Lithium exposure causes ferroptosis to induce unexplained miscarriage through non-canonical MBOAT1 pathway

Abstract

High incidence of unexplained miscarriage (UM) limits global human reproduction. Lithium (Li) batteries are widely used but rarely recycled, causing Li to spread into environments and enter human body. Notably, the healthy effects of Li exposure on UM are rarely explored. Herein, higher serum Li levels, villous tissue ferroptosis, and UM are associated. Exposure to high levels of Li causes placental ferroptosis to induce mouse miscarriage. Li exposure down-regulates ER levels to suppress ER-mediated METTL1 and MBOAT1 transcription, and also suppresses METTL1-mediated m7G modification on MBOAT1 mRNA to reduce MBOAT1 mRNA stability. Therefore, Li exposure causes ferroptosis in a GPX4-independent manner. Supplement with murine Mboat1 suppresses placental ferroptosis and relieves miscarriage in LiCl-exposed mouse model. This study discovers health risks and pathogenesis of Li exposure-induced UM, reveals potential biological targets, and also proposes an important Li exposure theme in energy-environment-health field, which deserves attention by global governments and enterprises.

Environmental Li pollution and health problems have been becoming an important concern worldwide. Here, the authors show that Li exposure causes trophoblast cell ferroptosis and thus induces unexplained miscarriage through a non-canonical MBOAT1 pathway.

Introduction

Approximately 15–25% pregnant women end with miscarriage, and 1–5% suffer from recurrent miscarriage in the world¹. Among them, 41.2% recurrent miscarriage patients have long-term anxiety, 8.6% have severe depression, and 1.4% even threaten life safety². The causes of miscarriage are very complex, including genetic factors, chromosomal abnormalities, infections, prothrombotic states, abnormal uterine anatomy, endocrine, metabolic, and reproductive immune diseases. However, there are still 50% miscarriage with unknown causes, which are generally called as unexplained miscarriage (UM)^3,4. As the key component in the placenta, trophoblast cells play vital roles in embryo implantation and healthy pregnancy^5,6 Trophoblast cell dysfunctions may induce miscarriage⁷. Notably, what might cause trophoblast cell dysfunctions and induce miscarriage are still largely elusive.

With the high-speed development of the economy, various types of batteries have been widely used in industries. Globally, lithium (Li) production reaches 100,000 tons in 2021, 256% more than in 2010, 74% of which are used for rechargeable Li batteries that power electronic products and electric vehicles⁸. However, at present, only 5% Li batteries are recovered, and up to 95% waste Li batteries are discarded into environments⁹. Moreover, Li is also present in global coals (average 12 mg/kg) and is released into the atmosphere through combustion¹⁰. Therefore, Li in energy batteries or coals is spread into various environments, including soil, atmosphere, and water. It has been detected that the levels of Li contamination in soils are 17.11–38.50 mg/kg in Beijing, China¹¹, 6.4–15 mg/kg in Europe¹², and 7–200 mg/kg in Romania^13,14. Meanwhile, approxiamately 5.5 × 10¹⁰g/year Li is released into the atmosphere during global coal combustion¹⁵. In aquatic environment, Li levels are detected as high as 6 mg/L in rivers of Chile, 180 μg/L in seawater¹⁶, and 0–150 μg/L in groundwater of the Baltic Sea, North Atlantic Ocean^17,18. The concentration of Li in drinking water is 2.8–219 μg/L in the USA, 0.03–14.83 μg/L in China, and 0.11–5.20 μg/L in Macedonia^17,18. In some cases, extreme higher levels of Li are found in surface waters in Argentina (520–620 mg/L)¹⁹, Bolivia (700–900 mg/L)²⁰, Chile (1400–1500 mg/L)²¹, and China (560–1300 mg/L)²². The widespread Li is inevitably ingested by the human body through air, food, and drinking water. It has been reported that the daily Li intake is 0.348–1.560 mg/day through food in different countries [China (1.560 ± 0.980 mg/day), USA (0.821 ± 0.684 mg/day), and Vienna (0.348 ± 0.290 mg/day)⁸; and the U.S. Environmental Protection Agency estimates the daily Li intake as 0.65–3.1 mg/day through foods and drinking water²³ under normal conditions. Due to environment-food transportation, human might ingest excessive Li. It has been reported that the levels of Li are 1.9–145 μg/L in human blood and 105–4600 μg/L in human urine in the population in northern Argentinean Ande with highly varying Li levels (5–1600 μg/L) in drinking water under real environmental conditions²⁴. Moreover, Li has been approved as a mood-stabilizing drug for treatment against manic disorders by the United States Food and Drug Administration (FDA) since 1970²⁵. Bipolar disorder patients might take 600–1200 mg/day Li₂CO₃ as medicine²⁶, giving 0.6–1.2 mM Li in their body fluids^8,26,27. It has been estimated that 12% women in pre-pregnancy and 2% women in pregnancy might take Li₂CO₃ as a mood stabilizer in HongKong, China²⁸. Increasing evidence has shown that Li exposure is associated with adverse reproduction outcomes. A prospective mother-child cohort in Argentinean Andes (N = 194) shows that Li levels in maternal blood and urine are inversely associated with fetal measurements²⁹. Meta-analysis of data from six international cohorts shows that Li exposure during the first trimester is associated with major malformations (OR 1.71, 95% CI: 1.07–2.72)³⁰. A Swedish population-based cohort study indicates that the use of Li during pregnancy is associated with a 2–3-fold increased risk of spontaneous preterm birth and cardiac malformations³¹. Mouse model assays confirmed that exposure to 375 mg/kg Li₂CO₃ on GD7.5 suppresses embryonic neurodevelopment in offspring³². A meta-analysis also indicates that Li exposure during the first trimester is associated with miscarriage (OR 3.77, 95% CI: 1.15–12.39)³³. Another prospective observational study in Israel (N = 183) reveals that there are more miscarriages (OR 1.94, 95% CI: 1.08–.48) in Li-exposed group compared with control group³⁴. Nonetheless, the association and causality between Li exposure and unexplained miscarriage are still largely unclear and should be fully investigated.

Ferroptosis is an iron-dependent and programmed cell death mode, which is different from the traditional apoptosis, cell necrosis, or pyroptosis³⁵. Ferroptosis is caused by the decreased levels of anti-lipid peroxidation or the increased levels of lipid peroxidation induced by excessive intracellular free Fe²⁺ levels^36,37. Both lipid peroxides and free Fe²⁺ (acted as a catalyst for Fenton reaction) are two major contributors to induce ferroptosis. Ferritin is the major iron storage protein, composed of ferritin heavy polypeptide 1 (FTH1) and ferritin light polypeptide 1. The degradation of FTH1 releases free Fe²⁺ to induce ferroptosis³⁸. Nuclear Receptor Coactivator 4 (NCOA4) functions as a selective autophagy adaptor that promotes ferritin autophagy degradation and induces cell ferroptosis³⁹. Knockdown of NCOA4 can reduce the levels of intracellular free Fe²⁺ and suppress ferroptosis^40,41. Ferroptosis plays important roles in the development of many human diseases, such as brain diseases (such as Alzheimer’s disease, Parkinson’s disease)⁴², heart (ischemia-reperfusion injury)⁴³, kidney degeneration⁴⁴, acute lung injury⁴⁵, and other pathological cell death^46,47, as well as reproductive diseases, such as male infertility, female endometriosis, preeclampsia, and miscarriage⁴⁸. In general, GPX4 (glutathione peroxidase 4) is a typical member in GPx (total glutathione peroxidases) family and acts as an antioxidant protein with glutathione peroxidase activity to suppress phospholipid oxidation and ROS (reactive oxygen species) production to suppress ferroptosis^49–52; and knockdown of GPX4 always induces cellular ferroptosis⁵³. Recently, it has been reported a novel ferroptosis suppressor MBOAT1 (membrane bound O-acyltransferase domain containing 1). MBOAT1 could selectively decrease cellular levels of PE-PUFA (polyunsaturated fatty acid), which is the preferred substrate for peroxidation to induce ferroptosis, and thus suppresses ferroptosis in a GPX4-independent manner⁵⁴. Therefore, whether Li exposure might trigger trophoblast cell ferroptosis to induce miscarriage, as well as through GPX4 or MBOAT1, is completely unknown and should be extensively explored.

Till now, as far as we known, the direct crosstalk and underlying mechanisms among Li exposure, ferroptosis, and unexplained miscarriage are still completely uncovered. For this aim, we collected serum and villous tissue samples from patients with unexplained miscarriage and their matched healthy control group, constructed a LiCl-exposed pregnant mouse model, and used LiCl-exposed human trophoblast Swan 71 cells. We find that high Li exposure, ferroptosis, and unexplained miscarriage are closely associated and causally regulated, which is dependent on non-canonical MBOAT1 but not classical GPX4. Moreover, Li exposure suppresses ER-mediated METTL1 (methyltransferase 1) or MBOAT1 transcription and suppresses METTL1-mediated m7G modification on MBOAT1 mRNA, both of which reduce MBOAT1 expression levels, and thus triggers trophoblast cell ferroptosis to induce miscarriage. Taken together, this study reveals biological mechanism and pathogenesis of Li exposure-induced unexplained miscarriage, giving an important example about the crosstalk among energy, environment, and healthy.

Results

Li internal exposure was associated with unexplained miscarriage

To investigate the correlation between Li internal exposure levels and unexplained miscarriage, we newly collected serum samples from unexplained miscarriage (UM) patients and their matched healthy control (HC) women (n = 80) (Fig. 1a). The baseline characteristic, clinical information, and lifestyle did not show significant differences between these two groups (Supplementary Table S1). Li levels in serum samples were detected by ICP-MS. The levels of Li were 4.70 ± 2.68 μg/L (mean ± SD) in the HC women group and 13.04 ± 4.61 μg/L (mean ± SD) in the UM patient group (Fig. 1b, readout and spectra graphs in source data). Li levels were significantly higher in UM vs HC group. In an unadjusted model, univariate logistic regression analysis showed that higher levels of Li in serum (OR = 1.930, 95% CI, 1.555–2.395) were associated with unexplained miscarriage (Fig. 1c, Supplementary Table S7). Based on directed acyclic graph analysis and previous literature^55–58, variables such as age, education, residence, smoking, and drinking were considered as potential confounders (Supplementary Fig. 1a). Multivariate logistic regression analysis by adjusting for all these confounders still showed that higher levels of Li in serum (OR = 2.210, 95% CI, 1.614–3.026) were associated with unexplained miscarriage (Fig. 1c, Supplementary Table S7). To further evaluate whether the estimated association differed among sub-populations, stratification analysis showed that the stratifying factors did not significantly alter the association between Li levels and unexplained miscarriage (all P for interaction >0.05, Fig. 1d). Therefore, these results showed that high serum Li levels were closely and positively associated with unexplained miscarriage.

Fig. 1. Higher levels of Li exposure induced miscarriage.

a Schematic diagram for HC and UM case control studies, created in BioRender (Huang, W. (2025) https://BioRender.com/7z5b66r). b The levels of Li (μg/L) in HC and UM women serum samples (n = 80). Data presented as mean values ± SD. c Univariate and multivariate logistic regression analysis of the association of Li levels in serum with miscarriage (n = 80). Data are presented as odds ratios (ORs) with 95% confidence intervals (CIs). d The estimated associations between Li levels and miscarriage by stratification analysis (all P for interaction >0.05) (n = 80). Data are presented as odds ratios (ORs) with 95% confidence intervals (CIs). (e) Schematic diagram of LiCl-exposed mouse model, created in BioRender (Huang, W. https://BioRender.com/g4az4gw). Pregnant mice were treated with 0, 3.6, 20, 100, or 200 mg/kg/d LiCl by oral gavage for continuous 13 days (each n = 6). f ICP-MS analysis of Li levels in blood samples of LiCl-exposed pregnant mice. Data presented as mean values ± SD. g The increase in body weight of Li-exposed pregnant mice with gestational days. h The increase in body weight of Li-exposed pregnant mice on D14 - the initial weight on D1 (each n = 6). Data presented as mean values ± SD. Representative images of mouse embryo resorption (indicated by red arrows) in LiCl-exposed mouse groups (i) and the average miscarriage rates of LiCl-exposed mice (j) (each n = 6). Data presented as mean values ± SD. Statistical significance was analyzed by logistic regression analysis (c, d), one-way analysis of variance (ANOVA) (f, h, j) or Student’ s t-test, Two-tailed (b). n.s., not significant. Source data are provided as a Source data file.

