL}–236$)	1.5	Inactive	Active (0.59) [14]	Inactive	Active (0.39) [100]	Active (0.33) [3.6]
PFHpA	52.2	0.918 (0–1,760)	$<1$	Inactive	Inactive	Inactive	Inactive	Active (0.35) [2.0]
PFPeA	32.6	$<\text{MDL}$ ($<\text{MDL}–455$)	$<1$	Unknown	Unknown	Unknown	Unknown	Active (0.013) [$<1$]
PFNA	30.4	$<\text{MDL}$ ($<\text{MDL}–\mathrm{1,480}$)	$<1$	Active (0.76) [2.3]	Active (0.47) [1.2]	Inactive	Inactive	Active (0.13) [$<1$]
PFBS	30.4	$<\text{MDL}$ ($<\text{MDL}–16.1$)	$<1$	Unknown	Unknown	Unknown	Unknown	Active (0.028) [$<1$]
PFDS	10.9	$<\text{MDL}$ ($<\text{MDL}–12.5$)	$<1$	Unknown	Unknown	Unknown	Unknown	Unknown
PFBA	4.35	$<\text{MDL}$ ($<\text{MDL}–155$)	$<1$	Unknown	Unknown	Unknown	Unknown	Active (0.0003) [$<1$]
PFDA	4.35	$<\text{MDL}$ ($<\text{MDL}–35.0$)	$<1$	Inactive	Active (0.29) [$<1$]	Inactive	Inactive	Active (0.033) [$<1$]
PFUnDA	0	$<\text{MDL}$ ($<\text{MDL}–<\text{MDL}$)	$<1$	Active (1) [4.2]	Inactive	Inactive	Inactive	Active (0.017) [$<1$]
PFDoDA	0	$<\text{MDL}$ ($<\text{MDL}–<\text{MDL}$)	$<1$	Unknown	Unknown	Unknown	Unknown	Active (0.0037) [$<1$]
N-MeFOSAA	0	$<\text{MDL}$ ($<\text{MDL}–<\text{MDL}$)	$<1$	Unknown	Unknown	Unknown	Unknown	Unknown
Polybrominated diphenyl ethers (PBDEs)
BDE-209	100	830 (34.8–13,000)	70	Unknown	Unknown	Inactive	Unknown	Unknown
BDE-99	100	124 (10.6–734)	11	Inactive	Active (0.63) [69]	Active (0.64) [52]	Inactive	Unknown
BDE-47	100	60.9 (6.55–1,470)	5.5	Inactive	Active (0.53) [31]	Active (1) [43]	Active (1) [100]	Unknown
BDE-100	100	26.9 (6.12–202)	2.4	Unknown	Unknown	Unknown	Unknown	Unknown
BDE-183	89.1	24.2 ($<\text{MDL}–817$)	2	Unknown	Unknown	Unknown	Unknown	Unknown
BDE-28	89.1	3.55 ($<\text{MDL}–104$)	$<1$	Unknown	Unknown	Unknown	Unknown	Unknown
BDE-153	84.8	18.9 ($<\text{MDL}–78.5$)	1.5	Inactive	Inactive	Active (0.46) [5.2]	Inactive	Unknown
BDE-154	80.4	10.6 ($<\text{MDL}–53.5$)	$<1$	Unknown	Unknown	Unknown	Unknown	Unknown
Organophosphate esters (OPEs)
TBOEP	100	15,300 (1,250–118,000)	65	Inactive	Active (1) [92]	Inactive	Inactive	Unknown
TCIPP	100	3,130 (675–139,000)	12	Inactive	Inactive	Inactive	Inactive	Unknown
TDCIPP	100	970 (220–6,440)	3.5	Inactive	Active (0.54) [2.9]	Active (0.84) [57]	Inactive	Unknown
TPHP	100	817 (238–10,600)	3.0	Active (0.74) [67]	Active (0.4) [2]	Active (0.47) [21]	Active (1) [76]	Unknown
TCEP	100	214 (2.31–3,170)	$<1$	Inactive	Inactive	Inactive	Inactive	Unknown
EHDPP	100	184 (4.65–2,480)	$<1$	Inactive	Active (0.54) [$>1$]	Active (0.42) [3.9]	Inactive	Unknown
IDDP	100	88.1 (0.699–612)	$<1$	Active (0.6) [5.4]	Active (0.64) [$<1$]	Active (0.5) [2.5]	Inactive	Unknown
TIBP	100	32.8 (5.28–804)	$<1$	Inactive	Active (0.45) [$<1$]	Inactive	Inactive	Unknown
TPP	100	11.1 (1.52–63.3)	$<1$	Unknown	Unknown	Unknown	Unknown	Unknown
CDPP	97.8	192 ($<\text{MDL}–\mathrm{11,500}$)	$<1$	Active (0.69) [15]	Inactive	Active (0.46) [6.3]	Active (0.72) [13]	Unknown
BPDP	97.8	88.4 ($<\text{MDL}–371$)	$<1$	Active (0.77) [5.6]	Active (0.52) [$<1$]	Active (0.52) [2.3]	Inactive	Unknown
TNBP	97.8	41.1 ($<\text{MDL}–\mathrm{1,060}$)	$<1$	Inactive	Inactive	Inactive	Active (0.76) [2.6]	Unknown
V6	95.7	23.2 ($<\text{MDL}–216$)	$<1$	Unknown	Unknown	Unknown	Unknown	Unknown
BDP	93.5	26.3 ($<\text{MDL}–494$)	$<1$	Unknown	Unknown	Unknown	Unknown	Unknown
TEHP	91.3	40.4 ($<\text{MDL}–\mathrm{1,680}$)	$<1$	Inactive	Inactive	Inactive	Active (0.68) [2.2]	Unknown
RDP	91.3	36.6 ($<\text{MDL}–\mathrm{3,270}$)	$<1$	Unknown	Unknown	Unknown	Unknown	Unknown
TMPP	91.3	28.8 ($<\text{MDL}–486$)	$<1$	Unknown	Unknown	Unknown	Unknown	Unknown
TEP	60.9	18.2 ($<\text{MDL}–416$)	$<1$	Inactive	Inactive	Inactive	Inactive	Unknown
TBPHP	19.6	$<\text{MDL}$ ($<\text{MDL}–10.7$)	$<1$	Unknown	Unknown	Unknown	Unknown	Unknown

