Identification of Neural-Relevant ToxCast High-Throughput Assay Intended Gene Targets: Applicability to Neurotoxicity and Neurotoxicant Putative Molecular Initiating Events

Abstract

The US EPA’s Toxicity Forecaster (ToxCast) is a suite of high-throughput in vitro assays to screen environmental toxicants and predict potential toxicity of uncharacterized chemicals. This work examines the relevance of ToxCast assay intended gene targets to putative molecular initiating events (MIEs) of neurotoxicants. This effort is needed as there is growing interest in the regulatory and scientific communities about developing new approach methodologies (NAMs) to screen large numbers of chemicals for neurotoxicity and developmental neurotoxicity. Assay gene function (GeneCards, NCBI-PUBMED) was used to categorize gene target neural relevance (1 = neural, 2 = neural development, 3 = general cellular process, 3A = cellular process critical during neural development, 4 = unlikely significance). Of 481 unique gene targets, 80 = category 1 (16.6%); 16 = category 2 (3.3%); 303 = category 3 (63.0%); 97 = category 3A (20.2%); 82 = category 4 (17.0%). A representative list of neurotoxicants (548) was researched (ex. PUBMED, PubChem) for neurotoxicity associated MIEs/Key Events (KEs). MIEs were identified for 375 compounds, whereas only KEs for 173. ToxCast gene targets associated with MIEs were primarily neurotransmitter (ex. dopaminergic, GABA) and ion channel (calcium, sodium, potassium) receptors. Conversely, numerous MIEs associated with neurotoxicity were absent. Oxidative stress (OS) mechanisms were 79.1% of KEs. In summary, 40% of ToxCast assay gene targets are relevant to neurotoxicity mechanisms. Additional receptor and ion channel subtypes and increased OS pathway coverage are identified for potential future assay inclusion to provide more complete coverage of neural and developmental neural targets in assessing neurotoxicity.

1. Introduction

The U.S. Environmental Protection Agency’s (EPA) Toxicity Forecaster (ToxCast^™ https://www.epa.gov/chemical-research/toxicity-forecasting) is a component of the EPA’s computational toxicology research program. This program was developed to address the demands of evaluating the safety of an expanding number of chemicals in the environment against the resource and time requirements for conventional defined end point toxicity testing (Dix et al., 2007). Initial chemical and assay selection focused on broad representation of structural classes and phenotypic outcomes, and the technical and economic feasibility of specific assay testing across a vast library of chemicals (Dix et al., 2007). The existing ToxCast chemical library contains over 4,500 compounds including pharmaceutical drugs, pesticides, food additives, and industrial solvents and by-products. Currently, ToxCast is a collection of over 700 high-throughput in vitro assays used to screen rapidly for potential toxic interactions with biological targets. The ToxCast assay suite includes primarily commercially available tests that heavily focus on liver enzymes, metabolic substrates, nuclear receptors, ion channels, adhesion molecules, and genotoxic and cytotoxic processes (Collins et al., 2008; Judson et al., 2010). Gene targeted assay end points include DNA binding, cell morphology, cell adhesion, chemotaxis, protease inhibition, inflammation, cytokine signaling, and transcription factor activation. Recently, functional assays for cytotoxicity, apoptosis, cellular respiration, gene expression, and neural network development and electrical activity have been added. Data generated using the ToxCast assay suite is publicly available and has been used to characterize and prioritize chemicals for in vivo testing, and identify endocrine disrupting (Reif et al., 2010; Filer et al., 2014), hepatotoxic (Liu et al., 2015), and genotoxic compounds (Chiu et al., 2018). Additionally, ToxCast data has been used to construct predictive quantitative in vitro–in vivo extrapolation (QIVIVE) models (Wetmore 2015, Sipes et al., 2017) and adverse outcome pathways (AOPs) (Bell et al., 2016, Oki and Edwards, 2016, Pittman et al., 2018).

The nervous system is highly complex and subject to perturbation and dysfunction from numerous environmental compounds. Because many environmental compounds are neuroactive, neurotoxicity and developmental neurotoxicity are important regulatory endpoints for the EPA. In addition, the public is also concerned about the potential for chemicals to disrupt nervous system function and/or development. However, neurotoxicity and developmental neurotoxicity data are lacking for thousands of environmental compounds (Judson et al., 2009; Strickland et al., 2018; Kosnik et al., 2020), which precludes understanding the potential hazards of exposure to these compounds. To begin to generate data specifically related to in vitro neuronal function, Strickland and coworkers (2018) screened the ToxCast Phase I and II library chemicals for alterations in neural function and viability in neural networks grown on microelectrode arrays. The chemicals consisted of 1056 individual compounds, which included 326 use categories (e.g., fungicides, pesticides, pharmaceuticals), and targeted 121 specific biological substrates associated with molecular initiating events (MIEs) (e.g., acetylcholinesterase inhibitor) and/or key events (KEs) (e.g., mitochondrial disruptor). These authors, and others, demonstrated that the chemical space which potentially impacts the nervous system, is vast and encompasses an unknown number of neurotoxic mechanisms (Richard et al., 2016; Iqubal et al., 2020; Ravichandran et al., 2021). Additionally, other researchers have reported deficiencies in the ability of the ToxCast assays to detect neurotoxicants (Shah and Greene, 2014; Silva et al., 2015), or that these methods have lower sensitivity for certain chemicals such as pharmaceuticals (Sirenko et al., 2019). Chemicals requiring metabolic activation, or repeated exposures to produce toxicity, may not be detected with existing in vitro screening batteries (Aylward and Hays, 2011; Knudsen et al., 2015; Silva et al., 2015). Together, these studies indicate that the current suite of assays in ToxCast may not cover the needed biological space to identify neurotoxicity hazard with sufficient breadth and/or granularity.

