Archives of Toxicology

The presence of N-nitrosamine drug substance-related impurities (NDSRIs) in pharmaceuticals represents a significant regulatory and safety challenge due to their classification as "cohort of concern" compounds. This paper describes the toxicological evaluation of N-Nitrosotrimetazidine (NTMZ), performed to refine the initial default acceptable intake (AI) limits of 18 to 26.5 ng/day established by regulatory authorities. The evaluation followed a tiered approach: NTMZ was first confirmed as mutagenic in vitro via the standard Ames test. To further investigate its genotoxic potential, two in vivo studies were conducted in Wistar and transgenic rats. Detection of DNA strand breaks in the liver and duodenum (comet assay) together with positive results in the cII mutation assay confirmed an in vivo mutagenic mode of action. Benchmark Dose (BMD) analysis of the transgenic rat data yielded a BMDL50 of 7 mg/kg/day in the male liver. To characterize long-term carcinogenic risk, a GLP-compliant 2-year carcinogenicity study was conducted in Wistar rats. Chronic exposure induced dose-dependent increases in liver tumors (hemangiosarcomas, hepatocellular carcinomas and adenomas) and intestinal tumors (adenomas and adenocarcinomas), leading to a Tumor Dose 50 (TD50) of 23 mg/kg/day in male rats. Benchmark dose analysis of tumor incidence identified a lowest BMDL10 of 2.6 mg/kg/day in females, which served as the basis for deriving an AI of 13 microg/day. This assessment demonstrates a strong predictive correlation between the BMD derived from the in vivo transgenic model, the BMDL10 and the final TD50 values obtained in the 2-year carcinogenicity study. These findings provided the scientific basis for establishing a conservative AI of 13 microg/person/day based on the BMDL10 and further support the regulatory acceptance and use of BMD-derived approaches for the evaluation of nitrosamine impurities.

Abstract Background. Acrylamide, a neurotoxicant in heated foods and smoke, is linked to occupational neuropathy, but evidence regarding chronic, low-level population exposure remains limited. We evaluated the association between acrylamide exposure biomarkers and peripheral neuropathy among U.S. adults. Methods. A total of 2,266 NHANES 2003-2004 participants (age >40) were analyzed. Exposure was assessed via hemoglobin adducts (HbAA/HbGA); neuropathy via monofilament testing >1 site). Survey-weighted logistic regression models adjusted for confounders. Sensitivity analyses included cubic splines, diabetes stratification, and multiple imputation. Results. Neuropathy prevalence was 15.5%. In adjusted models, neither adduct was associated with neuropathy (HbAA OR: 0.98, 95% CI: 0.82-1.17; HbGA OR: 0.91, 95% CI: 0.77-1.08). No dose-response gradient was observed. Expected risk factors (age, diabetes) showed strong associations, validating model sensitivity. The null result remained robust across sensitivity analyses, including a stricter outcome definition and multiple imputation (pooled OR: 0.97, 95% CI: 0.83-1.14). Conclusions. Acrylamide adducts were not associated with peripheral neuropathy in this national sample. General population levels (~55-70 pmol/g) lie well below established occupational no-observed-adverse-effect levels (~510 pmol/g) and clinical neuropathy thresholds (~6,000 pmol/g), providing a mechanistically coherent explanation for this null result.