Note: Chemical concentrations $<\text{MDL}$ in dust were substituted with the MDL divided by 2 for analyses and calculations of class sums. ACCs, activity concentrations at cutoff; BDE-28, polybrominated diphenyl ether–28; BDE-47, polybrominated diphenyl ether–47; BDE-99, polybrominated diphenyl ether–99; BDE-100, polybrominated diphenyl ether–100; BDE-153, polybrominated diphenyl ether–153; BDE-154, polybrominated diphenyl ether–154; BDE-183, polybrominated diphenyl ether–183; BDE-209, polybrominated diphenyl ether–209; BDP, bisphenol A bis(diphenyl phosphate); BPDP, tert-butylphenyl diphenyl phosphate; CDPP, cresyl diphenyl phosphate; EHDPP, 2-ethylhexyl diphenyl phosphate; FOSA, perfluorooctane sulfonamide; IDDP, isodecyl diphenyl phosphate; MDL, method detection limit; N-MeFOSAA, $n$-methyl perfluorooctane sulfonamidoacetic acid; PFBA, perfluorobutanoate; PFBS, perfluorobutane sulfonate; PFDA, perfluorodecanoate; PFDoDA, perfluorododecanoate; PFDS, perfluorodecane sulfonate; PFHpA, perfluoroheptanoate; PFHxA, perfluorohexanoate; PFHxS, perfluorohexane sulfonate; PFNA, perfluorononanoate; PFOA, perfluorooctanoate; PFOS, perfluorooctane sulfonate; PFPeA, perfluoropentanoate; PFUnDA, perfluoroundecanoate; $\text{PPAR}\mathrm{\gamma }$, peroxisome proliferator-activated receptor lowercase gamma 2; RPFs, relative potency factors; RDP, resorcinol bis(diphenyl phosphate); TBOEP, tris(2-butoxyethyl) phosphate; TBPHP, tris($p$-tert-butylphenyl) phosphate; TCEP, tris(2-chloroethyl) phosphate; TCIPP, tris(1-chloro-2-propyl) phosphate; TDCIPP, tris(1,3-dichloro-2-propyl) phosphate; TEHP, tris(2-ethylhexyl) phosphate; TEP, triethyl phosphate; TIBP, tri-iso-butyl phosphate; TMPP, tris(methylphenyl) phosphate; TNBP, tri-n-butyl phosphate; TPHP, triphenyl phosphate; TPP, tripropyl phosphate; unknown, the chemical did not have available screening data;V6, tetrakis(2-chloroethyl) dichloroisopentyl diphosphate.

^aAnalysis conducted by the same laboratory as assayed our dust samples (Besselink 2020).

Many of the 42 chemical analytes measured in this study were active in previous chemical screening assays (Besselink 2020; Huang et al. 2016) that employed similar gene reporter assays as those used in this study but not necessarily the same cell lines (except for TTR-${\mathrm{T}}_4$ interference) (Table 2). Twenty-nine percent of the analytes (12 of 42) were not screened in any of the five assays and therefore had no comparative data. Of the chemical analytes with available screening data, 87% (26 of 30) were found to be active in at least one of the five end points we measured in dust. Of all pairs of chemicals and four nuclear hormone receptor assays, 21% (36 of 168 pairs) exhibited evidence for activity, 36% (61 of 168) were classified as inactive, and 42% (71 of 168) were not analyzed.