Li exposure induced mouse miscarriage

Subsequently, we explored the causality between Li exposure and miscarriage. To this end, we constructed a LiCl-exposed pregnant mouse model by orally giving 0, 3.6, 20, 100, or 200 mg/kg/d LiCl (corresponding to 0, 1, 5.6, 28, and 56 -fold REED of Li) for continuous 13 days (Fig. 1e). Li levels in mouse blood were increased from 6 to 177 μg/L with increasing Li exposure doses (1–56 -fold REED of Li) (Fig. 1f). Body weights were increased with gestational days during pregnancy and exposure to ≥20 mg/kg/d LiCl (5.6-fold REED of Li) reduced the increase in body weight (Fig. 1g, h), and exposure to ≥100 mg/kg/d LiCl (28-fold REED of Li) increased embryo resorption and elevated miscarriage rates (Fig. 1i, j), indicating that exposure to higher levels of Li induced mouse miscarriage.

Sequencing data indicated that Li exposure significantly regulated ferroptosis

In order to explore which kind of phenotype might be involved in Li exposure-induced miscarriage, we made mRNA sequencing to select cell phenotype. Firstly, three pairs of random HC and UM villous tissues were used for high-throughput mRNA sequencing, giving 52 up-regulated mRNAs and 725 down-regulated mRNAs with difference >1.28-fold and p < 0.05 (Supplementary Fig. 2a, raw data in GSA-Human: HRA014470). GO and KEGG analysis of these differentially expressed mRNAs (DEGs) revealed that ferroptosis-related GO items (Response to lipid, Cellular response to metal ion, and Positive regulation of metabolic process) and KEGG pathways (Amino sugar and nucleotide sugar metabolism, and PI3K-Akt signaling pathway) were significantly enriched (Supplementary Fig. 2b, c). Secondly, 100 vs 0 mg/kg/d LiCl-exposed mouse placental tissues were also used for mRNA sequencing, giving 3401 up-regulated mRNAs and 2112 down-regulated mRNAs with difference >1.28-fold and p < 0.05 (Supplementary Fig. 2d, raw data in GSA: CRA032758). GO and KEGG analysis of these DEGs revealed that ferroptosis-related GO items (Cellular lipid metabolic process, Metal ion binding, Cell death, and Regulation of ferrous iron binding) and KEGG pathways (Glycine, serine, and threonine metabolism, and HIF-1 signaling pathway) were also significantly enriched (Supplementary Fig. 2e, f). Thirdly, human trophoblast cells in villous tissues play crucial roles in woman healthy pregnancy; and Swan 71 cells have been widely used as a trophoblast cell model in various miscarriage studies⁵⁹. In this study, we constructed a LiCl-exposed human trophoblast Swan 71 cell model (Supplementary Fig. 2g). Swan 71 cells were treated with 0 or 10 mM LiCl, and the intracellular Li levels were significantly elevated after Li exposure (Supplementary Fig. 2h). Then, LiCl-exposed Swan 71 cells, together with unexposed control cells, were also used for mRNA sequencing, giving 2848 up-regulated and 1718 down-regulated mRNAs with a difference >1.28-fold and p-value < 0.05 (Supplementary Fig. 2i, raw data in GSA-Human: HRA014470). GO and KEGG analysis of these DEGs also revealed that Li exposure altered ferroptosis-related GO items (Metal ion binding, Regulation of cell death, Intracellular lipid transport, and Ferritin complex) and KEGG pathways (PI3K-Akt signaling pathway, Alanine, aspartate and glutamate metabolism, and Ferroptosis) (Supplementary Fig. 2j, k). Collectively, all three sequencing data showed that ferroptosis was top regulated, indicating that ferroptosis might be involved in Li exposure-induced miscarriage.

Li exposure caused trophoblast cell ferroptosis

To experimentally validate this, in trophoblast cells, we found that Li exposure suppressed trophoblast cell viability (Supplementary Fig. 2l). To further assure whether ferroptosis was the primary cell death mode with Li exposure, we used various inhibitors in LiCl-exposed Swan 71 cells, including ferroptosis inhibitor Fer-1 (Ferrostatin-1), apoptosis inhibitor Z-VAD-fmk, pyroptosis inhibitor Ac-YVAD-cmk, and necroptosis inhibitors necrostatin-1 (Nec-1). Fer-1 could efficiently eliminate alkoxyl radicals or other rearranged products produced from lipid hydroperoxides in ferroptosis⁶⁰, which has been widely used as a ferroptosis inhibitor in various cellular and animal studies^61–63, as well as in our recent study⁵³. We found that co-treatment with only Fer-1, but not others, could obviously rescue the cell viability of LiCl-exposed Swan 71 cells (Fig. 2a), indicating that Li exposure caused trophoblast cell ferroptosis.

Fig. 2. Li exposure caused trophoblast cell ferroptosis.

a CCK8 assay analysis of cell viability of 10 mM LiCl-exposed human trophoblast Swan 71 cells with treatment with Fer-1, Z-VAD-fmk, Ac-YVAD-cmk, or Nec-1 for 0, 12, 24, 36, or 48 h (n = 3). Data presented as mean values ± SD. b–d Western blot analysis of FTH1 and NCOA4 protein levels in 0, 5, 10, or 20 mM LiCl-exposed Swan 71 cells and their relative quantification (n  =  3). Data presented as mean values ± SD. The relative levels of Fe²⁺ (e, f), SOD (g), ROS (h, i), MDA (j), and GPx activity (k) in 0, 5, 10, or 20 mM LiCl-exposed Swan 71 cells (n = 6). Data presented as mean values ± SD. The relative levels of Fe²⁺ (l–m), MDA (n), and GPx activity (o) in 10 mM LiCl-exposed Swan 71 cells with Fer-1 treatment (n = 6). Data presented as mean values ± SD. Statistical significance was analyzed by one-way analysis of variance (ANOVA) (a, c, d, f, g, i, j, k, m, n, o). n.s., not significant. Source data are provided as a Source data file.

Subsequently, we further explored how Li exposure caused ferroptosis. Li exposure reduced the levels of FTH1 protein and increased the levels of NCOA4 protein (Fig. 2b–d), and thus up-regulated the levels of intracellular free Fe²⁺ (Fig. 2e, f). In Fe²⁺ assays, treatment with FerroOrange showed Fe²⁺ red signal; whereas non-treatment did not show Fe²⁺ signal (Supplementary Fig. 2m). Therefore, Li exposure altered iron metabolism and increased Fe²⁺ levels, which further participated in the synthesis of lipid peroxidation lipoxygenases and Fenton reaction to produce hydroxyl radicals, thereby promoting ferroptosis^64,65. Meanwhile, Li exposure also reduced the levels of SOD (superoxide dismutase) and increased the levels of ROS (reactive oxygen species) (Fig. 2g–i), and thus up-regulated the levels of MDA (malondialdehyde) and reduced the levels of GPx (total glutathione peroxidases) activity (Fig. 2j, k), and eventually induced ferroptosis. GPx could degrade the toxic by-products of lipid peroxidation and suppress ferroptosis⁶⁶ Moreover, co-treatment of LiCl-exposed Swan 71 cells with Fer-1 could well rescue the levels of Fe²⁺, MDA, and GPx activity (Fig. 2l–o). Collectively, all these data confirmed that Li exposure caused trophoblast cell ferroptosis.

Li exposure, ferroptosis, and women unexplained miscarriage were positively associated

In HC and UM women villous tissues, we also detected the levels of several key indicators of ferroptosis (the levels of Fe²⁺, MDA, and GPx activity). Fe²⁺ and MDA levels were higher and GPx activity levels were lower in UM vs HC women villous tissues (Fig. 3a–c), indicating that the ferroptosis levels were higher in UM vs HC women villous tissues. Furthermore, multivariate logistic regression analysis by adjusting for all these confounders showed that the levels of Fe²⁺, MDA, and GPx activity were associated with miscarriage (Fig. 3d, Supplementary Table S7). Therefore, the levels of ferroptosis were higher in UM vs HC women villous tissues, and higher levels of ferroptosis in villous tissues were positively associated with unexplained miscarriage.

Fig. 3. Li exposure induced miscarriage by causing ferroptosis.

a–c The levels of Fe²⁺, MDA, and GPx activity in HC and UM women villous tissues (n = 80). Data presented as mean values ± SD. d Multivariate logistic regression analysis of the association of Fe²⁺, MDA, and GPx activity levels in villous tissues with miscarriage (n = 80). Data are presented as odds ratios (ORs) with 95% confidence intervals (CIs). e Pearson correlation analysis of the levels of Fe²⁺ in HC and UM villous tissues with serum Li levels (each n = 12). f The RCS curves of the levels of Fe²⁺ in villous tissues with serum Li levels in both HC and UM groups (total n = 24). RCS models were constructed based on the plot RCS and the knots were calculated using the 5^th, 50^th, or 95^th of their levels, with the median values as the reference. g–i The relative levels of Fe²⁺, MDA, and GPx activity in placental tissues of LiCl-exposed mouse groups (each n = 6). Data presented as mean values ± SD. j Schematic diagram of LiCl-exposed mouse model with Fer-1 treatment, created in BioRender (Huang, W. (2025) https://BioRender.com/br2lv8w). Pregnant mice were treated with saline or 100 mg/kg/d LiCl by oral gavage for continuous 13 days. Meanwhile, 100 mg/kg/d LiCl-exposed mice were also treated with Ferrostatin-1 (5 mg/kg/3 day) once per 3 days by intraperitoneal injection. Embryo resorption (indicated by red arrows, k) and the average miscarriage rates (l) in 100 mg/kg/d LiCl-exposed mice with Fer-1 treatment (each n = 6). Data presented as mean values ± SD. The relative levels of Fe²⁺, MDA, and GPx activity in placental tissues of 100 mg/kg/d LiCl-exposed mice with Fer-1 treatment (each n = 6). Data presented as mean values ± SD. Statistical significance was analyzed by logistic regression analysis (d), linear regression analysis (e, f), one-way analysis of variance (ANOVA) (g, h, i, l, m, n, o) or Student’ s t-test, Two-tailed (a–c). n.s., not significant. Source data are provided as a Source Data file.

Finally, the correlation between Li levels in serum and ferroptosis levels (as indicated by the levels of Fe²⁺, MDA, and GPx activity) in villous tissues were analyzed. Firstly, Pearson correlation analysis showed that Li levels were positive correlated with Fe²⁺ and MDA levels but negatively correlated with GPx activity levels in UM women villous tissues (Fig. 3e, Supplementary Fig. 3a, b). The datapoints in UM and HC groups were obviously separated. Secondly, restricted cubic spline (RCS) model analysis showed that the levels of serum Li levels were linearly associated with the levels of Fe²⁺ (P for non-linear = 0.274) and non-linearly associated with the levels of MDA and GPx activity (P for non-linear <0.05) in all HC and UM villous tissues (Fig. 3f, Supplementary Fig. 3c, d). These data indicated that Li internal exposure levels in serum were positively associated with ferroptosis in villous tissues. Taken together, these results confirmed that Li internal exposure, ferroptosis in villous tissues, and unexplained miscarriage were positively associated with each other.

Li exposure induced mouse miscarriage by causing placental ferroptosis

Furthermore, we explored the causality among Li exposure, ferroptosis, and miscarriage in the LiCl-exposed mouse model. Analysis of mouse placental tissues showed that the levels of Fe²⁺ and MDA were higher, whereas those of GPx activity were lower, in placental tissues of mice with exposure to higher levels of LiCl (≥20 mg/kg/d LiCl, 5.6-fold REED of Li, Fig. 3g–i), indicating that higher levels of Li exposure caused mouse placental ferroptosis. To further explore whether Li exposure induced miscarriage by causing placental ferroptosis, we constructed a miscarriage intervention model by intraperitoneally injecting Fer-1 in LiCl-exposed mice (Fig. 3j). Fer-1 treatment greatly reduced embryo resorption and miscarriage rates in Li-exposed mice (Fig. 3k, l). Meanwhile, this treatment also efficiently rescued the levels of Fe²⁺, MDA, and GPx activity in mouse placental tissues (Fig. 3m–o), indicating that the reduction in ferroptosis levels in mouse placental tissues could efficiently alleviate (i.e., reduce) mouse miscarriage. Taken together, these results showed that Li exposure induced mouse miscarriage through the elevated placental ferroptosis.

Li exposure caused trophoblast cell ferroptosis by down-regulating non-canonical MBOAT1 but not classical GPX4

Subsequently, we explore which molecule might be involved in Li exposure -caused trophoblast cell ferroptosis. GPX4 is a classical and typical protein to suppress ferroptosis. However, its mRNA and protein levels did not altered with Li exposure (Supplementary Fig. 4a, b). We constructed GPX4-overexpressed or—silenced Swan 71 cells (Supplementary Fig. 4c, d). Overexpression of GPX4 did not alter the levels of Fe²⁺, MDA, and GPx activity in LiCl-exposed trophoblast cells (column 4 vs 2 in Fig. 4a–c, Supplementary Fig. 4e). Meanwhile, knockdown of GPX4 did not either alter their levels (Supplementary Fig. 4f and column 4 vs 2 in Supplementary Fig. 4g–i). Therefore, Li exposure caused trophoblast cell ferroptosis not through the classical GPX4 pathway.