Integrating high-throughput in vitro data into next-generation risk assessment (NGRA) workflows requires screening strategies that yield quantitative potency estimates and mechanistically interpretable biological signals. Transcriptomic and morphological profiling are increasingly adopted for early-stage hazard identification by enabling triage of substances for resource-intensive follow-up and prioritizing candidates most likely to present meaningful risk. In this study, we aimed to characterize biological concordance and uncertainty by quantifying how well high-throughput transcriptomics (HTTr) and Cell Painting PLUS (CPP) bioactivity profiles recover target-relevant biological signals in immortalized human renal proximal tubule epithelial RPTEC/TERT1 cells using 313 reference chemicals with high-confidence target annotations. Through quality control procedures and biological activity filters we yielded 142 reference chemicals spanning 66 different targets, which were systematically evaluated for biological concentration-responses by HTTr and CPP. HTTr was evaluated using TXG-MAPr-based qualitative and quantitative gene network activity analysis. HTTr showed the most prominent activity for targets that were highest expressed in RPTEC/TERT1 cells. Active chemical-pairs showed strong gene network activity correlation albeit with different potencies. Similarly, the highest transcriptomic concordance was observed for reference chemicals acting in the same pathway, such as EGFR/MEK or PI3K/AKT/mTOR. CPP often showed high sensitivity primarily at the organelle level providing limited statistical power for chemical grouping. Collectively, the results support HTTr and CPP as complementary early-tier assays within an in vitro weight-of-evidence safety testing framework. Although CPP is suitable as a cost-effective screening modality, HTTr offers higher mechanistic resolution for mode-of-action inference in high-throughput bioactivity screening and therefore remains necessary for high-confidence mechanistic interpretation.

Liver tumors are the most common carcinogenic outcome observed in two-year bioassays used for regulatory risk assessment of chemicals and pharmaceuticals. Because these traditional long-term studies are resource-intensive and not feasible for most chemicals, there is a critical need to develop alternative approaches that can identify potential rodent liver carcinogens in short-term studies (e.g., [≤] 1 month). Biomarkers associated with molecular initiating events (MIEs) of early rodent liver tumorigenesis may enable the prediction of two-year outcomes from such studies. Here, we identified transcriptional changes associated with six key initiators of rat liver tumorigenesis: five transcription factors (TFs) associated with non-genotoxic modes of action (AhR, PXR, CAR, PPAR, and ER), and tumor suppressor protein p53, which is activated by diverse cellular stressors including genotoxicity. Three known chemical activators for each TF were administered orally once daily for five days at doses associated with tumorigenic responses in previous rat cancer bioassays. RNA-seq identified differentially expressed genes (DEGs) that were consistently regulated by at least two activators in wild-type but not in corresponding TF knockout animals. Gene set enrichment analysis of ranked differential expression results demonstrated enrichment of pathways reflecting established TF biology. When comparing the consensus DEGs for each TF to a library of chemically induced, liver-specific gene expression profiles, we found that in most cases the chemicals with the most similar profiles to the consensus lists were known to activate the corresponding TF. Collectively, these findings identified candidate gene sets that may have utility for monitoring the induction of carcinogenesis-associated MIEs in short-term studies.

Histological necrosis is the reference measure of hepatotoxicity but can only be obtained at terminal cull. A time course requires independent cohorts at every timepoint. Circulating alanine aminotransferase (ALT) can be sampled repeatedly in the same animal, and is released in liver injury. We asked whether serial ALT kinetics can be used to estimate histological necrosis in vivo. Forty-five fasted twelve-week-old male C57BL/6J mice received a single intraperitoneal dose of 350 mg/kg acetaminophen (paracetamol). Plasma ALT and microRNA-122 (miR-122) were measured serially from baseline to 48h as cumulative area-under-the-curve (AUC) to cull, against centrilobular necrosis on haematoxylin and eosin sections as reference (range 0 to 59%, mean 32%). A generalised additive model of necrosis on cumulative ALT AUC and time since dosing predicted necrosis with a leave-one-animal-out cross-validated mean absolute error of 7.7% necrotic area (95% CI 5.6 to 10.0; n=49). A panel of traditional regression and machine-learning models all gave equal or larger error, and adding an additional biomarker or regeneration information did not improve prediction. Serial ALT kinetics therefore provide a calibrated, longitudinal measure of hepatic necrosis in vivo and give a more stable estimate of within-group variance for study planning, while supporting reductions in animal use, because one serially-sampled cohort can replace separate cohorts at each timepoint.