For each chemical class, a few chemicals dominated the potency-weighted concentration profiles in dust. For example, TBOEP was responsible for a median 92% of the potency-weighted $\mathrm{\Sigma }\text{OPEs}$ for $\text{TR}\mathrm{\beta }$ suppression (Table 2) because it was the OPE with the highest potency toward that end point in Tox21 data and with the highest concentration in dust. By contrast, TBOEP and TCIPP (the two OPEs detected at the highest concentrations in dust) were designated as inactive toward AR suppression, so TDCIPP dominated the OPE profile for just the active chemicals. As another example, BDE-209 was detected at substantially higher concentrations in dust than any other measured PBDE, but other PBDEs had higher or better-known bioactivities in Tox21 screening assays and thus were weighted to contribute more to the total potency-weighted sum of the class for various assay end points than BDE-209. Figure 1 visualizes examples of these unweighted vs. potency-weighted chemical profiles.

Figure 1.

Comparison of profiles of chemicals with any vs. active vs. unknown designations for select pairs of assay end points and chemical classes, using Tox21 high-throughput screening data (Table 2) (Huang et al. 2016). (A) OPEs and androgen receptor antagonism, (B) PBDEs and estrogen receptor agonism, and (C) PBDEs and thyroid hormone receptor antagonism. Values below the limit of detection were replaced with half the limit. Unlabeled stacked bars represent chemicals that did not make appreciable contributions ($<4\%$) to the geometric mean concentrations of that chemical class. Note: BDE-47, decabromodiphenyl ether–47; BDE-99, decabromodiphenyl ether–99; BDE-209, decabromodiphenyl ether–209; BDP, bisphenol A bis(diphenylphosphate); CDPP, cresyl di phenyl phosphate; EHDPP, 2-ethylhexyl diphenyl phosphate; OPEs, organophosphate esters; PBDEs, polybrominated diphenyl ethers; RDP, resorcinol bis(diphenyl phosphate); TBOEP, tris(2-butoxyethyl) phosphate; TCIPP, tris (1-chloro-2-propyl) phosphate; TDCIPP, tris(1,3-dichloro-2-propyl) phosphate; TMPP, tris(2,4,6-trimethoxyphenyl)phosphine; TPHP, triphenyl phosphate; TPP, triphenylphosphine; V6, 2,2-bis(chloromethyl)trimethylene bis(bis(2-chloroethyl)phosphate).

The chemical analytes with the maximum individual potency for each of the five assay end points consisted of two PFAS, two OPEs, and a PBDE. Active OPEs tended to have both high median concentrations in our dust samples and high RPFs in the screening data compared with the other two chemical classes, whereas active PFAS tended to have lower median dust concentrations and a wide range of RPFs (Figure S3).

Unweighted Effects of Chemicals on Hormonal Activities of Dust

Table 3 presents models of associations between unweighted chemical class concentrations and hormonal activities of the dust extracts. For an interquartile range (IQR) increase in $\mathrm{\Sigma }\text{PFAS}$ concentration (from the 25th to the 75th percentile), dust extracts had 7.0% significantly higher levels of suppression of $\text{TR}\mathrm{\beta }$ activation [(95% confidence interval (CI): 1.7%, 13%), $p<0.01$] and 5.6% significantly higher AR suppression [(95% CI: 0.14%, 11%), $p<0.05$], adjusted for $\mathrm{\Sigma }\text{OPEs}$ and $\mathrm{\Sigma }\text{PBDEs}$. For an IQR increase in $\mathrm{\Sigma }\text{OPEs}$, dust extracts had an estimated 26% significantly higher $\text{TR}\mathrm{\beta }$ suppression [(95% CI: 3.8%, 52%), $p<0.05$], 25% significantly higher $\text{PPAR}\mathrm{\gamma }$ suppression [(95% CI: 7.2%, 46%), $p<0.01$], and 39% significantly higher interference of thyroid hormone transport [(95% CI: 7.0%, 79%), $p<0.05$], adjusted for $\mathrm{\Sigma }\text{PFAS}$ and $\mathrm{\Sigma }\text{PBDEs}$. Unweighted $\mathrm{\Sigma }\text{PBDEs}$ were not significantly associated with hormonal potencies of the extracted dust samples in this test system. As shown in Table 3, models using unweighted chemical concentrations explained 22%, 17%, 14%, and 11% of the variability in dust suppression of $\text{TR}\mathrm{\beta }$, $\text{PPAR}\mathrm{\gamma }$, TTR-${\mathrm{T}}_4$ binding, and AR, respectively.

