Metabolic dysregulation in follicular fluid is associated with adverse reproductive outcomes in women with decreased ovarian reserve.

The clinical characteristics of the IVF/ICSI participants are presented in Table  1 . A total of 426 women undergoing IVF treatment were enrolled in this study, following stringent inclusion and exclusion criteria as outlined in the experimental section (Materials and Methods). The cohort comprised 126 patients diagnosed with DOR and 300 NOR. Demographic analysis revealed comparable distributions between groups in terms of race, education level, infertility type, BMI, and duration of infertility ( P  > 0.05). However, significant between-group differences were identified in age, ovarian stimulation protocols, and fertilization methods ( P  < 0.05). Consistent with diagnostic criteria, the DOR cohort displayed markedly reduced AMH concentrations and AFC counts alongside elevated FSH levels relative to controls ( P  < 0.001), confirming the validity of our participant stratification. Meanwhile, the DOR group showed significant hormonal differences compared to NOR, including lower LH, P, and T but elevated E 2 (Fig.  2 ). Additionally, reproductive outcomes were significantly poorer in DOR patients, with lower numbers of retrieved oocytes, MII oocytes, number of 2PN zygote, high-quality embryo rates, clinical pregnancy rates, high-quality embryo rates, and live birth rates ( P  < 0.05). Table 1 The demographic and clinical characteristics of patients with DOR and NOR [Mean ± SD,M (P 25 ,P 75 ) or n (%)] DOR group NOR group P - value Age (year) 33.87 ± 3.62 31.55 ± 3.57 < 0.001 BMI (kg/m 2 ) 22.18 ± 3.09 21.72 ± 2.87 0.145 Duration of infertility (year) 3.70 ± 2.90 3.34 ± 2.60 0.230 AFC (n) 5.00 (3.00, 7.00) 12.00 (8.00, 16.00) < 0.001 Race 0.882  Han 121 (96.03%) 289 (96.33%)  Other 5 (3.97%) 11 (3.67%) Education level 0.499  Less than associate degree 61 (48.41%) 156 (52.00%)  Associate degree and above 65 (51.59%) 144 (48.00%) Type of infertility 0.455  Primary infertility 58 (46.03%) 150 (50%)  Secondary infertility 68 (53.97%) 150 (50%) Fertilization method 0.021  IVF 29 (23.02%) 103 (34.33%)  ICSI 97 (76.98%) 197 (65.67%) Ovarian stimulation protocols < 0.001  Ultra-long GnRH-a 39 (30.95%) 220 (73.33%)  GnRH antagonist 51 (40.48%) 73 (24.33%)  Others 36 (28.57%) 7 (2.33%) Oocytes retrieved (n) 5.00 (3.00, 8.00) 13.00 (9.00, 17.00) < 0.001 MII oocytes retrieved (n) 3.00 (1.00, 5.00) 6.00 (3.00, 10.00) < 0.001 Cleavage rate 91.71 ± 23.55 95.89 ± 9.57 0.056 Fertilization rate 88.29 ± 24.94 88.26 ± 17.66 0.991 Number of 2PN zygote (n) 4.00 (2.00, 5.00) 8.00 (5.00, 11.75) < 0.001 High-quality embryo rate 55.21 ± 34.40 62.38 ± 26.37 0.037 Number of embryos transferred (n) 0.73 ± 0.89 0.78 ± 0.83 0.590 Implantation rate 66.67(50.00, 93.75) 66.67(50.00, 85.71) 0.793 Clinical pregnancy rate 33(28.52%) 134(65.87%) < 0.001 Live birth rate 28(22.22%) 113(37.67%) 0.002 Miscarriage rate 7(21.21%) 25(20.16%) 0.738 DOR Decreased ovarian reserve, NOR Normal ovarian reserve, BMI Body mass index, AFC Antral follicle count Fig. 2 Basal reproductive hormones in serum between DOR and NOR. Serums hormones level of ( A ) FSH, ( B ) AMH, ( C ) LH, ( D ) T, ( E ) P, and ( F ) E 2 were displayed as violin plots with boxplots inside (median, interquartile range, and full data range). * p  < 0.05, ** p  < 0.01, *** p  < 0.001. FSH, follicle-stimulating hormone; AMH, anti-Müllerian hormone; LH, luteinizing hormone; T, testosterone; P, progesterone; E₂, estradiol The demographic and clinical characteristics of patients with DOR and NOR [Mean ± SD,M (P 25 ,P 75 ) or n (%)] DOR Decreased ovarian reserve, NOR Normal ovarian reserve, BMI Body mass index, AFC Antral follicle count Basal reproductive hormones in serum between DOR and NOR. Serums hormones level of ( A ) FSH, ( B ) AMH, ( C ) LH, ( D ) T, ( E ) P, and ( F ) E 2 were displayed as violin plots with boxplots inside (median, interquartile range, and full data range). * p  < 0.05, ** p  < 0.01, *** p  < 0.001. FSH, follicle-stimulating hormone; AMH, anti-Müllerian hormone; LH, luteinizing hormone; T, testosterone; P, progesterone; E₂, estradiol We employed LC–MS/MS to analyze follicular fluid samples from the DOR and NOR groups in both positive and negative ion modes, identifying a total of 4,261 metabolites. Following validation of the data using QC samples, 371 metabolites were retained for subsequent statistical analysis. To identify the best features for distinguishing between DOR and NOR groups, we used the Lasso regression, OPLS-DA model, and fold change data combined with a t-test for feature selection. Lasso regression analysis was conducted with tenfold cross-validation to determine the penalty parameter λ. Based on the lambda. Min criterion, the model corresponding to the minimum mean cross-validation error was selected (λ = 0.0274; Fig.  3 B). A total of 33 metabolites with non-zero coefficients were identified (Fig.  3 A). The OPLS-DA model demonstrated a distinct separation between the DOR and NOR groups along the predictive component (LV1), indicating clear differentiation in metabolic profiles. Based on VIP scores (VIP > 1),117 metabolites were extracted from the OPLS-DA model (Fig.  3 C). Volcano plot analysis showed that 128 metabolites differed significantly between the DOR and NOR groups (Fig.  3 D). Specifically, 10 metabolites were upregulated and 118 were downregulated in the DOR group compared to the NOR group (FDR p  1.2, p-test < 0.05). Ultimately, a