BACKGROUND Botanicals are widely consumed as herbal remedies and plant-derived products, but their complex and variable chemical composition presents major challenges for safety assessment. Although hepatotoxicity has been reported for several botanicals, the constituent compounds responsible and the underlying mechanisms often remain poorly characterised. OBJECTIVES We aimed to determine whether Cell Painting could identify candidate constituent drivers of botanical bioactivity and generate hypotheses about potential mechanisms of action in primary human hepatocytes. METHODS We profiled 13 botanical extracts and 85 known constituents across multiple concentrations using Cell Painting alongside assays of metabolic activity and membrane integrity. To interpret morphological phenotypes, we trained machine-learning models using a Cell Painting reference atlas of >1,000 chemicals linked to ToxCast/Tox21 molecular and cellular bioactivity annotations. RESULTS Cell Painting detected bioactivity in 11/13 botanical extracts and 46/85 constituents, compared with 9/13 extracts and 40/85 constituents in the cell-health assays, and generally detected activity at lower concentrations. For 9/13 extracts, at least one constituent was active at a lower concentration than the corresponding botanical extract, supporting the contribution of a subset of potent constituents to mixture-level responses. Aristolochia and goldenseal were the most potent extracts across Cell Painting and cell-health assays and contained highly active constituents. Reference-atlas models generated specific mechanistic hypotheses, including cytotoxicity-, endocrine-, and genotoxicity-associated responses for several botanicals and constituents. CONCLUSION These results support Cell Painting combined with machine learning as a sensitive approach for characterising cellular responses to botanicals, prioritising candidate constituent drivers of hepatotoxicity-associated phenotypes, and generating testable hypotheses about potential mechanisms of action.

Genetic variation impacts biological response to chemical and environmental exposures, contributing to differences in resistance or susceptibility. Forward genetic screens of genetically diverse cell populations can be used to identify the precise genetic variants that drive response variation. Cell lines from laboratory mouse genetic reference populations like the Diversity Outbred (DO) population are well powered for genetic mapping. Diverse stem cell panels are especially beneficial for studying early developmental effects of chemical exposures like triphenyl phosphate (TPHP), an organophosphate flame retardant linked to adverse developmental effects, including altered cell cycle in stem cells. Self-renewal and pluripotency are defining properties of stem cells during development; therefore, it is important to understand how chemicals like TPHP may alter these critical characteristics. To better understand the influence of genetic variation on TPHP exposure response, we utilized a genetically diverse panel of DO induced pluripotent stem cells (DO iPSCs). Our analyses characterize interline variation in gene expression through differential gene expression analysis and expression quantitative trait locus (eQTL) mapping. We identified genomic loci that contribute to variation in gene expression following TPHP exposure, including a regulatory hotspot on chromosome 15. Gene set enrichment highlighted several pathways affected by TPHP including lipid metabolism, steroid hormone signaling, and cell cycle. Our study shows that genetic variation modulates the effects of TPHP on gene expression and cell cycle in stem cells. Additionally, our work demonstrates that genetically diverse cell panels like the DO iPSC resource offer a tractable and scalable approach for identifying the genetic determinants of chemical toxicity.