Table 3.

Results of linear regression models of percentage differences in hormonal activities ($\mathrm{\mu }\mathrm{g}\text{-eq}/\mathrm{g}$ or $\text{ng-eq}/\mathrm{g}$) for an interquartile range (IQR) increase in summed concentrations $(\text{ng}/\mathrm{g})$ of three chemical classes in 46 dust samples: per- and polyfluoroalkyl substances (PFAS), organophosphate esters (OPEs), and polybrominated diphenyl ethers (PBDEs).

	Thyroid hormone receptor $\mathrm{\beta }$ antagonism ($\mathrm{\mu }\mathrm{g}\text{-eq}/\mathrm{g}$)				$\text{PPAR}\mathrm{\gamma }$ antagonism ($\mathrm{\mu }\mathrm{g}\text{-eq}/\mathrm{g}$)				Androgen receptor antagonism ($\mathrm{\mu }\mathrm{g}\text{-eq}/\mathrm{g}$)				Estrogen receptor $\mathrm{\alpha }$ agonism ($\text{ng-eq}/\mathrm{g}$)				Thyroid hormone transport interference ($\mathrm{\mu }\mathrm{g}\text{-eq}/\mathrm{g}$)
Covariate	$\mathrm{\Delta }\text{IQR}$ (%)	$p$	IQR	$n$	$\mathrm{\Delta }\text{IQR}$ (%)	$p$	IQR	$n$	$\mathrm{\Delta }\text{IQR}$ (%)	$p$	IQR	$n$	$\mathrm{\Delta }\text{IQR}$ (%)	$p$	IQR	$n$	$\mathrm{\Delta }\text{IQR}$ (%)	$p$	IQR	$n$
Model 1: unweighted effects of chemicals	($R^2=0.22$)				($R^2=0.17$)				($R^2=0.11$)				($R^2=0.0014$)				($R^2=0.14$)
$\sum$ PFAS	7.00**	0.01	267	15	1.06	0.6	267	15	5.58*	0.04	267	15	$–0.674$	0.8	267	15	1.86	0.6	267	15
$\sum$ OPEs	25.7*	0.02	29,000	8	25.2**	0.005	29,000	8	8.83	0.4	29,000	8	0.193	1	29,000	8	38.5*	0.01	29,000	8
$\sum$ PBDEs	$–3.93$	0.4	1,020	19	$–1.28$	0.7	1,020	19	$–3.8$	0.4	1,020	19	$–0.313$	1	1,020	19	$–4.22$	0.5	1,020	19
Model 2: potency-weighted effects of PFAS	($R^2=0.22$)				($R^2=0.013$)				($R^2=0.083$)				($R^2=0.039$)				($R^2=0.0014$)
Potency-weighted $\sum$ PFAS	27.5*	0.01	116	4	$–1.21$	0.7	57	4	No RPFsa	—	0	0	$–5.05$	0.2	13.2	1	0.755	0.8	44.1	13
$\sum$ PFAS of unknown potency	0.7	0.9	14.7	7	3.37	0.5	14.7	7	0.856	0.9	36.2	8	$–2.63$	0.7	14.7	7	Few detectsc	—	—	2
$\sum$ PFAS designated inactiveb	$–12.0$	0.2	218	4	0.182	1	172	4	5.11	0.3	261	7	1.88	0.8	261	7	NA	—	0	0
Model 3: potency-weighted effects of OPEs	($R^2=0.18$)				($R^2=0.18$)				($R^2=0.086$)				($R^2=0.016$)				($R^2=0.12$)
Potency-weighted $\sum$ OPEs	26.8#	0.08	81,500	7	$–2.22$	0.5	2,450	4	$–14.1$	0.3	5,500	6	$–4.14$	0.5	3,190	4	No RPFsa	—	0	0
$\sum$ OPEs of unknown potency	10.9*	0.02	189	6	1.97	0.6	189	6	12.7#	0.06	243	7	$–0.543$	0.9	189	6	34.4*	0.02	29,000	19
$\sum$ OPEs designated inactiveb	2.54	0.2	5,450	6	24.6**	0.004	28,600	9	7.2	0.5	28,600	6	1.59	0.9	28,400	9	NA	—	0	0
Model 4: potency-weighted effects of PBDEs	($R^2=0.18$)				($R^2=0.024$)				($R^2=0.022$)				($R^2=0.071$)				($R^2=0.00012$)
Potency-weighted $\sum$ PBDEs	20.2*	0.02	316	2	No RPFsa	—	0	0	$–2.78$	0.6	301	3	7.71#	0.08	173	1	No RPFsa	—	0	0
$\sum$ PBDEs of unknown potency	$–4.55$	0.3	963	5	0.210	1	963	5	4.41	0.5	66.2	4	$–1.59$	0.7	963	5	$–0.423$	0.9	1,020	8
$\sum$ PBDEs designated inactiveb	$–3.05$	0.8	14.5	1	4.82	0.3	160	3	$–2.16$	0.6	909	1	Collineard	—	94.1	2	NA	—	0	0

