Identification of DHX58 as a potential therapeutic target for ovarian cancer through multi-omics Mendelian randomization and transcriptomic data analysis

Abstract

Background

Ovarian cancer (OC), a prevalent malignancy of the female reproductive system, is significantly impacted by drug resistance, which reduces treatment efficacy and worsens patient survival. Therefore, identifying broader therapeutic targets is crucial for improving treatment outcomes in OC patients.

Methods

The research project utilized Mendelian randomization (MR) analysis and summary data-based Mendelian randomization (SMR) to identity potential treatment targets for OC. The eQTLGen collaboration provided the cis-expression quantitative trait locus data, while the Genotype-Tissue Expression (GTEx) project v8 provided the ovarian tissue eQTL data. Protein quantitative trait locus (pQTL) was generated by the INTERVAL, Fenland, and SCALLOP studies. The open-access genome-wide association study (ID: GCST90018888), involving a European population, provided the summary statistics for OC. Additionally, colocalization analyses were employed to determine wether the same SNPs drive the expression of genes and OC risk, and transcriptomic and Single-cell RNA sequencing data analysis was utilized to further investigate the targets clinical significance of these in OC. To evaluate its potential for drug development, drug prediction and molecular docking analyses were also carried out.

Results

In the SMR analysis, DHX58 was significantly associated with all three omics levels. The MR results indicated a correlation between elevated OC risk and DHX58 (OR = 1.300; 95% CI = 1.120–1.510; P = 0.000). Colocalization analysis further supported this finding (PP.H4 = 0.949). Our results revealed that DHX58 expression was significantly associated with patient prognosis, enriched immune-related pathways, and distinct immune cell infiltration patterns. PheWAS at the gene level revealed no significant correlation between DHX58 and additional phenotypes. The molecular docking results revealed strong interactions between DHX58 and other drugs and proteins with known structural information.

Conclusion

This study confirmed a significant correlation between DHX58 and OC, indicating that DHX58 could be a viable target for OC treatment. Drugs targeting DHX58 are likely to have higher success rates in clinical trials, providing new directions for future clinical and basic research.

Supplementary Information

The online version contains supplementary material available at 10.1186/s13048-026-02084-z.

Introduction

Ovarian cancer (OC) is a common malignant tumor of the female reproductive system. In 2020, OC was estimated to account for 3.7% of all cancer diagnoses and 4.7% of cancer-related deaths [1]. Currently, the incidence of OC is declining in Scandinavia and North America, whereas it is increasing in some regions of Eastern Europe and Asia [2]. Despite advancements in OC treatment, approximately 81% of OC patients are diagnosed at an advanced stage, and most develop recurrent disease that becomes resistant to chemotherapy, leading to a global 5-year survival rate of only approximately 50%[3]. Drug resistance remains a significant barrier in OC treatment and is a key factor in poor prognosis [4]. Therefore, identifying broader drug targets is crucial for improving treatment outcomes in OC patients.

In the process of drug development, accurately identifying potential therapeutic drugs for a disease and evaluating their impact on disease progression are essential. However, traditional drug development processes are limited by inefficiency and high costs. As a result, the integration of genomic technologies to accelerate the identification of novel targets for therapy has become a critical strategy to increase drug development efficiency [5]. By combining data from molecular quantitative trait loci (molQTL), such as expression quantitative trait loci (eQTL) or protein quantitative trait loci (pQTL), with data from genome-wide association studies (GWAS), causal inference methods can be used to identify target genes associated with disease risk variations [6]. Mendelian randomization (MR) infers causal links among exposures and outcomes by using genetic variations as instrumental variables, providing insights into drug efficacy while mitigating confounding biases inherent in traditional observational studies [7–11]. This method leverages genetic variants as instrumental variables to evaluate the causal effects of gene perturbation, enabling the prediction of both the efficacy and safety profiles of therapeutic targets. This strategy not only de-risks the drug development process but also facilitates reliable target prioritization [12], drug repurposing, and biomarker discovery [13, 14].

In this study, to directly infer causality between gene expression or protein levels associated with OC, we selected genetic variants related to eQTLs and pQTLs as instrumental variables (IVs). Relying solely on SMR may not be sufficient to identify reliable proteins in pathogenic pathways. To ensure the reliability of the results, subsequent MR, colocalization, HEIDI tests, differential gene expression analysis, and transcriptomic data analysis were conducted. Finally, to better evaluate possible therapeutic candidates for OC target genes, drug prediction and molecular docking were employed.