People are widely exposed to the antimicrobial cetylpyridinium chloride (CPC) via consumer products, but CPC is a mitochondrial toxicant with potency comparable to that of canonical mitotoxicants. CPC is largely unregulated despite growing usage, bioavailability, and ability to cross the blood-brain barrier. Previously, we showed, in several cell types at non-cytotoxic and exposure-relevant doses, CPC inhibits ATP and OCR, endpoints of the electron transport chain (ETC). Mitochondrial toxicity is linked to multiple diseases (e.g., diabetes, Parkinsons, myalgic encephalomyelitis), but CPC has not been studied epidemiologically, and little mechanistic information is available. To determine why OCR and ATP are hampered by CPC, we hypothesized that CPC inhibits individual ETC components, cardiolipin, or TCA enzymes. Here, we show that, in primary human skin cells, an immune mast cell model, and isolated mitochondria, CPC apparently inhibits multiple ETC Complexes. Detailed investigation pinpointed the mechanism to the distal end of ETC: Complex III-cytochrome C-Complex IV. Using multiple approaches, we show that CPC does not directly inhibit any of the Complexes (not even Complex I as earlier reported), nor TCA enzymes, nor coenzyme Q. Yet, we found that CPC exhibits mitotoxicity as potent as cyanide. Anionic lipid cardiolipin attracts cytochrome C to the inner mitochondrial membrane so that it may shuttle electrons from Complex III to IV. Despite not altering levels of cardiolipin, CPC hinders cytochrome C by electrostatically interfering with cardiolipin. To aid epidemiology, risk analysis, and predictive toxicology, we have determined the precise biochemical mechanism of action of this ubiquitous compound.

Tetrabromobisphenol A (TBBPA), is one the most widely produced brominated flame retardant, detected in human matrices including cord plasma, raising concern over its impact on embryonic development. Our previous work demonstrated that TBBPA disrupts craniofacial cartilage development in zebrafish; however, inhibition of bone morphogenetic protein (BMP) signaling did not rescue these defects, suggesting the involvement of alternative molecular mechanisms. Transcriptomic profiling revealed significant upregulation of aryl hydrocarbon receptor (Ahr) signaling genes, including cyp1a and cyp1c, as well as reactive oxygen species (ROS) responsive genes, such as nfe and gstp. Consistent with these findings, KEGG pathway enrichment analysis identified significant enrichment of pathways involved in xenobiotic metabolism, cytochrome P450-mediated metabolism, and molecular docking predicted stronger binding of TBBPA to Ahr2 than the Ahr2 agonist TCDD. Then we investigated the role of Ahr signaling and ROS in TBBPA-induced craniofacial cartilage defects using microinjection of Ahr2 translation-blocking morpholino and the antioxidant N-acetylcysteine (NAC), respectively. Immunohistochemistry confirmed concentration-dependent induction of Cyp1a and ROS at environmentally relevant concentrations (0.05;0.005 uM). Ahr2 knockdown using a translation-blocking morpholino rescued TBBPA-induced alterations in Cyp1a expression, ROS levels, DNA damage, the expression of chondrogenesis-related markers (Sox10 and Sox2), and epithelial to mesenchymal transition (EMT) markers (E-cadherin, N-cadherin, and Snail2). Furthermore, Ahr2 knockdown rescued several TBBPA-induced alterations in craniofacial cartilage parameters. Co-exposure of TBBPA with the antioxidant NAC rescued the ROS and DNA damage and rescued only ceratohyal cartilage length, whereas other cartilage parameters remained disrupted. These findings establish a causal role for the Ahr-Cyp1a axis in TBBPA-induced craniofacial cartilage developmental toxicity and identify ROS as an important downstream contributor, indicating that both ROS-dependent and ROS-independent Ahr mechanisms underlined the observed TBBPA induced craniofacial defects.

2,4-Dinitrochlorobenzene (DNCB) is a well-characterized skin sensitizer that has been widely used in immunological and toxicological research and, historically, in clinical immunotherapy. Although it is a well-investigated chemical, this is the first study focusing on the dose response behavior at low-level concentrations. The aim was to reveal potential hormetic effects due to its known Nrf2 inducting activity. Therefore, THP-1 cells were treated with low doses of DNCB and two endpoints were evaluated for hormetic responses: metabolic activity using a resazurin-based assay and immune activation by measuring CD86 and CD54 expression using flow cytometry. The results showed a significant hormetic effect on the metabolic endpoint at the lower cell density for both analyzed time points, and a hormetic tendency at the higher cell density. Metabolic activity increased to approximately 125% of the control at 0.05 micromolar DNCB. For the immunological endpoint a slight decrease in CD86 and CD54 surface marker expression was observed, up to -16% and up to -12% compared to control at 0.5 micromolar DNCB. These findings highlight the importance of including low dose concentrations when characterizing chemical dose-response relationships and evaluating toxicological risk.