Fig. 4. Li exposure caused ferroptosis by down-regulating MBOAT1.

a–c The levels of Fe²⁺, MDA, and GPx activity in 10 mM LiCl-exposed human trophoblast Swan 71 cells with overexpression of GPX4 (n = 6). Data presented as mean values ± SD. d RT-qPCR analysis of MBOAT1 mRNA levels in LiCl-exposed human trophoblast Swan 71 cells (n = 6). Data presented as mean values ± SD. e Western blot analysis of MBOAT1 protein levels in LiCl-exposed Swan 71 cells and its relative quantification (n  =  6). Data presented as mean values ± SD. The levels of Fe²⁺, MDA, and GPx activity in 10 mM LiCl-exposed Swan 71 cells with overexpression (f–h) or knockdown (i–k) of MBOAT1 (n = 6). l Western blot analysis of FTH1 and NCOA4 protein levels in Swan 71 cells with overexpression of MBOAT1 (n = 3). Data presented as mean values ± SD. m Western blot analysis of FTH1 and NCOA4 protein levels in Swan 71 cells with knockdown of MBOAT1 (n = 3). Data presented as mean values ± SD. The levels of SOD (n) and ROS (o) in Swan 71 cells with overexpression of MBOAT1 (n = 3). Data presented as mean values ± SD. The levels of SOD (p) and ROS (q) in Swan 71 cells with knockdown of MBOAT1 (n = 3). Data presented as mean values ± SD. Statistical significance was analyzed by one-way analysis of variance (ANOVA) (a–k) or Student’ s t-test, Two-tailed (n–q). n.s., not significant. Source data are provided as a Source data file.

To explore the molecule that was involved in Li exposure-caused ferroptosis, in the Venn diagram of three mRNA sequencing datasets (Supplementary Fig. 5a), we identified a top-down-regulated gene (MBOAT1) in the intersection. Recently, it has been reported that MBOAT1 is a novel protein to suppress ferroptosis in a GPX4-independent manner⁵². To test whether MBOAT1 might be involved in Li exposure-caused trophoblast cell ferroptosis, firstly, we found that Li exposure reduced MBOAT1 mRNA and protein levels (Fig. 4d, e). Subsequently, we constructed MBOAT1-overexpressed or -silenced trophoblast cells (Supplementary Fig. 5b, c). Co-overexpression of MBOAT1 restored the levels of intracellular Fe²⁺, MDA, and GPx activity in LiCl-exposed trophoblast cells (Fig. 4f–h, Supplementary Fig. 5d), showing that MBOAT1 overexpression reduced Li exposure-caused Swan 71 cell ferroptosis. In contrast, knockdown of MBOAT1 further exacerbated Li exposure-caused trophoblast cell ferroptosis, as indicated by further increasing Fe²⁺ and MDA levels and further decreasing GPx activity levels in LiCl-exposed trophoblast cells (Fig. 4i–k, Supplementary Fig. 5e). To further explore the reasons, we found that overexpression of MBOAT1 increased FTH1 protein levels and reduced NCOA4 protein levels (Fig. 4l, Supplementary Fig. 5f, g), resulting in the release of free Fe²⁺. In contrast, knockdown of MBOAT1 decreased FTH1 protein levels and increased NCOA4 protein levels (Fig. 4m, Supplementary Fig. 5h, i). Meanwhile, overexpression of MBOAT1 also increased the levels of SOD and reduced the levels of ROS (Fig. 4n, o, Supplementary Fig. 5j), resulting in lower MDA levels and higher GPx activity levels. In contrast, knockdown of MBOAT1 decreased the levels of SOD and increased the levels of ROS (Fig. 4p, q, Supplementary Fig. 5k). However, alteration of MBOAT1 did not significantly alter the mRNA and protein levels of GPX4 (Supplementary Fig. 5l–o). Collectively, these results confirmed that Li exposure caused trophoblast cell ferroptosis by down-regulating MBOAT1 in a GPX4-indeendent manner.

Li exposure suppressed ER-mediated MBOAT1 transcription

Since Li exposure down-regulated the mRNA levels of MBOAT1, subsequently, we explored its transcription. ER has been reported to be a transcription factor of MBOAT1⁵². Experimentally, overexpression of ER up-regulated, whereas knockdown of ER down-regulated, the mRNA and protein levels of MBOAT1 in Swan 71 cells (Fig. 5a, b, Supplementary Fig. 6a, b). Li exposure down-regulated the mRNA and protein levels of ER in a dose-dependent manner in Li-exposed Swan 71 cells (Fig. 5c, d). ChIP assays showed that ER bound to the promoter region of MBOAT1, and Li exposure weakened this binding (Fig. 5e). Dual-luciferase reporter assays demonstrated that ER showed transcription activity in wild-type (wt), but not mutant (mt), promoter region of MBOAT1, and Li exposure further suppressed ER transcription activity (Fig. 5f). In Li-exposed Swan 71 cells, co-overexpression of ER restored (i.e., increased) the mRNA and protein levels of MBOAT1 that were suppressed in Li-exposed trophoblast cells (Fig. 5g, h). Collectively, these results showed that Li exposure down-regulated ER expression levels and thus suppressed ER-mediated MBOAT1 transcription in Li-exposed trophoblast cells.

Fig. 5. Li exposure suppressed ER-mediated MBOAT1 transcription.

a MBOAT1 mRNA levels in Swan 71 cells with ER overexpression or knockdown (n = 3). Data presented as mean values ± SD. b MBOAT1 protein levels in Swan 71 cells with ER overexpression or knockdown and its relative quantification (n = 3). Data presented as mean values ± SD. c ER mRNA levels in LiCl-exposed human trophoblast Swan 71 cells (n = 6). Data presented as mean values ± SD. d ER protein levels in LiCl-exposed Swan 71 cells and its relative quantification (n = 6). Data presented as mean values ± SD. e ER ChIP assay analysis of MBOAT1 promoter region enriched by ER in 10 mM LiCl-exposed Swan 71 cells (n = 6). Data presented as mean values ± SD. f Dual-luciferase reporter assay analysis of the transcription activity of ER using wild-type (WT) or mutant (Mut) promoter sequence of MBOAT1 in 10 mM LiCl-exposed Swan 71 cells (n = 3). Data presented as mean values ± SD. g MBOAT1 mRNA levels in 10 mM LiCl-exposed Swan 71 cells with ER overexpression (n = 3). Data presented as mean values ± SD. h The protein levels of MBOAT1 in 10 mM LiCl-exposed Swan 71 cells with ER overexpression and its relative quantification (n = 3). Data presented as mean values ± SD. i, j The levels of the remained ER protein in 10 mM LiCl-exposed Swan 71 cells and with CHX (10 μM) treatment for 0–8 h and its relative quantification (n = 3). Data presented as mean values ± SD. k The root mean square deviation (RMSD) analysis of the protein structural stability of ER (UniProt ID: P03372) exposed to 10 mM LiCl (with NaCl as control) by molecular dynamics (MD) simulations using GROMACS within 100 ns. l The binding energy of LiCl-exposed or NaCl-exposed (as control group) ER-MBOAT1 promoter region was analyzed by gmx_MMPBSA v1.6.1 (single trajectory method). Data presented as mean values ± SD. Statistical significance was analyzed by one-way analysis of variance (ANOVA) (a–h) or Student’ s t-test, Two-tailed (a, b, e, f, j). Source data are provided as a Source data file.

We further explored how Li exposure reduced the expression levels of ER. In addition to that Li exposure down-regulated ER mRNA levels, protein stability assays also showed that Li exposure reduced ER protein stability (i.e., promoted ER protein degradation) (Fig. 5i, j). Therefore, we explored whether Li exposure might affect ER protein structure stability. The 3D and 2D structure of ER (UniProt ID: P03372) was constructed by AlphaFold3 (Supplementary Fig. 6c, d). The Ramachandran plot showed the strong overall quality of ER protein 3D structure (Supplementary Fig. 6e). Then, molecular dynamic (MD) simulations of LiCl-exposed ER (with NaCl-exposed ER as control) showed that the values of root mean square deviation (RMSD) of Li-exposed ER were higher and had more fluctuations (especially in 20–60 ns) than the control protein (Fig. 5k), showing that Li exposure might significantly reduced the structural stability of ER protein. In details, LiCl-exposed ER reduced the number of amino acid residues in α-helices (255 vs 263) and that in β-sheets (8 vs 16) and increased that in random coils (332 vs 316) relative to NaCl-exposed ER as control (Supplementary Fig. 6f, g), indicating that Li exposure might disrupt the ordered architecture of ER protein. Collectively, these results demonstrated that Li exposure might reduce the structural stability of ER protein, which might explain that Li exposure reduced ER protein stability and promoted its degradation.

Subsequently, we further investigated the effects of Li exposure on the binding between ER protein and MBOAT1 promoter region. The complex structures of DNA binding region (residues 185–250) of ER and MBOAT1 promoter region (5′-CTTAAAGGTCTCCCTGACCA-3′) were constructed by AlphaFold3 and optimized by MD simulations. The ER-MBOAT1 promoter complex contained 10 H-bonds, 4 salt bridges, and 1 π-cation interaction (Supplementary Fig. 6h), supporting that ER could bind with MBOAT1 promoter region. Then, we compared the structural difference between LiCl-exposed ER-MBOAT1 promoter region and NaCl-exposed ER-MBOAT1 promoter region as control group. Compared with control group, Li-exposed ER-MBOAT1 promoter region had less H-bonds (10 vs 14), less salt bridges (5 vs 6), and more hydrophobic interaction (1 vs 0) (Supplementary Fig. 6i, j), suggesting that Li exposure reduced the binding stability of ER with MBOAT1 promoter region. The average binding energy of Li-exposed ER-MBOAT1 promoter region was −86.68 kcal/mol, higher than −107.52 kcal/mol for the control group (Fig. 5l and Supplementary Table S8), indicating that Li exposure weakened the binding between ER and MBOAT1 promoter region, which might explain that Li exposure suppressed ER-mediated MBOAT1 transcription.

Li exposure reduced MBOAT1 mRNA stability at post-transcription level by down-regulating METTL1-mediated m7G modification on MBOAT1 mRNA

Subsequently, we also explored whether Li exposure might alter MBOAT1 mRNA stability at post-transcription level. Recently, m7G modification, mediated by methyltransferase METTL1, is discovered to regulate mRNA stability⁶⁷. Herein, we found that Li exposure down-regulated the mRNA and protein levels of METTL1 in a dose-dependent manner (Fig. 6a, b). Meanwhile, Li exposure also reduced the m7G modification levels on MBOAT1 mRNA, as well as MBOAT1 mRNA stability (Fig. 6c, d). Furthermore, overexpression of METTL1 up-regulated, whereas knockdown of METTL1 (Supplementary Fig. 7a, b) down-regulated, the levels of m7G modification on MBOAT1 mRNA (Supplementary Fig. 7c, d), MBOAT1 mRNA stability (Supplementary Fig. 7e, f), and MBOAT1 mRNA and protein levels (Supplementary Fig. 7g, h). In Li-exposed trophoblast cells, co-overexpression of METTL1 restored (i.e., increased) the levels of m7G modification on MBOAT1 mRNA, MBOAT1 mRNA stability, and MBOAT1 mRNA and protein levels (Fig. 6e–h). Collectively, these results confirmed that Li exposure down-regulated METTL1 expression levels and thus reduced m7G modification levels on MBOAT1 mRNA, which further reduced MBOAT1 mRNA stability and down-regulated MBOAT1 mRNA levels in Li-exposed trophoblast cells.

Fig. 6. Li exposure suppressed METTL1-mediated m7G modification on MBOAT1 mRNA and reduced MBOAT1 mRNA stability.

a METTL1 mRNA levels in LiCl-exposed human trophoblast Swan 71 cells (n = 6). Data presented as mean values ± SD. b METTL1 protein levels in LiCl-exposed Swan 71 cells and its relative quantification (n = 6). Data presented as mean values ± SD. c MeRIP analysis of the m7G modification levels on MBOAT1 mRNA in 10 mM LiCl-exposed Swan 71 cells (n = 6). Data presented as mean values ± SD. d The levels of the remained MBOAT1 mRNA in 10 mM LiCl-exposed Swan 71 cells with 5 μg/mL actinomycin D treatment for 0–8 h (n = 3). Data presented as mean values ± SD. e MeRIP analysis of the m7G modification levels on MBOAT1 mRNA in 10 mM LiCl-exposed Swan 71 cells with METTL1 overexpression (n = 3). Data presented as mean values ± SD. f The levels of the remained MBOAT1 mRNA in 10 mM LiCl-exposed Swan 71 cells with METTL1 overexpression (n = 3). Data presented as mean values ± SD. g MBOAT1 mRNA levels in 10 mM LiCl-exposed Swan 71 cells with METTL1 overexpression (n = 3). Data presented as mean values ± SD. h MBOAT1 protein levels in 10 mM LiCl-exposed Swan 71 cells with METTL1 overexpression and its relative quantification (n = 3). Data presented as mean values ± SD. i MBOAT1 protein levels in Swan 71 cells with METTL1 overexpression or knockdown and its relative quantification (n = 3). Data presented as mean values ± SD. j ER ChIP assay analysis of METTL1 promoter region enriched by ER in 10 mM LiCl-exposed Swan 71 cells (n = 6). Data presented as mean values ± SD. k Dual-luciferase reporter assay analysis of ER transcription activity using wild-type (WT) or mutant (Mut) promoter sequence of METTL1 in 10 mM LiCl-exposed Swan 71 cells (n = 3). Data presented as mean values ± SD. l METTL1 mRNA levels in 10 mM LiCl-exposed Swan 71 cells with ER overexpression (n = 3). Data presented as mean values ± SD. m METTL1 protein levels in 10 mM LiCl-exposed Swan 71 cells with ER overexpression and its relative quantification (n = 3). Data presented as mean values ± SD. n The binding energy of LiCl-exposed or NaCl-exposed (as control group) ER-METTL1 promoter region was analyzed by gmx_MMPBSA v1.6.1 (single trajectory method). Data presented as mean values ± SD. Statistical significance was analyzed by one-way analysis of variance (ANOVA) (a, b, e, g, h, i, l, m) or Student’s t-test, Two-tailed (c, d, f, i–k). Source data are provided as a Source data file.