Method

Exposure data

The eQTLs data were sourced from the eQTLGen Consortium (https://eqtlgen.org/). In general, the eQTLGen dataset includes information on 31,684 cis-eQTLs and 16,987 genes from sample blood taken from the majority of healthy Europeans. The original publication provided a detailed description of the data [15].

The ovarian tissue eQTLs data were obtained from The Genotype-Tissue Expression (GTEx) Project v8. This study was established to characterize the genetic influences on the human tissue transcriptome and relate these regulatory mechanisms to traits and diseases. RNA sequencing specimens coming from 49 tissues obtained from 838 the post-mortem donors were examined in this study. The primary paper provides a thorough explanation of the data [16].

The pQTL data were sourced from the INTERVAL study [17], Fenland study [18], and SCALLOP study [19]. The INTERVAL study included plasma samples from 3,622 healthy participants and identified 1,927 genotype-protein associations (pQTLs), including 1,104 trans-acting loci. The Fenland study measured 4,775 plasma protein targets from 10,708 individuals of European descent, identifying 10,674 genotype-protein associations (P < 1.004 × 10–11) across 2,548 genomic regions (1,097 novel). The SCALLOP study identified 451 pQTLs for 85 proteins in over 30,000 individuals. The original articles provide a complete description of the data.

Outcome data

The summary data for OC in our research were obtained from a publicly accessible GWAS (ID: GCST90018888), which conducted a GWAS on European individuals (Ncase = 1,588, Ncontrol = 244,932) and identified 24,137,758 independent SNPs. The IEU OpenGWAS project is the source of the summary statistics utilized in this investigation (https://gwas.mrcieu.ac.uk/).

Summary-data-based mendelian randomization and heidi analysis

Summary-data-based Mendelian Randomization (SMR) is a tool that applies GWAS and eQTL summary data to determine if gene expression mediates the effect size of an SNP on a phenotype and to investigate causal relationships between exposure factors (such as gene expression, protein levels, etc.) and diseases [20]. In this study, SMR was used to examine the associations between eQTLs, eQTL_GTEx_Ovary, pQTLs, and OC. Nevertheless, the observed associations in SMR do not just undoubtedly imply that gene expression and traits are influenced through the identical underlying caused variants because the top-associated cis-eQTL in linkage disequilibrium (LD) contains two distinct causal variants, one of which affects gene expression and the other affects trait variance. To address this, HEIDI testing was used to exclude the influence of LD [21]. When the HEIDI p-value > 0.05, it indicates that there is no significant heterogeneity in the associations between the instrumental variables (IVs) and the target gene, supporting the hypothesis of a single causal effect, and further validating the robustness of the MR results. SMR analysis in this study was performed using an online website and database [22] (https://yanglab.westlake.edu.cn/). This study involved secondary analysis of published data from human individuals and therefore did not require ethical approval. Figure 1 provides a schematic summary of the analysis.

Fig. 1.

Flow diagram of the analysis

Mendelian randomization

The robustness of the SMR results may be affected by horizontal pleiotropy, in which genetic variations impact the phenotype by the pathways independent of the target gene expression. To better control for horizontal pleiotropy and improve the reliability of the SMR results, we employed MR to further assess the causal association among eQTLs and OC. We used the R packages (TwoSampleMR) and R software (version 4.3.2). To investigate the causal connection among eQTLs and OC, we employed IVW, MR-Egger and weighted median. Since IVW regression ignores the intercept term and depends only on the outcome’s variance, it offers the most compelling estimates when IV’s directional pleiotropy is absent, making it the main analytical method because it has the highest statistical strength under the presumption that all instruments have validity [23]. When the results from the aforementioned MR methods are inconsistent, IVW is used as the primary result [24]. The first criterion for SNPs to be used as instrumental variables is the genome-wide significance threshold of P < 5 × 10^− 8. Additionally, we set the linkage disequilibrium (LD) threshold to r² < 0.001 to ensure the independence of the SNPs as instrumental variables [23]. SNPs with r² > 0.001 (clumping window of 10,000 kb) were excluded if they were correlated with other SNPs or had higher p-values. Finally, Cochran’s Q and MR-Egger intercept tests were performed for heterogeneity and pleiotropy tests to further improve the reliability and accuracy of causal inference [25].

