Ethics committee approval was obtained from Swiss ethics (Commission cantonale d’éthique de la recherche sur l’être humain) in 2022 for this nonclinical human study (2022-01567). Healthy women and men were recruited as participants in our study. Each participant signed a written informed consent form before inclusion in the study. The participants will be reimbursed for their time and inconvenience according to the Swiss guidelines.

The choice of solvents to be included in the study will be based on the amount annually placed on the European market and the number of products registered containing known glycol ethers. The selected organic solvents will be applied to various in vitro models to determine their neurotoxicity. They must be amphiphilic to be solubilized in the cell culture media. We established the following solvent selection criteria: (1) used or produced >1 metric ton per year, (2) incorporated in numerous industrial and commercial products, and (3) water solubility. This selection process involves consulting government databases and contacting different industry sectors. We will start the project concomitantly by testing 2 solvents of the P series, propylene glycol methyl ether (PGME), for which we have already developed a toxicokinetic model, and propylene glycol butyl ether. We will test 1 additional solvent from the E series with the in vitro test system, namely, ethylene glycol methyl ether (EGME), which has been banned for use in cosmetic products in Europe [29].

The study is organized into 5 work packages (WPs). The information workflow between the WPs is shown in Figure 2. All results collected from the abovementioned systems will contribute to refining the toxicokinetic model we previously developed for PGME [30]. The toxicokinetic parameters of the solvent and the metabolites will then be characterized in human volunteers after exposure to PGE vapors under controlled conditions. These results will be used to calibrate and expand our toxicokinetic model [30]. The toxicokinetic model will be constructed to predict brain concentrations of selected solvents and, consequently, will include a brain compartment to predict the target organ solvent and metabolite concentrations. BBB parameters such as barrier transport, transport of the solvent once in the brain, and solvent-brain binding will be incorporated. Solvent neurotoxicity may depend on the metabolic modifications of the substances; therefore, we will incorporate the parameters for the parent compound and the metabolites found in the in vitro liver system. The data necessary to build the model will be retrieved from peer-reviewed scientific literature for tissue:blood partition coefficients (PCs) [31] following the fit-for-purpose dose-response analysis approach. The toxicokinetic model should be able to predict human brain concentration for each of the tested solvents after inhalation exposure, given the air concentration of vapors and duration of exposure. The simulated human brain concentrations will then be compared with the no observed adverse effect concentrations (NOAECs) obtained from the neurotoxicity in vitro system (BrainSpheres). In addition, the toxicokinetic model will be used to predict solvent air concentrations that are unlikely to lead to brain concentration equal to or superior to the brain NOAECs using reverse dosimetry. The specific aims of each WP are summarized in Table 1.

Testing strategies are needed to evaluate the neurotoxicity of chemicals in a more cost-effective, efficient, and ethical manner. Participating in an international effort, we developed a 3D human-induced pluripotent stem cells (hiPSC)–derived brain model containing several subtypes of neurons, astrocytes, and oligodendrocytes [24]. This system allows the cells to reach a high level of differentiation and cellular maturation, exemplified by the presence of functional synapses and compact myelin. The presence of myelin is important for this project because solvents more easily target lipid-rich structures [32]. This 3D human brain model has already proven its usefulness for neurotoxicity testing [33-36]. We hypothesized that glycol ethers are neurotoxic. Therefore, we propose to take advantage of our hiPSC-derived BrainSpheres model to study the neurotoxicity induced by uncharacterized glycol ethers present on the market, which will be compared with the neurotoxicity of well-characterized solvents known to induce human encephalopathy.

Solvents are data-poor substances. It was originally hypothesized that they exert their toxic effects largely through nonspecific physicochemical effects that modulate membrane fluidity and perturb the hydrophobic force regulating macromolecular interactions [37]. However, recent evidence supports the view that solvents interact with lipophilic areas on protein receptors [38,39]. They have also been shown to induce lipid peroxidation, leading to mitochondrial dysfunction, failure of electron transport, and energy production [40,41]. In this study, omics (eg, proteomics, metabolomics, and lipidomics) technology will be used to decipher the mechanisms of glycol ether neurotoxicity and to identify potential biomarkers of toxic effects.

