Multi-omics pleiotropic association analyses reveal functionally relevant genes and druggable pathways for ovarian aging

Abstract

Background

Ovarian aging, marked by the gradual decline in both the number and quality of oocytes, significantly impacts women's reproductive lifespan and overall health. However, the biological mechanisms driving ovarian aging remain poorly understood and current treatment strategies are limited.

Results

We perform an integrative analysis using multi-omics summary data and genome-wide association studies for ovarian aging to identify molecular traits linked to ovarian aging. By applying Mendelian randomization and cross-omics association approaches, we prioritize key proteins, gene expressions, splicing events, and metabolites. Our analysis identifies conservation of key genes across species and cell types, with the mismatch repair gene MSH6 (MutS Homolog 6) emerging as a consistently prioritized candidate. Experimental validation shows that targeting DNA repair through PD-L1 (Programmed Death-Ligand 1) blockade may offer a potential therapeutic strategy to delay ovarian aging.

Conclusions

This study uncovers the multi-layered genetic and molecular architecture underlying ovarian aging. The identified molecular traits provide promising candidates for functional studies and suggest new avenues for developing therapies aimed at preserving ovarian function and preventing age-related decline.

Supplementary Information

The online version contains supplementary material available at 10.1186/s13059-026-04118-7.

Background

Ovarian aging is characterized by a decline in both the quality and quantity of oocytes, accompanied by hormonal change and diminished reproductive potential [1]. This process encompasses both physiological reproductive aging and pathological conditions, such as premature ovarian insufficiency (POI), defined as the loss of ovarian function before the age of 40 and affecting approximately 1% of women [2]. While physiological ovarian aging culminates in natural menopause, pathological ovarian aging can lead to infertility and reproductive disorders at much earlier ages. Beyond its impact on fertility, ovarian aging significantly affects women's long-term health, increasing the risk of conditions such as osteoporosis and cardiovascular diseases [3]. The mechanisms underlying ovarian aging are highly complex, involving genetic factors, immune dysregulation, inflammation, medications, and psychological influences, which collectively challenge efforts in prevention and treatment. To date, effective therapeutic options to delay ovarian aging remain limited.

Emerging evidence suggests that drug targets informed by human genetic studies have a higher likelihood of successful clinical translation [4, 5]. However, genome-wide association studies (GWAS) face inherent limitations, such as their inability to infer causal mechanism, which hinders the direct application of genetic data in drug development. Single-cell analysis is gaining prominence in identifying disease-associated cell types and can be combined with GWAS database to improve the accuracy and efficiency of therapeutic target discovery [6]. Age at natural menopause (ANM), which reflects the timing of ovarian follicle depletion, has been widely used as a measurable proxy phenotype for ovarian aging in large-scale genetic studies [7, 8]. Recent large-scale GWAS have identified hundreds of single nucleotide polymorphisms (SNPs) associated with ovarian aging, uncovering its genetic underpinnings [9]. Despite these advances, there remains a gap in comprehensively integrating genetic data with multi-omics layers for drug discovery related to ovarian aging.

Integrating GWAS data with multi-omics data can better understand the molecular mechanisms underlying genetic associations [10, 11]. To assess whether the effect of a genetic variant on a phenotype is mediated by molecular phenotypes, methods like SMR (summary-based Mendelian randomization) and HEIDI (heterogeneity in dependent instruments) have been developed [12]. Compared to other integrative methods, SMR and HEIDI features the ability to distinguish a pleiotropic model from a linkage model and has been widely used to identify the putative causal genes. OPERA, a Bayesian extension of SMR method, further enhances this approach by enabling joint analysis of GWAS summary statistics with xQTL summary statistics across multiple omics layers [13]. Both SMR and OPERA have been successfully applied to many complex traits to prioritize pleiotropically associated genes and proteins [14]. Recent advancements in high-throughput proteomic, transcriptomic, splicing genomics and metabolomics sequencing have enabled large-scale identifications of quantitative trait loci for proteins (pQTL), expression (eQTL), splicing (sQTL) and metabolites (mQTL) [15]. These multi-omics QTL datasets provide a valuable resource for integration into SMR and OPERA analyses, to decipher the molecular mechanisms underlying the genetic associations of ovarian aging.

In this study, we developed an analytical pipeline to integrate multi-omics data and prioritize molecular candidates that may inform future therapeutic research for ovarian aging (Fig. 1). We utilized QTL datasets and employed the SMR analysis to investigate the relationship between SNPs and ovarian aging through proteomics, transcriptomics, splicing omics, and metabolomics. Multi-omics integration techniques, including OPERA analysis and enrichment analyses were applied to uncover multi-omics evidence and explore comprehensive biological pathways related to ovarian aging. Subsequently, single-cell analysis was performed to identify conserved or species-specific prioritized genes in ovarian tissues across multiple species. Among the numerous identified target genes, we prioritized candidate genes based on functional analyses. Experimental validation was further conducted to evaluate their therapeutic potential. Notably, functional enrichment analysis based on multi-omics identifications highlighted DNA repair as the most significant pathway involved. By integrating single-cell data, we identified the highly conserved gene MSH6 as a promising candidate and PD-L1 blockade as a potential strategy, emphasizing their potential roles in addressing ovarian aging. This study demonstrates the power of integrating multi-omics data to translate human genetic insights into prioritized candidate targets for ovarian aging, providing a foundation for future research into potential therapeutic interventions.

Fig. 1.

A summary of the study design and workflow. The flow chart outlines the process for the selection of genetic variants and the corresponding analytical methods employed in the study. This figure was partly generated using Servier Medical Art, provided by Servier, licensed under a Creative Commons Attribution 3.0 unported license

Results

Integrative analysis of protein with ovarian aging identifies functionally relevant proteins

To prioritize proteins associated with ovarian aging, we applied the SMR approach to test the pleiotropic associations using ovarian aging GWAS summary data and pQTL data in large data sets from peripheral blood (Methods). Because ovarian aging cannot be directly measured in large populations, we used age at natural menopause (ANM)—defined as the age at the last spontaneous menstrual period followed by at least 12 consecutive months of amenorrhea—as a validated proxy phenotype (Methods). The GWAS summary data for ANM were sourced from the Reprogen consortium, comprising 201,323 individuals of European ancestry across 34 cohorts [9]. The summary statistics of SNPs on protein were from a pQTL analysis of 4,719 proteins in the deCODE project (n = 35,559). After quality control (Methods), we retained 1,814 proteins with at least a pQTL and included only SNPs available and with consistent alleles in the data sets used. Testing the associations between 1,814 proteins and ovarian aging revealed nine significant proteins at P_SMR < 2.75 × 10⁻⁵ correcting for 1,814 tests and P_HEIDI > 0.01. We used the threshold of 0.01 for HEIDI test as suggested in the justification in Wu et al. 2018 [16]. These identified proteins (encoded by genes such as GMPR, TMEM106A, DNER, FBLN1, GDI2, PLXNA1, ADH6, ADH5, and LY9) (Fig. 2a; Additional file 1: Table S1A). Among these, FBLN1, TMEM106A, PLXNA1, and DNER are associated with apoptosis and inflammation, while FBLN1, TMEM106A, and DNER specifically linked to apoptosis. Additionally, GDI2 was related to telomere length [17]. Notably, ADH5 emerged as an intriguing candidate, associated with aging through its metabolite formaldehyde, which implicates mechanisms such as mitochondrial-induced oxidative stress and DNA damage repair [18]. To validate these identified proteins, we performed the replication analysis using the pQTL summary data from the INTREVAL project and the GWAS summary data for ovarian aging from the UK Biobank separately [19, 20]. After quality control, there were 2 and 9 proteins available for replication in the respective datasets. Following multiple-testing correction of 2 tests for INTREVAL and 9 tests for UK Biobank, 2 and 6 were successfully replicated with P_SMR < 5.56 × 10⁻³ & P_HEIDI > 0.01 and P_SMR < 5.56 × 10⁻³ & P_HEIDI > 0.01(Additional file 1: Table S1A).

Fig. 2.

