Developmental neurotoxicity of thyroid hormone system-disrupting chemicals: a systems-level exploration using multi-omics approach

Abstract

There is increasing concern that thyroid hormone system–disrupting chemicals (THSDCs) may affect brain development during gestation and lactation. THSDCs comprise a wide range of natural and synthetic xenobiotics that activate diverse biological pathways. However, how disruption of specific molecular targets alters maternal thyroid hormone homeostasis and brain development in the offspring warrants further investigation. To address this question, this study investigates the effects of two THSDCs administered to pregnant rats from gestational day 6 through postnatal day 21: 5-propyl-2-thiouracil (PTU, 2.4 mg/kg/day), inhibitor of thyroid hormone synthesis, and pregnenolone-16α-carbonitrile (PCN, 300 mg/kg/day), inducer of hepatic enzymes involved in thyroid hormone metabolism. Circulating and brain thyroid hormone levels, enzymatic activities, and histopathology were assessed in dams and offspring. To further elucidate underlying mechanisms, multi-omics analyses combining proteomics, metabolomics, and spatial transcriptomics were performed on target organs including the thyroid gland, liver, and brain. Exposure to PTU resulted in severe thyroid hormone depletion in both serum and brain, accompanied by structural brain abnormalities, whereas PCN primarily induced hepatic enzyme activity with minimal effect on circulating thyroid hormone levels. Despite these distinct modes of action, multi-omics integration revealed convergent perturbations across molecular layers in the brain, particularly affecting energy metabolism and cytoskeletal organization with more pronounced effects observed following PTU exposure. Overall, multi-omics profiling enabled robust and highly sensitive identification of molecular signatures reflective of PCN exposure, without significant evidence of associated adverse toxicological effects. This approach highlights the value of multi-omics for mechanistic characterization and predictive assessment of THSDC-induced neurodevelopmental toxicity.

Supplementary Information

The online version contains supplementary material available at 10.1007/s00204-026-04303-4.

Introduction

Thyroid hormones (THs) are essential hormones that contribute to brain development in vertebrates, in both humans and rodents, due to their pleiotropic roles in promoting cellular proliferation, differentiation, and maturation across various neural and non-neural cell types (Alcaide Martin and Mayerl 2023). During gestation, the developing fetus is entirely dependent on maternal THs, which cross the placenta, until the fetal thyroid gland becomes fully functional in the later stages of pregnancy for humans and after birth for rats (Escobar et al. 2004).

Severe maternal TH deficiency in humans can result in stunted skeletal growth and irreversible neurodevelopmental impairments corresponding to cretinism syndrome (Segni 2017). Milder disruptions, such as maternal hypothyroxinemia defined as reduced circulating thyroxine (T4) with normal triiodothyronine (T3) and thyroid-stimulating hormone (TSH) levels, have been associated with decreased cognitive performance, lower IQ (Bélanger and Caron 2018), and an increased risk of neurodevelopmental disorders, including attention deficit hyperactivity disorder (ADHD) (Vermiglio et al. 2004) and autism spectrum disorders (ASD) (Thompson et al. 2018; Román et al. 2013).

Rodent models of maternal hypothyroidism have been instrumental in elucidating the role of THs in brain development, often through pharmacological induction of severe TH depletion using compounds such as 5-propyl-2-thiouracil (PTU) or methimazole (O’Shaughnessy et al. 2019). However, these models may not fully recapitulate the mechanisms by which environmental pollutants perturb thyroid function. Many environmental contaminants act through subtle and diverse pathways such as interfering with TH signaling, metabolism, or clearance, leading to altered circulating hormone levels without necessarily inducing overt developmental neurotoxicity (Street et al. 2024). For instance, exposure to hepatic enzyme inducers such as Phenobarbital has been shown to lower serum TH concentrations, yet often without corresponding structural or functional impairments in the developing brain, likely due to low-level effects and/or compensatory endocrine responses (Minami et al. 2023).

The limitations of current assessment strategies for thyroid hormone system-disrupting chemicals (THSDCs), lie particularly in the inadequacy of relying solely on serum TH measurements and most commonly thyroxine (T4), to predict neurodevelopmental outcomes. Multiple studies have demonstrated that changes in serum THs do not consistently correlate with TH concentrations in the brain, complicating the interpretation of systemic hormone levels as direct indicators of neurotoxicity (Marty et al. 2022).

To comprehensively characterize the mechanisms of action of THSDCs, it is essential to integrate classical toxicological endpoints including circulating and brain TH concentrations, key enzymatic activities, and histopathological evaluations, with systems-level approaches (Zekri et al. 2024). Specifically, omics technologies can provide powerful tools to capture dynamic changes in cellular pathways, offering a real-time, mechanistic understanding of how organisms respond to chemical stressors. This integrative strategy aims not only to detect early signs of toxicity but also to identify biomarkers of both exposure and effects (Bernhard et al. 2023).

This research reports the findings from a multiparametric pre- and postnatal development toxicity study in rats, investigating two THSDCs: PTU, an antithyroid agent that reduces THs synthesis by inhibiting thyroperoxidase (TPO) in the thyroid and impairs T4-to-T3 conversion through selective inhibition of type 1 deiodinase (DIO1) in target tissues; and 5-pregnen-3β-ol-20-one16α-carbonitrile (PCN) a synthetic steroid known to induce hepatic enzymes, particularly CYP3A and UGT2B in rodents, resulting in enhanced metabolism and clearance of THs. The objective of this work is to highlight an integrated toxicological approach that combines established reference parameters for THSDCs characterization with cutting-edge, high-sensitivity omics technologies. This holistic approach allows for a comprehensive assessment of chemical’s potential effects on the whole organism and contributes to the development of more refined tools for chemical screening and hazard identification.

Furthermore, these findings, in combination with our previously published data, particularly those in 2D and 3D in vitro models and metabolomics approach (Clavel Rolland et al. 2024), as well as advanced toxicokinetic studies in rat of the same compounds (Clavel Rolland et al. 2025), may ultimately support the development of more predictive models extrapolating for human health risk assessment.

Materials and methods

Test chemicals

PTU (5-propyl-2-thiouracil, CAS number: 2954-52-1, 98% purity), PCN (5- pregnen-3β-ol-20-one-16α-carbonitrile, CAS number: 1434-54-4, 97% purity), phenacetin, bupropion, urea, thiourea, dithiothreitol, 12-hydroxylauric acid, Dodecyl maltoside, and Methanol were obtained from Sigma-Aldrich (Saint-Quentin-Fallavier, France), Acetonitrile, formic acid, methanol, were obtained from Fischer Scientific (Illkirch, France). Trypsin at Promega (Charbonnières-les-Bains, France). The NeuN antibody was purchased from Sigma-Aldrich (Fontenay sous Bois, France).

Formol was purchased from Carlo Elba Reagents (Val de Reuil, France).

Animals, husbandry and mating

This project involving the use of laboratory animals has been authorized by the institutional ethical committee, the care and use of animals was conducted in accordance with the “Décret no 2013-18 du 1er février 2013 relatif à la protection des animaux utilisés à des fins scientifiques”: implementation into the French law of “The Directive 2010/63/EU of the European Parliament and the Council of 22 September 2010 on the protection of animals used for scientific purposes, by the Official Journal of the European Union, L276/33-79, 2010”. Sprague-Dawley Crl: CD (SD) presumed pregnant female rats (at gestational day (GD) 1 or 2) were purchased from Charles River Laboratories (Saint Germain Nuelles, France) at 11 to 13 weeks of age. The supplier used stock males from the same strain to mate with nulliparous females. All females used in the study were mated by independent males. The day on which evidence of mating was observed was designated as GD 0. Pregnant rats and dams with their litter were housed in suspended polycarbonate cages with controlled temperature (22 ± 2 °C), humidity (50 ± 20%), ventilation (10 to 20 air changes per hour), and photoperiod (12 h dark/ 12 h light from 6 am to 6 pm) throughout the study period. A commercially available pelleted rodent diet and filtered and softened local tap water were provided ad libitum throughout the study. Presumed pregnant rats were acclimated for 4 days prior to the initiation of the treatment. The animals were not subjected to overnight fasting prior to blood sampling or sacrifice.