total of 17 differentially regulated metabolites were identified. Specifically, metabolites such as acetamidopropanal, D-ribose, and oxoadipic acid were significantly upregulated in patients with DOR. Conversely, 14 metabolites, including 5a-Pregnane-3,20-dione, estrone glucuronide, and ribothymidine, were significantly downregulated (Fig.  3 E). Fig. 3 Identification of differential metabolomics profiles in the follicular fluid between DOR and NOR. A Lasso regression coefficient pathway plot. The x-axis represents the logarithm of the regularization parameter Log Lambda, and the y-axis represents the model coefficients. B Cross-validation error plot. The red curve indicates the cross-validation error at different Log(λ) values, with the gray and blue vertical lines marking the minimum error point lambda min and the one-standard-error point lambda 1-se. C OPLS-DA score plots. The purple circles represent NOR; the blue circles represent DOR. D Volcanoes show the distribution of different features. E Lollipop plot of differential metabolites Identification of differential metabolomics profiles in the follicular fluid between DOR and NOR. A Lasso regression coefficient pathway plot. The x-axis represents the logarithm of the regularization parameter Log Lambda, and the y-axis represents the model coefficients. B Cross-validation error plot. The red curve indicates the cross-validation error at different Log(λ) values, with the gray and blue vertical lines marking the minimum error point lambda min and the one-standard-error point lambda 1-se. C OPLS-DA score plots. The purple circles represent NOR; the blue circles represent DOR. D Volcanoes show the distribution of different features. E Lollipop plot of differential metabolites The pathway and enrichment analyses were conducted using the 17 selected differential metabolites with MetaboAnalyst 6.0, referencing the KEGG database. The enrichment analysis indicated that DOR-induced metabolic dysregulation primarily affected the following pathways: steroid hormone biosynthesis, retinol metabolism, ether lipid metabolism, pentose phosphate pathway, lipid acid metabolism, lysine degradation, pyrimidine metabolism, tryptophan metabolism, arachidonic acid metabolism, and purine metabolism (Fig.  4 A). Based on these identified metabolites, a metabolic pathway analysis (MetPA) was observed the dysregulation of the steroid hormone biosynthesis pathway was highly significant. Subsequent correlation analysis found that key steroid hormone-related metabolites, such as 5α-pregnane-3,20-dione and estrone glucuronide, showed significant interactions with other metabolic intermediates, suggesting their pivotal role in pathway disruption (Fig.  4 B). Fig. 4 A Pathway analysis of significantly different metabolites in DOR according to the KEGG pathway. The y-axis shows the name of each metabolic pathway corresponding to each bubble, while the x-axis displays the p -value. The larger the log10 ( p -value), the smaller the corresponding p-value. The larger the bubble, the greater the number of metabolites enriched in that pathway, with the bubble color indicating the magnitude of the p -value. B The metabolic correlation matrix plot illustrates the correlations between differential metabolites. Blue represents positive correlations, while red indicates negative correlations, with the intensity of the color reflecting the strength of the correlation. * p  < 0.05, ** p  < 0.01, *** p  < 0.001 A Pathway analysis of significantly different metabolites in DOR according to the KEGG pathway. The y-axis shows the name of each metabolic pathway corresponding to each bubble, while the x-axis displays the p -value. The larger the log10 ( p -value), the smaller the corresponding p-value. The larger the bubble, the greater the number of metabolites enriched in that pathway, with the bubble color indicating the magnitude of the p -value. B The metabolic correlation matrix plot illustrates the correlations between differential metabolites. Blue represents positive correlations, while red indicates negative correlations, with the intensity of the color reflecting the strength of the correlation. * p  < 0.05, ** p  < 0.01, *** p  < 0.001 Based on 17 differential metabolites, we developed a diagnostic model to distinguish between DOR and NOR patients and further identified key metabolic biomarkers. Eleven machine learning algorithms were used for classified multi-model comprehensive analysis to construct the optimal diagnostic model, and tenfold cross-validation was performed to evaluate model performance. The ROC curves presented in Fig.  5 demonstrate that the average AUC values ranged from 0.737 to 0.859, indicating a good predictive ability. The classif. glmnet model showed superior discriminatory performance, maintaining top rankings in both discrimination and precision-recall evaluations. After hyperparameter tuning, the model's training AUC improved to 0.886, with an AUPRC of 0.77, indicating strong class separation. Notably, the test set performance remained consistent (AUC = 0.874, AUPRC = 0.707), suggesting minimal overfitting. The glmnet model results identified 13 metabolites significantly contributing to DOR prediction. Ribothymidine demonstrated the highest negative contribution coefficient (−0.919), while acetamidopropanal demonstrated a positive contribution (0.479). Fig. 5 Development of the DOR diagnostic model by machine learning. A ROC curve of multi-model based on training sets. B Precision-recall