Colocalization analysis

We employed the R package coloc to perform the colocalization analysis of OC hazards genes that were significant across all three cohorts [26]. Colocalization analysis pertains to one among the five theories outlined: PPH0, the SNP shows no correlation with any trait; PPH1, the SNP shows a correlation with the expression of genes, yet it does not indicate a risk for OC; PPH2, the SNP shows a correlation with OC risk but not the expression of genes; PPH3, the SNP shows a correlation with both OC risk and the expression of genes but is influenced by distinct SNPs; and PPH4, the SNP shows a correlation with both OC risk and the expression of genes and is influenced by the same SNP [27]. PPH4 > 0.80 was used as the level of colocalization significance level. Possible therapeutic target genes could be those that colocalized with OC.

Transcriptomic data analysis

The TCGA database provided the OC data, which included RNA-seq data and clinical information about overall survival (OS), disease-specific survival (DSS), disease-free interval (DFI), progression-free interval (PFI) and survival outcomes including age and stage (https://gdc.cancer.gov/). The GTEx database provided the normal ovarian tissue data, with 88 °C samples (https://xenabrowser.net/). To correct for batch effects, all TPM values were first log2-transformed log2(TPM + 1). Batch correction between the TCGA and GTEx samples was then performed using the ComBat function from the sva R package, with the data source (TCGA vs. GTEx) specified as the batch variable. The effectiveness of the correction was evaluated using principal component analysis (PCA) before and after adjustment. The batch-corrected expression matrix was used in all subsequent analyses. The R package Survival and Survminer was utilized to survival analysis. To explore the biological pathways associated with DHX58, Gene Set Enrichment Analysis (GSEA) was performed. The c2.cp.kegg.v7.4.symbols.gmt subset was downloaded from the Molecular Signatures Database (MSigDB) to evaluate relevant pathways and molecular mechanisms. Enrichment scores (ES), normalized enrichment scores (NES), false discovery rates (FDR), and nominal p-values were calculated to determine statistical significance. To assess the correlation between DHX58 expression and tumor immune microenvironment, the CIBERSORT algorithm was employed to estimate the relative proportions of 22 immune cell types in OC samples. ImmuneScore-based stratification was conducted to examine survival differences across different levels of immune infiltration. Additionally, Pearson correlation analysis was performed to evaluate the relationships between DHX58 expression and specific immune cell subtypes.

Single-cell analysis

Single-cell RNA sequencing data were obtained from the Gene Expression Omnibus (GEO) under accession number GSE184880, comprising five normal ovarian tissue samples and seven OC samples. Quality control was performed using Seurat (v5.4). Cells expressing fewer than 500 genes or more than 4,000 genes were excluded to remove low-quality cells and potential doublets. Cells with mitochondrial gene expression exceeding 20% were also removed. After filtering, gene expression data were normalized and scaled, and highly variable genes were identified for downstream analyses. To correct for batch effects across samples, Harmony was applied based on the normalized expression matrix. PCA was performed on highly variable genes, and the Harmony-corrected embeddings were used for UMAP dimensionality reduction. Cells were clustered using a graph-based approach implemented in Seurat (Supplementary Figures S6). DHX58 expression across cell types and differences in cellular composition between normal and OC were assessed.

Phenome-wide association analysis

To assess the horizontal pleiotropy and possible adverse effects of the possible therapeutic targets, we performed a phenome-wide association study (PheWAS) on the AstraZeneca PheWAS Portal [28] (https://azphewas.com/). Data from approximately 15,500 binary and 1,500 continuous phenotypes from the UK Biobank were used in the initial investigation. The initial article presents the whole construction procedure [28]. This comprehensive PheWAS analysis aids in determining the efficacy and safety of medication targets as well as the genetic foundation of complex features.

Candidate drug prediction

We evaluated protein-drug combinations using the Drug Signatures Database (DSigDB, http://dsigdb.tanlab.org/DSigDBv1.0/) to determine whether genes that are targeted can be effective therapeutic targets [29]. DSigDB is a large database containing 22,527 gene sets and 17,389 different compounds associated with 19,531 genes, which is critical for determining whether the identified genes can be exploited as therapeutic targets. We uploaded the identified target genes to DSigDB and predicted candidate drugs that target these genes to evaluate their pharmacological activity.

Molecular docking

To evaluate the druggability of candidate drugs and look more closely at how they affect the target genes, molecular docking techniques were used in this study. This simulation method allows us to analyze the binding energy and interaction characteristics between the candidate drugs and target proteins at the atomic level. By screening ligands with excellent binding ability and ideal interaction patterns, we can prioritize those most likely to become effective drugs for subsequent experimental validation and guide the structural optimization of potential candidate drugs. We used AutodockVina 1.2.2 (http://autodock.scripps.edu/) for the molecular docking of ligands and target proteins [30]. Drug structure data were obtained from the PubChem compound database [31] (https://pubchem.ncbi.nlm.nih.gov/). Protein structure data were downloaded from the PDB (Protein Data Bank, http://www.rcsb.org/). The entire molecular docking process was visualized using Autodock Vina 1.2.2.