Drug-induced liver injury (DILI) remains a major challenge in drug development because existing in vitro models poorly predict clinical hepatotoxicity. We coupled patient-derived human liver organoids (HLOs) with high-content imaging and automated machine learning (ML) to detect hepatotoxicity-associated phenotypes by morphological profiling. HLOs from six DILI patients and one healthy donor (n = 7) retained donor-specific baseline phenotypes in hepatocyte maturity, metabolism, and morphology. ML models trained per line on MAD-robustized, well-level-aggregated median features achieved compound-level AUROCs of 0.88 (95% CI 0.86-0.90) to 0.91 (95% CI 0.89-0.93) and balanced accuracies of 0.81-0.84 in distinguishing 54 hepatotoxicants from 20 non-hepatotoxic controls. Averaging predictions across all seven lines at an optimized threshold gave 94.4% sensitivity (95% CI 84.6-98.8%) and 85.0% specificity (95% CI 62.1-96.8%), substantially outperforming HLO cytotoxicity assays and cancer cell line profiling. Lines were moderately-to-highly concordant (Pearson r = 0.66-0.85), and benchmark concentration modeling revealed reproducible inter-line differences in phenotypic sensitivity (median 4.0-fold, several compounds exceeding 20-fold). Exposure anchoring of benchmark concentrations to human-equivalent plasma concentrations, using in vitro mass-balance dosimetry and curated plasma free fractions, placed organoid points of departure within 100-fold of clinical Cmax for 60 of 66 compounds (23-fold residual offset, 95% CI 12-45), and placed two of the three misclassified negative controls an order of magnitude above clinical Cmax. Sulindac injury signatures were significantly enhanced by TNF co-treatment, demonstrating capacity to detect pro-inflammatory-mediated DILI. This platform preserves inter-line variability while enabling reproducible, scalable multi-line consensus hazard classification; because compounds were tested over a common nominal concentration range, the readout defines hepatocellular hazard, which must be combined with exposure as an independent input to estimate clinical DILI risk.

Peptide toxicity is a critical safety and developability parameter in peptide discovery and therapeutic development, yet relevant information remains fragmented across databases, literature resources, and curated datasets. Here, we present MAOMAO, an ontology-guided FAIR-oriented resource that integrates and harmonizes peptide toxicity data from 54 sources. MAOMAO contains 71,701 unique peptide sequences across seven toxicity-related endpoints, represented as 501,907 sequence endpoint combinations with endpoint-specific evidence states and explicit encoding of unavailable information. The resource combines standardized terminology, a hierarchical toxicity vocabulary, evidence-aware state resolution, provenance-aware metadata, 41 physicochemical descriptors, 10 protein language model representations, and a one-hot baseline. It provides endpoint-specific benchmark partitions across splitting strategies and random seeds, reusable with numerical representations. MAOMAO establishes a reusable framework for peptide toxicity research and data-driven toxicology.