curve of multi-model based on training sets. C Precision-recall curve for the glmnet model on training and test sets after hyperparameter tuning. D ROC curves for the glmnet model on training and test sets after hyperparameter tuning. E Feature coefficient value for the glmnet model Development of the DOR diagnostic model by machine learning. A ROC curve of multi-model based on training sets. B Precision-recall curve of multi-model based on training sets. C Precision-recall curve for the glmnet model on training and test sets after hyperparameter tuning. D ROC curves for the glmnet model on training and test sets after hyperparameter tuning. E Feature coefficient value for the glmnet model We further investigated the relationship between metabolites and ART outcomes. In the multivariate-adjusted generalized linear model analysis, 17 differential metabolites demonstrated significant associations with ART outcomes. Elevated levels of dUMP exhibited strong positive correlations with the number of oocytes retrieved, MII oocytes, and 2PN zygotes ( P  < 0.001). Conversely, increased acetylalanine levels showed significant negative correlations with multiple parameters, including cleavage rate, fertilization rate, high-quality embryo rate, and oocyte yield (Fig.  6 ). Notably, 5α-pregnane-3,20-dione displayed a unique bidirectional regulatory pattern: negatively associated with high-quality embryo rate while positively correlated with oocyte retrieval and the number 2PN zygotes. Furthermore, retinoyl beta-glucuronide selectively inhibited oocyte maturation, whereas 9,10-DHOME specifically reduced 2PN zygote formation. Elevated ribothymidine levels were significantly associated with improved pregnancy outcomes. For each standard deviation increase, the odds of clinical pregnancy and implantation increase by 1.02-fold (95% CI: 0.007–2.051) and 1.08-fold (95% CI: 0.101–2.10), respectively. Other metabolites did not show significant associations with pregnancy outcomes (Fig.  7 ). Fig. 6 The associations between follicular fluid metabolites and IVF outcomes based on GLM models. Forest plot of IVF outcomes between the two groups after adjusted for age, ovarian stimulation protocols, duration of infertility, and fertilization method: ( A ) cleavage rate, ( B ) fertilization rate, ( C ) high-quality embryo rate, ( D ) MII oocyte count, ( E ) number of 2PN zygotes, and ( F ) total oocytes retrieved. Data for count and proportional outcomes are presented as adjusted β (95% CI) Fig. 7 The associations between follicular fluid metabolites and pregnancy outcomes based on GLM models. Forest plot of IVF outcomes between the two groups after adjusted for age, ovarian stimulation protocols, duration of infertility, and fertilization method: ( A ) clinical pregnancy, ( B ) implantation, ( C ) live birth, and ( D ) miscarriage. Data for count and proportional outcomes are presented as adjusted β (95% CI) The associations between follicular fluid metabolites and IVF outcomes based on GLM models. Forest plot of IVF outcomes between the two groups after adjusted for age, ovarian stimulation protocols, duration of infertility, and fertilization method: ( A ) cleavage rate, ( B ) fertilization rate, ( C ) high-quality embryo rate, ( D ) MII oocyte count, ( E ) number of 2PN zygotes, and ( F ) total oocytes retrieved. Data for count and proportional outcomes are presented as adjusted β (95% CI) The associations between follicular fluid metabolites and pregnancy outcomes based on GLM models. Forest plot of IVF outcomes between the two groups after adjusted for age, ovarian stimulation protocols, duration of infertility, and fertilization method: ( A ) clinical pregnancy, ( B ) implantation, ( C ) live birth, and ( D ) miscarriage. Data for count and proportional outcomes are presented as adjusted β (95% CI) Mediation analysis was conducted to explore the mediating effect of differential metabolites in the causal pathway from DOR to ART outcomes. Mediation analysis indicates that the association between DOR and the number of retrieved oocytes is partially mediated by differential metabolites (Supplementary Table 1). A similar trend is observed in MII oocytes and 2PN zygotes. As shown in Fig.  8 , among the metabolites identified, 5α-pregnane-3,20-dione exerted the most pronounced mediating effect on oocyte developmental competence. Specifically, the proportion mediated by pregnane-3,20-dione was 60.2% for retrieved oocytes, 69.9% for MII oocytes, and 49.5% for 2PN zygotes. Ribothymidine also exhibited a substantial mediating influence, contributing 57.6% to retrieved oocytes, 62.8% to MII oocytes, and 48.8% to 2PN zygotes. Strikingly, ribothymidine was the only metabolite that demonstrated a significant mediation effect on pregnancy outcomes (Supplementary Table 1). However, we did not observe a significant mediating effect of differential metabolites on the associations between DOR and the cleavage rate, fertilization rate, high-quality embryo rate, clinical pregnancy, and biochemical pregnancy. Fig. 8 Mediation Analysis of follicular fluid differential metabolites linking DOR to ART outcomes (partial results displayed). The models were adjusted by age, ovarian stimulation protocols, duration of infertility, and fertilization method. Mediation Analysis of follicular fluid differential metabolites linking DOR to ART outcomes (partial results displayed). The models were adjusted by age, ovarian stimulation protocols, duration of infertility, and fertilization method.