Primary 3D hiPSC-derived brain cell cultures will be prepared and maintained as previously described [24]. This model contains neurons that form synapses, astrocytes, and oligodendrocytes myelinating the axons (Figure 3). Cytotoxicity will be determined by a resazurin assay after repeated exposure (7 d) to the selected glycol ethers (parent compounds and metabolites). Gene expression for cell type–specific genes, markers of synapses and myelin, and markers of cell stress will be quantified by quantitative reverse transcription polymerase chain reaction at concentrations of solvents under half-maximal effective concentration (EC50) for cytotoxicity. Immunostaining will be performed to assess the effects of solvents on synapses, myelin, and astrocyte reaction, and immunofluorescence will be quantified. NOAECs (Figure 2) will be determined for all tested endpoints, as previously shown for gene expression [42]. Brain cell cultures will also be exposed to the metabolites of PGME, propylene glycol butyl ether, and EGME and to the metabolites of the newly selected uncharacterized solvents, potentially produced by liver metabolism (WP3).

To establish the in vitro distribution kinetics of selected solvents necessary for toxicokinetic modeling, 3D brain cell cultures and medium will be collected 3, 6, 24, and 48 hours after the first exposure and after the last exposure of the repeated treatment. The solvent and its main metabolites (if relevant) will be quantified to establish a time course of disappearance from the medium and appearance in the cells as well as to assess the potential accumulation for the entire period of exposure. The fraction bound to culture plates’ plastic will be quantified after desorption. In silico modeling of glycol ethers in vitro distribution kinetics will then be developed. This model will be able to predict the change in cell-associated concentrations of solvents in BrainSpheres with time, as previously shown for amiodarone [43].

We previously established the zebrafish as a predictive vertebrate screening model to study the systemic circulation and tissue distribution of particulate drug carriers [44,45]. At 72 hours postfertilization, zebrafish embryos have a functional CNS and, presumably, a fully functional BBB. Anatomical structures such as the vascular endothelium can be visualized using transgenic fish lines expressing fluorescent proteins (Figure 4). Defined exposure of the zebrafish can be achieved by the simple addition of solvents to the fish incubation medium within a closed container. Other advantages of the model include the possibility of studying BBB functionality under physiological conditions in vivo and the high throughput. A well-known in vitro model for the human brain endothelium, hCMEC/D3 cell line (Figure 4) showing the formation of tight junctions and the expression of most transporters and receptors of the in vivo BBB [46], cultured in a transwell system, will also be used. Furthermore, extrapolation of in silico, in vitro, and zebrafish data to higher vertebrates seems feasible [47].

Zebrafish larvae are frequently used in developmental biology or toxicological studies. However, in this study, we will use zebrafish larvae exclusively to study BBB integrity and functionality [48]. Fluorescently labeled reference compounds (Figure 4) will be intravenously injected into the Duct of Cuvier, as markers of paracellular permeability (eg, fluorescein isothiocyanate dextran 70 or fluorescently labeled liposomes), substrates of drug export transporters (eg, rhodamine-123 as P-glycoprotein substrate), or nutrient transporters (eg, fluorescently labeled transferrin as a marker for receptor-mediated transcytosis). PGME will be the first reference compound to be tested because of its high water miscibility. To precisely assess exposure, analytical methods (gas chromatography-tandem mass spectrometry [GC-MS/MS]) will be used to determine the concentrations of solvents and their metabolites in zebrafish medium, in the headspace of closed incubation vessels and tissue samples (ie, zebrafish homogenates). Circulation, tissue distribution, and brain uptake of the reference compounds will be monitored by confocal laser scanning microscopy (live imaging of anesthetized fish embryos for up to 24 hours). The concentration-dependent toxicity of solvents or their metabolites will be monitored based on the viability and malformations of embryos. The integrity of vasculature will be visualized in transgenic zebrafish kdrl: enhanced green fluorescent protein embryos. The metabolic capacity of the zebrafish will be determined by quantifying potential metabolites (determined in WP3; Figure 2) in zebrafish tissue homogenates. The concentration-dependent toxicity to BBB and the coefficient of permeation of solvents will additionally be evaluated in the hCMEC/D3 cell line cultured in a transwell system.