SMR analysis results of pQTL, eQTL, sQTL, and mQTL on ovarian aging outcomes. a Volcano plot illustrating the effect of proteins on ovarian aging from the SMR analyses, with horizontal lines representing p-values after Bonferroni correction. b Volcano plot illustrating the effect of gene expression on ovarian aging from the SMR analyses, with horizontal lines representing p-values after Bonferroni correction. Semi-transparent dots with blue/red background show SMR results of eQTL from ovarian tissue. c Volcano plot illustrating the effect of gene splicing on ovarian aging from the SMR analyses, with horizontal lines representing p-values after Bonferroni correction. Semi-transparent dots with blue/red background show SMR results of sQTL from ovarian tissue. d Volcano plot illustrating the effect of metabolites on ovarian aging from the SMR analyses, with horizontal lines representing p-values after Bonferroni correction

Integrative analysis of gene expression and splicing with ovarian aging identifies new susceptibility genes

For transcriptome-wide analysis of ovarian aging, we conducted SMR analysis using the same GWAS summary statistics from the ReproGen consortium, and cis-eQTL data derived from the eQTLGen project (n = 31,684), comprising summary statistics for 19,942 genes (Methods). After quality control, including allele harmonization and restricting to genes with at least one significant cis-eQTL at P_eQTL < 5e-8, we retained 15,724 genes for the SMR analysis. We identified 106 genes significantly associated with ovarian aging, based on a Bonferroni-corrected significance threshold, P_SMR < 3.18 × 10⁻⁶ (adjusting for 15,724 tests) and P_HEIDI > 0.01. Among these, PNP, PGAP3, and OCIAD1 were the top candidates (Fig. 2b; Additional file 1: Table S2A). Functional categorization of these genes highlights their involvement in key mechanisms of ovarian aging, including DNA repair and maintenance (e.g., XRCC5, FANCM, RAD18, and RAD52), metabolic regulation (e.g., SIRT1, PGAP3), cell cycle regulation (e.g., CCT6A, PPP1CB), stress response (e.g., TNXA, IL9RP3), and transcriptional regulation (e.g., CWF19L1, L3MBTL3, LINC02728, and LCT-AS1). For instance, SIRT1 is well-studied for its role in oxidative stress, DNA damage repair, telomere shortening, and apoptosis, which significantly impact ovarian aging [21]. High consistency in effect sizes was observed in replication analysis using GWAS data of ovarian aging from UK Biobank. Of the 69 genes that passed quality control and were eligible for replication, 65 genes were replicated at P_SMR < 7.25 × 10^–4 (Bonferroni-corrected for 69 tests) and P_HEIDI > 0.01 (Additional file 1: Table S2B).

To investigate the potential contribution of alternative splicing in ovarian aging, we performed SMR analysis by integrating GWAS summary data from the Reprogen consortium with the largest available sQTL dataset for 12,794 splicing genes derived from BrainMeta v2 (n = 2,865; Methods). After quality control, we included 59,444 splicing events in the analysis. Our analysis identified 233 splicing events significantly associated with ovarian aging, spanning 91 unique genes, based on a Bonferroni-corrected thresholds of P_SMR < 8.41 × 10^–7 (adjusted for 59,444 tests) and P_HEIDI > 0.01. Notably, FNDC4, HELQ, and DIDO1 emerged as top candidate genes (Fig. 2c; Additional file 1: Table S3A). Similar to the genes identified using eQTL, the sQTL-identified genes are linked to processes such as cellular metabolism, DNA repair, epigenetic regulation, and stress responses. For example, RAD54L, FANCM, POLG, and HELQ contribute to DNA repair and genomic stability, while FNDC4, FTO, and HSD17B6 are associated with metabolism and energy regulation. Our identifications also include stress-response genes such as DDIT4L, DDAH2 and NBR1, which are critical for autophagy and protein degradation. Of the 233 identified associations, 79 passed quality control and were available in replication data of ovarian aging from UK Biobank, and 53 were successfully replicated at P_SMR < 6.32 × 10^–4 (Bonferroni-adjusted for 79 tests) and P_HEIDI > 0.01 (Additional file 1: Table S3B).

Integrative analysis of metabolite with ovarian aging reveals key metabolic features

To investigate metabolic associations with ovarian aging, we performed SMR analysis by integrating GWAS summary data from the Reprogen consortium with mQTL data from Canadian Longitudinal Study on Aging (CLSA) (N = 8,299), encompassing 1,091 metabolites and 309 metabolite ratios. After quality control, we retained 1,009 metabolic features with robust mQTL signals for SMR analysis (Methods). The SMR analysis identified 13 significant features associated with ovarian aging (P_SMR < 4.95 × 10⁻⁵, Bonferroni-adjusted for 1,009 tests; P_HEIDI > 0.01). These features spanned nucleotide, lipid, peptide, carbohydrate, and vitamin pathways. We performed two independent replication analyses using mQTL data from IEU OpenGWAS and ovarian aging GWAS data from the UK Biobank separately. After quality control, 8 and 13 candidate metabolite-trait associations were tested in each analysis. Among these, 3 and 7 replicated at genome-wide significance thresholds (P_SMR < 6.25 × 10⁻³, Bonferroni-adjusted for 8 tests; P_SMR < 3.85 × 10⁻³, Bonferroni-adjusted for 13 tests) and P_HEIDI > 0.01 (Fig. 2d; Additional file 1: Table S4A).

Since the sample size of metabolites is limited, we selected 40 metabolites with P_SMR < 0.001 and P_HEIDI > 0.01 retrievable from the Kyoto Encyclopedia of Genes and Genomes (KEGG) and Small Molecule Pathway Database (SMPDB) databases for metabolic enrichment analysis (Additional file 1: Table S4B). These analyses revealed 7 pathways in KEGG and 7 in SMPDB. Among them, alpha-linolenic acid and linoleic acid metabolism was the only pathway that remained significantly enriched after multiple testing correction (FDR = 8.81 × 10⁻⁶) (Additional file 1: Table S4C, Additional file 2: Fig. S1). Key metabolites included eicosapentaenoic acid, tetracosahexaenoic acid, docosahexaenoic acid, adrenic acid, and stearidonic acid. Notably, previous studies have suggested protective effects of arachidonic acid against aging and ovarian aging [22]. Several positive metabolites were linked to the SNP rs131805 in the SCO2 gene, which is involved in mitochondrial function and copper metabolism. Although a direct association between this SNP and ovarian aging has not been established, its role in oxidative stress due to mitochondrial dysfunction may contribute to ovarian aging [23]. Other notable metabolites include gamma-glutamyltyrosine (associated with oxidative stress and antioxidant defense mechanisms), 2'-deoxyuridine (linked to DNA repair and synthesis), citrate (involved in energy metabolism and mitochondrial function), and N-methyl-2-pyridone-5-carboxamide (associated with cellular stress responses and energy metabolism).

Identification of tissue-shared and tissue-specific genes using tissue-specific gene expression and splicing QTL data

To investigate tissue-specific regulatory signals associated with ovarian aging, we integrated ovary-derived molecular QTL data from the GTEx project (v8; N = 180). This included cis-eQTL data for 2,375 genes and sQTL for 3,880 splicing events. All QTL data underwent standardized quality control procedures (Methods). SMR analysis revealed 19 ovarian aging-associated genes (P_SMR < 2.11 × 10⁻⁵, Bonferroni-adjusted for 2,375 tests) and 22 splicing events (P_SMR < 1.29 × 10⁻⁵, Bonferroni-adjusted for 3,880 tests), all surpassing P_HEIDI > 0.01 thresholds. Among these, 3 genes and 7 splicing events overlapped with findings from discovery analysis (Fig. 2b, c; Additional file 2: Fig. S2 and Additional file 2: Fig. S3). Notably, NBR2, PLA2G4B and CBY1, were significantly associated with ovarian aging in both transcriptomic and splicing omics within ovarian tissue, highlighting their potential dual roles in ovarian aging. NBR2, adjacent to the tumor suppressor gene BRCA1, is a glucose starvation-induced lncRNA that interacts with AMPK to regulate energy metabolism and proliferation [24]. PLA2G4B, part of the arachidonic acid pathway, converts phosphatidylcholine (PC) into arachidonic acid, a precursor for key pro-inflammatory and anti-inflammatory mediators, and is linked to monocyte aging [25]. Additionally, CBY1 serves as a β-catenin-associated antagonist in the Wnt pathway, impacting cell proliferation [26]. Future availability of larger reproductive tissue-specific QTL datasets will help further refine and validate these tissue-specific associations.

