Polychlorinated biphenyls (PCBs) are environmental contaminants that can cause neurotoxicity. PCBs, such as PCB 95 (2,2′,3,5′,6-pentachlorobiphenyl), can be metabolized by cytochrome P450 enzymes into neurotoxic metabolites. To better understand how the metabolism of PCB 95 affects neurotoxic outcomes, we conducted a study on the disposition of PCB 95 in transgenic mouse models. The mice were given a single oral dose of PCB 95 (1.0 mg/kg) and were euthanized 24 h later for analysis. PCB 95 levels were highest in adipose tissue, followed by the liver, brain, and blood. Adipose tissue levels were significantly higher in wild-type (WT) mice than in Cyp2abfgs-null (KO) or CYP2A6-transgenic (KI) mice. We also observed genotype-dependent differences in the enrichment of aS-PCB 95 in female mice, with a less pronounced enrichment in KO than WT and KI mice. Ten hydroxylated PCB 95 metabolites were detected in blood and tissue across all exposure groups. The metabolite profiles differed across tissues, while sex and genotype-dependent differences were less pronounced. Total OH-PCB levels were highest in the blood, followed by the liver, adipose tissue, and brain. Total OH-PCB blood levels were lower in KO than in WT mice, while the opposite trend was observed in the liver. In male mice, total OH-PCB metabolite levels were significantly lower in KI than in WT mice in blood and the liver, while the opposite trend was observed in female mice. In conclusion, the study highlights the differences in the atropselective disposition of PCB 95 and its metabolites in different types of mice, demonstrating the usefulness of these transgenic mouse models for characterizing the role of PCB metabolism in PCB neurotoxicity.

Animal fat and iron-rich diets are risk factors for Parkinson’s disease (PD). The heterocyclic aromatic amines (HAAs) harman and norharman are neurotoxicants formed in many foods and beverages, including cooked meats, suggesting a role for red meat in PD. The structurally related carcinogenic HAAs 2-amino-1-methyl-6-phenylimidazo[4,5-b]pyridine (PhIP), 2-amino-3,8-dimethylmidazo[4,5-f]quinoxaline (MeIQx), and 2-amino-9H-pyrido[2,3-b]indole (AαC) also form in cooked meats. We investigated the cytotoxicity, DNA-damaging potential, and mitochondrial damage of HAAs and their genotoxic HONH-HAA metabolites in galactose-dependent SH-SY5Y cells, a human neuroblastoma cell line relevant for PD-related neurotoxicity. All HAAs and HONH-HAAs induced weak toxicity except HONH-PhIP, which was 1000-fold more potent than the other chemicals. HONH-PhIP DNA adduct formation occurred at 300-fold higher levels than adducts formed with HONH-MeIQx and HONH-AαC, assuming similar cellular uptake rates. PhIP-DNA adduct levels occurred at concentrations as low as 1 nM and were threefold or higher and more persistent in mitochondrial DNA than nuclear DNA. N-Acetyltransferases (NATs), sulfotransferases, and kinases catalyzed PhIP-DNA binding and converted HONH-PhIP to highly reactive ester intermediates. DNA binding assays with cytosolic, mitochondrial, and nuclear fractions of SH-SY5Y fortified with cofactors revealed that cytosolic AcCoA-dependent enzymes, including NAT1, mainly carried out HONH-PhIP bioactivation to form N-acetoxy-PhIP, which binds to DNA. Furthermore, HONH-PHIP and N-acetoxy-PhIP inhibited mitochondrial complex-I, -II, and -III activities in isolated SH-SY5Y mitochondria. Mitochondrial respiratory chain complex dysfunction and DNA damage are major mechanisms in PD pathogenesis. Our data support the possible role of PhIP in PD etiology.

