Integrative multi-omics analysis of druggable genes for therapeutic target identification in polycystic ovary syndrome Authors: Dan Xu, Dan Jia, Xiaohui Fang, Wansu Chen, Ying Liu, Qingxia Song, Xiudao Song Published: 2025 ABSTRACT BackgroundPolycystic ovary syndrome (PCOS) is a common endocrine disorders in women of reproductive age with limited targeted therapies. This study aimed to identify and prioritize potential druggable genes for PCOS through a purely computational, multi-omics approach.MethodWe conducted an integrated in silico analysis leveraging publicly available datasets. A multi-omics Mendelian randomization (MR) framework was applied to 2,888 druggable genes. Our workflow incorporated two-sample MR for initial discovery of gene-PCOS associations with false discovery rate (FDR) correction, followed by validation using summary-data-based MR (SMR) and colocalization. Tissue-specific effects were assessed via cis-expression quantitative trait loci (cis-eQTL) MR in ovary, uterus, and blood, and epigenetic regulation via methylation QTL (mQTL) MR. The ovarian microenvironment was characterized using CIBERSORT and single-cell RNA sequencing (scRNA-seq). We also assessed whether immune cell traits mediate the causal pathway between cis-eQTLs and PCOS. Finally, we predicted potential drug targets using the Comparative Toxicogenomics Database and thereby performed molecular docking simulations.ResultsWe identified seven FDR-significant druggable genes associated with PCOS risk: NRBP1, LGR6, GHRL, and IPP as risk-enhancing genes, and CBLN3, VIPR1, and TFRC as protective. SMR analysis confirmed that genetically predicted NRBP1 (OR = 1.5383, p = 8.67E-04), LGR6 (OR = 1.3442, p = 0.0107), GHRL (OR = 1.3391, p = 0.0413) increased PCOS risk, while CBLN3 (OR = 0.8605, p = 4.09E-04) decreased PCOS risk. Colocalization analysis prioritized CBLN3 (PP.H4 = 86.65%) and NRBP1 (PP.H4 = 78.46%) as high-confidence targets. CBLN3 expression was protective across tissues (blood: OR = 0.8586; ovary: OR = 0.8926; uterus: OR = 0.8713), while blood NRBP1 expression increased PCOS risk (OR = 2.1003). mQTL MR analysis revealed significant locus-specific causal effects for CBLN3 and NRBP1. Immune infiltration analysis revealed CBLN3 correlated positively with resting dendritic cells (r = 0.80), and inversely with eosinophils (r=-0.72) and neutrophils (r=-0.66), while NRBP1 associated positively with activated dendritic cells (r = 0.67) and inversely with resting CD4 T cells (r=-0.77). scRNA-seq identified four subtypes of PCOS theca cells, and the expression levels of CBLN3 and NRBP1 in four subtypes were explored. Mediation analysis suggested that the protective effect of CBLN3 may be partially mediated through NKT %lymphocytes. Primary compound screening identified nine high-affinity compounds binding both CBLN3 and NRBP1 (ΔG≤-5 kcal/mol).ConclusionsUsing a comprehensive bioinformatic framework, we prioritize CBLN3 as a novel protective target and NRBP1 as a risk-associated target with multi-omics validation, thus providing mechanistically informed candidates for the development of urgently needed PCOS therapeutics.Supplementary InformationThe online version contains supplementary material available at 10.1186/s13048-025-01889-8. Introduction Polycystic ovary syndrome (PCOS) represents the most prevalent endocrine disorder in reproductive-aged women, affecting 6–20% of this population globally [1, 2]. Characterized by adolescent-onset hyperandrogenism, insulin resistance, and ovarian dysfunction, its clinical manifestations encompass hirsutism, anovulation, and polycystic ovarian morphology [1]. PCOS confers lifelong risks for type 2 diabetes, cardiovascular diseases, and psychiatric comorbidities [3]. Aberrant folliculogenesis in PCOS causes anovulatory infertility and impaired oocyte and embryo quality [4], contributing to reproductive impairments including reduced pregnancy rates, diminished live births, and increased miscarriage risk [5]. Maternal PCOS further influences offspring metabolic health via epigenetic mechanisms [6]. Current management strategies (e.g., lifestyle interventions, metformin) primarily alleviate symptoms but fail to target core pathological mechanisms due to limited understanding of causal drivers. Although gene-environment interactions are established as core pathogenic mechanisms of PCOS [7], the precise underlying etiology of PCOS remains incompletely understood, representing a major obstacle in developing targeted and curative therapies. Therefore, there is an urgent, unmet clinical need for novel therapeutic strategies, particularly biologically targeted interventions aimed at the fundamental molecular drivers of PCOS. The recognition of PCOS as a complex, multi-systemic disorder with strong genetic, epigenetic, and environmental contributions necessitates a comprehensive multi-omics approach for mechanistic insights. Circulating biomolecules present in plasma, such as genes, proteins, and immune cells, offer a unique and minimally invasive window into the systemic pathophysiology of PCOS. Integrating these multi-omics data layers (e.g., genomics, transcriptomics, proteomics, metabolomics) holds immense promise for identifying convergent molecular networks and critical hubs dysregulated across PCOS phenotypes, potentially revealing master regulatory factors or pathways amenable to therapeutic intervention. However, these findings are inherently susceptible to confounding factors (e.g., age, ethnicity, medication use) and reverse causation bias, wherein disease progression influences biomarker levels rather than vice versa. Mendelian randomization (MR) overcomes these limitations by leveraging germline genetic variants as instrumental variables to infer causality [8]. This approach provides causal inference power comparable to randomized controlled trials (RCTs) while mitigating reverse causation bias [9, 10]. MR validity hinges on three core assumptions: genetic instruments must strongly associate with the exposure, remain independent of confounders, and affect outcomes exclusively via the target exposure. Recent MR studies have elucidated that BMI, TG, LDL-c, HDL-c and T2D were causally linked to PCOS [11]. Despite these advances, circulating genome drivers of PCOS remain underexplored through causal inference frameworks, limiting the identification of biologically coherent therapeutic targets. For drug-target MR, expression quantitative trait loci (eQTLs) regulating gene expression serve as instrumental variables, with cis-eQTLs proximal to target genes being preferentially selected. This approach has been successfully applied in metabolic dysfunction-associated steatotic liver disease [12] and estrogen receptor-positive breast cancer [13], validating its utility for identifying druggable targets. In the current study, we implemented an integrated multi-omics MR framework spanning 2,888 druggable genes. This approach synergistically combined causal inference through multi-tissue MR (analyzing blood, ovary, and uterus) and DNA methylation MR; rigorous validation via summary-data-based Mendelian randomization (SMR) and colocalization analysis; mechanistic deconvolution using immune infiltration profiling and single-cell RNA sequencing of ovarian cells; and computational drug