Study design

An overview of the study protocol is presented in Fig. 1. Two studies (the phase 1 including prenatal cohort and phase 2 for pre- and postnatal cohort) were conducted in accordance with the DNT study design (OECD TG 426) with additional refined parameters. In both studies, animals were randomly assigned to one of three test groups: control, PTU, or PCN dosing group. All doses were administered once daily to dams, orally, by gavage, in a volume of 10 mL/kg body weight/day. Dose volumes were calculated based on the animal’s most recently recorded body weight. Control animals received an equivalent volume of vehicle alone (aqueous solution of 0.5% methylcellulose 400). In phase 1, PTU and PCN were administered to 10 females/group, at 2.4 mg/kg/day and 300 mg/kg/day, respectively, from GD6 through GD21. The PTU dose level was selected based on published studies reporting marked thyroid hormone depletion (> 50%) and specific brain lesions, including periventricular heterotopia (O’Shaughnessy et al. 2019; Goodman and Gilbert 2007), allowing the investigation of associated molecular perturbations in the absence of overt systemic toxicity. The PCN dose level was selected based on internal data demonstrating robust induction of hepatic enzyme activity (≥ 2-fold increase in UDP-glucuronosyltransferase activity) in the absence of systemic toxicity. In phase 2, 8 females/group were administered with PTU at 2.4 mg/kg/day during gestation (GD6 to GD21) and 0.64 mg/kg/day during lactation (Lactation Day (LD)1 to LD21). An internal range-finding study with PTU showed no effect on fetal body weight but revealed a 39% reduction in body weight gain during lactation at a dose of 2.4 mg/kg/day. Therefore, a lower dose (0.64 mg/kg/day) was selected for the lactation period to limit excessive postnatal exposure while still allowing the assessment of thyroid hormone–related molecular effects. PCN was administered at 300 mg/kg/day from GD6 through LD21. Mortality, clinical signs, body weight, and food consumption were periodically monitored throughout the study. Gestation index, duration of gestation, number of implantation sites, and number of pups delivered were evaluated to assess the potential effects on reproductive performance. On Postnatal Day (PND) 4, litter sizes were adjusted to yield, as closely as possible, 4 males and 4 females per litter. Although all three groups (control, PTU, and PCN) were concomitantly studied, data sets of the PTU and PCN groups are separately presented in each figure of this work for clearer elucidation of each substance’s effects. Consequently, the same data sets for the control group are presented in each PTU and PCN experiment.

Fig. 1.

Schematic summary of the study protocol. A DNT rat study with additional refined parameters was performed with pregnant Sprague Dawley rats being either exposed during gestation only (GD6-GD21), or during gestation and lactation (GD6-PND21). Several parameters were studied for each tissue and blood sampling

Developmental landmarks and behavior

Neonates were observed individually daily for:

• pinnae detachment: The ears (pinnae) unfold and lift away from the head, marking a transition toward more mature external ear structure.

• righting reflex: A reflex allowing the animal to orient itself upright when placed on its back, indicating developing motor coordination.

• auditory startle response: A sudden movement or flinch in response to a loud sound, reflecting early auditory system function and neural responsiveness.

• and eye opening: The eyelids separate and open, enabling visual input and marking a key step in sensory development.

Observations were scheduled to start approximately one or two days before the projected day of onset and continued until present for all pups in the litter or at PND20.

Tissue and biological sampling

One hour after the final treatment of the dams, all samples were collected at approximately the same time of day during the morning, in a randomized order across dose groups, to mitigate the impact of circadian rhythm. Rats were maintained in a separate adjacent room to the necropsy room to prevent the effects of any stress on hormone levels.

In phase 1, on GD21, pregnant females in each group were exsanguinated under deep anesthesia with isoflurane for the collection of blood samples from the abdominal aorta. Sera were separated from the blood for hormone analyses via centrifugation. The sera and plasma were stored at -70 °C or below until analysis. Following blood collection, pregnant females underwent cesarean section and necropsy. Gravid uteri were excised and weighed, in addition the following organs were collected and weighed: thyroid glands of dams were bisected, with one lobe designated for histopathology and the other for qPCR. Livers were sectioned, and the left lateral lobes was allocated for histopathology, a portion of the median lobe for qPCR, and the remaining liver tissue for hepatotoxicity assessment. Organs were either fixed in 10% neutral-buffered formalin (NBF) for histopathology or frozen in liquid nitrogen and preserved at -70 °C or below for qPCR and hepatotoxicity analyses. Live fetuses were extracted from the uteri and weighed individually.

Following external examination, blood was collected from all live fetuses after decapitation. The collected blood was pooled by sex in each litter to reach 600µL per sample, processed for sera, and stored in a freezer (-70 °C or below). After blood sampling, the brain and liver were collected from selected fetuses for qPCR (3 fetuses/litter). Additionally, per litter, when possible, 2 livers were collected for omics, 5 fetuses’ brains for hormone quantification, and 2 other brains for omics, and were frozen in liquid nitrogen.

In phase 2, the day of completed parturition was designated as LD0. On LD21, the dams were exsanguinated under deep anesthesia with isoflurane for blood sample collection from the abdominal aorta. Sera were prepared and stored in the same manner as in phase 1. The liver and thyroid glands of the dams were weighed and stored for the same analyses as in phase 1.

As illustrated in Fig. 1, on PND4, the group size of each litter was standardized by random pup selection to 8 pups (4 males and 4 females, if possible). Extra pups were sacrificed by decapitation, and blood samples were collected in the same manner as for the fetuses. The brains were collected, frozen in liquid nitrogen, and stored at -70 °C or below for brain hormone analyses.

For pups sacrified on PND21, blood samples were collected from all pups, by the same method as noted above. In each litter, 1 brain, 2 livers (portions of the median lobe), and 4 thyroids samples were collected and frozen in liquid nitrogen for qPCR. 3 thyroid glands (dissected with the trachea), 3 brains, and 2 liver samples (left lateral lobes) were collected and fixed in 10% NBF for histopathology. In addition, 3 brain samples were collected for hormone analyses, and 2 liver samples for hepatotoxicity and were frozen in liquid nitrogen. For omics analysis 1 thyroid gland, liver (part of the median lobe), and 2 brain samples were selected and frozen in liquid nitrogen.

Serum TH and TSH analyses

Serum samples from dams, fetuses and pups were analyzed for the concentrations of T3, T4, and TSH. Serum samples were prepared in methanol. Thyroid hormones (T3 and T4) were extracted with a solid-phase extraction procedure with ethyl acetate and evaporated to dryness under nitrogen at 45 °C. Then samples were mixed with 100µL of H₂O/MeOH (60/40, v/v). T4 and T3 hormone concentrations were assessed by LC-MS/MS. The elution was performed using a mixture of solvent A (water containing 0.01% formic acid) and solvent B (acetonitrile containing 0.01% formic acid). Detection and quantification of T4, T3, and standards were carried out using Sciex T-Quad 6500 + system with electrospray (ESI) measured in positive mode and equipped with a column Thermo Accucore Phenyl-X 50*2.1 mm; 2.6 μm. T4 and T3 were detected by MRM (Multiple Reaction Monitoring) with the transitions 777.548/604.8 and 651.657/605.8 respectively.

The peak area of T4 and T3 and the theoretical concentrations of the calibration samples were fit to a linear function with 1/x weighting. The LLOQ for serum T4 and T3 were 0.38 µg/L and 0.015 µg/L, respectively.

TSH level measurements were performed using the Luminex MAP^® technology with a Bio-Plex^® 200 system (Bio-Rad) and Merck Millipore kits (Milliplex^® Map Kit Rat Merck, Ref.RPTMAG-86 K). The detection limit for TSH was 0.389 ng/mL, with inter- and intra-assay variation < 10%.

All statistical analysis were performed using GraphPad Prism version 10.1 for Windows (GraphPad Software, San Diego, CA, USA).

Brain TH analysis

The frozen brains were grounded and extracted 2 times with ACN + 0.1% HCOOH at a volume of 1/1 (1 ml for 1 g of brain). After centrifugation, the extracts were diluted by half with water + 0.1% HCOOH. The thyroid hormones in the brain homogenate were measured using the same method as described above for serum hormone level measurements.

Hepatic enzyme activity

Microsomal preparations were extracted from frozen liver samples of the maternal rats and PND21 pups, and were subjected to differential centrifugation with KCl-washing, and microsomal protein content was determined using Thermo Scientific’s Pierce BCA Protein Assay Kit. Data were expressed in an activity in pmol/min/mg protein. Specific CYP450 activities were evaluated by employing standards with the following biotransformations: CYP1A: Phenacetin to Acetaminophen, CYP2B: Buproprion to (2 S,3 S)-Hydroxybuproprion, CYP3A: Midazolam to α-hydroxymidazolam and CYP4A: 12-Lauric acid to 12-hydroxylauric acid. UHPLC-HRMS equipment was used for the detection of products of enzymatic reactions. Samples were diluted with 240 µL of water: methanol (90:10, v/v) containing 0.1% formic acid and centrifuged at 4400 rpm for 10 min at 5 °C. Elution was performed using a mixture of solvent A (water containing 0.01% formic acid) and solvent B (methanol containing 0.01% formic acid). Detection and quantification of enzyme products and standards were carried out using a Thermo Vanquish LC coupled with a QExactive + spectrometer equipped with electrospray and coupled with a column Phenomenex Luna Omega Polar C18 (50*2.1 mm; 1.6 μm). The amount of each enzyme product was determined using a standard calibration curve from 0.2 to 50 µg/L which was renewed before each experiment. The lower limit of quantification (LLOQ) for Acetaminophen, (2 S,3 S)-Hydroxybuproprion, α-hydroxymidazolam, and 12-hydroxylauric acid was established at 2 µg/L.