Results

Association of eQTL_eQTLgen with OC

We employed Summary-data-based Mendelian Randomization (SMR) as the primary analytical approach to investigate the causal relationship between eQTLs and OC (Supplementary Table 1). In the genome-wide association study (GWAS), we identified the top five genes (MTF2, FGFRL1, DHX58, DDX55, and HSPG2) with the smallest p-values, demonstrating the strongest statistical significance. These genes were located within or near loci relevant to OC (P_SMR < 0.05, P_HEIDI > 0.05) (Fig. 2A). Among the 11,112 eQTLs examined, 487 showed a causal association with OC (Supplementary Table 2). Specifically, MTF2 and ITGA5 exhibited a significant inverse causal relationship with OC risk (p < 0.001), whereas DHX58, DDX55, HSPG2, HID1, and MYD88 demonstrated a significant positive causal relationship (p < 0.001) (Fig. 2B).

Fig. 2.

SMR and HEIDE Analysis of molQTLs and OC and Intersection Analysis. A: Manhattan plot of blood eQTLs-OC SMR associations. B: Volcano plot of blood eQTLs-OC SMR positive findings. C: Circle heatmap of ovary tissue eQTLs-OC SMR positive findings. D: Polar bar plot of pQTLs-OC SMR positive findings. E: Venn plot of the intersection of analysis of blood eQTLs, ovary tissue eQTLs, pQTLs and OC SMR positive results. SMR=Summary-data-based Mendelian Randomization, OC=Ovarian cancer, eQTL=expression Quantitative Trait Locus, pQTL=protein Quantitative Trait Locus.

Association of eQTL_GTEx_Ovary with OC

Using SMR analysis, we further explored the causal relationship between eQTL_GTEx_Ovary and OC (Supplementary Table 3). Among 1,377 genes, 62 exhibited a causal association with OC (Fig. 2C), with 27 genes showing a positive causal relationship and 35 genes showing a negative causal relationship.

Association of pQTL with OC

We also applied SMR analysis to examine the causal relationship between pQTLs and OC (Supplementary Table 4). Among 2,218 pQTLs, 103 were found to be causally associated with OC (Supplementary Table 5, Fig. 2D), with 50 pQTLs exhibiting a positive causal relationship and 53 pQTLs exhibiting a negative causal relationship.

Identification of key genes

To identify key genes, we conducted an intersection analysis of eQTLs, eQTL_GTEx_Ovary, and pQTLs that were causally associated with OC in the three omics-based SMR analyses (Fig. 2E). This analysis identified DHX58 as the only gene consistently associated with OC risk across all three datasets (Table 1).

Table 1.

SMR analysis of DHX58 across multi-omics cohorts

Gene	qtl_name	Beta	OR	95%	P_value	P_HEIDI
DHX58	eQTLGen	0.2900	1.34	1.15-1.55	0.0002	0.1781
DHX58	eQTL_GTEx_Overy	0.2168	1.24	1.07-1.45	0.0049	0.6074
DHX58	pQTL_FELAND	0.2047	1.23	1.07-1.40	0.0028	0.4949

Mendelian randomization analysis of DHX58 and OC

Since SMR analysis is potentially sensitive to pleiotropy, we further employed MR to guarantee the soundness of our conclusions. Furthermore, after false discovery rate (FDR) correction, the IVW MR analysis confirmed that DHX58 was significantly associated with an increased risk of OC (OR = 1.30, 95% CI: 1.12–1.51, p = 0.0004, FDR = 0.0009). This association was further supported by the weighted median method (OR = 1.31, 95% CI: 1.13–1.52, p = 0.0003, FDR = 0.0006) and showed a consistent direction in the MR-Egger analysis (OR = 1. 43, 95% CI: 1.17–1.75, p = 0.0005) (Fig. 3A). We undertook sensitivity assessments to evaluate the robustness of our results. The results indicated no evidence of heterogeneity or pleiotropy (MR Egger: Q = 2.118, Q_pval = 0.3469, egger intercept: p = 0.4148) (Table 2).

Fig. 3.