A comprehensive circular Manhattan plot (Additional file 2: Fig. S2) illustrates the multi-omics associations, emphasizing the interconnected biological signals associated with ovarian aging. These findings underscore the coexistence of shared and tissue-specific effects, as well as omics-specific and omics-shared effects, reinforcing the importance of integrative approaches to provide insights into the molecular pathways of ovarian aging.

Identification of molecular mechanism through joint analysis of multi-omics QTL data

To uncover loci with pleiotropic associations across multiple molecular phenotypes, we conducted a joint analysis of multi-omics xQTL data (eQTLs, sQTLs, and pQTLs) and ovarian aging GWAS using OPERA. Due to the limited availability of metabolites with mQTLs, these were excluded from the analysis. We identified significant associations with a posterior probability of association (PPA) > 0.9 and P_HEIDI > 0.01, resulting in 145, 50, and 2 SNPs linked to one, two, and three omics types, respectively (Additional file 1: Table S5A-C). Notably, two SNPs (rs111637825 and rs799903) showed overlapping associations across all three omics layers. The variant rs111637825 colocalized with eQTL, sQTL, and pQTL signals for DHX58, suggesting coordinated regulation of gene expression, splicing, and protein abundance at this locus (joint PPA = 0.974633) (Fig. 3). The DHX58 gene (also known as LGP2) encodes an RNA helicase that is crucial for the immune response. It initiates downstream inflammation cascades, leading to the production of type I interferons (IFN) and inflammatory cytokines [27]. This suggests DHX58 may be relevant to ovarian aging–related molecular processes. The second variant, rs799903, was associated with multiple molecular phenotypes across nearby genes (joint PPA = 0.999994), including pQTL signals for TMEM106A, eQTL signals for CCDC200, and sQTL signals for NBR1, indicating that shared genetic variation at this locus may exert pleiotropic regulatory effects on different molecular layers.

Fig. 3.

Results of OPERA analysis. The locus zoom plot illustrates the consistent results from the OPERA analysis across pQTL, eQTL, and sQTL omics data. The plot highlights the pleiotropic associations between multiple molecular phenotypes and ovarian aging, showing overlapping significant loci from each omics layer

Prioritization of MSH6 expression associated with ovarian aging through single-cell analysis

After identifying 219 ovarian aging-related genes through multi-omics analyses, we next investigated their potential biological functions using Gene Ontology (GO) analysis. The top 20 enriched pathways, ranked by ascending p-value, are displayed in Fig. 4a. To evaluate cross-omics convergence, we examined pathway overlap across pQTL-, eQTL-, sQTL-, and mQTL-derived gene sets. Several biological processes were jointly enriched across three omics layers. Genes supported by pQTL, eQTL, and sQTL analyses collectively participated in pathways such as Purine salvage, Tyrosine metabolism, negative regulation of phosphorylation, and interstrand cross-link repair. Moreover, genes supported by eQTL, sQTL, and mQTL analyses converged on pathways including establishment of protein localization and the p53 transcriptional gene network, indicating metabolomic involvement in cellular stress-response processes. Among these, the DNA repair pathway emerged as the most significantly associated with ovarian aging. Specifically, we identified 24 DNA repair-related genes, including 17 from eQTL data (e.g., MSH6, MC1R, RAD52) and 16 from sQTL data (e.g., MSH5, POLG, RAD54L; Fig. 4b).

Fig. 4.

Results of single-cell analysis in transcriptomics. a Bar charts display the top 20 findings from the multi-omics co-enrichment analysis, conducted using Metascape (https://metascape.org), highlighting key molecular pathways involved in ovarian aging. b The network diagram illustrates the most significantly enriched pathway, DNA repair, showing the upstream and downstream genes associated with the genes identified in our research. Interactions with the highest confidence (minimum required interaction score greater than 0.9) are presented. The dot background represents eQTL data, while the diagonal stripe background represents sQTL data. Solid outlines denote ovary tissue, while dotted lines represent blood or brain tissue. c, d UMAP maps of single-cell data from human ovarian tissues. The bubble diagram shows the expression of eQTL identified through pathway analysis in various cell types, including granulosa cells (GCs) and oocytes, across different age groups in human ovarian tissue. e, f UMAP maps of single-cell data from monkey ovarian tissues. The bubble diagram shows the expression of eQTL identified through pathway analysis in various cell types, including GCs and oocytes, across different age groups in monkey ovarian tissue. g, h UMAP maps of single-cell data from mouse ovarian tissues. The bubble diagram shows the expression of eQTL identified through pathway analysis in various cell types, including GCs and oocytes, across different age groups in mouse ovarian tissue. i Gene expression of MSH6 in human ovarian GCs from patients with DOR (n = 40) and the control group (n = 40). j, k Correlation analysis of MSH6 expression in human ovarian GCs with serum AMH levels (n = 74) and AFC (n = 73)

We subsequently assessed whether eQTL have determinant effects on gene expression.

by examining the expression changes of 11 identified genes identified through eQTL analysis across different cell types in ovarian tissues using single-cell RNA-seq data. We included three published single-cell RNA sequencing (scRNA-seq) datasets (GSE255690, GSE130664, GSE232309) that were from humans, monkeys, and mice respectively, and comprised both young and old groups. We defined all ovarian cells into 7 major cell types, including granulosa cells (GCs), oocytes, theca and stroma cells(T&S), smooth muscle cells (SMCs), endothelial cells (Endo), monocytes (Mono), natural killer cells and T lymphocytes (NK&T) based on well-established markers (Methods)(Fig. 4c, e, g). Most genes identified by eQTL analysis did not display cell type specific expression. We next paid specific attention to the expression dynamics of the 11 genes in GCs and oocytes during aging (Fig. 4d, f, h). MSH6 (MutS homolog 6) was the only one conserved across humans, monkeys, and mice, showing consistent and significant downregulation in GCs with aging (Fig. 4d, f, h). MSH6 is a crucial component of the DNA mismatch repair family. It recruits other repair complexes, such as MutL and MutH dimers, to initiate repair signaling and correct base mismatches arising during DNA replication [28]. The observed downregulation of MSH6 in aged GCs aligns with previous eQTL findings that linked MSH6 expression to later menopause onset, suggesting its potential role in promoting ovarian longevity [28]. In contrast, other candidate genes, such as FANCM and RBBP8, did not exhibit consistent expression trends across these species. Similarly, no genes showed consistent expression changes in oocytes across the three species (Fig. 4d, f, h).

In addition, we found that its expression was very concentrated in the ovary (Additional file 2: Fig. S3). To further validate the specific expression patterns of MSH6 in GCs during aging, we analyzed MSH6 expression in GCs from women with diminished ovarian reserve (DOR). Baseline characteristics of DOR and control subjects are summarized in Additional file 1: Table S6. Compared to controls, women with DOR exhibited significantly lower serum anti-Müllerian hormone (AMH) levels, reduced antral follicle count (AFC), and elevated follicle-stimulating hormone (FSH) levels. Notably, MSH6 expression was markedly decreased in GCs of DOR patients. Correlation analysis revealed a positive association between MSH6 expression in human ovarian GCs and both serum AMH levels and AFC (Fig. 4i, j, k). Further analyses showed that MSH6 is broadly expressed in GCs during early follicular development, with peak expression at the antral follicle stage, followed by a sharp decline at the preovulatory stage (Additional file 2: Fig. S4a). This pattern suggests a potential regulatory role of MSH6 in folliculogenesis. Taken together, these findings demonstrate a decline in MSH6 expression in GCs during ovarian aging and highlight its potential role in modulating GCs function and the ovarian aging process.

MSH6 deficiency induces senescence in human ovarian granulosa cells

To investigate the role of MSH6 in maintaining GCs function, we knocked out the MSH6 gene in the human GCs line KGN (Fig. 5a). Senescence-associated β-galactosidase (SA-β-gal) activity, a marker of cellular senescence, was significantly elevated in MSH6 knockout (MSH6-KO) cells compared to controls, indicating that MSH6 depletion promoted GCs senescence (Fig. 5b). Senescent cells are characterized by the senescence-associated secretory phenotype (SASP). In MSH6-KO cells, the expression of SASP factors was markedly increased, including C-X-C motif chemokine ligand 1 (CXCL1), C-X-C motif chemokine ligand 2 (CXCL2), C-X-C motif chemokine ligand 8 (CXCL8), intercellular Adhesion Molecule 1 (ICAM1), and growth differentiation factor15 (GDF15) (Fig. 5c). Transcriptome analysis of MSH6-KO cells revealed 188 upregulated and 159 downregulated differential expressed genes (DEGs) compared to controls (Additional file 2: Fig. S4b). Gene Set Enrichment Analysis (GSEA) showed that DNA replication and repair processes were strongly suppressed, while immune and inflammatory pathways were activated in MSH6-KO cells (Fig. 5d). KEGG pathway analysis of DEGs corroborated these findings (Additional file 2: Fig. S4c and d). Consistent with these transcriptomic changes, we observed an increase in γ-H2AX, a marker of DNA damage, in MSH6-KO cells (Fig. 5e and Additional file 2: Fig. S4e). DNA damage triggers p53 and p21, leading to cell cycle arrest and cellular senescence.

