Declines in ovarian reserve associated with ambient ozone exposure: mediating role of lipid profile Authors: Yina Hu, Chenxi Zhao, Yanqing Deng, Abigail Solomon, Sydney Bell, Leizhen Xia, Lifeng Tian, Yan Zhao, Xiaoyan Ai, Haisheng Wu, Jialyu Huang Published: 2026 ABSTRACT BackgroundAmbient ozone pollution represents a growing threat to public health. However, evidence regarding its association with ovarian reserve and the mediating mechanisms remains limited and inconsistent.MethodsThis cross-sectional study included 8,233 women aged 20–45 who attended the Center for Reproductive Medicine at Jiangxi Maternal and Child Health Hospital between 2021 and 2023. Based on data from the Tracking Air Pollution in China project, ozone exposure was determined by assigning 10-km-resolution daily maximum 8-hour average ozone concentration to each participant’s residential address. Serum anti-Müllerian hormone (AMH) levels were measured to evaluate ovarian reserve, with six distinct exposure periods defined according to the stages of female follicular development. Participants’ lipid profiles were characterized by total cholesterol (TC), low-density lipoprotein cholesterol (LDL-C), high-density lipoprotein cholesterol and triglycerides. The ozone-AMH relationship was assessed using multivariable linear regression models, supplemented by stratified, mediation, and sensitivity analyses.ResultsEach interquartile range increase in ozone concentration was linked to an 8.14% (95% confidence intervals [CI]: 3.09% to 12.93%) decrease in AMH levels during period 4. Compared with women aged 20–29 years, this ozone-related decline was less pronounced among women over 35 (P for interaction = 0.004). Among the four lipid indicators, TC mediated 11.415% of the association between ozone and AMH levels, while LDL-C accounted for 7.38%. Similar findings were observed for period 6, while no significant linear associations were identified across the other periods.ConclusionHigher ambient ozone exposure is significantly linked to diminished ovarian reserve. This association is particularly evident during the developmental window from primary and secondary follicles to small antral follicle stage, with younger women showing greater susceptibility. To the best of current knowledge, this study provides the first evidence that alterations in lipid profiles may partially mediate this relationship, offering novel insights for protecting female reproductive health and informing environmental health interventions.Supplementary InformationThe online version contains supplementary material available at https://doi.org/10.1186/s12944-026-03003-3. Introduction As a major component of photochemical air pollution, ozone (O3) has raised serious public health concerns [1, 2]. Globally, ambient O3 exposure was responsible for approximately 365,222 deaths according to the Global Burden of Disease 2019 assessment [3]. Notably, severe surface O3 pollution has become increasingly prevalent across China [4], where the population-weighted average concentration ranks within the top 20% among 197 countries [5]. Extensive studies have demonstrated the harmful consequences of O3 exposure for respiratory and cardiovascular health, primarily driven by the systemic induction of oxidative stress and inflammation [6, 7]. In recent years, growing evidence also suggests potential associations between O3 exposure and adverse reproductive outcomes, including decreased sperm quality [8, 9], reduced fecundability [10], pregnancy complications and adverse birth outcomes [11]. Ovarian reserve plays a critical role in determining women’s fecundity. It refers to the remaining pool of primordial follicles capable of maturation [12]. Its biological decline with advancing age, coupled with the increasing trend of delayed childbearing, presents significant challenges to women’s reproductive health [13]. Beyond individual determinants, ovarian reserve is also affected by environmental exposures. Both epidemiological studies and mouse models have suggested that exposure to pyrethroid and organochlorine pesticides [14, 15], heavy metals [16], as well as endocrine-disrupting chemicals [17] are associated with decreased serum anti-Müllerian hormone (AMH) levels and/or ultrasound antral follicle count (AFC), suggesting the necessity of further identifying environmental risk factors for prevention. To date, epidemiological evidence linking ambient O3 exposure to ovarian reserve remains sparse and inconsistent [18–24]. Some studies reported that females exposed to higher levels of O3 tended to exhibit poorer ovarian reserve, reflected by declines in AFC or AMH levels [22–24]. In contrast, other studies reported no significant associations [18–21]. Such discrepant findings may partly be attributable to heterogeneity in sample size, O3 concentration, and insufficient control of other environmental variables (e.g., temperature and neighborhood greenness). Furthermore, the critical window of ovarian reserve susceptibility to O3 exposure remains poorly defined [22, 23], while the factors that mediate this hypothetical relationship have not yet been thoroughly investigated. Growing evidence supports the hypothesis that lipid metabolism is associated with both O3 exposure and ovarian reserve. Epidemiologic studies have linked ambient O3 exposure to adverse lipid alterations characterized by elevated total cholesterol (TC), increased low-density lipoprotein cholesterol (LDL-C), and reduced high-density lipoprotein cholesterol (HDL-C) [25]. Oxidative stress and chronic inflammatory responses are considered important mechanisms underlying these lipid alterations [26]. Within the female reproductive system, physiological lipid homeostasis provides critical substrates for oocyte energy metabolism and ovarian steroid hormone biosynthesis [27]. Conversely, disrupted lipid profiles correlate with a higher risk of diminished ovarian reserve (DOR) [28, 29]. This clinical link between dyslipidemia and reproductive decline aligns with guidelines from the American Society for Reproductive Medicine (ASRM) recommending lipid profile assessment in women with DOR [30]. Therefore, it is hypothesized that dysregulated lipid metabolism may partially mediate the relationship between O3 exposure and ovarian reserve. To address these gaps in current evidence, a large retrospective study involving 8,233 reproductive-aged women in Jiangxi Province, China, was conducted. It was hypothesized that higher ambient O3 exposure is linked to a decline in ovarian reserve, and that this association may vary across different exposure windows during follicle development. Furthermore, dysregulated lipid metabolism was hypothesized to act as a potential biological mediator of this relationship. This study offers novel insights into the reproductive implications of ambient O3 pollution, thereby informing targeted public health strategies to protect female fecundity, particularly among vulnerable populations. Methods Study population This cross-sectional study comprised women aged 20 to 45 years residing in Jiangxi Province, China, who underwent ovarian reserve evaluation at the provincial-level reproductive medicine center of Jiangxi Maternal and Child Health Hospital during the period from 2021 to 2023. For participants with multiple ovarian reserve assessments, only the initial measurement was considered, as these samples reflected their baseline status before any medical interventions or treatments. Initially, 10,607 individuals with detailed ovarian reserve measurement timing and residential data were included. Participants were then excluded if they were under 20 or over 45 