Finally, quantitative estimates of passive cellular uptake and BBB permeability of solvents and their metabolites will be provided based on computational modeling using physicochemical molecular descriptors according to the methods we previously established [26,49]. These methods provide very high throughput, allowing the screening of web-based chemical libraries.

Because the liver is the main organ responsible for metabolism and a large contributor to compound clearance, we will implement a system suitable for predicting the hepatic metabolism of solvents. In recent years, 3D liver cell models have been proposed as an alternative to less physiological 2D cell monolayers, and their applications have progressed substantially [50]. They are widely used for the assessment of hepatotoxicity [51-55]. An advantage of spheroids is that they overcome the limitation of rapid decline of drug-metabolizing enzyme activities in primary human hepatocyte suspension culture and cell lysates [56], such as microsomes and liver S9 fractions. In this study, we will use 3D liver cultures of the well-characterized human HepaRG cell line (Figure 5), which represent a promising model to evaluate hepatotoxicity and hepatic metabolism [57].

Determining the appropriate experimental test system (eg, cell plate, incubation time, and exposure concentration) will be an essential part of the development of the 3D model. Moreover, analytical methods (GC-MS/MS and liquid chromatography-tandem mass spectrometry) to detect and quantify the solvents and the metabolites formed must be developed to calculate hepatic metabolism and clearance. Then, proof of metabolic competence and maintenance of the 3D HepaRG cells will be carried out by assessing the metabolism of known P450 substrates. The presence and secretion of albumin as a specific hepatocyte marker will be assessed using immunostaining and enzyme-linked immunosorbent assay. Solvent- and metabolite-induced cytotoxicity will be assessed after 48 hours and 7 days (repeated) of exposure to determine the nontoxic concentration range for subsequent experiments. The metabolic abilities of the 3D HepaRG model and 3D primary human hepatocytes will be compared. Furthermore, the clearance data obtained from the 3D HepaRG model will be compared with the short-term clearance measured in the human liver cell lysate (S9 fractions). In addition, Michaelis-Menten-Kinetic parameters (V_max and Km) for the formation of the metabolites will be derived using the S9 fractions. These data will be used to build a physiologically based toxicokinetic (PBTK) model (Figure 2).

Human biomonitoring refers to monitoring exposure-related health risks by analyzing biological samples, usually blood and urine samples [58]. The biomonitoring limit values (BMLVs) are set to protect human populations against the potential toxic effects of chemical substances. These limit values account for all routes through which a chemical can enter the body. These are most often the inhalation and skin routes in occupational and environmental settings. Kinetic studies that provide absorption, biotransformation, and elimination rates as well as the absorption and elimination half-lives of the parent compound and its metabolites are necessary to set BMLVs. The apparent urinary elimination half-lives of the parent compound and its metabolites will later be used to develop a biomonitoring method. Sample collection time is crucial and is determined by the apparent elimination half-life of the chemical. Blood concentrations will be used to calibrate the air:blood PC for the toxicokinetic models.

We will recruit 4 participants for 2 of the selected solvents. All participants must meet the following criteria: they should be healthy individuals who do not smoke or use contraceptive hormones, do not consume alcohol, be aged between 18 and 65 years, have normal red blood cells and hemoglobin concentrations, maintain a BMI between 18 and 25, and should not be working with glycol ethers. Pregnant and breastfeeding women will be excluded from this study. Participants will be recruited using flyers and announcements distributed at the teaching hospital, university websites, and bulletin boards. All participants will sign a written informed consent form before being included in the study.