Fig. 5.

MSH6 knockout induced senescence of human ovarian granulosa cells. a Western blotting showing the efficacy of MSH6 knockout in the cells. b Representative images of SA-β-gal staining in MSH6-KO cells (scale bar: 500 μm) (left) and quantitative analysis (right), five randomly selected fields were examined for each sample. c Relative mRNA expression of SASP genes in the control and MSH6-KO cells. Fold changes were normalized to ACTB. d Gene set enrichment analysis (GSEA) terms of downregulated (blue) and upregulated (red) gene sets in MSH6-KO cells. e Effect of MSH6-KO on the expression of γ-H2AX, p53, p21, Bcl2, Bax, p-NF-κB/NF-κB, and p-ERK/ERK. Tubulin and vinculin were used as reference genes. f Cell proliferation curves of MSH6-KO and control KGN cells. The cell index was calculated using the RTCA software. g Flow cytometric analysis showing the distribution of cell cycle phases in control and MSH6-KO cells. h Western blotting showing PD-L1 expression in control and MSH6-KO cells. i Single-cell transcriptomic analysis of PD-L1 expression in human ovaries and granulosa cells (GCs) at three representative stages: young (18–28 years), middle (36–39 years), and old (47–49 years). j Western blotting (left) and gray value analysis (right) of PD-L1 protein expression in ovarian GCs from control (n = 6) and DOR patients (n = 6). k γ-H2AX expression in MSH6-KO cells after 24-h treatment with atezolizumab (0, 2.5 and 5 μg/mL). l RTCA analysis of proliferation in MSH6-KO cells treated with atezolizumab (0, 2.5 and 5 μg/mL). Data are expressed as Mean ± SEM. Unless otherwise stated, n = 3 independent biological replicates were performed for each experiment. P value was determined by two-tailed student’s t-test (i), one-way ANOVA (b, c, e, g, h, k) and two-way ANOVA (f, l). ns, not significant; * P value < 0.05; ** P value < 0.01; *** P value < 0.001

Protein levels of p53 and p21 significantly elevated in MSH6-KO cells (Fig. 5e and Additional file 2: Fig. S4f and g). Furthermore, MSH6 deficiency impaired cell proliferation and induced S-phase cell cycle arrest (Fig. 5f and g). In senescent cells, anti-apoptotic mechanisms are often activated to resist cell death. We found that the Bcl2/Bax ratio was significantly increased in the MSH6-KO group, indicating enhanced anti-apoptotic activity (Fig. 5e and Additional file 2: Fig. S4h).

Previously, we demonstrated that extracellular signal-regulated kinase 1/2 (ERK1/2) and nuclear factor kappa B (NF-κB) signaling pathways mediate proinflammatory responses in human ovarian GCs [29]. In MSH6-KO cells, we observed enhanced phosphorylation of ERK1/2 and NF-κB, further supporting the activation of inflammatory pathways (Fig. 5e and Additional file 2: Fig. S4i and j). Taken together, these findings indicate that MSH6 deficiency contributes to cellular senescence in GCs by inducing DNA damage, activating inflammatory pathways, and promoting anti-apoptotic mechanisms.

PD-L1 blockade mitigates granulosa cell senescence induced by MSH6 deficiency

Clinical evidence links mismatch repair (MMR) gene depletion to increased expression of the immune checkpoint protein programmed death-ligand (PD-L1) and improved efficiency to PD-1/L1 inhibitors [30–32]. Supporting these findings, we observed that PD-L1 expression was markedly upregulated in human GCs following MSH6 knockout (Fig. 5h). Notably, single-cell transcriptomic analysis of human ovaries further revealed a significant increase in PD-L1 expression in ovarian tissue and GCs with aging (Fig. 5i). Similarly, GCs from patients with DOR exhibited significantly elevated PD-L1 protein expression (Fig. 5j).A recent study found that blocking the PD-L1/PD-1 axis in naturally aged mice mitigates senescence-related phenotypes [33]. In line with this, we found Atezolizumab, a monoclonal antibody targeting PD-L1, partially ameliorated the MSH6 deficiency-induced senescence. This was evidenced by reduced DNA damage (Fig. 5k), enhanced cell proliferation (Fig. 5l), and decreased SA-βgal activity (Additional file 2: Fig. S4m). These results provide preliminary evidence supporting that PD-L1 blockade holds potential as a therapeutic strategy to counteract ovarian aging by alleviating GCs senescence.

Evaluation of drug targets and pathway identification

In addition to the PD-L1 blockade, we also explored other potential druggable targets by investigating the interactions between ovarian aging related proteins and existing drug targets currently available on the market. We analyzed the protein–protein interaction (PPI) network of ovarian aging-related proteins and their interactions with existing drug targets. We identified 5 priority proteins (GMPR, GDI2, PLXNA1, ADH5, and LY9) associated with 2 existing drug for ovarian aging (Additional file 1: Table S1B and Additional file 2: Fig. S5). Notably, most of the drug targets were linked to dasatinib, a medication primarily approved for the treatment of Philadelphia chromosome-positive (Ph +) leukemia [34]. We also investigated the druggability of 9 circulating proteins identified in SMR analysis using drug databases. None of these proteins were found to be established drug targets for ovarian aging. However, six of these proteins (GMPR, DNER, GDI2, ADH6, ADH5, LY9) have been targeted for drug development for other diseases (Additional file 1: Table S1C). For example, GMPR antagonist DARATUMUMAB is a monoclonal antibody primarily used for the treatment of multiple myeloma. It exerts its effect by targeting the CD38 protein [35]. In the future, it would be interesting to test whether therapeutic targeting of these proteins would delay ovarian aging.

Discussion

Through the integration of GWAS and multi-omics data, we identified several potential drug targets and biomarkers, underscoring the effectiveness of our comprehensive analytical framework. Through multi-omics pathway analysis, DNA repair emerged as the most significant contributor to ovarian aging. Using single-cell analysis of ovarian tissues and in vitro functional experiments, we pinpointed MSH6 as a key gene involved in this process. Importantly, our findings revealed that MSH6 could serve as a promising therapeutic target. Leveraging this insight, we proposed a potential intervention strategy using PD-L1 blockade to mitigate ovarian aging, supported by evidence that this approach alleviates senescence-related phenotypes in GCs. These findings highlight the translational potential of targeting MSH6 and associated pathways for clinical applications aimed at delaying ovarian aging.

Our analysis highlights DNA repair as the most significant pathway associated with ovarian aging, emphasizing its crucial role in correcting DNA damage caused by both endogenous and exogenous carcinogens, thus maintaining genomic stability [36, 37]. In tissues, the accumulation of DNA damage due to ineffective repair can lead to cellular senescence and functional decline, accelerating biological aging [9, 38, 39]. Consequently, abnormalities in DNA repair mechanisms may influence the aging process. MSH6, a member of the DNA mismatch repair MutS family, encodes a protein that forms a heterodimer with MSH2 to recognize and correct mismatched bases, nucleotide insertions, and deletions during DNA replication. Previous studies using whole-exome sequencing have identified missense mutations in MSH2, MSH5, MSH6, and MLH1 in patients with primary ovarian insufficiency (POI) [40], and GWAS analysis has suggests that mutations in MSH6 might be associated with earlier menopause [41]. Our findings demonstrate that MSH6 is enriched in ovarian GCs and exhibits a significant decline in expression with aging, a trend conserved across humans, non-human primates, and mice. This conservation underscores MSH6's potentially important role in involved in ovarian aging–related processes. We further investigated the functional consequences of MSH6 loss in GCs. MSH6-KO resulted in typical senescence-associated phenotypes, including elevated secretion of SASP factors and increased β-galactosidase activity. Loss of MSH6 triggered a DNA damage response that activated the p53/p21 pathway, leading to cell cycle arrest. These results suggest that MSH6 may play a role in preventing premature senescence in GCs, highlighting its potential as a therapeutic target in ovarian aging. However, the upstream regulatory mechanism underlying MSH6 downregulation in aging ovaries warrant further investigation to fully understand the molecular drivers of this process.