years of age (n = 33), lacked lipid profile measurements (n = 750), or presented with specific conditions such as abnormal chromosomal karyotypes, hyperprolactinemia, hypothyroidism, immunological disorders, or recurrent pregnancy loss (n = 1,542). Additionally, 49 participants missing essential covariate data were excluded. The final analytical sample included 8,233 participants, as illustrated in Figure S1. The study was approved by the Reproductive Medicine Ethics Committee of Jiangxi Maternal and Child Health Hospital. Assessment of ovarian reserve Serum AMH levels were utilized as a biomarker of ovarian reserve due to their high stability, reproducibility, and minimal fluctuation across the menstrual cycle [31]. Venous blood was collected from all participants, centrifuged to obtain serum, and immediately analyzed for AMH using a Cobas e 801 analyzer via an electrochemiluminescence immunoassay (Elecsys® AMH Plus, Roche Diagnostics). The quantifiable boundaries of the analysis ranged between 0.01 and 23 ng/mL, alongside a 0.03 ng/mL limit of quantification (LoQ). Quality control yielded repeatability coefficients of variation (CV) of 1.0–1.8% and intermediate precision CV of 2.7–4.4%. Measurements of serum lipids Four indicators were selected to characterize participants’ lipid profiles concurrently with AMH measurement, including triglycerides (TG), HDL-C, LDL-C and TC, all of which have been routinely measured in clinical practice and consistently linked to ovarian reserve in prior studies [29, 30, 32]. Fasting venous blood was drawn from each subject at their recruitment, followed by immediate centrifugation for serum separation. Homogeneous assays were utilized to determine the concentrations of serum LDL-C and HDL-C, while the assessment of TG and TC relied on conventional enzymatic colorimetric assays (GPO-PAP for TG and CHOD-PAP for TC) following the standard guidelines from Autobio (China). The analytical measuring ranges were 0.75–2.26 mmol/L for HDL-C, 0.10–13.00 mmol/L for LDL-C, 0.19-30.0 mmol/L for TC and 0.10–11.40 mmol/L for TG, with CVs for repeatability below 3%, 3%, 4% and 5%, respectively [33]. Dyslipidemia was defined according to the Chinese lipid management guideline as meeting any of the following criteria: [1] hypertriglyceridemia (TG ≥ 1.7 mmol/L); [2] reduced HDL-C (HDL-C < 1.0 mmol/L); [3] elevated LDL-C (LDL-C ≥ 3.4 mmol/L); or [4] hypercholesterolemia (TC ≥ 5.2 mmol/L) [34]. Exposure periods Ovarian development is a tightly regulated and cyclical process in which primordial follicles are recruited and subsequently progress through the primary, secondary, and small antral stages [35]. To identify potential sensitive windows of ozone exposure, six exposure periods were defined based on the established kinetic timeline of human folliculogenesis [36–38] (Fig. 1). P1 (2 years) reflected long-term exposure affecting the dormant primordial follicle pool, whereas P2 (1 year) represented exposure across the entire developmental trajectory from follicle recruitment to the small antral stage. To further capture stage-specific vulnerability, P3 (6 months) corresponded to the relatively slow initiation from primordial to primary follicles, P5 (4 months) covered the progression from primary to secondary follicles, and P6 (2 months) delineated the transition from secondary to small antral follicles. P4 (6 months) was additionally defined to encompass the full active growth phase from the primary to the small antral stage, thereby integrating the durations of P5 and P6. For each participant, the date of AMH measurement was used as the index date. The six exposure periods were defined retrospectively by tracing backward from this index date according to the above prespecified window length. Fig. 1The six exposure periods based on follicular development. Period 1: Two years before the formation of the small antral follicle; Period 2: One year from the primordial follicle to the small antral follicle; Period 3: Six months from primordial to primary follicle stage; Period 4: Six months from primary to small antral follicle stage; Period 5: Four months from primary to secondary follicle stage; Period 6: Two months from secondary to small antral follicle stage The six exposure periods based on follicular development. Period 1: Two years before the formation of the small antral follicle; Period 2: One year from the primordial follicle to the small antral follicle; Period 3: Six months from primordial to primary follicle stage; Period 4: Six months from primary to small antral follicle stage; Period 5: Four months from primary to secondary follicle stage; Period 6: Two months from secondary to small antral follicle stage Environment variable assessment Daily grid data for O3 and PM2.5 at a 10-km spatial resolution were obtained from the Tracking Air Pollution in China (TAP) dataset (http://tapdata.org.cn/). The dataset was generated by integrating meteorological fields, ground-based observations, satellite-based aerosol optical metrics, and chemical transport model (CTM) simulations via a multi-layer machine learning framework, as detailed in previous studies [39]. Specifically, daily maximum 8-hour average (MDA8) O3 concentrations were predicted using a three‑stage random forest model, achieving a high cross-validated R² of 0.84 [39]. Daily PM2.5 concentrations were estimated via out-of-bag cross-validation, yielding R² values in a range of 0.80–0.88 [40]. Daily average temperature at a 10 km spatial scale was collected from the ERA-5 dataset, the European Centre for Medium-Range Weather Forecasts (ECMWF) (https://cds.climate.copernicus.eu/datasets). To assess neighborhood greenness, the Normalized Difference Vegetation Index (NDVI) was calculated across a 400-meter buffer around each geocoded residence, referred to as NDVI400. NDVI data were obtained from MODIS MOD13Q1 v6.1 satellite imagery (250 m × 250 m spatial resolution; 16-day temporal resolution) via Google Earth Engine (https://earthengine.google.com/). For each participant, environmental exposures to O3, PM2.5, temperature, and NDVI400 were linked according to the geocoded residential address and the corresponding calendar dates of each exposure period. The average exposure level for each period was calculated as the mean of all available exposure estimates within that period. Covariates Potential covariates included: age (20–29, 30–34, or 35–45 years), education (lower than or equal to primary school, junior high school, senior high school or college, postsecondary or university, or equal to higher than postgraduate), employment status (employed, or unemployed), smoking (yes, or no), parity (0, or ≥ 1), gravida (0, or ≥ 1), menstrual cycle (regular, or irregular), duration of infertility, season of AMH measurement (spring, summer, autumn, or winter), warm season of AMH measurement (yes or no), and residential city. Infertility duration was defined as the time interval from the onset of regular unprotected intercourse with the intention to conceive to the date of first outpatient visit for fertility evaluation [41]. Covariates were selected based on biological plausibility and previous literature. Age, season, and residential city are related to both O3 exposure and AMH levels [42–44]. Education and employment status, as socioeconomic indicators, are associated with both residential location (and thus ambient O3 exposure levels) and health-related behaviors that may influence the ovarian reserve [45, 46]. Therefore, these factors are considered potential confounders. Smoking and reproductive factors, including parity, gravida, menstrual cycle regularity and infertility duration, were