Recently studies that used ToxCast and Tox21 assays and data to assess environmental chemicals for toxicity prediction and chemical prioritization were reviewed (Jeong et al., 2022). The target endpoint-mechanism focused analyses primarily included endocrine disruption, carcinogenicity, hepatotoxicity, immunotoxicity, developmental biology, cardiotoxicity, and cytotoxicity, but not neurotoxicity. To date, there has not been a comprehensive evaluation of how well the ToxCast assay intended gene targets align with potential xenobiotic targets that alter neurological function. The first goal of the current work was to examine and categorize the non-redundant biological intended gene targets included in ToxCast assays (US EPA Center for Computational Toxicology and Exposure, 2022, Invitro db version 3.5. DOI:https://clowder.edap-cluster.com/spaces/62bb560ee4b07abf29f88fef) for relevance to neural function. The second goal of our work was to use online databases and the published literature to compile a list of putative MIEs or KEs for a diverse chemical space of documented neurotoxicants. For the third goal of this work, areas of overlap between the ToxCast high-throughput assay suite intended gene targets and mechanisms of action were noted, and gaps between the two were identified. The results of this effort may help in the development of new approach methodologies (NAMs) for sensitive genomic targets (based on putative neurotoxicity AOPs) to detect and prioritize neurotoxic environmental chemicals.

2. Methods

2.1. ToxCast Assay Target Gene Classification

Because there are multiple ToxCast assays that interrogate the same biological endpoint, the individual assay endpoints (Supplemental Table 1. ToxCast In Vitro Assays) have been previously reduced to 481 non-redundant assay intended gene targets by the ToxCast program (Supplemental Table 2. Non-Redundant Gene Target List). In this manuscript, gene symbols are formatted as they are represented in the ToxCast database. Detailed information about the individual assays can be found at the ToxCast data download page https://clowder.edap-cluster.com/files/6215520fe4b039b22c7a7836. Intended “gene target” is used as a proxy designation for ToxCast assays with the same molecular target although the assay readouts may be different. As described by Judson et al. (2010) “Assays probed … genes either through direct interactions with the relevant protein or using a variety of indirect, downstream readouts of mRNA or protein levels.” In the work presented here, we examined the function of these non-redundant assay gene targets using the online databases National Center for Biotechnology Information (NCBI, https://www.ncbi.nlm.nih.gov/), and GeneCards: The Human Gene Database (https://www.genecards.org/). An internal subject-matter expert panel of seven EPA neurotoxicologists (each with 25 years of experience) created a relevance category scheme based on biomedical and genomic information in NCBI and GeneCards (see Supplemental Table 2) and decided if the ToxCast intended gene targets were related to neural function. This relevance category was an indication of a gene target’s importance in the nervous system as follows:

1 = Direct target relevant to neural function

2 = Target relevant to neurodevelopmental processes (i.e., may not be relevant in adult organism. ex: target has a role in neurodevelopment that if disrupted could result in altered neural differentiation or neonatal synaptogenesis)

3 = General cellular process found in all cells (not unique to neurons)

3A = General cellular process that may also have a specific function in neurodevelopment

4 = Not directly related to neural function

2.2. Neurotoxic Agents and MIE Identification:

An aggregated list of 548 documented or suspected neurotoxicants (Supplemental Table 3. Neurotoxic Agents_MIEs) was curated from two sources: Experimental and Clinical Neurotoxicology, Schaumburg, H.H. and Spencer, P.S. eds., (2000), and Casarett and Doull’s Toxicology: The Basic Science of Poisons, Unit V, Toxic Agents, Chapters 22– 24, “Toxic Effects of Pesticides”; “Toxic Effects of Metals” and “Toxic Effects of Solvents and Vapors” (2008). This list was composed of drugs, microbial toxins, pesticides, metals, inorganic compounds, organic solvents, food additives, venoms, biogenic substances, fuels, and industrial substances. Based on the cumulative weight of evidence a putative mechanism of action for these chemicals was ascertained by the expert panel from a manual literature curation of databases (PubChem, PubMed, CompTox Chemicals Dashboard, Google Scholar) employing chemical identifiers: IUPAC, Medical Subject Headings (MeSH), and brand names. The search included references for in vitro, in vivo, and clinical studies. The results were filtered with the search terms neurotoxicity, neurotoxicant, developmental neurotoxicity, toxicity, neural, mechanism, and mechanism of action. We attempted to determine a potential MIE for each entry using the adverse outcome pathway (AOP) framework. The MIE is defined as the initial chemical biomolecular injury that triggers a cascade of cellular dysfunctions (key events (KEs)) that results in an apical adverse outcome (AO) (Vinken, 2013; Villeneuve et al., 2014; Vinken and Blaauboer, 2017). However, some chemical searches yielded indeterminant initiating molecular targets (conflicting results or appropriate data not available). These compounds were assigned KEs taken from 2 or more published studies. We acknowledge that many chemicals have multiple in vivo mechanistic targets and identifying a single MIE/KE is challenging. Consequently, MIE and KE designations required unanimous agreement by the expert panel. Additionally, due to the vast number of possible assays and protocols that have been used to investigate neurotoxicants, identification of the most sensitive endpoint is virtually impossible. Thus, we attempted to identify consensus MIEs/KEs from review articles (when available). Then, each chemical’s MIEs or KEs were manually cross-referenced with the intended assay gene targets from Supplemental Table 2, and the appropriate ranking category was added to the MIE/KE table in Supplemental Table 3 (Neurotoxic Agents_MIEs).

MIE/KE Identification Workflow Summary:

1. Neurotoxicant Selection

2. Literature Curation

3. Search Term Filtering

4. Collection of Mechanistic Information

5. A. Identify MIE: Map to ToxCast intended gene target or,

   B. No MIE: Identify KE(s)

2.3. Areas for Enhanced/Expanded Neurotoxicological Coverage:

The Neurotoxic Agent List (Supplemental Table 3) was searched for common mechanisms of action (MIE and KE). Then, a comparison of these neurotoxic MIEs/KEs was made using Supplemental Table 3 (Neurotoxic Agents_MIEs) and Supplemental Table 2 (Non-Redundant Gene Target List), to identify areas where additional assay coverage may be beneficial for neurotoxicity screening.