Exposure to environmental tobacco smoke (ETS), which contains hundreds of toxic compounds, significantly increases the risk of developing many human diseases, including lung cancer. The most common method of assessing personal exposure to ETS-borne toxicants is by sampling sidestream smoke generated by a smoking machine through a sorbent tube or filter, followed by solvent extraction and instrumental analysis. However, the ETS sampled may not truly represent the ETS in the ambient environment, due to complicating factors from the smoke released by the burning end of the cigarette and from the absorption of the chemicals in the respiratory tract of the smoker. In this study, we developed and validated an alternative air sampling method involving breathing through a face mask to simultaneously determine personal exposure to 54 ETS-borne compounds, including polycyclic aromatic hydrocarbons, aromatic amines, alkaloids, and phenolic compounds in real smoking scenarios. The newly developed method was used to evaluate the risk associated with exposure to ETS released from conventional cigarettes (CCs) and that from novel tobacco products such as e-cigarettes (ECs) and heated tobacco products (HTPs), with the observation of cancer risk associated with exposure to ETS released from CCs significantly higher than that from ECs and HTPs. It is anticipated that this method offers a convenient and sensitive way to collect samples for assessing the health impacts of ETS exposure.

Unraveling the causes underlying polycyclic aromatic hydrocarbon phototoxicity is an essential step in understanding the harmful effects of these compounds in nature. Toward this end, we have studied the DNA interactions and photochemistry of N1-(anthracen-9-ylmethyl)ethane-1,2-diaminium dichloride in the presence and absence of NaF, KF, NaCl, KCl, NaBr, KBr, NaI, and KI (350 nm hν, pH 7.0). Exposing pUC19 plasmid to UV light in solutions containing 400 mM KCl formed significantly more direct strand breaks in DNA compared to no-salt control reactions. In contrast, NaCl increased DNA damage moderately, while the sodium(I) and potassium(I) fluoride, bromide, and iodide salts generally inhibited cleavage (I– > Br– > F–). A halide anion-induced heavy-atom effect was indicated by monitoring anthracene photodegradation and by employing the hydroxyl radical (•OH) probe hydroxyphenyl fluorescein (HPF). These studies revealed that among no-salt controls and the eight halide salts, only NaCl and KCl enabled the anthracene to photosensitize the production of high levels of DNA-damaging reactive oxygen species (ROS). Pre-irradiation of N1-(anthracen-9-ylmethyl)ethane-1,2-diaminium dichloride at 350 nm increased the amounts of chloride salt-induced •OH detected by HPF in subsequent anthracene photoactivation experiments. Taking into consideration that •OH and other highly reactive ROS are extremely short-lived, this result suggests that the pre-irradiation step might lead to the formation of oxidized anthracene photoproducts that are exceedingly redox-active. The fluorometric probes HPF and Singlet Oxygen Sensor Green revealed that KCl concentrations ranging from 150 to 400 mM and from 100 to 400 mM, respectively, enhanced N1-(anthracen-9-ylmethyl)ethane-1,2-diaminium dichloride photosensitized •OH and singlet oxygen (1O2) production over no-salt controls. Considering the relatively high levels of Na+, K+, and Cl– ions that exist in the environment and in living organisms, our findings may be relevant to the phototoxic effects exhibited by anthracenes and other polycyclic hydrocarbons in vivo.