considered precision factors that are biologically related to AMH but have no plausible association with ambient O3 exposure [47–50]; including them in regression models reduces residual variance and improves precision without introducing confounding bias. All these covariates have been utilized in prior research on air pollution and ovarian reserve [23, 51]. Statistical analysis For descriptive analyses, categorical and continuous variables were reported as counts (%) and means ± standard deviations (SD), respectively. Quantile-based summaries were used to characterize the distribution of ambient O3 exposure across exposure windows. To delineate the exposure-response patterns for O3 indices and ovarian reserve, restricted cubic splines (RCS) were constructed. Exposure windows exhibiting clear nonlinearity were primarily presented based on spline results, whereas windows showing a predominantly linear pattern were subsequently quantified using multivariate linear regression with log-transformed AMH levels as the outcome. Statistical assessments of ambient O3 concentrations involved both continuous modeling (per interquartile range [IQR] increase) and categorical analysis using quartiles (Q1-Q4), with Q1 serving as the reference. For categorical analyses, effect estimates were expressed as percent changes in AMH levels, computed as 100 × [exp(β)-1], along with their 95% confidence intervals (CIs). The single-exposure model (main model) adjusted for age, educational attainment, employment status, smoking habits, parity, gravidity, menstrual cycle regularity, duration of infertility, year and season of AMH assessment, whether the assessment occurred during the warm season, and residential city. Both ‘season’ and ‘warm season indicator’ were adjusted to finely control for season conditioning [51], because O3 effects are often warm season specific and AMH may also vary seasonally [52]; the former controls for within season factors, while the latter captures warm season effect modification. Given the intercorrelations among environmental variables, robustness to co-exposure confounding was additionally evaluated using dual-exposure models, in which temperature, PM2.5, or NDVI400 was added to the main model separately (one at a time) to reduce multicollinearity. These variables have been shown to influence ambient O3 levels [53, 54], and their exposures are also linked to ovarian reserve [20, 55, 56]. Multicollinearity in the two‑exposure models was assessed using generalized variance inflation factors (GVIF). Subgroup analyses were also applied to investigate potential effect modification on the relationship linking ambient O3 exposure to ovarian reserve. The stratifying variables included body mass index (BMI) (< 18.5, 18.5–23.9, ≥ 24 kg/m2), age (20–29, 30–34, or 35–45 years) and education level (≤ or > junior high school). Formal interaction effects were evaluated using likelihood ratio tests that compared the regression models with and without the O3 × modifier interaction term. Mediation analyses were conducted to examine whether lipid profiles mediates the O3-AMH association. The analyses focused on exposure periods identified as critical windows in both RCS and primary regression analyses (periods 4 and 6). Two complementary sets of mediators were evaluated. First, lipid indicators, namely HDL-C, LDL-C, TG, and TC, were assessed as continuous variables. Second, binary dyslipidemia phenotypes were assessed, including overall dyslipidemia and its component abnormalities. For each mediator, two regression models were specified: (1) Mediator model, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:M={\alpha\:}_{0}+{\alpha\:}_{1}{O}_{3}+{\alpha\:}_{2}C+{\varepsilon\:}_{M}$$\end{document}; (2) Outcome model, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:\mathrm{l}\mathrm{o}\mathrm{g}\left(AMH\right)={\beta\:}_{0}+{\beta\:}_{1}{O}_{3}+{{\beta\:}_{2}M+\:\beta\:}_{3}C+{\varepsilon\:}_{Y}$$\end{document}. The C is the vector of covariates (as defined in Sect. 2.6), and ε denotes random error. All models were estimated using logistic regression (for binary mediators) or ordinary least squares regression (for continuous mediators). To evaluate indirect effects, bias-corrected bootstrapping with 1,000 replications was performed to derive 95% CIs. Mediation analysis assumes: (i) temporal ordering (exposure → mediator → outcome); (ii) no unmeasured confounding of the exposure-outcome, exposure-mediator and mediator-outcome relationships; (iii) no measurement error; and (iv) no exposure-mediator interaction. Owing to the cross‑sectional design, assumption (i) is only partially satisfied (O3 exposure preceded enrollment, but lipids and AMH were measured concurrently). The robustness of the primary findings was further evaluated through a series of sensitivity analyses. First, logistic regression was performed by transforming AMH into a binary outcome (1 = AMH < 1.1 ng/mL, 0 = AMH ≥ 1.1 ng/mL) [57]. Second, the warm season indicator was excluded from the regression model, to examine the possibility of overadjustment. Third, given that the duration of infertility is also downstream of AMH, adjustment for this variable was not performed in the regression model. Fourth, analyses were restricted to women with male‑factor infertility only (n = 3,487). Fifth, analyses were restricted to 6,481 participants without DOR (defined as AMH < 1.1 ng/mL) [57] to reduce potential bias from underlying reproductive dysfunction. Sixth, analyses were restricted to 7,144 individuals reporting a regular menstrual cycle. Seventh, the main model was additionally adjusted for short-term ambient O3 and temperature exposures to account for possible environmental confounding. Eighth, as an alternative modeling approach, a mixed-effects linear regression model was fitted with residential city specified as a random intercept rather than as a fixed covariate. Finally, the robustness of the observed mediation effect to unmeasured confounding was assessed using mediational E-values [58]. All statistical computations were conducted using R software (version 4.3.2). For all hypothesis testing, statistical significance was defined by a two-sided P-value below 0.05. Results Characteristics of participants and ozone exposure The baseline characteristics of the 8,233 female participants are presented in Table 1. The mean age was 31.7 years, with the largest subgroup (39.1%) aged 30–34 years. A majority had completed at least a high school education (63.8%) and were employed (86.5%). Most participants reported a prior pregnancy (61.9%) and regular menstrual cycles (86.9%). The mean serum AMH concentration was 3.0 ng/mL (SD = 2.1). The prevalence of dyslipidemia among participants ranged from 11.4% (hypertriglyceridemia, n = 936) to 18.8% (elevated LDL-C, n = 1547). Table 1Characteristics of study participantsVariablesDataAMH (ng/ml ± SD)3.0 ± 2.1Age (years ± SD)31.7 ± 4.9Age, n (%) 20–29 years2707 (32.9) 30–34 years3219 (39.1) 35–45 years2307 (28.0)Educational level, n (%) Lower than or equal to primary school330 (4.0) Junior high school2646 (32.1) Senior high school or college1408 (17.1) Postsecondary or university3557 (43.2) Equal to or higher than postgraduate292 (3.5)Employment status, n (%) Unemployed1113 (13.5) Employed7120 (86.5)Smoking, n (%) No8064 (97.9) Yes169 (2.1)Parity, n (%) 05118 (62.2) ≥ 13115 (37.8)Gravida, n (%) 03135 (38.1) ≥ 15098 (61.9)Menstrual cycles, n (%) Regular7155 (86.9) Irregular1078 (13.1)Duration of infertility (years)3.5 ± 3.0Body mass index (kg/m2 ± SD)22.1 ± 3.1Body mass index, n (%) Underweight (< 18.5)811 (9.9) Normal weight (18.5–24.0)5332 (64.8) Overweight