3. Results

3.1. ToxCast Assay Gene Target Classification-Neurotoxicity Relevance

The 2,505 ToxCast assay suite endpoints (606 are not gene related) were previously reduced to 481 unique intended gene targets (see Supplemental Table 2). After extensive manual review by the expert panel, eighty targets (16.6%) were assigned a relevance category of 1 (Table 1. Category 1 Gene Targets), indicating that the intended gene targets coded information with direct relationship to neural function. Many of these diverse gene targets involve various neurotransmitter receptors or ion channels and are known to be targets of neurotoxic chemicals. Specific examples of category 1 targets include the following: Early Growth Response 1 (EGR1) is an immediate early gene and transcription factor involved in neurotransmission and neuronal plasticity and is integral to learning and memory function (Veyrac et al., 2014; Duclot and Kabbaj, 2017). Glutamate signaling through glutamate receptors (Gria1, Grik1, Grin1, Grm1, Grm5) is the primary excitatory neurotransmitter in the human nervous system and is required for proper neural function (Willard and Koochekpour, 2013). Conversely, glutamate excitotoxicity is associated with neurodegenerative (Lau and Tymianski, 2010) and neuropsychiatric disease (Myers et al., 2019). Nitric Oxide Synthase 1 (Nos1) is constitutively expressed in the brain and produces nitric oxide via L-arginine. It mediates neural functions that include central control of blood pressure, long-term potentiation and inhibition, and neurogenesis (Arami et al., 2017; Solanki et al., 2022). Tubulin (TUBA1A) represents a structural neural gene target of the ToxCast assay suite. The alpha-tubulin is a primary component of neural microtubules that provide the cytoskeletal structure necessary for neuron development, maturation, and maintenance (Breuss et al., 2017; Buscaglia et al., 2020). In addition, microtubules facilitate dendritic spine changes that underlie synaptic plasticity (Aiken et al., 2017). See Table 1 for the complete list of category 1 genes.

Table 1.

Category 1 Gene Targets

Gene Family	Gene Symbol1	Number of Neurotoxicants2
Cholinergic Receptor	CHRM1, CHRM2, Chrm3, CHRM4, CHRM5, CHRNA2, Chrna7	149
Glutamate Receptor	Gria1, Grik1, Grin1, Grm1, Grm5	93
GABA Receptor	Gabbr1, GABRA1, GABRA5, Gabra6	62
Sodium Channel	Scn1a	60
Calcium Channel	Cacna1a, Cacna1b	59
Serotonin Receptor	Htr1a, HTR2A, HTR2C, HTR3A, Htr4, HTR5A, HTR6, HTR7	57
Dopamine Receptor	DRD1, DRD2, DRD4, DRD5	56
Acetylcholinesterase	ACHE	52
Potassium Channel	KCNH2, Kcnj1, KCNK1, Kcnn1	32
Adrenergic Receptor	Adra1a, Adra1b, Adra2a, Adra2b, ADRA2C, ADRB1, ADRB2, ADRB3	16
Nitric Oxide Synthase 1	Nos1	7
Glycine Receptor	Glra1	6
Opioid Receptor	OPRD1, Oprk1, OPRL1, OPRM1	6
Histamine Receptor	HRH1, Hrh2, Hrh3	5
Tubulin	TUBA1A 1	5
Monoamine Oxidase A, B	Maoa, Maob	4
Solute Carrier Family	Slc18a2, SLC6A2, Slc6a3, SLC6A4	2
Adenosine Receptor	ADORA1, ADORA2A, ADORA2B	1
Neuropeptide Y	NPY, NPY1R, NPY2R	1
Arginine Vasopresin Receptor	AVPR1A	0
Beta-Site APP-Cleaving Enzyme 1	BACE1	0
Cholecystokinin Receptor	Cclar, Cckbr	0
Chemokine (C-C motif) Ligand	CCL2	0
Catechol-O-methyltransferase	Comt	0
Early Growth Response 1	EGR1	0
Neurotrophic Tyrosine Kinase Receptor	NTRK1	0
Neurotensin Receptor	NTSR1	0
Sigma Non-Opioid Intracellular Receptor	Sigmar1	0
Tachykinin Receptor	Tacr1, TACR2, Tacr3	0

¹Intended gene symbols in BOLD are associated with a neurotoxicant MIE/KE listed in Supplemental Table 3. Plain text gene symbols did not match to a specific neurotoxicant.

²Number of neurotoxicants indicates the chemicals from Supplemental Table 3 that have been associated with the noted gene family and gene symbols. (See section 3.2 Neurotoxicant MIE/KE Overview below.)

Categorization of the intended gene targets indicated that 16 gene targets (3.3%) were relevant to neurodevelopmental processes (Category 2; Table 2. Category 2 Gene Targets). Whereas a relatively small number of genes were placed in this category, they are essential to developmental processes. Alpha-Fetoprotein (AFP) is a fetal yolk sac plasma protein that is used as an indicator of neural tube defects and other neurological anomalies (Schieving et al., 2014). Activated v-akt Murine Thymoma Viral Oncogene Homolog (AKT1) is a critical component of growth factor signaling that regulates neural development, survival, and differentiation (Cheng et al., 2013). The Iodothyronine Deiodinases (DIO1, DIO2, DIO3) control the availability of thyroxine (T4) through the conversion of T4 to 3,3’−5-triiodothyronine (T3), as well as the inactivation of both hormones, and thus modulate thyroid signaling (THRA, thra.L, thraa) (Köhrle and Frädrich, 2022). It is well documented that thyroid hormone is involved in neural cell migration, differentiation, synaptogenesis, myelination and signaling (Bernal, 2005; Horn and Heuer, 2010; Stepien and Huttner,2019). In addition, thyroid hormone alters the expression of the sonic hedgehog (SHH) protein that is critical for brain growth and morphogenesis (Desouza et al., 2011; Memi et al., 2018). The ephrin receptors (EPHA1, EPHA2, EPHB1) are tyrosine kinase receptors that regulate axon guidance and are involved in neurogenesis and differentiation during nervous system development (Huot, 2004; Xu and Henkemeyer, 2012; Cramer and Miko, 2016). Dual-Specificity Tyrosine-(Y)-Phosphorylation Regulated Kinase (DYRK1A) controls neural differentiation in developing brain (Arbones et al., 2019). Nuclear Receptor Subfamily 6 Group A Member 1 (NR6A1) is an intracellular transcription factor that has been linked to differentiation and development of the forebrain and depressive behavior in mice (Tan et al., 2022). Paired Box Gene 6 (PAX6) is a transcription factor involved in regulating neural patterning, differentiation, proliferation and oculogenesis (Osumi et al., 2008; Georgala et al., 2011). Lastly, Protein Phosphatase 3 Catalytic Subunit Alpha (PPP3CA) protein is an alpha isoform of a subunit of calcineurin which regulates turnover of synaptic vesicles and has been implicated in neurodevelopmental disorders (Myers et al., 2017).