Human metabolism converts inorganic arsenic to the pentavalent methylated species MMA(V) and DMA(V), the forms most people excrete, and the forms long read as the end of a detoxification pathway. Whether a transporter sets how much of these metabolites reaches the genome has not been tested in a mammalian cell. We expressed human AQP3, AQP7, AQP9 or AQP10 in HEK293T and MRC5-SV40 cells and measured gamma-H2AX by flow cytometry across dose series of As(V), MMA(V) and DMA(V), pairing every aquaporin with a GFP-Tubulin control and an untransfected mock acquired in the same replicate. As(V) was inactive in HEK293T cells and only weakly active in MRC5-SV40 cells to 20 micromolar, and both methylated species damaged DNA only in the millimolar range, DMA(V) being the more potent of the two in both cell lines. Against that weak baseline, AQP9 and AQP10 raised DMA(V)-induced gamma-H2AX in HEK293T cells by roughly 17 percentage points over the matched control, more than doubling the damage the same exposure produced in control cells, whereas AQP3 and AQP7 changed it not at all. AQP9 alone remained active with MMA(V). The ranking held in MRC5-SV40 fibroblasts at one-sixth the size, and within single wells the damage rose with the amount of AQP9 a cell carried while the control was flat. Aquaglyceroporins therefore discriminate among arsenic species, and AQP9 and AQP10 turn a weakly genotoxic metabolite into a substantially more genotoxic one.

Biodiversity is globally threatened by chemical pollution, yet toxicity data remain unavailable for millions of species and tens of thousands of chemicals, severely limiting our ability to assess ecological impacts. Here we present TRIDENT-2, a multimodal artificial intelligence model for predicting chemical toxicity across evolutionarily diverse eukaryotic species. Trained on 560,780 toxicity assays spanning 82,775 chemicals, 6,793 species, and multiple exposure scenarios, TRIDENT-2 accurately predicts toxicity across Eukaryota with an average median absolute error ranging from 1.76 to 3.80. By jointly learning from chemical, biological, and experimental information, it remains accurate across broad chemical and taxonomic distances, allowing for toxicity assessment for species and chemicals beyond the current experimental evidence. Our findings demonstrate that artificial intelligence can help overcome longstanding data limitations in ecotoxicology, paving the way for improved decision-making and reducing chemical impacts on biodiversity and ecosystems.

Acute radiation syndrome remains a major consequence of high-dose ionizing radiation exposure, and effective pharmacological countermeasures are limited. Nitro-fatty acids possess anti-inflammatory and cytoprotective properties, suggesting potential effectiveness in radiation injury. Here, we investigated the radioprotective effects of 10-nitro-oleic acid (10-NO2OA) in mouse models of total body irradiation (TBI). C57BL/6 mice were exposed to either sublethal (4 Gy) or LD50-level (7.5 Gy) gamma irradiation and treated with 10-NO2OA according to pre-irradiation, post-irradiation, or combined treatment regimens. Survival, hematological responses, bone marrow cells (BMCs)/colony-forming units (CFU), plasma granulocyte colony-stimulating factor (G-CSF), NRF2-responsive signaling, and intestinal responses were evaluated. Administration of 10-NO2OA before and after irradiation significantly improved 30-day survival following 7.5 Gy TBI (57% vs. 20% in vehicle-treated controls). In contrast, post-irradiation treatment alone was ineffective, whereas pre-irradiation treatment alone produced a non-significant trend toward improved survival. The protective effect was absent in BALB/c mice, indicating strain dependence. Although 10-NO2OA did not prevent long-term radiation-induced cytopenia, it increased neutrophil and platelet counts, elevated plasma G-CSF levels shortly after irradiation, and transiently enhanced BMC counts. In irradiated BMCs, 10-NO2OA activated NRF2-responsive gene expression in vitro and in vivo. Exploratory intestinal analyses revealed improved macroscopic appearance and partial restoration of irradiation-suppressed cytokine and growth factor production. In conclusion, 10-NO2OA improves survival following LD50-level gamma irradiation and is associated with modulation of hematopoietic, cytoprotective, and intestinal responses. These findings identify 10-NO2OA as a promising candidate radioprotective agent and support a multifactorial mechanism of action.