prioritization screening compound-target interactions. Methods Data sources and instrument selection Cis-eQTL data for all genes were acquired from the eQTLGen Consortium, exclusively derived from whole blood samples (https://www.eqtlgen.org/cis-eqtls.html) [14]. In this genome-wide drug target discovery analysis, we implemented stringent criteria for instrumental variable (IV) selection using cis-eQTLs [15]: (1) genome-wide significance threshold: SNPs highly correlated with plasma genes were identified based on a significance threshold of p < 5E-08; (2) To ensure the inclusion of independent SNPs and mitigate the impact of linkage disequilibrium (LD) on the results, a LD parameter (r2) of 0.1 and a genetic distance of 10,000 kb were set; and (3) Exclusion of variants with potential weak instrument bias (F-statistic >10). According to predefined selection criteria, a total of 15,695 cis-eQTLs identified from eQTLGen Consortium, were selected as IVs. The druggable genes are defined as a set of genes encoding proteins with potential to be modulated by a drug-like small molecule based on sequence and structural similarity to the targets of existing drugs [16]. It comprises 1,427 genes that are efficacy targets of approved drugs or clinical-phase drug candidates, 682 genes encoding proteins with known bioactive drug-like small molecule binders or those with high sequence similarity to approved drug targets, and 2,370 genes encoding proteins with more distant similarity to approved targets. Finally, we obtained the cis-eQTL data in the blood for only 2,888 genes out of 4,479 druggable genes. Genome-wide association study (GWAS) summary statistics for PCOS encompassing 2,214 cases and 267,780 controls from FinnGen biobank (v12) (https://www.finngen.fi/en), and GCST90044902 encompassing 797 cases and 140,558 controls were obtained from the IEU OpenGWAS platform (https://gwas.mrcieu.ac.uk/). A schematic representation of the study design was shown in Fig. 1. We hereby explicitly state that the present work is a purely computational, bioinformatic investigation. No new patient samples were collected, nor were any laboratory experiments conducted specifically for this study. All analyses and findings are derived from the secondary analysis and integration of the following publicly available datasets. Fig. 1Flowchart of the study design Flowchart of the study design Two-sample Mendelian Randomization (MR) analysis We employed a two-sample MR framework to investigate causal relationships between cis-eQTLs and PCOS, with steiger filtering applied to eliminate reverse causation bias. cis-eQTLs were considered as exposures and PCOS as the outcome, excluding instruments with F-statistics < 10 to mitigate weak instrument bias. Primary causal effect estimates were generated using the inverse-variance weighted (IVW) method, with significance defined as p < 0.05. Robustness was validated through sensitivity analyses: (1) pleiotropy assessment (MR-Egger intercept test); (2) heterogeneity evaluation (Cochran’s Q statistic). Odds ratios (ORs) with 95% confidence intervals (CIs) were calculated to estimate causal effects. In the cis-eQTL MR analysis, genes that did not satisfy the horizontal pleiotropy and heterogeneity criteria were excluded from the further analysis. Analyses were conducted using R packages TwoSampleMR (v0.5.7), ggplot2 (v3.5.2), and foreach (v1.5.2). Summary-data-based Mendelian Randomization (SMR) analysis To assess potential causal relationships between genetically predicted gene expression and PCOS, we performed Summary-data-based Mendelian Randomization (SMR) analysis using the SMR software (version 1.3.1; https://yanglab.westlake.edu.cn/software/smr/) [17]. This analysis utilized blood cis-eQTL summary statistics obtained from the eQTLGen Consortium as the exposure and summary statistics for PCOS as the outcome. Blood gene expression data from the Genotype-Tissue Expression (GTEx) project V8 (https://gtexportal.org/) [18] were employed for validation purposes. The most strongly associated cis-eQTL (serving as IVs) for each gene was selected based on a genome-wide significance threshold of 5E-08. To evaluate potential horizontal pleiotropy, we applied the Heterogeneity in Dependent Instruments (HEIDI) test as a sensitivity analysis. Associations yielding a HEIDI test p-value (pHEIDI) ≤ 0.05 were considered likely influenced by pleiotropy and were consequently excluded from further consideration. Only associations demonstrating a nominal SMR p-value < 0.05 and evidence against pleiotropy (pHEIDI >0.05) were retained for subsequent analysis. DNA Methylation Quantitative Trait Loci (mQTL) analysis To investigate the causal effects of DNA methylation on PCOS, we utilized summary statistics for DNA methylation quantitative trait loci (mQTLs) from the GoDMC database (version 1.0; http://www.GoDMC.org.uk) [19]. This resource aggregates genome-wide mQTL data derived from 39,851 individuals of European ancestry. mQTL were extracted and subjected to IV selection using a three-tiered filtering approach: (1) genome-wide significance threshold (p < 5E-08), (2) LD clumping within 10,000 kb windows and r2 < 0.1, and (3) verification against Illumina EPIC array CpG annotations. This multi-step filtering strategy yielded two independent mQTL instruments for the CBLN3 gene (corresponding to CpG sites cg05142272 and cg08016929) and five independent mQTL instruments for the NRBP1 gene (corresponding to CpG sites cg05102552, cg05385684, cg12243333, cg15478930, and cg20888435). Co-localization analysis To investigate potential shared genetic causality, we performed Bayesian co-localization analysis testing whether cis-eQTL signals (summary statistics per gene after LD clumping from the eQTLGen consortium) and PCOS GWAS signals (summary statistics from the FinnGen R12 release) at specific loci are driven by the same underlying causal variant [20]. Analysis was conducted using R packages data.Table (1.16.4), dplyr (1.1.4), coloc (5.2.3), and foreach (1.5.2), employing the approximate Bayes factor method with prior probabilities set to p1 = 1E − 04 (cis-eQTL association), p2 = 1E − 04 (PCOS association), and p12 = 1E − 05 (association with both). This evaluates five competing hypotheses (H₀: no association; H₁: cis-eQTL only; H₂: PCOS only; H₃: both traits with two independent causal variants; H₄: both traits with one shared causal variant). Results were interpreted primarily using the posterior probability for H₄ (PP.H₄), where PP.H₄ ≥ 0.75 indicated strong evidence for co-localization [21]. The co-localization analysis was conducted within a fixed 100 kb window upstream and downstream of the transcriptional start site of each gene. Tissue-specific Expression Quantitative Trait Locus (eQTL) analysis for MR Given that cis-regulatory elements have more direct and specific biological effects than trans-regulatory elements, we utilized cis-eQTLs from human blood, ovary, and uterus tissue derived from the genotype tissue expression data (GTEx analysis V10). Genetic variants were initially filtered using a significance threshold of p < 1E-05. LD clumping was then performed (100 kb window, r² threshold 0.3) to ensure independence of IVs. We further employed a two-sample to investigate causal relationships between cis-eQTLs