UDP-glycosyltransferase (UGT) activity with T4 as substrate was determined by quantifying the formation of T4-glucuronide (T4-Gluc). T4-Gluc was extracted using a liquid-liquid extraction procedure. Briefly, 500 µL of ethyl acetate was added to each sample, followed by centrifugation at 14,000 rpm for 5 min. The supernatant was then evaporated to dryness under nitrogen at 45 °C. The dried residue was reconstituted in 1000 µL of H2O/ACN/formic acid (80/20/0.1, v/v/v). Chromatographic separation was achieved using a Phenomenex Kinetex EVO C18 column (50 × 2.1 mm, 1.7 μm particle size). The mobile phase consisted of solvent A (acetonitrile with 0.1% formic acid) and solvent B (water with 0.1% formic acid). Detection and quantification were performed using a Sciex QTRAP 6500 + mass spectrometer equipped with an electrospray ionization source operating in positive mode. T4-Gluc was detected using multiple reaction monitoring (MRM) with the transition 953.685 → 777.6. The peak area of T4-Gluc and the theoretical concentrations of the calibration samples were fitted to a linear function with 1/x² weighting. The lower limit of quantification (LLOQ) for the assay was established at 10 µg/L.

mRNA expression levels by quantitative real-time qPCR

Total RNA was isolated from the frozen tissues of individual control and treated animals using RNeasy Mini kits (Qiagen). RNA was quantified and controlled using the Quant-iTTM RiboGreenTM RNA Assay Kit (Invitrogen) on the Infinite F500 machine (Tecan). RNA integrity and quality were assessed based on the ribosomal RNA electrophoretic profiles using a Fragment Analyzer 5300 (Agilent). Total RNA was reverse transcribed using the Reverse Transcription Master Mix kit (Fluidigm/Standard BioTools). Pre-amplification of cDNA was performed using the Pre-Amp Master Mix kit (Fluidigm/Standard BioTools) and TaqMan probes for genes of interest (Applied Biosystems). Q-PCR reactions were performed on pre-amplified samples using the Standard TaqMan Gene expression assay (Applied Biosystems) on a BioMark machine (Fluidigm/Standard BioTools). 2 reference genes were used to normalize target gene expression data for quantitative calculations (reference genes with Ct standard deviation > 0.5 was not used for the calculation). Rnase-free water was used as a template for negative control. The list of test genes in the different target organs is presented in the Table 1 of supplementary data.

Histology, immunohistochemistry, and histopathology

The liver, thyroid, and brain sections were randomly selected for histopathological examination and were fixed in 4% formalin for 72 h after which they were dehydrated, cleaned, and embedded. The embedded samples were sectioned into 5 μm thick pieces and stained with hematoxylin-eosin (Roche HE600 stainer). A Leica microscope (DM 2000 LED Germany) was used to observe the histological structures. Besides, immunohistochemistry labeling was carried out with the Leica Bond RX stainer. The slides were dewaxed and pre-treated with Bond ER1 (citrate buffer pH 6) for 10 min. After blocking with 3–4% H2O2 (provided in Leica Microsystems kit, Nanterre, France) the slides were treated with the primary antibodies (Anti NeuN) for 30 min. Then these slides were incubated with the HRP-conjugated secondary antibodies (Bond Polymer Refine Detection: post primary 8 min. and polymer 8 min) and detected with DAB. Histopathological evaluation was performed by a board certified pathologist and a peer review was conducted by a pathologist experienced in developmental neurotoxicity studies.

Sample preparation for metabolomic analysis

On PND21, pup tissues were processed as follows: The whole thyroids, 5 mm square cubes of liver tissue, and longitudinal 1-mm thick sections from the entire brain were collected using a matrix specifically adapted to the developmental stage of the rat to ensure precise slicing at the brain center, and were frozen in liquid nitrogen. For fetuses on GD21, whole livers and brains were collected and frozen in liquid nitrogen. All samples were stored at -70 °C. Metabolites were extracted in 0.6 mL of methanol. Chemical homogenization was completed by mechanical grinding. After centrifugation of samples (13 000 g, 15 min, 4 °C), the supernatant was removed, dried using a SpeedVAC concentrator (SVC100H, SAVANT, Thermo Fisher Scientific, Illkirch, France), resuspended in 80 µL of a 20:80 acetonitrile-H2O mixture (HPLC grade, Merck Millipore) and stored at − 20 °C until LC-MS analysis.

Sample preparation for proteomics analysis

Whole livers and brains were collected from fetuses on GD21 and frozen in liquid nitrogen. Proteins were extracted in 0.2 mL or RIPA buffer (Pierce) in the presence of protease inhibitors (Complete from Roche) and phosphatase inhibitors (PhoStop from Roche). Chemical homogenization was completed by mechanical grinding. After centrifugation of samples (13,000g, 15 min, 4 °C), a volume of supernatant corresponding to 100 µg of protein was loaded onto a 10% urea polyacrylamide gel. After penetration of the proteins into the gel, electrophoresis was stopped, and the lane was cut at 40 kDa into 2 parts, allowing for the separation of high- and low-molecular-weight proteins for optimizing the proteomic analysis. Each gel portion underwent proteolytic in-gel digestion with trypsin. After centrifugation, the supernatants containing the digested peptides were vacuum-dried and resuspended in 50 µL of a 20:80 acetonitrile-H2O mixture and stored at − 20 °C until LC-MS analysis.

Untargeted metabolomic analysis

Chromatographic analysis was performed with the DIONEX Ultimate 3000 HPLC system coupled to a chromatographic column (Phenomenex Synergi 4 u Hydro-RP 80A 250_3.0 mm) set at 40°C and a flow rate of 0.9 mL/min. Chromatographic separation was achieved in 25 min using a gradient of mobile phase A (0.1% formic acid in water) and mobile phase B (0.1% formic acid in acetonitrile). MS analysis was carried out on a Thermo Scientific Exactive Plus Orbitrap mass spectrometer. The heated electrospray ionization source (HESI II) was used in positive and negative ion modes. The instrument was operated in full scan mode from m/z 50 to m/z 1000. Post-treatment of data was performed using the MZmine2 version 2.9.1 (http://mzmine.github.io/) and Compound Discoverer software 3.3 (Thermo Fisher Scientific, France). Metabolites were identified using the Human Metabolome Data Base (HMDB) and MZ cloud databases. Metabolite annotation was performed by m/z matching (Δ < 8 ppm) and MS/MS spectral matching against reference databases, corresponding to level 2 of the Metabolomics Standards Initiative (MSI) confidence scale (Sumner et al. 2007).

Untargeted proteomic analysis

The samples were analyzed by using an ESI-Q Exactive Plus mass spectrometer incorporating a Thermo Scientific Exactive Plus Orbitrap mass spectrometer coupled to a Nano LC Ultimate 3000 RSL C system (Thermo Fisher Scientific, France). An EASY-Spray C18 column of 25 cm×75 μm and 2 μm in diameter was connected and coupled to the system with an EASY-Spray source operating at 40 °C. Separation was achieved at a flow rate of 0.3 µL/min with a linear gradient from 5% to 45% of solvent B (80% acetonitrile, 20% water, 0.1% formic acid) over 120 min, where solvent A was 0.1% formic acid in water. Full-scan mass spectra were measured from 350 to 1500 m/z with an Automatic Gain Control Target. All raw MS data files were analyzed using Proteome Discoverer software 2.1 (Thermo Fisher Scientific, France) with the Sequest HT search engine, against a database of protein sequences (UniProtKB).

Spatial transcriptomic

Spatial transcriptomics were performed on 4 PND21 pups per group in two brain regions: corpus callosum and cerebellum. The nanoString GeoMx® Digital Spatial Profiler was used and the protocol followed the one described in (Goralski et al. 2024). All subsequent steps were performed using RNase-free conditions. Slides were processed using Leica automation platform. They were baked for 1 h at 65°C and de-paraffinized. Then, Antigen retrieval was performed in target retrieval reagent for 20 min at 100 °C. Slides were incubated in 1.0 μg/mL proteinase K for 15 min at 37 °C. Slides were then incubated with GeoMx WTA assay probe cocktail overnight in a hybridization oven at 37 °C. Following probe incubation, slides were washed with stringent washes (equal parts formamide and 4 × SSC buffer) at 37 °C twice for 25 min each. Then slides were incubated with morphology markers GFAP, MBP, and NeuN and nuclei marker Syto83 (marker information provided in supplementary data, Table 2) at 4 °C overnight. Slides were loaded onto the nanoString GeoMx® DSP instrument. Slides were fluorescently scanned and were imported onto the DSP platform. Region of Interest (ROIs) were generated using the polygon tool which aligned with the lesions of interest: periventricular heterotopia in the corpus callosum, cortex, and persistent external granular layer in the cerebellum. Each ROI was segmented into several areas of illumination (AOIs): “GFAP + ” for astrocytes, “MBP + ” for oligodendrocytes, and “NeuN + ” for neurons. Probe identities in each segment were captured via UV illumination in a 96-well plate and sequenced on Illumina NGS platform. Once data were acquired, quality control and data normalization were performed using the GeoMx DSP analysis suite following the guidelines provided by NanoString.