Li exposure also suppressed ER-mediated METTL1 transcription

To correlate ER and METTL1, predicted by PROMO, ER might also be a transcription factor of METTL1. Experimentally, overexpression of ER up-regulated, whereas knockdown of ER down-regulated, the mRNA and protein levels of METTL1 (Fig. 6i, Supplementary Fig. 7i). ChIP assays showed that ER bound to the promoter region of METTL1, and Li exposure weakened this binding (Fig. 6j). Dual-luciferase reporter assays showed that ER showed transcription activity in wild-type (wt) but not in mutant (mut) promoter region of METTL1, and Li exposure suppressed its transcription activity (Fig. 6k). In Li-exposed Swan 71 cells, overexpression of ER restored (i.e., increased) the mRNA and protein levels of METTL1 that were suppressed in Li-exposed Swan 71 cells (Fig. 6l, m). Collectively, these results showed that Li exposure suppressed ER-mediated METTL1 transcription in Li-exposed trophoblast cells.

Subsequently, we further investigated the effects of Li exposure on the binding between ER protein and METTL1 promoter region. Firstly, the complex structures of DNA binding region (residues 185–250) of ER and METTL1 promoter region (5′-ATGTTGCAATGACCC-3′) were constructed by AlphaFold3 and optimized by MD simulations. The complex contained 8 H-bonds, 3 salt bridges, and 1 π-cation interaction in ER-METTL1 promoter complex (Supplementary Fig. 7j), supporting that ER could bind with METTL1 promoter region. Then, we compared the structural difference between LiCl-exposed ER-METTL1 promoter region and NaCl-treated ER-MBOAT1 promoter region as control group. Compared with control group, Li-exposed ER-METTL1 promoter region had same H-bonds (8 vs 8), less salt bridges (3 vs 5), same π-cation interaction (1 vs 1), less hydrophobic interaction (1 vs 2), and more π-π stacking interactions (2 vs 0) (Supplementary Fig. 7k, l). The average binding energy of LiCl-exposed ER-METTL1 promoter region was −81.93 kcal/mol, higher than −97.36 kcal/mol for NaCl-expsoed control group (Fig. 6n and Supplementary Table S9), indicating that Li exposure weakened the binding between ER and METTL1 promoter region, which might explain that Li exposure suppressed ER-mediated METTL1 transcription.

Li exposure caused ferroptosis by suppressing ER/METTL1/MBOAT1 axis

Subsequently, we explored whether Li exposure caused ferroptosis through ER/METTL1/MBOAT1 axis. Overexpression of ER, METTL1, or MBOAT1 decreased the levels of intracellular Fe²⁺ and MDA and increased the levels of GPx activity; whereas knockdown of any of them increased the levels of Fe²⁺ and MDA and decreased the levels of GPx activity (Fig. 7a–c, Supplementary Fig. 8a–c), indicating that this ER/METTL1/MBOAT1 axis suppressed trophoblast cell ferroptosis. In Li-exposed trophoblast cells, overexpression of ER, METTL1, or MBOAT1 rescued (i.e., reduced) Li exposure-caused trophoblast cell ferroptosis, as indicated by recovering the levels of intracellular Fe²⁺, MDA, and GPx activity (Fig. 7d–f, Supplementary Fig. 8 d). Collectively, these results demonstrated that Li exposure caused trophoblast cell ferroptosis through suppressing ER/METTL1/MBOAT1 axis.

Fig. 7. Li exposure caused ferroptosis by suppressing ER/METTL1/MBOAT1 axis.

a–c The levels of Fe²⁺, MDA, and GPx activity in human trophoblast Swan 71 cells with overexpression or knockdown of ER, METTL1, or MBOAT1 (n = 3). Data presented as mean values ± SD. d–f The levels of Fe²⁺, MDA, and GPx activity in 10 mM LiCl-exposed Swan 71 cells with overexpression of ER, METTL1, or MBOAT1 (n = 3). Data presented as mean values ± SD. Data presented as mean values ± SD. Statistical significance was analyzed by one-way analysis of variance (ANOVA) (a–f) or Student’ s t-test, Two-tailed (a–c). Source data are provided as a Source data file.

ER/METTL1/MBOAT1 axis was suppressed in UM vs HC women villous tissues

To validate the cellular mechanisms in women villous tissues, we detected the expression levels of ER, METTL1, and MBOAT1, as well as the levels of m7G modification on MBOAT1 mRNA, in both UM and HC women villous tissues. The protein and mRNA levels of ER, METTL1, and MBOAT1 were lower in UM vs HC group (Fig. 8a–c). ChIP assays showed that the levels of the promoter region of METTL1 or MBOAT1 enriched by ER were lower in UM vs HC group (Fig. 8d, e). Pearson correlation analysis showed that the protein levels of ER were positively correlated with those of METTL1 or MBOAT1 in UM villous tissues (Fig. 8f, g), indicating that ER-mediated transcription of METTL1 or MBOAT1 was attenuated in UM vs HC women villous tissues. Meanwhile, the levels of m7G modification on MBOAT1 mRNA were also lower in UM vs HC group (Fig. 8h). Pearson correlation analysis showed that the protein levels of ER or METTL1 were positively correlated with the levels of m7G modification on MBOAT1 mRNA in UM villous tissues (Fig. 8i, j), indicating that METTL1-mediated m7G modification on MBOAT1 mRNA was also lower in UM vs HC villous tissues. Collectively, these results showed that the cellular mechanisms in Li-exposed human trophoblast cells were consistent with those in UM women villous tissues.

Fig. 8. ER/METTL1/MBOAT1 axis was suppressed in UM vs HC women villous tissues.

a, b The protein levels of ER, METTL1, and MBOAT1 in HC and UM women villous tissues and their relative quantification (each n = 12). Data presented as mean values ± SD. c The mRNA levels of ER, METTL1, and MBOAT1 in HC and UM women villous tissues (each n = 12). Data presented as mean values ± SD. d ER ChIP assay analysis of MBOAT1 promoter region enriched by ER in HC and UM women villous tissues (n = 6). Data presented as mean values ± SD. e ER ChIP assay analysis of METTL1 promoter region enriched by ER in HC and UM women villous tissues (n = 6). Data presented as mean values ± SD. f, g Pearson correlation analysis of the protein levels of ER with those of METTL1 or MBOAT1 in HC and UM women villous tissues (each n = 12). h MeRIP analysis of the m7G modification levels on MBOAT1 mRNA in HC and UM women villous tissues (n = 6). Data presented as mean values ± SD. i, j Pearson correlation analysis of the protein levels of ER or METTL1 with the m7G levels on MBOAT1 mRNA in HC and UM women villous tissues (n = 6). Statistical significance was analyzed by linear regression analysis (f, g, i, j), or Student’ s t-test, Two-tailed (b–e, h). Source data are provided as a Source data file.

Subsequently, Pearson correlation analysis showed that serum Li levels were negatively correlated with the protein levels of ER, METTL1, or MBOAT1 in UM villous tissues (Supplementary Fig. 9a–c). Moreover, the protein levels of ER, METTL1, or MBOAT1 were negatively correlated with the levels of Fe²⁺ and MDA and positively correlated with those of GPx activity in UM villous tissues (Supplementary Fig. 9d–l). The datapoints in both UM and HC groups were obviously separated (Supplementary Fig. 9a–l). Combined with the cellular results, these data implied that Li exposure might down-regulate this ER/METTL1/MBOAT1 axis and thus cause ferroptosis in UM vs HC villous tissues, which further induced miscarriage.

Murine Er/Mettl1/Mboat1 axis was also suppressed in LiCl-exposed mouse placental tissues

Next, we further validated the cellular mechanisms in LiCl-exposed mouse miscarriage models. UCSC-BLAST analysis showed that the sequences of ER, METTL1, and MBOAT1 were conserved in human, mouse, dog, rhesus monkey, and cow (Supplementary Table S10). Firstly, the protein and mRNA levels of murine Er, Mettl1, and Mboat1 were all lower in Li-exposed mouse placental tissues (Fig. 9a–c). ChIP assays showed that the levels of the promoter region of murine Mettl1 or Mboat1 enriched by Er were lower in Li-exposed mouse placental tissues (Fig. 9d, e). The levels of m7G modification on murine Mboat1 mRNA were also lower in Li-exposed mouse placental tissues (Fig. 9f). Pearson correlation analysis showed that (1) the protein levels of Er were positively correlated with those of Mettl1 or Mboat1 (Fig. 9g, h), (2) the protein levels of Er or Mettl1 were also positively correlated with the levels of m7G modification on Mboat1 mRNA (Supplementary Fig. 10a, b), and (3) the protein levels of Er, Mettl1, or Mboat1 were negatively correlated with the levels of Fe²⁺ and MDA and positively correlated with those of GPx activity (Supplementary Fig. 10c–k) in Li-exposed mouse placental tissues. Combined with the cellular results, these data implied that Li exposure down-regulated this Er/Mettl1/Mboat1 axis and thus caused ferroptosis in mouse placental tissues, which further induced mouse miscarriage.

Fig. 9. Murine Er/Mettl1/Mboat1 axis was suppressed in LiCl-exposed mouse placental tissues.

a, b The protein levels of murine Er, Mettl1, and Mboat1 in LiCl-exposed mouse placental tissues (each n = 6) and their relative quantification. Data presented as mean values ± SD. c The mRNA levels of murine Er, Mettl1, and Mboat1 in LiCl-exposed mouse placental tissues (each n = 6). Data presented as mean values ± SD. d Er ChIP assay analysis of the levels of Mettl1 promoter region enriched by Er in LiCl-exposed mouse placental tissues (each n = 6). Data presented as mean values ± SD. e Er ChIP assay analysis of the levels of Mboat1 promoter region enriched by Er in LiCl-exposed mouse placental tissues (each n = 6). Data presented as mean values ± SD. f MeRIP analysis of the m7G modification levels on Mboat1 mRNA in LiCl-exposed mouse placental tissues (each n = 6). Data presented as mean values ± SD. g, h Pearson correlation analysis of the protein levels of Er with those of Mettl1 or Mboat1 in LiCl-exposed mouse placental tissues (each n = 6). Statistical significance was analyzed by linear regression analysis (g, h), one-way analysis of variance (ANOVA) (b, c) or Student’ s t-test, Two-tailed (d–f). Source data are provided as a Source data file.

Overexpression of Mboat1 alleviated placental ferroptosis and reduced miscarriage in LiCl-exposed mouse model

Finally, to validate the roles of Mboat1 in Li exposure-induced miscarriage and to explore the potential approach for miscarriage treatment, we constructed a miscarriage treatment model in which LiCl-exposed mice were intraperitoneally injected with Mboat1 overexpression plasmid (Supplementary Fig. 11a). Overexpression of Mboat1 reduced embryo resorption and miscarriage rates in LiCl-exposed mice (Fig. 10a, b). Overexpression of Mboat1 also restored (i.e., increased) the protein levels of Mboat1, restored (i.e., decreased) the levels of Fe²⁺ and MDA and (i.e., increased) the levels of GPx activity in placental tissues of Li-exposed mice (Fig. 10c–g). These results indicated that overexpression of Mboat1 could efficiently reduce placental ferroptosis and alleviate mouse miscarriage in Li-exposed mouse model, providing an efficient target for treatment against Li exposure-induced unexplained miscarriage.