and PCOS, with steiger filtering applied to eliminate reverse causation bias. Processing single-cell RNA sequencing data of PCOS This analysis represents an in silico re-analysis of a publicly available single-cell RNA sequencing (scRNA-seq) dataset from Zenodo (https://zenodo.org/record/7942968) [22], comprising theca cell profiles from 5 PCOS patients and 5 controls. Raw data processing, quality control, and subsequent analyses were performed using Seurat (v5.1.0) in R. Genes detected in fewer than three cells were excluded from the analysis. Cell-level quality control was rigorously applied to remove low-quality cells and potential doublets. Low-quality cells were excluded based on the following criteria: gene detection count (n_Feature) >200 and < 8000, or total transcript count (n_Count) >1000. Furthermore, cells exhibiting a high proportion of mitochondrial gene expression (>20%) or a significant presence of red blood cell genes (>5%) were also discarded. To minimize the impact of multiplets, DoubletFinder (v2.0) was employed on each sample individually, and only confidently predicted singlets were retained for downstream analysis. The count matrix was then log-normalized, and 3,000 highly variable features were selected using the Seurat pipeline. The matrix constructed from these highly variable genes was subsequently scaled and used for Principal Component Analysis (PCA). The Reciprocal PCA method implemented in the Seurat package was used to correct for batch effects across samples. A k-nearest neighbour graph was generated for all cells for subsequent unsupervised clustering using the Louvain algorithm, and Uniform Manifold Approximation and Projection (UMAP) was used to visualize the projection of all cells in a low-dimensional space. CIBERSORT analysis Immune cell infiltration patterns within granulosa cells from PCOS patients versus healthy controls were characterized using the CIBERSORT algorithm on transcriptomic profiles from the GSE34526 dataset comprising 7 patients with PCOS and 7 healthy controls. This analysis quantified relative abundances of 22 immune cell subtypes. Differential infiltration between PCOS patient and control groups was assessed using the Mann-Whitney U test, followed by Spearman’s correlation analysis to evaluate associations between hub gene expression and immunocyte fractions. Prediction of potential drug and molecular docking To assess the therapeutic potential of candidate genes, we identified putative drug compounds through interrogation of the Comparative Toxicogenomics Database (CTD), which catalogs chemical-gene interactions [23]. Molecular docking simulations were subsequently performed to evaluate binding affinities and intermolecular interactions between prioritized compounds and their target proteins. Compound structures were retrieved in SDF format from the PubChem Compound Database (https://pubchem.ncbi.nlm.nih.gov/). Protein structures of CBLN3 (UniProt/AlphaFold ID: Q6UW01) and NRBP1 (UniProt/AlphaFold ID: Q9UHY1), were retrieved from the AlphaFold Protein Structure Database (https://alphafold.ebi.ac.uk/) with corresponding accession identifiers. The global per-residue confidence scores (pLDDT) for CBLN3 and NRBP1 are 75.57 and 75.38, respectively, indicating “Confident” predictions. Docking analyses were performed using the CB-Dock2 web server (https://cadd.labshare.cn/cb-dock2/) [24], employing its blind docking algorithm to predict ligand binding poses and calculate binding affinity scores (ΔG in kcal/mol) using the default Vina scoring function. The top 5 poses for each ligand were retained for analysis. Mediation analysis A two-step MR mediation analysis was performed to assess whether immune cell traits mediate the causal pathway between cis-eQTLs and PCOS. GWAS summary statistics for 731 immune traits were obtained from the GWAS Catalog (accession range: GCST90001391-GCST90002121), derived from 3,757 European individuals [25]. These traits encompassed four measurement types (absolute cell counts [AC], median fluorescence intensity [MFI], morphological parameters [MP], relative cell counts [RC]) across seven immune cell panels (B cell, conventional dendritic cell [cDC], T cell maturation stages, monocyte, myeloid cell, TBNK, T regulatory cell [Treg]). First, causal effects of immune cell traits on PCOS were estimated using SNPs significantly associated with immune traits (P < 5E-08) as IVs. Subsequently, causal effects of cis-eQTLs on those immune cell traits exhibiting significant causal relationships with PCOS were evaluated. Where both effects were significant, immune traits were tested as potential mediators in the pathway from cis-eQTLs to PCOS. The Steiger test was implemented to mitigate reverse causality concerns. The proportion mediated by immune cells was calculated via the following formula (beta 1 × beta 2/beta all), where beta 1 represents the effect of cis-eQTLs on immune cells, beta 2 represents the effect of immune cells on PCOS, and beta all represents the effect of cis-eQTLs on PCOS. Results Mendelian randomization identified seven druggable genes significantly associated with polycystic ovary syndrome risk Utilizing a MR framework, we assessed the causal effects of 2,888 druggable genes on PCOS risk. Primary MR analysis (Fig. 2A) identified the top 10 genes associated with PCOS risk ranked by p-value, with NRBP1 exhibiting the strongest association (p = 1.61E-09). Following stringent correction for multiple testing using the False Discovery Rate (FDR), seven druggable genes demonstrated significant evidence for a causal relationship with PCOS risk (Fig. 2B, C; Table 1). To ensure the robustness of the causal associations identified at FDR significance, we performed a suite of sensitivity analyses for the seven FDR-significant genes. The results of these analyses, including tests for horizontal pleiotropy (MR-Egger intercept, MR-PRESSO) and heterogeneity (Cochran’s Q), are summarized in Table 2. Sensitivity analyses found no substantial evidence of horizontal pleiotropy or heterogeneity, supporting the robustness of these causal inferences. Leave-one-out sensitivity analysis confirmed that no single SNP was driving the observed associations for any of the seven FDR-significant genes (Supplementary Data 1). Directionally, four genes (NRBP1, LGR6, GHRL, IPP) exhibited positive causal effects, indicating an increased risk of PCOS, while three genes (CBLN3, VIPR1, TFRC) showed negative causal effects, suggesting a potential protective role against PCOS development. The complete list of genetic instruments (SNPs) used for all druggable genes, along with their F-statistics, is provided in Supplementary Table S3. Fig. 2Mendelian randomization analysis of druggable genes for Polycystic Ovary Syndrome (PCOS) Risk. A Volcano plots display the causal effect of 1,882 druggable genes on PCOS risk, evaluated using an inverse-variance weighted (IVW) method in a two-sample MR framework. The dashed horizontal line represents the threshold for nominal significance (p < 0.05). The top 10 genes with the smallest p-values are explicitly labeled. Manhattan plot (B) and forest plot (C) illustrate the seven druggable genes significantly associated with polycystic ovary syndrome risk after False Discovery Rate (FDR) correction. The dashed