Statistical analyses

For in-life parameters, TH and TSH measurements in serum and brain, mRNA quantification, and hepatic enzyme activity, means and standard deviations were calculated for each group and each sex using Pristima software version 7.5.2 Build 9 (Xybion Corp.). The group means were compared at least at the 5% level of significance. All data were tested for normality. If data were not normally distributed, data were then transformed to normalized or analyzed with a non-parametric test. Data were analyzed with ANOVA and then the group means were intercompared to the control group using the Dunnett test.

Statistical analyses of the proteomic and metabolomic datasets were performed using MetaboAnalyst 4.0 (https://www.metaboanalyst.ca) and R (version 2025.05.1). Data were normalized, mean-centered, and scaled to unit variance prior to analysis.

For univariate testing, each metabolite/protein feature was assessed using a two-sided Student’s t-test, and significance was determined after Bonferroni correction for multiple comparisons.

For multivariate analysis, Principal Component Analysis (PCA) and Partial Least Squares–Discriminant Analysis (PLS-DA) models were constructed in MetaboAnalyst. Cross-validation to evaluate classification performance and reduce overfitting was performed for PLS-DA. Altered omic patterns were further visualized using heatmaps and UpSet plots generated in R.

Results

1. Evaluation of PTU effects.

1. PTU: General condition of dams.

No PTU-related clinical signs were observed through palpation or cage-side observation during the gestation or lactation periods. No variation in body weight was detected either in gestation or in lactation. The mean body weight gain was slightly decreased during the PND3-7 period, although not statistically significant, while maternal food consumption decreased significantly by 18% from GD13 to PND21 exclusively in the phase 2 cohort. However, no treatment-related alterations were observed in cesarean section parameters in maternal rats (Supplementary data Table 3).

• (b)PTU: Serum TH and TSH levels in dams.

On GD21, PTU exposure significantly decreased serum T3, T4, and free T4 (fT4) levels by 75%, 84%, and 23% respectively, with a significant increase in serum TSH level (+ 349%) but no changes in free T3 (fT3) level (Table 1). In PND21 dams, PTU exposure also significantly decreased serum free T4 (fT4) level by 19%, concomitant with a significant increase in serum TSH level (+ 186%) but no variation was measured in T4, T3, and fT3 levels. This lack of effect at PND21 is likely attributable to the PTU dose reduction during lactation from 2.4 to 0.64 mg/kg/day.

Table 1.

Thyroid hormone levels in serum and brain of dams and their offspring following PTU treatment. Thyroid hormones T3, T4, their free fraction fT4 and fT3, and TSH were quantified in dams at GD21 and PND21 and in their offspring at GD21, PND4, and PND21. The results shown are the percentage change compared to the control. For dams n = 8. For serum measurements in offspring, n = 20 at GD21, n = 8 at PND4, n = 8 at PND21, for brain measurements in offspring, n = 50 at GD21, n = 30 at PND4, n = 24 at PND21. Significant increase in thyroid hormone levels between treated and controls is indicated in orange, while decrease is shown in green. Dunnett LSD test: *p-value < 0.01; **p-value < 0.001; ***p-value < 0.0001

• (c)PTU: Thyroid pathological changes in dams.

PTU treatment significantly increased absolute (+ 109%) and relative thyroid gland weights (+ 100%) in dams at GD21, whereas no variation was observed at PND21. This lack of effect at PND21 is likely attributable to the PTU dose reduction during lactation from 2.4 to 0.64 mg/kg/day (Supplementary data Table 3). Correspondingly, moderate follicular cell hypertrophy was observed in all pregnant dams (GD21), while slight to moderate hypertrophy was observed in all lactating dams (PND21) (Supplementary data Table 4). qPCR analysis on thyroid tissue in dams at GD21 and PND21 revealed that mRNA levels decreased for the TSH receptor TshR (-46% and − 22%), deiodinase 3 Dio3 (-93% and − 70%), and thyroid hormone transporter Mct8 (-53% and − 39%), while mRNA levels increased in Tpo (+ 49% and + 31%), respectively. NIS symporter expression was elevated (+ 652%) exclusively in GD21 dams (Table 2).

Table 2.

mRNA quantification in thyroid and brain tissues following PTU treatment. Transcriptional expression levels were determined using quantitative real-time PCR. The results shown are the percentage change compared to the control. Genes that are significantly upregulated are shown in orange, while downregulated genes are shown in green. n=10 for dams (GD21), n=8 for dams (PND21), n=10 pups (GD21), n=10 pups (PND21). Dunnett LSD test: *p-value<0.01

• (d)PTU: Liver pathological changes in dams.

No significant findings were documented for liver weight or during histopathological examination of the liver tissue in dams at GD21 and PND21. qPCR revealed hepatic enzymes CYP2b1 mRNA level was increased (+ 240%) while UGT1A6 (-54%) and deiodinase 1 Dio1 (-60%) were decreased exclusively in GD21 dams (Supplementary data, Table 6a).

• (e)PTU: General condition of offspring.

The mean body weight of PND21 pups exposed to PTU exhibited a significant decrease both in males (-27%) and females (-30%) compared to controls. Pups from the PTU group demonstrated a 3-day delay in righting reflex and eyes opening developmental tests, and they also showed no reaction in the auditory startle response, suggesting their deafness (Supplementary data Fig. 2).

• (f)PTU: Serum and brain thyroid hormone levels in offspring.

Fetuses (GD21) from dams exposed to PTU showed significantly decreased serum T3 (-77%), T4 (-99%), and fT4 (-15%) levels (Table 1). There was also a significant increase in serum TSH level by 202%. PND4 pups exhibited significantly decreased serum T3 (-92%), T4 (-100%), and fT4 (-25%) levels, with a significant compensatory increase in serum TSH level by 477%. For PND21 pups in the PTU group, serum T3 (-43%), T4 (-86%), and fT4 (-47%) levels were significantly decreased with significant increase in serum TSH level by 306%.

Effects of PTU on brain T4 levels in offspring are shown in Table 1. Brain T4 level was significantly decreased in GD21 (-47%), PND4 (-92%), and PND21 (-89%).

• (g)PTU: Thyroid pathological changes in offspring.

The PTU treatment resulted in a significant increase in absolute (+ 89%) and relative weights (+ 171%) of thyroid in male pups (PND21), while only relative thyroid weight increased (+ 60%) in females (Supplementary data, Table 5). Correspondingly, in the thyroids of PTU-exposed PND21 pups, slight to moderate follicular cell hypertrophy was observed in all males, whereas hypertrophy was minimal to slight in all females (Supplementary data, Table 4). In PND21 pups, the thyroid mRNA levels of NIS (+ 324%) and thyroglobulin Tg (+ 48%) were elevated, while those of Dio3 (-88%) and MCT8 (-64%) were reduced (Table 2).

• (h)PTU: Liver pathological changes in offspring.

Rarefaction (glycogen store depletion) was observed in the liver of PND21 pups in both sexes likely due to reduced food consumption. This outcome could explain the body weight reduction in these pups. No alterations were detected in the mean liver weight of the pups compared to controls. However, mRNA levels were significantly decreased for hepatic enzymes CYP1A1 (-85%), CYP2B1 (-40%), CYP4A1 (-44%), UGT1A6 (-82%), UGT2B1 (-40%) as well as Dio1 (-62%), and were increased for CYP3A23 (+ 91%) and UGT1A1 (+ 100%) (Supplementary data, Table 6a).

• (i)PTU: Brain pathological changes in offspring.

No variation was detected in pups’ mean brain weight (Supplementary data, Table 5). In the PND21 pup brains, periventricular heterotopia was observed in 2 out of 8 males and 5 out of 8 females (Supplementary, data Table 4). The heterotopia consisted of a cluster of neuronal cells, confirmed by anti-NeuN immunohistochemistry, which was located in the corpus callosum in the periventricular area (Fig. 2). Additionally, in both sexes, all pups from the PTU group exhibited persistent external granular layer in the cerebellum area (Supplementary, data Table 4), which is also an accumulation of immature neurons. The brain mRNA levels of the Camk4 gene coding for serine/threonine protein kinase and transcription factor coding gene Klf9 were both decreased either at GD21 or PND21 by 15% and 17% for Camk4 and 24% and 28% for Klf9, respectively. All the other mRNA levels were only varying in PND21 pups, with decreases observed in Dio1 (-60%), Dio3 (-55%), a transcriptional corepressor for thyroid hormones Hairless gene Hr (-48%), a calcium-binding protein Parvalbumin (-39%), and the myelin proteolipid protein plp1 (-20%), while the Ntf3 gene coding for neurotrophic factor was increased by 35% (Table 2).

Fig. 2.

Neuron alterations in pups brain following PTU treatment. NeuN staining showed specifically misplaced neurons at the end of lactation (PND21) in pups treated with PTU. a periventricular heterotopia in the corpus callosum in the cerebrum and b persistence of the external granular layer in the cerebellum

• (2)Evaluation of PCN effects.