A total of 946 women undergoing ART treatment at the Center of Reproductive Medicine, Fujian Maternity and Child Health Hospital were recruited for this study. After applying inclusion and exclusion criteria, 300 women were assigned to the normal ovarian reserve (NOR) group, and 126 to the DOR group. The study was approved by the Clinical Research and Ethics Committee of Fujian Maternity and Child Health Hospital (Ethics Number: 2023KYLLR01035). The detailed workflow is shown in Fig.  1 . Fig. 1 Study diagram Study diagram The inclusion criteria of the DOR group were as follows [ 18 ]: Anti-Müllerian hormone (AMH) < 1.1 ng/mL; basic follicle stimulating hormone (bFSH) ≥ 10 IU/L for two consecutive menstrual cycles; antral follicle count (AFC) of both ovaries < 5–7. Any one of the above three items can be diagnosed as DOR. In the control group, patients with normal ovarian reserve were treated with IVF/ICSI only because of male infertility or tubal factor infertility. The inclusion criteria of the control group were as follows: normal ovulation and regular menstrual cycles; normal FSH and luteinizing hormone (LH) levels. All subjects in this experiment should meet the following exclusion criteria: Endometriosis; history of ovarian surgery, polycystic ovary syndrome (PCOS), ovarian cysts, and other diseases that may affect the ovarian reserve function; chromosome abnormalities; endocrine diseases such as hyperthyroidism and diabetes. Follicular fluid was collected during oocyte retrieval, centrifuged after excluding contaminated samples, and the supernatant was stored at −80 °C. Isotope-labeled standards (ISTD) working solutions were prepared in the final concentration of 30,45,15,30,15,15 μg/mL for succinic acid-2,2,3,3-d4, cholic acid-2,2,3,4,4-d5, DL-tryptophan-2,3,3-d3, DL-methionine-3,3,4,4-d4, L-phenylalanine(ring-d5), and choline chloride (trimethyl-d9) in MeOH, respectively. Transfer an appropriate amount of the experimental sample and add 100 μL ISTD working solution plus 400 μL MeOH for protein precipitation. After vortex and centrifugation, the supernatant was vacuum-dried and reconstituted in 150 μL of 80% (vol/vol) H 2 O/MeOH, filter the supernatant with a 0.22 μm membrane and transfer into the detection bottle for LC–MS detection. Serial dilutions of quality control (QC) samples were prepared by mixing equal aliquots of all FF samples. The same procedures as those applied for sample analysis were used to dispose of and examine the QC samples. LC analysis was performed on a Thermo Vanquish UHPLC system (Thermo Fisher Scientific, USA) with an ACQUITY UPLC®HSST3 column (150 X 2.1 mm,1.8 μm) maintained at 40 °C, with a flow rate of 0.25 mL/min and an injection volume of 2 μL. For LC-ESI (+)-MS, the mobile phases consisted of (B2) 0.1% formic acid in acetonitrile (v/v) and (A2)0.1% formic acid in water (v/v), with the following gradient: 0 ~ 1 min, 2% B2; 1 ~ 9 min, 2% ~ 50% B2; 9 ~ 12 min, 50% ~ 98% B2; 12 ~ 13.5 min, 98% B2, 13.5 ~ 14 min, 98% ~ 2% B2; 14 ~ 20 min, 2% B2. For LC-ESI (-)-MS, the analytes were carried out with (B3) acetonitrile and (A3) ammonium formate (5 mM), with the following gradient: 0 ~ 1 min, 2% B3; 1 ~ 9 min, 2% ~ 50% B3; 9 ~ 12 min, 50% ~ 98% B3; 12 ~ 13.5 min,98% B3; 13.5 ~14 min, 98% ~ 2% B3, 14 ~ 17 min, 2% B3. Mass spectrometric detection was performed on an Orbitrap Exploris 120 (Thermo Fisher Scientific, USA) equipped with an ESI source in Full MS-ddMS2 mode. Parameters were as follows: sheath gas, 30 arb; aux gas flow,10 arb; spray voltage, + 3.5 kV/− 2.5 kV (ESI +/−); capillary temperature, 325 °C; MS1 range, m/z 100–1000; resolution, 60,000 FWHM; MS/MS resolution, 15 000 FWHM; data-dependent scans per cycle, 4; normalized collision energy, 30%; and dynamic exclusion, automatic. Raw data were processed using XCMS for feature detection, retention time alignment, and feature extraction. The metabolites were identified by accurate mass (< 30 ppm) and MS/MS data, which were matched with HMDB ( http://www.hmdb.ca ), massbank ( http://www.massbank.jp ), LipidMaps ( http://www.lipidmaps.org ), mzcloud ( https://www.mzcloud.org ), and KEGG ( http://www.genome.jp/kegg ). Data normalization was performed using the robust LOESS signal correction (QC-RLSC) method. After normalization, only ion peaks with relative standard deviations (RSDs) less than 30% in QC were kept to ensure proper metabolite identification. At the treatment cycle, women