horizontal line represents the FDR significance threshold (q < 0.05) Mendelian randomization analysis of druggable genes for Polycystic Ovary Syndrome (PCOS) Risk. A Volcano plots display the causal effect of 1,882 druggable genes on PCOS risk, evaluated using an inverse-variance weighted (IVW) method in a two-sample MR framework. The dashed horizontal line represents the threshold for nominal significance (p < 0.05). The top 10 genes with the smallest p-values are explicitly labeled. Manhattan plot (B) and forest plot (C) illustrate the seven druggable genes significantly associated with polycystic ovary syndrome risk after False Discovery Rate (FDR) correction. The dashed horizontal line represents the FDR significance threshold (q < 0.05) Table 1MR estimates from inverse-variance weighted and weighted median methods for the seven FDR-significant druggable genesGeneInverse-variance weightedWeighted medianp valueFDRp valueFDRNRBP11.61E-094.64E-066.67E-050.0963LGR61.75E-075.06E-040.00161CBLN36.03E-070.00177.50E-050.0722VIPR17.26E-070.00210.00631GHRL7.73E-070.00220.00221TFRC8.60E-060.02480.00111IPP1.25E-050.03620.05941 Table 1MR estimates from inverse-variance weighted and weighted median methods for the seven FDR-significant druggable genesGeneInverse-variance weightedWeighted medianp valueFDRp valueFDRNRBP11.61E-094.64E-066.67E-050.0963LGR61.75E-075.06E-040.00161CBLN36.03E-070.00177.50E-050.0722VIPR17.26E-070.00210.00631GHRL7.73E-070.00220.00221TFRC8.60E-060.02480.00111IPP1.25E-050.03620.05941 MR estimates from inverse-variance weighted and weighted median methods for the seven FDR-significant druggable genes Table 2Sensitivity analyses for the seven FDR-significant druggable genes identified by Mendelian randomizationGeneHeterogeneity (Cochran’s Q)PleiotropyIVW_QIVW_pvalueMR Egger_QMR Egger_ p valueEgger interceptEgger intercept seEgger intercept pvalMR-PRESSO global test p valueNRBP112.980.839712.380.82700.02020.02620.44970.863LGR621.250.443719.360.49830.02820.02050.18480.513CBLN376.410.103076.290.0897−0.00330.01070.75930.105VIPR146.380.456845.840.43700.00940.01300.47380.471GHRL6.9720.64006.1910.62590.02730.03090.40240.692TFRC31.750.578631.670.53320.00380.01380.78560.653IPP18.910.706417.210.75170.02360.01810.20580.643 Table 2Sensitivity analyses for the seven FDR-significant druggable genes identified by Mendelian randomizationGeneHeterogeneity (Cochran’s Q)PleiotropyIVW_QIVW_pvalueMR Egger_QMR Egger_ p valueEgger interceptEgger intercept seEgger intercept pvalMR-PRESSO global test p valueNRBP112.980.839712.380.82700.02020.02620.44970.863LGR621.250.443719.360.49830.02820.02050.18480.513CBLN376.410.103076.290.0897−0.00330.01070.75930.105VIPR146.380.456845.840.43700.00940.01300.47380.471GHRL6.9720.64006.1910.62590.02730.03090.40240.692TFRC31.750.578631.670.53320.00380.01380.78560.653IPP18.910.706417.210.75170.02360.01810.20580.643 Sensitivity analyses for the seven FDR-significant druggable genes identified by Mendelian randomization Multi-omics analysis of seven druggable genes using summary-data-based MR analysis (SMR) and colocalization analyses To validate the MR-identified druggable genes, we integrated multi-omics analyses encompassing SMR, external validation, and colocalization. SMR analysis of genetically predicted gene expression for the seven FDR-significant genes revealed significant causal associations with PCOS risk at the abundance level under stringent criteria (p < 0.05 and PHEIDI > 0.05): NRBP1 (OR = 1.5383, 95% CI = 1.1939–1.9820, p = 8.67E-04), LGR6 (OR = 1.3442, 95% CI = 1.0709–1.6873, p = 0.0107), and GHRL (OR = 1.3391, 95% CI = 1.0116–1.7728, p = 0.0413) exhibited risk-increasing effects,. whereas CBLN3 demonstrated a protective effect (OR = 0.8605, 95% CI = 0.7917–0.9353, p = 4.09E-04) (Fig. 3A). Further validation using GTEx v8 eQTLs confirmed associations for NRBP1 (OR = 1.9119, 95% CI = 1.2717–2.8746, p = 0.0018), LGR6 (OR = 1.8234, 95% CI = 1.0744–3.0947, p = 0.026), and CBLN3 (OR = 0.8621, 95% CI = 0.7931–0.9372, p = 4.94E-04) (Fig. 3B). Colocalization analysis provided strong evidence for shared causal variants between PCOS susceptibility and both CBLN3 (PP.H4 = 86.65%) and NRBP1 (PP.H4 = 78.46%), exceeding the prespecified threshold (Fig. 3C). To visualize the co-localization results, we plotted the loci (Fig. 4). Collectively, this convergent multi-omics evidence prioritizes CBLN3 and NRBP1 as high-confidence therapeutic targets for PCOS. Fig. 3Multi-omics analysis of seven druggable genes using Summary-data-based MR (SMR) analysis and colocalization analyses. Forest Plot of summary-data-based MR (SMR) results for druggable genes associated with PCOS risk using blood eQTL summary statistics obtained from the eQTLGen Consortium (A) and Genotype-Tissue Expression (GTEx) project V8 (B). C Heatmap visualization of colocalization analysis assessing shared causal variants between PCOS risk and eQTLs of seven significant druggable genes Multi-omics analysis of seven druggable genes using Summary-data-based MR (SMR) analysis and colocalization analyses. Forest Plot of summary-data-based MR (SMR) results for druggable genes associated with PCOS risk using blood eQTL summary statistics obtained from the eQTLGen Consortium (A) and Genotype-Tissue Expression (GTEx) project V8 (B). C Heatmap visualization of colocalization analysis assessing shared causal variants between PCOS risk and eQTLs of seven significant druggable genes Fig. 4Regional genomic plots for conventional colocalization analysis. Colocalizations between PCOS GWAS dataset E4_PCOS and (A) blood CBLN3; (B) blood NRBP1 Regional genomic plots for conventional colocalization analysis. Colocalizations between PCOS GWAS dataset E4_PCOS and (A) blood CBLN3; (B) blood NRBP1 Tissue-specific Expression Quantitative Trait Locus (eQTL) analysis for MR Tissue-specific MR analysis using GTEx v10 expression quantitative trait loci revealed distinct organ-dependent causal effects of CBLN3 and NRBP1 expression on PCOS pathogenesis (Fig. 5A). Genetically predicted elevation of CBLN3 expression demonstrated consistent protective effects across multiple tissues, significantly reducing PCOS risk in whole blood (OR = 0.8586, 95% CI = 0.8138–0.9059, p = 2.45E-08), ovarian tissue (OR = 0.8926, 95% CI = 0.8509–0.9363, p = 3.21E-06), and uterine tissue (OR = 0.8713, 95% CI = 0.8128–0.9339, p = 1.01E-04). Conversely, increased NRBP1 expression in whole blood substantially elevated disease susceptibility (OR = 2.1003, 95% CI = 1.6673–2.6457, p = 2.97E-10), highlighting its tissue-specific risk-promoting role. Sensitivity analyses provided robust evidence supporting the primary MR assumptions, showing no significant horizontal pleiotropy or heterogeneity (Table S11). Critically, these tissue-specific causal findings were validated using an independent PCOS GWAS dataset (ID: GCST90044902) (Fig. 5B). Fig. 5Tissue-specific causal effects of genetically regulated gene expression on polycystic ovary syndrome risk revealed by Mendelian randomization analysis Forest plot of MR estimates for CBLN3 and NRBP1 expression across tissues using PCOS GWAS dataset E4_PCOS (A) and GCST90044902 (B) Tissue-specific causal effects of genetically regulated gene expression on polycystic ovary syndrome risk revealed by Mendelian randomization analysis Forest