1. PCN: General condition of dams.

No PCN-related clinical signs were observed through palpation or cage-side observation during the gestation or lactation periods. No variation in mean body weight, body weight gain, and food consumption were detected either during gestation or lactation. No treatment-related alterations were observed in cesarean section parameters and reproductive performance in maternal rats (Supplementary data, Table 3).

• (b)PCN: Serum thyroid hormone in dams.

In GD21 dams, PCN exposure significantly decreased mean serum T4 levels by 16% (Table 3). TSH was significantly increased by 100% and no statistical changes were observed in T3, fT4 and fT3 levels. In PND21 dams TSH was significantly increased by 192% compared to controls, but no variation was observed for the other hormone levels.

Table 3.

Thyroid hormone levels in serum and brains of dams and their offspring following PCN treatment. Thyroid hormones T3, T4, their free fraction fT4 and fT3, and TSH were quantified in dams at GD21 and PND21 and in their offspring at GD21, PND4, and PND21. The results shown are percentage change compared to the control. For dams n = 8. For serum measurements in offspring, n = 20 at GD21, n = 8 at PND4, n = 8 at PND21, for brain measurements in offspring, n = 50 at GD21, n = 30 at PND4, n = 24 at PND21. Significant increase in thyroid hormone levels between treated and controls is indicated in orange, while decrease is shown in green. Dunnett LSD test: *p-value < 0.01

• (c)PCN: Thyroid pathological changes in dams.

In the thyroid gland, no absolute or relative thyroid weight variation was detected (Supplementary data, Table 3). However, minimal to slight follicular cell hypertrophy was observed in 4 of 9 GD21 dams and 4 of 7 PND21 dams of the PCN group (Supplementary data, Table 4). Also, NIS mRNA level increased (+ 117%) at GD21 exclusively, and TshR (-29%) and MCT8 (-31%) were decreased exclusively at PND21 (Table 4).

Table 4.

mRNA quantification in thyroid and liver tissues following PCN treatment. Transcriptional expression levels were determined using quantitative real-time PCR. The results shown are the percentage change compared to the control. Genes that are significantly upregulated are shown in orange, while downregulated genes are shown in green. n=10 for dams (GD21), n=8 for dams (PND21), n=10 pups (GD21), n=10 pups (PND21). Dunnett LSD test: *p-value<0.01

• (d)PCN: Liver pathological changes in dams.

Absolute and relative mean liver weight was increased but not significantly in GD21 dams and significantly increased (+ 22%) in PND21 dams (Supplementary data, Table 3). No pathological observations were recorded in the liver of GD21 and PND21 dams (Supplementary data, Table 4). However, PCN exposure significantly increased the activity of CYP3A by 1360% and UGT by 210% in GD21 dams and by 1240% and 90% in PND21 dams, respectively (Fig. 3). These findings correlate to mRNA levels as Ugt1a1 (+ 611% and + 334%) and Ugt2b1 (+ 296% and + 261%) significantly increased in GD21 and PND21 dams respectively, but CYP3a23 mRNA could not be measured (Table 4). Additionally, Cyp2b1 mRNA levels were also significantly increased in GD21 and PND21 dams (+ 2493% and + 1240% respectively) compared to controls as well as CYP1A1 (+ 244% and + 716%). Cyp4a1 mRNA level was significantly decreased in GD21 and PND21 dams (-42% and − 43% respectively). Sult2a2 (+ 424%) and Dio3 (+ 141%) increased and Dio1 (-32%) decreased only in GD21 dams compared to controls.

Fig. 3.

Hepatic enzyme activity following PCN treatment. Enzyme activities were determined by combining enzyme substrate quantification and protein quantification. Values are expressed as mean ± standard deviation, control data are displayed in grey and PCN data in orange, percentage change are indicated in red. n = 10 for GD21 dams, n = 8 for PND21 dams and n = 16 for PND21 pups. *p-value < 0.01 and ***p-value < 0.0001

• (e)PCN: General condition in offspring.

The PCN exposure did not impact mean body weight (Supplementary data, Table 5) and no delays on developmental landmark tests were observed (Supplementary data, Fig. 1).

• (f)PCN: Serum and brain thyroid hormone levels in offspring.

No variation was observed in serum or brain TH levels and serum TSH level in GD21, PND4, and PND21 pups (Table 3).

• (g)PCN: Thyroid and Brain pathological changes in offspring.

PCN exposure did not impact the absolute and relative thyroid gland and brain weights of PND21 pups (Supplementary data Table 5) compared to controls. Correspondingly, no pathological observation was recorded in both sexes. In the thyroid glands, no hypertrophy of follicular cells was observed in this group. Periventricular heterotopia as well as persistence of the external granular layer were not observed in the PND21 pups of the PCN group (Supplementary data Table 4).

• (h)PCN: Liver pathological changes in offspring.

The PCN exposure did not impact on the absolute and relative liver weight and no pathological observation was recorded in both sexes. However, PND21 pups showed a 90% increase in CYP3A activity with no variation in UGT activity (Fig. 3). This CYP3A induction was also shown with Cyp3a23 mRNA level increase in both GD21 and PND21 pups by 4713% and 212%, respectively, while Ugt1a1 was increased in both GD21 and PND21 pups by 280% and 51%, respectively, and UGT2b1 was increased by 414% only in GD21 pups. Additionally, Sult2a2 increased by 892% only in GD21 pups and Sult1e1 by 530% only in PND21 pups (Table 4).

• (3)Multi-omics approach on PTU and PCN.

A comparative analysis of PTU and PCN exposure in perinatal rats reveals distinct yet overlapping statistically significant metabolic and proteomic alterations across the thyroid gland, liver, and brain tissues, suggesting both shared and compound-specific mechanisms of toxicity.

1. Discrimination of exposure with statistical modeling.

The aim of statistical analysis in metabolomics or proteomics investigations was to assess the overall impact of chemical exposure (PTU or PCN) on systemic homeostasis and to identify potential biomarkers associated with such exposure. In this study, omics analyses were conducted on tissues collected from juvenile rats. We chose to illustrate the statistical analysis of the omics data using only the results obtained from fetal brain tissue collected at GD21 in rats exposed to either PTU or PCN, as these data are highly representative. However, metabolomic analyses of GD21 liver as well as PND21 liver, brain, and thyroid tissues, along with proteomic analyses of GD21 brain and liver, were also conducted using the same analytical pipeline and statistical workflow. These other results are presented in Supplementary data, Fig. 2.

Initial data processing enabled the detection of 1571 putatively annotated metabolic features in the PTU-exposed group and 1508 putatively annotated metabolic features in the PCN-exposed group. Each feature was tentatively associated with metabolites predicted from the MS1 spectra through databases matching (all data are presented in supplementary data, Excels files, metabo). Unsupervised statistical analysis was then performed using Principal Component Analysis (PCA), as illustrated in Fig. 4 (left panel). The PCA plots distinguish the exposed (in green) and control groups (in red), with each dot corresponding to a biological replicate. PCA revealed a clear separation between exposed and non-exposed groups for both PTU (Fig. 4a) and PCN (Fig. 4b) along the first two principal components, both explaining approximately 30% of total variance. These results indicated that exposure to PTU or PCN induces substantial alterations in the brain metabolome.

Fig. 4.

Multivariate statistical analysis of fetal rat brain metabolome following exposure to PTU or PCN. Brains from GD21 rats exposed to PTU (a) or PCN (b) were analyzed by metabolomics. The left panels show Principal Component Analysis (PCA) score plots (PC1 vs. PC2), highlighting separation between exposed and control groups. The right panels show Partial Least Squares–Discriminant Analysis (PLS-DA) score plots together with the corresponding performance table summarizing cross-validated classification accuracy, explained variance (R²), and predictive ability (Q²)

To further explore these differences and identify discriminant features, a supervised multivariate analysis using Partial Least Squares Discriminant Analysis (PLS-DA) was conducted for both exposure conditions. The performance and reliability of the PLS-DA models were evaluated through cross-validation, as summarized in the associated table (Fig. 4a and b). The Q² values were high (≈ 0.8 for both PTU and PCN), indicating strong predictive ability and robustness. The corresponding R² values were close to 1, demonstrating that the model components effectively explained the variance structure of the data. Classification accuracy was also high (close to 1), supporting the strong discriminatory power of the models.

• (b)Identification of metabolic discriminant biomarkers.