Recent advances in single-cell and spatial transcriptomics have provided additional insights into the cellular and spatial organization of ovarian aging. For example, a recent study by Wu et al. combined single-cell and spatial transcriptomic analyses to characterize the spatiotemporal landscape of human ovarian aging and identified FOXP1 as a transcriptional regulator of ovarian senescence through repression of CDKN1A [42]. In the present study, we incorporated the single-cell dataset generated in that work to examine candidate gene expression across ovarian cell types. Our analysis revealed that MSH6 shows a conserved age-associated downregulation in granulosa cells across human, primate, and mouse datasets, a pattern that is also consistent with the differential expression results reported in Wu et al., where MSH6 is significantly downregulated in aged granulosa cells. While the study by Wu et al. highlights spatially organized senescence-associated transcriptional programs, our work focuses on prioritizing genetically supported molecular drivers of ovarian aging through integrative multi-omics analyses. Future studies integrating genetic prioritization with spatial transcriptomic approaches will be valuable to further resolve the spatial organization of MSH6-associated aging programs within ovarian follicles.

Additionally, we identified several genes previously linked to ovarian aging and age-related diseases. Sirtuin 1 (SIRT1), an NAD + -dependent deacetylase, is a key regulator of various cellular processes, including metabolism, mitochondrial homeostasis, autophagy, DNA repair, apoptosis, and the balance between oxidative and antioxidative states, as well as aging [43, 44]. Resveratrol, a small polyphenol identified as a SIRT1 activator, shows promise as an anti-aging therapeutic by enhancing antioxidant capacity, suppressing inflammation, and extending the lifespan of lower organisms. However, further experiments and clinical trials are necessary to validate its potential as a drug target [45].

Our SMR analysis also uncovered novel, high-confidence associations between genes and ovarian aging. Proinflammatory stress, a key molecular event in ovarian aging, was highlighted by these findings [46]. One such gene, DNER, a novel non-canonical Notch ligand identified in proteomics, was shown to promote macrophage secretion of IFNγ by activating the Notch1-NFκB signaling axis, contributing to ovarian aging through pro-inflammatory stress [47]. In addition to the identified genes, non-coding RNAs also play an important role in fine-tuning gene expression during ovarian aging [46]. For instance, specific miRNAs expressed differently in ovarian follicular fluid, such as hsa-miR-190b and hsa-miR-99b-3p, have been suggested as potential biomarkers of oocyte quality [48]. In our study, we identified several miRNAs and lncRNAs, such as RP11-242D8, that are associated with ovarian aging. Although research on these is still in its infancy, our findings provide a foundation for exploring these molecules as new biomarkers and mechanistic contributors for ovarian aging. Further investigation is required to validate these potential markers and explore their clinical applications. Metabolite also play an important in ovarian aging. Significant changes in energy metabolism, particularly in TCA cycle, oxidative phosphorylation, lipid metabolism, glutamine metabolism and adenosine rescue pathway, have been observed in individuals with ovarian aging [49]. One such metabolite, citric acid, emerged as a key component in our metabolomics analysis. While previous studies have shown a reduction in TCA cycle, such as citric acid, in an age-dependent manner in oocytes [50], our findings suggest that citric acid can promote ovarian aging. This could be attributed to its accumulation, which may result from disruptions in energy metabolism, though the exact mechanism requires further exploration.

This study offers several significant advantages. First, to the best of our knowledge, it represents the most extensive and comprehensive integrative analysis of omics data conducted on ovarian aging to date. The study’s breadth, depth, and methodological rigor enabled the robust identification of promising targets for developing screening biomarkers and potential therapeutic interventions for ovarian aging. Second, the methodology employed is highly rigorous. To prevent violations of fundamental MR assumptions and minimize confounding bias, we excluded xQTLs located in the Major Histocompatibility Complex (MHC) region or far from their corresponding genes. Consistent findings across replicated MR analyses using ovarian xQTL data significantly reduced the likelihood of false positives. Lastly, to explore the significance and functionality of candidate targets, we integrated bioinformatics approaches, including enrichment analysis, protein–protein interaction (PPI) network assessments, and single-cell analyses. Furthermore, we validated our analytical models through in vitro and in vivo experiments. These efforts provide novel insights and a solid foundation for the discovery of new drug targets for ovarian aging.

Our study also has several limitations. First, ANM was used as a proxy for ovarian aging; although widely adopted in genetic studies, it is an indirect measure influenced by systemic aging processes and environmental factors, and does not distinguish between physiological and pathological ovarian decline. Second, most xQTL datasets were derived from non-ovarian tissues, primarily blood and brain, due to their larger sample sizes and greater statistical power. Ovarian-specific QTL datasets were limited in sample size or unavailable for certain omics layers. Given the well-established tissue specificity of gene regulation, these constraints may limit the interpretation of our findings in the context of ovary-specific biological mechanisms. future availability of large-scale reproductive tissue-specific QTL datasets will be important for refining and validating these findings. Third, our analytical strategy was intentionally conservative to minimize false positive findings, but this may have reduced statistical power and sensitivity. Limited probe coverage in the SMR analysis constrained the range of genes evaluated. In addition, stringent multiple testing correction and the conservative HEIDI filtering approach may have excluded loci with moderate effects or complex linkage disequilibrium structures. As a result, some true associations may have been missed. Lastly, our vitro experiments focused exclusively on the MSH6 gene, which may have caused us to overlook the contributions of other potentially significant genes. This narrow focus might have resulted in an incomplete understanding of the broader biological processes and missed potential interactions or regulatory mechanisms that could provide additional insights into ovarian aging.

Conclusions

In summary, our study supports the development of novel candidate biomarkers and potential targeted intervention strategies associated with ovarian aging through a multi-omics analysis framework. Notably, PD-L1 blockade alleviates DNA damage caused by MSH6 deficiency, a newly prioritized association of ovarian aging in this study, highlighting a promising therapeutic avenue that warrants further experimental investigation to delay ovarian aging. Overall, this research advances our understanding of the genetic mechanisms underlying ovarian aging, providing a foundation for future clinical approaches and therapeutic developments.

Methods

Study design and data resources (participants)

Figure 1 illustrates the overall design of this study, which integrates multiple data sources and analytical approaches. Four sets of exposures were utilized to comprehensively explore the genetic and molecular mechanisms of ovarian aging. Detailed information regarding all QTL and GWAS datasets employed in this study is provided in Table 1 and Methods section. These datasets include genetic, transcriptomic, proteomic, and metabolomic profiles, offering a robust foundation for identifying candidate biomarkers and potential therapeutic targets.

Table 1.

Information of QTL and GWAS datasets

	Data subtype	Resource	Sample size	Reference
xQTL (Discovery)	cis-pQTL	deCODE	35,559	https://www.decode.com
	cis-eQTL	eQTLGen Consortium	31,684	https://www.eqtlgen.org/
	cis-sQTL	BrainMeta v2	2,865	https://www.nature.com/articles/s41588-022-01154-4
	cis-mQTL	CLSA	8,299	https://www.clsa-elcv.ca/
xQTL (Replication)	cis-pQTL	INTERVAL	3,301	https://doi.org/10.1038/s41586-018-0175-2
	cis-eQTL	GTEx Consortium	180–283	https://www.gtexportal.org/home/
	cis-sQTL	GTEx Consortium	180–283	https://www.gtexportal.org/home/
	cis-mQTL	MRCIEU GWAS	7,468	https://gwas.mrcieu.ac.uk/datasets/met-a-360/
	cis-mQTL	MRCIEU GWAS	24,770	https://gwas.mrcieu.ac.uk/datasets/met-c-849/
	cis-mQTL	MRCIEU GWAS	291	https://gwas.mrcieu.ac.uk/datasets/ebi-a-GCST90026105/, https://gwas.mrcieu.ac.uk/datasets/ebi-a-GCST90026065/, https://gwas.mrcieu.ac.uk/datasets/ebi-a-GCST90026160/, https://gwas.mrcieu.ac.uk/datasets/ebi-a-GCST90026052/, https://gwas.mrcieu.ac.uk/datasets/ebi-a-GCST90026046/, https://gwas.mrcieu.ac.uk/datasets/ebi-a-GCST90026044/
GWAS (Discovery)	Ovarian aging	Reprogen	201,323	https://www.reprogen.org/
GWAS (Replication)	Ovarian aging	UK Biobank	143,819	https://gwas.mrcieu.ac.uk/datasets/ukb-b-17422/