A Forest plot of the causal link between DHX58 and OC. B Regional plot of colocalization evidence of DHX58 and OC

Table 2.

Sensitive analysis results for MR study

Exposure	Horizontal pleiotropy		Heterogencity
	Egger intercept	P_value	Q	P_value
DHX58	-0.0363	0.4148	2.118	0.3469

Colocalization analysis

Previous studies suggest that significant MR results may arise from strong linkage disequilibrium (LD) with a nearby causal SNP, rather than a direct causal effect, potentially leading to false-positive conclusions [32]. To assess whether the exposure and outcome share the same causal SNP, we performed colocalization analysis [26]. Additionally, studies have shown that proteins validated through both MR and colocalization analyses are more likely to be viable drug targets and significantly increase the success rate of drug approval [33].Our colocalization analysis provided strong evidence supporting the colocalization of DHX58 with OC (PP.H4 = 0.949), suggesting that DHX58 is a promising candidate drug target (Table 3, Fig. 3B).

Table 3.

Colocalization analysis of DHX58

Gene	PP.H0	PP.H1	PP.H2	PP.H3	PP.H4
DHX58	0.000	0.047	0.000	0.000	0.949

Transcriptomic data analysis

To explore the clinical relevance of DHX58 in OC, we conducted a comprehensive analysis integrating survival evaluation, GSEA and immune microenvironment assessment. We first performed batch effect correction on OC and normal ovarian tissue samples from the TCGA and GTEx databases (Supplementary Fig. S1–S2). Analysis of differential expression showed that DHX58 expression was markedly elevated in OC tissues relative to normal ovarian tissues (Fig. 4A). GSEA further indicated that high DHX58 expression was significantly associated with the Toll-like receptor (TLR) signaling pathway (ES = -0.6720, NES = -2.1365, FDR = 0.0019) (Fig. 4B, Supplementary Table 6). Kaplan-Meier survival analysis revealed that high DHX58 expression was significantly correlated with DSS (p < 0.0001) (Fig. 2D), DFS (p < 0.0001) (Fig. 4E), and PFS (p < 0.0001) (Fig. 4F). Although high DHX58 expression was associated with poorer OS, the difference was not statistically significant (p = 0.09) (Fig. 4C).

Fig. 4.

Violin plot of DHX58 expression in normal and OC samples. B GSEA of KEGG pathways based on DHX58 expression C-R Kaplan–Meier survival analyses. C-F Survival outcomes based on DHX58 expression. Significant associations were observed with DSS (D), DFI (E), and PFI (F) (p < 0.001), but not with OS (C, p > 0.05). G-J Survival outcomes based on ImmuneScore. Significant associations were observed with OS (G), DSS (H), DFI (I), and PFI (J) (all p < 0.001). K-N Survival analysis of DHX58 expression within the high ImmuneScore subgroup. Significant associations were observed for DSS (L) and DFI (M) (p < 0.001), but not for OS (K) or PFI (N) (p > 0.05). O-R Survival analysis of DHX58 expression within the low ImmuneScore subgroup. Significant associations were observed with OS (O), DSS (P), DFI (Q), and PFI (R) (all p < 0.001).. S Barplot showing the proportion of 22 immune cell subsets in OC samples. T Violin plot showed the ratio differentiation of 22 kinds of immune cells between OC samples with high or low DHX58 expression. U Correlation heatmap of immune cell subsets and each tiny box indicating the p value of correlation between two kinds of cells. V Scatter plot showing the correlation between DHX58 expression and 12 immune cell types. OS=Overall Survival, DSS=Disease-Specific Survival, DFI= Disease-free interval, PFI=Progression-free interval

To further explore the relationship between immune infiltration and prognosis, we stratified patients based on ImmuneScore. Patients with high ImmuneScore exhibited significantly poorer OS, DSS, DFS, and PFS (all p < 0.0001) (Fig. 4G–J), suggesting that increased immune infiltration may be associated with adverse prognosis in OC. Subgroup analysis within the high ImmuneScore group revealed that DHX58 expression was significantly associated with DSS (p < 0.0001) (Fig. 4L), with worse DSS observed in patients with high DHX58 expression. In contrast, for DFS, patients with high DHX58 expression appeared to have better outcomes compared to those with low expression (p < 0.0001) (Fig. 4M), suggesting a divergent effect of DHX58 depending on the clinical endpoint. However, no significant association was observed between DHX58 expression and OS (p = 0.5) (Fig. 4K) or PFS (p = 0.16) (Fig. 4N). Conversely, in low ImmuneScore group, low DHX58 expression was significantly associated with better OS (p < 0.0001) (Fig. 2O), DSS (p < 0.0001) (Fig. 4P), DFS (p < 0.0001) (Fig. 4Q), and PFS (p < 0.0001) (Fig. 4R).