PLS-DA Variable Importance in Projection (VIP) scores were used to rank all detected metabolites according to their contribution to the differentiation between exposed and control groups. From this ranking, the most discriminant metabolites under PTU and PCN exposure were identified. A further selection step was performed to retain only those features for which annotation was confirmed at level 2 of the Metabolomics Standards Initiative (MSI) (Sumner et al. 2007). The complete list of metabolomic features is available in the supplementary material (Excel files, metabo). Table 5a and b display the metabolites identified in fetal (GD21) rat brain tissue following in utero exposure to PTU or PCN, respectively. Heatmaps of peak relative intensities together with fold change-values, illustrate that several metabolites were commonly affected by both treatments. Of these, cis, cis-9,12-Octadecadienoyl-L-carnitine and palmitoylcarnitine abundance, both markedly decreased in respect to controls, indicating disrupted mitochondrial β-oxidation and impaired lipid metabolism, which are critical for brain energy supply during development. Daucic acid was also upregulated in both conditions, especially in the PCN group, possibly reflecting altered lipid-related pathways. PTU exposure resulted in more pronounced changes, in addition of the marked accumulation of PTU itself. Also, increased levels of vaccenyl carnitine and decreased levels of alpha-tocotrienol, 2-aminooctadeca-4,14-diene-1,3-diol, and 5-sulfosalicylic acid, pointed to disturbances in fatty acid metabolism, oxidative stress, and detoxification. In contrast, PCN exposure led to reductions in tetradecanoylcarnitine, palmitoleoylethanolamide, methyl 3-oxo-12-oleanen-28-oate, leukotriene D5, and 4-oxoretinal, alongside a moderate increase in 2-(methylthio)-3 H-phenoxazin-3-one, indicating alterations in peroxisomal activity, inflammation, and retinoid metabolism. These findings highlight distinct yet converging metabolic disruptions in the developing brain with potential implications for neurodevelopmental outcomes.

Table 5.

Discriminant metabolites identified in fetal rat brain (GD21) following in utero exposure to PTU (a) or PCN (b). Metabolites were selected based on PLS-DA VIP scores and retained only when annotation was confirmed at level 2 of the MSI. Heatmaps display relative peak intensities (green = low, red = high) across control and treated groups, accompanied by fold-change values relative to controls. N = 10 per group

Metabolomic analyses were also performed on GD21 fetuses’ liver as well as PND21 pups’ liver, brain, and thyroid tissues, using the same analytical workflow. The results are presented in the Supplementary data (Excel files, metabo). Briefly, the most discriminating deregulated metabolites in the thyroid gland revealed that both PTU and PCN exposure led to increased levels of acyl-carnitines, including dodecanoylcarnitine, arachidonoylcarnitine, and oleoylcarnitine, suggesting a disruption of fatty acid metabolism. A consistent finding across both treatments was the downregulation of sphingosine, a key component of sphingolipid signaling pathways. In addition, PTU specifically increased iodide (I⁻) levels, reflecting its inhibitory impact on thyroid hormone biosynthesis. In the liver, both treatments induced alterations of acyl-carnitines and other metabolites associated with impaired β-oxidation. Notably, PCN exposure was associated with mitochondrial dysfunction, as with decreased S-3-oxodecanoyl cysteamine, and redox imbalance, evidenced by reduced pantothenic acid and elevated methionine levels.

• (c)Comparative metabolomics analysis between PTU and PCN exposure.

Figure 5a shows upset plots of the distribution of the top 50 metabolites selected by PLS-DA (obtain in the previous section) across liver, brain, and thyroid tissues in rat at GD21 and PND21 following exposure to either PTU or PCN. The data shows both treatment-specific and shared metabolic features alterations. In the liver and thyroid gland, most deregulated metabolites are unique to specific treatment and timepoint combinations, indicating distinct metabolic responses. In contrast, the brain exhibits a higher degree of overlap between PTU and PCN-induced changes, with 17 common metabolites identified at GD21, and a different set of 17 shared metabolites at PND21, suggesting timepoint-specific but convergent metabolic disruptions in the brain. Figure 5b displays a heatmap of the most relevant brain metabolites selected by PLS-DA at GD21 and PND21 for PTU and PCN, and merged together. Overall 109 metabolic features are displayed and clustered in the heatmap. At GD21, fetuses exposed to either PTU or PCN exhibit a distinct metabolomic signature (cluster D) compared to controls, characterized by widespread downregulation of a large group of metabolites. Conversely, another group of metabolites (cluster C) shows pronounced upregulation, more evident in PCN-treated animals than in those exposed to PTU. At PND21, the common deregulation between PTU and PCN exposure is less uniform as most of clusters of metabolites are specifically deregulated in PTU-treated animals compared to controls (cluster A being upregulated and cluster B downregulated), indicating that PTU exerts a more robust and specific impact on brain metabolism at this timepoint.

Fig. 5.

Metabolomic alterations induced by PTU and PCN in liver, brain, and thyroid tissues at different developmental stages. a Upset plots representing the distribution of the top 50 discriminant metabolites in liver, brain, and thyroid tissues of rat at GD21 and PND21 following in utero exposure to PTU or PCN. Each bar represents the number of unique or shared altered metabolites across the different treatment-timepoint combinations. b Heatmap of the most relevant brain metabolites identified by PLS-DA, representing their abundance profiles across treatments (PCN or PTU) and timepoints (GD21 and PND21). Metabolites are hierarchically clustered, and color scale indicates standardized abundance (Z-score)

• (d)Identification of protein discriminant biomarkers.

All detected proteins were subjected to statistical comparison between exposed and control groups using two-tailed Student’s t-tests. The resulting p-values were used to rank proteins according to the magnitude of their differential abundance. Proteins showing significant alterations, defined by an adjusted p-value < 0.1 and a minimum fold change of 2, under PTU and PCN exposure are illustrated in Fig. 6. A total of 159 proteins were considered in both the PTU and PCN groups. The complete list of protein features, along with their associated statistical parameters, are available in the Supplementary Material (Excel files, proteo). Figure 6 illustrates the most relevant differentially expressed proteins in fetal (GD21) rat brain tissue following in utero exposure to PTU or PCN. Differential expressions between control and treated samples were assessed using volcano plot analysis, based on log2 fold change (FC) and adjusted p-values. Proteins that were upregulated are highlighted in red, whereas downregulated proteins are shown in blue.

Fig. 6.

Protein biomarkers identification in the brain of fetal rats. Proteins are presented in a volcano plot with a significance measure on the Y-axis (− log10(p-value)) and effect size on the X-axis (logarithmized fold-change). Each dot corresponds to an identified protein, blue dots are less represented, and red dots are more represented in the treated group compared to the control. N = 5

In the PTU group, reduced expression of key cytoskeletal and motility-related proteins, such as SPTBN2, Myosin-9, and Myosin-11, suggests compromised axonal stability, neurite outgrowth, and neuronal connectivity. Decreased levels of SERCA2 point to disruptions in calcium-dependent signaling, while reduced 60 S ribosomal protein L3 may indicate suppressed protein synthesis and impaired neurogenesis. Conversely, elevated levels of ARF4, ARL2, and Syntaxin-1B suggest altered vesicle trafficking and synaptic activity. Upregulation of NME2 and enolase-phosphatase E1 points to metabolic adaptation, while higher expression of proteasome activator complex subunit 1 may reflect elevated protein turnover in response to cellular stress. Overall, these changes indicate disruptions in structural organization, signaling, and proteostasis.

In the PCN group, downregulation of Copine-1 and EEF1D suggests impaired calcium signaling and reduced protein synthesis, potentially affecting neuronal maturation and stress responses. In contrast, upregulation of proteasome activator subunit 1 and 26 S proteasome regulatory subunit 9 indicates increased proteolysis. Elevated levels of gelsolin, coactosin-like protein, and α-SNAP point to enhanced actin cytoskeleton remodeling and vesicular trafficking, processes essential for neurite outgrowth and synaptic plasticity. Overall, these alterations suggest increased proteolytic activity and impaired structural organization.

Proteomic analysis revealed that, although PTU and PCN exposures affected distinct sets of proteins, many of these are involved in common biological pathways, indicating converging disruptions in key processes underlying brain development.

• (e)Spatial transcriptomics.

To gain deeper insight into brain alterations in juvenile rats following chemical treatment, a spatial transcriptomic approach was employed. Spatial transcriptomics combines high-resolution imaging and immunofluorescence to visualize tissue architecture and cell types (Fig. 7). After imaging, selected regions of interest (ROIs) were segmented, and gene expression was analyzed using digital barcoding while preserving spatial context. RNA transcripts were localized and quantified for spatially resolved expression profiling. Statistical analyses were performed to identify significant differences between groups (Supplementary data, Fig. 3).

Fig. 7.

Immunofluorescence in periventricular heterotopia region in PTU condition. Immunofluorescence using antibody NeuN (Neurons), GFAP (Astrocytes), and MBP (Oligodendrocytes) for the staining and segmentation in the periventricular heterotopia region located in the corpus callosum

The first region of interest selected for spatial transcriptomic analysis in the brain was the periventricular heterotopia (PVH) located in the corpus callosum, identifiable histologically by a cluster of ectopic neurons (Figs. 2 and 7) after PTU exposure. Over 450 gene pathways were found to be deregulated in neurons of this region in PTU-treated animals (Supplementary Data, Table 7), including those involved in GABAergic interneuron function and glutamate metabolism. Neuron-specific deregulated pathways also included key regulators of synaptic plasticity. Although PVH lesions appear as clusters of misplaced neurons, spatial transcriptomics revealed gene expression alterations in surrounding glial populations. In astrocytes, 250 gene pathways were deregulated in PTU group, notably those related to cytoskeletal dynamics including tubulin post-translational modifications and the Rho, RhoA, and Ras GTPase signaling pathways which are critical for actin remodeling. These pathways are essential for astrocyte morphological plasticity which support neuronal migration. In oligodendrocytes, 46 gene pathways were deregulated in PTU group, with a predominant involvement in cholesterol and lipid metabolism, processes directly linked to myelination. Moreover, pathways associated with oxidative stress were consistently deregulated across all examined cell types.