Omics methods have been envisioned to complement traditional toxicological testing for chemical risk assessment, in which identifying adverse effects is a critical step. However, the high dimensionality of transcriptomic data has historically led to reliance on context-dependent analysis. Liberality is a quantitative index that reduces genome-scale data dimensionality, with its changes reflecting underlying biological phenomena. In this study, we measured liberality in mouse liver RNA sequencing (RNA-Seq) datasets from studies in which mice were exposed to the environmental contaminant 2,3,7,8-tetrachlorodibenzo-p-dioxin (TCDD) every 4 days for 28 or 92 days, comparing dose-liberality relationships. Both 28- and 92-day TCDD treatments increased liberality but exhibited different dose-liberality relationships. Analysis of genes contributing to liberality revealed that longer exposure duration induced more extensive alterations in transcriptomic architecture. These findings suggest that liberality may serve as an unbiased metric to assess the extent of treatment-induced transcriptome perturbation.

The muscle-relaxing effects of botulinum neurotoxin (BoNT) serotypes A and B are widely utilized in clinical and aesthetic medicine. While the mouse bioassay remains the gold standard for evaluating the potency of pharmaceutical BoNT preparations, a non-animal alternative would be preferable to comply with European Directive 2010/63/EU. The Binding and Cleavage (BINACLE) assay, which measures BoNT activity based on the receptor-binding and proteolytic properties of the toxin, represents a promising alternative. Previous studies showed that this method allows reliable and sensitive activity determination. Here, we describe a collaborative study involving ten international laboratories, which aimed to characterize the applicability of the BINACLE assay for the potency testing of pharmaceutical BoNT products. Our results demonstrate that the BINACLE assay enables reproducible relative potency determinations for BoNT/A, the most pharmaceutically relevant serotype, with variability within acceptable limits. Reproducible relative potency calculations were also demonstrated for BoNT/B, while the data highlighted opportunities for refinement of the BoNT/B BINACLE protocol to further enhance the robustness of the method. Overall, this work confirms the utility of the BINACLE assay as a reliable and sensitive method for determining BoNT product potency, thereby substantiating its applicability as an alternative to the mouse bioassay.

Environmental factors including chemical exposures are important contributors to Parkinsons disease (PD). Nearly all well-validated chemicals involved in PD affect mitochondria, and the great majority of those identified inhibit mitochondrial complex I, causing ATP depletion and oxidative stress. We hypothesized that inhibition of mitochondrial complex III would also cause dopaminergic neurotoxicity. Using Caenorhabditis elegans to evaluate the in vivo effects of complex III-inhibiting pesticides antimycin A and pyraclostrobin, we found that both caused selective dopaminergic neurotoxicity. We evaluated exacerbation of dopaminergic neurotoxicity by the presence of -synuclein, and pdr-1/PRKN and pink-1/PINK1 mutant backgrounds and found increased neurotoxicity for pdr-1. Complex III inhibition caused a more-oxidized cellular environment in those neurons and pharmacological and genetic antioxidant interventions rescued neurotoxicity, but energetic rescue attempts did not. Finally, optogenetic production of superoxide anion specifically at complex III caused dopaminergic neuronal damage. Thus, redox stress at complex III following chemical exposure causes dopaminergic neurotoxicity in vivo in C. elegans.