underwent specific IVF treatment protocols based on age, infertility diagnosis, and ovarian response: (a) long luteal-phase gonadotropin-releasing hormone (GnRH) agonist; (b) GnRH antagonist; and (c) others, such as a minimal stimulation IVF protocol. When more than two follicles matured, women underwent human chorionic gonadotropin (hCG) injection. Oocyte retrieval was performed 36 h after the hCG injection using transvaginal ultrasound guidance, and in vitro fertilization or intracytoplasmic sperm injection procedures were performed based on the male semen condition. All oocyte retrievals were to be autologous. After oocyte retrieval, the embryos were cultured in vitro for 72 h. If embryos were formed, 1–2 embryos were selected for fresh embryo transplantation, and the remaining embryos were cryopreserved. All procedures were conducted by qualified and experienced embryologists in our department's embryology laboratory. Fertilization was determined to be normal when two polar nuclei and two pronuclei (2PN) appeared in the fertilized oocyte 16–18 h after insemination. The 2PN cleavage zygotes are those normal fertilized oocytes that can continue to divide after fertilization. Embryos were classified as high-quality if they had 4–5 cells on day 2, 7–10 cells on day 3, and fragmentation less than 10%. Fertilization rate was calculated by dividing the number of normal fertilized oocytes by the number of MII oocytes. The cleavage rate was defined as the number of 2PN cleavage zygotes divided by the number of normal fertilized oocytes. The high-quality embryo rate was the ratio of the number of high-quality embryos to the number of 2PN cleavage zygotes. Implantation success was defined as a serum β-hCG concentration of more than 10 IU/L on day 14 after embryo transfer. Clinical pregnancy was defined as the presence of an intrauterine pregnancy confirmed by ultrasound 3–4 weeks after embryo transfer. Live birth was defined as the delivery of a live neonate on or after 28 weeks of gestation. Continuous variables were summarized as the median with interquartile range and were compared using the Wilcoxon rank-sum test. Categorical variables were expressed as numbers and percentages and were compared using the Chi-square tests or Fisher’s exact probability method. LASSO regression with tenfold cross-validation was used to determine the optimal regularization parameter (λ) and identify metabolites with non-zero coefficients. Differential metabolites were analyzed by constructing Orthogonal Partial Least Squares Discriminant Analysis (OPLS-DA) models, with variable importance in projection (VIP) values calculated for each metabolite. Differential Metabolites were identified using P   1.2, and VIP value > 1. Kyoto Encyclopedia of Genes and Genomes (KEGG) pathway enrichment analysis was performed on the differential metabolites. After feature selection, we constructed models using eleven machine learning algorithms, including extreme gradient boosting (XGB), random forest (RF), decision tree (DT), logistic regression (LR), K-nearest neighbor (KNN), support vector machine (SVM), naive Bayes (NBC), and multilayer perceptron (MLP). Data were randomly split into training and test sets (7:3), and models were trained and validated using tenfold cross-validation to prevent overfitting. Random hyperparameter tuning was performed to optimize model performance, evaluated by the area under the ROC curve (AUC) and the area under the precision-recall curve (AUPRC). Generalized linear models (GLMs) were used to evaluate the associations between differential metabolites and ART outcomes, adjusted for age, ovarian stimulation protocols, duration of infertility, and fertilization method. A log link was used for count outcomes, a logit link with binomial distribution for proportional outcomes, and a logit link with binary distribution for binary outcomes. Mediation analysis was conducted to explore the mediating effect of differential metabolites in the causal pathway from DOR to ART outcomes. The endpoints included the number of oocytes retrieved, MII oocytes retrieved, fertilization rate, and clinical pregnancy. We assessed the direct effect (ADE), average causal mediation effect (ACME), and the total effect of each metabolite on ART outcome. The statistical analyses and modeling process were conducted by using R version 4.0.5, and a two-sided P -value < 0.05 was regarded as statistically significant.