Spatial transcriptomic analyses were also conducted on the persistent external granular layer (EGL) in the cerebellum following PTU exposure (Supplementary data, Fig. 4). Several of the deregulated pathways overlapped with those identified in the PVH region (Supplementary data, Table 7). However, unique to the EGL, pathways involved in protein maturation, localization, and degradation were markedly altered in both astrocytes and neurons. Concurrently, stress-response pathways were activated, including mTOR, AMPK, and ATM signaling.

For the PCN group, spatial transcriptomic analysis was only performed in the cerebellum region (Supplementary Data, Table 8). Over 160 gene pathways were found to be deregulated in neurons. Some pathways overlapped with those affected by PTU, including those related to synaptic plasticity, cytoskeletal organization, and stress adaptation. Notably, unlike in the PTU group, no pathways associated with protein maturation were altered in either neurons or astrocytes. Instead, extracellular matrix organization pathways were specifically enriched.

Discussion

This study is part of a broader effort to comprehensively characterize the mechanisms of action of THSDCs. Although this topic has been extensively investigated by the scientific community, each experimental model and methodological approach provides distinct and/or complementary insights. To address the complex issue of thyroid disruption, we therefore adopted a cross-disciplinary strategy with a specific emphasis on neurodevelopmental toxicity.

Our work began with the identification of molecular initiating events and molecular signatures that could be identified in several in vitro models (Clavel Rolland et al. 2024). This was followed by the integration of toxicokinetic analyses to better characterize inter-model and compound-specific bioavailability to target organs (Clavel Rolland et al. 2025). In the present study, we extended this framework using innovative omics-based approaches to investigate molecular signatures of two THSDCs in rats. These datasets are intended to support the development of physiologically based pharmacokinetic (PBPK) models, thereby improving the prediction of dose–response relationships across in vitro and in vivo systems and enhancing human health risk extrapolation. Furthermore, identifying molecular concordance between in vitro and in vivo models may contribute to the refinement and reduction of animal use in toxicological research.

Specifically, the current study assessed the impact of two THSDCs, PTU and PCN, on thyroid function, brain development, and systemic toxicity in a pre- and postnatal rat model. PTU, a well-characterized TH disruptor, served as a positive control due to its extensively documented effects. In contrast, PCN has been less studied, with existing data primarily focused on its role as a Pregnane X Receptor (PXR) agonist (Taneja et al. 2019) and its potential to disrupt estrogen metabolism (Zhu et al. 2000). To our knowledge, this is the first study to evaluate PCN’s impact on thyroid function and neurodevelopment in vivo.

PTU exposure of dams resulted in a pronounced reduction of serum T4 (-84%) and T3 (-75%) and increase in TSH (+ 349%) levels in GD20 dams, whereas no significant alterations in T4 and T3 and lower increase of TSH (+ 186%) were observed in PND21 dams (Table 1). This lower alteration at PND21, is likely due to the dose reduction during lactation (2.4 mg/kg/day during gestation (GD6 to GD21) and 0.64 mg/kg/day during lactation (LD1 to LD21). Despite the reduced PTU dose that we used during lactation, thyroid hormone levels in pups remained clearly disrupted. Indeed, in embryos and pups, T4 (-99%, -100%, -86%), T3 (-77%, -92%, -43%), and fT4 (-15%, -25%, -47%) concentrations were significantly decreased at GD21, PND4 and PND21 respectively. Regarding TSH, increases were consistent across timepoints. However, this rise was particularly pronounced in dams at GD21 (+ 349%), possibly reflecting a compensatory mechanism to sustain TH supply to the developing embryos. Indeed, pups may exhibit greater vulnerability as PTU can cross the placenta efficiently (Mortimer et al. 1997), thereby reducing maternal TH production, and thus the maternal supply to the fetus, as well as directly impairing fetal TH synthesis. (Hassan et al. 2017).

These TH disruptions were accompanied by histopathological changes in the thyroid gland as follicular cell hypertrophy was observed in all dams and pups at all timepoint. This indicates that, although the reduction in TH levels was mitigated by dose adjustment in dams, morphological alterations in the thyroid gland persisted.

In addition, molecular alterations were detected in both dams and pups with decreased expression of TshR, Dio3, and MCT8, and increased NIS and TPO mRNA levels in thyroid tissues, consistent with previous reports (O’Shaughnessy et al. 2018; O’Shaughnessy and Gilbert 2020). These gene expression changes represent relevant molecular biomarkers indicative of overactive thyroid (hypothyroidism).

As emphasized by Gilbert et al., a quantitative understanding of the relationship between serum and brain TH levels is essential, as serum T4 alone is an inadequate predictor of brain morphological outcomes (Gilbert et al. 2020; O’Shaughnessy et al. 2018). In this study (Table 1), PTU moderately reduced brain T4 levels in offspring at late gestation (by 47%), whereas the decrease was significantly greater during lactation (92% at PND4 and 89% at PND21). Indeed, in rats, brain development occurs predominantly after birth (Clancy et al. 2001) and involves key processes such as neuronal migration, synaptogenesis, and myelination, all of which are highly dependent on thyroid hormones (Koibuchi and Yen 2016). As a result, the need for thyroid hormones is greater during the postnatal period, which may explain why PTU has a more substantial impact on brain TH levels during lactation compared to gestation.

Measuring TH levels in offspring serum or brain alone may be insufficient to assess developmental neurotoxicity due to variability from sampling and experimental conditions. Thus, incorporating brain histopathology can significantly enhance the mechanistic understanding of chemical effects. A well-documented consequence of PTU exposure is the disruption of neuronal migration, leading to the formation of periventricular heterotopia (Goodman and Gilbert 2007; Hassan et al. 2017; O’Shaughnessy et al. 2019). In the present study (Fig. 2), this cell migration failure was observed but more frequently in female pups (5 out of 8) compared to males (2 out of 8). In addition, another neuronal migration-related abnormality was observed in the cerebellum, where all PTU-exposed pups exhibited a persistent external granular layer (Fig. 2), consistent with previously reported studies (Shimokawa et al. 2014; Fauquier et al. 2011). These findings suggest that, for screening purposes, the former alteration may be preferentially selected, as it is distributed throughout the cerebellum, unlike periventricular heterotopia, which is confined to a specific region and therefore more challenging to sample. However, a recent study from Ogata et al. (Ogata et al. 2024) reported that the persistence of the persistent external granular layer is transient and no longer detectable by PND28, whereas periventricular heterotopia represents a permanent structural abnormality.

At the molecular level, expression of genes directly regulated by T3 was decreased for Parvalbumin, Hr, and Plp1 transcripts, and increased for Ntf3 in PTU-exposed pups. These findings suggest impaired neuronal differentiation, myelination, and neurotrophic support, consistent with TH’s critical role in brain maturation (Alcaide Martin & Mayerl 2023). More specifically, we found decreased expression of Camk4 which is involved in biogenesis of the cytoskeleton of neurons and migration (Li et al. 2004) or Klf9 which is involved in maturation of Purkinje cells and elimination of granular layer of cerebellum (Avci et al. 2012). These expression changes are consistent with the histopathological alterations observed and may serve as relevant molecular biomarkers, particularly for the early detection of subtle toxicological effects preceding overt disruption of cellular organization.

Finally, delays in several developmental landmarks were observed (Supplementary data, Fig. 1), notably in the righting reflex and eye opening. These delays may result either directly from PTU exposure or indirectly from the associated reduction in body weight gain (- 28%) observed in PTU-treated animals. Indeed, delayed weight gain in rat pups is frequently associated with postponed maturation of motor and behavioral reflexes. These parameters are commonly used as an indicator of general toxicity or perinatal stress (Naik et al. 2015). However, the most pronounced effect was on the auditory startle response, as none of the pups from the PTU-treated group exhibited any hearing response. Indeed, TH deficiency during critical periods of development can disrupt cochlear structure and function. In rats, the greater epithelial ridge normally regresses during the first two postnatal weeks to form the inner sulcus, allowing proper suspension of the tectorial membrane, a process tightly regulated by T3 (Ng et al. 2013). In hypothyroid conditions, the greater epithelial ridge persistence has been observed and may contribute to auditory deficits (Sharlin et al. 2018).

In this study, we confirmed that PCN induced significant hepatic enzyme induction with CYPs UGTs and SULTs mRNA elevated in both dams and pups. Specifically, CYP3A enzyme activity increased by 1360% and 1240% in dams at GD21 and PND21. Additionally, UGT enzyme activity was elevated in dams, showing a 210% increase at GD21 and 90% increase at PND21. These findings were accompanied by a significant increase in liver weight in PND21 dams (22%). This is consistent with previous finding in rat (Vansell and Klaassen 2002) and indicate efficient activation of xenobiotic metabolism pathways. However, in pups, only a 90% increase in CYP3A activity was observed at PND21, with no changes in UGT activity. As demonstrated in our previous study (Clavel Rolland et al. 2025), PCN is not excreted through maternal milk, which likely explains the limited exposure of pups during lactation and, consequently, the minimal impact of PCN on hepatic enzyme activity.