Note: Assay activities were log-transformed in the models, but the estimates are transformed and presented as the percentage difference in activity for an IQR increase in the chemical covariate. —, not available; IQR, interquartile range; $n$, number of chemicals contributing to the sum for that covariate; NA, not applicable; $\text{PPAR}\mathrm{\gamma }$, peroxisome proliferator-activated receptor lowercase gamma 2; RPF, relative potency factor; $\mathrm{\Delta }\text{IQR}$, change in interquartile range. ^#, $p<0.10$; *, $p<0.05$; **, $p<0.01$.

^aNone had RPFs available, so this covariate could not be included in the model.

^bDesignated as inactive in Tox21 assays of the chemicals (for antagonism/agonism) or in the exact luciferase assays by the laboratory (for transport interference).

^cThe few PFAS in this unknown assay classification were too infrequently detected and, thus, were excluded from the model.

^dThis covariate was very collinear (Spearman $r=0.9$), so it was excluded from the model.

Potency-Weighted Effects of Chemicals on Hormonal Activities of Dust

In models with potency-weighted chemical concentrations (Table 3), an IQR increase in potency-weighted $\mathrm{\Sigma }\text{PBDEs}$ in dust extracts was associated with 20% significantly higher levels of $\text{TR}\mathrm{\beta }$ suppression [(95% CI: 3.3%, 40%), $p<0.05$] and 7.7% higher $\text{ER}\mathrm{\alpha }$ activation [(95% CI: $-0.91\%$, 17%), $p<0.08$] with borderline statistical significance, adjusted for sums of PBDEs with inactive or unknown potencies.

For an IQR increase in potency-weighted $\mathrm{\Sigma }\text{PFAS}$, dust extracts had 28% significantly higher suppression of $\text{TR}\mathrm{\beta }$ [(95% CI: 5.4%, 54%), $p<0.01$], adjusted for sums of PFAS with inactive or unknown potencies. There were no PFAS with available RPFs for AR suppression, and sums of PFAS with inactive or unknown potencies did not yield significant associations for that end point.

The $\mathrm{\Sigma }\text{OPEs}$ with unknown potencies had 11% significantly higher $\text{TR}\mathrm{\beta }$ suppression [(95% CI: 1.6%, 21%), $p<0.05$] for an IQR increase in concentration, adjusted for potency-weighted and inactive $\mathrm{\Sigma }\text{OPEs}$. In that model, the potency-weighted $\mathrm{\Sigma }\text{OPEs}$ had a similar effect estimate as the unweighted $\mathrm{\Sigma }\text{OPEs}$ did on $\text{TR}\mathrm{\beta }$ inhibition in the previous model, with borderline statistical significance [27% per IQR increase (95% CI: $-3.1\%$, 66%), $p=0.08$]. An IQR increase in $\mathrm{\Sigma }\text{OPEs}$ with unknown potencies also had 13% higher suppression of AR, with borderline significance [(95% CI: $-0.53\%$, 28%), $p=0.06$]. For TTR-${\mathrm{T}}_4$ binding interference, $\mathrm{\Sigma }\text{OPEs}$ with unknown potencies (which consisted of all OPE analytes) in its own model had a similar impact as unweighted $\mathrm{\Sigma }\text{OPEs}$ in the previous model with all three chemical classes [34% (95% CI: 5.3%, 72%), $p<0.05$]. Finally, there was only one significant association between any inactive chemicals and an end point. For an IQR increase in $\mathrm{\Sigma }\text{OPEs}$ designated as inactive in Tox21 data, dust extracts had 25% higher suppression of $\text{PPAR}\mathrm{\gamma }$ [(95% CI: 7.5%, 45%), $p<0.01$].

Each chemical class individually explained substantial amounts of variation in $\text{TR}\mathrm{\beta }$ or $\text{PPAR}\mathrm{\gamma }$ inhibition (Table 3). The models for PFAS, OPEs, and PBDEs explained between 18% and 22% of the variability in $\text{TR}\mathrm{\beta }$ antagonism in dust extracts. The model for OPEs explained 18% of the variability for $\text{PPAR}\mathrm{\gamma }$ suppression and 12% for TTR-${\mathrm{T}}_4$ binding interference. Variability in $\text{ER}\mathrm{\alpha }$ activation was not as highly explained (at most 7.1%).