Fig. 10. Miscarriage treatment by therapeutic up-regulation of murine Mboat1 in LiCl-exposed mouse model.

a, b Embryo resorption (indicated by red arrows) and the average miscarriage rates in 100 mg/kg/d LiCl-exposed mice with Mboat1 supplement (each n = 6). Data presented as mean values ± SD. c, d The protein levels of murine Er, Mettl1, and Mboat1 in placental tissues of 100 mg/kg/d LiCl-exposed mice with Mboat1 supplement (each n = 6) and their relative quantification. Data presented as mean values ± SD. e–g The relative levels of Fe²⁺, MDA, and GPx activity in placental tissues of 100 mg/kg/d LiCl-exposed mice with Mboat1 supplement (each n = 6). Data presented as mean values ± SD. h The proposed mechanism, created in BioRender (Huang, W. (2025) https://BioRender.com/k7glvye). Statistical significance was analyzed by one-way analysis of variance (ANOVA) (b–g). Source data are provided as a Source data file.

Discussion

With widely global application of Li batteries, environmental Li pollution and health problems have been becoming one important concern worldwide, a new prominent major problem in the energy-environment-health field. Preliminary epidemiological studies have shown that Li exposure during early pregnancy might be associated with an increased risk of adverse pregnancy outcomes, such as fetal measurements²⁹, fetal malformations (such as cardiovascular anomalies) and preterm birth^30,31. Till now, there are only few observational studies to imply that Li exposure might be associated with miscarriage^33,34. Whether and how Li exposure during pregnancy triggers unexplained miscarriage remain largely unclear. In this study, we aim to explore novel pathways and mechanisms of UM that might be induced by environmental pollutants. Therefore, we exclude miscarriage women with the known causes, including bipolar disorder patients. Li content in these women was originated from food source and environmental exposure under the inadvertent environment. Li levels in serum samples were detected as 4.70 ± 2.68 μg/L (mean ± SD) in HC group and 13.04 ± 4.61 μg/L (mean ± SD) in UM group (Fig. 1b). The levels were in the range of Li levels (7-28 μg/L Li in serum) in normal population⁶⁸. Epidemiological analysis shows that serum Li levels are significantly higher in UM vs HC women and higher levels of Li are positively associated with unexplained miscarriage. However, bipolar disorder patients might take 600–1200 mg/day Li₂CO₃ as medicine²⁶, giving 0.6–1.2 mM Li in their body fluids^8,27. Moreover, LiCl-exposed mouse model confirms that exposure to higher levels of Li induces mouse miscarriage. Next question is how Li exposure induces miscarriage. We detect the levels of several key indicators of ferroptosis in LiCl-exposed trophoblast cells, in HC and UM women villous tissues, and in LiCl-exposed mouse placental tissues, and find that Li exposure induces miscarriage by causing ferroptosis. Moreover, reduction in mouse placental ferroptosis by treating with ferroptosis inhibitor Fer-1 could efficiently relieve mouse miscarriage in Li-exposed mouse model. Other study also showed that Li treatment reduces brain Tau protein levels and attenuates iron efflux, leading to iron accumulation in mice⁶⁹. Exposure of male Balb/c mice to 10, 50, or 100 ppm LiCl in drinking water for 5 weeks results in significant toxic effects and oxidative stress in mouse testicular tissues⁷⁰. In addition, exposure to 20 mg/kg Li₂CO₃ causes renal injury in mice, as indicated by the reduced antioxidant enzyme capacity and increased ROS levels⁷¹. Therefore, we conclude that Li exposure causes ferroptosis to induce miscarriage based on the comprehensive epidemiological analysis, mouse model assays, and cellular mechanistic studies, discovering the emerging Li exposure as a new risk factor for unexplained miscarriage in the energy-environment-health field.

GPX4 is essential for maintaining lipid homeostasis in cells and preventing accumulation of phospholipid oxidation and ROS production⁷²; and inactivation of GPX4 may lead to their accumulation and ferroptosis^42,73. Recently, it has been newly reported that MBOAT1 suppresses ferroptosis by reducing the levels of cellular polyunsaturated fatty acid⁵⁴ in a GPX4-independent manner. In this study, we find that Li exposure causes trophoblast cell ferroptosis, which is not dependent on the classical GPX4 but dependent on the non-canonical MBOAT1. The regulatory mechanisms are proposed (Fig. 10h). At transcription level, Li exposure down-regulates ER expression levels, suppresses ER-mediated transcription of METTL1 and MBOAT1, and thus down-regulates their mRNA levels. At post-transcription level, METTL1 performs m7G modification on MBOAT1 mRNA and enhances its mRNA stability; and the down-regulated METTL1 reduces MBOAT1 mRNA stability. Therefore, Li exposure down-regulates MBOAT1 mRNA levels at both transcription and post-transcription levels, reducing MBOAT1 expression levels. Subsequently, the down-regulated MBOAT1 causes trophoblast cell ferroptosis and further induces miscarriage. The cellular mechanisms are consistent with those in UM patient villous tissues and in LiCl-exposed mouse placental tissues. Possibly, higher levels of Li in UM vs HC villous tissues reduce MBOAT1 mRNA levels and thus promote ferroptosis in UM vs HC villous tissues, eventually inducing miscarriage. Furthermore, supplement with Mboat1 could effectively suppress ferroptosis in LiCl-exposed trophoblast cells and inhibit placental ferroptosis to alleviate miscarriage in LiCl-exposed mouse model. It should be noted that ferroptosis (including the reduced levels of GPx activity) is not originated from GPX4 but from the suppressed MBOAT1 and the elevated oxidative stress. Therefore, this study provides approach and essential mechanisms by which Li exposure leads to ferroptosis to induce miscarriage.

M7G modification on mRNAs can regulate mRNA stability; and the abnormal M7G modification has been found to be associated with various diseases, such as tumors and Alzheimer ‘s disease^74–77. Herein, we find for the first time that m7G modification levels on MBOAT1 mRNA are significantly lower in UM vs HC women villous tissues, and its levels are negatively associated with unexplained miscarriage. Moreover, Li exposure down-regulates METTL1 expression levels and m7G modification levels on MBOAT1 mRNA, which down-regulates MBOAT1 expression levels and induces ferroptosis. In LiCl-exposed mouse model, Li exposure also down-regulates the m7G modified levels on MBOAT1 mRNA and thus reduces MBOAT1 expression levels, which results in mouse placental ferroptosis and induces miscarriage. In addition to m7G modification, we also find that m6A modification is also closely associated with unexplained miscarriage. For example, m6A modification promotes the RNA stability of lnc-HZ01, lnc-HZ09, and lnc-HZ14, and up-regulates their expression levels and ultimately inhibits trophoblast proliferation (by lnc-HZ01) and migration/invasion (by lnc-HZ09), and causes trophoblast pyroptosis (by lnc-HZ14), respectively, further inducing unexplained miscarriage^78–80 Moreover, environmental BaP (benzo(a)pyrene or BPDE (benzo(a)pyrene-7,8-dihydrodiol-9,10-epoxide) exposure also up-regulates m6A modification levels. Therefore, RNA modifications are also crucial factors for healthy pregnancy and various environmental factors might alter their modification levels. The relationship among environmental toxicants, the types and levels of RNA modifications, and unexplained miscarriage should be further and extensively investigated.

Recently, increasing studies have uncovered the molecular mechanisms of unexplained miscarriage^1,78–84. Fructose-1,6-diphosphate has been reported to prevent miscarriage by inducing COX-2 macrophage differentiation in the decidua⁸⁴. Also, increasing succinate levels by intravenous injection during the first trimester reduces the risk of recurrent miscarriage⁸³. In our recent studies, we have found that BaP/BPDE exposure leads to ferroptosis in vascular endothelial cells, inhibits their angiogenesis, and thus induces miscarriage⁵³; and supplement with murine Gpx4 could suppress ferroptosis, recover angiogenesis, and alleviate BaP-induced miscarriage in mouse mode⁵³. Hypoxia leads to trophoblast cell ferroptosis and induces miscarriage⁴⁸; and knockdown of either murine lnc-hz06 or Ncoa4 could suppress ferroptosis and alleviate miscarriage in hypoxic mouse model⁴⁸. BaP/BPDE exposure also inhibits trophoblast cell migration/invasion, suppresses homologous recombination repair, or induces pyroptosis, any of which might induce miscarriage^79,80,85. Knockdown of Nlrp3 could efficiently reduce placenta pyroptosis⁷⁹ and knockdown of murine Ahr recovers homologous recombination repair⁸⁵ in placental tissues, both of which alleviate miscarriage in BaP-exposed mice. Polystyrene nanoplastics (PS-NPs) inhibit trophoblast cell migration/invasion and migrasome formation or lead to trophoblast cell apoptosis, which induce miscarriage^86,87. Supplementing with murine Sox2 or Rock1 could rescue migration/invasion and migrasome formation⁸⁷ and supplement with Bcl-2 reduces apoptosis⁸⁶, both of which alleviate miscarriage in PS-NPs-exposed pregnant mouse model. Copper exposure leads to mouse placental cuproptosis and causes miscarriage⁸⁸; and treatment with TTM (a cuproptosis inhibitor) suppresses placental cuproptosis and alleviates miscarriage in CuCl₂-exposed mouse model⁸⁸. In addition to these toxicants, in this study, we find that exposure to higher levels of Li leads to trophoblast cell ferroptosis to induce miscarriage; and supplement with Fer-1 or murine Mboat1 could efficiently reduce placental ferroptosis and alleviate mouse miscarriage in a Li-exposed mouse model, providing targets for effective treatment against unexplained miscarriage caused by excessive Li exposure. Recently, a study shows that Li administration in lactating rat dams alters thyroid hormones and blood urea in both dams and pups, which could be reversed by iodine supplement⁸⁹, providing a potential approach for treatment against Li exposure-induced miscarriage by iodine supplement. All these studies enrich the pathogenesis and their corresponding therapeutic approach for various toxicant-induced unexplained miscarriage.

To recycle and reuse Li batteries and to reduce environmental Li exposure are important topics in energy-environment-health field, which should be addressed by global governments, enterprises, and hospitals. The transport pathway and organ enrichment of Li in human remain unclear and need to be explored urgently. In addition to the current study, Li exposure might also lead to other dysfunctions of human trophoblast cells and induce other adverse pregnancy outcomes, such as fetal development. A study has shown that rat infants exposed to Li via breast milk from rat mothers results in the reduction in blood thyroxine and elevation in blood urea nitrogen, showing adverse effects on thyroid and kidney functions in rat infants⁹⁰. Li inhibits iodine uptake by thyroid follicles, which further suppresses the iodination of tyrosine, the cleavage of thyroglobulin, and the production of thyroid hormone. Notably, thyroid functions of rat infants could be significantly improved through breast-feeding by supplementing iodine to rat mother’s diet⁹¹. Therefore, to explore clinical approaches to reduce or eliminate the adverse effects of excessive Li exposure during pregnancy is particularly important for patients and their infants. Therefore, to detect Li levels or other metal ion levels in serum might be concerned in annual routine physical examination. Moreover, the doses of Li exposure in mouse and cell assays are higher than real environmental exposure doses. Frankly, it could not exclude the possibility that women might ingest excessive Li in areas with high Li pollution, with Li occupational exposure or occupational poisoning, or with possible long-term cumulative exposure. The short-term and high-dose Li-exposed animal and cell models could reflect or predict, but might not completely consistent with, the long-term low-dose effects of Li exposure on miscarriage in human cohort, which should be further studied. Although we have find that exposure to higher levels of Li could induce unexplained miscarriage by causing ferroptosis, it cannot exclude other cell dysfunctions and the interactions of Li exposure with other environmental toxicants. In general, environmental toxicants might have their corresponding receptors. For example, polycyclic aromatic hydrocarbons⁸⁵, bisphenol analogues⁹², and dioxins^93,94 might activate their common receptor AhR. AhR is typical transcription factor and activates the transcription of a group of target genes to perform down-stream signaling and functions⁸⁵. To identify the common receptors in various toxicant-induced miscarriage is also an important topic. CCK8 assays show that co-treatment with ferroptosis inhibitor could partially recover Li exposure-suppressed cell viability (Fig. 2a), indicating that there might be other cell death modes in LiCl-exposed trophoblast cells. Finally, the treatment against unexplained miscarriage should be further explored for better clinical application.

Based on the epidemiological studies, mouse model, and cellular assays, we conclude that Li exposure causes trophoblast cell ferroptosis and thus induces unexplained miscarriage. Mechanistically, Li exposure down-regulates ER expression levels and thus suppresses ER-mediated METTL1 and MBOAT1 transcription. Li exposure further suppresses METTL1-mediated m7G modification on MBOAT1 mRNA and thus reduces MBOAT1 mRNA stability. Therefore, Li exposure down-regulates MBOAT1 expression levels at both transcription and post-transcription levels, resulting in ferroptosis and inducing miscarriage. The mechanisms are consistent with those in LiCl-exposed trophoblast cells, in LiCl-exposed mouse model, and in UM patient villous tissues, which contain higher levels of Li relative to HC tissues. Supplement with murine Mboat1 effectively suppresses placental ferroptosis and relieves miscarriage in LiCl-exposed mouse model. This study not only discovers the healthy risk effects of Li exposure and the pathogenesis of Li exposure-induced unexplained miscarriage, but also reveals potential biological targets for treatment against unexplained miscarriage, and also proposes an important Li exposure theme in the energy-environment-health field, which deserves critical attention by governments, enterprises, and hospitals.