The CIBERSORT algorithm was employed to analyze the composition of tumor-infiltrating immune cells to further validate the association between DHX58 expression and the tumor immune microenvironment. A comprehensive immune cell profile was constructed, characterizing 22 distinct immune cell subsets in OC samples (Fig. 4S). Comparative analysis between the high DHX58 and low DHX58 expression groups demonstrated significant differences in the infiltration levels of plasma cells, CD8 + T cells, CD4+ naïve T cells, T cells follicular helper, NK cells resting, monocytes, macrophages, dendritic cells resting, mast cells activated and neutrophils, suggesting a potential immunomodulatory role of DHX58 (Fig. 4T). The immune cell correlations are shown in Fig. 4U. Further correlation analysis confirmed that DHX58 expression exhibited the strongest positive correlation with macrophage M1 infiltration (r = 0.31, p = 1.08e-10). In contrast, its correlation with macrophages M0 (r = -0.15, p = 1.43e-03) was relatively weak (Fig. 4V).

Single-cell analysis reveals cellular heterogeneity and cell type–specific DHX58 expression in OC

After quality control, harmony integration, and clustering, a total of 17 distinct cell clusters were identified and visualized using UMAP (Fig. 5A, Supplementary Fig. S3–S5). Cell type annotation revealed a complex and heterogeneous cellular landscape encompassing immune, stromal, and epithelial compartments (Fig. 5B and C). Immune cells, including Treg cells, CD4⁺ T cells, and myeloid cells, constituted a substantial fraction of the OC microenvironment, whereas stromal populations such as fibroblasts and endothelial cells were present at relatively lower proportions. These findings indicate extensive remodeling of both immune and stromal compartments during OC progression. Visualization of DHX58 expression across the UMAP embedding demonstrated pronounced heterogeneity, with expression confined to specific cell populations rather than uniformly distributed across all cells (Fig. 5D). This pattern highlights a clear degree of cell type specificity. When stratified by tissue type, DHX58-positive cells were observed in both normal and OC samples. However, differences in expression density and spatial distribution were evident between conditions. In OC tissues, DHX58 expression appeared more prominent within Treg cells, CD4⁺ T cells, NK cells, and myeloid cells compared with normal tissues, suggesting that DHX58 expression may be modulated by the tumor microenvironment.

Fig. 5.

Single-cell transcriptomic landscape of OC reveals cell type–specific expression of DHX58. A: UMAP visualization of single cells from normal ovarian tissue and OC. B: UMAP projection of single cells with colors indicating annotated cell types. C: Stacked bar plot showing the proportion of different cell types in normal and OC. D: UMAP plot showing the expression pattern of DHX58 across different cell types in normal and tumor tissues.

Phenome-wide association study

To assess potential side effects of targeting DHX58, we performed a phenotype-wide association study (PheWAS) using the PheWAS Portal database. No significant direct associations were identified between DHX58 and other phenotypes (Fig. 6, Supplementary Table 7). These results further support the reliability of our findings and suggest that, if DHX58 is a viable therapeutic target, its likelihood of causing adverse drug reactions or pleiotropic effects is minimal.

Fig. 6.

Binary traits PheWAS association with DHX58

Candidate drug prediction

To evaluate the potential of DHX58 as a drug target, we conducted a comprehensive assessment of its protein-drug interaction properties. Using the Enrichr platform, which integrates the DSigDB drug database, we identified potential candidate drugs that may interact with DHX58. The top four candidate drugs were Suloctidil, TPEN, Zinc sulfate, and Cupric oxide (Supplementary Table 8).

Molecular docking analysis

To assess the binding affinity and druggability of the candidate compounds for DHX58, we conducted molecular docking simulations. The 3D structures of the target drugs were retrieved from PubChem, and docking simulations were performed using AutoDock Vina v.1.2.2 to evaluate interactions such as binding sites, hydrogen bonding, and electrostatic interactions, while also calculating binding free energy. Based on binding energy and toxicity data, two candidate drugs were selected for further analysis (Fig. 7, Table 4). Notably, DHX58 exhibited the strongest binding affinity with TPEN, with a binding energy of − 5.08 kcal/mol, suggesting its potential as a lead compound for future drug development.