Table 2.

Category 2 Gene Targets

Gene Family	Gene Symbol1	Number of Neurotoxicants2
Alpha-Fetoprotein	AFP	0
v-akt Murine Thymoma Viral Oncogene Homolog	AKT1	0
Iodothyronine Deiodinase	DIO1, DIO2, DIO3	0
Dual-Specificity Tyrosine-(Y)-Phosphorylation Regulated Kinase	DYRK1A	0
Ephrin Receptor A1, A2	EPHA1, EPHA2	0
Ephrin Receptor B1	EPHB1	0
Nuclear Receptor Subfamily 6 Group A Member 1	NR6A1	0
Paired Box Gene 6	PAX6	0
Protein Phosphatase 3 Catalytic Subunit Alpha	PPP3CA	1
Sonic Hedgehog	SHH	0
Thyroid Hormone Receptor Alpha, Alpha L Homeolog, Alpha a	THRA, thra.L, thraa	3

¹Intended gene symbols in BOLD are associated with a neurotoxicant MIE/KE listed in Supplemental Table 3. Plain text gene symbols did not match to a specific neurotoxicant.

²Number of neurotoxicants indicates the chemicals from Supplemental Table 3 that have been associated with the noted gene family and gene symbols. (See section 3.2 Neurotoxicant MIE/KE Overview below.)

The largest subset of genes (303; 63.0%) was assigned to relevance category 3, which represented general cellular processes important to all cells. Due to the large number of intended gene targets, the complete list is found in Supplemental Table 4. Category 3 and Category 3A Genes. The genes assigned the category of 3 were arguably the most diverse and covered the largest biological space. The 10 most represented assay gene targets included nuclear receptors, phosphodiesterases, protein tyrosine phosphatases, estrogen receptors, interleukins, caspases, matrix metallopeptidases, chemokines, mitogen-activated protein kinases and ATP-binding cassette (ABC) transporters (see Supplemental Table 5. Category 3/3A Gene Targets for the number of genes for each family/per ranking category). In total, the general cellular process genes were associated with 235 of the 548 neurotoxicants we examined (42.9%).

The review of the assay target genes revealed a circumstance that required an additional designation. Within the category of genes involved in general cellular function (category 3), there was a group that was deemed critical for correct neural development. These 97 gene targets represented 20.1% of all the genes (32% of category 3) and were placed in category 3A (Supplemental Table 4. Category 3_Category 3A Genes). As an example, the retinoic acid receptors (RARs) are nuclear receptors that initiate or repress transcription of proteins involved in embryonic organ development. Retinoic acid and related retinoid pathways control differentiation, proliferation, apoptosis, and cell cycle exit in varied cells including kidneys, retina, and heart (Rhinn and Dolle, 2012; Gutierrez-Mazariegos et al., 2014). However, retinoic acid signaling specifically regulates neurogenin 2, which controls neural determination (Ma et al., 1996; Lee et al., 2009; Das et al., 2014), and therefore RXRA, RXRB and RXRG were placed in category 3A. In summation (category 1 +2 +3A), the neural-relevant genes represented approximately 40% of all ToxCast gene targets.

Eighty-two intended gene targets (17.0 %) were placed in category 4 (Supplemental Table 2. Non-Redundant Gene Target List), which were characterized as not having directly relevant biological significance for neural function. The cytochrome P450 family which primarily controls drug metabolism, steroid and cholesterol synthesis was the largest component of the rank 4 genes (35.4 %). Immune response and cell-cycle processes, as well as metabolic and electrolyte homeostasis genes were also included in this category. This does not mean that these processes are not involved in certain neurotoxic reactions from chemicals. Rather, the category means that an active chemical paired to an in vitro assay for these genes would not lead to an immediate concern for neural function or correct neural development.

3.2. Neurotoxicant MIE/KE Overview (Identification)

The list of chemicals we examined that have been associated with neurologic signs or symptoms has 548 unique chemical entries. The neurotoxicants were sub-divided into six broad categories based on use: chemical-industrial, drug-medication, metal, pesticide, solvent-volatile or toxin (Supplemental Table 3). The associated MIEs varied by chemical category, although many were common to all chemical groups. In addition, mechanisms of action could only be narrowed to KEs for some chemicals in each category. Specific MIEs for 375 compounds were identified and 173 compounds could only be associated with KE pathways. As shown in Table 1, neurotoxicant MIE/KEs were associated with neurotransmitter receptors, ion channels (Cacna1a, Cacna1b KCNH2, Kcnj1, KCNK1, Kcnn1, Scn1a), structural components (tubulin), second messenger synthesis (nitric oxide), solute carriers, and enzymes essential to neurotransmitter degradation. Additionally, MIE/KEs were also associated with the tyrosine kinase and thyroid hormone homeostasis (Table 2).

For some neurotoxicants, the purported mechanism of action could not be related to a particular gene in the current ToxCast library. These chemicals are listed in Supplemental Table 6 (Neurotoxicant MIEs without Corresponding Gene Targets). Some examples include cannabis, acting via cannabinoid receptors (Grotenhermen, 2003, 2004), capsaicin acting via TRPV1 Receptors (Davies, et al., 2010; Nagy et al., 2014), or lobeline acting via the vesicular monoamine transporter (Dwoskin and Crooks, 2002). Other known neurotoxicants such as n-hexane (via the metabolite 2,5-hexandione) can produce protein adducts, resulting in neurotoxicity. Such mechanisms of toxicity are not currently linked to specific intended gene targets in ToxCast.