The participants will be exposed to a single glycol ether for 4 hours under controlled conditions in an exposure chamber (12 m³). PGE concentrations will be set at or below the Swiss occupational exposure level (OEL) if one exists. In the absence of an OEL, we will rely on existing OELs for other propylene glycols. The parent compound (free and conjugated) and the oxidative metabolites (free and conjugated) of the selected glycol ethers will be monitored in blood, urine, and exhaled air samples. These are noninvasive methods used for human participants, and the results will be used in WP5 to estimate brain concentrations. All compounds will be quantified using capillary gas (parent compound in blood, urine, and exhaled air) or liquid (metabolites in blood and urine) chromatograms with tandem mass spectroscopy detection.

PBTK models can be used to estimate human brain concentrations. The risk of neurotoxic effects can be estimated by comparing the predicted solvent-brain concentrations with the NOAEC obtained from the in vitro models. Mathematical models such as PBTK models can be used to predict the ADME of a chemical and its metabolites. In these PBTK models, the body is represented by 1 or more compartments. Each compartment represents 1 or more tissues that are kinetically homogeneous, that is, that have similar perfusion rates and an assumed similar substance solubility. PBTK models are described by a set of parameters that define the compartments and a set of mass balance differential equations for each compartment.

A previously developed toxicokinetic model for PGME with metabolism that is assumed to follow Michaelis-Menten kinetics calibrated for different age groups serves as the basis for our development [30,31,59]. We aim to modify this previously developed toxicokinetic model to include a separate compartment for the brain using BBB flux rates obtained from in vivo, in vitro, and in silico models (WP2). In addition, we will implement PC obtained from empirical human experiments (WP4) and metabolic parameters assessed in a hepatocyte assay (WP3). The toxicokinetic models will be able to simulate not only acute but also chronic exposures; therefore, both short-term and long-term exposures can be explored in silico. We will develop the toxicokinetic model into a physiologically based pharmacokinetic model based on the existing inhalation-only toxicokinetic model originally developed for PGME [40] and build it in the Berkeley Madonna software or equivalent. We will model the brain as a single compartment with direct contact with the blood flow and where organic solvent uptake will be assumed to be diffusion limited, which is in line with other physiologically based pharmacokinetic models [60]. Values for physiological parameters (volume of vascular brain, as fraction of brain volume [FVvb], volume of extravascular brain, as fraction of brain volume [FVevb], volume fraction of brain tissue [FVB; as percent of body weight], BBB surface [Sh] in cm², fraction of cardiac output in brain at rest [BFbrainrest]/cardiac output in brain at light work [BFbrain]) required to build the TK model are from the scientific literature. Depending on the substance, values of chemical-specific parameters such as the pulmonary retention (Rpulm), central:air PC (Pca), blood:air PC (Pba), and brain tissue:vascular brain PC (Pevb_vb) are either taken from the literature or estimated in silico. Since the partitioning of organic compounds between human tissue homogenate and blood is a function of water and lipid content of tissues and the n-octanol:water PC (Kow), PCs are estimated in silico based on LogKow. Kinetic coefficients needed for each organic solvent included in this study will be found in WP3 for liver metabolism (Michaelis-Menten parameters [V_max and Km]), WP2 for BBB uptake (BBB permeability-surface area product [PS]). The fraction unbound in blood (Fu_blood) will be estimated based on the fraction unbound in plasma (Fu_plasma) and the blood-to-plasma ratio (Rb), and the fraction unbound in brain (Fu_brain) will be considered when modeling each solvent as only the free fraction is able to distribute to different tissues and is biologically active. The model will be calibrated by comparing the predicted and actual urinary organic solvent concentrations obtained from the controlled human experiments (WP4). Both the free and total organic solvent concentrations (free+conjugated) will be obtained for calibration.