or obese (≥ 24.0)2090 (25.4)Ln (HDL-C) (mmol/L ± SD)0.32 (0.27)Low HDL-C, n (%)1511 (18.4)Ln (LDL-C) (mmol/L ± SD)0.98 (0.28)Elevated LDL-C, n (%)1547 (18.8)Ln (TG) (mmol/L ± SD)-0.09 (0.52)Hypertriglyceridemia, n (%)936 (11.4)Ln (TC) (mmol/L ± SD)1.48 (0.20)Hypercholesterolemia, n (%)1532 (18.6)Season of ovarian reserve indicators, n (%) Spring (March-May)1640 (19.9) Summer (June-August)2610 (31.7) Autumn (September-November)2235 (27.1) Winter (December-February)1748 (21.2)Warm season of indicator measurement, n (%) Yes (May-October)4210 (51.1) No (November-the following April)4023 (48.9)Year of AMH measurement, n (%) 20212926 (35.5) 20222470 (30.0) 20232837 (34.5)Abbreviation AMH antimüllerian hormone Table 1Characteristics of study participantsVariablesDataAMH (ng/ml ± SD)3.0 ± 2.1Age (years ± SD)31.7 ± 4.9Age, n (%) 20–29 years2707 (32.9) 30–34 years3219 (39.1) 35–45 years2307 (28.0)Educational level, n (%) Lower than or equal to primary school330 (4.0) Junior high school2646 (32.1) Senior high school or college1408 (17.1) Postsecondary or university3557 (43.2) Equal to or higher than postgraduate292 (3.5)Employment status, n (%) Unemployed1113 (13.5) Employed7120 (86.5)Smoking, n (%) No8064 (97.9) Yes169 (2.1)Parity, n (%) 05118 (62.2) ≥ 13115 (37.8)Gravida, n (%) 03135 (38.1) ≥ 15098 (61.9)Menstrual cycles, n (%) Regular7155 (86.9) Irregular1078 (13.1)Duration of infertility (years)3.5 ± 3.0Body mass index (kg/m2 ± SD)22.1 ± 3.1Body mass index, n (%) Underweight (< 18.5)811 (9.9) Normal weight (18.5–24.0)5332 (64.8) Overweight or obese (≥ 24.0)2090 (25.4)Ln (HDL-C) (mmol/L ± SD)0.32 (0.27)Low HDL-C, n (%)1511 (18.4)Ln (LDL-C) (mmol/L ± SD)0.98 (0.28)Elevated LDL-C, n (%)1547 (18.8)Ln (TG) (mmol/L ± SD)-0.09 (0.52)Hypertriglyceridemia, n (%)936 (11.4)Ln (TC) (mmol/L ± SD)1.48 (0.20)Hypercholesterolemia, n (%)1532 (18.6)Season of ovarian reserve indicators, n (%) Spring (March-May)1640 (19.9) Summer (June-August)2610 (31.7) Autumn (September-November)2235 (27.1) Winter (December-February)1748 (21.2)Warm season of indicator measurement, n (%) Yes (May-October)4210 (51.1) No (November-the following April)4023 (48.9)Year of AMH measurement, n (%) 20212926 (35.5) 20222470 (30.0) 20232837 (34.5)Abbreviation AMH antimüllerian hormone Characteristics of study participants Abbreviation AMH antimüllerian hormone The estimated O3 concentrations for participants differed significantly across exposure periods, as characterized in Table S1. Participants’ mean O3 exposure ranged between 88.67 (± 13.81) and 91.93 (± 11.50) µg/m³. The Q50 values of O3 exposure for the six periods were determined to be 91.01, 90.16, 91.42, 87.29, 88.64, and 90.25 µg/m³, respectively. During period 6, O3 levels exhibited the widest variability, spanning from 47.05 to 148.62 µg/m3. Correlation analyses revealed that O3 levels covaried significantly with ambient temperature, PM2.5, and NDVI400 across all exposure windows (Table S2). Exposure-response relationship between ambient O3 exposure and ovarian reserve RCS analysis was conducted to assess the exposure–response association between ambient O₃ exposure and ovarian reserve across various exposure periods (Fig. 2). The results revealed a significant monotonic linear association between higher ambient O3 levels and lower AMH concentration during periods 4 (P overall = 0.009, P nonlinear = 0.649) and 6 (P overall = 0.002, P nonlinear = 0.164). The association observed in period 1 was also significant (P overall < 0.05) but followed a nonlinear, inverted V-shaped pattern (P nonlinear = 0.010). No significant O3-AMH associations were identified across the period 2, 3 and 5. Fig. 2Exposure-response curves between ambient O3 exposure and AMH. Panels (a-f) show the exposure-response relationships between ambient O3 exposure and AMH for Periods 1-6, respectively. The curves (the red solid lines) with 95% CI (red dashed lines) are estimated by adjusted nonlinear effect models with restricted cubic splines. Covariates included age (20–29, 30–34, or 35–45 years), education (lower than or equal to primary school, junior high school, senior high school or college, postsecondary or university, or equal to higher than postgraduate), employment status (employed or unemployed), body mass index (underweight, normal weight, overweight or obese), smoking status (yes or no), parity (0 or ≥ 1), gravida (0 or ≥ 1), menstrual cycle (regular or irregular), duration of infertility (years), season of AMH measurement (spring, summer, autumn, or winter), warm season of AMH measurement (yes or no), and residential city. The histograms (gray bars) present the distributions of the ambient O3 exposure among the studied population. For the exposure-response curves, please refer to the right y-axis; for the distributions, please refer to the left y-axis. Abbreviation: AMH, antimüllerian hormone; CI, confidence interval Exposure-response curves between ambient O3 exposure and AMH. Panels (a-f) show the exposure-response relationships between ambient O3 exposure and AMH for Periods 1-6, respectively. The curves (the red solid lines) with 95% CI (red dashed lines) are estimated by adjusted nonlinear effect models with restricted cubic splines. Covariates included age (20–29, 30–34, or 35–45 years), education (lower than or equal to primary school, junior high school, senior high school or college, postsecondary or university, or equal to higher than postgraduate), employment status (employed or unemployed), body mass index (underweight, normal weight, overweight or obese), smoking status (yes or no), parity (0 or ≥ 1), gravida (0 or ≥ 1), menstrual cycle (regular or irregular), duration of infertility (years), season of AMH measurement (spring, summer, autumn, or winter), warm season of AMH measurement (yes or no), and residential city. The histograms (gray bars) present the distributions of the ambient O3 exposure among the studied population. For the exposure-response curves, please refer to the right y-axis; for the distributions, please refer to the left y-axis. Abbreviation: AMH, antimüllerian hormone; CI, confidence interval Association between ambient O3 exposure and ovarian reserve Table 2 presents the linear regression results of the O3–AMH association after adjustment for potential confounders in the main model, with statistical significance observed during periods 4, 5, and 6. Specifically, in period 4, higher ambient O3 concentration (per IQR increase) was correlated with an 8.14% reduction in AMH levels (95% CI: 3.09% to 12.93%; P = 0.002). Relative to the Q1 quartile, participants exhibited significant AMH reductions in Q3 (-8.18%, 95% CI: -13.73% to -2.26%) and Q4 (-12.03%, 95% CI: -19.27% to -4.15%). Similar associations were observed in period 6, with an 8.3% (95% CI: 2.48% to 13.77%; P = 0.006) decline in AMH per IQR increase in O₃ concentration. Compared to the Q1 quartile, a significant reduction was observed in Q2 (-7.34%, 95% CI: -12.34% to -2.04%), whereas reductions in Q3 (P = 0.054) and Q4 (P = 0.082) approached statistical significance. Period 5 also demonstrated a significant negative association, with each IQR increase in O3 linked to a 5.75% decrease in AMH levels (P = 0.037), while comparisons across quartile groups did not reach statistical significance. Table 2Association of ambient O3 exposure levels with AMH across different exposure windowsExposure windowQuartilesPercentage change (95% CI) P Period 1Per IQR change1.86 (-2.92, 6.87)0.452QuartilesQ1ReferenceQ21.82 (-3.34, 7.26)0.497Q35.86 (-0.93, 13.12)0.092Q4-0.20 (-8.06, 8.33)0.962Period 2Per IQR change-4.30 (-8.79, 0.41)0.073QuartilesQ1ReferenceQ2-2.17 (-7.08, 3.00)0.403Q3-0.42 (-6.24, 5.76)0.891Q4-5.05 (-12.39, 2.91)0.207Period 3Per IQR change0.13 (-5.05, 5.59)0.963QuartilesQ1ReferenceQ2-2.37 (-7.66, 3.22)0.398Q30.45 (-6.10, 7.46)0.896Q40.30 (-7.93, 9.27)0.945Period 4Per IQR change-8.14 (-12.93, -3.09) 0.002 QuartilesQ1ReferenceQ2-2.48 (-7.35, 2.64)0.336Q3-8.18 (-13.73, -2.26) 0.007 Q4-12.03 (-19.27, -4.15) 0.003 Period 5Per IQR change-5.75 (-10.86, -0.35) 0.037 QuartilesQ1ReferenceQ21.67 (-3.59, 7.22)0.541Q3-2.25 (-8.48, 4.39)0.497Q4-2.27 (-10.26, 6.44)0.598Period 6Per IQR change-8.30 (-13.77, -2.48) 0.006 QuartilesQ1ReferenceQ2-7.34 (-12.34, -2.04) 0.007 Q3-6.79 (-13.22, 0.11)0.054Q4-7.74 (-15.76, 1.04)0.082Abbreviation AMH antimüllerian hormone, CI Confidence interval, IQR interquartile rangea Analyzed with multivariate linear models. Models were adjusted for age (20–29, 30–34, or 35–45 years), education (lower than or equal to primary school, junior high school, senior high school or college, postsecondary or university, or equal to higher than postgraduate), employment status (employed or unemployed), body mass index (underweight, normal weight, overweight or obese), smoking status (yes or no), parity (0 or ≥ 1), gravida (0 or ≥ 1), menstrual cycle (regular or irregular), duration of infertility (years), season of AMH measurement (spring, summer, autumn, or winter), warm season of AMH measurement (yes or no), and residential city. P values were two-sided and multiple comparisons were not adjusted. The P values in bold are < 0.05 Table 2Association of ambient O3 exposure levels with AMH across different exposure windowsExposure windowQuartilesPercentage change (95% CI) P Period 1Per IQR change1.86 (-2.92, 6.87)0.452QuartilesQ1ReferenceQ21.82 (-3.34, 7.26)0.497Q35.86 (-0.93, 13.12)0.092Q4-0.20 (-8.06, 8.33)0.962Period 2Per IQR change-4.30 (-8.79, 0.41)0.073QuartilesQ1ReferenceQ2-2.17 (-7.08, 3.00)0.403Q3-0.42 (-6.24, 5.76)0.891Q4-5.05 (-12.39, 2.91)0.207Period 3Per IQR change0.13 (-5.05, 5.59)0.963QuartilesQ1ReferenceQ2-2.37 (-7.66, 3.22)0.398Q30.45 (-6.10, 7.46)0.896Q40.30 (-7.93, 9.27)0.945Period 4Per IQR change-8.14 (-12.93, -3.09) 0.002 QuartilesQ1ReferenceQ2-2.48 (-7.35, 2.64)0.336Q3-8.18 (-13.73, -2.26) 0.007 Q4-12.03 (-19.27, -4.15) 0.003 Period 5Per IQR change-5.75 (-10.86, -0.35) 0.037 QuartilesQ1ReferenceQ21.67 (-3.59, 7.22)0.541Q3-2.25 (-8.48, 4.39)0.497Q4-2.27 (-10.26, 6.44)0.598Period 6Per IQR change-8.30 (-13.77, -2.48) 0.006 QuartilesQ1ReferenceQ2-7.34 (-12.34, -2.04) 0.007 Q3-6.79 (-13.22, 0.11)0.054Q4-7.74 (-15.76, 1.04)0.082Abbreviation AMH antimüllerian hormone, CI Confidence interval, IQR interquartile rangea Analyzed with multivariate linear models. Models were adjusted for age (20–29, 30–34, or 35–45 years), education (lower than or equal to primary school, junior high school, senior high school or college, postsecondary or university, or equal to higher than postgraduate), employment status (employed or unemployed), body mass index (underweight, normal weight, overweight or obese), smoking status (yes or no), parity (0 or ≥ 1), gravida (0 or ≥ 1), menstrual cycle (regular or irregular), duration of infertility (years), season of AMH measurement (spring, summer, autumn, or winter), warm season of AMH measurement (yes or no), and residential city. P values were two-sided and multiple comparisons were not adjusted. The P values in bold are < 0.05 Association of ambient O3 exposure levels with AMH across different exposure windows Abbreviation AMH antimüllerian hormone, CI Confidence interval, IQR interquartile range a Analyzed with multivariate linear models. Models were adjusted for age (20–29, 30–34, or 35–45 years), education (lower than or equal to primary school, junior high school, senior high school or college, postsecondary or university, or equal to higher than postgraduate), employment status (employed or unemployed), body mass index (underweight, normal weight, overweight or obese), smoking status (yes or no), parity (0 or ≥ 1), gravida (0 or ≥ 1), menstrual cycle (regular or irregular), duration of infertility (years), season of AMH measurement (spring, summer, autumn, or winter), warm season of AMH measurement (yes or no), and residential city. P values were two-sided and multiple comparisons were not adjusted. The P values in bold are < 0.05 Based on the main model, double-exposure models were further constructed by separately including temperature, PM2.5, and NDVI400 (Fig. 3). The associations in periods 4 and 6 remained stable after adjusting for any of the three environmental covariates, demonstrating the robustness of the O3–AMH relationship during these periods. The association in period 5 turned to be non-significant after further adjustment for temperature or PM2.5. In two-exposure models, all GVIF values for O3 and the co‑exposures (NDVI, PM2.5, temperature) ranged from 1.18 to 3.21, indicating that multicollinearity did not materially affect the model estimates (Table S3). Fig. 3Percentage changes of AMH associated with an IQR increase in ambient O3 exposure across different exposure periods. In each period, the fully adjusted percentage change with its 95% CI (black dot with error bar) estimated by the single-exposure model of O3 (the main model) is compared with the estimates of corresponding double-exposure models (shown in grey). Covariates included age (20–29, 30–34, or 35–45 years), education (lower than or equal to primary school, junior high school, senior high school or college, postsecondary or university, or equal to higher than postgraduate), employment status (employed or unemployed), body mass index (underweight, normal weight, overweight or obese), smoking status (yes or no), parity (0 or ≥ 1), gravida (0 or ≥ 1), menstrual cycle (regular or irregular), duration of infertility (years), season of AMH measurement (spring, summer, autumn, or winter), warm season of AMH measurement (yes or no), and residential city. An asterisk represents P < 0.05. Abbreviation: AMH, antimüllerian hormone; CI, confidence interval; IQR, interquartile range; NDVI: average neighborhood normalized difference vegetation index (NDVI) within a 400 m radius of residence; PM2.5: Particulate Matter with a diameter of 2.5 μm or less Percentage changes of AMH associated with an IQR increase in ambient O3 exposure across different exposure periods. In each period, the fully adjusted percentage change with its 95% CI (black dot with error bar) estimated by the single-exposure model of O3 (the main model) is compared with the estimates of corresponding double-exposure models (shown in grey). Covariates included age (20–29, 30–34, or 35–45 years), education (lower than or equal to primary school, junior high school, senior high school or college, postsecondary or university, or equal to higher than postgraduate), employment status (employed or unemployed), body mass index (underweight, normal weight, overweight or obese), smoking status (yes or no), parity (0 or ≥ 1), gravida (0 or ≥ 1), menstrual cycle (regular or irregular), duration of infertility (years), season of AMH measurement (spring, summer, autumn, or winter), warm season of AMH measurement (yes or no), and residential city. An asterisk represents P < 0.05. Abbreviation: AMH, antimüllerian hormone; CI, confidence interval; IQR, interquartile range; NDVI: average neighborhood normalized difference vegetation index (NDVI) within a 400 m radius of residence; PM2.5: Particulate Matter with a diameter of 2.5 μm or less Effect modification by age, BMI and educational level As shown in Table 3, stratified analysis indicated that age significantly modified the relationship between ambient O3 and AMH. During period 4, a more pronounced O3-related