However, the largest class of KEs for the neurotoxicants that were examined involved oxidative stress (OS) mechanisms (137 of 173; 79.1 %). OS is a multifactorial process that involves a complex network of pathways, and there are not specific OS genes that code for OS per se. However, numerous genes are associated with the molecular events that lead to the imbalance of oxidation and antioxidation processes. Several intended target genes (32) are not in a causal pathway but may be involved in a cascade of activation or deactivation that results in or from OS (Supplemental Table 7. Oxidative Stress Pathway Genes). The remaining KEs predominantly targeted DNA and RNA pathways: DNA adduction, intercalation, methylation, DNA/RNA alkylation and protein synthesis inhibition. In addition, for a small subset of compounds (captan, folpet, captafol, acetone, picaridin, DEET) neurotoxicity was equivocal based upon published clinical or experimental data and the US EPA Comp Tox Chemicals Dashboard (https://comptox.epa.gov/dashboard).

3.3. Relationship Between ToxCast Assay Intended Target Genes and MIEs

The overlap between neurotoxicant identified MIEs and associated intended gene targets varied greatly by our relevance classification. As anticipated, the highest concordance was attributed to the category 1 direct neural relevance genes, matching to 375 of 548 (68.4%) neurotoxicants. The category 2 genes which are related to neural development were only matched at 0.55% (3 of 548) of compounds. However, the general cellular process genes (63% of assay endpoint targets), corresponded with 235 of 548 (42.9%) neurotoxicants. Finally, the category 4 genes matched to neurotoxicant MIEs by less than 1 percent (0.18%), thus confirming these genes are not uniquely important for neural function.

The distribution of the category 1 gene targets (Table1) associated with neurotoxicants (in descending order) is as follows: cholinergic receptors (muscarinic = 68, nicotinic = 81); glutamate receptors (ionotropic = 32; metabotropic = 61); gamma-aminobutyric acid, GABA, receptors A and B (62); sodium channels (60) and calcium channels (59); serotonin receptors (57); dopamine receptors (56); acetylcholinesterase (52); potassium channels (32); adrenergic receptors (16); nitric oxide synthase 1 (7); glycine and opioid receptors and solute carrier family (6 each); histamine and tubulin (5 each); monoamine oxidase A and B (4) and neuropeptide Y (1). The adenosine receptors (ADORA1, ADORA2A) and the early growth response 1 (EGR1) gene were directly associated with 1 or no neurotoxicants, respectively. However, based on the published literature, they potentially could be associated with chemicals that may be involved in oxidative stress pathways (Ramkumar et al., 2001; Sousa et al., 2008; Guo et al., 2023). In addition, 13 gene targets that were categorized as being relevant to neural function were found not to be potential/putative targets of any chemical on the list used for this review. These results show that including neurotransmitter receptors, ion channels, and enzymes important for second messenger systems and neurotransmitter synthesis/degredation is important when developing screening batteries for neurotoxicity.

For the genes targets ranked in category 2 (Table 2), the ephrin receptors (A1 and A2) and the thyroid hormone receptors alpha, alpha L homeolog, and alpha A were associated with targets for 153 and 154 neurotoxicants, respectively. As indicated with the category 1 genes targets, the linkages to neurotoxicant MIE/KEs was due to involvement with OS. An exception to the OS linkages was the pesticide kepone (chlordecone) which impacts thyroid and estrogenic pathways, as well as multiple ATPases (Costa et al., 2008; Multigner et al., 2016). Additionally, a third gene family in category 2 was protein phosphatase 3, which was identified as a possible MIE target for the drug cantharidin (Li and Casida, 1992; Ren et al., 2021).

3.31. MIEs Covered in ToxCast Gene Targets

Many neurotoxicologically relevant MIEs were covered in the intended gene targets, and often involved neuronal receptors. There are assays specific for adrenergic (Adra1a, Adra1b, Adra2a, Adra2b, Adra2c, Adrb1, Adrab2, Adrab3), cholinergic (AChE, BCHE, CHRM1, CHRM2, CHRM3, CHRM4, CHRM5, Chrna2, Chrna7), dopaminergic (DRD1, DRD2, DRD4, DRD5), GABA (Gabbr1, GABRA1, GABRA5, GABRA6), glycine (GLI1, GLI1, GLI3, Glra1), glutamatergic (Gria1, Grik1, Grin1, Grm1, Grm5), histamine (HRH1, Hrh2, Hrh3), opioid (OPRD1 Oprk1 OPRL1, OPRM1) and serotonergic (Htr1a, Htr2a, Htr2c, Htr3a, Htr4, Htr5a, Htr6, Htr7) receptors. Ion channel assays are also present, but less well represented in the ToxCast assay suite. There are two calcium (Cacna1a, Cacna1b), one sodium (Scn1a) and three potassium (KCNH2, Kcnj1, Kcnn1) voltage-gated ion channels. The peptide related genes somatostatin receptor 1, oxytocin receptor and nuclear receptor subfamily 2 group F member 6 (Sstr1, Oxtr, NR2F6) are also included. However, from a neurobiological perspective, many additional receptors and ion channels exist in the nervous system as potential targets for neurotoxicants.

3.32. Neurotoxicological Coverage Gaps

It is important to recognize that the existing ToxCast battery of assays was designed to screen for multiple types of toxicity using in vitro assays. This approach is different than one that would focus on test batteries designed to screen for specific pathways or types of toxicity. Additionally, the assays need to be available in a high-throughput format. These criteria result in gaps for components of specific toxicity pathways, of which neurotoxicity is only one example. Examination of the mechanisms of action of known neurotoxicants can guide the development of screening batteries with more complete coverage for the different types of neurotoxicity.