plot of MR estimates for CBLN3 and NRBP1 expression across tissues using PCOS GWAS dataset E4_PCOS (A) and GCST90044902 (B) Mendelian randomization analysis of DNA Methylation Quantitative Trait Loci (mQTLs) for CBLN3 and NRBP1 in relation to PCOS risk MR analysis of DNA mQTLs revealed significant locus-specific causal effects of CBLN3 and NRBP1 methylation on PCOS risk (Fig. 6). Specifically, a genetically predicted per unit increase in methylation at the CBLN3 CpG site cg05142272 demonstrated a significant causal effect of increasing PCOS risk (OR = 1.3043, 95% CI = 1.0190–1.6696, p = 0.0349). For the NRBP1 gene, increased methylation at cg05102552 was causally associated with elevated PCOS risk (OR = 1.2000, 95% CI = 1.1004–1.3085, p = 3.70E-05). Conversely, genetically predicted increases in methylation at four distinct NRBP1 loci exhibited significant protective effects, causally decreasing PCOS risk: cg05385684 (OR = 0.6040, 95% CI = 0.4771–0.7647, p = 2.80E-05), cg12243333 (OR = 0.2468, 95% CI = 0.1111–0.5483, p = 5.92E-04), cg15478930 (OR = 0.6020, 95% CI = 0.4518–0.8020, p = 5.52E-04), and cg20888435 (OR = 0.2098, 95% CI = 0.0881–0.4995, p = 4.19E-04). Sensitivity analyses provided robust evidence against significant horizontal pleiotropy (Table S8), supporting the validity of these causal inferences. These findings establish hypermethylation at the CBLN3 cg05142272 locus as a risk factor for PCOS, while revealing bidirectional causal effects of methylation (both risk-increasing and protective) at distinct loci within the NRBP1 gene, highlighting the complexity and gene-specific nature of epigenetic regulation in PCOS pathogenesis. Fig. 6Mendelian randomization analysis reveals divergent effects of CBLN3 and NRBP1 DNA methylation on PCOS risk Mendelian randomization analysis reveals divergent effects of CBLN3 and NRBP1 DNA methylation on PCOS risk Quantification of immune infiltration and correlation analysis Given established links between immune dysregulation and PCOS pathogenesis [26], we employed the CIBERSORT algorithm to quantify immune cell infiltration in the GSE34526 cohort. Comparative analysis revealed significantly reduced activated CD4 memory T cells in PCOS samples versus controls (Fig. 7A). Spearman correlation analysis demonstrated distinct immunomodulatory signatures: CBLN3 expression positively correlated with resting dendritic (r = 0.80, p = 0.0057) but inversely correlated with eosinophils (r = −0.72, p = 0.0194) and neutrophils (r = −0.66, p = 0.0440) (Fig. 7B). Similarly, NRBP1 expression showed positive associations with B cells naive (r = 0.79, p = 0.0098) and Dendritic cells activated (r = 0.67, p = 0.0394), while exhibiting negative correlations with resting CD4 memory T cells (r = −0.77, p = 0.0089) and T cells gamma delta (r = −0.66, p = 0.0440) (Fig. 7C). These findings delineate specific immune cell interactions for the CBLN3 and NRBP1. Fig. 7The landscape of immune cell infiltration and correlation analysis in GSE34526. A Analysis of immune-cell proportion comparisons between sepsis and healthy controls by CIBERSORT. Spearman correlation analysis of CIBERSORT between infiltrating immune cells and identified druggable genes CBLN3 (B) and NRBP1 (C) The landscape of immune cell infiltration and correlation analysis in GSE34526. A Analysis of immune-cell proportion comparisons between sepsis and healthy controls by CIBERSORT. Spearman correlation analysis of CIBERSORT between infiltrating immune cells and identified druggable genes CBLN3 (B) and NRBP1 (C) Single-cell RNA sequencing to identify distinct theca cell subpopulations and annotations in PCOS To independently validate the cell-type-specific expression of our MR-prioritized genes CBLN3 and NRBP1, we performed a secondary analysis of a publicly available scRNA-seq dataset from Zenodo containing PCOS and control theca cells. Following rigorous quality control, doublet removal, batch effect correction and manual marker annotation, we retained a total of 13,957 high-quality theca cells for downstream analysis. These cells were divided into four main subpopulations (C1–C4) through unsupervised clustering (Fig. 8A), each exhibiting distinct gene expression characteristics (Fig. 8B). Differential abundance analysis indicated notable shifts in theca subpopulation proportions between control and PCOS conditions (Fig. 8C). Specifically, the Theca C2 subpopulation was more abundant in control samples, whereas Theca C1 and Theca C4 were significantly enriched in PCOS samples, suggesting a potential association with the disease pathophysiology. Expression analysis of key genes of interest, CBLN3 and NRBP1, revealed distinct patterns across the subpopulations. CBLN3 expression was elevated in cluster C2 (Fig. 8D), whereas NRBP1 was highly expressed in clusters C1 and C3 (Fig. 8E). Pathway analysis delineated distinct functional specializations among the four theca cell subpopulations, implicating their coordinated roles in PCOS pathophysiology (Fig. 8F). Theca C1 was significantly enriched in pathways governing ovarian follicle development, female gamete generation, and response to external stimuli, suggesting a primary role in folliculogenesis. In contrast, Theca C2 was specifically associated with regulation of sterol transport, sterol homeostasis, and positive regulation of lipid transport, underscoring its function in maintaining sterol and lipid metabolic balance. Theca C3 exhibited a unique profile enriched in positive regulation of fibroblast migration, steroid biosynthetic process, and endothelial cell chemotaxis. Most notably, Theca C4 was robustly enriched for processes related to cellular response to estrogen, positive regulation of blood circulation, and neutrophil chemotaxis.Fig. 8Single-cell RNA sequencing identification of the expression patterns of CBLN3 and NRBP1 in theca cells. A UMAP projection of 13,957 high-quality theca cells after quality control and integration, colored by the four identified subpopulations (C1-C4). B Heatmap displaying the marker genes across the four theca cell subpopulations (C1-C4). C Bar plot showing the proportional abundance of each theca cell subpopulation (C1-C4) in control versus PCOS samples. D The expression distribution of CBLN3 in four theca cell subpopulations (C1-C4). E The expression distribution of NRBP1 in four theca cell subpopulations (C1-C4). F Bar plot of pathway analysis for each theca cell subpopulation Single-cell RNA sequencing identification of the expression patterns of CBLN3 and NRBP1 in theca cells. A UMAP projection of 13,957 high-quality theca cells after quality control and integration, colored by the four identified subpopulations (C1-C4). B Heatmap displaying the marker genes across the four theca cell subpopulations (C1-C4). C Bar plot showing the proportional abundance of each theca cell subpopulation (C1-C4) in control versus PCOS samples. D The expression distribution of CBLN3 in four theca cell subpopulations (C1-C4). E The expression distribution of NRBP1 in four theca cell subpopulations (C1-C4). F Bar plot of pathway analysis for each theca cell subpopulation Mediation analysis Building upon evidence linking immune dysregulation to PCOS pathogenesis, we performed a two-sample MR mediation analysis to assess the mediating role of