Effects of Room Factors on Hormonal Activities of Dust

In secondary, albeit statistically underpowered, modeling, common spaces had 82%, 63%, and 98% significantly higher dust levels of $\text{TR}\mathrm{\beta }$ [(95% CI: 7.7%, 210%), $p<0.05$], $\text{PPAR}\mathrm{\gamma }$ [(95% CI: 8.3%, 140%), $p<0.05$], and TTR-${\mathrm{T}}_4$ interference [(95% CI: 0.11%, 290%), $p=0.05$] respectively than classrooms did, adjusted for renovation materials status. Renovation status did not achieve statistically significant effects (Table S4).

Discussion

Hormonal Activities of Dust Samples

In our study of chemical mixtures and bioactivities in dust in buildings across a university, we found that all extracted dust samples interfered with at least two hormone receptors or hormone transporters. Every sample suppressed the activation of $\text{PPAR}\mathrm{\gamma }$, disruption of which has been implicated in obesity, diabetes, and other metabolic disorders based on in vivo and in vitro research (Grimaldi et al. 2015). The vast majority of samples also disrupted estrogen and androgen receptors, which play important roles in reproductive health and development (Delfosse et al. 2015; Ren and Guo 2013; Shanle and Xu 2011). Finally, all extracted dust samples interfered with action of thyroid hormones, either through suppression of the receptor or displacement of ${\mathrm{T}}_4$ from a serum transport protein. Disruption of thyroid hormones has been established as a mechanism for impaired development, metabolism, brain function (Grimaldi et al. 2015; Ishihara et al. 2003; Zoeller et al. 2002), and cardiovascular health (Alexander et al. 2017). Hormonal effects of the dust samples were observed with as little as 4–$18\mathrm{\mu }\mathrm{g}$ of extracted dust across the five cell-based assays, which cannot be translated to hormonal effects of actual absorbed human doses but does occur at amounts relevant to external exposure (a person ingests an estimated average $\mathrm{20,000}\mathrm{\mu }\mathrm{g}$ of indoor dust per day) (U.S. EPA 2017).

Effects of Chemicals on Hormonal Activities of Dust

Many of the individual chemicals we measured in the dust samples were also active toward the hormone receptors when tested in isolation by the laboratory (Besselink 2020) or by Tox21 (Huang et al. 2016), and we found that chemical profiles significantly explained the hormonal activities of dust extracts in the cell-based assays. The potency of dust to inhibit $\text{TR}\mathrm{\beta }$ activation was significantly explained by all three classes (PFAS, OPEs, and PBDEs), which have been associated with thyroid disruption and developmental impairment in human epidemiologic studies (Boas et al. 2012; Czerska et al. 2013; Doherty et al. 2019b; Liew et al. 2018; Linares et al. 2015; Ren and Guo 2013; Vuong et al. 2018; Wang et al. 2020). The OPEs were also significantly associated with thyroid-interfering activities in dust via substantial displacement of ${\mathrm{T}}_4$ thyroid hormone from the TTR transporter, which prevents the thyroid hormone from being delivered to essential brain issues or the placenta and makes ${\mathrm{T}}_4$ more readily excreted (Grimm et al. 2013). OPEs have been under-investigated as potential disruptors of thyroid hormone binding to that transporter, unlike PFAS and PBDEs, but two recent studies have reported ${\mathrm{T}}_4$ displacement from TTR by certain OPEs such as TPHP and TCIPP (Rosenmai et al. 2021; Zhao et al. 2017).

$\text{PPAR}\mathrm{\gamma }$ suppression by dust extracts was also significantly associated with OPE concentrations. Some OPEs were linked in epidemiologic studies to obesity and related metabolic effects, although more human research is needed (Boyle et al. 2019; Cano-Sancho et al. 2017; Wang et al. 2019). There were statistically significant or close-to-significant associations of PFAS and OPE levels in dust with AR inhibition, and in epidemiologic studies, PFAS may be associated with birth size (Darrow et al. 2013; Xiao et al. 2020) and OPEs with brain development (Doherty et al. 2019a, 2019b) and infertility (Carignan et al. 2018; Meeker and Stapleton 2010).

There was some evidence that $\text{ER}\mathrm{\alpha }$ activation by dust extracts was associated with the potency-weighted concentration of all PBDEs, which was driven by BDE-47. Epidemiologic studies have linked PBDEs to adverse effects on fertility, pregnancy outcomes, and development (Choi et al. 2019; Czerska et al. 2013; Linares et al. 2015; Mumford et al. 2015; Ren and Guo 2013). Of all the end points, $\text{ER}\mathrm{\alpha }$ agonism had the least amount of variability explained by the chemical classes, perhaps because there are so many other unmeasured chemicals in dust that influence estrogenicity (including unmeasured FRs, phthalate plasticizers, bisphenol plasticizers, chlorinated pesticides, and polychlorinated biphenyls) (Shanle and Xu 2011; Vandermarken et al. 2016).