Numerous genes and MIEs/KEs associated with the neurotoxicant compound list (Supplemental Table 3) are not included in the current ToxCast assay targets. Some examples include cannabinoid receptors (CB1, CB2, GPR55, anandamide (ANA/AEA), 2-arachidonoylglycerol (2-AG)) and trace amine-associated receptors (TAAR1, TAAR2, TAAR3, TAAR4, TAAR5, TAAR6, TAAR7, TAAR8, TAAR9). In addition, there are also ion channel targets that are covered incompletely (voltage-gated sodium, potassium, and calcium channels) or not included in the current in vitro test battery. For example, there are up to 10 different gene products associated with voltage-gated sodium channels, but only one (Scn1a) is covered by the current ToxCast assay suite, while transient receptor potential channels (TRPC, TRPV, TRPVL, TRPA, TRPM, TRPS, TRPN, TRPP and TRPML), toll-like receptors (TLRs), and ryanodine receptors (RYR2, RYR3) are not covered at all. Lastly, although oxidative stress associated genes are found in the ToxCast assay targets (see Supplemental Table 7), specific genes targeting key components of critical steps in REDOX/Energy homeostasis are missing. Whereas direct measurement of reactive species is almost impossible due to very brief half-lives, the by-products of damage to biomacromolecules and the status of antioxidants are routinely assessed using existing methods. In the current ToxCast assay suite of 2,505 endpoints, 4 assays are designated Oxidative Stress Target Family Subtype and are mapped to 2 genes, catalase (CAT) and the growth arrest and DNA-damage-inducible, alpha (GADD45A) (Supplemental Table 1). CAT is a first line defense antioxidant gene commonly used to indicate OS (Ighodaro and Akinloye, 2018), and GADD45A is a gene induced by DNA damage and stressful growth arrest (Moskalev et al., 2012, Salvador et al., 2013). An additional two assay endpoint intended gene targets, nuclear factor, erythroid 2-like 2 (NFE2L2) and nuclear respiratory factor 1 (NRF1), upregulate cellular antioxidant responses (Sant et al., 2017, Sies et al., 2017), but are classified as Family Subtypes nuclear respiratory factors and basic leucine zipper, respectively (Supplemental Table 1). However, there are many gene families including superoxide dismutase, Coenzyme Q, transferrin, peroxiredoxin, metallothionein, glutathione and glutathione peroxidase (SOD, COQ, TF, PRDX, MT, GSH, GPX) represented in the REDOX literature (Czerska et al., 2015; Pisoschi and Pop, 2015; Demirci-Çekiç et al., 2022) that are not included in the current ToxCast screening battery. These examples (receptors, second messengers, ion channels, oxidative stress) can serve as guidance for areas for assay development that could be added to future ToxCast assays.

4. Discussion and Conclusions

This work reviews the ToxCast assay suite intended gene targets and categorizes their relevance to neural function based on review by an expert panel of neurotoxicologists. Additionally, we examined a wide-ranging list of putative neurotoxic compounds, encompassing a diverse chemical space, and attempted to determine MIEs and KEs that were relevant to their neurotoxicity using a literature review. The intersection of the intended gene targets and MIEs/KEs informs the range of coverage of the current ToxCast assays for detecting known neurotoxic compounds.

An examination of ToxCast gene targets and neurotoxic agent MIEs/KEs demonstrated that the current suite is representative of a large percentage of relevant biological space for the chemical list we examined. However, the coverage is limited for some potential MIEs/KEs. A significant number of compounds are believed to be neurotoxic due to the induction of OS pathways and mitochondrial bioenergetics (which are interlinked). While several genes peripherally related to OS/mitochondrial bioenergetics are represented in the ToxCast assays (Supplemental Table 7), an incalculable number of MIEs and KEs may underlie these pathways (Sayre et al., 2008; Neal and Richardson, 2018). Consequently, development of additional high throughput assays or adding existing assays to detect OS (traditional assays such as NADH-Ubiquinone reductase, Gamma glutamylcysteine synthetase, superoxide dismutase, catalase, total antioxidant substances etc and/or genomic targets such as Cat, Dhcr24, Gpx1, Gss, Sod1, Nnt, Ucp2,… etc) and mitochondrial bioenergetics (traditional assays such as mitochondrial respiration, complex I to V enzymes and/or genomic targets such as Ndufa5, Nuufa6, Ndufb5, Sdhb, Sdhc, Sdhd, Cox5b, Cox6a1, Cox7a21, Atp5a1, Atp5b, Atp5g2, etc…) may enhance the detection sensitivity of future screening efforts for neurotoxic chemicals.

Previously Chushak et al. (2018) analyzed an earlier version (October 2015 release) of the Tox21/ToxCast database target genes. Using data from 1050 chemicals in the ToxCast library, they identified 216 (of 656) high-throughput assays which had 342 protein targets comprising 123 unique genes associated with neural function. The frequency of a protein target matching to active chemicals varied from a low of two to greater than 450 chemicals. The top 22 (>100 active chemicals) genes/proteins were all included in the current analysis (Table 3. Chushak and Current Categorization Neural Function Genes). It should be noted that although seven genes, CCL2, EGR1, SLC6A3, OPRM1, HTR7, SCN1A, DRD1, were designated as direct targets relevant to neurotoxicity by both studies, there were differences among the remaining 15 (see Table 3.).

Table 3.

Chushak and Current Categorization Neural Function Genes

Gene Symbol	Gene_Protein Name	Chushak Category1	Current Categorization
ESR1	Estrogen Receptor 1 (alpha)	1	3A
HLA-DRA	HLA class II Histocompatibility Antigen, DR alpha	1	3
CCL2	C-C motif Chemokine	1	1
JUN	Jun Proto-Oncogene	1	3
TGFB1	Transforming Growth Factor beta-1	1	3A
EGR1	Early Growth Response 1	1	1
SLC6A3	Sodium-Dependent Dopamine Transporter	1	1
RARA	Retinoic Acid Receptor alpha	1	3A
TSPO	Translocator Protein	1	3A
PAX6	Paired Box 6	1	2
RXRB	Retinoic Acid Receptor RXR-beta	1	3A
ICAM1	Intercellular Adhesion Molecule 1	1	3
PLAT	Tissue-Type Plasminogen Activator	1	3
IL6	Interleukin-6	1	3
AR	Androgen Receptor	1	3A
OPRM1	Mu-type opioid Receptor	1	1
HIF1A	Hypoxia Inducible Factor 1 alpha subunit	1	3
HTR7	3A-Hydroxytryptamine Receptor 7	1	1
EGFR	Epidermal Growth Factor Receptor	1	3A
SCN1A	Sodium Voltage-Gated Channel alpha subunit 1	1	1
PGR	Progesterone Receptor	1	3
DRD1	Dopamine Receptor D1	1	1

¹Chushak “1” = association with term(s) “neurological”, “synapse”, and “axon” in Gene Ontology search.