xQTL datasets and the selection of instrumental variables

In this study, we utilized a comprehensive collection of xQTL datasets (Table 1) datasets to explore genetic, transcriptomic, proteomic, and metabolomic associations with ovarian aging. For proteome-wide QTL (pQTL) data, we drew upon 1881 cis-pQTLs from the deCODE study, which included data for 4,719 different blood proteins measured in 35,559 Icelandic individuals [51]. To validate our findings, we incorporated replication data from the INTERVAL study, comprising 1,478 cis-pQTLs in a sample of 3,301 participants [19]. Expression QTL (eQTL) data were primarily sourced from eQTLGen, a resource encompassing genetic data on blood gene expression from 31,684 individuals across 37 datasets with 19,942 cis-eQTLs [52]. To extend these findings to ovarian-specific contexts, we utilized eQTL data from the GTEx project v8 (https://www.gtexportal.org), which included data from 180 ovarian tissue samples. This dual-layer approach allowed us to examine eQTL associations at both systemic and tissue-specific levels [53]. For cis-sQTL data, we analyzed summary statistics for 12,794 splicing genes derived from BrainMeta v2, which were based on 2,865 brain cortex samples [54]. Ovarian-specific sQTL data were also incorporated from the GTEx project (https://www.gtexportal.org) to provide further context on alternative splicing events relevant to ovarian tissues [53]. Metabolome-wide QTL (mQTL) data were obtained from the CLSA study, which included information on 1,091 metabolites and 309 metabolite ratios measured in 8,299 individuals [55]. To ensure robust replication of our findings, additional mQTL datasets were retrieved from IEU OpenGWAS project (https://gwas.mrcieu.ac.uk/), including studies such as met-a-360, met-c-849 and various datasets from the ebi-a series (ebi-a-GCST90026105, ebi-a-GCST90026065, ebi-a-GCST90026160, ebi-a-GCST90026052, ebi-a-GCST90026046, ebi-a-GCST90026044) [56]. All data used in this study are from public databases, with no sample overlap, and no ethical considerations were required. The original QTL analyses adjusted for relevant demographic and technical covariates, including age, sex, genotype principal components, and batch effects. Except for the ovarian tissue–specific eQTL and sQTL data derived from GTEx, most QTL datasets were generated from mixed-sex populations, as provided in the original resources. Detailed information on the sample composition, omic analysis and genotyping protocols can be found in the original publication.

GWAS summary statistics

Ovarian aging, also known as female reproductive aging, is mainly defined by the loss of quantity and quality of oocytes or follicular pools [1]. The terminal stage of ovarian aging is marked by menopause, which establishes a clear physiological landmark through the irreversible depletion of primordial follicle reserves [57]. Age at natural menopause (ANM), defined as the age at the last naturally occurring menstrual period followed by at least 12 consecutive months of amenorrhea. ANM serves as a widely used proxy for ovarian aging in genetic studies [7, 8]. For this study, we leveraged GWAS data on ANM to explore genetic associations with ovarian aging. Specifically, we utilized summary statistics from the largest GWAS meta-analysis conducted to date. This comprehensive dataset was sourced from the REPROGEN Consortium and included data from 200,000 women across 34 European cohorts, as reported by Ruth et al. [9]. This powerful dataset has identified 290 genetic determinants of ovarian aging, providing a valuable resource for studying the genetic basis of ovarian aging. For the replication cohort, we utilized dataset from 143,819 European-ancestry individuals in the UK Biobank. The data were collected via the ACE touchscreen questionnaire: "How old were you when your periods stopped?" [20]. The GWAS summary statistics was derived from the IEU OpenGWAS database under the identifier ukb-b-17422 (https://gwas.mrcieu.ac.uk/datasets/ukb-b-17422/).

Single-cell transcriptome analysis

We analyzed single-cell RNA sequencing (scRNA-seq) data from ovarian samples of human, primate, and mouse to investigate molecular changes associated with ovarian aging. The data were obtained from publicly available datasets: human (GEO: GSE255690), primate (GEO: GSE130664), and mouse (GEO: GSE232309).

To ensure cross-species comparability, genes were mapped to their one-to-one orthologs using Ensembl BioMart (version 104), creating a unified gene set for analysis. The Seurat package (version 4.3.0) was employed for data processing and analysis.Cells with mitochondrial gene percentages exceeding 20% were considered low-quality and subsequently removed from further analysis. Additionally, cells exhibiting excessively high nCount (total UMI counts per cell) and nFeature (number of detected genes per cell) were identified as doublets and excluded.Then, the data were normalized using NormalizeData with the LogNormalize method, followed by identification of variable features with FindVariableFeatures (selection.method = "vst", nfeatures = 2000). The dataset was then scaled using ScaleData to regress out unwanted sources of variation. Dimensionality reduction was performed via principal component analysis (PCA) with RunPCA, retaining the top 30 principal components based on an elbow plot. To address batch effects within each species, we applied the Harmony algorithm (version 1.0.0) to the PCA embeddings, using donor and batch as grouping variables to correct for technical variability while preserving biological differences. The resulting Harmony-corrected embeddings were then used for UMAP visualization, enabling us to represent the cells in a two-dimensional space that reflects biological similarities rather than technical artifacts. Clustering was subsequently conducted on these embeddings with Seurat’s FindNeighbors (k.param = 20) and FindClusters (resolution = 0.5), where the resolution was chosen to balance the identification of distinct ovarian cell types.

Cell type annotation was based on differentially expressed genes (DEGs) identified for each cluster using FindAllMarkers (Wilcoxon rank-sum test, min.logfc.threshold = 0.25, adjusted p-value < 0.05 via Bonferroni correction). DEGs were compared to established marker genes for ovarian cell types from published studies [51–53]. Marker gene expression was visualized on UMAP plots and dot plots to confirm cluster-specific expression.A comprehensive list of marker genes and their expression patterns is provided in Additional file 1: Table S7A-C, with supporting visualizations in Additional file 2: Fig. S6a-c.

For differential expression analysis between aging and young cell populations, we used FindMarkers with the Wilcoxon rank-sum test (min.logfc.threshold = 0.25) and applied FDR correction for multiple testing, setting the significance threshold at adjusted p < 0.05. This analysis was performed within each cell type to identify age-related gene expression changes.

Visualization of the results was done using the scRNAtoolVis and plot1cell packages, providing insights into the molecular differences between aging and young cell populations within the ovaries. All analyses were conducted in R (version 4.3.1).

SMR and HEIDI analysis

The Summary-data-based Mendelian Randomization (SMR) method was employed in this study to integrate genome-wide association studies (GWAS) with xQTL data, leveraging the Mendelian randomization framework. For the SMR analysis, cis-QTLs were selected as instrumental variables (IVs) through a rigorous filtering process: (i) SNPs were required to be robustly associated with gene expression levels (P < 5 × 10⁻⁸)within ± 1000 kb upstream and downstream of the target gene; (ii) SNPs were filtered to ensure minimal linkage disequilibrium (LD) (R² < 0.001, distance = 10,000 kb), indicating that each selected SNP is largely independent of the others; (ⅲ)SNPs with allele frequency differences exceeding 0.2 between any two datasets (including LD reference samples, QTL summary data, and outcome summary data) were excluded. The threshold for statistical significance in SMR analysis is set at p < 0.05, signifying a robust association between genetic variants and molecular traits such as gene expression, protein, gene splicing, metabolites associated with ovarian aging. To ensure robustness and minimize false discoveries, a Bonferroni correction was applied to adjust the significance thresholds in the SMR tests. To further validate the findings and differentiate pleiotropy from linkage, the SMR analysis was complemented by the heterogeneity in dependent instruments (HEIDI) test. This approach offers more robust discrimination between pleiotropic and linkage effects, reduces potential biases due to LD, and lowers the large sample size requirements often seen in standard MR methods. Molecular phenotypes that passed both the Bonferroni-adjusted SMR significance criteria and the HEIDI test (P > 0.01) were considered functionally relevant and likely to contribute to the biological mechanisms underlying ovarian aging. All analyses were conducted using SMR software (SMR v1.0.3) [12].