Methods

Chemicals and reagents

LiCl (99.99% purity) was purchased from Aladdin (7447-41-8). LiCl was dissolved in ddH₂O to make 5 M LiCl stock. DMSO was from Sigma-Aldrich (D2650). Ferrostatin-1 (Fer-1), Acetyl-Tyr-Val-Ala-Asp-chloromethyl ketone (Ac-YVAD-cmk), Z-Val-Ala-Asp (OMe)-fluoromethyl ketone (Z-VAD-fmk), and necrostatin-1 (Nec-1) were purchased from AbMole (M2698, M8340, M3143 and M2315).

Villous tissues and serum samples

In this study, unexplained miscarriage (UM) patients and healthy control (HC) women who had elective miscarriage to terminate their unwanted pregnancies in 25–30 years old with 6–10 weeks gestation were recruited from the affiliated Shenzhen maternity and child healthcare hospital of Southern Medical University, as the methods described previously^78,81,82. Any women with clinically known causes of miscarriage were excluded, such as cervical incompetence, chromosome abnormalities, endocrine or metabolic diseases, virus or bacterial infections, as described previously^78,81,82. Bipolar disorder patients were also excluded because they might take 600–1200 mg/day Li₂CO₃ as medicine²⁶, giving 0.6–1.2 mM Li in their body fluids^8,26,27. The UM group did not have a previous successful pregnancy, and the HC group did not report any previous miscarriages. Finally, 80 UM and 80 HC women were enrolled in this study. Miscarriage is defined by the World Health Organization (WHO) as the loss of an embryo or fetus weight <500 g or before 20 weeks of gestation, including spontaneous loss or personally artificial miscarriage⁹⁵. HC and UM women did not received any treatment. The characteristic of these HC and UM women were listed in Supplementary Table S1, including baseline characteristic (age, residence, education, body mass index), clinical information (gestational days, RBC, WBC, Hb), and lifestyle (smoking, drinking) in the period of 3 months before miscarriage operation. All the information was obtained from medical records. A piece of villous tissue with dimensions of approximate 2 × 0.5 × 0.5 cm³ was manually dissected from the fetal side of placenta and was cleared of maternal decidua. After washing with sterile saline, villous tissue samples were immediately frozen in liquid nitrogen and stored at −80 °C before RNA or protein extraction. Peripheral blood samples were collected on the same day of miscarriage operation and stored in BD Vacutainer SST™ tubes. Serum samples were isolated within 30 min by centrifugation and were stored in aliquot at −80 °C until further use. The experiment protocols were approved by the Ethics Committee of the Affiliated Shenzhen maternity and child healthcare hospital of Southern Medical University. Written informed consents were collected from all the participants before enrollment.

Determination of Li levels in serum, blood and cell samples by inductively coupled plasma-mass spectrometry (ICP-MS)

Human and mouse blood samples were collected and centrifuged at 3500 rpm for 10 min, and serum samples were isolated and stored at −80 °C. Human serum or mouse blood samples, as well as the collected Swan 71 cells, were diluted 20-fold with ddH₂O, digested with 1% nitric acid, and then sonicated for 5 min. Metal elements in serum, blood, or cell samples were analyzed using kinetic energy discrimination (KED)-based Thermo iCAP Q inductively coupled plasma mass spectrometer (ICP-MS) with an octopole based collision/reaction cell (Thermo Fisher Technologies, USA). The limits of quantification (LOQs) for Li was 0.019 μg/L. The quality control samples and the standard reference samples were analyzed repeatedly every 80 serum samples. The spike recoveries for the internal standard metals ranged from 82 to 117%. Calibration standard curves showed correlation coefficients >0.999 for the instrument responses and the known levels of trace elements. For quality control, the standards and a random sample were re-measured for every batch of 20 samples with relative standard deviations (RSDs) < 10 %. The values of the samples with measurements <LOQ were replaced with LOQ/√2, as commonly used in prior studies^96–99. Li levels in serum or blood samples were expressed as μg/L and those in cell samples were expressed as ng/10⁶ cells.

Mice and experimental design

LiCl-exposed mouse model was constructed as the similar methods described previously^53,54,78. Pregnant C57BL/6 mice (Charles River Company, Beijing, China) were randomly divided into different groups (n = 6 per group). The appearance of a vaginal copulation plug was considered as day 1 (D1) of pregnancy, which was further confirmed by weighting mouse body every day¹⁰⁰. The pregnant mice were treated with LiCl everyday. To explore the effects of real environmental exposure dose (REED) of Li on miscarriage, we calculated the actual daily intake dose of Li in human body. One study showed that Li intake dose was approximately 3.674 mg/day in the residents under normal condition in Canary Islands, Spain¹⁰¹ (i.e., 0.05 mg/kg/day Li based on an average adult weight of 70 kg). This dose could be considered as a representative of human real intake dose under the inadvertent environment. Since different species might have different sensitivity to chemicals, FDA recommends body surface area conversion coefficients among different species^102,103. According to body surface area coefficient (mouse/human = 12), this REED of Li corresponds to 0.6 mg/kg/day Li element in mouse (i.e., 3.6 mg/kg/day LiCl). Clinically, bipolar disorder patients might take 600–1200 mg/day Li₂CO₃²⁶, corresponding to 100–200 mg/kg/day LiCl in mouse based on this body surface area coefficient (mouse/human = 12). Mouse model assays showed that exposure of C57BL/6 mice to 200 mg/kg LiCl for 5 days causes the inflammation in colon of mice¹⁰⁴. Considering the doses used in literature, short lifespan of mice, and the feasibility of mouse model assays, we chose 0, 3.6, 20, 100, or 200 mg/kg/day LiCl in this pregnant mouse model, corresponding to 0 (control), 1-, 5.6-, 28-, or 56-fold REED of Li (Supplementary Table S2).

Model 1, LiCl-exposed pregnant mice model was constructed: ① control group, treated with the same volume of saline solution, ② 1-fold REED of Li group, treated with 3.6 mg/kg/d LiCl, ③ 5.6-fold REED of Li group, treated with 20 mg/kg/d LiCl, ④ 28-fold REED of Li group, treated with 100 mg/kg/d LiCl, ⑤ 56-fold REED of Li group, treated with 200 mg/kg/d LiCl. Mice were daily given LiCl or equal volume of saline solution by oral gavage from D1 to D13. Model 2, we also constructed a miscarriage intervention model by treating LiCl-exposed pregnant mice with Fer-1 (ferroptosis inhibitor): ① control group, treated with the same volume of saline solution, ② treated with 100 mg/kg/d LiCl, ③ treated with 100 mg/kg/d LiCl and DMSO, ④ treated with 100 mg/kg/d LiCl and 5 mg/kg/3 d Fer-1 in corn oil. We selected 100 mg/kg/day LiCl because this dose was enough to exhibit obvious miscarriage and ferroptosis phenotypes. This dose was related with clinical treatment dose and public health and was also good for investigation of miscarriage invention and treatment. Mice were daily given LiCl or equal volume of saline solution by oral gavage from D1 to D13. Meanwhile, 5 mg/kg Fer-1 in DMSO or equal volume DMSO was intraperitoneally injected into mice once per three days from D1 to D13. Model 3, we also constructed another miscarriage intervention model by treating LiCl-exposed pregnant mice with pcDNA3.1-Mboat1 (murine Mboat1): ① control group, treated with the same volume of saline solution, ② treated with 100 mg/kg/d LiCl, ③ treated with 100 mg/kg/d LiCl and 10 mg/kg/3 d pcDNA3.1, ④ treated with 100 mg/kg/d LiCl and 10 mg/kg/3 d pcDNA3.1-Mboat1 (murine Mboat1). Mice were daily given LiCl or equal volume of saline solution by oral gavage from D1 to D13. Meanwhile, pcDNA3.1 (empty vector as control) or pcDNA3.1-Mboat1 (plasmid overexpressing murine Mboat1) in saline solution were also intraperitoneally injected into mice once per 3 days from D1 to D13. For three models, mice were euthanized by intraperitoneal injection with nembutal (100 mg/kg) on D14 for collection of uterus. The embryo resorption was identified as smaller or darker appearance relative to the viable and pink healthy embryos⁵⁵. The miscarriage rate in each mouse and the average miscarriage rate in each group were calculated by (the number of adsorbed embryos)/(the total number of normal embryos and the adsorbed embryos)⁵⁵. Placental tissues were collected and the RNAs and proteins were extracted from a random placenta in each mouse for RT-qPCR and Western blot analysis, respectively. Blood samples were also collected from mice in Model 1 mice for ICP-MS assays. The animal project has been authorized by Ethics Committee of the Eighth Affiliated Hospital of Sun Yat-sen University.

Cell culture

First-trimester human trophoblast Swan 71 cells, which were immortalized by human telomerase, were constructed by Gil Mor’s group at Yale University¹⁰⁵ and were received as gifts. Swan 71 cells were cultured in DMEM/F12 medium (GIBCO, Invitrogen) supplemented with 10% FBS (GIBCO) at 37 °C in a humidified atmosphere containing 5% CO₂. Li levels have been detected as 0.274–20.9 μM in human blood and 0.015–0.663 mM in human urine²⁴ under real environmental exposure and as 1–5 mM in body fluids of bipolar disorder patients taking Li₂CO₃ as medicine^8,27. It has been reported that human breast cancer cells (MCF-7) exposed to 50 or 100 mM LiCl induced apoptosis by up-regulating GSK-3β, caspase-2, Bax, and cleaved caspase-7 and by down-regulating anti-apoptotic proteins (Akt, β-catenin, Bcl-2, and cyclin D1)²⁷. Treatment of MCF-7 or MDA-MB-231 breast cancer cells with 20 or 30 mM LiCl increased GSK-3β protein levels and DNA damages and suppressed cell viability¹⁰⁶. Based on the doses used in literature and pre-experiments, in this study, we constructed a LiCl-exposed human trophoblast cell model by treating Swan 71 cells with 0, 5, 10, or 20 mM LiCl for 48 h.

High-throughput mRNA sequencing and data processing

Three pairs of random HC and UM villous tissues and 100 vs 0 mg/kg/d LiCl-exposed mouse placental tissues were collected and used for high-throughput mRNA sequencing. Human Swan 71 cells (5 × 10⁶ cells) treated with 10 mM LiCl, together with the equal number of untreated control cells, were also used for mRNA sequencing. High-throughput mRNA sequencing was performed on HiSeq 2000 sequencing platform (BGI-Shenzhen) according to the BGI commercial standard process (https://www.bgi.com/)^78,81,82. Briefly, total RNAs were extracted by Trizol reagent (Thermo Fisher Scientific). The process included the removal of rRNA, synthesis of double-stranded cDNA, end repair, degradation of one strand, and enrichment of the other strand by quantitative reverse transcription PCR (RT-qPCR). The library quality was confirmed by sequencing. The differentially expressed mRNAs (DEGs)with differences >1.28-fold and p < 0.05 were generated from read counts using the online bioinformatic platform Dr. Tom provided by BGI (biosys.bgi.com). The DEGs were searched in the NCBI database (Gene Bank, Homo sapiens, GRCh38.p14) to determine their genome loci. These intersected DEGs with differences >1.28-fold and p < 0.05 were used for gene ontology (GO) and KEGG analysis to generate GO and KEGG plots, respectively^107,108.

Cell transfection

Human trophoblast Swan 71 cells were transfected with overexpression plasmid or siRNA to overexpress or knockdown some certain genes, respectively. Empty vector pcDNA3.1 (Catalog No. V790-20) was purchased from Thermo Fisher Scientific Company. cDNAs that were used for construction of overexpression plasmid of ER (pcDNA3.1-ER), METTL1 (pcDNA3.1-METTL1), MBOAT1 (pcDNA3.1-MBOAT1), and GPX4 (pcDNA3.1-GPX4) were synthesized and constructed into pcDNA3.1 vector by Addgene (Supplementary Table S3). The corresponding RNA sequences were obtained from National Center for Biotechnology Information (NCBI) database (Gene Bank 2013). Empty vector pcDNA3.1 was used as negative control. Si-ER, si-METTL1, si-MBOAT1, si-GPX4, and si-NC (negative control) were customized by Thermo Fisher (sequences in Supplementary Table S4). Human trophoblast Swan 71 cells (1 × 10⁶ cells/well) were seeded in 6-well plates and cultured to 80% confluence. Trophoblast cells were transfected with 1 μg plasmids or 50 nM siRNAs in turbofect transfection reagent (R0531, Thermo Scientific) for 24 h according to the manufacturer’s protocols. The transfection efficiencies were validated by RT-qPCR.