As some examples, the paired box 6 (PAX6) gene is essential for embryonic brain and eye development (Ochi et al., 2022) and was given the category of 2 in this study. The ESR1, TGFB1, RARA, RXRB, TSPO, AR and EGFR genes are widely distributed across tissue types but may also have specific neurodevelopmental roles (category 3A). Cumulatively, these genes are categorized as direct targets of neurotoxicants. In contrast, the remaining seven, HLA-DRA, JUN, ICAM1, PLAT, IL6, HIF1A and PGR, were identified as having “neurological function” by Chushak but were classified as general cellular processes found in all cells (category 3) in our assessment. This perceived discrepancy is likely the result of different analysis methods. The Chushak report performed a Gene Ontology database search using the terms neurological, synapse, or axon. Thus, a gene was included if it had an association with one of the three words. Here, gene function was ascertained from the online databases National Center for Biotechnology Information or GeneCards: The Human Gene Database (see 2.1 ToxCast Assay Target Gene Classification). Next, a relevance scale was applied to the results to characterize further a target gene’s potential to disrupt neural function. Consequently, it is not unexpected that all findings are not in concurrence. However, it should be noted that the results for 15 of 22 genes (68%) directly correspond.

Selected neurotransmitter receptor assays are represented in the ToxCast assay suite. Glutamate, the most abundant neurotransmitter in the human CNS, has three distinct ionotropic receptors, AMPA, NMDA and kainite, and three groups of metabotropic (mGluR) receptors. There are a combined 9 assays detecting Gria1, Grik1, Grin1, Grm1 and Grm5 receptor binding. Because there are 24 recognized glutamate receptor subtypes (Reiner and Levitz, 2018), the sensitivity to detect interactions with glutamate receptors is potentially not complete. In addition, the lack of receptor heterogeneity in the existing assays limits the discovery of mechanistic information, as receptor isoform combinations are integral to glutamatergic function (Du et al., 2020; McCullock and Kammermeier, 2021). As with glutamate receptors, other receptor types (GABAergic, nicotinic, etc) and ion channels (Na⁺, K⁺, Ca⁺⁺) are comprised of multiple different subunits and exhibit considerable functional and pharmacological heterogeneity (Olsen and Sieghart, 2009; Kuang et al., 2015; Zamponi et al., 2015; Zoli et al., 2015; Noreng et al., 2021; Sallard et al., 2021). Depending on the cross reactivity of a chemical with the multiple forms of various receptors, the existing coverage of neurotransmitter and ion channel receptor targets may reduce the neurotoxicity screening ability, either qualitatively (ability to detect a response) or quantitatively (detecting a response, but at a different potency), of the current set of assays.

Earlier studies have suggested limitations exist in the current ToxCast assays to detect specific neurotoxicants. In 2014, Valdivia and coworkers tested compounds from the ToxCast libraries using primary cortical cultures on microelectrode arrays (MEAs). They reported that 36 of 56 (64%) substances which tested negative in the ToxCast Novascreen ion channel assays were positive for changes in neuronal mean firing rate recorded on MEAs. It should be noted that 20 of the 36 (~55%) were recognized neurotoxic compounds, and included cholinesterase inhibitors, GABA_A antagonists and modulators of sodium channel kinetics. In particular, compounds acting on voltage-gated sodium channels, especially pyrethroids, were not readily detected by the relevant ToxCast assays. There is only a single sodium channel gene target, Scn1, in the existing test suite. Thus, the lack of sensitivity to detect pyrethroids, which modify sodium channel function (Soderlund et al., 2002), and exhibit use-dependent effects is not unexpected. (Soderlund, 2010;2020; Kadala et al., 2011) The following year, another report illustrated that the ToxCast battery does not contain some targets associated with neurotoxicity MIEs, and that these gaps in the biological coverage can produce conflicting conclusions. This comparison between the published literature for endosulfan and methidathion (documented in vivo neurotoxicants) and ToxCast assay results indicated positive and negative findings were endpoint dependent (Silva et al., 2015). Concordance was investigated for neural, estrogenic, androgenic, and developmental toxicity. However, both chemicals were inactive in the assays existing at the time the review was conducted. Because endosulfan is a noncompetitive GABA_A antagonist, its negative effects may be due to the target of the in vitro assays, which is specific to the agonist receptor binding site (Silva et al., 2015). Methidathion requires metabolic activation to a neurotoxic metabolite, and it is well known that metabolic activation is not included in many in vitro assays and has been identified as an area for future development (DeGroot et al., 2018). More recently, Paul Friedman et al. (2020) reported discrepancies for organophosphate and carbamate pesticides between ToxCast assay results and previously published findings. The case study compared point of departure (POD) values of high-throughput new approach methodologies (NAMs) derived data from “the 50th (POD_{NAM, 50}) and the 95th (POD_{NAM, 95}) percentile credible interval estimates for the steady-state plasma concentration used in in vitro to in vivo extrapolation of administered equivalent doses” to POD values from traditional in vivo neurotoxicity testing. Of the 448 chemicals that were analyzed, 48 compounds (~11%) had a calculated log₁₀ POD ratio₉₅ < 0 where “a log₁₀POD ratio of less than zero indicates that the POD_NAM is greater than the POD_traditional” (Paul Friedman et al., 2020). Chemical domain analysis of the 48 compounds found 24 were specifically associated with chemical structure features of carbamate or organophosphate pesticides. Thus, in addition to metabolic activation, the complexity of the in vivo situation makes detection of certain chemical classes with sufficient sensitivity to be “health protective” an area for improvement of the assay battery NAMs. The ability of a battery of NAMs to be health protective using general measures of toxicity (as opposed to specific types of apical toxicity) has recently been identified as a future consideration for risk assessment (Browne et al., 2024). However, this concept is beyond the scope of the present work, which focused on the current single gene targets and how they may be relevant for neurotoxicity/developmental neurotoxicity.