Metabolic pathway analysis

Metabolic pathway analysis was performed using the web based MetaboAnalyst 5.0. (https://www.metaboanalyst.ca/). The functional enrichment and pathway analysis modules were employed to identify relevant metabolite groups or pathways associated with the biological processes of ovarian aging. For this analysis, we utilized two comprehensive databases: SMPDB and the KEGG database. The significance threshold for pathway analysis was set at 0.10 to capture potentially important pathways related to ovarian aging. Given the sensitivity of metabolic changes, we applied a stringent threshold of 0.001 to screen for effective metabolites involved in these pathways, ensuring that even minor but biologically significant alterations in metabolites were considered. This approach allowed for a more precise identification of metabolic pathways that may play a crucial role in the regulation of ovarian aging.

OPERA: jointly analyzes GWAS and multi-omics QTL

To identify pleiotropic associations between multiple molecular phenotypes and ovarian aging, we utilized the OPERA (Omics PleiotRopic Association) software [13]. This approach jointly analyzes genome-wide association study (GWAS) data alongside multi-omics QTL datasets, including pQTL, eQTL, and sQTL data (mQTL is excluded because of the limited availability of its metabolites). OPERA identifies pleiotropic molecular phenotypes underlying complex traits by integrating GWAS and multi-omics xQTL data within a Bayesian framework. The method first estimates genome-wide prior probabilities of association patterns from quasi-independent loci, then computes locus-specific PPAs (posterior probability of association) for molecular interactions by combining likelihoods with these priors. High-confidence associations (PPA > 0.9) undergo multi-exposure HEIDI testing to exclude linkage-driven signals, retaining only pleiotropic interactions. This two-stage strategy balances computational efficiency with robustness, leveraging summary statistics and LD references to prioritize shared causal variants across omics layers.

The GWAS summary data were from the largest available GWAS meta-analyses REPROGEN Consortium. The plasma pQTL summary statistics were from the deCODE study and the blood eQTL summary data were from the eQTLGen study. The sQTL summary data were generated from BrainMeta v2. All IVs have been rigorously screened during SMR analysis. To ensure robust findings, associations identified by OPERA were considered significant only if they met two stringent criteria: a HEIDI test P-value (P_HEIDI) greater than 0.01, confirming pleiotropy rather than linkage, and a joint PPA exceeding 0.9. These thresholds provided high confidence in identifying loci that play pleiotropic roles in the regulation of ovarian aging. All the analysis is carried out in OPERA software (https://github.com/wuyangf7/OPERA).

Multi-omics joint enrichment analysis

We conducted a multi-omics joint enrichment analysis using the online platform Metascape (https://metascape.org), which integrates over 40 gene function annotation databases, such as Gene Ontology (GO), KEGG, UniProt, and DrugBank, among others [58]. The analysis was performed with a minimum overlap of three genes and a p-value cutoff of 0.05 to ensure statistical significance. For the mQTL-derived metabolites, the corresponding genes were annotated based on the gene loci represented by the associated SNPs. The genes identified from pQTL, eQTL, sQTL, and mQTL datasets were subsequently integrated. After removing duplicates, a total of 219 genes associated with ovarian aging were obtained for downstream functional enrichment analysis. To visualize the results, we created a network diagram using the software Cytoscape, which allowed for a clear representation of the relationships between the identified molecular pathways and their potential roles in ovarian aging. This approach enabled us to gain deeper insights into the functional enrichment and biological processes associated with the molecular phenotypes under investigation.

Single-cell analysis

We analyzed single‐cell RNA‐sequencing (scRNA‐seq) data from ovarian tissues of human, non‐human primate, and mouse to characterize molecular signatures of ovarian aging. Raw count matrices were retrieved from the Gene Expression Omnibus (GEO: GSE255690 for human, GSE130664 for primate, and GSE232309 for mouse). To enable cross‐species comparison, gene identifiers were mapped to one‐to‐one orthologs via Ensembl BioMart (release 104), yielding a unified gene set.

All downstream analyses were performed in R (v4.3.1) using Seurat (v4.3.0). Cells with > 20% mitochondrial transcripts were excluded to remove low‐quality captures. Potential doublets were identified and discarded by filtering out cells with extreme total UMI counts (nCount_RNA) or detected gene numbers (nFeature_RNA), based on dataset‐specific thresholds. Remaining counts were log-normalized (NormalizeData(method = "LogNormalize")), and the top 2,000 highly variable genes were selected (FindVariableFeatures(selection.method = "vst", nfeatures = 2000)). To mitigate confounding sources of variation, we performed linear scaling and regression against nCount_RNA and percentage of mitochondrial reads (ScaleData). Dimensionality reduction was carried out via principal component analysis (PCA) using the top features (RunPCA), and the first 30 PCs were retained according to the “elbow” criterion. To correct for batch effects within each species, we applied Harmony (v1.0.0) to the PCA embeddings, specifying donor and batch as covariates. Harmony‐corrected embeddings were then projected into two dimensions using UMAP for visualization (RunUMAP).

Graph‐based clustering was performed on the Harmony embeddings (FindNeighbors(k.param = 20)followed by FindClusters(resolution = 0.5)), optimizing resolution to distinguish known ovarian cell populations without over‐fragmentation. Cluster‐specific marker genes were identified with FindAllMarkers (Wilcoxon rank‐sum test; log₂ fold‐change > 0.25; Bonferroni‐adjusted P < 0.05) and annotated by comparison to established ovarian cell markers [51–53]. Marker expression was verified via UMAP feature plots and dot plots (Additional file 2: Fig. S6a-c), and a comprehensive list is provided in Additional file 1: Table S7A-C.

To detect age‐associated transcriptional changes, we conducted pairwise differential expression analyses between young and aged cells within each annotated cell type using FindMarkers (Wilcoxon test;FDR < 0.05). Visualization of differentially expressed genes and cell‐type compositions utilized the scRNAtoolVis and plot1cell packages.

Isolation of human granulosa cells

Mural granulosa cells were isolated from 46 patients diagnosed with DOR and 46 healthy women undergoing standard clinical in vitro fertilization (IVF) protocols at the Center for Reproductive Medicine of Peking University Third Hospital. The DOR diagnosis was made if patients met at least two of the following three criteria: a summation of bilateral AFC ≤ 7; AMH ≤ 1.1 ng/ml; and basal follicle-stimulating hormone (FSH) ≥ 10 IU/L on the second or third day of a spontaneous menstrual cycle [59].The control group consisted of women receiving IVF treatment because of male factor infertility. Mural granulosa cells were isolated from follicular fluid aspirated during oocyte retrieval using Ficoll density gradient centrifugation [29].

Construction and treatment of MSH6-KO granulosa cells

To construct and treat MSH6-knockout (KO) granulosa cells, CRISPR/Cas9 lentiviral single-guide RNA (sgRNA) was used to knock out MSH6 expression in KGN cells. The sgRNA sequences targeting MSH6 were designed using the online tool available at http://crispor.gi.ucsc.edu, with the following sequences: TACTGCAACATTTGATGGGA and sgRNA2: GAGGTACTGCAACATTTGAT. The sgRNA Oligos were annealed and ligated into the LentiCRISPRv2 vector. KGN cells were infected with the lentiviral supernatant for 24 h in the presence of polybrene (Beyotime, 1:1000). Cells were selected with 1 μg/mL puromycin (Gibco). For the treatment of MSH6-KO KGN cells, different concentrations of atezolizumab (MCE, HY-P9904) were added to the culture medium for 24 h.

RNA isolation and real-time PCR

Total RNA was extracted from granulosa cells using TRIzol reagent (Invitrogen). cDNA was synthesized using a reverse transcription kit (TOYOBO). Quantitative polymerase chain reaction (PCR) analysis was performed using SYBR Green PCR master mix (Applied Biosystems, A25777) according to the manufacturer’s instruction. Primers sequences are shown in Additional file 1: Table S8.