Activation of hepatic enzymes, particularly UGTs, can enhance TH metabolism in the liver. However, PCN exposure led to only a modest reduction in maternal serum T4 levels (-16%) at GD21, accompanied by marked TSH induction at GD21 (+ 100%) and PND21 (+ 191%). Histological analysis revealed minimal to slight follicular cell hypertrophy in maternal thyroid glands at both timepoints, without any change in thyroid weight. In offspring, no histopathological alterations were detected in the thyroid gland, nor were there any significant changes in serum or brain TH levels.

Importantly, PCN exposure did not lead to notable developmental abnormalities in the offspring, as no brain histopathological alterations were observed. Furthermore, there were no reductions in body weight or delays in developmental landmarks. These findings are consistent with previous research indicating that mild TH suppression caused by hepatic enzyme inducers typically does not lead to overt neurodevelopmental toxicity (Street et al. 2024; Minami et al. 2023). This supports the notion that PCN’s mode of action, primarily through hepatic TH clearance, does not elicit the severe disruptions in thyroid hormone homeostasis observed with PTU.

To gain a deeper understanding of the molecular effects of THSDCs, we conducted a multi-omics analysis. This integrative approach offers a systems-level perspective on how two representative THSDCs perturb the transcriptomic, proteomic, and metabolic landscapes of target tissues. Our findings collectively highlight that THSDCs induce toxicity through coordinated disruptions across multiple biological layers, with effects on energy metabolism, redox homeostasis, cytoskeletal organization, and proteostasis.

For instance, several findings point toward an integrated disruption of protein homeostasis and synaptic machinery (Table 5, Fig. 6, Supplementary Tables 7 and 8):

• In proteomics, both PTU and PCN induced expression of proteasome components (e.g., PA28, 26S subunits), indicating enhanced protein degradation, possibly compensating for misfolded or oxidatively damaged proteins.

• In spatial transcriptomics, PTU-exposed neurons and astrocytes in the cerebellum showed marked deregulation of protein localization, maturation, and degradation pathways, as well as stress-response signaling (ATM, mTOR), consistent with a maladaptive unfolded protein response.

• Additionally, upregulation of vesicle trafficking proteins such as Syntaxin-1B and ARF4 in PTU proteomics suggests altered synaptic vesicle dynamics, likely impairing neurotransmission efficiency.

These multi-level disruptions in protein handling mechanisms may synergize to impair synapse formation and plasticity. Such findings align with studies showing synaptic gene vulnerability to TH disturbance and oxidative stress (Berghuis et al. 2015; Zoeller et al. 2002).

In addition, at the metabolomic level (Table 5), both PTU and PCN exposure significantly decreased levels of multiple acyl-carnitines (e.g., palmitoylcarnitine, tetradecanoylcarnitine) in the fetal brain, thyroid gland, and liver, as well as downregulation of sphingosine. Indeed, a decrease in mitochondrial activity leads to reduced fatty acid production, which in turn limits sphingosine synthesis, as fatty acids are required for its formation. These observations were echoed at the proteomic level (Fig. 6), where key mitochondrial regulators such as NME2 (a nucleoside diphosphate kinase involved in ATP homeostasis) were upregulated. This lipid metabolism dysfunction signature is reflected in the spatial transcriptomic data (Supplementary Table 7), where oligodendrocytes in PTU-treated animals showed strong deregulation of cholesterol and lipid metabolism pathways, likely impairing myelin biosynthesis (Mebarek et al. 2019; Van Meer et al. 2008). Since myelination requires high metabolic input and lipid turnover, these multi-layered alterations suggest that PTU and PCN can compromise white matter integrity. These findings further implicate glial cells in mediating TH-dependent neurodevelopment. Notably, oligodendrocyte alterations underscore the vulnerability of myelination processes to endocrine disruption, potentially leading to long-lasting connectivity deficits (Barres and Raff 1999; Zoeller et al. 2002). Previous studies have already linked acyl-carnitine depletion with impaired neuronal maturation and cognitive deficits in hypothyroid models (Bernal 2015; Mori et al. 2022).

Finally, it is known that structural alterations to the developing brain are a hallmark of TH deficiency, our multi-omics results point to cytoskeletal instability as a key convergent target:

• Proteomics revealed significant downregulation of structural proteins such as SPTBN2, Myosin-9, and Myosin-11 under PTU exposure, suggesting impaired neuronal architecture, axon guidance, and neurite extension.

• Transcriptomic analyses of astrocytes in the PTU group showed deregulation of the Rho/Ras GTPase pathways, critical regulators of actin cytoskeleton remodeling, and tubulin post-translational modification pathways, affecting glial morphology and migration.

• PCN-exposed cerebellum showed transcriptomic enrichment in extracellular matrix remodeling and cell adhesion pathways, supporting an alteration of the physical scaffold required for synaptogenesis and migration.

Thus, despite acting through different molecular mediators, both compounds impair cellular architecture (Willoughby et al. 2013; Lavado-Autric et al. 2003). This vulnerability of the cytoskeleton and the surrounding scaffolding environment may, at least in part, underlie the impairments in neuronal migration commonly observed in the context of TH disruption.

This study demonstrates that severe TH suppression caused by PTU results in significant neurodevelopmental impairments, whereas mild TH alterations associated with PCN-mediated hepatic enzyme induction do not lead to overt developmental toxicity. However, despite the distinct profiles of PTU and PCN, overlapping omics signatures were observed in the developing brain, particularly at GD21. This convergence was less evident at PND21, where PTU exposure exerts more pronounced effects, consistent with its persistent suppression of circulating and brain TH levels. These observations reinforce the importance of investigating the association between systemic and cerebral TH depletion and brain molecular responses.

Our integrated multi-omics framework provided deep mechanistic insights by revealing multiscale biomarkers with potential applications in predictive toxicology and regulatory screening. Robust statistical modeling, combined with histologically guided spatial transcriptomics, enhances confidence in the biological relevance of observed molecular alterations. This work is consistent with the comprehensive omics integration addressing thyroid disruption achieved by Canzler et al. (Canzler et al. 2025). In contrast, the present work specifically aims to focus on neurodevelopmental impairments arising from TH disruption. The application of this framework revealed both TH-dependent and TH-independent modes of action, as illustrated by the absence of brain TH level variations in PCN-exposed pups despite molecular changes.

While the study illustrates the power of an integrated, multiparametric toxicological approach, some limitations should be acknowledged. The spatial transcriptomic analyses of key neuroanatomical structures, such as the periventricular heterotopia and external granular layer, highlighted region- and cell-type-specific transcriptional reprogramming, particularly under PTU exposure. However, the analysis was limited to selected brain areas and lacked single-cell resolution, potentially masking subtle cell-type-specific effects. Future integration with single-cell RNA sequencing will be critical to refine these findings.

Also, dose levels were selected to be sufficiently high to elicit clear thyroid hormone–related adverse outcomes and robust molecular signatures, thereby enabling mechanistic investigations, rather than to directly reflect human exposure levels. However, future phases should include dose–response studies with doses selected to reflect human exposure, in order to determine thresholds of adversity and to assess the emergence of omics biomarkers. Additionally, the limited placental transfer of PCN may have restricted neonatal exposure, complicating interpretation of its apparent lack of neurodevelopmental toxicity. It remains to be clarified whether this reflects its mode of action as a hepatic enzyme inducer, or simply insufficient TH suppression in critical periods of brain development.

To address these uncertainties, future studies should employ graded exposure levels of several enzyme inducers and systematically incorporate toxicokinetic analyses to refine internal exposure assessment. Moreover, this study aimed to identify candidate biomarkers and associated signatures through omics analyses in target tissues. Extending these analyses to more accessible matrices, such as blood, may facilitate cross-species comparisons with biomonitoring data and strengthen their translational relevance.

In summary, our results support the growing consensus that mild TH disruptions (i.e., < 40% reduction) are unlikely to cause structural brain abnormalities, as opposed to severe depletion models (O'Shaughnessy et al., 2018; Marty et al. 2022). However, the implications of subtle molecular alterations warrant further investigation. This study also reaffirms the inadequacy of serum TH levels alone as predictors of neurodevelopmental outcomes, as emphasized by O’Shaughnessy et al. (2018). Hence, brain-focused assessments, including the combination of molecular and morphometric endpoints, are essential particularly for compounds inducing subtle endocrine disruption.

The integration of omics investigations such as spatial transcriptomics, proteomics, and metabolomics can enhance mechanistic understanding of THSDC-induced effects by enabling the identification of early biomarkers, thresholds of adversity, and potential compensatory responses. These advances can significantly refine risk assessment frameworks for endocrine-disrupting chemicals.

Supplementary Information

Below is the link to the electronic supplementary material.

Data availability

All data are available in the supplementary data.