Cell viability

Cell viability was evaluated using Cell Counting Kit-8 (CCK8, ab228554, Abcam, Cambridge, UK)¹⁰⁹. Cells (5 × 10³ cells per well, three replicates for each group) were seeded in 96-well plates, treated with 0, 5, 10, or 20 mM LiCl for 0, 12, 24, 36, or 48 h. In another assays, Swan 71 cells (5 × 10³ cells per well, three replicates for each group) were seeded in 96-well plates, were treated with 10 μM Fer-1, 50 μM pyroptosis inhibitor Acetyl-Tyr-Val-Ala-Asp-chloromethyl ketone (Ac-YVAD-cmk), 50 μM apoptosis inhibitor Z-Val-Ala-Asp (OMe)-fluoromethyl ketone (Z-VAD-fmk), or 20 μM necroptosis inhibitors necrostatin-1 (Nec-1) and co-treated with 0 or 10 mM LiCl for 0, 12, 24, 36, 48, or 72 h. Afterward, 10 μL incubation reagent and 90 μL DMEM/F12 medium were added according to the manufacturer’s protocols. The 96-well plate was fully covered with tin foil to avoid light. Subsequently, the plate was incubated at 37 °C for 1 h. The absorbance was detected at 450 nm using a microplate reader (Bio-Rad, Hercules, CA, USA) with cell culture medium as background. The proliferation of cells was expressed as the change in absorbance at 450 nm. All experiments were replicated thrice.

ROS (reactive oxygen species) measurement

The levels of intracellular ROS were measured using Cellular ROS Assays (ROS, ab113851, Abcam, Cambridge, UK)¹¹⁰. Swan 71 cells were exposed to 0, 5, 10, or 20 mM LiCl or overexpressed or silenced MBOAT1. Cells (5 × 10³ cells per well) were incubated with 10 μM 2′,7′-Dichlorodihydrofluorescein diacetate (DCFH-DA) at 37 °C for 30 min in the dark. DCFH-DA is generally used to detect the generation of reactive oxygen intermediates and to assess the overall oxidative stress in toxicological phenomenon¹¹¹. The cells were thoroughly washed thrice with PBS to remove excessive DCFH-DA. Subsequently, the resulting cells were visualized under a fluorescence microscope (Leica, Germany). DCFDA fluorescence was excited at 480 nm and detected at 525 nm for all the samples, and quantified using Image J 1.43 U (NIH Image J System, Bethesda, MD). The unexposed Swan 71 cells were used as control cells. All experiments were replicated thrice. The levels of ROS were expressed as the mean fluorescence intensity and were normalized against the levels in the control group.

Detection of SOD levels

The levels of SOD (superoxide dismutase) were determined using the Total Superoxide Dismutase Assay Kit with WST-8 (S0101, Beyotime, China)^112,113. Swan 71 cells (1 × 10⁷ cells) were exposed to 0, 5, 10, or 20 mM LiCl or overexpressed or silenced MBOAT1. Cells were washed thrice with ice-cold PBS (4 °C) and then incubated in 200 μL SOD sample preparation buffer. After complete cell lysis, 20 μL supernatant, 160 μL WST-8/enzyme working solution, and 20 μL reaction start solution were incubated in a 96-well plate for 30 min at 37 °C. The absorbance was measured at 450 nm on an enzyme labeling instrument. The unexposed Swan 71 cells were used as control cells. All experiments were replicated thrice. The SOD levels were calculated as mU/mg total proteins in each sample and were normalized against the levels in the control group.

Lipid peroxidation assay (MDA)

The levels of MDA (malondialdehyde) were determined in 0, 5, 10, or 20 mM LiCl-exposed Swan 71 cells, in 10 mM LiCl-exposed Swan 71 cells with co-treatment with 10 μM Fer-1, in 10 mM LiCl-exposed Swan 71 cells with co-overexpression or co-knockdown of GPX4, ER, METTL1, or MBOAT1, in HC and UM women villous tissues, in LiCl-exposed pregnant mouse placental tissues, and in placental tissues of LiCl-exposed pregnant mouse model with treatment with Fer-1 or Mboat1 overexpression plasmid. Lipid peroxidation was assessed by detecting the levels of MDA using Lipid Peroxidation MDA Assay Kit (ab118970, Abcam, UK) according to the manufacturer’s instructions¹¹⁴. Cells (4 × 10⁷ cells) or tissue samples (10 mg) were homogenized in 400 μL IP lysis buffer (containing 10 μL 100 mM PMSF) on ice for 30 min for cell lysis. The lysates were collected and centrifugated for 10 min at 12,000 g at 4 °C. The protein concentration was determined by Pierce BCA Protein Assay Kit (Pierce). Supernatant (300 μL) was incubated with 600 μL MDA assay detection solution at 100 °C for 15 min. After centrifugation at 1000 g for 10 min, 200 μL supernatant were loaded into a new 96-well plate (three replicates per group), and the absorbance was measured at 532 nm using a microplate reader (Bio-Rad, Hercules, CA, USA), with lysis buffer as background. Microplate manager was used to prepare the standard curve. Swan 71 cells without exposure to LiCl were used as control cells. All experiments were replicated for six times. The MDA levels were calculated as nmol/mg total proteins in each sample and were normalized against the levels in control group.

Total glutathione peroxidase activity (GPx)

The levels of total GPx (glutathione peroxidases) activity were determined in 0, 5, 10, or 20 mM LiCl-exposed Swan 71 cells, in 10 mM LiCl-exposed Swan 71 cells with co-treatment with 10 μM Fer-1, in 10 mM LiCl-exposed Swan 71 cells with co-overexpression or co-knockdown of GPX4, ER, METTL1, or MBOAT1, in HC and UM women villous tissues, in LiCl-exposed pregnant mouse placental tissues, and in placental tissues of LiCl-exposed pregnant mouse model with treatment with Fer-1 or Mboat1 overexpression plasmid. Total activity of GPx was detected using Total Glutathione Peroxidase Assay Kit (ab102530, Abcam, Cambridge, UK)^115,116. Briefly, cells (3 × 10⁷ cells per well) or tissue samples (10 mg) were homogenized in 400 μL IP lysis buffer (containing 10 μL 100 mM PMSF) on ice for 30 min for cell lysis. Lysates were collected and centrifugated for 10 min at 12,000 g at 4 °C. Then, the lysate supernatant, GPx detection solution (2 μL 62.5 mM NAPDH, 2 μL 75 mM GSH, and 1 μL glutathione reductase), and 35 μL glutathione peroxidase detection buffer were mixed and incubated in a 96-well plate at 25 °C for 15 min (three replicates per group). Next, 10 μL 30 mM peroxide reagent solution were added into these wells. Total GPx activity was determined by recording the absorbance at 340 nm on a microplate reader (Bio-Rad, Hercules, CA, USA) with assay buffer as background. Swan 71 cells without exposure to LiCl were used as control cells. All experiments were replicated for six times. Total GPx activity was expressed as mU/mg total proteins in each sample and was normalized against the levels in control group.

Free Fe²⁺ levels in human trophoblast Swan 71 cells

Human trophoblast Swan 71 cells were treated with 0, 5, 10, or 20 mM LiCl, with 0 or 10 mM LiCl and 10 μM Fer-1, 1 μg plasmids (pcDNA3.1-GPX4, pcDNA3.1-MBOAT1, pcDNA3.1-ER and pcDNA3.1-METTL1), or 50 nM siRNA (si-GPX4, si-MBOAT1, si-ER and si-METTL1). Free Fe²⁺ levels were detected using FerroOrange (F347, Dojindo, Japan)^117,118. Cells (5 × 10³ cells per well) were incubated with 1 μM FerroOrange (final concentration) at 37 °C for 30 min. Then, fluorescence was measured at 580 nm (excited at 543 nm) on a fluorescence microscope (Leica, Germany). Three digital images per well were recorded. Fluorescence signals on digital photographs of cells were analyzed using ImageJ. Swan 71 cells without exposure to LiCl were used as control cells. The levels of free Fe²⁺ were expressed as the mean fluorescence intensity and were normalized against the levels in control group. Swan 71 cells without FerroOrange treatment were considered as baseline assays to exclude the potential autofluorescence. All experiments were replicated thrice.

Free Fe²⁺ levels in tissue samples

Free Fe²⁺ levels in HC and UM women villous tissues, in LiCl-exposed pregnant mouse placental tissues, and in placental tissues of LiCl-exposed pregnant mouse model with treatment with Fer-1 or Mboat1 overexpression plasmid were determined using an Iron Assay Kit (Abcam, ab83366), as the similar methods described previously^53,59,119. Tissue samples (10 mg) were homogenized in 100 μL iron assay buffer containing iron reducer in an ice bath. After centrifugation at 4 °C at 16,000 g for 10 min, the supernatant (60 μL) was collected, added into 96-well plates, and replenished to 100 μL with iron assay buffer. Five μL assay buffer was added for Fe²⁺ iron assays, and 5 μL iron reducer was added to total iron assays. These samples were incubated at 37 °C for 30 min. Then, 100 μL Iron Probe was added per well, and the samples were incubated at 37 °C for 60 min to determine the absorbance at 593 nm with a microplate reader (Bio-Rad, Hercules, CA, USA). The iron assay buffer was used as background. HC women villous tissues or mouse placental tissues without LiCl exposure were used as control group. A standard linear curve of free Fe²⁺ levels vs absorbance was derived from the standard solution provided in this kit (R² > 0.9). The levels of Fe²⁺ in each of tissue sample were calculated based on this standard curve. Free Fe²⁺ levels were expressed as nmol/mg tissues and were normalized against the levels in control group. Each sample was replicated independently thrice.

Quantitative reverse transcription PCR (RT-qPCR)

Total RNAs were extracted from Swan 71 cells, human villous tissues, and mouse placental tissues using Trizol (Invitrogen, Carlsbad, USA)^120,121. RNA quality and quantity were assessed using a NanoDrop 2000 UV spectrophotometer (Thermo Fischer Scientific, Waltham, USA), and RNA integrity number (RIN) were assessed on an Agilent 2100 Bioanalyzer (Agilent Technologies, Palo Alto, CA, USA). The concentration of RNA was in 500–1500 ng/μL and the RIN ≥ 9. The RNA purity was high as the A260/280 values were in 1.8–2.2. The isolated RNAs (800 ng) were converted into cDNAs using the first-strand cDNA synthesis kit (Invitrogen). Then, cDNAs were amplified using a 20-μL SYBR Green Supermix (Takara, Kyoto, Japan). The RT-qPCR program was described as follows: pre-denaturation at 95 °C for 30 s, cycling reaction at 95 °C for 10 s for 40 cycles, dissolution at 95 °C for 15 s, 60 °C for 60 s, and then 95 °C for 15 s. The sequences of the specific primers were shown in Supplementary Table S5. The mRNA levels of GAPDH were not changed in our experiments and were used as normalization internal standard for all mRNA detection, as shown in literature and our recent studies^79,80. All experiments were replicated for three or six times. The levels of mRNAs were expressed as 2^-ΔΔCt, where Ct was cycle threshold, ΔCt = testing gene (Ct) - average GAPDH (Ct), and ΔΔCt = sample group Δ(Ct)—average control group Δ(Ct).

RNA stability analysis

Human trophoblast Swan 71 cells (1 × 10⁶ cells/well) were treated with 0 or 10 mM LiCl, with 50 nM si-METTL1 vs NC, with 1 μg pcDNA3.1-METTL1 plasmids vs vector, or with 10 mM LiCl + 1 μg pcDNA3.1-METTL1 plasmids or 1 μg pcDNA3.1-vector plasmids. These cells were seeded in a 6-well plate for 24 h. Then, the cells were further treated with 5 μg/mL actinomycin D (Sigma-Aldrich) to block mRNA transcription. After 0, 2, 4, 6, 8, or 10 h, RNAs were extracted from cells, and the levels of MBOAT1 mRNA were analyzed by RT-qPCR assays. GAPDH mRNA was used as the normalization internal standard.

Western blot analysis

Total proteins were extracted from human trophoblast Swan 71 cells, human villous tissues, and mouse placental tissues using RIPA lysis buffer (Thermo Fisher Scientific) and quantified using Pierce BCA Protein Assay Kit (Pierce). Proteins (10–30 μg/well, equal amounts within group but different amounts among groups for better comparison) were separated on 10% or 12% SDS-PAGE gel and transferred on an equilibrated polyvinylidene difluoride membrane (PVDF, Amersham Biosciences, Buckinghamshire, UK). After blocking with 5% bovine serum albumin (BSA, Sigma-Aldrich) in TBST (10 mM Tris-HCl, 150 mM NaCl, and 0.1% Tween 20) for 1 h at room temperature, the membrane was incubated with primary antibody overnight at 4 °C. The primary antibodies contained anti-GPX4 (ET1706-45, HUABIO, dilution 1:1000), anti-MBOAT1 (DF15819, Affinity, dilution 1:1000), anti-METTL1 (11525-MM05, Sino Biological, dilution 1:1000), anti-ER (R380695, Zen BioScience, dilution 1:1000), anti-FTH1 (381204, Zen BioScience, dilution 1:1000), anti-NCOA4 (680442, Zen BioScience, dilution 1:1000), and anti-GAPDH (5174, CST, dilution 1:1000). The PVDF membrane was washed thrice with TBST and incubated with secondary antibody for 1 h at room temperature. The secondary antibodies included goat anti-rabbit IgG (ab205718, Abcam, dilution 1:10000) and goat anti-mouse IgG (ab6789, Abcam, dilution 1:10000). Afterward, PVDF membrane was washed thrice, and proteins were detected by enhanced chemiluminescence (Amersham Corporation, Arlington Heights, IL, USA). The intensity of each band was quantified by Image J. All experiments were replicated for three or six times. The value of each band density in experimental and control groups was normalized to that of its corresponding GAPDH band (loading control, ratio to GAPDH%).