Decreased ovarian reserve (DOR) is a prevalent condition in female fertility, impacting an estimated 10–35% of women [ 1 ]. The pathogenesis of DOR remains multifactorial and incompletely understood, primarily characterized by a diminished follicular pool and compromised oocyte quality [ 2 ]. In recent years, women with DOR have constituted a growing proportion of assisted reproductive technology (ART) patients [ 3 ]. According to the Society for Assisted Reproductive Technology (SART), 26% of in vitro fertilization (IVF) cycles (approximately 181,536 cycles) were diagnosed with DOR [ 4 ]. However, IVF outcomes rely heavily on oocyte quality and embryonic developmental competence, making ART potentially suboptimal for addressing infertility in DOR patients. Even with IVF-ET treatment, the poor prognosis associated with DOR cannot be fully overcome, as these patients still face increased risks of miscarriage, poor ovarian response, reduced oocyte yield, higher cycle cancellation rates, and lower pregnancy and live birth rates [ 5 , 6 ]. Consequently, DOR remains a significant challenge in reproductive medicine. Follicular fluid (FF) constitutes a complex mixture of cytokines, proteins, steroids, and metabolic factors that collectively create an optimal niche for oocyte maturation [ 7 ]. Critical to this process is the selective uptake and accumulation of specific metabolites within the oocyte, which fundamentally governs its developmental competence [ 8 ]. Given that female reproductive success is profoundly dependent on this precisely coordinated follicular-oocyte crosstalk, disruptions in these metabolic interactions frequently precipitate systemic reproductive-metabolic dysfunction [ 9 ]. In this context, a deeper understanding of metabolic imbalances in FF may serve as an effective, non-invasive tool for assessing the oocyte microenvironment, thereby potentially enhancing oocyte development and improving pregnancy outcomes in IVF. This mechanistic insight positions follicular fluid metabolomic analysis as a transformative non-invasive strategy, enabling comprehensive evaluation of oocyte microenvironmental quality, identification of developmental competence biomarkers, and guidance of personalized interventions to optimize IVF outcomes. Therefore, FF has emerged as the biospecimen of choice for elucidating the pathophysiological mechanisms underlying DOR-associated infertility. Metabolomics is a powerful tool that enables the simultaneous quantification of diverse small molecules (< 1000 Da), including amino acids, fatty acids, carbohydrates, and other cellular metabolites [ 9 , 10 ]. Unlike transcriptomic and proteomic analyses, metabolomics provides more proximal functional readouts of phenotypic alterations by capturing dynamic biochemical endpoint products [ 11 , 12 ]. This methodology has become pivotal in deciphering disease mechanisms and discovering clinically actionable biomarkers. Emerging evidence indicates that fertility impairment reciprocally induces systemic metabolic dysregulation, establishing a bidirectional pathophysiology loop [ 13 ]. Consequently, integrated metabolomic profiling of follicular fluid in DOR patients could elucidate aberrant metabolic networks and signature metabolites, thereby delineating mechanistic pathways and revealing potential therapeutic targets. Despite the great potential of metabolomics, few studies have investigated metabolic changes in follicular fluid in DOR, and most are constrained by small sample sizes [ 14 , 15 ]. Moreover, most research focuses on targeted analysis of specific metabolites in follicular fluid, leaving the potential of untargeted metabolomics underexplored [ 16 , 17 ]. This study employed liquid chromatography-tandem mass spectrometry (LC–MS) metabolomics to analyze metabolic profiles of follicular fluid from women with DOR and those with normal ovarian reserve. The investigation further explored the associations between differential metabolites and IVF outcomes. The findings were anticipated to provide new insights for early identification of DOR, development of potential biomarkers, and precision medicine-based interventions for ovarian reserve disorders.

The metabolic profile of follicular fluid serves as a critical biological indicator of the oocyte microenvironment, with its dynamic changes directly reflecting both oocyte developmental potential and microenvironmental homeostasis [ 19 ]. Through untargeted metabolomic profiling, our study identified significant alterations in follicular fluid metabolic profiles between patients with DOR and NOR, which were consistent with published results [ 20 – 22 ]. The DOR group exhibited significant metabolic dysregulation, characterized by disrupted such as steroid hormone biosynthesis, pentose phosphate pathway, lysine degradation, arachidonic acid metabolism, and purine metabolism. Importantly, our findings suggested that such metabolic disturbances may compromise oocyte quality and embryonic developmental competence. To our knowledge, this is the first mediation analysis to establish a causal relationship between follicular fluid metabolites and ART outcomes. Furthermore, ribothymidine and 5α-pregnane-3,20-dione were identified as a potential novel biomarker for DOR. Numerous clinical data have demonstrated that DOR adversely affects reproductive outcomes in IVF [ 23 ]. Consistent with the literature [ 5 , 24 ], this research showed that young women with DOR experienced not only a reduction in the number of retrieved oocytes and MII oocytes but also decreased embryo quality and lower pregnancy rates. Several underlying causes have been connected to adverse reproductive outcomes. On one hand, the diminished oocyte pool in DOR patients results in fewer viable embryos per retrieval cycle, consequently limiting pregnancy success rates [ 24 ]. On the other hand, impaired oocyte developmental competence associated with follicular depletion in DOR may further exacerbate reproductive