We found that $\text{PPAR}\mathrm{\gamma }$ suppression was significantly correlated with both $\text{TR}\mathrm{\beta }$ suppression and ${\mathrm{T}}_4$ transport interference by dust extracts. This indicates that chemicals can be bioactive against several hormone receptors and transporters. Regulatory assessments often treat each chemical and mechanism separately, but interactions are more complex in our bodies (Kienzler et al. 2016). The same chemical can bind multiple different receptors (at different affinities), different chemicals can influence signaling of the same receptor, different receptors can contribute to the same adverse health outcome (Delfosse et al. 2015; Grimaldi et al. 2015), and the presence of multiple chemicals can jointly elicit greater-than- or less-than-expected effects (Kollitz et al. 2018; Orton et al. 2014).

In secondary analyses, we found that common spaces had significantly higher activities against $\text{TR}\mathrm{\beta }$, $\text{PPAR}\mathrm{\gamma }$, and ${\mathrm{T}}_4$ transport in dust than classrooms. Given the higher utilization and foot traffic in common areas, the more potent activities may indicate that people track in, and bring in personal products with, hormone-disrupting contaminants.

Method Evaluation of Potency-Weighted Concentrations of Chemical Classes

Potency-weighted chemical concentration assessments helped us better characterize the impacts of chemical mixtures on hormonal activities of dust. For example, the unweighted sum of all PBDEs in dust did not have statistically significant effects on the assay end points. However, one dominant chemical, BDE-209, contributed a median 70% to the PBDE concentrations in dust but was only designated as unknown or inactive in the chemical screening assays. When we instead conducted models with potency-weighted sums of bioactive PBDEs based on Tox21 data (which excluded BDE-209), we found significant or near-significant associations with interference of $\text{TR}\mathrm{\beta }$ and $\text{ER}\mathrm{\alpha }$. Therefore, in the unweighted models, the null effect of BDE-209 was masking the effect of individual PBDEs that had lower dust concentrations but might be more potent in the screening assays. In addition, the potency-weighted exposure models showed suggestive evidence that the sum of OPEs with missing Tox21 screening data was associated with AR suppression. We did not observe this association based on the unweighted concentration of all OPEs, likely because two specific OPEs heavily dominated the chemical profiles for all OPEs but were classified as inactive in the Tox21 database. In addition, the sum of OPEs with unknown activities was significantly associated with dust suppression of $\text{TR}\mathrm{\beta }$. These findings support the importance of efforts to continue high-throughput screening of chemicals in cell-based assays and of using screening results to weight chemical concentrations in samples by chemical potencies.

High-throughput screening data will become increasingly useful as future studies analyze samples for more and more chemicals, such as with nontargeted laboratory approaches (Hilton et al. 2010). Potency-weighted chemical exposure indicators can a) act as more biologically relevant covariates in models; b) improve the explanatory power of statistical models to prioritize causative chemicals of concern; c) reduce the dimensions of mixture data; and d) parse the contributions of chemicals with known bioactivities vs. with missing data vs. with inactivity when assayed in isolation.

However, models of the unweighted chemical exposures are still also useful for several reasons. The unweighted sum of a chemical class does not exclude chemicals that did not reach an active designation when tested in isolation but that may enhance (or reduce) mixture effects when present with other chemicals (Kienzler et al. 2016; Kollitz et al. 2018; Orton et al. 2014). This could be one possible explanation for why the concentrations of OPEs designated as inactive in Tox21 screening assays had statistically significant associations with $\text{PPAR}\mathrm{\gamma }$ antagonism in dust samples. That result could also be partly explained by the fact that the inactive-designated OPEs are sometimes correlated with other unmeasured FRs or phthalate plasticizers (Bi et al. 2018; Castorina et al. 2017; Kassotis et al. 2019; Kollitz et al. 2018) or could be influenced by the small error that could be introduced from using designations based on a $\text{PPAR}\mathrm{\gamma }$ antagonism assay with a different reporter gene (beta-lactamase) and human cell line tissue type (embryonic kidney) than our dust assays. However, another previous study that measured $\text{PPAR}\mathrm{\gamma }$ antagonism of OPEs using the same luciferase reporter gene assays and human osteosarcoma cell lines as our study was consistent in classifying the four Tox21-inactive chemicals found in our samples at the highest concentrations as inactive (Suzuki et al. 2013). The Tox21 data on the other three nuclear hormone receptor assays used the same luciferase reporter. The modeling of unweighted chemical concentrations has the advantage of not introducing error in calculating potency-weighted sums for chemical classes, although we do in effect have to assume that each chemical has equal potency.