immune cell signatures in causal pathways from MR-prioritized genes (CBLN3 and NRBP1) to PCOS. Initial screening identified 37 immunophenotypes with putative causal effects on PCOS (IVW: p < 0.05), including 11 B-cell, 7 TBNK, 6 dendritic cell, and 13 other phenotypes (Fig. 9A), with no evidence of significant horizontal pleiotropy and heterogeneity (Table S13). Genetically proxied CBLN3 expression showed significant associations with eight distinct immunophenotypes (Fig. 9B) without evidence of significant horizontal pleiotropy (MR-Egger intercept: p > 0.05) (Table S15), the strongest being with NKT %lymphocyte (OR = 1.1536, 95% CI = 1.1018–1.2079, FDR = 4.10E-08). Similarly, genetically proxied NRBP1 expression correlated positively with seven immunophenotypes (Fig. 9B), with the strongest effect observed for CD27 on IgD- CD38dim (OR = 1.2987, 95% CI = 1.1422–1.4766, FDR = 0.0024) and no evidence of significant horizontal pleiotropy and heterogeneity detected (Table S17). Mediation analysis revealed that CBLN3 exerted a protective effect on PCOS risk through NKT %T cell (Fig. 9C) and NKT %lymphocyte (Fig. 9D), while NRBP1 increased PCOS risk via elevated CD24 on IgD + CD24+ (Fig. 9E). Notably, NKT %lymphocyte mediated 14.38% of the total causal effect of CBLN3 on PCOS (mediated effect: −0.0203; p = 0.0258). Validation using an independent GWAS dataset (ID: GCST90044902) confirmed that CBLN3 was causal relationship with increased PCOS risk (IVW-MR OR = 0.9433, 95% CI = 0.9058–0.9823; p = 0.0048) NKT %lymphocyte was inversely associated with PCOS risk (IVW p < 0.05; Fig. 10A), with no horizontal pleiotropy (MR-Egger intercept: p > 0.05) or heterogeneity (Cochran’s Q: p > 0.05). The proportion of CBLN3’s total effect mediated by NKT %lymphocyte increased to 24.35% (mediated effect: −0.0142, p = 0.0153; Fig. 10B), reinforcing its role as a key mediator. Fig. 9Mendelian randomization analysis reveals immune cell mediators of CBLN3 and NRBP1 effects on polycystic ovary syndrome risk. A Circular heatmap displays the 37 immune cell phenotypes causally associated with PCOS risk. B Forest Plot of MR results for the significant associations between genetically CBLN3 and NRBP1 and immunophenotypes. Schematic diagram of NKT %T cell (C), NKT %lymphocyte (D), and CD24 on IgD + CD24+ (E) mediation effect Mendelian randomization analysis reveals immune cell mediators of CBLN3 and NRBP1 effects on polycystic ovary syndrome risk. A Circular heatmap displays the 37 immune cell phenotypes causally associated with PCOS risk. B Forest Plot of MR results for the significant associations between genetically CBLN3 and NRBP1 and immunophenotypes. Schematic diagram of NKT %T cell (C), NKT %lymphocyte (D), and CD24 on IgD + CD24+ (E) mediation effect Fig. 10Validation analysis for the role of NKT %lymphocyte in the association between CBLN3 and PCOS risk (GWAS ID: GCST90044902). A Forest Plot of MR results for the significant associations between genetically NKT %lymphocyte and PCOS risk. B Schematic diagram of NKT %lymphocyte mediation effect Validation analysis for the role of NKT %lymphocyte in the association between CBLN3 and PCOS risk (GWAS ID: GCST90044902). A Forest Plot of MR results for the significant associations between genetically NKT %lymphocyte and PCOS risk. B Schematic diagram of NKT %lymphocyte mediation effect Screening of candidate compounds and molecular docking Computational drug discovery targeting the MR-validated therapeutic candidates CBLN3 and NRBP1 was performed through Comparative Toxicogenomics Database (CTD) screening and molecular docking. Primary CTD analysis identified 32 candidate compounds targeting CBLN3 and 54 targeting NRBP1, with nine dual-target compounds demonstrating high-affinity binding to both proteins (binding energy ≤ −5.0 kcal/mol) (Table 3). Molecular docking simulations revealed specific binding poses and critical amino acid interactions for these compounds with CBLN3 (Fig. 11) and NRBP1 (Fig. 12), providing a computational structural model for potential target engagement and establishing a foundation for future experimental validation. Table 3Molecular docking results of CBLN3 protein (AlphaFold ID: Q6UW01), NRBP1 (AlphaFold ID: Q9UHY1) and drugsDrugPubChem IDVina scoreCBLN3NRBP1Acetaminophen1983−5.4−5.7Aflatoxin B1186,907−6.7−7.7Benzo(a)pyrene2236−8.2−8.3Bisphenol A6623−6.3−6.4Chlorpyrifos2730−5.5−5.1Folic Acid135,398,658−7.6−7.6Okadaic Acid446,512−8.5−9.2Rotenone6758−7.0−7.6Triptonide65,411−7.3−7.5 Table 3Molecular docking results of CBLN3 protein (AlphaFold ID: Q6UW01), NRBP1 (AlphaFold ID: Q9UHY1) and drugsDrugPubChem IDVina scoreCBLN3NRBP1Acetaminophen1983−5.4−5.7Aflatoxin B1186,907−6.7−7.7Benzo(a)pyrene2236−8.2−8.3Bisphenol A6623−6.3−6.4Chlorpyrifos2730−5.5−5.1Folic Acid135,398,658−7.6−7.6Okadaic Acid446,512−8.5−9.2Rotenone6758−7.0−7.6Triptonide65,411−7.3−7.5 Molecular docking results of CBLN3 protein (AlphaFold ID: Q6UW01), NRBP1 (AlphaFold ID: Q9UHY1) and drugs Fig. 11Docking analysis of amino acid residues of CBLN3 protein (AlphaFold ID: Q6UW01) and Acetaminophen (A), Aflatoxin B1 (B), Benzo(a)pyrene (C), Bisphenol A (D), Chlorpyrifos (E), Folic Acid (F), Okadaic Acid (G), Rotenone (H), Triptonide (I) Docking analysis of amino acid residues of CBLN3 protein (AlphaFold ID: Q6UW01) and Acetaminophen (A), Aflatoxin B1 (B), Benzo(a)pyrene (C), Bisphenol A (D), Chlorpyrifos (E), Folic Acid (F), Okadaic Acid (G), Rotenone (H), Triptonide (I) Fig. 12Docking analysis of amino acid residues of NRBP1 (AlphaFold ID: Q9UHY1) and Acetaminophen (A), Aflatoxin B1 (B), Benzo(a)pyrene (C), Bisphenol A (D), Chlorpyrifos (E), Folic Acid (F), Okadaic Acid (G), Rotenone (H), Triptonide (I) Docking analysis of amino acid residues of NRBP1 (AlphaFold ID: Q9UHY1) and Acetaminophen (A), Aflatoxin B1 (B), Benzo(a)pyrene (C), Bisphenol A (D), Chlorpyrifos (E), Folic Acid (F), Okadaic Acid (G), Rotenone (H), Triptonide (I) Discussion This study establishes a multi-omics MR framework, integrating genomic, epigenomic, and scRNA-seq, to systematically identify CBLN3 and NRBP1 as high-priority therapeutic targets for PCOS. MR analysis of 2,888 druggable genes revealed seven FDR-significant targets, with NRBP1 exhibiting the strongest risk effect and CBLN3 demonstrating robust protective effects; these findings were robustly validated through SMR, colocalization, mQTL MR, and independent GWAS replication. Single-cell sequencing identified four subtypes of PCOS theca cells, and CBLN3 expressed predominantly in the Theca C2 subpopulation and NRBP1 enriched in Theca C1 and Theca C3. Significantly, computational drug screening identified nine high-affinity lead compounds (ΔG≤−5.0 kcal/mol) targeting both proteins. Our work not only provides multi-dimensional causal insights into PCOS pathogenesis but also, for the first time, delivers a systematic roadmap bridging causal human genetics to targeted drug discovery, offering novel avenues for precision therapy in PCOS. CBLN3 (Cerebellin 3 precursor), belonging to the precerebellin family, is a secreted glycoprotein characterized molecularly by a cerebellin motif and a C-terminal C1q domain [27]. CBLN3 is incapable of forming stable homomeric oligomers and cannot be secreted alone; it requires binding to CBLN1 to evade the endoplasmic reticulum