decline in AMH was observed among women younger than 35 years. Specifically, an IQR increase in O3 corresponded to an 11.05% (95% CI: 5.30% to 16.46%) reduction in AMH for women aged 20–29, compared with a non-significant association for those aged 35–45. This age-related effect was not observed in period 6. Additionally, no significant effect modification was detected for BMI or educational level, as evidenced by all interaction P values > 0.05. Table 3Percentage changes in AMH associated with an interquartile range (IQR) increase in ambient O3 exposure in Periods 4 and 6, stratified by age, body mass index and education levelSubgroupPeriod 4Period 6Percentage change, % (95% CI)P forinteraction aPercentage change, % (95% CI)P forinteraction aAge0.0040.001 20–29 years-11.05 (-16.46, -5.30)-8.11 (-14.63, -1.09) 30–34 years-7.40 (-13.08, -1.34)-11.62 (-18.11, -4.61) 35–45 years3.03 (-5.58, 12.43)0.39 (-9.53, 11.40)Body mass index0.8550.735 Underweight(< 18.5)-8.37 (-14.29, -2.04)-7.29 (-14.02, -0.04) Normal weight(18.5–24.0)-15.73 (-26.84, -2.95)-6.53 (-20.20, 9.48) Overweight or obese (≥ 24.0)-3.77 (-14.08, 7.78)-12.52 (-23.74, 0.35)Education level0.1290.405 ≤Junior high school-4.63 (-12.42, 3.84)-9.83 (-18.22, -0.57) > Junior high school-9.95 (-15.89, -3.59)-6.65 (-13.55, 0.81)a Formal interaction tests were performed using likelihood ratio tests comparing models with and without the O₃ × modifier product term. Both models included O₃, the modifier, and the covariates used in the main model Table 3Percentage changes in AMH associated with an interquartile range (IQR) increase in ambient O3 exposure in Periods 4 and 6, stratified by age, body mass index and education levelSubgroupPeriod 4Period 6Percentage change, % (95% CI)P forinteraction aPercentage change, % (95% CI)P forinteraction aAge0.0040.001 20–29 years-11.05 (-16.46, -5.30)-8.11 (-14.63, -1.09) 30–34 years-7.40 (-13.08, -1.34)-11.62 (-18.11, -4.61) 35–45 years3.03 (-5.58, 12.43)0.39 (-9.53, 11.40)Body mass index0.8550.735 Underweight(< 18.5)-8.37 (-14.29, -2.04)-7.29 (-14.02, -0.04) Normal weight(18.5–24.0)-15.73 (-26.84, -2.95)-6.53 (-20.20, 9.48) Overweight or obese (≥ 24.0)-3.77 (-14.08, 7.78)-12.52 (-23.74, 0.35)Education level0.1290.405 ≤Junior high school-4.63 (-12.42, 3.84)-9.83 (-18.22, -0.57) > Junior high school-9.95 (-15.89, -3.59)-6.65 (-13.55, 0.81)a Formal interaction tests were performed using likelihood ratio tests comparing models with and without the O₃ × modifier product term. Both models included O₃, the modifier, and the covariates used in the main model Percentage changes in AMH associated with an interquartile range (IQR) increase in ambient O3 exposure in Periods 4 and 6, stratified by age, body mass index and education level Underweight (< 18.5) Normal weight (18.5–24.0) a Formal interaction tests were performed using likelihood ratio tests comparing models with and without the O₃ × modifier product term. Both models included O₃, the modifier, and the covariates used in the main model Two mediation analyses were conducted to examine whether lipid profiles mediate the O3–AMH relationship. Among the four indicators, TC consistently served as a mediator during both period 4 and period 6, accounting for 11.415% and 9.668% of the total effect, respectively (Fig. 4). Similarly, hypercholesterolemia (defined as TC ≥ 5.2 mmol/L) contributed 0.756% to the mediating effect in period 4 (Fig. 5). Regarding LDL-C, it mediated 7.38% of the O3-AMH relationship during period 4, and showed a borderline significant mediating effect of 4.45% during period 6 (P = 0.074) (Fig. 4). Elevated LDL-C consistently mediated this association across both periods 4 and 6, with mediating effects of 7.52% and 5.22%, respectively (Fig. 5). Defined as the presence of at least one condition including low HDL-C, elevated LDL-C, hypertriglyceridemia, or hypercholesterolemia [34], dyslipidemia also mediated a significant portion (2.534%) of the total association during period 4 (Fig. 5). In contrast, the proportion mediated (PM) values for HDL-C (e.g., -0.046% in period 4; − 0.966% in period 6) failed to reach statistical significance in both periods, indicating that HDL-C did not act as a meaningful mediator in the O3-AMH association. The negative indirect effect suggests that any potential mediation is either absent or inconsistent, and these estimates should not be interpreted as genuine mediating pathways. Fig. 4Mediation analyses of whether the associations between O3 and AMH were mediated by lipid profile. Panels a and b correspond to Period 4 and Period 6, respectively. All of the potential mediators were treated as continuous variables. Covariates included age (20–29, 30–34, or 35–45 years), education (lower than or equal to primary school, junior high school, senior high school or college, postsecondary or university, or equal to higher than postgraduate), employment status (employed or unemployed), body mass index (underweight, normal weight, overweight or obese), smoking status (yes or no), parity (0 or ≥ 1), gravida (0 or ≥ 1), menstrual cycle (regular or irregular), duration of infertility (years), season of AMH measurement (spring, summer, autumn, or winter), warm season of AMH measurement (yes or no), and residential city. Abbreviation: ACME: average causal mediation effect; ADE: average direct effect; AMH, antimüllerian hormone; CI: confidence interval; HDL-C: high-density lipoprotein cholesterol; LDL-C: low-density lipoprotein cholesterol; PM: proportion of mediation effect; TC: total cholesterol; TG: triglyceride Mediation analyses of whether the associations between O3 and AMH were mediated by lipid profile. Panels a and b correspond to Period 4 and Period 6, respectively. All of the potential mediators were treated as continuous variables. Covariates included age (20–29, 30–34, or 35–45 years), education (lower than or equal to primary school, junior high school, senior high school or college, postsecondary or university, or equal to higher than postgraduate), employment status (employed or unemployed), body mass index (underweight, normal weight, overweight or obese), smoking status (yes or no), parity (0 or ≥ 1), gravida (0 or ≥ 1), menstrual cycle (regular or irregular), duration of infertility (years), season of AMH measurement (spring, summer, autumn, or winter), warm season of AMH measurement (yes or no), and residential city. Abbreviation: ACME: average causal mediation effect; ADE: average direct effect; AMH, antimüllerian hormone; CI: confidence interval; HDL-C: high-density lipoprotein cholesterol; LDL-C: low-density lipoprotein cholesterol; PM: proportion of mediation effect; TC: total cholesterol; TG: triglyceride Fig. 5Mediation analyses of whether the associations between O3 and AMH were mediated by lipid profile. Panels a and b correspond to Period 4 and Period 6, respectively. All of the potential mediators were treated as binary variables. Covariates included age (20–29, 30–34, or 35–45 years), education (lower than or equal to primary school, junior high school, senior high school or college, postsecondary or university, or equal to higher than postgraduate), employment status (employed or unemployed), body mass index (underweight, normal weight, overweight or obese), smoking status (yes or no), parity (0 or ≥ 1), gravida (0 or ≥ 1), menstrual cycle (regular or irregular), duration of infertility (years), season of AMH measurement (spring, summer, autumn, or winter), warm season of AMH measurement (yes or no), and residential city. Abbreviation: ACME: average causal mediation effect; ADE: average direct effect; AMH, antimüllerian hormone; CI: confidence interval; HDL-C: high-density lipoprotein cholesterol; LDL-C: low-density lipoprotein cholesterol; PM: proportion of mediation effect Mediation analyses of whether the associations between O3 and AMH were mediated by lipid profile. Panels a and b correspond to Period 4 and Period 6, respectively. All of the potential mediators were treated as binary variables. Covariates included age (20–29, 30–34, or 35–45 years), education (lower than or equal to primary school, junior high school, senior high school or college, postsecondary or university, or equal to higher than postgraduate), employment status (employed or unemployed), body mass index (underweight, normal weight, overweight or obese), smoking status (yes or no), parity (0 or ≥ 1), gravida (0 or ≥ 1), menstrual cycle (regular or irregular), duration of infertility (years), season of AMH measurement (spring, summer, autumn, or winter), warm season of AMH measurement (yes or no), and residential city. Abbreviation: ACME: average causal mediation effect; ADE: average direct effect; AMH, antimüllerian hormone; CI: confidence interval; HDL-C: high-density lipoprotein cholesterol; LDL-C: low-density lipoprotein cholesterol; PM: proportion of mediation effect Sensitivity analyses Figure S2 shows the ambient O3-AMH associations of the main model across different periods, following several additional analyses: without adjustment for warm season or duration of infertility, restricting analysis to women with male‑factor infertility only, restricting to individuals without DOR, including only those with a regular menstrual cycle, additionally adjusting for short-term exposure to ambient O3 and temperature, and using a mixed linear regression model with residential city as a random intercept. The ambient O3-AMH associations in periods 4 and 6 were consistently significant across all sensitivity analyses. When dichotomizing AMH into a binary variable using a cutoff of 1.1 ng/mL, logistic regression yielded odds ratios (ORs) > 1 for all exposure windows (ranging from 1.04 to 1.29), with a statistically significant association was observed in period 2 (P = 0.024) and marginally significant associations during period 4 (OR = 1.25, P = 0.056) and period 6 (OR = 1.25, P = 0.101) (Table S4). In addition, the mediational E-values of LDL-C, elevated LDL-C, hypercholesterolemia and dyslipidemia ranged from 1.05 to 1.11 (Table S5). These E-values suggest that an unmeasured confounder would need a joint association strength of 5%–11% above the null with both the mediator and the outcome to fully account for the observed indirect effect. Discussion This large cross-sectional study conducted in Jiangxi Province, China, demonstrated a negative association between ambient ozone exposure and ovarian reserve, specifically during the critical follicular development window spanning from the primary and secondary stages to the small antral stage. This susceptibility was particularly heightened among women younger than 35 years. Notably, no prior studies have explored the mediating mechanisms underlying this relationship. This study offers initial insights into how dysregulated lipid profiles play a partial mediating role, providing a novel theoretical basis for future public health strategies designed to preserve female fertility. Limited research has examined the association between ambient ozone exposure and ovarian reserve, with their findings remaining inconsistent [18–24]. Two studies from Poland and South Korea failed to detect significant associations [18, 20]. However, the relatively advanced mean age of participants (36.6 ± 4.2 years) in the Korean study may have attenuated detectable ozone-related effects [18], whereas the low average ozone concentration (44.80 ± 18.21 µg/m³) in the Polish study may limit the ability to detect an association [20]. Two additional studies conducted in China reported similar findings. Research from Shandong and Shanxi provinces reported relatively high ozone concentrations (96.4–109.6 µg/m³ and 120 µg/m³) yet no significant associations between ozone exposure and ovarian reserve were detected [19, 21]. Notably, these studies did not adjust for the potential influence of additional environmental factors. In contrast, three studies in Tianjin, Hubei, and Sichuan, China consistently reported that higher ozone exposure was associated with reduced ovarian reserve, despite differences in sample size (2,212, 6,008 and 2,815, respectively) and ozone concentrations (100.56, 88.4–92.6 and 91.1 µg/m³, respectively) across these studies [22–24]. This heterogeneity highlights the need for large-scale studies across diverse geographic regions to clarify region-specific exposure–response patterns and to evaluate the generalizability of ozone-related effects on ovarian reserve. This study included 8,233 women from Jiangxi Province and investigated the association between ozone exposure (88.67–91.93 µg/m³) and AMH levels. To identify potential windows of susceptibility, six distinct periods of ovarian development were defined. Notably, the inverse ozone–AMH association remained robust in two-pollutant models with additional adjustment for ambient temperature, PM2.5, and NDVI400, further supporting the reliability of the observed findings. By defining six distinct exposure windows, this study identified that ovarian reserve was most sensitive to ozone exposure during narrower exposure windows including period 4 (primary to small antral follicle) and period 6 (secondary to small antral follicle), rather than broader ones. This finding aligns with previous research conducted in Shanxi, China, which observed elevated vulnerability to sulphur dioxide (SO2) exposure during the phase when primary follicles develop into preantral follicless (β = −0.01; 95% CI: −0.016 to − 0.002) [19]. Similarly, another study involving 631 women in the northeastern United States suggested that cumulative heat exposure from the preantral to preovulatory follicular stages may exert greater adverse effects [59]. These findings consistently reflect a stage-specific vulnerability during folliculogenesis. The developmental shift from primary or secondary follicles into early antral follicles is characterized by rapid granulosa cell proliferation, increased metabolic demand and active follicular selection [60, 61]. These processes may render developing follicles more susceptible to oxidative or endocrine disruption induced by environmental exposures [62]. In addition, because preantral and early antral follicles constitute the principal source of granulosa cell-derived AMH, ozone-related injury during these specific stages is more readily captured through a subsequent decline in AMH levels [31]. Experimental studies further support ozone-induced toxicity by demonstrating oxidative ovarian injury, granulosa cell dysfunction, and impaired fertility following exposure [63, 64]. Taken together, these insights highlight that minimizing ozone exposure could be particularly important for preserving ovarian reserve, particularly during the six months preceding conception. Stratified analysis indicated greater vulnerability to ambient ozone among women under 35 years. Age 35 represents a critical transition in follicular development, as both follicle quantity and quality decline more rapidly thereafter, following a biphasic pattern [65]. This accelerated, age-related deterioration potentially obscures the adverse impacts of air pollution on ovarian reserve [66]. Conversely, younger females possess more metabolically active follicles with higher vascular demands, potentially facilitating pollutant entry and thereby increasing susceptibility to ozone-induced