Aromatic aldehydes are ubiquitous in humans’ everyday life. As aldehydes, they can form imines (Schiff bases) with amino groups of skin proteins, leading to immune response-triggered allergic contact dermatitis. Many known aromatic aldehydes are considered as weak or nonsensitizers, but others like atranol and chloratranol, two components of the fragrance oak moss absolute, show strong sensitization potency. This large discrepancy in potency and, in particular, the underlying reaction mechanisms are only little understood so far. To reduce this knowledge gap, our chemoassay employing glycine-para-nitroanilide (Gly-pNA) as an amino model nucleophile was applied to 23 aromatic aldehydes. The determined Gly-pNA second-order rate constants for imine formation (k1 ≤ 2.85 L·mol–1·min–1) and the imine stability constant (K ≤ 333 L·mol–1) are on the lower end of the known amino reactivity scale for aldehydes, confirming many aromatic aldehydes as less potent sensitizers in line with animal and human data. The substantially higher sensitization potency of atranol and chloratranol, in turn, is reflected by their unique reaction chemistry profiles, inter alia, identifying them as cross-linkers able to form thermodynamically more stable epitopes with skin proteins (despite low formation kinetics, k1). The discussion further includes a comparison of experimentally determined k1 values with computed reactivity data (Taft σ*), the impact of the substitution pattern of the aryl ring on the reactivity with Gly-pNA, and analytically determined adduct patterns. Overall, this work provides new insights into the reaction of aromatic aldehydes with amino groups under aqueous conditions and fosters a better understanding of the chemistry underlying skin sensitization.

Drosophila shares maximum homology with the human disease-causing genes and thus has been employed to evaluate the toxicity of numerous compounds. Further, its distinguishable developmental stages, easy rearing, and short lifespan make it a perfect model organism to study toxicological properties of any new compound. The current study evaluates the toxic effect of a coumarin-based organic fluorescent dye, 7-hydroxy-4-methyl-8-((4-(2-oxo-2H-chromen-3-yl)thiazol-2-ylimino)methyl)-2H-chromen-2-one (CTC), using Drosophila melanogaster as a model organism by studying different behavioral, screening, and staining techniques using Oregon-R flies. For toxicity assessment, one control fly group was compared with various flies that had been subjected to fed CTC dye orally of different concentrations (0.5, 1, 2.5, and 5 μg/mL). The 3rd instar larvae were checked for the larvae crawling assay. The crawling assay demonstrates that the speed and path of the treated larvae are almost equal to the control ones, which signifies the non-neurotoxic property of CTC. Trypan blue assay further suggested that the dye does not cause any major damage to the gut. Phalloidin staining revealed that the actin composition remains unaltered even after the CTC treatment, while the DAPI staining experiment indicates that CTC does not cause any nuclear damage to fly gut cells. However, at a concentration of 5 μg/mL, CTC causes developmental delay. The flies hatched after larval treatment of CTC do not show any structural defects, suggesting clearly that CTC is also nongenotoxic to Drosophila. The current studies propose CTC as a noncytotoxic and nongenotoxic dye to track actin protein in the model organism D. melanogaster.

The search for chemical hit material is a lengthy and increasingly expensive drug discovery process. To improve it, ligand-based quantitative structure–activity relationship models have been broadly applied to optimize primary and secondary compound properties. Although these models can be deployed as early as the stage of molecule design, they have a limited applicability domain─if the structures of interest differ substantially from the chemical space on which the model was trained, a reliable prediction will not be possible. Image-informed ligand-based models partly solve this shortcoming by focusing on the phenotype of a cell caused by small molecules, rather than on their structure. While this enables chemical diversity expansion, it limits the application to compounds physically available and imaged. Here, we employ an active learning approach to capitalize on both of these methods’ strengths and boost the model performance of a mitochondrial toxicity assay (Glu/Gal). Specifically, we used a phenotypic Cell Painting screen to build a chemistry-independent model and adopted the results as the main factor in selecting compounds for experimental testing. With the additional Glu/Gal annotation for selected compounds we were able to dramatically improve the chemistry-informed ligand-based model with respect to the increased recognition of compounds from a 10% broader chemical space.

The correction to this paper is to Figure 16. The published version had the wrong chemical structure of isoalantolactone (IAL) in Figure 16B. The correct Figure 16 and its legend appear below. Author contributions have been added. Figure 16. Conjugation of GSH with AL (A) or IAL (B). X.L., H.L., and Y.G. equally contributed to the work. This article has not yet been cited by other publications. Figure 16. Conjugation of GSH with AL (A) or IAL (B).