dysfunction [ 25 ], but the exact regulatory mechanisms governing this process remain unclear. Therefore, the precise pathophysiological mechanisms underlying these poor pregnancy outcomes remain to be elucidated. Follicular fluid microenvironment as a master regulator linking ovarian follicle dynamics, oocyte developmental potential, and post-fertilization embryonic fitness [ 26 ]. Previous studies have identified significant associations between serum metabolic profiles and IVF outcomes in women with DOR [ 22 ]. Emerging evidence positions targeted metabolomics investigations have revealed that DOR patients exhibit markedly reduced levels of 15 oxylipin metabolites in follicular fluid compared to controls [ 21 ]. Importantly, among these differentially expressed metabolites, nine demonstrated significant positive correlations with oocyte yield and fertilization rates, while one showed a strong association with high-quality embryo formation. Similarly, untargeted metabolomics analysis has also reported significantly reduced levels of tryptophan (TRP) and its indole metabolites in the DOR group, with the TRP concentration in FF showing a positive correlation with the usable embryo rate [ 27 ]. These findings demonstrated that metabolic dysregulation might constitute a key mechanistic contributor to the compromised reproductive performance observed in DOR patients. Our research provides new evidence that dUMP, ribothymidine, acetylalanine, and 5α-pregnane-3,20-dione exhibit significant associations with oocyte yield and embryo quality. However, it remains elusive as to how the above metabolites play a role in FF and which metabolic regulation channels influence ovarian reserve function and oocyte quality, which warrants further exploration. While follicular fluid metabolite disruption appears to impair reproductive outcomes in DOR, further research is needed to establish whether these changes are causative or consequential. In this study, mediation analysis initially identified 5a-Pregnane-3,20-dione, ribothymidine, D-ribose, and dUMP as key metabolic mediators linking DOR to impaired oocyte quality and embryonic development. This indicated that follicular fluid metabolic dysregulation might contribute to poor reproductive outcomes. However, only ribothymidine showed a significant mediatory effect on pregnancy outcomes. This likely reflects the multifactorial regulation of pregnancy establishment, where endometrial receptivity, immune-endocrine interactions, and maternal systemic factors collectively outweigh follicular fluid metabolic influences [ 28 ]. Therefore, we speculate that metabolic disturbances in follicular fluid are significant factors contributing to poor reproductive outcomes in DOR patients, likely by reducing both the quantity and quality of oocytes, potentially leading to clinical pregnancy and live birth failures. In the present study, ribothymidine was identified as a potential novel biomarker for DOR. Our results were consistent with previous studies [ 20 ] that reported reduced ribothymidine levels in the follicular fluid of patients with DOR. Besides that, ribothymidine mediated over 50% of the total effect on oocyte developmental competence. However, the precise mechanisms by which ribothymidine exerts its effects remain unclear, despite the significant implications of these associations. It is well established that protein synthesis was essential for oocyte proliferation, differentiation, and maturation. As a post-transcriptionally modified nucleoside in tRNA, ribothymidine may influence oocyte development by modulating the rate of protein synthesis. In addition, ribothymidine synthesis depends on S-adenosylmethionine (SAM) as the methyl donor [ 29 , 30 ]. There is evidence that disturbances in methylation levels can lead to imbalanced gene expression and affect oocyte meiosis and fertilization by disrupting epigenetic regulatory mechanisms [ 31 ]. It is speculated that ribothymidine may influence oocyte development through epigenetic mechanisms. Remarkably, ribothymidine not only exhibited a strong correlation with oocyte quality and embryonic development in this study but also showed high diagnostic potential for DOR. With an increasing number of studies on ribothymidine in oocytes, it may become a promising target for elucidating the pathogenesis and developing treatments for DOR. Steroid hormone biosynthesis is one of the key signaling pathways related to reproductive regulation [ 32 , 33 ]. Our study found that steroid hormone levels in the serum of DOR patients were significantly abnormal. Specifically, E 2 levels were increased while P and T levels were decreased, in line with previous studies [ 34 , 35 ]. Likewise, several steroid hormone-related metabolites in the follicular fluid were decreased, including 5α-pregnane-3,20-dione, estrone glucuronide and estrone sulfate. This finding was concordant with the results of the KEGG pathway analysis, which further confirmed that dysfunction in steroid metabolism is a key metabolic feature of DOR. The decreased P and T levels observed in patients with DOR may be attributed to the reduced number of AFC [ 36 ]. Interestingly, E 2 levels were elevated alongside an increase in FSH, which may be attributed to a dysregulation of the Hypothalamic-Pituitary-Ovarian (HPO) axis [ 37 ]. It has been reported that elevated FSH stimulates the aromatase-mediated conversion of androstenedione to estradiol [ 38 ], which may account for the increased serum E 2 and decreased T levels in DOR patients. Meanwhile, we observed a decline in conjugated estrogen metabolites such as estrone sulfate and estrone glucuronide, implying impaired estrogen metabolism and clearance, potentially leading to elevated levels of free estrogens and further disruption of the HPO axis feedback loop. It is noteworthy that polyamine metabolism was also associated with steroid hormone synthesis [ 39 ]. Animal studies shown that inhibition of polyamine synthesis in