Fig. 7.

Docking results of Suloctidil. B: Docking results of TPEN

Table 4.

Molecular docking results of DHX58 with drug molecules

Drug	PubChem ID	Target	PDB ID	Binding energy (kcal/mol)
Suloctidil	657255	DHX58	3EQT	-4.64
TPEN	5519	DHX58	3EQT	-5.08

Discussion

Our study utilized large publicly available genetic datasets and employed MR and SMR to investigate the causal relationships between eQTL, eQTL_Ovary_GTEx, and pQTLs with OC. Additionally, we validated the impact of candidate target genes on OC patient by integrated tumor data analysis. Through the prognostic significance of DHX58 in OC and its potential association with immune infiltration, we found that DHX58 is significantly upregulated in OC, and DHX58 was enriched in the Toll-like receptor (TLR) signaling pathway. Notably, our results revealed that high expression of DHX58 expression is correlated with poorer DSS, DFS, and PFS. However, its relationship with OS did not reach statistical significance (p = 0.09). This may be due to the broader nature of OS as a clinical endpoint, which incorporates non-cancer-related causes of death. While immune infiltration is often associated with favorable outcomes in cancer, our results indicate that a higher ImmuneScore correlates with poorer prognosis in OC. This apparent contradiction may reflect the fact that not all immune cells are functionally beneficial—consistent with prior findings that the composition and activation state of the tumor-infiltrating immune cells play a critical role in shaping clinical outcomes [34]. DHX58 consistently exhibited a negative prognostic effect in the low ImmuneScore group across all survival endpoints, while showing a more complex pattern in the high ImmuneScore group, particularly for DFI. The robustness of our results was ensured through MR analysis and integrated tumor data analysis. In addition, single-cell RNA sequencing analysis was performed to delineate the cellular context of DHX58 expression within the OC microenvironment, providing cell-type–specific insights that complement the bulk-level findings. Finally, we predicted potential drugs targeting DHX58 and conducted molecular docking analysis, further supporting its therapeutic potential.

DHX58, a member of the RIG-I (DDX58)-like receptor family, also known as LGP2, belongs to the RNA helicase family. As a key intracellular sensor, it plays a role in recognizing invading RNA viruses within the host innate immune system. Due to its RNA-binding ability, DHX58 is classified as a non-classical RNA-binding protein (RBP) [35]. The RNA helicase family includes DEAD/DExH-box domain-containing helicases (DDX/DHX), which are crucial in RNA binding and unwinding and play essential roles in various RNA metabolic processes, such as recognition, modification, splicing, transport, degradation, and translation [36, 37]. Some family members are also categorized as interferon (IFN)-stimulated genes (ISGs), whose expression is regulated by IFN stimulation. Certain members, such as DDX58 (RIG-I) [38, 39]. and DHX58 (LGP2) [40, 41], promote IFN production. However, the precise role of DHX58 in the RIG-I-like receptor signaling pathway remains controversial [42]. Some studies suggest that DHX58 inhibits this pathway through multiple mechanisms, such as acting as a competitive receptor that blocks viral dsRNA binding to RIG-I and interfering with RIG-I–Mavs signaling [43, 44]. Conversely, other studies have proposed that DHX58 may activate the RIG-I pathway, enhancing the innate immune response in neuroblastoma and inhibiting tumor progression [45].