Hexafluoropropylene oxide dimer acid (GenX), a replacement for legacy per- and polyfluoroalkyl substances (PFAS), is increasingly detected in the environment, yet its chronic toxicity remains poorly characterized. Current safety assessments rely largely on short-term, high-dose studies that may not capture the biological consequences of long-term, low-dose exposure. To address this gap, we employed 3D human liver (HepG2/C3A) spheroids cultured in a continuously rotating bioreactor system (ClinoStar) to systematically evaluate dose- and time-dependent mRNA changes in response to GenX under environmentally relevant conditions. Spheroids were exposed to GenX (0.08-50 M, spanning environmentally relevant to mechanistically informative concentrations) for acute (4 days) and chronic (4 weeks) durations, followed by genome-wide TempO-Seq transcriptomic profiling and benchmark dose (BMD) modeling. GenX elicited pronounced non-monotonic mRNA changes in acute exposure conditions, with the greatest number of differentially expressed genes (DEGs) observed at an intermediate concentration (0.4 M). In contrast, chronic exposure exhibited a generally concentration-dependent increase in DEGs, except for the 10 M condition, indicating a more consistent dose-response relationship than acute exposure. Notably, acute and chronic exposures elicited qualitatively distinct mRNA changes with low concordance across matched concentrations, demonstrating that exposure duration was a major determinant of mRNA changes. Acute low-dose GenX exposure preferentially modulated mRNA encoding components of cell cycle-related pathways, whereas acute higher dose exposures suppress mRNA levels of the constituents of lipid metabolic pathways and increase expression of mRNA encoding proteins involved in stress- and toxicity-associated signaling. Chronic exposure revealed a different pattern of changes in mRNA expression not observed under acute exposure conditions, including suppression of cellular components involved in lipid-related pathways at the lowest concentration tested. At higher concentrations, mRNA levels of components of multiple metabolic pathways were altered. Benchmark dose modeling identified a significantly lower transcriptomic point of departure (tPOD) for chronic exposure as compared to acute exposure, suggesting increased cellular sensitivity to prolonged GenX exposure and supporting the relevance of chronic models for human exposure assessment. Collectively, these findings demonstrate that GenX elicits time-dependent and non-monotonic changes in mRNA levels of human liver (HepG2/C3A) spheroids, with distinct responses depending on the exposure duration and dose. This study, therefore, highlights the importance of incorporating chronic, human-relevant in vitro models and transcriptomic endpoints into PFAS risk assessment and suggests that conventional short-term assays may underestimate the biological impact of sustained low-dose exposure. Key message (Impact of the study)This study provides systematic comparisons of short term (4 day) versus longer term (4 weeks), environmentally relevant GenX exposure in human liver spheroids, revealing non-monotonic, time-dependent changes in mRNA levels encoding cellular components of lipid metabolism-related pathways with potential implications for appropriate dose and time exposure parameters for use in New Approach Methods to be applied in risk assessment.

Experimental assessment of developmental neurotoxicity (DNT) is costly, time-consuming, and difficult to scale across large chemical inventories. New approach methodologies (NAMs), including computational approaches based on machine learning, deep learning, and large language models (LLMs), offer new ways to screen and prioritize chemicals. However, these emerging computational approaches have largely focused on predicting toxicity endpoints, with less attention to the mechanisms underlying those predictions. For DNT assessment, understanding this mechanistic basis is equally important. In this study, we integrated LLM inference with the adverse outcome pathway (AOP) framework to connect DNT predictions with plausible mechanisms. We constructed an AOP-informed benchmark linking ChEMBL target-activity measurements for 5,861 compounds to 14 AOPs representing five neurodevelopmental outcomes and evaluated ten LLMs for the prediction of molecular initiating events (MIEs), key events (KEs), key event relationships (KERs), and adverse outcomes (AOs), as well as the generation of mechanistic explanations. Explanation quality was evaluated using both a naive LLM judge and a rubric-based LLM judge. Across the zero-shot models, the highest observed macro-F1 scores were 0.818 for MIEs and 0.463 for AOs, but only 0.056 for KEs and 0.034 for KERs, identifying downstream pathway reconstruction as the principal challenge. Few-shot prompting substantially improved performance, with Gemini 3.8 Flash under four-shot prompting achieving macro-F1 scores of 0.956 for AOs, 0.813 for MIEs, 0.492 for KEs, and 0.457 for KERs. Both LLM-judge configurations indicated improved mechanistic explanation quality with few-shot examples, although strong outcome prediction did not consistently ensure recovery of the recorded pathway. Our results demonstrate the potential of AOP-informed LLMs to support DNT assessment by linking predicted neurodevelopmental outcomes to plausible biological mechanisms and enabling predictions to be evaluated at both the outcome and pathway levels.