retention signal via a “hide-and-run” mechanism for co-secretion [28, 29]. The two proteins reciprocally regulate each other’s degradation and secretion, evidenced by the near absence of CBLN3 in CBLN1 knockout mice and a significant increase (~ 6-fold) in brain CBLN1 levels in CBLN3 knockout mice [29, 30]. RT-PCR confirmed that CBLN3 transcripts can be detected in the rat endocrine pancreas, adrenal zona glomerulosa, pituitary gland, thyroid gland, testes, ovaries, and other tissues, suggesting its possible involvement in metabolism or hormone regulation [31, 32]. CBLN3’s expression in peripheral tissues like pancreatic islets (inhibiting insulin secretion) [33] and osteoblasts (regulating differentiation) [34] hints at broader endocrine/metabolic functions. Research on CBLN1 in PCOS is relatively limited. Clinical studies show that in PCOS patients undergoing ovarian stimulation via the GnRH antagonist protocol, the concentration of CBLN1 in follicular fluid is significantly higher than in their own serum, suggesting that CBLN1 is primarily synthesized locally within the follicle, thereby participating in the regulation of steroid hormone production and influencing follicular growth [35]. Bioinformatic analyses have identified CBLN1 as a key diagnostic biomarker for PCOS correlated with immune microenvironment alterations (e.g., M2 macrophage infiltration), indicating its potential involvement in disease pathogenesis through the modulation of the immune microenvironment [36]. This study provides the first robust genetic evidence from MR analyses supporting a protective role for elevated CBLN3 expression against PCOS risk, with this association being significant across blood, ovary, and uterus tissues, positioning CBLN3 as a novel cross-tissue regulator potentially mitigating PCOS pathogenesis. This core MR finding is strongly corroborated by multiple lines of evidence: immune cell correlation analyses suggested its influence on innate immunity (evidenced by negative correlations with eosinophils/neutrophils and a positive correlation with activated dendritic cells); strong colocalization evidence coupled with mediation analysis provide genetic support for an immune-mediated mechanism underlying CBLN3’s effect. Single-cell transcriptomics further revealed that CBLN3 is specifically enriched in the Theca C2 subpopulation, a cell state characterized by sterol homeostasis and lipid transport pathways, providing a plausible cellular basis for its protective metabolic function. Interestingly, for CBLN3, higher gene body methylation (cg05142272) was associated with increased PCOS risk, while its gene expression was associated with decreased risk, aligning with the canonical model where DNA methylation often negatively regulates gene expression. Collectively, the strong concordance of the above evidence suggests that CBLN3 influences PCOS pathophysiology, likely via metabolic-immune crosstalk within a specific theca cell compartment. This positions CBLN3 as a compelling priority target candidate warranting further mechanistic and therapeutic investigation for PCOS. NRBP1 (Nuclear Receptor Binding Protein 1), a highly conserved adaptor protein ubiquitously expressed across human cell types, regulates diverse biological processes including the Wnt/β-catenin signaling pathway, vesicular trafficking, autophagy, cell proliferation/differentiation, and interactions with nuclear receptors [37]. Crucially, NRBP1 exhibits context-dependent and even opposing functions across different cancer types and molecular subtypes; for instance, it acts as a tumor suppressor in most breast cancers but promotes metastasis in triple-negative breast cancer [38–41], underscoring that its biological impact is dictated by specific cellular and signaling milieus. Furthermore, NRBP1 promoter hypomethylation, which attenuates binding of the transcriptional repressor TFAP2A leading to its upregulation, has been associated with increased gout risk [42]. Significantly, despite these established roles in other pathologies, the expression patterns, regulatory mechanisms, and functional contribution of NRBP1 in PCOS remain entirely unexplored. Given the dysregulation of cAMP-dependent signaling in theca cells of PCOS ovaries [43, 44], and prior reports implicating NRBP1 in regulating the ERK-CREB-BDNF signaling axis [45], NRBP1 may contribute to the hyperandrogenism characteristic of PCOS by interfering with the cAMP signaling pathway in theca cells. This study establishes for the first time that elevated NRBP1 increases PCOS risk through MR and SMR analyses, with colocalization confirming a causal association at its genomic locus. Critically, spatially distinct epigenetic regulation was observed: hypermethylation at CpG site cg05102552 activates NRBP1 transcription, whereas hypermethylation at cg05385684, cg12243333, cg15478930, and cg20888435 significantly reduces PCOS risk, suggesting its localization within a silencer/repressor element and indicating that targeting specific CpG sites may yield opposing clinical outcomes. Importantly, the opposing effect directions of DNA methylation and NRBP1 expression on PCOS risk observed in our MR analyses strongly imply that DNA methylation likely influences PCOS susceptibility by regulating the expression of NRBP1. scRNA-seq data crucially localized NRBP1 overexpression specifically to disease-associated theca cell subpopulations (Theca C1 and Theca C3), which are functionally linked to ‘follicle development’ and ‘steroid biosynthetic process’, respectively. This provides a coherent cellular mechanism for its risk effects, potentially involving aberrant folliculogenesis and the excessive androgen production in the PCOS ovary. Collectively, these findings implicate NRBP1 as a key modulator in PCOS pathogenesis. Future studies will validate NRBP1 expression levels (mRNA/protein) in PCOS-affected ovarian and uterine tissues and utilize specific NRBP1 knockout models to assess resulting metabolic and reproductive phenotypes. Accumulating evidence has documented the critical role of inflammatory immune mechanisms in the pathogenesis of PCOS. The significant reduction in activated CD4⁺ memory T cells within ovarian granulosa cells observed in our current study aligns with prior histological findings by Wu et al. [46], who reported a 60% decrease in CD45RO⁺ cells (activated/memory T lymphocytes) in the theca layer of PCOS follicles. This convergence strongly suggests that T lymphocytes play pivotal roles in local pathological mechanisms of PCOS [47]. Notably, CD4⁺ cells were identified in androgenized ovaries, particularly within the granulosa cell layer of cysts exhibiting high VCAM-1 expression [48]. Bioinformatics analyses have further indicated that activated CD4⁺ memory T cells are significantly decreased in PCOS and show positive correlation with CBLN1 expression [36]. Crucially, our analysis provides the evidence: NRBP1 expression demonstrates a negative causal relationship with resting CD4⁺ memory T cells and T cells gamma delta. Moreover, mediation analysis reveals that the protective effect of CBLN3 on PCOS risk operates through NKT %lymphocyte. NKT cells, as innate-like T lymphocytes with semi-invariant TCRs restricted by CD1d, serve as a bridge between innate and adaptive immunity [49]. Recent MR studies indicate that NKT cell activity centers influence PCOS susceptibility via metabolic pathways [50], which complements our findings on the causal pathway linking CBLN3 to PCOS risk through NKT cells. Collectively, integrating histological evidence and our genetic causal inference, we propose that dysregulation of T-cell subsets-particularly CD4⁺ memory T cells and NKT cells-in terms of local infiltration and quantitative control represents a core pathological feature of PCOS. The molecular docking results, while computationally derived and requiring experimental validation, serve as a proof-of-concept for the druggability of NRBP1 and CBLN3. In the current study, molecular docking suggested that several compounds, including acetaminophen, aflatoxin B1, benzo(a)pyrene, bisphenol A, chlorpyrifos, folic acid, okadaic acid, rotenone, and triptonide, could exhibit high-affinity binding to NRBP1 and CBLN3. This computationally predicts that these proteins might be modulated by diverse chemicals, potentially linking them to PCOS progression. These in silico findings resonate with existing epidemiological and mechanistic evidence. Computer simulation studies suggest that acetaminophen can bind to the aromatase receptor and has been proposed as a potential alternative to letrozole for ovulation induction [51]. Benzo[a]pyrene, as a polycyclic aromatic hydrocarbon pollutant, directly interferes with ovarian folliculogenesis and angiogenesis, effects mechanistically linked to hyperandrogenism and insulin resistance in PCOS [52]. Numerous human cross-sectional studies, animal experiments, and systematic reviews consistently indicate that bisphenol A exposure is strongly associated with PCOS [53–55]. Folic acid optimizes pregnancy outcomes by enhancing insulin sensitivity [56] and vascular endothelial function [57]. Notably, rotenone, a mitochondrial complex I inhibitor, shares mechanistic parallels with metformin which suppresses key androgen-synthesizing enzymes to reduce ovarian testosterone levels, providing novel evidence for shared metabolic targets between environmental toxins and therapeutic agents [58]. Collectively, these in silico findings position the NRBP1 and CBLN3 as a central hub integrating environmental exposures with metabolic-reproductive disruptions in PCOS, thereby revealing new avenues for mechanistic research and targeted interventions. This study established a multi-omics causal inference framework by integrating GWAS, epigenome-wide, and single-cell transcriptomic data, effectively addressing confounding biases inherent in conventional observational studies and significantly enhancing the reliability of causal inferences. This cross-omics integration strategy not only strengthens the biological plausibility of instrumental variables but also provides novel insights into the potential mechanistic pathways linking exposures to outcomes. Despite rigorous methodological design, several limitations warrant cautious interpretation. First, although we controlled for known confounders through multiple approaches including MR-Egger regression, genetic instruments may still influence outcomes through unmeasured biological pathways. Second, current analyses focused on European populations, potentially limiting the generalizability of our conclusions to African, Hispanic, and other populations. Finally, although the causal relationships inferred from MR were validated bioinformatically, it has not been validated through in vivo or in vitro experiments. Conclusions Through a purely computational, multi-omics approach, this study indicates that CBLN3 and NRBP1 are respectively a potential protective gene and a risk-enhancing gene for PCOS. These genes exhibit multifaceted roles across genetic, epigenetic, tissue-specific, and immunomodulatory dimensions, and are thus highlighted as priority candidate therapeutic targets. This discovery not only broadens the understanding of PCOS pathogenesis but also provides novel molecular leads for biomarker screening and drug development. Subsequent work should further expand sample sizes, validate related pathways through in vivo and in vitro experiments, and advance the translation of causal mechanisms towards clinical application. Supplementary Information Supplementary Material 1: Table S1: Details of the GWAS used in the study. Table S2: eQTL for druggable genes. Table S3: Data used in the MR analysis to assess the causal effects of druggable genes on PCOS. Table S4: The causal effects of druggable Genes on PCOS risk. Table S5: SMR analysis to assess the causal effects of druggable genes from eQTLGen consortium on PCOS risk. Table S6: SMR analysis to assess the causal effects of druggable genes from GTEx (V8) on PCOS risk. Table S7: Data used in the MR analysis to assess CBLN3 and NRBP1 gene methylation on PCOS risk. Table S8: MR analysis to assess CBLN3 and NRBP1 gene methylation on PCOS risk. Table S9: Co-localization analysis. Table S10: Data used from tissue-specific expression quantitative trait locus analysis for MR. Table S11: Tissue-specific expression quantitative trait locus analysis for MR. Table S12: Data used in the MR analysis to assess the causal effects of immunophenotypes on PCOS. Table S13: The causal effects of immunophenotypes on PCOS risk. Table S14: Data used in the MR analysis to assess the causal effects of CBLN3 on immunophenotypes. Table S15: The causal effects of CBLN3 on immunophenotypes associated with PCOS. Table S16: Data used in the MR analysis to assess the causal effects of NRBP1 on immunophenotypes. Table S17: The causal effects of NRBP1 on immunophenotypes associated with PCOS. Supplementary Material 1: Table S1: Details of the GWAS used in the study. Table S2: eQTL for druggable genes. Table S3: Data used in the MR analysis to assess the causal effects of druggable genes on PCOS. Table S4: The causal effects of druggable Genes on PCOS risk. Table S5: SMR analysis to assess the causal effects of druggable genes from eQTLGen consortium on PCOS risk. Table S6: SMR analysis to assess the causal effects of druggable genes from GTEx (V8) on PCOS risk. Table S7: Data used in the MR analysis to assess CBLN3 and NRBP1 gene methylation on PCOS risk. Table S8: MR analysis to assess CBLN3 and NRBP1 gene methylation on PCOS risk. Table S9: Co-localization analysis. Table S10: Data used from tissue-specific expression quantitative trait locus analysis for MR. Table S11: Tissue-specific expression quantitative trait locus analysis for MR. Table S12: Data used in the MR analysis to assess the causal effects of immunophenotypes on PCOS. Table S13: The causal effects of immunophenotypes on PCOS risk. Table S14: Data used in the MR analysis to assess the causal effects of CBLN3 on immunophenotypes. Table S15: The causal effects of CBLN3 on immunophenotypes associated with PCOS. Table S16: Data used in the MR analysis to assess the causal effects of NRBP1 on immunophenotypes. Table S17: The causal effects of NRBP1 on immunophenotypes associated with PCOS. Supplementary Material 2. Data 1: MR plots for the seven FDR-significant druggable genes. Supplementary Material 2. Data 1: MR plots for the seven FDR-significant druggable genes. Supplementary Material 3. Supplementary Code Supplementary Material 3. Supplementary Code