7,8-Dihydro-8-oxo-2′-deoxyguanosine (8-OxodGuo) is a ubiquitous DNA damage formed by oxidation of 2′-deoxyguanosine. In this study, plasmid DNA containing 8-OxodGuo located in three mutational hot spots of human cancers, codons 248, 249, and 273 of the Tp53 tumor suppressor gene, was replicated in HEK 293T cells. 8-OxodGuo was only a weak block of replication, and the bypass was largely error-free. The mutations (1–5%) were primarily G → T transversions, and the mutation frequency was generally lower than that of the chemically related Fapy·dG. A unique 8-OxodGuo mutation spectrum was observed at each site, as reflected by replication in translesion synthesis (TLS) polymerase- or hPol λ-deficient cells. In codon 248 (CG*G) and 249 (AG*G), where G* denotes 8-OxodGuo, hPol η and hPol ζ carried out largely error-free bypass of the lesion, whereas hPol κ and hPol ι were involved mostly in error-prone TLS, resulting in G → T mutations. 8-OxodGuo bypass in codon 273 (CG*T) was unlike the other two sites, as hPol κ participated in the mostly error-free bypass of the lesion. Yet, in all three sites, including codon 273, simultaneous deficiency of hpol κ and hPol ι resulted in reduction of G → T transversions. This indicates a convincing role of these two TLS polymerases in error-prone bypass of 8-OxodGuo. Although the dominant mutation was G → T in each site, in codon 249, and to a lesser extent in codon 248, significant semi-targeted single-base deletions also occurred, which suggests that 8-OxodGuo can initiate slippage of a base near the lesion site. This study underscores the importance of sequence context in 8-OxodGuo mutagenesis in human cells. It also provides a more comprehensive comparison between 8-OxodGuo and the sister lesion, Fapy·dG. The greater mutagenicity of the latter in the same sequence contexts indicates that Fapy·dG is a biologically significant lesion and biomarker on par with 8-OxodGuo.

Recently, many electronic cigarettes (ECIGs) manufacturers have begun offering e-liquids, known as “nicotine salts”. These salts that have started gaining big popularity among users can be formed by adding weak acid to e-liquid mixtures consisting of propylene glycol (PG), vegetable glycerin (VG), flavors, and nicotine. The latter can exist in two forms: monoprotonated (mp) and freebase (fb) based on the pH of the matrix. Over the years, the determination of the fraction of fb was found important to policymakers as the prevalence of this form in ECIGs has been associated with the harshness sensory of inhalable aerosols. Liquid–liquid extraction (LLE), 1H NMR, and Henderson–Hasselback have been developed to deduce the fraction of fb; however, these methods were found to be time-consuming and have shown some challenges mainly due to the presence of a non-aqueous matrix consisting of PG and VG. This paper presents a quick non-aqueous pH measurement-based method that allows a quick determination of the fraction fb by just measuring the pH and the dielectric constant of the e-liquid. Then, by inputting these values into an established mathematical relationship, the fraction fb can be deduced. The relationship between pH, dielectric constant, and fb relies on knowing the values of the acidity dissociation constants of nicotine, which were determined for the first time in various PG/VG mixtures using a non-aqueous potentiometric titration. To validate the proposed method, the fraction fb was determined for commercials and lab-made nicotine salts utilizing the pH and LLE methods. The variation between the two methods was (<8.0%) for commercial e-liquids and lab-made nicotine salts containing lactic acid and salicylic acid. A larger discrepancy of up to 22% was observed for lab-made nicotine salts containing benzoic acid, which can be attributed to the stronger affinity of benzoic acid to toluene in the LLE method.