vivo significantly reduced serum P and LH levels [ 40 – 42 ]. Conversely, decreased spermidine content in granulosa cells led to elevated E 2 levels [ 41 ]. Our research further supported this finding, in which the polyamine metabolite N1-acetylspermine was found to be significantly downregulated alongside hormonal dysregulation. This suggested that disorders of polyamine metabolism may further exacerbate hormonal imbalance. Overall, these coordinated alterations suggested a close interplay between steroid hormone dysregulation and local follicular metabolic disturbances. Notably, the intrafollicular endocrine environment critically determines oocyte developmental potential. In this study, 5α-pregnane-3,20-dione was identified as the most critical metabolite that affected oocyte developmental competence. As a bioactive progesterone metabolite, 5α-pregnane-3,20-dione plays a critical role in follicular development and luteal function [ 43 ]. Interestingly, our results showed that 5α-pregnane-3,20-dione was positively correlated with the number of retrieved oocytes but negatively correlated with the rate of high-quality embryos. Progesterone is a product secreted by ovarian granulosa cells, and the concentration of progesterone increases with the development of follicles [ 44 ]. Previous studies found that progesterone exerted a dose-dependent effect on follicular development and ovulation [ 45 ]. Specifically, low levels promoted follicle growth in mice, whereas excessive accumulation could impair oocyte quality by activating progesterone receptor-mediated apoptotic pathways or disrupting mitochondrial function [ 45 – 47 ]. Related studies also indicated that excessive levels of progestin-like substances promoted premature luteinization of granulosa cells and increased oxidative stress, ultimately affecting mitochondrial function and chromatin stability in oocytes [ 47 – 49 ]. This suggested that 5α-pregnane-3,20-dione might influence both the quantity and quality of oocytes through a dose-dependent biphasic regulatory mechanism. Therefore, we proposed that 5α-pregnane-3,20-dione might serve as an important molecular marker affecting oocyte developmental potential. In the FF of DOR patients, we also detected a decrease in nucleotide intermediates such as dUMP and adenine. They are involved in DNA and RNA synthesis [ 50 ], and their downregulation may directly interfere with nucleic acid synthesis, chromosomal stability, and the developmental capacity of early embryos within oocytes [ 51 ]. It is suggested that physiological processes in the follicular fluid of DOR patients may be in a low-activity or dormant state. In contrast, we observed a significant upregulation of D-ribose levels in the FF of DOR patients, which exhibited a negative correlation with both the number of retrieved oocytes and the number of 2PN zygotes. D-ribose is a fundamental component of the ribose backbone in RNA molecules and also serves as a key intermediate in the pentose phosphate pathway (PPP) [ 52 ]. This metabolic pathway not only supplies the necessary five-carbon sugars for nucleotide synthesis but also generates nicotinamide adenine dinucleotide phosphate (NADPH), which is essential for maintaining cellular redox homeostasis. The elevation of D-ribose might reflect a compensatory response to oxidative stress or impaired nucleotide turnover. However, excessive accumulation of free D-ribose can lead to non-enzymatic glycation of proteins and nucleic acids, forming advanced glycation end-products (AGEs) [ 53 ]. These AGEs could lead to meiotic delays and impair mitochondrial integrity, spindle formation, and chromosome congression failure [ 54 ]. These findings suggest that energy metabolism dysfunction and nucleotide metabolic disturbances may be key pathological mechanisms underlying poor reproductive outcomes in DOR. While this study provides novel insights into the metabolic profile and molecular regulatory mechanisms underlying DOR, several limitations should be acknowledged. First, as a single-center investigation, our findings require validation through larger-scale, multicenter prospective studies to ensure robustness and generalizability. Second, while differential metabolites were identified, further functional validation through in vitro or in vivo experiments is still required to confirm the underlying mechanisms. Third, the present study did not collect detailed information on lifestyle factors, including smoking, alcohol consumption, and dietary habits. It has been acknowledged that these factors exert an influence on oxidative stress, neuroendocrine function, energy metabolism, and related pathways [ 55 , 56 ]. Consequently, the potential for unmeasured confounders to influence the observed metabolomic associations cannot be discounted. In subsequent studies, it is advised that these factors be considered in order to minimise confounding and enhance the identification of metabolic alterations in DOR. Lastly, the current analysis was confined to metabolomics, integrating multi-omics approaches, including genomics, transcriptomics, and proteomics, could offer a more comprehensive understanding of DOR pathogenesis. Future research should prioritize collaborative multicenter efforts, experimental validation, and multi-omics integration to further elucidate the molecular underpinnings of DOR.

This metabolomics study found significant metabolic disturbances in the follicular fluid of patients with DOR. These metabolic alterations likely contribute to DOR pathogenesis through specific metabolic pathways and may mediate adverse reproductive outcomes. Furthermore, we identified ribothymidine and 5α-pregnane-3,20-dione were potential biomarkers for DOR and are significantly correlated with ART outcomes. These findings provide new insights into DOR pathogenesis and establish a theoretical foundation for developing clinical diagnostic tools and therapeutic targets.

Supplementary Material 1. Supplementary Material 1.