Our findings demonstrated a significant association between elevated DHX58 expression and the TLR signaling pathway. As pattern recognition receptors, TLRs play a pivotal role in identifying pathogens and initiating innate immune responses [46].Extensive research on the role of TLR signaling in cancer has revealed dual effects, with both tumor-suppressive and tumor-promoting functions depending on the specific TLR subtype and tumor context [47, 48]. Activation of TLRs can enhance anti-tumor immunity by stimulating immune cells or directly inducing tumor cell apoptosis. However, TLRs also contribute to tumor progression. For instance, in lung cancer, TLR4 activation by lipopolysaccharide (LPS) promotes the secretion of immunosuppressive cytokines, facilitating immune evasion [49]. Similarly, in breast cancer, LPS stimulation of TLR4-expressing tumor cells enhances proliferation through the upregulation of IL-8 and IL-6 production [50, 51]. Importantly, our single-cell RNA sequencing analysis provided cell-type–specific insights into the biological context of DHX58 expression within the OC microenvironment. DHX58 expression was predominantly enriched in immune cell populations rather than malignant epithelial cells, particularly within myeloid-lineage cells, supporting a functional role of DHX58 in regulating tumor–immune interactions. This observation is consistent with our bulk transcriptomic and immune deconvolution analyses, which revealed a negative association between DHX58 expression and patient survival outcomes that was further modulated by immune infiltration levels. Specifically, CIBERSORT profiling showed that high DHX58 expression is accompanied by increased infiltration of monocytes, macrophages, resting dendritic cells, neutrophils and activated mast cells. Among them, macrophages, dendritic cells and neutrophils are differentiated from myeloid-derived suppressor cells (MDSCs) [52]. MDSCs serve as facilitators of tumorigenesis by promoting an immunosuppressive microenvironment [53]. Integrating single-cell–level localization with bulk immune profiling, we propose that DHX58 may influence the balance between antitumor immunity and immune tolerance in OC through modulation of TLR-related signaling pathways in immune cells. Rather than universally enhancing immune activation, DHX58 may preferentially regulate specific TLR subtypes in myeloid cells, thereby promoting the production of immunosuppressive cytokines and recruitment of pro-tumorigenic immune populations. This mechanism could explain why increased immune infiltration, particularly in the context of high DHX58 expression, is associated with poorer clinical outcomes. Collectively, these findings—consistent with previous reports—suggest that DHX58 is not only a prognostic biomarker but also a potential therapeutic target; its inhibition might remodel the TLR/RIG‑I axis and reinvigorate antitumor immune responses in OC.

PheWAS analysis indicated minimal potential side effects of DHX58, significantly reducing the likelihood of bias caused by pleiotropy. Our study revealed that the small molecules, TPEN and Suloctidil have strong binding affinities with DHX58, suggesting that DHX58 might be a viable potential target for cancer therapy.

Our study has several strengths. First, we performed cross-validation across multiple omics layers, considering a gene as a druggable target only when it demonstrated significance across all three omics datasets. This approach enhances the reliability of our findings and reduces false positives, potentially increasing the success rate of clinical trials. Second, we combined multiple analytical methods, including SMR analysis, MR analysis, HEIDI testing, colocalization analysis, integrated tumor data analysis, and PheWAS, to further improve the robustness of drug target selection. Lastly, drug prediction and molecular docking results provided additional support for the potential of DHX58 as a drug target.

However, our study has certain limitations. First, as our analysis was based on European population data, the generalizability of our findings to other ethnic groups remains uncertain. Future studies should incorporate more diverse populations to enhance the applicability of our results. Second, while MR analysis provides important insights into potential causal relationships, its assumptions may not fully align with real-world clinical trial conditions, requiring cautious interpretation and application of MR findings. Third, despite our efforts to use reverse MR for validation in MR, we were unable to include this analysis due to the lack of appropriate instrumental variables. Finally, our study is primarily computational and lacks experimental validation. Although the multi-omics evidence strongly supports the role of DHX58 in OC, functional validation in vitro and in vivo is essential to confirm its mechanistic role. Future studies should explore CRISPR/Cas9-mediated gene editing in OC cell lines, RNA interference approaches, and xenograft mouse models to investigate the effect of DHX58 on tumor growth, immune cell recruitment, and cytokine expression within the tumor microenvironment.

Conclusion

Multi-omics Mendelian randomization analysis and integrated tumor data analysis found a strong correlation between DHX58 expression and increased OC risk. These findings underscore DHX58 as a promising therapeutic target, warranting further clinical and mechanistic investigations. However, further studies are needed to elucidate the underlying mechanisms of this association.

Data Availability

All data supporting the findings of this study are included in this article and its supplementary materials.

Authors' contributions

Yanbin Chen: Conceptualization; Data interpretation; Writing-Original draft preparation; Software; Resources.Yanbin Chen, Fengyan Ma, Xifeng Huang, Zhemin Zhuang, Likun Lin, Yuming Liao: Methodology, Investigation; Supervision.Jinhong Wang: Conceptualization; Writing-review & editing; Funding acquisition.

Funding Statement

This work was supported by the GuangDong Basic and Applied Basic Research Foundation (Grant No. 2024A1515220033). Medical Scientific Research Foundation of Guangdong Province of China (Grant No. A2024154).

Data availability

All data supporting the findings of this study are included in this article and its supplementary materials.

Declarations

Ethics approval and consent to participate

All datasets complied with the ethical guidelines established in their original studies, so no additional ethical approval was required for this reanalysis.

Consent for publication

All authors approved the final version and agreed to be responsible for the study.

Competing interests

The authors declare no competing interests.