Nongenotoxic (NGTX) carcinogens induce cancer via other mechanisms than direct DNA damage. A recognized mode of action for NGTX carcinogens is induction of oxidative stress, a state in which the amount of oxidants in a cell exceeds its antioxidant capacity, leading to regenerative proliferation. Currently, carcinogenicity assessment of environmental chemicals primarily relies on genetic toxicity end points. Since NGTX carcinogens lack genotoxic potential, these chemicals may remain undetected in such evaluations. To enhance the predictivity of test strategies for carcinogenicity assessment, a shift toward mechanism-based approaches is required. Here, we present an adverse outcome pathway (AOP) network for chemically induced oxidative stress leading to (NGTX) carcinogenesis. To develop this AOP network, we first investigated the role of oxidative stress in the various cancer hallmarks. Next, possible mechanisms for chemical induction of oxidative stress and the biological effects of oxidative damage to macromolecules were considered. This resulted in an AOP network, of which associated uncertainties were explored. Ultimately, development of AOP networks relevant for carcinogenesis in humans will aid the transition to a mechanism-based, human relevant carcinogenicity assessment that involves a substantially lower number of laboratory animals.

Here, we reported a spontaneous reaction between anticancer drug doxorubicin and GTP or dGTP. Incubation of doxorubicin with GTP or dGTP at 37 °C or above yields a covalent product: the doxorubicin-GTP or -dGTP conjugate where a covalent bond is formed between the C14 position of doxorubicin and the 2-amino group of guanine. Density functional theory calculations show the feasibility of this spontaneous reaction. Fluorescence imaging studies demonstrate that the doxorubicin-GTP and -dGTP conjugates cannot enter nuclei although they rapidly accumulate in human SK-OV-3 and NCI/ADR-RES cells. Consequently, the doxorubicin-GTP and -dGTP conjugates are less cytotoxic than doxorubicin. We also demonstrate that doxorubicin binds to ATP, GTP, and other nucleotides with a dissociation constant (Kd) in the sub-millimolar range. Since human cells contain millimolar levels of ATP and GTP, these results suggest that doxorubicin may target ATP and GTP, energy molecules that support essential processes in living organisms.

The health risks associated with glyphosate (GLY) have recently received increasing attention. However, its potential vascular toxic effects in occupationally exposed populations remain unclear. This study assessed the effects of GLY on human aortic vascular smooth muscle cells (HAVSMCs) and the relationship between GLY and atherosclerosis. The results demonstrate that GLY induces a relatively larger and more flattened cell morphology, which is typical of cellular senescence and promotes senescence-associated β-galactosidase activity, as well as the expression of p53, p21, and p16 proteins in HAVSMCs. Regarding toxic effects, GLY induces the accumulation of reactive oxygen species, DNA damage, and mitochondrial damage in HAVSMCs. Mechanistically, the nuclear factor erythroid 2-related factor 2–Kelch-like ECH-associated protein 1 pathway is activated in response to oxidative stress produced by GLY. In an in vivo model, GLY led to dyslipidemia and macrophage recruitment in zebrafish vasculature. In conclusion, our results demonstrate that GLY induces vascular toxicity and may be a potential risk for atherosclerosis. These findings highlight the need for concern about cardiovascular risk in occupational populations chronically exposed to GLY.

Overdose of acetaminophen, a widely used antipyretic and analgesic drug, is one of the leading causes of drug-induced acute liver injury in the United States and worldwide. Phase-I metabolism of acetaminophen generates the toxic N-acetyl-p-benzoquinone imine (NAPQI) intermediate. Reactions of NAPQI with a wide range of biomolecules cause increased oxidative stress, endoplasmic reticulum (ER) stress, inflammation, and mitochondrial dysfunction, some of the cellular events contributing toward liver toxicity. Previously, we evaluated the potential of an FDA-approved, ER stress-modulating antihypertensive drug, Wytensin (trans-guanabenz, E-GA), as an antidote for acetaminophen hepatotoxicity. E-GA prevented elevation of the liver enzyme alanine aminotransferase (ALT), even when administered up to 6 h after acetaminophen overdose, and exhibited synergistic analgesic interactions. However, the commercially available guanabenz exists solely as a trans-isomer and suffers from sedative side effects resulting from the inhibition of central α2A-adrenergic receptors in locus coeruleus. Here, we studied the utility of the relatively unexplored cis-isomer of guanabenz as a treatment option for acetaminophen-induced liver toxicity. cis(Z)-Guanabenz acetate (Z-GA) lacks interaction with α2A-adrenoreceptors and is thus devoid of sedative, blood-pressure-lowering side effects of E-GA. Treatment of mice with Z-GA (10 mg/kg) before acetaminophen overdose and up to 6 h post APAP administration prevented liver injury and suppressed the elevation of serum ALT levels. Mechanistically, hepatoprotective effects of both isomers are similar and partly attributed to attenuation of the ER stress and oxidative stress in the liver. The results of this study suggest that Z-GA may be a safer, effective antidote for the clinical management of acute liver injury resulting from acetaminophen overdose. It also raises a tantalizing possibility of a prophylactic combination of the geometric isomer of the approved drug guanabenz with acetaminophen in a clinical setting.