Comparison with Previous Literature

Only two previous studies quantified bioactivities of indoor dust in some of the same cell-based assays with comparable units. Vandermarken et al. (2016) measured estrogenic activity in dust in 12 kindergartens in Belgium using a similar luciferase cell-based assay and found a median potency of $1.34\text{ng-estradiol}/\mathrm{g}\text{-dust}$ (range: 0.426–8.71). These bioactivities are comparable to, if not slightly less potent than, our study samples ($\text{median}=1.76\text{ng-estradiol}/\mathrm{g}\text{-dust}$; range: 0.287–22.0). Suzuki et al. (2013) used three equivalent assays of 13 dust samples pooled from 66 homes in five countries. They reported results as micrograms of dust per well needed to achieve 5% of the maximal response of the reference compound; for comparison, we converted our results to micrograms per well to achieve an effect above the LOQ. Our results were similar for $\text{ER}\mathrm{\alpha }$ activation [our study: $\text{median}=32.0\mathrm{\mu }\mathrm{g}/\text{well}$ (range among active: 3.66–201); Suzuki et al. (2013): 39 (12–120)]; AR inhibition [our study: 56.2 (16.1–172); Suzuki et al. (2013): 72 (38–120)]; and $\text{PPAR}\mathrm{\gamma }$ inhibition [our study: 29.3 (5.20–121); Suzuki et al. (2013): 70 (11–120)].

The chemical concentrations in dust from university common spaces, offices, and classrooms in the present study had some differences compared with previous studies mostly conducted in homes. Levels of legacy PFOA and PFOS tended to be lower in our study, whereas a common replacement, PFHxA, was higher, reflective of the substitution of chemicals within the PFAS class over time. Legacy PBDE levels in dust in our study also tended to be lower in alignment with the phase-out of these FRs over time. Finally, dust concentrations of several OPEs in our study were higher than or comparable to previous studies. So, the buildings sampled in our study may reflect more modern profiles of building material chemicals in indoor environments. Further discussion and data visualization on the measured chemical concentrations are described in another paper (Young et al. 2021).

Strengths and Limitations

The present study was novel in the use of cell-based assays of indoor dust as an inexpensive, rapid, holistic, health-driven method to evaluate toxic loads in buildings from chemicals. To our knowledge, this is the first study to measure several important chemical classes and to use Tox21 data to develop potency-weighted concentrations of chemicals in dust to identify important chemical contributors. A key strength is also the summary of bioactivities across a diverse range of hormonal activity end points in the largest sample size of indoor dust samples ($n=46$). We used units relative to reference compounds (in micrograms of reference compound equivalents per gram of dust) that can be used as benchmarks in future studies. Finally, luciferase cell-based assays did not measure only receptor binding but, rather, actual transcriptional effects due to activation or suppression of receptors upon binding.

Study limitations include that many unmeasured hormone-disrupting chemicals in the dust may have influenced the assay results (Mitro et al. 2016; Shanle and Xu 2011). Given the sample size, we had too limited statistical power to be able to develop complex models with many covariates at once. In addition, the cell-based assays were conducted on dust samples collected in a different split of each room than the dust samples for the chemical analyses were (although collected at the same time), so the natural heterogeneity of dust may have limited the explanatory power of our models.

The gene reporter assays from the Tox21 data used different cell types (although still mostly human) compared with our assays. Chemicals can sometimes have slightly different effects in different cell lines depending on the species and tissue type (Sonneveld et al. 2006; Windal et al. 2005). Suzuki et al. (2013) measured $\text{ER}\mathrm{\alpha }$ agonism, $\text{PPAR}\mathrm{\gamma }$ antagonism, and AR antagonism of some of our analytes in the exact same assays and human cell lines as our study, and they had discordance with the Tox21 data for 6 of 31 chemical–assay pairs. Acknowledging some level of uncertainty, high-throughput chemical screening data are still useful for understanding large patterns in the types of chemicals contributing to bioactivities in complex environmental mixtures. Although a causal link between cell-based assays and human health effects has not been determined, the assays are useful for identifying key chemical characteristics that indicate potential endocrine-disrupting activity, and some studies have found that results from in vitro assays of well-studied chemicals accurately reflect known health effects (La Merrill et al. 2020; Rotroff et al. 2013; Schenk et al. 2010).

Conclusions

The present study found that all indoor dust samples were hormonally active and that PFAS, PBDEs, and OPEs significantly contributed to hormonal activities. Public high-throughput chemical screening data was useful for incorporating both potency and exposure into enhanced evaluations of chemical drivers of hormonal activities in dust. Because dust is a complex, hormonally potent mixture of many endocrine-disrupting compounds, we need more research to identify important contaminants and evaluate interventions to reduce them indoors.