ovarian damage [67]. Consistent results have also been observed in previous studies. Studies conducted in Tianjin and Hubei similarly reported greater ozone-related reductions in AMH levels among women under 35 and 30 years, respectively [22, 23]. Collectively, these findings identify younger women as a population particularly vulnerable to ozone-associated decline in ovarian reserve, offering valuable insights for targeted prevention and protective interventions. Although the precise mechanism through which ozone influences ovarian reserve remains unclear, this study provides initial evidence that alterations in lipid metabolism may partially mediate this association. In the present study, mediation analysis revealed that dyslipidemia accounted for 2.534% of the total effect in period 4, with TC and LDL-C contributing the largest proportions among all evaluated mediators. Growing evidence links ozone exposure to adverse alterations in lipid profiles. A prospective multi-center study across 31 provinces in China reported that high ozone concentrations were associated with elevated lipid levels, while cleaner air policies improved lipid metabolism [68]. Similarly, findings from the China Multi-Ethnic Cohort indicated that for every 10 µg/m³ increase in the three-year average of ozone exposure, TC and LDL-C rose by 1.34% and 9.46%, respectively, while HDL-C decreased by 5.16% [25]. Despite these associations, the underlying mechanisms remain unclear. One hypothesis suggests that ozone exposure induces the hyperactivation of the hypothalamic–pituitary–adrenal (HPA) axis, thereby upregulating stress hormone secretion, which in turn promotes lipolysis and metabolic dysregulation [69]. Another proposed mechanism is that ozone induces oxidative stress and inflammatory cascades, elevating circulating inflammatory mediators that disrupt lipid homeostasis [70, 71]. Emerging evidence also suggests that lipids play a crucial role in female fecundity. Prior research by Calonge et al. demonstrated that dysregulated lipid profiles in both plasma and follicular fluid negatively affect ovarian response, characterized by a diminished count of retrieved mature oocytes [72]. A study using depression-like mice demonstrated that disrupted lipid metabolism mediates and potentially contributes to ovarian insufficiency induced by psychological stress [73]. These effects may reflect the critical role of lipids in oocyte development. Lipids are essential for ovarian energy metabolism and steroid hormone biosynthesis [27], and their dysregulation may further compromise oocyte quality through mechanisms such as vascular damage and cellular lipotoxicity [74]. In summary, abnormal lipid profiles may serve as an intermediate pathway linking ambient ozone exposure to the depletion of ovarian reserve. Ozone exposure could disrupt lipid metabolism through activation of neurohormonal stress responses and induction of redox imbalance. The resulting dyslipidemia may impair oocyte maturation and accelerate follicular aging, thereby diminishing ovarian reserve. Nevertheless, lipid profiles may capture not only a biological intermediate pathway, but also broader metabolic and lifestyle patterns linked to diet, physical activity, and regional context. Therefore, the underlying mechanisms remain complex and insufficiently characterized, warranting further investigation for causality. Strengths and limitations This study has several notable strengths. First, the analysis is based on a large sample of 8,233 women from Jiangxi Province in eastern China, which enhances statistical reliability and provides valuable regional data, enriching the current evidence base. Second, ozone exposure was estimated using the TAP dataset, which integrates multi-source data to generate high–spatiotemporal-resolution MDA8 ozone concentrations, thereby improving exposure accuracy. Third, in contrast to existing research that concentrated on the direct linkage between ozone and ovarian reserve, the present study fills a crucial knowledge gap by unveiling the mediating role of lipid profiles. Finally, the consistency between the two-pollutant and single-pollutant models further strengthens the validity and robustness of the findings. This study has several limitations. First, its cross-sectional design has inherent bias, necessitating cautious interpretation. For example, exposure was assigned using the residential address recorded at recruitment, and residential mobility as well as time spent away from home could not be fully captured, which may have introduced non-differential exposure misclassification. Second, as participants were recruited from infertility centers, generalizing the results to the broader reproductive-aged females warrants caution. Nevertheless, existing studies have suggested that AMH levels in infertile women under 40 years of age are comparable to those in their peers without infertility [75]. Third, despite a large population enrolled, this study was conducted in a single geographic region, which may constrain the external validity of the results. Therefore, further studies in other geographic settings or multi-region datasets are warranted to validate these findings. Fourth, AFC data were missing for most participants and therefore could not be analyzed as a complementary outcome. Although AMH is a well-established and cycle-independent marker, the absence of AFC limited our potential to fully assess the ovarian reserve. Fifth, the study period (2021–2023) partially overlapped with the COVID-19 pandemic, during which human activities and air pollution profiles may have been altered. Nevertheless, the main findings remain consistent with studies from pre-pandemic to pandemic (2018–2021) [23], suggesting that pandemic‑related biases, if any, are unlikely to have substantially affected the study conclusions. Sixth, age at menarche was not available for most participants and therefore was not adjusted in the regression models. Although it may be associated with ovarian reserve, it is more likely to act as a precision variable rather than a confounder of the ambient ozone–AMH association. Thus, its omission may have affected statistical precision but is unlikely to introduce substantial confounding bias. Seventh, although adjustment was made for PM2.5 and NDVI400 in the co-exposure model, other air pollutants, including NO2, SO2, and CO, were not included because high-resolution exposure data were not consistently available during the study period. Therefore, residual confounding by other environmental exposures cannot be fully excluded. Further studies incorporating comprehensive multi-pollutant exposure assessment are warranted. Finally, the concurrent measurement of lipids and AMH precludes definitive causal ordering, and future prospective studies with repeated measurements are needed to validate the mediating role of lipid profiles. Conclusion This population-based study demonstrates a negative association between ambient ozone exposure and ovarian reserve that is partially mediated by altered lipid profiles. The relationship is particularly pronounced during the transition from primary and secondary to small antral follicle stages, with women under 35 showing greater susceptibility. These findings indicate that ambient ozone exposure represents a potentially modifiable environmental risk factor in female fertility preservation. This insight underscores the clinical utility of integrating air-pollution awareness into fertility counseling and risk stratification, while also informing targeted public health strategies. Supplementary Information Supplementary Material 1. Supplementary Material 1.