The US Food and Drug Administration (FDA) regulatory process often involves several reviewers who focus on sets of information related to their respective areas of review. Accordingly, manufacturers that provide submission packages to regulatory agencies are instructed to organize the contents using a structure that enables the information to be easily allocated, retrieved, and reviewed. However, this practice is not always followed correctly; as such, some documents are not well structured, with similar information spreading across different sections, hindering the efficient access and review of all of the relevant data as a whole. To improve this common situation, we evaluated an artificial intelligence (AI)-based natural language processing (NLP) methodology, called Bidirectional Encoder Representations from Transformers (BERT), to automatically classify free-text information into standardized sections, supporting a holistic review of drug safety and efficacy. Specifically, FDA labeling documents were used in this study as a proof of concept, where the labeling section structure defined by the Physician Label Rule (PLR) was used to classify labels in the development of the model. The model was subsequently evaluated on texts from both well-structured labeling documents (i.e., PLR-based labeling) and less- or differently structured documents (i.e., non-PLR and Summary of Product Characteristic [SmPC] labeling.) In the training process, the model yielded 96% and 88% accuracy for binary and multiclass tasks, respectively. The testing accuracies observed for the PLR, non-PLR, and SmPC testing data sets for the binary model were 95%, 88%, and 88%, and for the multiclass model were 82%, 73%, and 68%, respectively. Our study demonstrated that automatically classifying free texts into standardized sections with AI language models could be an advanced regulatory science approach for supporting the review process by effectively processing unformatted documents.

Arsenic exposure can perturb gut microbiota and their metabolic functions. We exposed C57BL/6 mice to 1 ppm arsenic in drinking water and investigated whether arsenic exposure affects the homeostasis of bile acids, a group of key microbiome-regulated signaling molecules of microbiome–host interactions. We found that arsenic exposure differentially changed major unconjugated primary bile acids and consistently decreased secondary bile acids in the serum and liver. The relative abundance of Bacteroidetes and Firmicutes was associated with the bile acid level in serum. This study demonstrates that arsenic-induced gut microbiota dysbiosis may play a role in arsenic-perturbed bile acid homeostasis.

Despite an exponential increase in PFAS research over the past two decades, the mechanisms behind how PFAS cause adverse health effects are still poorly understood. Protein interactions are considered a significant driver of bioaccumulation and subsequent toxicity from re-exposure; however, most of the available literature is limited to legacy PFAS. We utilized microcalorimetric and spectroscopic methods to systematically investigate the binding between human serum albumin (HSA) and perfluorocarboxylic acids (PFCAs) of varying chain lengths and their nonfluorinated fatty acid (FA) counterparts. The results reveal the optimal chain length for significant PFCA–HSA binding and some fundamental interactions, i.e., the polar carboxylic head of PFCA is countered by ionizable amino acids such as arginine, and the fluorocarbon tails stabilized by hydrophobic residues like leucine and valine. Additionally, fluorine’s unique polarizability contributes to PFCA’s stronger binding affinities relative to the corresponding fatty acids. Based on these observations, we posit that PFCAs likely bind to HSA in a “cavity-filling” manner, provided they have an appropriate size and shape to accommodate the electrostatic interactions. The results reported herein widen the pool of structural information to explain PFAS bioaccumulation patterns and toxicity and support the development of more accurate computational modeling of protein–PFAS interactions. TOC graphic created with Biorender.com.