Protein stability assays

ER protein stability was measured in Swan 71 cells treated with cycloheximide (CHX), an inhibitor of protein synthesis¹²². Swan 71 cells were treated with 10 μM CHX (Abcam, ab120093) within 10 h to block protein synthesis. Detection of remaining protein levels in cells could well affect the degradation of this protein. For LiCl-exposed cells with co-treatment with CHX, Swan 71 cells were firstly treated with 10 mM LiCl for 48 h and then treated with 10 μM CHX for another 0, 2, 4, 6, or 8 h. After treatment, cells were harvested, and the proteins in cell lysate supernatant were analyzed by western blotting. The protein band intensity was quantiffed by ImageJ. These experiments were independently performed thrice.

ChIP (chromatin immunoprecipitation) assay

ChIP assays were performed using EZ-Magna ChIP™ Chromatin Immunoprecipitation Kit (Millipore), as described previously^78,81. Briefly, trophoblast cells (2 × 10⁷ cells) were digested using trypsin; and human villous tissues (0.2 g) and mouse placental tissues (0.2 g) were cutting into very small pieces by using tissue scissors. These tissue samples and cells were cross-linked in PBS containing 1% formaldehyde at room temperature for 15 min, and then quenched with 125 mM glycine for 5 min. Then, these samples were lysed in cell lysis/wash buffer (150 mM NaCl, 5 mM EDTA [pH = 7.5 at 25 °C], 50 mM Tris-HCl [pH 7.5], 0.5% NP-40) containing protease inhibitor for 10 min on ice and then sonicated to generate DNA fragmentation with length of 200–500 bp in shearing buffer (1% SDS, 10 mM EDTA [pH = 8.0], 50 mM Tris-HCl [pH = 8.0 at 25 °C]) containing protease inhibitor using a Diagenode BioruptorPlus sonicator (30 s on and 30 s off for 12 cycles). Then, DNA fragments were incubated with ER antibody (13258S, CST, dilution 1:1000) overnight at 4 °C, with equal weight of IgG antibody (ab172730, Abcam, dilution 1:200) as negative control. Protein A/G magnetic beads were then added and incubated at 4 °C for 4 h to form bead-protein-DNA complex. Subsequently, beads were washed with cell lysis/wash buffer at 4 °C for six times and then washed twice with cold TE buffer (Invitrogen, 12090015). The cross-linking was terminated by treating cells with elution buffer (100 mM NaHCO₃ and 1% SDS) on a shaker at room temperature for 15 min. Subsequently, the cross-link of antibody-bound chromatin complexes were dissociated at 65 °C with 100 mM NaCl overnight to release free DNA fragments. Each sample was treated with 50 ng/μL RNase A at 37 °C for 30 min and then with 10 mM Proteinase K at 45 °C for 1 h. Then, the immunoprecipitated DNA fragments were extracted with Min-Elute PCR purification kit (Qiagen, 28004). The promoter regions of MBOAT1, METTL1, murine Mboat1, and murine Mettl1 were amplified by RT-qPCR with specific primers (sequences in Supplementary Table S5).

Luciferase reporter assay

Luciferase reporter assays were performed as described previously^82,123. Briefly, wild-type (wt, 5′-GAGTGTGGATGACCCTTGGT-3′) or mutant (mut, 5′-CTCACACCTACTGGGAACCA-3′) sequence in MBOAT1 promoter region or wt (5′-GGTAAAGAGGCGCGGGGA-3′) or mut (5′-CCATTTCTCCGCGCCCCT-3′) sequence in METTL1 promoter region was fused into luciferase pGL3-basic reporter vector (Promega, Madison, USA) to construct pmirGLO-MBOAT1-wt/-mut or pmirGLO-METTL1-wt/-mut (sequences in Supplementary Table S6). Human trophoblast Swan 71 cells were seeded into 24-well plates and were co-transfected with 100 ng pmirGLO-MBOAT1-wt/-mut or pmirGLO-METTL1-wt/-mut and exposed to 10 mM LiCl for 48 h according to the manufacturer’s instructions. Cells were lyzed using passive lysis buffer (Promega Corporation) and the firefly luciferase activity in each well was measured using Dual-Luciferase Reporter Assay System (Promega) according to the manufacturer’s protocols.

Structural and energy analysis of Li-exposed ER and Li-exposed ER bound to MBOAT1 or METTL1 promoter region

Selection of ER region and MBOAT1 or METTL1 promoter region: Human protein structure ESR1_HUMAN (UniProt ID: P03372) was selected from the UniProt database. The DNA binding domain of ER contained amino acid residues from 185 to 250. JASPAR database showed the highest-scoring DNA fragment to bind with ER: MBOAT1 promoter region 5′-CTTAAAGGTCTCCCTGACCA-3′ and METTL1 promoter region 5′-ATGTTGCAATGACCC-3′.

Molecular dynamics (MD) simulations: The structure of ER protein was constructed by Alphafold3. The 3D structure was visualized by PyMOL v2.5.4 an d its 2D structures was analyzed using PDBsum. The structures of ER bound to MBOAT1 or METTL1 promoter region in the presence of 10 mM LiCl (with 10 mM NaCl as control group) were analyzed by MD simulations (within 100 ns) using GROMACS 2020.6. Protein parameters and topology files were generated using Amber03 force field. In simulation box, periodic boundary conditions were applied and ER protein or ER-DNA complexes centered in a cubic box (minimum 1.0 nm edge distance) with TIP3P water as solvent. Charge neutrality was maintained by replacing solvent with 10 mM NaCl or LiCl. Energy minimization was conducted via the steepest descent algorithm to eliminate steric clashes and optimize solvent orientation.

The molecular dynamics pre-equilibration was conducted in two consecutive phases. Firstly, NVT ensemble equilibration (constant particle number, volume, and temperature) involved a 100 ps simulation at 300 K, with the goal of allowing the system to adapt to the set temperature and ensuring a uniform temperature distribution. Secondly, NPT ensemble equilibration (constant particle number, pressure, and temperature) followed with a 100 ps simulation at 1 bar, aiming to enable the system to adapt to the specified pressure, thereby stabilizing the density of the system.

Production MD simulations were executed under isothermal-isobaric conditions (300 K, 1 bar) within 100 ns using the leapfrog algorithm. For post-simulation, trajectories were aligned to the protein backbone, and root mean square deviation (RMSD) was analyzed.

Ramachandran Plot Assessment: The Ramachandran plot (generated using the PyMod module in PyMOL v2.5.4) evaluated the quality of the 3D structural model of ER protein. This model exhibited strong overall quality with 75.5% residues in the most favored regions (red), 15.9% in additional allowed regions (yellow), 2.5% in generously allowed regions (light yellow), and 6.1% in disallowed regions (white).

ER Protein 2D structure annotation: The secondary structures of ER or LiCl (or NaCl)-exposed ER were analyzed using PDBsum. Sequences were represented by single-letter amino acid codes. Secondary structural elements included α-helices (labeled H1, H2, etc., depicted as purple cylinders), β-sheets (strands labeled A, B, etc., shown as purple arrows), and conformational motifs (β-turns, γ-turns, and β-hairpins).

3D structural analysis and visualization: The 3D structures of ER or ER-DNA complex were constructed by AlphaFold3; and the detailed 3D structures and binding interface analysis were visualized by PyMOL v2.5.4. The 3D structure of ER protein contained α-helices in red, β-sheets in yellow, and random coils in green. The 3D structure of ER-DNA complex contained stable interactions, such as H-bonds (formed between polar donors (e.g., N-H) and acceptors (e.g., O, N), energy range in 10–40 kJ/mol), salt bridges (formed between oppositely charged residues, distance ≤5.5 Å, energy range in 3–13 kJ/mol), hydrophobic interactions (aggregation of hydrophobic groups), π–π Stacking (weak interactions of aromatic rings between electron-rich and electron-deficient systems), π-cation interactions (interactions between π-systems (e.g., aromatic rings) and cations (e.g., positively charged residues)).

Binding Energy Calculation: The binding energies (kcal/mol) of ER-MBOAT1 or METTL1 promoter region were analyzed by gmx_MMPBSA v1.6.1 (single trajectory method).

MeRIP (Methylated RNA Immunoprecipitation) assay

N7-methyladenosine (m7A RNA methylation) modification on MBOAT1 and Mboat1 mRNA was determined by MeRIP-qPCR assays using BersinBio™ RIP Immunoprecipitation (RIP) Kit (BersinBio) according to the manufacturer’s instructions. RNAs were extracted from trophoblast cells (2 × 10⁷), human villous tissues (0.2 g), or mouse placental tissues (0.2 g) in lysis buffer containing protease inhibitor and RNAase inhibitor. Then, the RNA samples were treated with DNaseI and sonicated to generate RNA fragments in 200–500 nt. The fragmented RNAs were incubated with m7G-methyladenosine antibody (4141–13, MBL, dilution 1:200), with IgG as negative control, overnight at 4 °C in IP binding buffer; and then the mixtures were incubated with Protein A/G magnetic beads (HY-K0202, MedChemExpress) for another 4 h at 4 °C. Afterward, the beads were washed thrice with IP washing buffer. The m7G-modified RNAs were eluted and extracted from beads for RT-qPCR analysis with the specific primers (sequences in Supplementary Table S5). One tenth of total RNAs was used as Input.

Statistical analysis

All experiments were replicated thrice independently with similar results. The measured or calculated data (including control group) were presented as mean ± SD. SPSS26.0 or R software was used to analyze the data, and GraphPad Prism 8.0 as well as R software was used for figure generation. In cellular assays, the number n = 3 or 6 indicated that all experiments were replicated for three or six times independently. Data in other groups were normalized against that in control group. In animal model, the number n = 6 indicated 6 mice in each group. In villous tissue assays, n = 50 indicated 50 pair of HC and UM women tissue or serum samples. Levene’s test was used to evaluate the homogeneity of variance of the data. Student’s t-test was used to compare differences between two groups. ANOVA and least significant difference (LSD) test were used to compare differences among multiple groups. The qualitative data were compared using the Chi-square test. Eighty pairs HC and UM women were randomly selected for detection of Li levels in serum samples and the levels of Fe²⁺, MDA, and GPx in villous tissues. Then, these parameters were used for linear regression analysis (n = 12). The restricted cubic spline (RCS) curve was plotted by R software. When p < 0.05, the difference was considered to be statistically significant.

Reporting summary

Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.

Author contributions

Y.S., W.H., and H.Z.: Designed the study. Y.S., and W.H.: Performed most of the experiments. Y.S., W.H., and H.Z.: Wrote the draft manuscript. Q.F., and X.Y.: Molecular docking. Y.L., X.C., and J.X.: Important help and advice. P.L., S.X., and Y.C.: Helped for statistical analysis. Q.L. and M. W.: Contributed to the tissue preparation. G.G. and D.Z.: Co-design topic and guidance. H.Z.: Conceptualization, methodology, funding acquisition, investigation, supervision.

Peer review

Peer review information

Nature Communications thanks Condon Lau, Ahmed Irfan and the other, anonymous, reviewer(s) for their contribution to the peer review of this work. A peer review file is available.

Data availability

The raw sequence data have been deposited in the Genome Sequence Archive¹²⁴ in National Genomics Data Center¹²⁵, China National Center for Bioinformation/Beijing Institute of Genomics, Chinese Academy of Sciences (GSA: CRA032758 (https://ngdc.cncb.ac.cn/gsa/browse/CRA032758); GSA-Human: HRA014470 (https://ngdc.cncb.ac.cn/gsa-human/browse/HRA014470)) that are publicly accessible at https://ngdc.cncb.ac.cn/gsa. All data and materials presented in this manuscript are available from the corresponding author (H. Zhang) upon a request under a completed Material Transfer Agreement. Any additional information required to reanalyze the data reported in this work paper is available from the corresponding author upon request. Source data are provided with this paper.

Competing interests

The authors declare no competing interests.

Supplementary information

The online version contains supplementary material available at 10.1038/s41467-025-67026-7.