Our evaluation of ToxCast assay coverage relative to neurotoxicity has focused on those assays that are linked to specific genes. However, it should be noted that there are numerous assays in ToxCast that are not linked to a gene, and many of those assays are relevant to the nervous system. These are cellular assays where the readouts can be influenced by compound effects on any one or more of multiple targets. For example, compound effects on function of neural networks grown on microelectrode arrays (MEAs) following exposure during formation of the networks, or acute exposure are reported as the CCTE_Shafer_MEA_dev (Frank et al., 2017) and CCTE_Shafer_MEA_acute assays (Strickland et al., 2018), respectively (see Supplemental Table 1 for assay details). Acute activity of cortical neural networks grown on MEAs can be influenced by actions on voltage-gated ion channels, neurotransmitter receptors, and transporters. For example, as noted above, this assay reported effects of pyrethroids that were not detected by the assay linked to the voltage-gated sodium channel (Valdivia et al., 2014). During network formation, activity can be impacted by actions on any of these targets as well as processes that influence neurite outgrowth and synaptogenesis, among others. There are also data from other assays (cell migration, neurite area, radial glia migration) that were designed to inform about the potential developmental neurotoxicity hazard of compounds, which also cannot be linked to specific genes (Sachana et al., 2021). It is possible that these assays cover some of the gaps identified in this manuscript. However, the extent to which these gaps have been sufficiently covered will require additional analysis. Finally, ToxCast includes many cytotoxicity assays for a variety of different cell types. Cytotoxicity in many cases also cannot be linked to a specific gene and may rather be the result of generalized disruption of cellular machinery and/or cell stress. Judson and colleagues (2016) determined that a “cytotoxicity burst” where nonspecific disruption of multiple endpoints occurs in conjunction with cytotoxicity can be described for many compounds in ToxCast. In cases where an assay “hit” occurs in neural cells at concentrations below this “cytotoxicity burst” an indication of a selectivity of the compound for the nervous system may be justified.

Oxidative Stress (OS) was the predominant KE identified in the neurotoxicant list derived from the reference libraries of neurotoxic chemicals (Spencer and Schaumburg, 2000; Casarett, 2008). This may be related to the multiple insults/pathways that can result in OS, making detection of its effects evident after many initiating mechanisms. Although ToxCast has added a cellular respiration test (Hallinger et al., 2020), currently there is not a suite of specific gene targeted assays for OS in the battery. By definition OS is a complex molecular process involving multiple components and substrates that results in an imbalance of reactive oxygen species and antioxidants that can lead to mitochondrial dysfunction and cellular damage (Kodavanti, 1999; Kodavanti et al., 2011; Bhatti et al., 2017; Lushchak et al., 2018; Lushchak and Storey, 2021). Oxidative stress is known to induce neurodegenerative disease and developmental neurotoxicity (Abdollahi et al., 2004; Nishimura et al., 2021). The brain is particularly vulnerable to OS. For example, adrenalin, dopamine and serotonin are prone to auto-oxidation and subsequent generation of reactive oxygen species, thereby creating increased neural OS sensitivity (Cobley et al., 2018). Dopamine oxidation can induce mitochondrial dysfunction and neuroinflammation that damages dopaminergic neurons of the substantia nigra in the pathogenesis of Parkinson’s disease (Yoo et al., 2003; Guo et al, 2018). Additionally, Nishimura et al. (2021) document OS as a common KE in numerous DNT adverse outcome pathways in vertebrate in vivo studies. Consequently, OS is an integral component of the neurotoxicity induced by many environmental chemicals. Although 32 ToxCast intended gene targets have the potential to be associated with downstream OS pathways, apart from CAT, GADD45A, NRF1, and NFE2L2, they are not directly involved in REDOX/Energy homeostasis. Mitochondrial membranes are especially vulnerable to chemical insult and baseline toxicity may increase permeability, triggering dysfunction at lower concentrations than other biomolecular targets (Vinken and Blaauboer, 2017, Escher et al., 2020). Our analysis indicates that the addition of tests explicitly targeting OS mechanisms would enhance the coverage of the ToxCast battery to detect OS mediated neurotoxicity.

5. Conclusion

These results demonstrate that a large percentage (~40%) of the intended gene targets of the existing ToxCast assays are relevant to neural and developmental neural function. However, due to the cellular and biological diversity of the nervous system, this analysis indicated several areas where expanded coverage would be beneficial to increase coverage for potential neurotoxicity/developmental neurotoxicity screening in ToxCast. Specifically, increased coverage for additional receptor subtypes, ion channels and mechanisms of oxidative stress have been identified as needs for future development. Among the more important targets for development would be additional assays that reflect the diversity of different subtypes of important neurotransmitter receptors such as glutamate, GABA, and nicotinic receptors, as well as important ion channels including the voltage-gated sodium, calcium, and potassium channels. Finally, assays that evaluate additional targets critical to generation and mitigation of oxidative stress as identified before, including superoxide dismutase, Coenzyme Q, transferrin, peroxiredoxin, metallothionein, glutathione and glutathione peroxidase, are needed to provide greater coverage of the biology underlying this important endpoint. While the use of high-throughput screening and NAMs have shown promise to reduce animal testing, additional work is still needed to assure coverage of the biological and known neurotoxicant MIE/KE space.

HIGHLIGHTS.

• We evaluated ToxCast assay gene targets for relevance to neurotoxicity.

• 40% of 481 targets are relevant to neural function; 3.3% are specific to neurodevelopment.

• Gene targets were compared to molecular initiating and key events for 548 neurotoxicants.

• This work identifies missing neuro-relevant MIE/KE targets in existing screening programs.