Senescence-associated β-galactosidase staining

Cells were seeded into 6-well plates at a density of 2.5 × 10⁵ cells per well and cultured for 24 h. After incubation, the cells were fixed and stained using the Senescence β-Galactosidase Staining Kit (Cell Signaling, #9860) according to the manufacturer’s instructions. The stained cells were visualized under a microscope, and the percentage of SA-β-gal-positive staining areas was quantified using ImageJ software.

Transcriptomic analysis

RNA-seq libraries were prepared and sequenced by CapitalBio Technology (Beijing, China) using paired-end sequencing with a 150 bp read length on the Illumina NovaSeq platform (Illumina). The human reference genome (hg38 version) was used for alignment. Sequencing quality was assessed with FastQC (v0.11.5), and low-quality data were filtered using NGSQC (v2.3.3). Clean reads were then aligned to the reference genome with HISAT2 (v2.1.0) using default parameters. Differential gene expression analysis was performed using DESeq (v1.28.0) to identify differentially expressed genes (DEGs) between samples, with a fold change ≥ 1.5 (|log2FC|≥ 0.585) and a p-value ≤ 0.05. Gene set enrichment analysis (GSEA) was conducted using GSEA software (http://software.broadinstitute.org/gsea/index.jsp and KEGG pathway enrichment analysis of DEGs was performed with the clusterProfiler R package.

Western blot analysis

Cells were lysed directly with RIPA buffer (Beyotime) and incubated on ice for 30 min. Protein concentrations were quantified using the BCA assay (Thermo Fisher). Total protein extracts from each sample were separated on 12% SDS-PAGE gels. The primary antibodies were applied according to the provided recommendations: MSH6 (Proteintech, No.66172), γH2AX (Abcam, No.ab11174), p21 (Cell Signaling, No.2947), p53 (Cell Signaling, No.9282), Bcl2 (Cell Signaling, No.15071), Bax (Cell Signaling, No.5023), p-NF-κB (Cell Signaling, No.3033), NF-κB (Cell Signaling, No.8242), p-ERK1/2(Cell Signaling, No.4370), ERK1/2(Cell Signaling, No.9102), PD-L1 (Cell Signaling, No.13684), Tubulin (Cell Signaling, No.2144), Vinculin (1:1000; Abcam, No.ab207440). We determined the intensity of bands by ImageJ.

Real-time cell analysis

Cell proliferation was assessed using real-time cell analysis (RTCA) on the xCELLigence platform. Each well of E-plates (Agilent) was pre-filled with 80 μL of culture media. A cell suspension containing 5000 cells in 120 μL of culture medium was then added to each well, which had previously been used to measure background impedance. The cells were incubated for 30 min at 37 ℃ in a 5% CO2 incubator to allow for initial cell adhesion at the bottom of each well. The RTCA software was used to analyze the cell index values, providing a continuous measurement of cell proliferation.

Cell cycle analysis

Cell cycle phase analysis was performed using the Cell Cycle Assay Kit (Elabscience, E-CK-A351). Cells were first collected by centrifugation and resuspended in 0.3 mL of phosphate-buffered saline (PBS). Then, 1.2 mL of absolute ethanol was added to the cells, and they were fixed overnight at −20 °C. After fixation, cells were centrifuged and resuspended in 1 mL of PBS. The suspension was left at room temperature for 15 min before being collected by centrifugation. The cells were resuspended in 100 μL of RNase A Reagent and incubated at 37 °C for 30 min. Following this, 400 μL of propidium iodide (PI) reagent was added, and the cells were incubated at 4 °C for 30 min. Finally, the cell cycle distribution was analyzed using a Beckman FACS flow cytometer, with data processed using Kaluza Flow Analysis Software (Beckman Coulter).

Protein‑protein interaction (PPI) and druggability evaluation

To investigate the potential interactions between the proteins identified in our analysis, we constructed a protein–protein interaction (PPI) network using the STRING database. Additionally, we sought to explore the interactions between these ovarian aging-associated proteins and existing drug targets, particularly focusing on disease-modifying drugs for ovarian aging. For this, we retrieved a list of nine such drugs from a recent review, along with their corresponding drug targets, using the DrugBank database (https://www.drugbank.ca) [60].

To further assess the druggability of the proteins identified in our study, we cross-referenced them with current medications and potential drugs targeting these proteins. We utilized multiple databases for this analysis, including DGIdb (https://dgidb.org/), ChEMBL (https://www.ebi.ac.uk/chembl/), and DrugBank (https://go.drugbank.com/). All PPI analyses were performed using STRING database version 11.5 (https://string-db.org/), with a minimum required interaction score of 0.4 to ensure the reliability of identified interactions. These analyses provided insight into potential therapeutic avenues for targeting ovarian aging and informed further exploration of the druggability of the identified proteins.

Authors’ contributions

Conceptualization: L.M., Y.Z., Y.W., Y.Y., J.W. Methodology: L.M., Y.Z., Y.W., Y.Y., J.W. Data curation: X.L., S.S., C.L., M.Z., Y.L., W.L., X.J. Formal analysis: X.L., S.S., C.L., M.Z., Y.L., W.L. Visualization: X.L., S.S., C.L., M.Z. Investigation: L.M., Y.Z., Y.W., Y.Y., J.W., X.L., S.S., G.W., H.Y., L.W., Y.C., H.-M.C. Writing – original draft: X.L., S.S., C.L., M.Z. Writing – review & editing: L.M., Y.Z., Y.W., Y.Y., J.W., W.Z., L.H., Y.M., X.D. Supervision: L.M., Y.Z., Y.W., Y.Y., J.W. Funding acquisition: L.M., Y.Z., Y.W., Y.Y. All authors read and approved the manuscript. Author order was assigned based on relative contribution to experimental design and manuscript drafting, with mutual consensus.

Funding

This study was supported by the National Key Research and Development Project of China (2024YFC2706600), the National Natural Science Foundation of China (82522038, 82288102), the Fundamental Research Funds for the Central Universities, the 1·3·5 project for disciplines of excellence, West China Hospital, Sichuan University (ZYYC24006), the Beijing Natural Science Foundation (Z230013), and the Key Clinical Projects of Peking University Third Hospital (BYSY2022043).

Data availability

Availability of Data and Materials The datasets supporting the conclusions of this article are available in public repositories. The pQTL datasets were obtained from the deCODE study portal [61] as described in Ferkingstad et al. [51] and the INTERVAL study [62] as described in Sun et al. [19]. The eQTL datasets were obtained from the eQTLGen Consortium [63] as described in Võsa et al. [52] and the GTEx Project v8 [64] as described in [65]. The sQTL datasets were obtained from BrainMeta [66] as described in [54] and the GTEx Project v8 [64] as described in [65]. The mQTL datasets were obtained from the CLSA [67] as described in [55] and the IEU OpenGWAS database [68] as described in [69–71] (Accession IDs: met-a-360, met-c-849, ebi-a-GCST90026105, ebi-a-GCST90026065, ebi-a-GCST90026160, ebi-a-GCST90026052, ebi-a-GCST90026046, and ebi-a-GCST90026044). GWAS summary statistics were obtained from the REPROGEN Consortium [72] as described in [9], and the UK Biobank (Accession ID: ukb-b-17422) [68] as described in [20]. The single-cell transcriptomic datasets were obtained from the Gene Expression Omnibus (GEO) [73] under accession numbers GSE255690 [42], GSE130664 [74], and GSE232309 [75]. The source code used in this study is publicly available on GitHub and archived in Zenodo under the MIT License, an OSI-approved open-source license, at https://github.com/XuanLian-ops/ovarian-aging-multiomics-SMR-OPERA and https://doi.org/10.5281/zenodo.20085006 [76, 77].

The analyses were performed using publicly available software including SMR (https://yanglab.westlake.edu.cn/software/smr/#Overview) [12], OPERA (https://github.com/wuyangf7/OPERA) [13].

Declarations

Ethics approval and consent to participate

All summarized statistics utilized in the MR analyses were derived from previously published studies, for which ethical approval and informed consent had been obtained by the original investigators. The experimental study was approved by the Medical Science Research Ethics Committee of Peking University Third Hospital (Approval No. 2024–020-01) and was conducted in accordance with the principles of the Declaration of Helsinki. Written informed consent was obtained from all participants prior to participation in the study.

Consent for publication

Not applicable.

Competing interests

The authors declare no competing interests.