Smoking is a risk factor for bladder cancer (BC), although the specific chemicals responsible for BC remain uncertain. Considerable research has focused on aromatic amines (AAs), including o-toluidine (o-tol), o-anisidine (o-anis), 2-naphthylamine (2-NA), and 4-aminobiphenyl (4-ABP), which are linked to human BC based on elevated BC incidence in occupationally exposed factory workers. These AAs arise at nanogram levels per combusted cigarette. The unambiguous identification of AAs, particularly low-molecular-weight monocyclic AAs in tobacco smoke extracts, by liquid chromatography−mass spectrometry (LC−MS) is challenging due to their poor performance on reversed-phase columns and co-elution with isobaric interferences from the complex tobacco smoke matrix. We employed a tandem liquid–liquid and solid-phase extraction method to isolate AAs from the basic fraction of tobacco smoke condensate (TSC) and utilized high-field asymmetric waveform ion mobility spectrometry (FAIMS) coupled to high-resolution accurate mass (HRAM) Orbitrap LC−MS2 to assay AAs in TSC. The employment of FAIMS greatly reduced sample complexity by removing precursor co-isolation interfering species at the MS1 scan stage, resulting in dramatically improved signal-to-noise of the precursor ions and cleaner, high-quality MS2 spectra for unambiguous identification and quantification of AAs in TSC. We demonstrate the power of LC/FAIMS/MS2 by characterizing and quantifying two low-molecular-weight carcinogenic AAs, o-tol and o-anis, in TSC, using stable isotopically labeled internal standards. These results demonstrate the power of FAIMS in trace-level analyses of AA carcinogens in the complex tobacco smoke matrix.

The French Lentil & Leek Crumbles frozen food product was recently recalled due to reports of gastrointestinal issues. So far, 393 adverse illness complaints and 133 hospitalizations have been reported from consumption of this food, and the tara (Tara spinosa) protein flour ingredient is hypothesized to be responsible. A multipronged approach resulted in identification of (S)-(−)-baikiain in tara as a compound of interest due to its abundance, possible metabolic fate, and close resemblance to irreversible inhibitors of L-pipecolate oxidase. Oral administration of baikiain in ND4 mice showed a statistically significant increase in blood ALT levels and a reduction in liver GSH.

The Ames test is a gold standard mutagenicity assay that utilizes various Salmonella typhimurium strains with and without S9 fraction to provide insights into the mechanisms by which a chemical can mutate DNA. Multitask deep learning is an ideal framework for developing QSAR models with multiple end points, such as the Ames test, as the joint training of multiple predictive tasks may synergistically improve the prediction accuracy of each task. This work investigated how toxicology domain knowledge can be used to handcraft task groupings that better guide the training of multitask neural networks compared to a naïve ungrouped multitask neural network developed on a complete set of tasks. Sixteen S. typhimurium ± S9 strain tasks were used to generate groupings based on mutagenic and metabolic mechanisms that were reflected in correlation data analyses. Both grouped and ungrouped multitask neural networks predicted the 16 strain tasks with a higher balanced accuracy compared with single task controls, with grouped multitask neural networks consistently featuring incremental increases in predictivity over the ungrouped approach. We conclude that the main variable driving these performance improvements is the general multitask effect with mechanistic task groupings acting as an enhancement step to further concentrate synergistic training signals united by a common biological mechanism. This approach enables incorporation of toxicology domain knowledge into multitask QSAR model development allowing for more transparent and accurate Ames mutagenicity prediction.